diff options
Diffstat (limited to 'llvm')
167 files changed, 5968 insertions, 4346 deletions
diff --git a/llvm/docs/MLGO.rst b/llvm/docs/MLGO.rst index 965a21b..bf3de11 100644 --- a/llvm/docs/MLGO.rst +++ b/llvm/docs/MLGO.rst @@ -508,7 +508,7 @@ embeddings can be computed and accessed via an ``ir2vec::Embedder`` instance. .. code-block:: c++ - const ir2vec::Embedding &FuncVector = Emb->getFunctionVector(); + ir2vec::Embedding FuncVector = Emb->getFunctionVector(); Currently, ``Embedder`` can generate embeddings at three levels: Instructions, Basic Blocks, and Functions. Appropriate getters are provided to access the diff --git a/llvm/include/llvm/ADT/Bitset.h b/llvm/include/llvm/ADT/Bitset.h index b1e539e..1d4cbf8 100644 --- a/llvm/include/llvm/ADT/Bitset.h +++ b/llvm/include/llvm/ADT/Bitset.h @@ -38,14 +38,27 @@ class Bitset { static constexpr unsigned NumWords = (NumBits + BitwordBits - 1) / BitwordBits; -protected: using StorageType = std::array<BitWord, NumWords>; - -private: StorageType Bits{}; protected: - constexpr Bitset(const StorageType &B) : Bits{B} {} + constexpr Bitset(const std::array<uint64_t, (NumBits + 63) / 64> &B) { + if constexpr (sizeof(BitWord) == sizeof(uint64_t)) { + for (size_t I = 0; I != B.size(); ++I) + Bits[I] = B[I]; + } else { + unsigned BitsToAssign = NumBits; + for (size_t I = 0; I != B.size() && BitsToAssign; ++I) { + uint64_t Elt = B[I]; + // On a 32-bit system the storage type will be 32-bit, so we may only + // need half of a uint64_t. + for (size_t offset = 0; offset != 2 && BitsToAssign; ++offset) { + Bits[2 * I + offset] = static_cast<uint32_t>(Elt >> (32 * offset)); + BitsToAssign = BitsToAssign >= 32 ? BitsToAssign - 32 : 0; + } + } + } + } public: constexpr Bitset() = default; diff --git a/llvm/include/llvm/ADT/ImmutableSet.h b/llvm/include/llvm/ADT/ImmutableSet.h index 017585a4..310539f 100644 --- a/llvm/include/llvm/ADT/ImmutableSet.h +++ b/llvm/include/llvm/ADT/ImmutableSet.h @@ -21,7 +21,9 @@ #include "llvm/ADT/iterator.h" #include "llvm/Support/Allocator.h" #include "llvm/Support/Compiler.h" +#include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/Signals.h" #include <cassert> #include <cstdint> #include <functional> diff --git a/llvm/include/llvm/Analysis/DependenceAnalysis.h b/llvm/include/llvm/Analysis/DependenceAnalysis.h index a2ca695..18a8f8a 100644 --- a/llvm/include/llvm/Analysis/DependenceAnalysis.h +++ b/llvm/include/llvm/Analysis/DependenceAnalysis.h @@ -160,7 +160,7 @@ public: /// getDVEntry - Returns the DV entry associated with a regular or a /// SameSD level - DVEntry getDVEntry(unsigned Level, bool isSameSD) const; + DVEntry getDVEntry(unsigned Level, bool IsSameSD) const; /// getDirection - Returns the direction associated with a particular /// common or SameSD level. @@ -234,7 +234,7 @@ public: /// dumpImp - For debugging purposes. Dumps a dependence to OS with or /// without considering the SameSD levels. - void dumpImp(raw_ostream &OS, bool SameSD = false) const; + void dumpImp(raw_ostream &OS, bool IsSameSD = false) const; protected: Instruction *Src, *Dst; @@ -282,8 +282,8 @@ public: /// getDVEntry - Returns the DV entry associated with a regular or a /// SameSD level. - DVEntry getDVEntry(unsigned Level, bool isSameSD) const { - if (!isSameSD) { + DVEntry getDVEntry(unsigned Level, bool IsSameSD) const { + if (!IsSameSD) { assert(0 < Level && Level <= Levels && "Level out of range"); return DV[Level - 1]; } else { diff --git a/llvm/include/llvm/Analysis/IR2Vec.h b/llvm/include/llvm/Analysis/IR2Vec.h index 81409df..6bc51fe 100644 --- a/llvm/include/llvm/Analysis/IR2Vec.h +++ b/llvm/include/llvm/Analysis/IR2Vec.h @@ -533,21 +533,20 @@ protected: /// in the IR instructions to generate the vector representation. const float OpcWeight, TypeWeight, ArgWeight; - // Utility maps - these are used to store the vector representations of - // instructions, basic blocks and functions. - mutable Embedding FuncVector; - mutable BBEmbeddingsMap BBVecMap; - mutable InstEmbeddingsMap InstVecMap; - - LLVM_ABI Embedder(const Function &F, const Vocabulary &Vocab); + LLVM_ABI Embedder(const Function &F, const Vocabulary &Vocab) + : F(F), Vocab(Vocab), Dimension(Vocab.getDimension()), + OpcWeight(ir2vec::OpcWeight), TypeWeight(ir2vec::TypeWeight), + ArgWeight(ir2vec::ArgWeight) {} - /// Function to compute embeddings. It generates embeddings for all - /// the instructions and basic blocks in the function F. - void computeEmbeddings() const; + /// Function to compute embeddings. + Embedding computeEmbeddings() const; /// Function to compute the embedding for a given basic block. + Embedding computeEmbeddings(const BasicBlock &BB) const; + + /// Function to compute the embedding for a given instruction. /// Specific to the kind of embeddings being computed. - virtual void computeEmbeddings(const BasicBlock &BB) const = 0; + virtual Embedding computeEmbeddings(const Instruction &I) const = 0; public: virtual ~Embedder() = default; @@ -556,23 +555,27 @@ public: LLVM_ABI static std::unique_ptr<Embedder> create(IR2VecKind Mode, const Function &F, const Vocabulary &Vocab); - /// Returns a map containing instructions and the corresponding embeddings for - /// the function F if it has been computed. If not, it computes the embeddings - /// for the function and returns the map. - LLVM_ABI const InstEmbeddingsMap &getInstVecMap() const; - - /// Returns a map containing basic block and the corresponding embeddings for - /// the function F if it has been computed. If not, it computes the embeddings - /// for the function and returns the map. - LLVM_ABI const BBEmbeddingsMap &getBBVecMap() const; + /// Computes and returns the embedding for a given instruction in the function + /// F + LLVM_ABI Embedding getInstVector(const Instruction &I) const { + return computeEmbeddings(I); + } - /// Returns the embedding for a given basic block in the function F if it has - /// been computed. If not, it computes the embedding for the basic block and - /// returns it. - LLVM_ABI const Embedding &getBBVector(const BasicBlock &BB) const; + /// Computes and returns the embedding for a given basic block in the function + /// F + LLVM_ABI Embedding getBBVector(const BasicBlock &BB) const { + return computeEmbeddings(BB); + } /// Computes and returns the embedding for the current function. - LLVM_ABI const Embedding &getFunctionVector() const; + LLVM_ABI Embedding getFunctionVector() const { return computeEmbeddings(); } + + /// Invalidate embeddings if cached. The embeddings may not be relevant + /// anymore when the IR changes due to transformations. In such cases, the + /// cached embeddings should be invalidated to ensure + /// correctness/recomputation. This is a no-op for SymbolicEmbedder but + /// removes all the cached entries in FlowAwareEmbedder. + virtual void invalidateEmbeddings() { return; } }; /// Class for computing the Symbolic embeddings of IR2Vec. @@ -580,7 +583,7 @@ public: /// representations obtained from the Vocabulary. class LLVM_ABI SymbolicEmbedder : public Embedder { private: - void computeEmbeddings(const BasicBlock &BB) const override; + Embedding computeEmbeddings(const Instruction &I) const override; public: SymbolicEmbedder(const Function &F, const Vocabulary &Vocab) @@ -592,11 +595,15 @@ public: /// embeddings, and additionally capture the flow information in the IR. class LLVM_ABI FlowAwareEmbedder : public Embedder { private: - void computeEmbeddings(const BasicBlock &BB) const override; + // FlowAware embeddings would benefit from caching instruction embeddings as + // they are reused while computing the embeddings of other instructions. + mutable InstEmbeddingsMap InstVecMap; + Embedding computeEmbeddings(const Instruction &I) const override; public: FlowAwareEmbedder(const Function &F, const Vocabulary &Vocab) : Embedder(F, Vocab) {} + void invalidateEmbeddings() override { InstVecMap.clear(); } }; } // namespace ir2vec diff --git a/llvm/include/llvm/AsmParser/LLToken.h b/llvm/include/llvm/AsmParser/LLToken.h index d976d40e..6de99fe 100644 --- a/llvm/include/llvm/AsmParser/LLToken.h +++ b/llvm/include/llvm/AsmParser/LLToken.h @@ -488,27 +488,28 @@ enum Kind { SummaryID, // ^42 // String valued tokens (StrVal). - LabelStr, // foo: - GlobalVar, // @foo @"foo" - ComdatVar, // $foo - LocalVar, // %foo %"foo" - MetadataVar, // !foo - StringConstant, // "foo" - DwarfTag, // DW_TAG_foo - DwarfAttEncoding, // DW_ATE_foo - DwarfVirtuality, // DW_VIRTUALITY_foo - DwarfLang, // DW_LANG_foo - DwarfCC, // DW_CC_foo - EmissionKind, // lineTablesOnly - NameTableKind, // GNU - FixedPointKind, // Fixed point - DwarfOp, // DW_OP_foo - DIFlag, // DIFlagFoo - DISPFlag, // DISPFlagFoo - DwarfMacinfo, // DW_MACINFO_foo - ChecksumKind, // CSK_foo - DbgRecordType, // dbg_foo - DwarfEnumKind, // DW_APPLE_ENUM_KIND_foo + LabelStr, // foo: + GlobalVar, // @foo @"foo" + ComdatVar, // $foo + LocalVar, // %foo %"foo" + MetadataVar, // !foo + StringConstant, // "foo" + DwarfTag, // DW_TAG_foo + DwarfAttEncoding, // DW_ATE_foo + DwarfVirtuality, // DW_VIRTUALITY_foo + DwarfLang, // DW_LANG_foo + DwarfSourceLangName, // DW_LNAME_foo + DwarfCC, // DW_CC_foo + EmissionKind, // lineTablesOnly + NameTableKind, // GNU + FixedPointKind, // Fixed point + DwarfOp, // DW_OP_foo + DIFlag, // DIFlagFoo + DISPFlag, // DISPFlagFoo + DwarfMacinfo, // DW_MACINFO_foo + ChecksumKind, // CSK_foo + DbgRecordType, // dbg_foo + DwarfEnumKind, // DW_APPLE_ENUM_KIND_foo // Type valued tokens (TyVal). Type, diff --git a/llvm/include/llvm/CodeGen/SlotIndexes.h b/llvm/include/llvm/CodeGen/SlotIndexes.h index 1e270b4..780bd82 100644 --- a/llvm/include/llvm/CodeGen/SlotIndexes.h +++ b/llvm/include/llvm/CodeGen/SlotIndexes.h @@ -467,16 +467,27 @@ class raw_ostream; return getMBBRange(mbb).first; } - /// Returns the last index in the given basic block number. + /// Returns the index past the last valid index in the given basic block. SlotIndex getMBBEndIdx(unsigned Num) const { return getMBBRange(Num).second; } - /// Returns the last index in the given basic block. + /// Returns the index past the last valid index in the given basic block. SlotIndex getMBBEndIdx(const MachineBasicBlock *mbb) const { return getMBBRange(mbb).second; } + /// Returns the last valid index in the given basic block. + /// This index corresponds to the dead slot of the last non-debug + /// instruction and can be used to find live-out ranges of the block. Note + /// that getMBBEndIdx returns the start index of the next block, which is + /// also used as the start index for segments with phi-def values. If the + /// basic block doesn't contain any non-debug instructions, this returns + /// the same as getMBBStartIdx.getDeadSlot(). + SlotIndex getMBBLastIdx(const MachineBasicBlock *MBB) const { + return getMBBEndIdx(MBB).getPrevSlot(); + } + /// Iterator over the idx2MBBMap (sorted pairs of slot index of basic block /// begin and basic block) using MBBIndexIterator = SmallVectorImpl<IdxMBBPair>::const_iterator; diff --git a/llvm/include/llvm/IR/ConstantFPRange.h b/llvm/include/llvm/IR/ConstantFPRange.h index 930c6f9..4a54caa 100644 --- a/llvm/include/llvm/IR/ConstantFPRange.h +++ b/llvm/include/llvm/IR/ConstantFPRange.h @@ -200,6 +200,12 @@ public: /// with another range. The resultant range is guaranteed to include the /// elements of both sets, but may contain more. LLVM_ABI ConstantFPRange unionWith(const ConstantFPRange &CR) const; + + /// Calculate absolute value range. + LLVM_ABI ConstantFPRange abs() const; + + /// Calculate range of negated values. + LLVM_ABI ConstantFPRange negate() const; }; inline raw_ostream &operator<<(raw_ostream &OS, const ConstantFPRange &CR) { diff --git a/llvm/include/llvm/IR/PatternMatch.h b/llvm/include/llvm/IR/PatternMatch.h index 2e31fe5..7934277 100644 --- a/llvm/include/llvm/IR/PatternMatch.h +++ b/llvm/include/llvm/IR/PatternMatch.h @@ -2184,6 +2184,18 @@ inline PtrToIntSameSize_match<OpTy> m_PtrToIntSameSize(const DataLayout &DL, return PtrToIntSameSize_match<OpTy>(DL, Op); } +/// Matches PtrToAddr. +template <typename OpTy> +inline CastOperator_match<OpTy, Instruction::PtrToAddr> +m_PtrToAddr(const OpTy &Op) { + return CastOperator_match<OpTy, Instruction::PtrToAddr>(Op); +} + +/// Matches PtrToInt or PtrToAddr. +template <typename OpTy> inline auto m_PtrToIntOrAddr(const OpTy &Op) { + return m_CombineOr(m_PtrToInt(Op), m_PtrToAddr(Op)); +} + /// Matches IntToPtr. template <typename OpTy> inline CastOperator_match<OpTy, Instruction::IntToPtr> diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.h b/llvm/include/llvm/IR/RuntimeLibcalls.h index 93183bc..307cc66 100644 --- a/llvm/include/llvm/IR/RuntimeLibcalls.h +++ b/llvm/include/llvm/IR/RuntimeLibcalls.h @@ -15,6 +15,7 @@ #define LLVM_IR_RUNTIME_LIBCALLS_H #include "llvm/ADT/ArrayRef.h" +#include "llvm/ADT/Bitset.h" #include "llvm/ADT/Sequence.h" #include "llvm/ADT/StringTable.h" #include "llvm/IR/CallingConv.h" @@ -54,8 +55,22 @@ static inline auto libcall_impls() { static_cast<RTLIB::LibcallImpl>(RTLIB::NumLibcallImpls)); } +/// Manage a bitset representing the list of available libcalls for a module. +class LibcallImplBitset : public Bitset<RTLIB::NumLibcallImpls> { +public: + constexpr LibcallImplBitset() = default; + constexpr LibcallImplBitset( + const std::array<uint64_t, (RTLIB::NumLibcallImpls + 63) / 64> &Src) + : Bitset(Src) {} +}; + /// A simple container for information about the supported runtime calls. struct RuntimeLibcallsInfo { +private: + /// Bitset of libcalls a module may emit a call to. + LibcallImplBitset AvailableLibcallImpls; + +public: explicit RuntimeLibcallsInfo( const Triple &TT, ExceptionHandling ExceptionModel = ExceptionHandling::None, @@ -129,6 +144,14 @@ struct RuntimeLibcallsInfo { return getLibcallName(RTLIB::MEMMOVE); } + bool isAvailable(RTLIB::LibcallImpl Impl) const { + return AvailableLibcallImpls.test(Impl); + } + + void setAvailable(RTLIB::LibcallImpl Impl) { + AvailableLibcallImpls.set(Impl); + } + /// Return the libcall provided by \p Impl static RTLIB::Libcall getLibcallFromImpl(RTLIB::LibcallImpl Impl) { return ImplToLibcall[Impl]; @@ -225,8 +248,7 @@ private: /// Return true if the target has sincosf/sincos/sincosl functions static bool hasSinCos(const Triple &TT) { - return TT.isGNUEnvironment() || TT.isOSFuchsia() || - (TT.isAndroid() && !TT.isAndroidVersionLT(9)); + return TT.isGNUEnvironment() || TT.isOSFuchsia() || TT.isAndroid(); } static bool hasSinCos_f32_f64(const Triple &TT) { diff --git a/llvm/include/llvm/Target/GlobalISel/Combine.td b/llvm/include/llvm/Target/GlobalISel/Combine.td index e2b7a5e..3d21f52 100644 --- a/llvm/include/llvm/Target/GlobalISel/Combine.td +++ b/llvm/include/llvm/Target/GlobalISel/Combine.td @@ -695,6 +695,7 @@ def constant_fold_fabs : constant_fold_unary_fp_op_rule<G_FABS>; def constant_fold_fsqrt : constant_fold_unary_fp_op_rule<G_FSQRT>; def constant_fold_flog2 : constant_fold_unary_fp_op_rule<G_FLOG2>; def constant_fold_fptrunc : constant_fold_unary_fp_op_rule<G_FPTRUNC>; +def constant_fold_fpext : constant_fold_unary_fp_op_rule<G_FPEXT>; // Fold constant zero int to fp conversions. class itof_const_zero_fold_rule<Instruction opcode> : GICombineRule < @@ -713,6 +714,7 @@ def constant_fold_fp_ops : GICombineGroup<[ constant_fold_fsqrt, constant_fold_flog2, constant_fold_fptrunc, + constant_fold_fpext, itof_const_zero_fold_si, itof_const_zero_fold_ui ]>; diff --git a/llvm/include/llvm/Target/TargetSelectionDAG.td b/llvm/include/llvm/Target/TargetSelectionDAG.td index 7bc90d4..774063b 100644 --- a/llvm/include/llvm/Target/TargetSelectionDAG.td +++ b/llvm/include/llvm/Target/TargetSelectionDAG.td @@ -116,7 +116,7 @@ def SDTIntBinOp : SDTypeProfile<1, 2, [ // add, and, or, xor, udiv, etc. SDTCisSameAs<0, 1>, SDTCisSameAs<0, 2>, SDTCisInt<0> ]>; def SDTIntShiftOp : SDTypeProfile<1, 2, [ // shl, sra, srl - SDTCisSameAs<0, 1>, SDTCisInt<0>, SDTCisInt<2> + SDTCisSameAs<0, 1>, SDTCisInt<0>, SDTCisInt<2>, SDTCisSameNumEltsAs<0, 2> ]>; def SDTIntShiftPairOp : SDTypeProfile<2, 3, [ // shl_parts, sra_parts, srl_parts SDTCisInt<0>, SDTCisSameAs<1, 0>, diff --git a/llvm/include/llvm/Transforms/Scalar/GVN.h b/llvm/include/llvm/Transforms/Scalar/GVN.h index 74a4d6c..bc0f108 100644 --- a/llvm/include/llvm/Transforms/Scalar/GVN.h +++ b/llvm/include/llvm/Transforms/Scalar/GVN.h @@ -28,6 +28,7 @@ #include <cstdint> #include <optional> #include <utility> +#include <variant> #include <vector> namespace llvm { @@ -322,11 +323,6 @@ private: }; LeaderMap LeaderTable; - // Block-local map of equivalent values to their leader, does not - // propagate to any successors. Entries added mid-block are applied - // to the remaining instructions in the block. - SmallMapVector<Value *, Value *, 4> ReplaceOperandsWithMap; - // Map the block to reversed postorder traversal number. It is used to // find back edge easily. DenseMap<AssertingVH<BasicBlock>, uint32_t> BlockRPONumber; @@ -402,9 +398,9 @@ private: void verifyRemoved(const Instruction *I) const; bool splitCriticalEdges(); BasicBlock *splitCriticalEdges(BasicBlock *Pred, BasicBlock *Succ); - bool replaceOperandsForInBlockEquality(Instruction *I) const; - bool propagateEquality(Value *LHS, Value *RHS, const BasicBlockEdge &Root, - bool DominatesByEdge); + bool + propagateEquality(Value *LHS, Value *RHS, + const std::variant<BasicBlockEdge, Instruction *> &Root); bool processFoldableCondBr(BranchInst *BI); void addDeadBlock(BasicBlock *BB); void assignValNumForDeadCode(); diff --git a/llvm/include/llvm/Transforms/Utils/Local.h b/llvm/include/llvm/Transforms/Utils/Local.h index 3f5f427..9acfd87 100644 --- a/llvm/include/llvm/Transforms/Utils/Local.h +++ b/llvm/include/llvm/Transforms/Utils/Local.h @@ -452,6 +452,11 @@ LLVM_ABI unsigned replaceDominatedUsesWith(Value *From, Value *To, LLVM_ABI unsigned replaceDominatedUsesWith(Value *From, Value *To, DominatorTree &DT, const BasicBlock *BB); +/// Replace each use of 'From' with 'To' if that use is dominated by the +/// given instruction. Returns the number of replacements made. +LLVM_ABI unsigned replaceDominatedUsesWith(Value *From, Value *To, + DominatorTree &DT, + const Instruction *I); /// Replace each use of 'From' with 'To' if that use is dominated by /// the given edge and the callback ShouldReplace returns true. Returns the /// number of replacements made. @@ -464,6 +469,12 @@ LLVM_ABI unsigned replaceDominatedUsesWithIf( LLVM_ABI unsigned replaceDominatedUsesWithIf( Value *From, Value *To, DominatorTree &DT, const BasicBlock *BB, function_ref<bool(const Use &U, const Value *To)> ShouldReplace); +/// Replace each use of 'From' with 'To' if that use is dominated by +/// the given instruction and the callback ShouldReplace returns true. Returns +/// the number of replacements made. +LLVM_ABI unsigned replaceDominatedUsesWithIf( + Value *From, Value *To, DominatorTree &DT, const Instruction *I, + function_ref<bool(const Use &U, const Value *To)> ShouldReplace); /// Return true if this call calls a gc leaf function. /// diff --git a/llvm/include/llvm/Transforms/Utils/MisExpect.h b/llvm/include/llvm/Transforms/Utils/MisExpect.h index e9fba47..2d35d92 100644 --- a/llvm/include/llvm/Transforms/Utils/MisExpect.h +++ b/llvm/include/llvm/Transforms/Utils/MisExpect.h @@ -22,8 +22,7 @@ #include "llvm/IR/Instructions.h" #include "llvm/IR/LLVMContext.h" -namespace llvm { -namespace misexpect { +namespace llvm::misexpect { /// checkBackendInstrumentation - compares PGO counters to the thresholds used /// for llvm.expect and warns if the PGO counters are outside of the expected @@ -34,8 +33,8 @@ namespace misexpect { /// /// \param I The Instruction being checked /// \param RealWeights A vector of profile weights for each target block -void checkBackendInstrumentation(Instruction &I, - const llvm::ArrayRef<uint32_t> RealWeights); +void checkBackendInstrumentation(const Instruction &I, + ArrayRef<uint32_t> RealWeights); /// checkFrontendInstrumentation - compares PGO counters to the thresholds used /// for llvm.expect and warns if the PGO counters are outside of the expected @@ -48,8 +47,8 @@ void checkBackendInstrumentation(Instruction &I, /// \param I The Instruction being checked /// \param ExpectedWeights A vector of the expected weights for each target /// block, this determines the threshold values used when emitting diagnostics -void checkFrontendInstrumentation(Instruction &I, - const ArrayRef<uint32_t> ExpectedWeights); +void checkFrontendInstrumentation(const Instruction &I, + ArrayRef<uint32_t> ExpectedWeights); /// veryifyMisExpect - compares RealWeights to the thresholds used /// for llvm.expect and warns if the PGO counters are outside of the expected @@ -58,8 +57,8 @@ void checkFrontendInstrumentation(Instruction &I, /// \param I The Instruction being checked /// \param RealWeights A vector of profile weights from the profile data /// \param ExpectedWeights A vector of the weights attatch by llvm.expect -void verifyMisExpect(Instruction &I, ArrayRef<uint32_t> RealWeights, - const ArrayRef<uint32_t> ExpectedWeights); +void verifyMisExpect(const Instruction &I, ArrayRef<uint32_t> RealWeights, + ArrayRef<uint32_t> ExpectedWeights); /// checkExpectAnnotations - compares PGO counters to the thresholds used /// for llvm.expect and warns if the PGO counters are outside of the expected @@ -72,11 +71,10 @@ void verifyMisExpect(Instruction &I, ArrayRef<uint32_t> RealWeights, /// \param I The Instruction being checked /// \param ExistingWeights A vector of profile weights for each target block /// \param IsFrontend A boolean describing if this is Frontend instrumentation -void checkExpectAnnotations(Instruction &I, - const ArrayRef<uint32_t> ExistingWeights, +void checkExpectAnnotations(const Instruction &I, + ArrayRef<uint32_t> ExistingWeights, bool IsFrontend); -} // namespace misexpect -} // namespace llvm +} // namespace llvm::misexpect #endif diff --git a/llvm/lib/Analysis/DependenceAnalysis.cpp b/llvm/lib/Analysis/DependenceAnalysis.cpp index 1f0da8d1..8d20b0e 100644 --- a/llvm/lib/Analysis/DependenceAnalysis.cpp +++ b/llvm/lib/Analysis/DependenceAnalysis.cpp @@ -275,7 +275,7 @@ bool Dependence::isAnti() const { // if no subscript in the source or destination mention the induction // variable associated with the loop at this level. // Leave this out of line, so it will serve as a virtual method anchor -bool Dependence::isScalar(unsigned level, bool isSameSD) const { return false; } +bool Dependence::isScalar(unsigned level, bool IsSameSD) const { return false; } //===----------------------------------------------------------------------===// // FullDependence methods @@ -351,38 +351,38 @@ bool FullDependence::normalize(ScalarEvolution *SE) { // getDirection - Returns the direction associated with a particular common or // SameSD level. -unsigned FullDependence::getDirection(unsigned Level, bool isSameSD) const { - return getDVEntry(Level, isSameSD).Direction; +unsigned FullDependence::getDirection(unsigned Level, bool IsSameSD) const { + return getDVEntry(Level, IsSameSD).Direction; } // Returns the distance (or NULL) associated with a particular common or // SameSD level. -const SCEV *FullDependence::getDistance(unsigned Level, bool isSameSD) const { - return getDVEntry(Level, isSameSD).Distance; +const SCEV *FullDependence::getDistance(unsigned Level, bool IsSameSD) const { + return getDVEntry(Level, IsSameSD).Distance; } // Returns true if a particular regular or SameSD level is scalar; that is, // if no subscript in the source or destination mention the induction variable // associated with the loop at this level. -bool FullDependence::isScalar(unsigned Level, bool isSameSD) const { - return getDVEntry(Level, isSameSD).Scalar; +bool FullDependence::isScalar(unsigned Level, bool IsSameSD) const { + return getDVEntry(Level, IsSameSD).Scalar; } // Returns true if peeling the first iteration from this regular or SameSD // loop level will break this dependence. -bool FullDependence::isPeelFirst(unsigned Level, bool isSameSD) const { - return getDVEntry(Level, isSameSD).PeelFirst; +bool FullDependence::isPeelFirst(unsigned Level, bool IsSameSD) const { + return getDVEntry(Level, IsSameSD).PeelFirst; } // Returns true if peeling the last iteration from this regular or SameSD // loop level will break this dependence. -bool FullDependence::isPeelLast(unsigned Level, bool isSameSD) const { - return getDVEntry(Level, isSameSD).PeelLast; +bool FullDependence::isPeelLast(unsigned Level, bool IsSameSD) const { + return getDVEntry(Level, IsSameSD).PeelLast; } // Returns true if splitting loop will break the dependence. -bool FullDependence::isSplitable(unsigned Level, bool isSameSD) const { - return getDVEntry(Level, isSameSD).Splitable; +bool FullDependence::isSplitable(unsigned Level, bool IsSameSD) const { + return getDVEntry(Level, IsSameSD).Splitable; } // inSameSDLoops - Returns true if this level is an SameSD level, i.e., @@ -691,7 +691,7 @@ void Dependence::dump(raw_ostream &OS) const { dumpImp(OS); unsigned SameSDLevels = getSameSDLevels(); if (SameSDLevels > 0) { - OS << "! / assuming " << SameSDLevels << " loop level(s) fused: "; + OS << " / assuming " << SameSDLevels << " loop level(s) fused: "; dumpImp(OS, true); } } @@ -706,13 +706,13 @@ void Dependence::dump(raw_ostream &OS) const { // For debugging purposes. Dumps a dependence to OS with or without considering // the SameSD levels. -void Dependence::dumpImp(raw_ostream &OS, bool isSameSD) const { +void Dependence::dumpImp(raw_ostream &OS, bool IsSameSD) const { bool Splitable = false; unsigned Levels = getLevels(); unsigned SameSDLevels = getSameSDLevels(); bool OnSameSD = false; unsigned LevelNum = Levels; - if (isSameSD) + if (IsSameSD) LevelNum += SameSDLevels; OS << " ["; for (unsigned II = 1; II <= LevelNum; ++II) { diff --git a/llvm/lib/Analysis/IR2Vec.cpp b/llvm/lib/Analysis/IR2Vec.cpp index 1794a60..85b5372 100644 --- a/llvm/lib/Analysis/IR2Vec.cpp +++ b/llvm/lib/Analysis/IR2Vec.cpp @@ -153,11 +153,6 @@ void Embedding::print(raw_ostream &OS) const { // Embedder and its subclasses //===----------------------------------------------------------------------===// -Embedder::Embedder(const Function &F, const Vocabulary &Vocab) - : F(F), Vocab(Vocab), Dimension(Vocab.getDimension()), - OpcWeight(::OpcWeight), TypeWeight(::TypeWeight), ArgWeight(::ArgWeight), - FuncVector(Embedding(Dimension)) {} - std::unique_ptr<Embedder> Embedder::create(IR2VecKind Mode, const Function &F, const Vocabulary &Vocab) { switch (Mode) { @@ -169,110 +164,85 @@ std::unique_ptr<Embedder> Embedder::create(IR2VecKind Mode, const Function &F, return nullptr; } -const InstEmbeddingsMap &Embedder::getInstVecMap() const { - if (InstVecMap.empty()) - computeEmbeddings(); - return InstVecMap; -} - -const BBEmbeddingsMap &Embedder::getBBVecMap() const { - if (BBVecMap.empty()) - computeEmbeddings(); - return BBVecMap; -} - -const Embedding &Embedder::getBBVector(const BasicBlock &BB) const { - auto It = BBVecMap.find(&BB); - if (It != BBVecMap.end()) - return It->second; - computeEmbeddings(BB); - return BBVecMap[&BB]; -} +Embedding Embedder::computeEmbeddings() const { + Embedding FuncVector(Dimension, 0.0); -const Embedding &Embedder::getFunctionVector() const { - // Currently, we always (re)compute the embeddings for the function. - // This is cheaper than caching the vector. - computeEmbeddings(); - return FuncVector; -} - -void Embedder::computeEmbeddings() const { if (F.isDeclaration()) - return; - - FuncVector = Embedding(Dimension, 0.0); + return FuncVector; // Consider only the basic blocks that are reachable from entry - for (const BasicBlock *BB : depth_first(&F)) { - computeEmbeddings(*BB); - FuncVector += BBVecMap[BB]; - } + for (const BasicBlock *BB : depth_first(&F)) + FuncVector += computeEmbeddings(*BB); + return FuncVector; } -void SymbolicEmbedder::computeEmbeddings(const BasicBlock &BB) const { +Embedding Embedder::computeEmbeddings(const BasicBlock &BB) const { Embedding BBVector(Dimension, 0); // We consider only the non-debug and non-pseudo instructions - for (const auto &I : BB.instructionsWithoutDebug()) { - Embedding ArgEmb(Dimension, 0); - for (const auto &Op : I.operands()) - ArgEmb += Vocab[*Op]; - auto InstVector = - Vocab[I.getOpcode()] + Vocab[I.getType()->getTypeID()] + ArgEmb; - if (const auto *IC = dyn_cast<CmpInst>(&I)) - InstVector += Vocab[IC->getPredicate()]; - InstVecMap[&I] = InstVector; - BBVector += InstVector; - } - BBVecMap[&BB] = BBVector; -} - -void FlowAwareEmbedder::computeEmbeddings(const BasicBlock &BB) const { - Embedding BBVector(Dimension, 0); + for (const auto &I : BB.instructionsWithoutDebug()) + BBVector += computeEmbeddings(I); + return BBVector; +} + +Embedding SymbolicEmbedder::computeEmbeddings(const Instruction &I) const { + // Currently, we always (re)compute the embeddings for symbolic embedder. + // This is cheaper than caching the vectors. + Embedding ArgEmb(Dimension, 0); + for (const auto &Op : I.operands()) + ArgEmb += Vocab[*Op]; + auto InstVector = + Vocab[I.getOpcode()] + Vocab[I.getType()->getTypeID()] + ArgEmb; + if (const auto *IC = dyn_cast<CmpInst>(&I)) + InstVector += Vocab[IC->getPredicate()]; + return InstVector; +} + +Embedding FlowAwareEmbedder::computeEmbeddings(const Instruction &I) const { + // If we have already computed the embedding for this instruction, return it + auto It = InstVecMap.find(&I); + if (It != InstVecMap.end()) + return It->second; - // We consider only the non-debug and non-pseudo instructions - for (const auto &I : BB.instructionsWithoutDebug()) { - // TODO: Handle call instructions differently. - // For now, we treat them like other instructions - Embedding ArgEmb(Dimension, 0); - for (const auto &Op : I.operands()) { - // If the operand is defined elsewhere, we use its embedding - if (const auto *DefInst = dyn_cast<Instruction>(Op)) { - auto DefIt = InstVecMap.find(DefInst); - // Fixme (#159171): Ideally we should never miss an instruction - // embedding here. - // But when we have cyclic dependencies (e.g., phi - // nodes), we might miss the embedding. In such cases, we fall back to - // using the vocabulary embedding. This can be fixed by iterating to a - // fixed-point, or by using a simple solver for the set of simultaneous - // equations. - // Another case when we might miss an instruction embedding is when - // the operand instruction is in a different basic block that has not - // been processed yet. This can be fixed by processing the basic blocks - // in a topological order. - if (DefIt != InstVecMap.end()) - ArgEmb += DefIt->second; - else - ArgEmb += Vocab[*Op]; - } - // If the operand is not defined by an instruction, we use the vocabulary - else { - LLVM_DEBUG(errs() << "Using embedding from vocabulary for operand: " - << *Op << "=" << Vocab[*Op][0] << "\n"); + // TODO: Handle call instructions differently. + // For now, we treat them like other instructions + Embedding ArgEmb(Dimension, 0); + for (const auto &Op : I.operands()) { + // If the operand is defined elsewhere, we use its embedding + if (const auto *DefInst = dyn_cast<Instruction>(Op)) { + auto DefIt = InstVecMap.find(DefInst); + // Fixme (#159171): Ideally we should never miss an instruction + // embedding here. + // But when we have cyclic dependencies (e.g., phi + // nodes), we might miss the embedding. In such cases, we fall back to + // using the vocabulary embedding. This can be fixed by iterating to a + // fixed-point, or by using a simple solver for the set of simultaneous + // equations. + // Another case when we might miss an instruction embedding is when + // the operand instruction is in a different basic block that has not + // been processed yet. This can be fixed by processing the basic blocks + // in a topological order. + if (DefIt != InstVecMap.end()) + ArgEmb += DefIt->second; + else ArgEmb += Vocab[*Op]; - } } - // Create the instruction vector by combining opcode, type, and arguments - // embeddings - auto InstVector = - Vocab[I.getOpcode()] + Vocab[I.getType()->getTypeID()] + ArgEmb; - // Add compare predicate embedding as an additional operand if applicable - if (const auto *IC = dyn_cast<CmpInst>(&I)) - InstVector += Vocab[IC->getPredicate()]; - InstVecMap[&I] = InstVector; - BBVector += InstVector; + // If the operand is not defined by an instruction, we use the + // vocabulary + else { + LLVM_DEBUG(errs() << "Using embedding from vocabulary for operand: " + << *Op << "=" << Vocab[*Op][0] << "\n"); + ArgEmb += Vocab[*Op]; + } } - BBVecMap[&BB] = BBVector; + // Create the instruction vector by combining opcode, type, and arguments + // embeddings + auto InstVector = + Vocab[I.getOpcode()] + Vocab[I.getType()->getTypeID()] + ArgEmb; + if (const auto *IC = dyn_cast<CmpInst>(&I)) + InstVector += Vocab[IC->getPredicate()]; + InstVecMap[&I] = InstVector; + return InstVector; } // ==----------------------------------------------------------------------===// @@ -695,25 +665,17 @@ PreservedAnalyses IR2VecPrinterPass::run(Module &M, Emb->getFunctionVector().print(OS); OS << "Basic block vectors:\n"; - const auto &BBMap = Emb->getBBVecMap(); for (const BasicBlock &BB : F) { - auto It = BBMap.find(&BB); - if (It != BBMap.end()) { - OS << "Basic block: " << BB.getName() << ":\n"; - It->second.print(OS); - } + OS << "Basic block: " << BB.getName() << ":\n"; + Emb->getBBVector(BB).print(OS); } OS << "Instruction vectors:\n"; - const auto &InstMap = Emb->getInstVecMap(); for (const BasicBlock &BB : F) { for (const Instruction &I : BB) { - auto It = InstMap.find(&I); - if (It != InstMap.end()) { - OS << "Instruction: "; - I.print(OS); - It->second.print(OS); - } + OS << "Instruction: "; + I.print(OS); + Emb->getInstVector(I).print(OS); } } } diff --git a/llvm/lib/Analysis/InstructionSimplify.cpp b/llvm/lib/Analysis/InstructionSimplify.cpp index d1977f0..4e38626 100644 --- a/llvm/lib/Analysis/InstructionSimplify.cpp +++ b/llvm/lib/Analysis/InstructionSimplify.cpp @@ -671,12 +671,12 @@ Value *llvm::simplifyAddInst(Value *Op0, Value *Op1, bool IsNSW, bool IsNUW, /// This is very similar to stripAndAccumulateConstantOffsets(), except it /// normalizes the offset bitwidth to the stripped pointer type, not the /// original pointer type. -static APInt stripAndComputeConstantOffsets(const DataLayout &DL, Value *&V, - bool AllowNonInbounds = false) { +static APInt stripAndComputeConstantOffsets(const DataLayout &DL, Value *&V) { assert(V->getType()->isPtrOrPtrVectorTy()); APInt Offset = APInt::getZero(DL.getIndexTypeSizeInBits(V->getType())); - V = V->stripAndAccumulateConstantOffsets(DL, Offset, AllowNonInbounds); + V = V->stripAndAccumulateConstantOffsets(DL, Offset, + /*AllowNonInbounds=*/true); // As that strip may trace through `addrspacecast`, need to sext or trunc // the offset calculated. return Offset.sextOrTrunc(DL.getIndexTypeSizeInBits(V->getType())); @@ -853,10 +853,12 @@ static Value *simplifySubInst(Value *Op0, Value *Op1, bool IsNSW, bool IsNUW, return W; // Variations on GEP(base, I, ...) - GEP(base, i, ...) -> GEP(null, I-i, ...). - if (match(Op0, m_PtrToInt(m_Value(X))) && match(Op1, m_PtrToInt(m_Value(Y)))) + if (match(Op0, m_PtrToIntOrAddr(m_Value(X))) && + match(Op1, m_PtrToIntOrAddr(m_Value(Y)))) { if (Constant *Result = computePointerDifference(Q.DL, X, Y)) return ConstantFoldIntegerCast(Result, Op0->getType(), /*IsSigned*/ true, Q.DL); + } // i1 sub -> xor. if (MaxRecurse && Op0->getType()->isIntOrIntVectorTy(1)) diff --git a/llvm/lib/Analysis/TargetLibraryInfo.cpp b/llvm/lib/Analysis/TargetLibraryInfo.cpp index 6e92766..813632c 100644 --- a/llvm/lib/Analysis/TargetLibraryInfo.cpp +++ b/llvm/lib/Analysis/TargetLibraryInfo.cpp @@ -740,11 +740,6 @@ static void initializeLibCalls(TargetLibraryInfoImpl &TLI, const Triple &T, TLI.setAvailable(LibFunc_fgets_unlocked); } - if (T.isAndroid() && T.isAndroidVersionLT(21)) { - TLI.setUnavailable(LibFunc_stpcpy); - TLI.setUnavailable(LibFunc_stpncpy); - } - if (T.isPS()) { // PS4/PS5 do have memalign. TLI.setAvailable(LibFunc_memalign); diff --git a/llvm/lib/AsmParser/LLLexer.cpp b/llvm/lib/AsmParser/LLLexer.cpp index f6937d3..50d1d47 100644 --- a/llvm/lib/AsmParser/LLLexer.cpp +++ b/llvm/lib/AsmParser/LLLexer.cpp @@ -982,6 +982,7 @@ lltok::Kind LLLexer::LexIdentifier() { DWKEYWORD(ATE, DwarfAttEncoding); DWKEYWORD(VIRTUALITY, DwarfVirtuality); DWKEYWORD(LANG, DwarfLang); + DWKEYWORD(LNAME, DwarfSourceLangName); DWKEYWORD(CC, DwarfCC); DWKEYWORD(OP, DwarfOp); DWKEYWORD(MACINFO, DwarfMacinfo); diff --git a/llvm/lib/AsmParser/LLParser.cpp b/llvm/lib/AsmParser/LLParser.cpp index 5589966..380b192 100644 --- a/llvm/lib/AsmParser/LLParser.cpp +++ b/llvm/lib/AsmParser/LLParser.cpp @@ -4740,6 +4740,10 @@ struct DwarfLangField : public MDUnsignedField { DwarfLangField() : MDUnsignedField(0, dwarf::DW_LANG_hi_user) {} }; +struct DwarfSourceLangNameField : public MDUnsignedField { + DwarfSourceLangNameField() : MDUnsignedField(0, UINT32_MAX) {} +}; + struct DwarfCCField : public MDUnsignedField { DwarfCCField() : MDUnsignedField(0, dwarf::DW_CC_hi_user) {} }; @@ -4998,6 +5002,25 @@ bool LLParser::parseMDField(LocTy Loc, StringRef Name, DwarfLangField &Result) { } template <> +bool LLParser::parseMDField(LocTy Loc, StringRef Name, + DwarfSourceLangNameField &Result) { + if (Lex.getKind() == lltok::APSInt) + return parseMDField(Loc, Name, static_cast<MDUnsignedField &>(Result)); + + if (Lex.getKind() != lltok::DwarfSourceLangName) + return tokError("expected DWARF source language name"); + + unsigned Lang = dwarf::getSourceLanguageName(Lex.getStrVal()); + if (!Lang) + return tokError("invalid DWARF source language name" + Twine(" '") + + Lex.getStrVal() + "'"); + assert(Lang <= Result.Max && "Expected valid DWARF source language name"); + Result.assign(Lang); + Lex.Lex(); + return false; +} + +template <> bool LLParser::parseMDField(LocTy Loc, StringRef Name, DwarfCCField &Result) { if (Lex.getKind() == lltok::APSInt) return parseMDField(Loc, Name, static_cast<MDUnsignedField &>(Result)); @@ -5836,9 +5859,12 @@ bool LLParser::parseDICompileUnit(MDNode *&Result, bool IsDistinct) { if (!IsDistinct) return tokError("missing 'distinct', required for !DICompileUnit"); + LocTy Loc = Lex.getLoc(); + #define VISIT_MD_FIELDS(OPTIONAL, REQUIRED) \ - REQUIRED(language, DwarfLangField, ); \ REQUIRED(file, MDField, (/* AllowNull */ false)); \ + OPTIONAL(language, DwarfLangField, ); \ + OPTIONAL(sourceLanguageName, DwarfSourceLangNameField, ); \ OPTIONAL(producer, MDStringField, ); \ OPTIONAL(isOptimized, MDBoolField, ); \ OPTIONAL(flags, MDStringField, ); \ @@ -5860,12 +5886,23 @@ bool LLParser::parseDICompileUnit(MDNode *&Result, bool IsDistinct) { PARSE_MD_FIELDS(); #undef VISIT_MD_FIELDS + if (!language.Seen && !sourceLanguageName.Seen) + return error(Loc, "missing one of 'language' or 'sourceLanguageName', " + "required for !DICompileUnit"); + + if (language.Seen && sourceLanguageName.Seen) + return error(Loc, "can only specify one of 'language' and " + "'sourceLanguageName' on !DICompileUnit"); + Result = DICompileUnit::getDistinct( - Context, DISourceLanguageName(language.Val), file.Val, producer.Val, - isOptimized.Val, flags.Val, runtimeVersion.Val, splitDebugFilename.Val, - emissionKind.Val, enums.Val, retainedTypes.Val, globals.Val, imports.Val, - macros.Val, dwoId.Val, splitDebugInlining.Val, debugInfoForProfiling.Val, - nameTableKind.Val, rangesBaseAddress.Val, sysroot.Val, sdk.Val); + Context, + language.Seen ? DISourceLanguageName(language.Val) + : DISourceLanguageName(sourceLanguageName.Val, 0), + file.Val, producer.Val, isOptimized.Val, flags.Val, runtimeVersion.Val, + splitDebugFilename.Val, emissionKind.Val, enums.Val, retainedTypes.Val, + globals.Val, imports.Val, macros.Val, dwoId.Val, splitDebugInlining.Val, + debugInfoForProfiling.Val, nameTableKind.Val, rangesBaseAddress.Val, + sysroot.Val, sdk.Val); return false; } diff --git a/llvm/lib/BinaryFormat/Dwarf.cpp b/llvm/lib/BinaryFormat/Dwarf.cpp index 969047a..55fa2df 100644 --- a/llvm/lib/BinaryFormat/Dwarf.cpp +++ b/llvm/lib/BinaryFormat/Dwarf.cpp @@ -893,6 +893,8 @@ StringRef llvm::dwarf::AttributeValueString(uint16_t Attr, unsigned Val) { return DefaultedMemberString(Val); case DW_AT_APPLE_enum_kind: return EnumKindString(Val); + case DW_AT_language_name: + return SourceLanguageNameString(static_cast<SourceLanguageName>(Val)); } return StringRef(); diff --git a/llvm/lib/Bitcode/Reader/MetadataLoader.cpp b/llvm/lib/Bitcode/Reader/MetadataLoader.cpp index a4d1b83..cdcf7a8 100644 --- a/llvm/lib/Bitcode/Reader/MetadataLoader.cpp +++ b/llvm/lib/Bitcode/Reader/MetadataLoader.cpp @@ -1867,12 +1867,18 @@ Error MetadataLoader::MetadataLoaderImpl::parseOneMetadata( // distinct. It's always distinct. IsDistinct = true; + const auto LangVersionMask = (uint64_t(1) << 63); + const bool HasVersionedLanguage = Record[1] & LangVersionMask; + auto *CU = DICompileUnit::getDistinct( - Context, DISourceLanguageName(Record[1]), getMDOrNull(Record[2]), - getMDString(Record[3]), Record[4], getMDString(Record[5]), Record[6], - getMDString(Record[7]), Record[8], getMDOrNull(Record[9]), - getMDOrNull(Record[10]), getMDOrNull(Record[12]), - getMDOrNull(Record[13]), + Context, + HasVersionedLanguage + ? DISourceLanguageName(Record[1] & ~LangVersionMask, 0) + : DISourceLanguageName(Record[1]), + getMDOrNull(Record[2]), getMDString(Record[3]), Record[4], + getMDString(Record[5]), Record[6], getMDString(Record[7]), Record[8], + getMDOrNull(Record[9]), getMDOrNull(Record[10]), + getMDOrNull(Record[12]), getMDOrNull(Record[13]), Record.size() <= 15 ? nullptr : getMDOrNull(Record[15]), Record.size() <= 14 ? 0 : Record[14], Record.size() <= 16 ? true : Record[16], diff --git a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp b/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp index 7ed140d..0ca55a26 100644 --- a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp +++ b/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp @@ -2108,7 +2108,13 @@ void ModuleBitcodeWriter::writeDICompileUnit(const DICompileUnit *N, assert(N->isDistinct() && "Expected distinct compile units"); Record.push_back(/* IsDistinct */ true); - Record.push_back(N->getSourceLanguage().getUnversionedName()); + auto Lang = N->getSourceLanguage(); + Record.push_back(Lang.getName()); + // Set bit so the MetadataLoader can distniguish between versioned and + // unversioned names. + if (Lang.hasVersionedName()) + Record.back() ^= (uint64_t(1) << 63); + Record.push_back(VE.getMetadataOrNullID(N->getFile())); Record.push_back(VE.getMetadataOrNullID(N->getRawProducer())); Record.push_back(N->isOptimized()); diff --git a/llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp b/llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp index bc0bb34..f0f0861 100644 --- a/llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp +++ b/llvm/lib/CodeGen/AsmPrinter/DwarfExpression.cpp @@ -587,10 +587,12 @@ bool DwarfExpression::addExpression( emitUnsigned(LeftShift); emitOp(dwarf::DW_OP_shl); } - emitOp(dwarf::DW_OP_constu); - emitUnsigned(RightShift); - emitOp(OpNum == dwarf::DW_OP_LLVM_extract_bits_sext ? dwarf::DW_OP_shra - : dwarf::DW_OP_shr); + if (RightShift) { + emitOp(dwarf::DW_OP_constu); + emitUnsigned(RightShift); + emitOp(OpNum == dwarf::DW_OP_LLVM_extract_bits_sext ? dwarf::DW_OP_shra + : dwarf::DW_OP_shr); + } // The value is now at the top of the stack, so set the location to // implicit so that we get a stack_value at the end. diff --git a/llvm/lib/CodeGen/CodeGenPrepare.cpp b/llvm/lib/CodeGen/CodeGenPrepare.cpp index eb73d01b..4320b1d 100644 --- a/llvm/lib/CodeGen/CodeGenPrepare.cpp +++ b/llvm/lib/CodeGen/CodeGenPrepare.cpp @@ -3194,7 +3194,7 @@ struct ExtAddrMode : public TargetLowering::AddrMode { case ScaledRegField: return ScaledReg; case BaseOffsField: - return ConstantInt::get(IntPtrTy, BaseOffs); + return ConstantInt::getSigned(IntPtrTy, BaseOffs); } } @@ -6100,7 +6100,7 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr, // Add in the Base Offset if present. if (AddrMode.BaseOffs) { - Value *V = ConstantInt::get(IntPtrTy, AddrMode.BaseOffs); + Value *V = ConstantInt::getSigned(IntPtrTy, AddrMode.BaseOffs); if (ResultIndex) { // We need to add this separately from the scale above to help with // SDAG consecutive load/store merging. @@ -6226,7 +6226,7 @@ bool CodeGenPrepare::optimizeMemoryInst(Instruction *MemoryInst, Value *Addr, // Add in the Base Offset if present. if (AddrMode.BaseOffs) { - Value *V = ConstantInt::get(IntPtrTy, AddrMode.BaseOffs); + Value *V = ConstantInt::getSigned(IntPtrTy, AddrMode.BaseOffs); if (Result) Result = Builder.CreateAdd(Result, V, "sunkaddr"); else diff --git a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp index 906d62a3..b425b95 100644 --- a/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp +++ b/llvm/lib/CodeGen/GlobalISel/CombinerHelper.cpp @@ -1728,6 +1728,7 @@ static APFloat constantFoldFpUnary(const MachineInstr &MI, Result.clearSign(); return Result; } + case TargetOpcode::G_FPEXT: case TargetOpcode::G_FPTRUNC: { bool Unused; LLT DstTy = MRI.getType(MI.getOperand(0).getReg()); diff --git a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp index 3f6813e..90c60d4 100644 --- a/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp +++ b/llvm/lib/CodeGen/GlobalISel/GISelValueTracking.cpp @@ -344,6 +344,22 @@ void GISelValueTracking::computeKnownBitsImpl(Register R, KnownBits &Known, Known = KnownBits::mul(Known, Known2); break; } + case TargetOpcode::G_UMULH: { + computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts, + Depth + 1); + computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts, + Depth + 1); + Known = KnownBits::mulhu(Known, Known2); + break; + } + case TargetOpcode::G_SMULH: { + computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts, + Depth + 1); + computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts, + Depth + 1); + Known = KnownBits::mulhs(Known, Known2); + break; + } case TargetOpcode::G_SELECT: { computeKnownBitsMin(MI.getOperand(2).getReg(), MI.getOperand(3).getReg(), Known, DemandedElts, Depth + 1); diff --git a/llvm/lib/CodeGen/RegisterCoalescer.cpp b/llvm/lib/CodeGen/RegisterCoalescer.cpp index ebfea8e..e17a214 100644 --- a/llvm/lib/CodeGen/RegisterCoalescer.cpp +++ b/llvm/lib/CodeGen/RegisterCoalescer.cpp @@ -2051,6 +2051,12 @@ bool RegisterCoalescer::joinCopy( } if (CP.getNewRC()) { + if (RegClassInfo.getNumAllocatableRegs(CP.getNewRC()) == 0) { + LLVM_DEBUG(dbgs() << "\tNo " << TRI->getRegClassName(CP.getNewRC()) + << "are available for allocation\n"); + return false; + } + auto SrcRC = MRI->getRegClass(CP.getSrcReg()); auto DstRC = MRI->getRegClass(CP.getDstReg()); unsigned SrcIdx = CP.getSrcIdx(); diff --git a/llvm/lib/DebugInfo/DWARF/DWARFDie.cpp b/llvm/lib/DebugInfo/DWARF/DWARFDie.cpp index edc69a3..212a0c0 100644 --- a/llvm/lib/DebugInfo/DWARF/DWARFDie.cpp +++ b/llvm/lib/DebugInfo/DWARF/DWARFDie.cpp @@ -149,7 +149,8 @@ static void dumpAttribute(raw_ostream &OS, const DWARFDie &Die, if (!Name.empty()) WithColor(OS, Color) << Name; else if (Attr == DW_AT_decl_line || Attr == DW_AT_decl_column || - Attr == DW_AT_call_line || Attr == DW_AT_call_column) { + Attr == DW_AT_call_line || Attr == DW_AT_call_column || + Attr == DW_AT_language_version) { if (std::optional<uint64_t> Val = FormValue.getAsUnsignedConstant()) OS << *Val; else diff --git a/llvm/lib/IR/AsmWriter.cpp b/llvm/lib/IR/AsmWriter.cpp index ae086bcd..0bc877d 100644 --- a/llvm/lib/IR/AsmWriter.cpp +++ b/llvm/lib/IR/AsmWriter.cpp @@ -2370,10 +2370,16 @@ static void writeDICompileUnit(raw_ostream &Out, const DICompileUnit *N, Out << "!DICompileUnit("; MDFieldPrinter Printer(Out, WriterCtx); - Printer.printDwarfEnum("language", - N->getSourceLanguage().getUnversionedName(), - dwarf::LanguageString, - /* ShouldSkipZero */ false); + auto Lang = N->getSourceLanguage(); + if (Lang.hasVersionedName()) + Printer.printDwarfEnum( + "sourceLanguageName", + static_cast<llvm::dwarf::SourceLanguageName>(Lang.getName()), + dwarf::SourceLanguageNameString, + /* ShouldSkipZero */ false); + else + Printer.printDwarfEnum("language", Lang.getName(), dwarf::LanguageString, + /* ShouldSkipZero */ false); Printer.printMetadata("file", N->getRawFile(), /* ShouldSkipNull */ false); Printer.printString("producer", N->getProducer()); diff --git a/llvm/lib/IR/ConstantFPRange.cpp b/llvm/lib/IR/ConstantFPRange.cpp index 7509188..fba6942 100644 --- a/llvm/lib/IR/ConstantFPRange.cpp +++ b/llvm/lib/IR/ConstantFPRange.cpp @@ -391,3 +391,23 @@ ConstantFPRange ConstantFPRange::unionWith(const ConstantFPRange &CR) const { return ConstantFPRange(minnum(Lower, CR.Lower), maxnum(Upper, CR.Upper), MayBeQNaN | CR.MayBeQNaN, MayBeSNaN | CR.MayBeSNaN); } + +ConstantFPRange ConstantFPRange::abs() const { + if (isNaNOnly()) + return *this; + // Check if the range is all non-negative or all non-positive. + if (Lower.isNegative() == Upper.isNegative()) { + if (Lower.isNegative()) + return negate(); + return *this; + } + // The range contains both positive and negative values. + APFloat NewLower = APFloat::getZero(getSemantics()); + APFloat NewUpper = maxnum(-Lower, Upper); + return ConstantFPRange(std::move(NewLower), std::move(NewUpper), MayBeQNaN, + MayBeSNaN); +} + +ConstantFPRange ConstantFPRange::negate() const { + return ConstantFPRange(-Upper, -Lower, MayBeQNaN, MayBeSNaN); +} diff --git a/llvm/lib/Target/AArch64/AArch64Combine.td b/llvm/lib/Target/AArch64/AArch64Combine.td index 639ddcb..ecaeff7 100644 --- a/llvm/lib/Target/AArch64/AArch64Combine.td +++ b/llvm/lib/Target/AArch64/AArch64Combine.td @@ -350,7 +350,7 @@ def AArch64PostLegalizerLowering // Post-legalization combines which are primarily optimizations. def AArch64PostLegalizerCombiner : GICombiner<"AArch64PostLegalizerCombinerImpl", - [copy_prop, cast_of_cast_combines, + [copy_prop, cast_of_cast_combines, constant_fold_fp_ops, buildvector_of_truncate, integer_of_truncate, mutate_anyext_to_zext, combines_for_extload, combine_indexed_load_store, sext_trunc_sextload, diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp index 31b3d18..7294f3e 100644 --- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp +++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp @@ -16249,7 +16249,9 @@ SDValue AArch64TargetLowering::LowerDIV(SDValue Op, SelectionDAG &DAG) const { bool Negated; uint64_t SplatVal; - if (Signed && isPow2Splat(Op.getOperand(1), SplatVal, Negated)) { + // NOTE: SRAD cannot be used to represent sdiv-by-one. + if (Signed && isPow2Splat(Op.getOperand(1), SplatVal, Negated) && + SplatVal > 1) { SDValue Pg = getPredicateForScalableVector(DAG, DL, VT); SDValue Res = DAG.getNode(AArch64ISD::SRAD_MERGE_OP1, DL, VT, Pg, Op->getOperand(0), @@ -30034,7 +30036,9 @@ SDValue AArch64TargetLowering::LowerFixedLengthVectorIntDivideToSVE( bool Negated; uint64_t SplatVal; - if (Signed && isPow2Splat(Op.getOperand(1), SplatVal, Negated)) { + // NOTE: SRAD cannot be used to represent sdiv-by-one. + if (Signed && isPow2Splat(Op.getOperand(1), SplatVal, Negated) && + SplatVal > 1) { EVT ContainerVT = getContainerForFixedLengthVector(DAG, VT); SDValue Op1 = convertToScalableVector(DAG, ContainerVT, Op.getOperand(0)); SDValue Op2 = DAG.getTargetConstant(Log2_64(SplatVal), DL, MVT::i32); @@ -30606,6 +30610,43 @@ AArch64TargetLowering::LowerVECTOR_DEINTERLEAVE(SDValue Op, assert(OpVT.isScalableVector() && "Expected scalable vector in LowerVECTOR_DEINTERLEAVE."); + if (Op->getNumOperands() == 3) { + // aarch64_sve_ld3 only supports packed datatypes. + EVT PackedVT = getPackedSVEVectorVT(OpVT.getVectorElementCount()); + Align Alignment = DAG.getReducedAlign(PackedVT, /*UseABI=*/false); + SDValue StackPtr = + DAG.CreateStackTemporary(PackedVT.getStoreSize() * 3, Alignment); + + // Write out unmodified operands. + SmallVector<SDValue, 3> Chains; + for (unsigned I = 0; I < 3; ++I) { + SDValue Ptr = + DAG.getMemBasePlusOffset(StackPtr, PackedVT.getStoreSize() * I, DL); + SDValue V = getSVESafeBitCast(PackedVT, Op.getOperand(I), DAG); + Chains.push_back( + DAG.getStore(DAG.getEntryNode(), DL, V, Ptr, MachinePointerInfo())); + } + + Intrinsic::ID IntID = Intrinsic::aarch64_sve_ld3_sret; + EVT PredVT = PackedVT.changeVectorElementType(MVT::i1); + + SmallVector<SDValue, 7> Ops; + Ops.push_back(DAG.getNode(ISD::TokenFactor, DL, MVT::Other, Chains)); + Ops.push_back(DAG.getTargetConstant(IntID, DL, MVT::i64)); + Ops.push_back(DAG.getConstant(1, DL, PredVT)); + Ops.push_back(StackPtr); + + // Read back and deinterleave data. + SDVTList VTs = DAG.getVTList(PackedVT, PackedVT, PackedVT, MVT::Other); + SDValue LD3 = DAG.getNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops); + + SmallVector<SDValue, 3> Results; + Results.push_back(getSVESafeBitCast(OpVT, LD3.getValue(0), DAG)); + Results.push_back(getSVESafeBitCast(OpVT, LD3.getValue(1), DAG)); + Results.push_back(getSVESafeBitCast(OpVT, LD3.getValue(2), DAG)); + return DAG.getMergeValues(Results, DL); + } + // Are multi-register uzp instructions available? if (Subtarget->hasSME2() && Subtarget->isStreaming() && OpVT.getVectorElementType() != MVT::i1) { @@ -30647,6 +30688,42 @@ SDValue AArch64TargetLowering::LowerVECTOR_INTERLEAVE(SDValue Op, assert(OpVT.isScalableVector() && "Expected scalable vector in LowerVECTOR_INTERLEAVE."); + if (Op->getNumOperands() == 3) { + // aarch64_sve_st3 only supports packed datatypes. + EVT PackedVT = getPackedSVEVectorVT(OpVT.getVectorElementCount()); + SmallVector<SDValue, 3> InVecs; + for (SDValue V : Op->ops()) + InVecs.push_back(getSVESafeBitCast(PackedVT, V, DAG)); + + Align Alignment = DAG.getReducedAlign(PackedVT, /*UseABI=*/false); + SDValue StackPtr = + DAG.CreateStackTemporary(PackedVT.getStoreSize() * 3, Alignment); + + Intrinsic::ID IntID = Intrinsic::aarch64_sve_st3; + EVT PredVT = PackedVT.changeVectorElementType(MVT::i1); + + SmallVector<SDValue, 7> Ops; + Ops.push_back(DAG.getEntryNode()); + Ops.push_back(DAG.getTargetConstant(IntID, DL, MVT::i64)); + Ops.append(InVecs); + Ops.push_back(DAG.getConstant(1, DL, PredVT)); + Ops.push_back(StackPtr); + + // Interleave operands and store. + SDValue Chain = DAG.getNode(ISD::INTRINSIC_VOID, DL, MVT::Other, Ops); + + // Read back the interleaved data. + SmallVector<SDValue, 3> Results; + for (unsigned I = 0; I < 3; ++I) { + SDValue Ptr = + DAG.getMemBasePlusOffset(StackPtr, PackedVT.getStoreSize() * I, DL); + SDValue L = DAG.getLoad(PackedVT, DL, Chain, Ptr, MachinePointerInfo()); + Results.push_back(getSVESafeBitCast(OpVT, L, DAG)); + } + + return DAG.getMergeValues(Results, DL); + } + // Are multi-register zip instructions available? if (Subtarget->hasSME2() && Subtarget->isStreaming() && OpVT.getVectorElementType() != MVT::i1) { diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp index 24bef82..8e35ba7 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp @@ -15,6 +15,7 @@ #include "AMDGPU.h" #include "AMDGPUTargetMachine.h" #include "SIModeRegisterDefaults.h" +#include "llvm/ADT/SetVector.h" #include "llvm/Analysis/AssumptionCache.h" #include "llvm/Analysis/ConstantFolding.h" #include "llvm/Analysis/TargetLibraryInfo.h" @@ -27,6 +28,7 @@ #include "llvm/IR/InstVisitor.h" #include "llvm/IR/IntrinsicsAMDGPU.h" #include "llvm/IR/PatternMatch.h" +#include "llvm/IR/ValueHandle.h" #include "llvm/InitializePasses.h" #include "llvm/Pass.h" #include "llvm/Support/KnownBits.h" @@ -106,6 +108,7 @@ public: bool FlowChanged = false; mutable Function *SqrtF32 = nullptr; mutable Function *LdexpF32 = nullptr; + mutable SmallVector<WeakVH> DeadVals; DenseMap<const PHINode *, bool> BreakPhiNodesCache; @@ -242,6 +245,8 @@ public: Value *emitSqrtIEEE2ULP(IRBuilder<> &Builder, Value *Src, FastMathFlags FMF) const; + bool tryNarrowMathIfNoOverflow(Instruction *I); + public: bool visitFDiv(BinaryOperator &I); @@ -281,28 +286,21 @@ bool AMDGPUCodeGenPrepareImpl::run() { BreakPhiNodesCache.clear(); bool MadeChange = false; - Function::iterator NextBB; - for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; FI = NextBB) { - BasicBlock *BB = &*FI; - NextBB = std::next(FI); - - BasicBlock::iterator Next; - for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; - I = Next) { - Next = std::next(I); - - MadeChange |= visit(*I); - - if (Next != E) { // Control flow changed - BasicBlock *NextInstBB = Next->getParent(); - if (NextInstBB != BB) { - BB = NextInstBB; - E = BB->end(); - FE = F.end(); - } - } + // Need to use make_early_inc_range because integer division expansion is + // handled by Transform/Utils, and it can delete instructions such as the + // terminator of the BB. + for (BasicBlock &BB : reverse(F)) { + for (Instruction &I : make_early_inc_range(reverse(BB))) { + if (!isInstructionTriviallyDead(&I, TLI)) + MadeChange |= visit(I); } } + + while (!DeadVals.empty()) { + if (auto *I = dyn_cast_or_null<Instruction>(DeadVals.pop_back_val())) + RecursivelyDeleteTriviallyDeadInstructions(I, TLI); + } + return MadeChange; } @@ -422,7 +420,7 @@ bool AMDGPUCodeGenPrepareImpl::replaceMulWithMul24(BinaryOperator &I) const { Value *NewVal = insertValues(Builder, Ty, ResultVals); NewVal->takeName(&I); I.replaceAllUsesWith(NewVal); - I.eraseFromParent(); + DeadVals.push_back(&I); return true; } @@ -496,10 +494,10 @@ bool AMDGPUCodeGenPrepareImpl::foldBinOpIntoSelect(BinaryOperator &BO) const { FoldedT, FoldedF); NewSelect->takeName(&BO); BO.replaceAllUsesWith(NewSelect); - BO.eraseFromParent(); + DeadVals.push_back(&BO); if (CastOp) - CastOp->eraseFromParent(); - Sel->eraseFromParent(); + DeadVals.push_back(CastOp); + DeadVals.push_back(Sel); return true; } @@ -895,7 +893,7 @@ bool AMDGPUCodeGenPrepareImpl::visitFDiv(BinaryOperator &FDiv) { if (NewVal) { FDiv.replaceAllUsesWith(NewVal); NewVal->takeName(&FDiv); - RecursivelyDeleteTriviallyDeadInstructions(&FDiv, TLI); + DeadVals.push_back(&FDiv); } return true; @@ -1302,10 +1300,7 @@ it will create `s_and_b32 s0, s0, 0xff`. We accept this change since the non-byte load assumes the upper bits within the byte are all 0. */ -static bool tryNarrowMathIfNoOverflow(Instruction *I, - const SITargetLowering *TLI, - const TargetTransformInfo &TTI, - const DataLayout &DL) { +bool AMDGPUCodeGenPrepareImpl::tryNarrowMathIfNoOverflow(Instruction *I) { unsigned Opc = I->getOpcode(); Type *OldType = I->getType(); @@ -1330,6 +1325,7 @@ static bool tryNarrowMathIfNoOverflow(Instruction *I, NewType = I->getType()->getWithNewBitWidth(NewBit); // Old cost + const TargetTransformInfo &TTI = TM.getTargetTransformInfo(F); InstructionCost OldCost = TTI.getArithmeticInstrCost(Opc, OldType, TTI::TCK_RecipThroughput); // New cost of new op @@ -1360,7 +1356,7 @@ static bool tryNarrowMathIfNoOverflow(Instruction *I, Value *Zext = Builder.CreateZExt(Arith, OldType); I->replaceAllUsesWith(Zext); - I->eraseFromParent(); + DeadVals.push_back(I); return true; } @@ -1370,8 +1366,7 @@ bool AMDGPUCodeGenPrepareImpl::visitBinaryOperator(BinaryOperator &I) { if (UseMul24Intrin && replaceMulWithMul24(I)) return true; - if (tryNarrowMathIfNoOverflow(&I, ST.getTargetLowering(), - TM.getTargetTransformInfo(F), DL)) + if (tryNarrowMathIfNoOverflow(&I)) return true; bool Changed = false; @@ -1436,7 +1431,7 @@ bool AMDGPUCodeGenPrepareImpl::visitBinaryOperator(BinaryOperator &I) { if (NewDiv) { I.replaceAllUsesWith(NewDiv); - I.eraseFromParent(); + DeadVals.push_back(&I); Changed = true; } } @@ -1492,7 +1487,7 @@ bool AMDGPUCodeGenPrepareImpl::visitLoadInst(LoadInst &I) { Value *ValTrunc = Builder.CreateTrunc(WidenLoad, IntNTy); Value *ValOrig = Builder.CreateBitCast(ValTrunc, I.getType()); I.replaceAllUsesWith(ValOrig); - I.eraseFromParent(); + DeadVals.push_back(&I); return true; } @@ -1534,7 +1529,7 @@ bool AMDGPUCodeGenPrepareImpl::visitSelectInst(SelectInst &I) { Fract->takeName(&I); I.replaceAllUsesWith(Fract); - RecursivelyDeleteTriviallyDeadInstructions(&I, TLI); + DeadVals.push_back(&I); return true; } @@ -1822,7 +1817,7 @@ bool AMDGPUCodeGenPrepareImpl::visitPHINode(PHINode &I) { } I.replaceAllUsesWith(Vec); - I.eraseFromParent(); + DeadVals.push_back(&I); return true; } @@ -1903,7 +1898,7 @@ bool AMDGPUCodeGenPrepareImpl::visitAddrSpaceCastInst(AddrSpaceCastInst &I) { auto *Intrin = B.CreateIntrinsic( I.getType(), Intrinsic::amdgcn_addrspacecast_nonnull, {I.getOperand(0)}); I.replaceAllUsesWith(Intrin); - I.eraseFromParent(); + DeadVals.push_back(&I); return true; } @@ -2000,16 +1995,10 @@ bool AMDGPUCodeGenPrepareImpl::visitFMinLike(IntrinsicInst &I) { Value *Fract = applyFractPat(Builder, FractArg); Fract->takeName(&I); I.replaceAllUsesWith(Fract); - - RecursivelyDeleteTriviallyDeadInstructions(&I, TLI); + DeadVals.push_back(&I); return true; } -static bool isOneOrNegOne(const Value *Val) { - const APFloat *C; - return match(Val, m_APFloat(C)) && C->getExactLog2Abs() == 0; -} - // Expand llvm.sqrt.f32 calls with !fpmath metadata in a semi-fast way. bool AMDGPUCodeGenPrepareImpl::visitSqrt(IntrinsicInst &Sqrt) { Type *Ty = Sqrt.getType()->getScalarType(); @@ -2030,18 +2019,6 @@ bool AMDGPUCodeGenPrepareImpl::visitSqrt(IntrinsicInst &Sqrt) { if (ReqdAccuracy < 1.0f) return false; - // FIXME: This is an ugly hack for this pass using forward iteration instead - // of reverse. If it worked like a normal combiner, the rsq would form before - // we saw a sqrt call. - auto *FDiv = - dyn_cast_or_null<FPMathOperator>(Sqrt.getUniqueUndroppableUser()); - if (FDiv && FDiv->getOpcode() == Instruction::FDiv && - FDiv->getFPAccuracy() >= 1.0f && - canOptimizeWithRsq(FPOp, FDiv->getFastMathFlags(), SqrtFMF) && - // TODO: We should also handle the arcp case for the fdiv with non-1 value - isOneOrNegOne(FDiv->getOperand(0))) - return false; - Value *SrcVal = Sqrt.getOperand(0); bool CanTreatAsDAZ = canIgnoreDenormalInput(SrcVal, &Sqrt); @@ -2065,7 +2042,7 @@ bool AMDGPUCodeGenPrepareImpl::visitSqrt(IntrinsicInst &Sqrt) { Value *NewSqrt = insertValues(Builder, Sqrt.getType(), ResultVals); NewSqrt->takeName(&Sqrt); Sqrt.replaceAllUsesWith(NewSqrt); - Sqrt.eraseFromParent(); + DeadVals.push_back(&Sqrt); return true; } diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp index 73b2660..5407566 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.cpp @@ -468,6 +468,38 @@ void RegBankLegalizeHelper::lowerUnpackBitShift(MachineInstr &MI) { MI.eraseFromParent(); } +void RegBankLegalizeHelper::lowerUnpackMinMax(MachineInstr &MI) { + Register Lo, Hi; + switch (MI.getOpcode()) { + case AMDGPU::G_SMIN: + case AMDGPU::G_SMAX: { + // For signed operations, use sign extension + auto [Val0_Lo, Val0_Hi] = unpackSExt(MI.getOperand(1).getReg()); + auto [Val1_Lo, Val1_Hi] = unpackSExt(MI.getOperand(2).getReg()); + Lo = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Lo, Val1_Lo}) + .getReg(0); + Hi = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Hi, Val1_Hi}) + .getReg(0); + break; + } + case AMDGPU::G_UMIN: + case AMDGPU::G_UMAX: { + // For unsigned operations, use zero extension + auto [Val0_Lo, Val0_Hi] = unpackZExt(MI.getOperand(1).getReg()); + auto [Val1_Lo, Val1_Hi] = unpackZExt(MI.getOperand(2).getReg()); + Lo = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Lo, Val1_Lo}) + .getReg(0); + Hi = B.buildInstr(MI.getOpcode(), {SgprRB_S32}, {Val0_Hi, Val1_Hi}) + .getReg(0); + break; + } + default: + llvm_unreachable("Unpack min/max lowering not implemented"); + } + B.buildBuildVectorTrunc(MI.getOperand(0).getReg(), {Lo, Hi}); + MI.eraseFromParent(); +} + static bool isSignedBFE(MachineInstr &MI) { if (GIntrinsic *GI = dyn_cast<GIntrinsic>(&MI)) return (GI->is(Intrinsic::amdgcn_sbfe)); @@ -654,6 +686,8 @@ void RegBankLegalizeHelper::lower(MachineInstr &MI, } case UnpackBitShift: return lowerUnpackBitShift(MI); + case UnpackMinMax: + return lowerUnpackMinMax(MI); case Ext32To64: { const RegisterBank *RB = MRI.getRegBank(MI.getOperand(0).getReg()); MachineInstrBuilder Hi; diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h index 7affe5a..d937815 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h +++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeHelper.h @@ -123,6 +123,7 @@ private: void lowerSplitTo32(MachineInstr &MI); void lowerSplitTo32Select(MachineInstr &MI); void lowerSplitTo32SExtInReg(MachineInstr &MI); + void lowerUnpackMinMax(MachineInstr &MI); }; } // end namespace AMDGPU diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp index f413bbc..7392f4b 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp @@ -522,6 +522,22 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST, .Uni(S64, {{Sgpr64}, {Sgpr64, Sgpr32, Sgpr32}, S_BFE}) .Div(S64, {{Vgpr64}, {Vgpr64, Vgpr32, Vgpr32}, V_BFE}); + addRulesForGOpcs({G_SMIN, G_SMAX}, Standard) + .Uni(S16, {{Sgpr32Trunc}, {Sgpr32SExt, Sgpr32SExt}}) + .Div(S16, {{Vgpr16}, {Vgpr16, Vgpr16}}) + .Uni(S32, {{Sgpr32}, {Sgpr32, Sgpr32}}) + .Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}}) + .Uni(V2S16, {{SgprV2S16}, {SgprV2S16, SgprV2S16}, UnpackMinMax}) + .Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}}); + + addRulesForGOpcs({G_UMIN, G_UMAX}, Standard) + .Uni(S16, {{Sgpr32Trunc}, {Sgpr32ZExt, Sgpr32ZExt}}) + .Div(S16, {{Vgpr16}, {Vgpr16, Vgpr16}}) + .Uni(S32, {{Sgpr32}, {Sgpr32, Sgpr32}}) + .Div(S32, {{Vgpr32}, {Vgpr32, Vgpr32}}) + .Uni(V2S16, {{SgprV2S16}, {SgprV2S16, SgprV2S16}, UnpackMinMax}) + .Div(V2S16, {{VgprV2S16}, {VgprV2S16, VgprV2S16}}); + // Note: we only write S1 rules for G_IMPLICIT_DEF, G_CONSTANT, G_FCONSTANT // and G_FREEZE here, rest is trivially regbankselected earlier addRulesForGOpcs({G_IMPLICIT_DEF}).Any({{UniS1}, {{Sgpr32Trunc}, {}}}); diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h index d0c6910..93e0efd 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h +++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.h @@ -212,6 +212,7 @@ enum LoweringMethodID { VccExtToSel, UniExtToSel, UnpackBitShift, + UnpackMinMax, S_BFE, V_BFE, VgprToVccCopy, diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp index 557d87f..56807a4 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPURegisterBankInfo.cpp @@ -5053,16 +5053,18 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const { // // vdst, srcA, srcB, srcC const SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>(); + + bool UseAGPRForm = !Subtarget.hasGFX90AInsts() || + Info->selectAGPRFormMFMA(MinNumRegsRequired); + OpdsMapping[0] = - Info->getMinNumAGPRs() >= MinNumRegsRequired - ? getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI) - : getVGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI); + UseAGPRForm ? getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI) + : getVGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI); OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI); OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI); OpdsMapping[4] = - Info->getMinNumAGPRs() >= MinNumRegsRequired - ? getAGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI) - : getVGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI); + UseAGPRForm ? getAGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI) + : getVGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI); break; } case Intrinsic::amdgcn_mfma_scale_f32_16x16x128_f8f6f4: @@ -5115,11 +5117,21 @@ AMDGPURegisterBankInfo::getInstrMapping(const MachineInstr &MI) const { case Intrinsic::amdgcn_smfmac_f32_32x32x64_bf8_fp8: case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_bf8: case Intrinsic::amdgcn_smfmac_f32_32x32x64_fp8_fp8: { + Register DstReg = MI.getOperand(0).getReg(); + unsigned DstSize = MRI.getType(DstReg).getSizeInBits(); + unsigned MinNumRegsRequired = DstSize / 32; + const SIMachineFunctionInfo *Info = MF.getInfo<SIMachineFunctionInfo>(); + bool UseAGPRForm = Info->selectAGPRFormMFMA(MinNumRegsRequired); + // vdst, srcA, srcB, srcC, idx - OpdsMapping[0] = getAGPROpMapping(MI.getOperand(0).getReg(), MRI, *TRI); + OpdsMapping[0] = UseAGPRForm ? getAGPROpMapping(DstReg, MRI, *TRI) + : getVGPROpMapping(DstReg, MRI, *TRI); + OpdsMapping[2] = getVGPROpMapping(MI.getOperand(2).getReg(), MRI, *TRI); OpdsMapping[3] = getVGPROpMapping(MI.getOperand(3).getReg(), MRI, *TRI); - OpdsMapping[4] = getAGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI); + OpdsMapping[4] = + UseAGPRForm ? getAGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI) + : getVGPROpMapping(MI.getOperand(4).getReg(), MRI, *TRI); OpdsMapping[5] = getVGPROpMapping(MI.getOperand(5).getReg(), MRI, *TRI); break; } diff --git a/llvm/lib/Target/AMDGPU/GCNRegPressure.cpp b/llvm/lib/Target/AMDGPU/GCNRegPressure.cpp index ef63acc..71494be 100644 --- a/llvm/lib/Target/AMDGPU/GCNRegPressure.cpp +++ b/llvm/lib/Target/AMDGPU/GCNRegPressure.cpp @@ -905,7 +905,7 @@ bool GCNRegPressurePrinter::runOnMachineFunction(MachineFunction &MF) { OS << ":\n"; SlotIndex MBBStartSlot = LIS.getSlotIndexes()->getMBBStartIdx(&MBB); - SlotIndex MBBEndSlot = LIS.getSlotIndexes()->getMBBEndIdx(&MBB); + SlotIndex MBBLastSlot = LIS.getSlotIndexes()->getMBBLastIdx(&MBB); GCNRPTracker::LiveRegSet LiveIn, LiveOut; GCNRegPressure RPAtMBBEnd; @@ -931,7 +931,7 @@ bool GCNRegPressurePrinter::runOnMachineFunction(MachineFunction &MF) { } } else { GCNUpwardRPTracker RPT(LIS); - RPT.reset(MRI, MBBEndSlot); + RPT.reset(MRI, MBBLastSlot); LiveOut = RPT.getLiveRegs(); RPAtMBBEnd = RPT.getPressure(); @@ -966,14 +966,14 @@ bool GCNRegPressurePrinter::runOnMachineFunction(MachineFunction &MF) { OS << PFX " Live-out:" << llvm::print(LiveOut, MRI); if (UseDownwardTracker) - ReportLISMismatchIfAny(LiveOut, getLiveRegs(MBBEndSlot, LIS, MRI)); + ReportLISMismatchIfAny(LiveOut, getLiveRegs(MBBLastSlot, LIS, MRI)); GCNRPTracker::LiveRegSet LiveThrough; for (auto [Reg, Mask] : LiveIn) { LaneBitmask MaskIntersection = Mask & LiveOut.lookup(Reg); if (MaskIntersection.any()) { LaneBitmask LTMask = getRegLiveThroughMask( - MRI, LIS, Reg, MBBStartSlot, MBBEndSlot, MaskIntersection); + MRI, LIS, Reg, MBBStartSlot, MBBLastSlot, MaskIntersection); if (LTMask.any()) LiveThrough[Reg] = LTMask; } diff --git a/llvm/lib/Target/AMDGPU/GCNRegPressure.h b/llvm/lib/Target/AMDGPU/GCNRegPressure.h index a9c58bb..898d1ff 100644 --- a/llvm/lib/Target/AMDGPU/GCNRegPressure.h +++ b/llvm/lib/Target/AMDGPU/GCNRegPressure.h @@ -313,8 +313,8 @@ public: /// reset tracker to the end of the \p MBB. void reset(const MachineBasicBlock &MBB) { - reset(MBB.getParent()->getRegInfo(), - LIS.getSlotIndexes()->getMBBEndIdx(&MBB)); + SlotIndex MBBLastSlot = LIS.getSlotIndexes()->getMBBLastIdx(&MBB); + reset(MBB.getParent()->getRegInfo(), MBBLastSlot); } /// reset tracker to the point just after \p MI (in program order). diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp index 730be69..80e985d 100644 --- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp +++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp @@ -103,52 +103,52 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM, addRegisterClass(MVT::Untyped, V64RegClass); addRegisterClass(MVT::v3i32, &AMDGPU::SGPR_96RegClass); - addRegisterClass(MVT::v3f32, TRI->getVGPRClassForBitWidth(96)); + addRegisterClass(MVT::v3f32, &AMDGPU::VReg_96RegClass); addRegisterClass(MVT::v2i64, &AMDGPU::SGPR_128RegClass); addRegisterClass(MVT::v2f64, &AMDGPU::SGPR_128RegClass); addRegisterClass(MVT::v4i32, &AMDGPU::SGPR_128RegClass); - addRegisterClass(MVT::v4f32, TRI->getVGPRClassForBitWidth(128)); + addRegisterClass(MVT::v4f32, &AMDGPU::VReg_128RegClass); addRegisterClass(MVT::v5i32, &AMDGPU::SGPR_160RegClass); - addRegisterClass(MVT::v5f32, TRI->getVGPRClassForBitWidth(160)); + addRegisterClass(MVT::v5f32, &AMDGPU::VReg_160RegClass); addRegisterClass(MVT::v6i32, &AMDGPU::SGPR_192RegClass); - addRegisterClass(MVT::v6f32, TRI->getVGPRClassForBitWidth(192)); + addRegisterClass(MVT::v6f32, &AMDGPU::VReg_192RegClass); addRegisterClass(MVT::v3i64, &AMDGPU::SGPR_192RegClass); - addRegisterClass(MVT::v3f64, TRI->getVGPRClassForBitWidth(192)); + addRegisterClass(MVT::v3f64, &AMDGPU::VReg_192RegClass); addRegisterClass(MVT::v7i32, &AMDGPU::SGPR_224RegClass); - addRegisterClass(MVT::v7f32, TRI->getVGPRClassForBitWidth(224)); + addRegisterClass(MVT::v7f32, &AMDGPU::VReg_224RegClass); addRegisterClass(MVT::v8i32, &AMDGPU::SGPR_256RegClass); - addRegisterClass(MVT::v8f32, TRI->getVGPRClassForBitWidth(256)); + addRegisterClass(MVT::v8f32, &AMDGPU::VReg_256RegClass); addRegisterClass(MVT::v4i64, &AMDGPU::SGPR_256RegClass); - addRegisterClass(MVT::v4f64, TRI->getVGPRClassForBitWidth(256)); + addRegisterClass(MVT::v4f64, &AMDGPU::VReg_256RegClass); addRegisterClass(MVT::v9i32, &AMDGPU::SGPR_288RegClass); - addRegisterClass(MVT::v9f32, TRI->getVGPRClassForBitWidth(288)); + addRegisterClass(MVT::v9f32, &AMDGPU::VReg_288RegClass); addRegisterClass(MVT::v10i32, &AMDGPU::SGPR_320RegClass); - addRegisterClass(MVT::v10f32, TRI->getVGPRClassForBitWidth(320)); + addRegisterClass(MVT::v10f32, &AMDGPU::VReg_320RegClass); addRegisterClass(MVT::v11i32, &AMDGPU::SGPR_352RegClass); - addRegisterClass(MVT::v11f32, TRI->getVGPRClassForBitWidth(352)); + addRegisterClass(MVT::v11f32, &AMDGPU::VReg_352RegClass); addRegisterClass(MVT::v12i32, &AMDGPU::SGPR_384RegClass); - addRegisterClass(MVT::v12f32, TRI->getVGPRClassForBitWidth(384)); + addRegisterClass(MVT::v12f32, &AMDGPU::VReg_384RegClass); addRegisterClass(MVT::v16i32, &AMDGPU::SGPR_512RegClass); - addRegisterClass(MVT::v16f32, TRI->getVGPRClassForBitWidth(512)); + addRegisterClass(MVT::v16f32, &AMDGPU::VReg_512RegClass); addRegisterClass(MVT::v8i64, &AMDGPU::SGPR_512RegClass); - addRegisterClass(MVT::v8f64, TRI->getVGPRClassForBitWidth(512)); + addRegisterClass(MVT::v8f64, &AMDGPU::VReg_512RegClass); addRegisterClass(MVT::v16i64, &AMDGPU::SGPR_1024RegClass); - addRegisterClass(MVT::v16f64, TRI->getVGPRClassForBitWidth(1024)); + addRegisterClass(MVT::v16f64, &AMDGPU::VReg_1024RegClass); if (Subtarget->has16BitInsts()) { if (Subtarget->useRealTrue16Insts()) { @@ -180,7 +180,7 @@ SITargetLowering::SITargetLowering(const TargetMachine &TM, } addRegisterClass(MVT::v32i32, &AMDGPU::VReg_1024RegClass); - addRegisterClass(MVT::v32f32, TRI->getVGPRClassForBitWidth(1024)); + addRegisterClass(MVT::v32f32, &AMDGPU::VReg_1024RegClass); computeRegisterProperties(Subtarget->getRegisterInfo()); diff --git a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h index b7dbb59..2c1a13c 100644 --- a/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h +++ b/llvm/lib/Target/AMDGPU/SIMachineFunctionInfo.h @@ -1202,6 +1202,12 @@ public: unsigned getMinNumAGPRs() const { return MinNumAGPRs; } + /// Return true if an MFMA that requires at least \p NumRegs should select to + /// the AGPR form, instead of the VGPR form. + bool selectAGPRFormMFMA(unsigned NumRegs) const { + return !MFMAVGPRForm && getMinNumAGPRs() >= NumRegs; + } + // \returns true if a function has a use of AGPRs via inline asm or // has a call which may use it. bool mayUseAGPRs(const Function &F) const; diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td index 7cfd059..6500fce 100644 --- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td +++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td @@ -964,14 +964,12 @@ class MAIFrag<SDPatternOperator Op, bit HasAbid = true, bit Scaled = false> : Pa class CanUseAGPR_MAI<ValueType vt> { code PredicateCode = [{ return !Subtarget->hasGFX90AInsts() || - (!SIMachineFunctionInfo::MFMAVGPRForm && - MF->getInfo<SIMachineFunctionInfo>()->getMinNumAGPRs() >= - }] # !srl(vt.Size, 5) # ");"; + MF->getInfo<SIMachineFunctionInfo>()->selectAGPRFormMFMA( + }] # !srl(vt.Size, 5) # ");"; code GISelPredicateCode = [{ return !Subtarget->hasGFX90AInsts() || - (!SIMachineFunctionInfo::MFMAVGPRForm && - MF.getInfo<SIMachineFunctionInfo>()->getMinNumAGPRs() >= + MF.getInfo<SIMachineFunctionInfo>()->selectAGPRFormMFMA( }] # !srl(vt.Size, 5) # ");"; } diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp index 2a40fb9..83c7def 100644 --- a/llvm/lib/Target/ARM/ARMISelLowering.cpp +++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp @@ -42,7 +42,6 @@ #include "llvm/CodeGen/CallingConvLower.h" #include "llvm/CodeGen/ComplexDeinterleavingPass.h" #include "llvm/CodeGen/ISDOpcodes.h" -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/CodeGen/MachineBasicBlock.h" #include "llvm/CodeGen/MachineConstantPool.h" #include "llvm/CodeGen/MachineFrameInfo.h" diff --git a/llvm/lib/Target/Hexagon/HexagonPatterns.td b/llvm/lib/Target/Hexagon/HexagonPatterns.td index a0acfcf..85ce944 100644 --- a/llvm/lib/Target/Hexagon/HexagonPatterns.td +++ b/llvm/lib/Target/Hexagon/HexagonPatterns.td @@ -699,35 +699,20 @@ def: OpR_RR_pat<C2_cmpgtp, setgt, i1, I64>; def: OpR_RR_pat<C2_cmpgtup, setugt, i1, I64>; def: OpR_RR_pat<C2_cmpgtp, RevCmp<setlt>, i1, I64>; def: OpR_RR_pat<C2_cmpgtup, RevCmp<setult>, i1, I64>; -def: OpR_RR_pat<A2_vcmpbeq, seteq, i1, V8I8>; def: OpR_RR_pat<A2_vcmpbeq, seteq, v8i1, V8I8>; -def: OpR_RR_pat<A4_vcmpbgt, RevCmp<setlt>, i1, V8I8>; def: OpR_RR_pat<A4_vcmpbgt, RevCmp<setlt>, v8i1, V8I8>; -def: OpR_RR_pat<A4_vcmpbgt, setgt, i1, V8I8>; def: OpR_RR_pat<A4_vcmpbgt, setgt, v8i1, V8I8>; -def: OpR_RR_pat<A2_vcmpbgtu, RevCmp<setult>, i1, V8I8>; def: OpR_RR_pat<A2_vcmpbgtu, RevCmp<setult>, v8i1, V8I8>; -def: OpR_RR_pat<A2_vcmpbgtu, setugt, i1, V8I8>; def: OpR_RR_pat<A2_vcmpbgtu, setugt, v8i1, V8I8>; -def: OpR_RR_pat<A2_vcmpheq, seteq, i1, V4I16>; def: OpR_RR_pat<A2_vcmpheq, seteq, v4i1, V4I16>; -def: OpR_RR_pat<A2_vcmphgt, RevCmp<setlt>, i1, V4I16>; def: OpR_RR_pat<A2_vcmphgt, RevCmp<setlt>, v4i1, V4I16>; -def: OpR_RR_pat<A2_vcmphgt, setgt, i1, V4I16>; def: OpR_RR_pat<A2_vcmphgt, setgt, v4i1, V4I16>; -def: OpR_RR_pat<A2_vcmphgtu, RevCmp<setult>, i1, V4I16>; def: OpR_RR_pat<A2_vcmphgtu, RevCmp<setult>, v4i1, V4I16>; -def: OpR_RR_pat<A2_vcmphgtu, setugt, i1, V4I16>; def: OpR_RR_pat<A2_vcmphgtu, setugt, v4i1, V4I16>; -def: OpR_RR_pat<A2_vcmpweq, seteq, i1, V2I32>; def: OpR_RR_pat<A2_vcmpweq, seteq, v2i1, V2I32>; -def: OpR_RR_pat<A2_vcmpwgt, RevCmp<setlt>, i1, V2I32>; def: OpR_RR_pat<A2_vcmpwgt, RevCmp<setlt>, v2i1, V2I32>; -def: OpR_RR_pat<A2_vcmpwgt, setgt, i1, V2I32>; def: OpR_RR_pat<A2_vcmpwgt, setgt, v2i1, V2I32>; -def: OpR_RR_pat<A2_vcmpwgtu, RevCmp<setult>, i1, V2I32>; def: OpR_RR_pat<A2_vcmpwgtu, RevCmp<setult>, v2i1, V2I32>; -def: OpR_RR_pat<A2_vcmpwgtu, setugt, i1, V2I32>; def: OpR_RR_pat<A2_vcmpwgtu, setugt, v2i1, V2I32>; def: OpR_RR_pat<F2_sfcmpeq, seteq, i1, F32>; @@ -1213,12 +1198,6 @@ def: OpR_RI_pat<S2_asl_i_r, Shl, i32, I32, u5_0ImmPred>; def: OpR_RI_pat<S2_asr_i_p, Sra, i64, I64, u6_0ImmPred>; def: OpR_RI_pat<S2_lsr_i_p, Srl, i64, I64, u6_0ImmPred>; def: OpR_RI_pat<S2_asl_i_p, Shl, i64, I64, u6_0ImmPred>; -def: OpR_RI_pat<S2_asr_i_vh, Sra, v4i16, V4I16, u4_0ImmPred>; -def: OpR_RI_pat<S2_lsr_i_vh, Srl, v4i16, V4I16, u4_0ImmPred>; -def: OpR_RI_pat<S2_asl_i_vh, Shl, v4i16, V4I16, u4_0ImmPred>; -def: OpR_RI_pat<S2_asr_i_vh, Sra, v2i32, V2I32, u5_0ImmPred>; -def: OpR_RI_pat<S2_lsr_i_vh, Srl, v2i32, V2I32, u5_0ImmPred>; -def: OpR_RI_pat<S2_asl_i_vh, Shl, v2i32, V2I32, u5_0ImmPred>; def: OpR_RR_pat<S2_asr_r_r, Sra, i32, I32, I32>; def: OpR_RR_pat<S2_lsr_r_r, Srl, i32, I32, I32>; diff --git a/llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp b/llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp index ba70c9e..97379d7 100644 --- a/llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp +++ b/llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp @@ -3677,7 +3677,7 @@ bool MipsAsmParser::expandBranchImm(MCInst &Inst, SMLoc IDLoc, MCStreamer &Out, Out, STI)) return true; - if (IsLikely) { + if (IsLikely && MemOffsetOp.isExpr()) { TOut.emitRRX(OpCode, DstRegOp.getReg(), ATReg, MCOperand::createExpr(MemOffsetOp.getExpr()), IDLoc, STI); TOut.emitRRI(Mips::SLL, Mips::ZERO, Mips::ZERO, 0, IDLoc, STI); diff --git a/llvm/lib/Target/Mips/MipsInstrInfo.td b/llvm/lib/Target/Mips/MipsInstrInfo.td index eff80e5..21d8ded 100644 --- a/llvm/lib/Target/Mips/MipsInstrInfo.td +++ b/llvm/lib/Target/Mips/MipsInstrInfo.td @@ -855,6 +855,16 @@ def calltarget : Operand<iPTR> { def imm64: Operand<i64>; +def ConstantImmAsmOperandClass : AsmOperandClass { + let Name = "ConstantImm"; + let PredicateMethod = "isConstantImm"; + let RenderMethod = "addImmOperands"; +} + +def ConstantImm64: Operand<i64> { + let ParserMatchClass = ConstantImmAsmOperandClass; +} + def simm19_lsl2 : Operand<i32> { let EncoderMethod = "getSimm19Lsl2Encoding"; let DecoderMethod = "DecodeSimm19Lsl2"; @@ -2947,10 +2957,10 @@ def : MipsInstAlias<"nor\t$rs, $imm", (NORImm GPR32Opnd:$rs, GPR32Opnd:$rs, let hasDelaySlot = 1, isCTI = 1 in { def BneImm : MipsAsmPseudoInst<(outs GPR32Opnd:$rt), - (ins imm64:$imm64, brtarget:$offset), + (ins ConstantImm64:$imm64, brtarget:$offset), "bne\t$rt, $imm64, $offset">; def BeqImm : MipsAsmPseudoInst<(outs GPR32Opnd:$rt), - (ins imm64:$imm64, brtarget:$offset), + (ins ConstantImm64:$imm64, brtarget:$offset), "beq\t$rt, $imm64, $offset">; class CondBranchPseudo<string instr_asm> : @@ -2978,7 +2988,7 @@ def BGTUL: CondBranchPseudo<"bgtul">, ISA_MIPS2_NOT_32R6_64R6; let isCTI = 1 in class CondBranchImmPseudo<string instr_asm> : - MipsAsmPseudoInst<(outs), (ins GPR32Opnd:$rs, imm64:$imm, brtarget:$offset), + MipsAsmPseudoInst<(outs), (ins GPR32Opnd:$rs, ConstantImm64:$imm, brtarget:$offset), !strconcat(instr_asm, "\t$rs, $imm, $offset")>; def BEQLImmMacro : CondBranchImmPseudo<"beql">, ISA_MIPS2_NOT_32R6_64R6; diff --git a/llvm/lib/Target/RISCV/RISCVFeatures.td b/llvm/lib/Target/RISCV/RISCVFeatures.td index 40c05e8..5ceb477 100644 --- a/llvm/lib/Target/RISCV/RISCVFeatures.td +++ b/llvm/lib/Target/RISCV/RISCVFeatures.td @@ -1520,6 +1520,8 @@ def HasVendorXqcics : Predicate<"Subtarget->hasVendorXqcics()">, AssemblerPredicate<(all_of FeatureVendorXqcics), "'Xqcics' (Qualcomm uC Conditional Select Extension)">; +def NoVendorXqcics + : Predicate<"!Subtarget->hasVendorXqcics()">; def FeatureVendorXqcicsr : RISCVExperimentalExtension<0, 4, "Qualcomm uC CSR Extension">; @@ -1823,6 +1825,11 @@ def TuneConditionalCompressedMoveFusion def HasConditionalMoveFusion : Predicate<"Subtarget->hasConditionalMoveFusion()">; def NoConditionalMoveFusion : Predicate<"!Subtarget->hasConditionalMoveFusion()">; +def TuneHasSingleElementVecFP64 + : SubtargetFeature<"single-element-vec-fp64", "HasSingleElementVectorFP64", "true", + "Certain vector FP64 operations produce a single result " + "element per cycle">; + def TuneMIPSP8700 : SubtargetFeature<"mips-p8700", "RISCVProcFamily", "MIPSP8700", "MIPS p8700 processor">; diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoXqci.td b/llvm/lib/Target/RISCV/RISCVInstrInfoXqci.td index f2724c41..5e1d07a 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfoXqci.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoXqci.td @@ -1571,35 +1571,42 @@ def : QCIMVCCIPat<SETUGE, QC_MVGEUI, uimm5nonzero>; } let Predicates = [HasVendorXqcicli, IsRV32] in { -def : QCILICCPat<SETEQ, QC_LIEQ>; -def : QCILICCPat<SETNE, QC_LINE>; def : QCILICCPat<SETLT, QC_LILT>; def : QCILICCPat<SETGE, QC_LIGE>; def : QCILICCPat<SETULT, QC_LILTU>; def : QCILICCPat<SETUGE, QC_LIGEU>; -def : QCILICCIPat<SETEQ, QC_LIEQI, simm5>; -def : QCILICCIPat<SETNE, QC_LINEI, simm5>; def : QCILICCIPat<SETLT, QC_LILTI, simm5>; def : QCILICCIPat<SETGE, QC_LIGEI, simm5>; def : QCILICCIPat<SETULT, QC_LILTUI, uimm5>; def : QCILICCIPat<SETUGE, QC_LIGEUI, uimm5>; -def : QCILICCPatInv<SETNE, QC_LIEQ>; -def : QCILICCPatInv<SETEQ, QC_LINE>; def : QCILICCPatInv<SETGE, QC_LILT>; def : QCILICCPatInv<SETLT, QC_LIGE>; def : QCILICCPatInv<SETUGE, QC_LILTU>; def : QCILICCPatInv<SETULT, QC_LIGEU>; -def : QCILICCIPatInv<SETNE, QC_LIEQI, simm5>; -def : QCILICCIPatInv<SETEQ, QC_LINEI, simm5>; def : QCILICCIPatInv<SETGE, QC_LILTI, simm5>; def : QCILICCIPatInv<SETLT, QC_LIGEI, simm5>; def : QCILICCIPatInv<SETUGE, QC_LILTUI, uimm5>; def : QCILICCIPatInv<SETULT, QC_LIGEUI, uimm5>; } // Predicates = [HasVendorXqcicli, IsRV32] +// Prioritize Xqcics over these patterns. +let Predicates = [HasVendorXqcicli, NoVendorXqcics, IsRV32] in { +def : QCILICCPat<SETEQ, QC_LIEQ>; +def : QCILICCPat<SETNE, QC_LINE>; + +def : QCILICCIPat<SETEQ, QC_LIEQI, simm5>; +def : QCILICCIPat<SETNE, QC_LINEI, simm5>; + +def : QCILICCPatInv<SETNE, QC_LIEQ>; +def : QCILICCPatInv<SETEQ, QC_LINE>; + +def : QCILICCIPatInv<SETNE, QC_LIEQI, simm5>; +def : QCILICCIPatInv<SETEQ, QC_LINEI, simm5>; +} // Predicates = [HasVendorXqcicli, NoVendorXqcics, IsRV32] + let Predicates = [HasVendorXqcics, IsRV32] in { // (SELECT X, Y, Z) is canonicalised to `(riscv_selectcc x, 0, NE, y, z)`. // These exist to prioritise over the `Select_GPR_Using_CC_GPR` pattern. diff --git a/llvm/lib/Target/RISCV/RISCVInstrPredicates.td b/llvm/lib/Target/RISCV/RISCVInstrPredicates.td index 6d86aff..3658817 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrPredicates.td +++ b/llvm/lib/Target/RISCV/RISCVInstrPredicates.td @@ -14,6 +14,10 @@ // otherwise. def VLDSX0Pred : MCSchedPredicate<CheckRegOperand<3, X0>>; +// This scheduling predicate is true when subtarget feature TuneHasSingleElementVecFP64 +// is enabled. +def SingleElementVecFP64SchedPred : FeatureSchedPredicate<TuneHasSingleElementVecFP64>; + // Returns true if this is the sext.w pattern, addiw rd, rs1, 0. def isSEXT_W : TIIPredicate<"isSEXT_W", diff --git a/llvm/lib/Target/RISCV/RISCVProcessors.td b/llvm/lib/Target/RISCV/RISCVProcessors.td index 17a7948..e86431f 100644 --- a/llvm/lib/Target/RISCV/RISCVProcessors.td +++ b/llvm/lib/Target/RISCV/RISCVProcessors.td @@ -338,7 +338,8 @@ def SIFIVE_X390 : RISCVProcessorModel<"sifive-x390", FeatureStdExtZvl1024b, FeatureVendorXSiFivecdiscarddlone, FeatureVendorXSiFivecflushdlone], - SiFiveIntelligenceTuneFeatures>; + !listconcat(SiFiveIntelligenceTuneFeatures, + [TuneHasSingleElementVecFP64])>; defvar SiFiveP400TuneFeatures = [TuneNoDefaultUnroll, TuneConditionalCompressedMoveFusion, diff --git a/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td b/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td index 3e07eff..f863392a 100644 --- a/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td +++ b/llvm/lib/Target/RISCV/RISCVSchedSiFive7.td @@ -317,7 +317,6 @@ multiclass SiFive7WriteResBase<int VLEN, ProcResourceKind VL, ProcResourceKind VS, ProcResourceKind VCQ, SiFive7FPLatencies fpLatencies, - bit isFP64Throttled = false, bit hasFastGather = false> { // Branching @@ -832,29 +831,56 @@ multiclass SiFive7WriteResBase<int VLEN, // 13. Vector Floating-Point Instructions foreach mx = SchedMxListF in { foreach sew = SchedSEWSet<mx, isF=1>.val in { - defvar Cycles = !if(!and(isFP64Throttled, !eq(sew, 64)), - SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c, - SiFive7GetCyclesDefault<mx>.c); - defvar Lat8 = !if(!and(isFP64Throttled, !eq(sew, 64)), Cycles, 8); - defvar VA = !if(!and(isFP64Throttled, !eq(sew, 64)), VA1, VA1OrVA2); defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxListF, isF=1>.c; - let Latency = Lat8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in { - defm : LMULSEWWriteResMXSEW<"WriteVFALUV", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFALUF", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFMulV", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFMulF", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFMulAddV", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFMulAddF", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFRecpV", [VCQ, VA1], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFCvtIToFV", [VCQ, VA1], mx, sew, IsWorstCase>; - } - defvar Lat4 = !if(!and(isFP64Throttled, !eq(sew, 64)), Cycles, 4); - let Latency = Lat4, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in { - defm : LMULSEWWriteResMXSEW<"WriteVFSgnjV", [VCQ, VA], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFSgnjF", [VCQ, VA], mx, sew, IsWorstCase>; - // min max require merge - defm : LMULSEWWriteResMXSEW<"WriteVFMinMaxV", [VCQ, VA1], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFMinMaxF", [VCQ, VA1], mx, sew, IsWorstCase>; + if !eq(sew, 64) then { + defvar SingleElementCycles = SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c; + foreach SchedWriteName = ["WriteVFALUV", "WriteVFALUF", "WriteVFMulV", "WriteVFMulF", + "WriteVFMulAddV", "WriteVFMulAddF"] in + defm : LMULSEWWriteResMXSEWVariant<SchedWriteName, SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], !add(SingleElementCycles, 7), [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1OrVA2], 8, [0, 1], [1, !add(1, SiFive7GetCyclesDefault<mx>.c)], + mx, sew, IsWorstCase>; + foreach SchedWriteName = ["WriteVFRecpV", "WriteVFCvtIToFV"] in + defm : LMULSEWWriteResMXSEWVariant<SchedWriteName, SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], !add(SingleElementCycles, 7), [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, SiFive7GetCyclesDefault<mx>.c)], + mx, sew, IsWorstCase>; + foreach SchedWriteName = ["WriteVFSgnjV", "WriteVFSgnjF"] in + defm : LMULSEWWriteResMXSEWVariant<SchedWriteName, SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], !add(SingleElementCycles, 3), [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1OrVA2], 4, [0, 1], [1, !add(1, SiFive7GetCyclesDefault<mx>.c)], + mx, sew, IsWorstCase>; + foreach SchedWriteName = ["WriteVFMinMaxV", "WriteVFMinMaxF"] in + defm : LMULSEWWriteResMXSEWVariant<SchedWriteName, SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], !add(SingleElementCycles, 3), [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1], 4, [0, 1], [1, !add(1, SiFive7GetCyclesDefault<mx>.c)], + mx, sew, IsWorstCase>; + } else { + let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, SiFive7GetCyclesDefault<mx>.c)] in { + defm : LMULSEWWriteResMXSEW<"WriteVFALUV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFALUF", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFMulV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFMulF", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFMulAddV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFMulAddF", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFRecpV", [VCQ, VA1], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFCvtIToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + } + let Latency = 4, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, SiFive7GetCyclesDefault<mx>.c)] in { + defm : LMULSEWWriteResMXSEW<"WriteVFSgnjV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFSgnjF", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; + // min max require merge + defm : LMULSEWWriteResMXSEW<"WriteVFMinMaxV", [VCQ, VA1], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFMinMaxF", [VCQ, VA1], mx, sew, IsWorstCase>; + } } } } @@ -892,19 +918,28 @@ multiclass SiFive7WriteResBase<int VLEN, // Widening foreach mx = SchedMxListW in { foreach sew = SchedSEWSet<mx, isF=0, isWidening=1>.val in { - defvar Cycles = !if(!and(isFP64Throttled, !eq(sew, 32)), - SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c, - SiFive7GetCyclesDefault<mx>.c); defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxListW>.c; - let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in - defm : LMULSEWWriteResMXSEW<"WriteVFWCvtIToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + defvar DefaultCycles = SiFive7GetCyclesDefault<mx>.c; + if !eq(sew, 32) then { + defvar SingleElementCycles = SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c; + defm : LMULSEWWriteResMXSEWVariant<"WriteVFWCvtIToFV", SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, DefaultCycles)], + mx, sew, IsWorstCase>; + } else { + let Latency = 8, + AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, DefaultCycles)] in + defm : LMULSEWWriteResMXSEW<"WriteVFWCvtIToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + } } } foreach mx = SchedMxListFW in { foreach sew = SchedSEWSet<mx, isF=1, isWidening=1>.val in { - defvar Cycles = SiFive7GetCyclesDefault<mx>.c; + defvar DefaultCycles = SiFive7GetCyclesDefault<mx>.c; defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxListFW, isF=1>.c; - let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in { + let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, DefaultCycles)] in { defm : LMULSEWWriteResMXSEW<"WriteVFWALUV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; defm : LMULSEWWriteResMXSEW<"WriteVFWALUF", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; defm : LMULSEWWriteResMXSEW<"WriteVFWMulV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; @@ -912,11 +947,19 @@ multiclass SiFive7WriteResBase<int VLEN, defm : LMULSEWWriteResMXSEW<"WriteVFWMulAddV", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; defm : LMULSEWWriteResMXSEW<"WriteVFWMulAddF", [VCQ, VA1OrVA2], mx, sew, IsWorstCase>; } - defvar CvtCycles = !if(!and(isFP64Throttled, !eq(sew, 32)), - SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c, - SiFive7GetCyclesDefault<mx>.c); - let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, CvtCycles)] in - defm "" : LMULSEWWriteResMXSEW<"WriteVFWCvtFToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + if !eq(sew, 32) then { + defvar SingleElementCycles = SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c; + defm : LMULSEWWriteResMXSEWVariant<"WriteVFWCvtFToFV", SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, DefaultCycles)], + mx, sew, IsWorstCase>; + } else { + let Latency = 8, + AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, DefaultCycles)] in + defm : LMULSEWWriteResMXSEW<"WriteVFWCvtFToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + } } defvar Cycles = SiFive7GetCyclesDefault<mx>.c; defvar IsWorstCase = SiFive7IsWorstCaseMX<mx, SchedMxListFW>.c; @@ -933,13 +976,23 @@ multiclass SiFive7WriteResBase<int VLEN, } foreach mx = SchedMxListFW in { foreach sew = SchedSEWSet<mx, isF=1, isWidening=1>.val in { - defvar Cycles = !if(!and(isFP64Throttled, !eq(sew, 32)), - SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c, - SiFive7GetCyclesNarrowing<mx>.c); defvar IsWorstCase = SiFive7IsWorstCaseMXSEW<mx, sew, SchedMxListFW, isF=1>.c; - let Latency = 8, AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, Cycles)] in { - defm : LMULSEWWriteResMXSEW<"WriteVFNCvtIToFV", [VCQ, VA1], mx, sew, IsWorstCase>; - defm : LMULSEWWriteResMXSEW<"WriteVFNCvtFToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + defvar DefaultCycles = SiFive7GetCyclesNarrowing<mx>.c; + if !eq(sew, 32) then { + defvar SingleElementCycles = SiFive7GetCyclesOnePerElement<mx, sew, VLEN>.c; + foreach SchedWriteName = ["WriteVFNCvtIToFV", "WriteVFNCvtFToFV"] in + defm : LMULSEWWriteResMXSEWVariant<SchedWriteName, SingleElementVecFP64SchedPred, + // Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, SingleElementCycles)], + // Not Predicated + [VCQ, VA1], 8, [0, 1], [1, !add(1, DefaultCycles)], + mx, sew, IsWorstCase>; + } else { + let Latency = 8, + AcquireAtCycles = [0, 1], ReleaseAtCycles = [1, !add(1, DefaultCycles)] in { + defm : LMULSEWWriteResMXSEW<"WriteVFNCvtIToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + defm : LMULSEWWriteResMXSEW<"WriteVFNCvtFToFV", [VCQ, VA1], mx, sew, IsWorstCase>; + } } } } @@ -1499,7 +1552,6 @@ multiclass SiFive7ReadAdvance { /// eventually be supplied by different SchedMachineModels. multiclass SiFive7SchedResources<int vlen, bit extraVALU, SiFive7FPLatencies fpLatencies, - bit isFP64Throttled, bit hasFastGather> { defm SiFive7 : SiFive7ProcResources<extraVALU>; @@ -1527,8 +1579,7 @@ multiclass SiFive7SchedResources<int vlen, bit extraVALU, : SiFive7WriteResBase<vlen, SiFive7PipeA, SiFive7PipeB, SiFive7PipeAB, SiFive7IDiv, SiFive7FDiv, SiFive7VA1, SiFive7VA1OrVA2, SiFive7VL, SiFive7VS, - SiFive7VCQ, fpLatencies, isFP64Throttled, - hasFastGather>; + SiFive7VCQ, fpLatencies, hasFastGather>; //===----------------------------------------------------------------------===// // Bypass and advance @@ -1560,7 +1611,6 @@ class SiFive7SchedMachineModel<int vlen> : SchedMachineModel { bit HasExtraVALU = false; SiFive7FPLatencies FPLatencies; - bit IsFP64Throttled = false; bit HasFastGather = false; string Name = !subst("Model", "", !subst("SiFive7", "", NAME)); @@ -1587,7 +1637,6 @@ def SiFive7VLEN512Model : SiFive7SchedMachineModel<512> { def SiFive7VLEN1024X300Model : SiFive7SchedMachineModel<1024> { let HasExtraVALU = true; let FPLatencies = SiFive7LowFPLatencies; - let IsFP64Throttled = true; let HasFastGather = true; } @@ -1596,7 +1645,6 @@ foreach model = [SiFive7VLEN512Model, SiFive7VLEN1024X300Model] in { let SchedModel = model in defm model.Name : SiFive7SchedResources<model.VLEN, model.HasExtraVALU, model.FPLatencies, - model.IsFP64Throttled, model.HasFastGather>; } diff --git a/llvm/lib/Target/RISCV/RISCVScheduleV.td b/llvm/lib/Target/RISCV/RISCVScheduleV.td index 01a4308..d11b446 100644 --- a/llvm/lib/Target/RISCV/RISCVScheduleV.td +++ b/llvm/lib/Target/RISCV/RISCVScheduleV.td @@ -128,6 +128,22 @@ multiclass LMULWriteResMXVariant<string name, SchedPredicateBase Pred, IsWorstCase>; } +multiclass LMULSEWWriteResMXSEWVariant<string name, SchedPredicateBase Pred, + list<ProcResourceKind> predResources, + int predLat, list<int> predAcquireCycles, + list<int> predReleaseCycles, + list<ProcResourceKind> noPredResources, + int noPredLat, list<int> noPredAcquireCycles, + list<int> noPredReleaseCycles, + string mx, int sew, bit IsWorstCase> { + defm "" : LMULWriteResVariantImpl<name, name # "_" # mx # "_E" # sew, Pred, predResources, + predLat, predAcquireCycles, + predReleaseCycles, noPredResources, + noPredLat, noPredAcquireCycles, + noPredReleaseCycles, + IsWorstCase>; +} + // Define multiclasses to define SchedWrite, SchedRead, WriteRes, and // ReadAdvance for each (name, LMUL) pair and for each LMUL in each of the // SchedMxList variants above. Each multiclass is responsible for defining diff --git a/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp b/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp index e8c849e..28a1690 100644 --- a/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVLegalizePointerCast.cpp @@ -46,7 +46,6 @@ #include "SPIRVSubtarget.h" #include "SPIRVTargetMachine.h" #include "SPIRVUtils.h" -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/Intrinsics.h" diff --git a/llvm/lib/Target/SPIRV/SPIRVMergeRegionExitTargets.cpp b/llvm/lib/Target/SPIRV/SPIRVMergeRegionExitTargets.cpp index 20f03b0..60d39c9 100644 --- a/llvm/lib/Target/SPIRV/SPIRVMergeRegionExitTargets.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVMergeRegionExitTargets.cpp @@ -19,7 +19,6 @@ #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/Analysis/LoopInfo.h" -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/IR/Dominators.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/Intrinsics.h" diff --git a/llvm/lib/Target/SPIRV/SPIRVStripConvergentIntrinsics.cpp b/llvm/lib/Target/SPIRV/SPIRVStripConvergentIntrinsics.cpp index 278ad7c..e621bcd44 100644 --- a/llvm/lib/Target/SPIRV/SPIRVStripConvergentIntrinsics.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVStripConvergentIntrinsics.cpp @@ -14,7 +14,6 @@ #include "SPIRV.h" #include "SPIRVSubtarget.h" #include "SPIRVUtils.h" -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/Intrinsics.h" #include "llvm/Transforms/Utils/Cloning.h" diff --git a/llvm/lib/Target/SPIRV/SPIRVStructurizer.cpp b/llvm/lib/Target/SPIRV/SPIRVStructurizer.cpp index 1811492..5b149f8 100644 --- a/llvm/lib/Target/SPIRV/SPIRVStructurizer.cpp +++ b/llvm/lib/Target/SPIRV/SPIRVStructurizer.cpp @@ -16,7 +16,6 @@ #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/Analysis/LoopInfo.h" -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/IR/CFG.h" #include "llvm/IR/Dominators.h" #include "llvm/IR/IRBuilder.h" diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 9580ade..1cfcb1f 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -28,7 +28,6 @@ #include "llvm/Analysis/BlockFrequencyInfo.h" #include "llvm/Analysis/ProfileSummaryInfo.h" #include "llvm/Analysis/VectorUtils.h" -#include "llvm/CodeGen/IntrinsicLowering.h" #include "llvm/CodeGen/LivePhysRegs.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" diff --git a/llvm/lib/Target/X86/X86ISelLoweringCall.cpp b/llvm/lib/Target/X86/X86ISelLoweringCall.cpp index 3bc46af..6dd43b2 100644 --- a/llvm/lib/Target/X86/X86ISelLoweringCall.cpp +++ b/llvm/lib/Target/X86/X86ISelLoweringCall.cpp @@ -547,7 +547,7 @@ unsigned X86TargetLowering::getAddressSpace() const { static bool hasStackGuardSlotTLS(const Triple &TargetTriple) { return TargetTriple.isOSGlibc() || TargetTriple.isOSFuchsia() || - (TargetTriple.isAndroid() && !TargetTriple.isAndroidVersionLT(17)); + TargetTriple.isAndroid(); } static Constant* SegmentOffset(IRBuilderBase &IRB, diff --git a/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp b/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp index cfdfd94..5066a99 100644 --- a/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp +++ b/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp @@ -1033,19 +1033,17 @@ private: }; } // namespace -namespace llvm { template <> -struct DenseMapInfo<typename CallsiteContextGraph< +struct llvm::DenseMapInfo<typename CallsiteContextGraph< ModuleCallsiteContextGraph, Function, Instruction *>::CallInfo> : public DenseMapInfo<std::pair<Instruction *, unsigned>> {}; template <> -struct DenseMapInfo<typename CallsiteContextGraph< +struct llvm::DenseMapInfo<typename CallsiteContextGraph< IndexCallsiteContextGraph, FunctionSummary, IndexCall>::CallInfo> : public DenseMapInfo<std::pair<IndexCall, unsigned>> {}; template <> -struct DenseMapInfo<IndexCall> +struct llvm::DenseMapInfo<IndexCall> : public DenseMapInfo<PointerUnion<CallsiteInfo *, AllocInfo *>> {}; -} // end namespace llvm namespace { diff --git a/llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp b/llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp index ac41fdd..2d5cb82 100644 --- a/llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp +++ b/llvm/lib/Transforms/IPO/WholeProgramDevirt.cpp @@ -372,9 +372,7 @@ struct VTableSlot { } // end anonymous namespace -namespace llvm { - -template <> struct DenseMapInfo<VTableSlot> { +template <> struct llvm::DenseMapInfo<VTableSlot> { static VTableSlot getEmptyKey() { return {DenseMapInfo<Metadata *>::getEmptyKey(), DenseMapInfo<uint64_t>::getEmptyKey()}; @@ -393,7 +391,7 @@ template <> struct DenseMapInfo<VTableSlot> { } }; -template <> struct DenseMapInfo<VTableSlotSummary> { +template <> struct llvm::DenseMapInfo<VTableSlotSummary> { static VTableSlotSummary getEmptyKey() { return {DenseMapInfo<StringRef>::getEmptyKey(), DenseMapInfo<uint64_t>::getEmptyKey()}; @@ -412,8 +410,6 @@ template <> struct DenseMapInfo<VTableSlotSummary> { } }; -} // end namespace llvm - // Returns true if the function must be unreachable based on ValueInfo. // // In particular, identifies a function as unreachable in the following diff --git a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp index 9b272c4..3ddf182 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineAndOrXor.cpp @@ -28,6 +28,10 @@ using namespace PatternMatch; #define DEBUG_TYPE "instcombine" +namespace llvm { +extern cl::opt<bool> ProfcheckDisableMetadataFixes; +} + /// This is the complement of getICmpCode, which turns an opcode and two /// operands into either a constant true or false, or a brand new ICmp /// instruction. The sign is passed in to determine which kind of predicate to @@ -1272,7 +1276,8 @@ Value *InstCombinerImpl::foldEqOfParts(Value *Cmp0, Value *Cmp1, bool IsAnd) { static Value *foldAndOrOfICmpsWithConstEq(ICmpInst *Cmp0, ICmpInst *Cmp1, bool IsAnd, bool IsLogical, InstCombiner::BuilderTy &Builder, - const SimplifyQuery &Q) { + const SimplifyQuery &Q, + Instruction &I) { // Match an equality compare with a non-poison constant as Cmp0. // Also, give up if the compare can be constant-folded to avoid looping. CmpPredicate Pred0; @@ -1306,9 +1311,12 @@ static Value *foldAndOrOfICmpsWithConstEq(ICmpInst *Cmp0, ICmpInst *Cmp1, return nullptr; SubstituteCmp = Builder.CreateICmp(Pred1, Y, C); } - if (IsLogical) - return IsAnd ? Builder.CreateLogicalAnd(Cmp0, SubstituteCmp) - : Builder.CreateLogicalOr(Cmp0, SubstituteCmp); + if (IsLogical) { + Instruction *MDFrom = + ProfcheckDisableMetadataFixes && isa<SelectInst>(I) ? nullptr : &I; + return IsAnd ? Builder.CreateLogicalAnd(Cmp0, SubstituteCmp, "", MDFrom) + : Builder.CreateLogicalOr(Cmp0, SubstituteCmp, "", MDFrom); + } return Builder.CreateBinOp(IsAnd ? Instruction::And : Instruction::Or, Cmp0, SubstituteCmp); } @@ -3396,13 +3404,13 @@ Value *InstCombinerImpl::foldAndOrOfICmps(ICmpInst *LHS, ICmpInst *RHS, /*IsLogical*/ false, Builder)) return V; - if (Value *V = - foldAndOrOfICmpsWithConstEq(LHS, RHS, IsAnd, IsLogical, Builder, Q)) + if (Value *V = foldAndOrOfICmpsWithConstEq(LHS, RHS, IsAnd, IsLogical, + Builder, Q, I)) return V; // We can convert this case to bitwise and, because both operands are used // on the LHS, and as such poison from both will propagate. - if (Value *V = foldAndOrOfICmpsWithConstEq(RHS, LHS, IsAnd, - /*IsLogical=*/false, Builder, Q)) { + if (Value *V = foldAndOrOfICmpsWithConstEq( + RHS, LHS, IsAnd, /*IsLogical=*/false, Builder, Q, I)) { // If RHS is still used, we should drop samesign flag. if (IsLogical && RHS->hasSameSign() && !RHS->use_empty()) { RHS->setSameSign(false); diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp index 56194fe..4c9b10a 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp @@ -2202,6 +2202,11 @@ Instruction *InstCombinerImpl::visitPtrToInt(PtrToIntInst &CI) { return commonCastTransforms(CI); } +Instruction *InstCombinerImpl::visitPtrToAddr(PtrToAddrInst &CI) { + // FIXME: Implement variants of ptrtoint folds. + return commonCastTransforms(CI); +} + /// This input value (which is known to have vector type) is being zero extended /// or truncated to the specified vector type. Since the zext/trunc is done /// using an integer type, we have a (bitcast(cast(bitcast))) pattern, diff --git a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h index e01c145..218aaf9 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineInternal.h +++ b/llvm/lib/Transforms/InstCombine/InstCombineInternal.h @@ -143,6 +143,7 @@ public: Instruction *visitUIToFP(CastInst &CI); Instruction *visitSIToFP(CastInst &CI); Instruction *visitPtrToInt(PtrToIntInst &CI); + Instruction *visitPtrToAddr(PtrToAddrInst &CI); Instruction *visitIntToPtr(IntToPtrInst &CI); Instruction *visitBitCast(BitCastInst &CI); Instruction *visitAddrSpaceCast(AddrSpaceCastInst &CI); diff --git a/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp b/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp index 5c747bb..9815644 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombinePHI.cpp @@ -1069,27 +1069,22 @@ struct LoweredPHIRecord { }; } // namespace -namespace llvm { - template<> - struct DenseMapInfo<LoweredPHIRecord> { - static inline LoweredPHIRecord getEmptyKey() { - return LoweredPHIRecord(nullptr, 0); - } - static inline LoweredPHIRecord getTombstoneKey() { - return LoweredPHIRecord(nullptr, 1); - } - static unsigned getHashValue(const LoweredPHIRecord &Val) { - return DenseMapInfo<PHINode*>::getHashValue(Val.PN) ^ (Val.Shift>>3) ^ - (Val.Width>>3); - } - static bool isEqual(const LoweredPHIRecord &LHS, - const LoweredPHIRecord &RHS) { - return LHS.PN == RHS.PN && LHS.Shift == RHS.Shift && - LHS.Width == RHS.Width; - } - }; -} // namespace llvm - +template <> struct llvm::DenseMapInfo<LoweredPHIRecord> { + static inline LoweredPHIRecord getEmptyKey() { + return LoweredPHIRecord(nullptr, 0); + } + static inline LoweredPHIRecord getTombstoneKey() { + return LoweredPHIRecord(nullptr, 1); + } + static unsigned getHashValue(const LoweredPHIRecord &Val) { + return DenseMapInfo<PHINode *>::getHashValue(Val.PN) ^ (Val.Shift >> 3) ^ + (Val.Width >> 3); + } + static bool isEqual(const LoweredPHIRecord &LHS, + const LoweredPHIRecord &RHS) { + return LHS.PN == RHS.PN && LHS.Shift == RHS.Shift && LHS.Width == RHS.Width; + } +}; /// This is an integer PHI and we know that it has an illegal type: see if it is /// only used by trunc or trunc(lshr) operations. If so, we split the PHI into diff --git a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp index 3704ad7..860f8f7 100644 --- a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp +++ b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp @@ -600,9 +600,7 @@ static ShadowMapping getShadowMapping(const Triple &TargetTriple, int LongSize, !IsRISCV64 && !IsLoongArch64 && !(Mapping.Offset & (Mapping.Offset - 1)) && Mapping.Offset != kDynamicShadowSentinel; - bool IsAndroidWithIfuncSupport = - IsAndroid && !TargetTriple.isAndroidVersionLT(21); - Mapping.InGlobal = ClWithIfunc && IsAndroidWithIfuncSupport && IsArmOrThumb; + Mapping.InGlobal = ClWithIfunc && IsAndroid && IsArmOrThumb; return Mapping; } diff --git a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp index 3f7003d..e5935f4 100644 --- a/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp +++ b/llvm/lib/Transforms/Scalar/DFAJumpThreading.cpp @@ -148,19 +148,16 @@ public: class DFAJumpThreading { public: - DFAJumpThreading(AssumptionCache *AC, DominatorTree *DT, LoopInfo *LI, + DFAJumpThreading(AssumptionCache *AC, DomTreeUpdater *DTU, LoopInfo *LI, TargetTransformInfo *TTI, OptimizationRemarkEmitter *ORE) - : AC(AC), DT(DT), LI(LI), TTI(TTI), ORE(ORE) {} + : AC(AC), DTU(DTU), LI(LI), TTI(TTI), ORE(ORE) {} bool run(Function &F); bool LoopInfoBroken; private: void - unfoldSelectInstrs(DominatorTree *DT, - const SmallVector<SelectInstToUnfold, 4> &SelectInsts) { - // TODO: Have everything use a single lazy DTU - DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Lazy); + unfoldSelectInstrs(const SmallVector<SelectInstToUnfold, 4> &SelectInsts) { SmallVector<SelectInstToUnfold, 4> Stack(SelectInsts); while (!Stack.empty()) { @@ -168,7 +165,7 @@ private: std::vector<SelectInstToUnfold> NewSIsToUnfold; std::vector<BasicBlock *> NewBBs; - unfold(&DTU, LI, SIToUnfold, &NewSIsToUnfold, &NewBBs); + unfold(DTU, LI, SIToUnfold, &NewSIsToUnfold, &NewBBs); // Put newly discovered select instructions into the work list. llvm::append_range(Stack, NewSIsToUnfold); @@ -181,7 +178,7 @@ private: std::vector<BasicBlock *> *NewBBs); AssumptionCache *AC; - DominatorTree *DT; + DomTreeUpdater *DTU; LoopInfo *LI; TargetTransformInfo *TTI; OptimizationRemarkEmitter *ORE; @@ -389,6 +386,22 @@ inline raw_ostream &operator<<(raw_ostream &OS, const PathType &Path) { return OS; } +/// Helper to get the successor corresponding to a particular case value for +/// a switch statement. +static BasicBlock *getNextCaseSuccessor(SwitchInst *Switch, + const APInt &NextState) { + BasicBlock *NextCase = nullptr; + for (auto Case : Switch->cases()) { + if (Case.getCaseValue()->getValue() == NextState) { + NextCase = Case.getCaseSuccessor(); + break; + } + } + if (!NextCase) + NextCase = Switch->getDefaultDest(); + return NextCase; +} + namespace { /// ThreadingPath is a path in the control flow of a loop that can be threaded /// by cloning necessary basic blocks and replacing conditional branches with @@ -401,6 +414,10 @@ struct ThreadingPath { ExitVal = V->getValue(); IsExitValSet = true; } + void setExitValue(const APInt &V) { + ExitVal = V; + IsExitValSet = true; + } bool isExitValueSet() const { return IsExitValSet; } /// Determinator is the basic block that determines the next state of the DFA. @@ -583,44 +600,8 @@ struct AllSwitchPaths { BasicBlock *getSwitchBlock() { return SwitchBlock; } void run() { - StateDefMap StateDef = getStateDefMap(); - if (StateDef.empty()) { - ORE->emit([&]() { - return OptimizationRemarkMissed(DEBUG_TYPE, "SwitchNotPredictable", - Switch) - << "Switch instruction is not predictable."; - }); - return; - } - - auto *SwitchPhi = cast<PHINode>(Switch->getOperand(0)); - auto *SwitchPhiDefBB = SwitchPhi->getParent(); - VisitedBlocks VB; - // Get paths from the determinator BBs to SwitchPhiDefBB - std::vector<ThreadingPath> PathsToPhiDef = - getPathsFromStateDefMap(StateDef, SwitchPhi, VB, MaxNumPaths); - if (SwitchPhiDefBB == SwitchBlock || PathsToPhiDef.empty()) { - TPaths = std::move(PathsToPhiDef); - return; - } - - assert(MaxNumPaths >= PathsToPhiDef.size() && !PathsToPhiDef.empty()); - auto PathsLimit = MaxNumPaths / PathsToPhiDef.size(); - // Find and append paths from SwitchPhiDefBB to SwitchBlock. - PathsType PathsToSwitchBB = - paths(SwitchPhiDefBB, SwitchBlock, VB, /* PathDepth = */ 1, PathsLimit); - if (PathsToSwitchBB.empty()) - return; - - std::vector<ThreadingPath> TempList; - for (const ThreadingPath &Path : PathsToPhiDef) { - for (const PathType &PathToSw : PathsToSwitchBB) { - ThreadingPath PathCopy(Path); - PathCopy.appendExcludingFirst(PathToSw); - TempList.push_back(PathCopy); - } - } - TPaths = std::move(TempList); + findTPaths(); + unifyTPaths(); } private: @@ -812,6 +793,69 @@ private: return Res; } + // Find all threadable paths. + void findTPaths() { + StateDefMap StateDef = getStateDefMap(); + if (StateDef.empty()) { + ORE->emit([&]() { + return OptimizationRemarkMissed(DEBUG_TYPE, "SwitchNotPredictable", + Switch) + << "Switch instruction is not predictable."; + }); + return; + } + + auto *SwitchPhi = cast<PHINode>(Switch->getOperand(0)); + auto *SwitchPhiDefBB = SwitchPhi->getParent(); + VisitedBlocks VB; + // Get paths from the determinator BBs to SwitchPhiDefBB + std::vector<ThreadingPath> PathsToPhiDef = + getPathsFromStateDefMap(StateDef, SwitchPhi, VB, MaxNumPaths); + if (SwitchPhiDefBB == SwitchBlock || PathsToPhiDef.empty()) { + TPaths = std::move(PathsToPhiDef); + return; + } + + assert(MaxNumPaths >= PathsToPhiDef.size() && !PathsToPhiDef.empty()); + auto PathsLimit = MaxNumPaths / PathsToPhiDef.size(); + // Find and append paths from SwitchPhiDefBB to SwitchBlock. + PathsType PathsToSwitchBB = + paths(SwitchPhiDefBB, SwitchBlock, VB, /* PathDepth = */ 1, PathsLimit); + if (PathsToSwitchBB.empty()) + return; + + std::vector<ThreadingPath> TempList; + for (const ThreadingPath &Path : PathsToPhiDef) { + for (const PathType &PathToSw : PathsToSwitchBB) { + ThreadingPath PathCopy(Path); + PathCopy.appendExcludingFirst(PathToSw); + TempList.push_back(PathCopy); + } + } + TPaths = std::move(TempList); + } + + // Two states are equivalent if they have the same switch destination. + // Unify the states in different threading path if the states are equivalent. + void unifyTPaths() { + llvm::SmallDenseMap<BasicBlock *, APInt> DestToState; + for (ThreadingPath &Path : TPaths) { + APInt NextState = Path.getExitValue(); + BasicBlock *Dest = getNextCaseSuccessor(Switch, NextState); + auto StateIt = DestToState.find(Dest); + if (StateIt == DestToState.end()) { + DestToState.insert({Dest, NextState}); + continue; + } + + if (NextState != StateIt->second) { + LLVM_DEBUG(dbgs() << "Next state in " << Path << " is equivalent to " + << StateIt->second << "\n"); + Path.setExitValue(StateIt->second); + } + } + } + unsigned NumVisited = 0; SwitchInst *Switch; BasicBlock *SwitchBlock; @@ -822,11 +866,11 @@ private: }; struct TransformDFA { - TransformDFA(AllSwitchPaths *SwitchPaths, DominatorTree *DT, + TransformDFA(AllSwitchPaths *SwitchPaths, DomTreeUpdater *DTU, AssumptionCache *AC, TargetTransformInfo *TTI, OptimizationRemarkEmitter *ORE, SmallPtrSet<const Value *, 32> EphValues) - : SwitchPaths(SwitchPaths), DT(DT), AC(AC), TTI(TTI), ORE(ORE), + : SwitchPaths(SwitchPaths), DTU(DTU), AC(AC), TTI(TTI), ORE(ORE), EphValues(EphValues) {} bool run() { @@ -1002,19 +1046,16 @@ private: SmallPtrSet<BasicBlock *, 16> BlocksToClean; BlocksToClean.insert_range(successors(SwitchBlock)); - { - DomTreeUpdater DTU(*DT, DomTreeUpdater::UpdateStrategy::Lazy); - for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths()) { - createExitPath(NewDefs, TPath, DuplicateMap, BlocksToClean, &DTU); - NumPaths++; - } - - // After all paths are cloned, now update the last successor of the cloned - // path so it skips over the switch statement - for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths()) - updateLastSuccessor(TPath, DuplicateMap, &DTU); + for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths()) { + createExitPath(NewDefs, TPath, DuplicateMap, BlocksToClean, DTU); + NumPaths++; } + // After all paths are cloned, now update the last successor of the cloned + // path so it skips over the switch statement + for (const ThreadingPath &TPath : SwitchPaths->getThreadingPaths()) + updateLastSuccessor(TPath, DuplicateMap, DTU); + // For each instruction that was cloned and used outside, update its uses updateSSA(NewDefs); @@ -1118,7 +1159,7 @@ private: } // SSAUpdater handles phi placement and renaming uses with the appropriate // value. - SSAUpdate.RewriteAllUses(DT); + SSAUpdate.RewriteAllUses(&DTU->getDomTree()); } /// Clones a basic block, and adds it to the CFG. @@ -1335,28 +1376,13 @@ private: return It != ClonedBBs.end() ? (*It).BB : nullptr; } - /// Helper to get the successor corresponding to a particular case value for - /// a switch statement. - BasicBlock *getNextCaseSuccessor(SwitchInst *Switch, const APInt &NextState) { - BasicBlock *NextCase = nullptr; - for (auto Case : Switch->cases()) { - if (Case.getCaseValue()->getValue() == NextState) { - NextCase = Case.getCaseSuccessor(); - break; - } - } - if (!NextCase) - NextCase = Switch->getDefaultDest(); - return NextCase; - } - /// Returns true if IncomingBB is a predecessor of BB. bool isPredecessor(BasicBlock *BB, BasicBlock *IncomingBB) { return llvm::is_contained(predecessors(BB), IncomingBB); } AllSwitchPaths *SwitchPaths; - DominatorTree *DT; + DomTreeUpdater *DTU; AssumptionCache *AC; TargetTransformInfo *TTI; OptimizationRemarkEmitter *ORE; @@ -1399,7 +1425,7 @@ bool DFAJumpThreading::run(Function &F) { << "candidate for jump threading\n"); LLVM_DEBUG(SI->dump()); - unfoldSelectInstrs(DT, Switch.getSelectInsts()); + unfoldSelectInstrs(Switch.getSelectInsts()); if (!Switch.getSelectInsts().empty()) MadeChanges = true; @@ -1421,7 +1447,7 @@ bool DFAJumpThreading::run(Function &F) { } #ifdef NDEBUG - LI->verify(*DT); + LI->verify(DTU->getDomTree()); #endif SmallPtrSet<const Value *, 32> EphValues; @@ -1429,13 +1455,15 @@ bool DFAJumpThreading::run(Function &F) { CodeMetrics::collectEphemeralValues(&F, AC, EphValues); for (AllSwitchPaths SwitchPaths : ThreadableLoops) { - TransformDFA Transform(&SwitchPaths, DT, AC, TTI, ORE, EphValues); + TransformDFA Transform(&SwitchPaths, DTU, AC, TTI, ORE, EphValues); if (Transform.run()) MadeChanges = LoopInfoBroken = true; } + DTU->flush(); + #ifdef EXPENSIVE_CHECKS - assert(DT->verify(DominatorTree::VerificationLevel::Full)); + assert(DTU->getDomTree().verify(DominatorTree::VerificationLevel::Full)); verifyFunction(F, &dbgs()); #endif @@ -1450,7 +1478,9 @@ PreservedAnalyses DFAJumpThreadingPass::run(Function &F, LoopInfo &LI = AM.getResult<LoopAnalysis>(F); TargetTransformInfo &TTI = AM.getResult<TargetIRAnalysis>(F); OptimizationRemarkEmitter ORE(&F); - DFAJumpThreading ThreadImpl(&AC, &DT, &LI, &TTI, &ORE); + + DomTreeUpdater DTU(DT, DomTreeUpdater::UpdateStrategy::Lazy); + DFAJumpThreading ThreadImpl(&AC, &DTU, &LI, &TTI, &ORE); if (!ThreadImpl.run(F)) return PreservedAnalyses::all(); diff --git a/llvm/lib/Transforms/Scalar/EarlyCSE.cpp b/llvm/lib/Transforms/Scalar/EarlyCSE.cpp index 0f8cc6c..2afa7b7 100644 --- a/llvm/lib/Transforms/Scalar/EarlyCSE.cpp +++ b/llvm/lib/Transforms/Scalar/EarlyCSE.cpp @@ -108,7 +108,7 @@ struct SimpleValue { // of instruction handled below (UnaryOperator, etc.). if (CallInst *CI = dyn_cast<CallInst>(Inst)) { if (Function *F = CI->getCalledFunction()) { - switch ((Intrinsic::ID)F->getIntrinsicID()) { + switch (F->getIntrinsicID()) { case Intrinsic::experimental_constrained_fadd: case Intrinsic::experimental_constrained_fsub: case Intrinsic::experimental_constrained_fmul: @@ -154,9 +154,7 @@ struct SimpleValue { } // end anonymous namespace -namespace llvm { - -template <> struct DenseMapInfo<SimpleValue> { +template <> struct llvm::DenseMapInfo<SimpleValue> { static inline SimpleValue getEmptyKey() { return DenseMapInfo<Instruction *>::getEmptyKey(); } @@ -169,8 +167,6 @@ template <> struct DenseMapInfo<SimpleValue> { static bool isEqual(SimpleValue LHS, SimpleValue RHS); }; -} // end namespace llvm - /// Match a 'select' including an optional 'not's of the condition. static bool matchSelectWithOptionalNotCond(Value *V, Value *&Cond, Value *&A, Value *&B, @@ -509,9 +505,7 @@ struct CallValue { } // end anonymous namespace -namespace llvm { - -template <> struct DenseMapInfo<CallValue> { +template <> struct llvm::DenseMapInfo<CallValue> { static inline CallValue getEmptyKey() { return DenseMapInfo<Instruction *>::getEmptyKey(); } @@ -524,8 +518,6 @@ template <> struct DenseMapInfo<CallValue> { static bool isEqual(CallValue LHS, CallValue RHS); }; -} // end namespace llvm - unsigned DenseMapInfo<CallValue>::getHashValue(CallValue Val) { Instruction *Inst = Val.Inst; @@ -580,9 +572,7 @@ struct GEPValue { } // namespace -namespace llvm { - -template <> struct DenseMapInfo<GEPValue> { +template <> struct llvm::DenseMapInfo<GEPValue> { static inline GEPValue getEmptyKey() { return DenseMapInfo<Instruction *>::getEmptyKey(); } @@ -595,8 +585,6 @@ template <> struct DenseMapInfo<GEPValue> { static bool isEqual(const GEPValue &LHS, const GEPValue &RHS); }; -} // end namespace llvm - unsigned DenseMapInfo<GEPValue>::getHashValue(const GEPValue &Val) { auto *GEP = cast<GetElementPtrInst>(Val.Inst); if (Val.ConstantOffset.has_value()) diff --git a/llvm/lib/Transforms/Scalar/GVN.cpp b/llvm/lib/Transforms/Scalar/GVN.cpp index 638952a..3a8ade8 100644 --- a/llvm/lib/Transforms/Scalar/GVN.cpp +++ b/llvm/lib/Transforms/Scalar/GVN.cpp @@ -170,9 +170,7 @@ struct llvm::GVNPass::Expression { } }; -namespace llvm { - -template <> struct DenseMapInfo<GVNPass::Expression> { +template <> struct llvm::DenseMapInfo<GVNPass::Expression> { static inline GVNPass::Expression getEmptyKey() { return ~0U; } static inline GVNPass::Expression getTombstoneKey() { return ~1U; } @@ -188,8 +186,6 @@ template <> struct DenseMapInfo<GVNPass::Expression> { } }; -} // end namespace llvm - /// Represents a particular available value that we know how to materialize. /// Materialization of an AvailableValue never fails. An AvailableValue is /// implicitly associated with a rematerialization point which is the @@ -2084,13 +2080,6 @@ bool GVNPass::processNonLocalLoad(LoadInst *Load) { return Changed; } -static bool hasUsersIn(Value *V, BasicBlock *BB) { - return any_of(V->users(), [BB](User *U) { - auto *I = dyn_cast<Instruction>(U); - return I && I->getParent() == BB; - }); -} - bool GVNPass::processAssumeIntrinsic(AssumeInst *IntrinsicI) { Value *V = IntrinsicI->getArgOperand(0); @@ -2149,85 +2138,7 @@ bool GVNPass::processAssumeIntrinsic(AssumeInst *IntrinsicI) { } Constant *True = ConstantInt::getTrue(V->getContext()); - bool Changed = false; - - for (BasicBlock *Successor : successors(IntrinsicI->getParent())) { - BasicBlockEdge Edge(IntrinsicI->getParent(), Successor); - - // This property is only true in dominated successors, propagateEquality - // will check dominance for us. - Changed |= propagateEquality(V, True, Edge, false); - } - - // We can replace assume value with true, which covers cases like this: - // call void @llvm.assume(i1 %cmp) - // br i1 %cmp, label %bb1, label %bb2 ; will change %cmp to true - ReplaceOperandsWithMap[V] = True; - - // Similarly, after assume(!NotV) we know that NotV == false. - Value *NotV; - if (match(V, m_Not(m_Value(NotV)))) - ReplaceOperandsWithMap[NotV] = ConstantInt::getFalse(V->getContext()); - - // If we find an equality fact, canonicalize all dominated uses in this block - // to one of the two values. We heuristically choice the "oldest" of the - // two where age is determined by value number. (Note that propagateEquality - // above handles the cross block case.) - // - // Key case to cover are: - // 1) - // %cmp = fcmp oeq float 3.000000e+00, %0 ; const on lhs could happen - // call void @llvm.assume(i1 %cmp) - // ret float %0 ; will change it to ret float 3.000000e+00 - // 2) - // %load = load float, float* %addr - // %cmp = fcmp oeq float %load, %0 - // call void @llvm.assume(i1 %cmp) - // ret float %load ; will change it to ret float %0 - if (auto *CmpI = dyn_cast<CmpInst>(V)) { - if (CmpI->isEquivalence()) { - Value *CmpLHS = CmpI->getOperand(0); - Value *CmpRHS = CmpI->getOperand(1); - // Heuristically pick the better replacement -- the choice of heuristic - // isn't terribly important here, but the fact we canonicalize on some - // replacement is for exposing other simplifications. - // TODO: pull this out as a helper function and reuse w/ existing - // (slightly different) logic. - if (isa<Constant>(CmpLHS) && !isa<Constant>(CmpRHS)) - std::swap(CmpLHS, CmpRHS); - if (!isa<Instruction>(CmpLHS) && isa<Instruction>(CmpRHS)) - std::swap(CmpLHS, CmpRHS); - if ((isa<Argument>(CmpLHS) && isa<Argument>(CmpRHS)) || - (isa<Instruction>(CmpLHS) && isa<Instruction>(CmpRHS))) { - // Move the 'oldest' value to the right-hand side, using the value - // number as a proxy for age. - uint32_t LVN = VN.lookupOrAdd(CmpLHS); - uint32_t RVN = VN.lookupOrAdd(CmpRHS); - if (LVN < RVN) - std::swap(CmpLHS, CmpRHS); - } - - // Handle degenerate case where we either haven't pruned a dead path or a - // removed a trivial assume yet. - if (isa<Constant>(CmpLHS) && isa<Constant>(CmpRHS)) - return Changed; - - LLVM_DEBUG(dbgs() << "Replacing dominated uses of " - << *CmpLHS << " with " - << *CmpRHS << " in block " - << IntrinsicI->getParent()->getName() << "\n"); - - // Setup the replacement map - this handles uses within the same block. - if (hasUsersIn(CmpLHS, IntrinsicI->getParent())) - ReplaceOperandsWithMap[CmpLHS] = CmpRHS; - - // NOTE: The non-block local cases are handled by the call to - // propagateEquality above; this block is just about handling the block - // local cases. TODO: There's a bunch of logic in propagateEqualiy which - // isn't duplicated for the block local case, can we share it somehow? - } - } - return Changed; + return propagateEquality(V, True, IntrinsicI); } static void patchAndReplaceAllUsesWith(Instruction *I, Value *Repl) { @@ -2526,39 +2437,28 @@ void GVNPass::assignBlockRPONumber(Function &F) { InvalidBlockRPONumbers = false; } -bool GVNPass::replaceOperandsForInBlockEquality(Instruction *Instr) const { - bool Changed = false; - for (unsigned OpNum = 0; OpNum < Instr->getNumOperands(); ++OpNum) { - Use &Operand = Instr->getOperandUse(OpNum); - auto It = ReplaceOperandsWithMap.find(Operand.get()); - if (It != ReplaceOperandsWithMap.end()) { - const DataLayout &DL = Instr->getDataLayout(); - if (!canReplacePointersInUseIfEqual(Operand, It->second, DL)) - continue; - - LLVM_DEBUG(dbgs() << "GVN replacing: " << *Operand << " with " - << *It->second << " in instruction " << *Instr << '\n'); - Instr->setOperand(OpNum, It->second); - Changed = true; - } - } - return Changed; -} - -/// The given values are known to be equal in every block +/// The given values are known to be equal in every use /// dominated by 'Root'. Exploit this, for example by replacing 'LHS' with /// 'RHS' everywhere in the scope. Returns whether a change was made. -/// If DominatesByEdge is false, then it means that we will propagate the RHS -/// value starting from the end of Root.Start. -bool GVNPass::propagateEquality(Value *LHS, Value *RHS, - const BasicBlockEdge &Root, - bool DominatesByEdge) { +/// The Root may either be a basic block edge (for conditions) or an +/// instruction (for assumes). +bool GVNPass::propagateEquality( + Value *LHS, Value *RHS, + const std::variant<BasicBlockEdge, Instruction *> &Root) { SmallVector<std::pair<Value*, Value*>, 4> Worklist; Worklist.push_back(std::make_pair(LHS, RHS)); bool Changed = false; - // For speed, compute a conservative fast approximation to - // DT->dominates(Root, Root.getEnd()); - const bool RootDominatesEnd = isOnlyReachableViaThisEdge(Root, DT); + SmallVector<const BasicBlock *> DominatedBlocks; + if (const BasicBlockEdge *Edge = std::get_if<BasicBlockEdge>(&Root)) { + // For speed, compute a conservative fast approximation to + // DT->dominates(Root, Root.getEnd()); + if (isOnlyReachableViaThisEdge(*Edge, DT)) + DominatedBlocks.push_back(Edge->getEnd()); + } else { + Instruction *I = std::get<Instruction *>(Root); + for (const auto *Node : DT->getNode(I->getParent())->children()) + DominatedBlocks.push_back(Node->getBlock()); + } while (!Worklist.empty()) { std::pair<Value*, Value*> Item = Worklist.pop_back_val(); @@ -2606,9 +2506,9 @@ bool GVNPass::propagateEquality(Value *LHS, Value *RHS, // using the leader table is about compiling faster, not optimizing better). // The leader table only tracks basic blocks, not edges. Only add to if we // have the simple case where the edge dominates the end. - if (RootDominatesEnd && !isa<Instruction>(RHS) && - canReplacePointersIfEqual(LHS, RHS, DL)) - LeaderTable.insert(LVN, RHS, Root.getEnd()); + if (!isa<Instruction>(RHS) && canReplacePointersIfEqual(LHS, RHS, DL)) + for (const BasicBlock *BB : DominatedBlocks) + LeaderTable.insert(LVN, RHS, BB); // Replace all occurrences of 'LHS' with 'RHS' everywhere in the scope. As // LHS always has at least one use that is not dominated by Root, this will @@ -2618,12 +2518,14 @@ bool GVNPass::propagateEquality(Value *LHS, Value *RHS, auto CanReplacePointersCallBack = [&DL](const Use &U, const Value *To) { return canReplacePointersInUseIfEqual(U, To, DL); }; - unsigned NumReplacements = - DominatesByEdge - ? replaceDominatedUsesWithIf(LHS, RHS, *DT, Root, - CanReplacePointersCallBack) - : replaceDominatedUsesWithIf(LHS, RHS, *DT, Root.getStart(), - CanReplacePointersCallBack); + unsigned NumReplacements; + if (const BasicBlockEdge *Edge = std::get_if<BasicBlockEdge>(&Root)) + NumReplacements = replaceDominatedUsesWithIf( + LHS, RHS, *DT, *Edge, CanReplacePointersCallBack); + else + NumReplacements = replaceDominatedUsesWithIf( + LHS, RHS, *DT, std::get<Instruction *>(Root), + CanReplacePointersCallBack); if (NumReplacements > 0) { Changed = true; @@ -2682,26 +2584,45 @@ bool GVNPass::propagateEquality(Value *LHS, Value *RHS, // If the number we were assigned was brand new then there is no point in // looking for an instruction realizing it: there cannot be one! if (Num < NextNum) { - Value *NotCmp = findLeader(Root.getEnd(), Num); - if (NotCmp && isa<Instruction>(NotCmp)) { - unsigned NumReplacements = - DominatesByEdge - ? replaceDominatedUsesWith(NotCmp, NotVal, *DT, Root) - : replaceDominatedUsesWith(NotCmp, NotVal, *DT, - Root.getStart()); - Changed |= NumReplacements > 0; - NumGVNEqProp += NumReplacements; - // Cached information for anything that uses NotCmp will be invalid. - if (MD) - MD->invalidateCachedPointerInfo(NotCmp); + for (const auto &Entry : LeaderTable.getLeaders(Num)) { + // Only look at leaders that either dominate the start of the edge, + // or are dominated by the end. This check is not necessary for + // correctness, it only discards cases for which the following + // use replacement will not work anyway. + if (const BasicBlockEdge *Edge = std::get_if<BasicBlockEdge>(&Root)) { + if (!DT->dominates(Entry.BB, Edge->getStart()) && + !DT->dominates(Edge->getEnd(), Entry.BB)) + continue; + } else { + auto *InstBB = std::get<Instruction *>(Root)->getParent(); + if (!DT->dominates(Entry.BB, InstBB) && + !DT->dominates(InstBB, Entry.BB)) + continue; + } + + Value *NotCmp = Entry.Val; + if (NotCmp && isa<Instruction>(NotCmp)) { + unsigned NumReplacements; + if (const BasicBlockEdge *Edge = std::get_if<BasicBlockEdge>(&Root)) + NumReplacements = + replaceDominatedUsesWith(NotCmp, NotVal, *DT, *Edge); + else + NumReplacements = replaceDominatedUsesWith( + NotCmp, NotVal, *DT, std::get<Instruction *>(Root)); + Changed |= NumReplacements > 0; + NumGVNEqProp += NumReplacements; + // Cached information for anything that uses NotCmp will be invalid. + if (MD) + MD->invalidateCachedPointerInfo(NotCmp); + } } } // Ensure that any instruction in scope that gets the "A < B" value number // is replaced with false. // The leader table only tracks basic blocks, not edges. Only add to if we // have the simple case where the edge dominates the end. - if (RootDominatesEnd) - LeaderTable.insert(Num, NotVal, Root.getEnd()); + for (const BasicBlock *BB : DominatedBlocks) + LeaderTable.insert(Num, NotVal, BB); continue; } @@ -2789,11 +2710,11 @@ bool GVNPass::processInstruction(Instruction *I) { Value *TrueVal = ConstantInt::getTrue(TrueSucc->getContext()); BasicBlockEdge TrueE(Parent, TrueSucc); - Changed |= propagateEquality(BranchCond, TrueVal, TrueE, true); + Changed |= propagateEquality(BranchCond, TrueVal, TrueE); Value *FalseVal = ConstantInt::getFalse(FalseSucc->getContext()); BasicBlockEdge FalseE(Parent, FalseSucc); - Changed |= propagateEquality(BranchCond, FalseVal, FalseE, true); + Changed |= propagateEquality(BranchCond, FalseVal, FalseE); return Changed; } @@ -2814,7 +2735,7 @@ bool GVNPass::processInstruction(Instruction *I) { // If there is only a single edge, propagate the case value into it. if (SwitchEdges.lookup(Dst) == 1) { BasicBlockEdge E(Parent, Dst); - Changed |= propagateEquality(SwitchCond, Case.getCaseValue(), E, true); + Changed |= propagateEquality(SwitchCond, Case.getCaseValue(), E); } } return Changed; @@ -2942,8 +2863,6 @@ bool GVNPass::processBlock(BasicBlock *BB) { if (DeadBlocks.count(BB)) return false; - // Clearing map before every BB because it can be used only for single BB. - ReplaceOperandsWithMap.clear(); bool ChangedFunction = false; // Since we may not have visited the input blocks of the phis, we can't @@ -2955,11 +2874,8 @@ bool GVNPass::processBlock(BasicBlock *BB) { for (PHINode *PN : PHINodesToRemove) { removeInstruction(PN); } - for (Instruction &Inst : make_early_inc_range(*BB)) { - if (!ReplaceOperandsWithMap.empty()) - ChangedFunction |= replaceOperandsForInBlockEquality(&Inst); + for (Instruction &Inst : make_early_inc_range(*BB)) ChangedFunction |= processInstruction(&Inst); - } return ChangedFunction; } diff --git a/llvm/lib/Transforms/Scalar/NewGVN.cpp b/llvm/lib/Transforms/Scalar/NewGVN.cpp index 3c1a8ba..80aa98d 100644 --- a/llvm/lib/Transforms/Scalar/NewGVN.cpp +++ b/llvm/lib/Transforms/Scalar/NewGVN.cpp @@ -434,10 +434,6 @@ private: int StoreCount = 0; }; -} // end anonymous namespace - -namespace llvm { - struct ExactEqualsExpression { const Expression &E; @@ -449,8 +445,9 @@ struct ExactEqualsExpression { return E.exactlyEquals(Other); } }; +} // end anonymous namespace -template <> struct DenseMapInfo<const Expression *> { +template <> struct llvm::DenseMapInfo<const Expression *> { static const Expression *getEmptyKey() { auto Val = static_cast<uintptr_t>(-1); Val <<= PointerLikeTypeTraits<const Expression *>::NumLowBitsAvailable; @@ -493,8 +490,6 @@ template <> struct DenseMapInfo<const Expression *> { } }; -} // end namespace llvm - namespace { class NewGVN { diff --git a/llvm/lib/Transforms/Utils/CanonicalizeFreezeInLoops.cpp b/llvm/lib/Transforms/Utils/CanonicalizeFreezeInLoops.cpp index 2190dcd..a87822c 100644 --- a/llvm/lib/Transforms/Utils/CanonicalizeFreezeInLoops.cpp +++ b/llvm/lib/Transforms/Utils/CanonicalizeFreezeInLoops.cpp @@ -84,10 +84,6 @@ public: bool run(); }; -} // anonymous namespace - -namespace llvm { - struct FrozenIndPHIInfo { // A freeze instruction that uses an induction phi FreezeInst *FI = nullptr; @@ -103,7 +99,9 @@ struct FrozenIndPHIInfo { bool operator==(const FrozenIndPHIInfo &Other) { return FI == Other.FI; } }; -template <> struct DenseMapInfo<FrozenIndPHIInfo> { +} // namespace + +template <> struct llvm::DenseMapInfo<FrozenIndPHIInfo> { static inline FrozenIndPHIInfo getEmptyKey() { return FrozenIndPHIInfo(DenseMapInfo<PHINode *>::getEmptyKey(), DenseMapInfo<BinaryOperator *>::getEmptyKey()); @@ -124,8 +122,6 @@ template <> struct DenseMapInfo<FrozenIndPHIInfo> { }; }; -} // end namespace llvm - // Given U = (value, user), replace value with freeze(value), and let // SCEV forget user. The inserted freeze is placed in the preheader. void CanonicalizeFreezeInLoopsImpl::InsertFreezeAndForgetFromSCEV(Use &U) { diff --git a/llvm/lib/Transforms/Utils/Local.cpp b/llvm/lib/Transforms/Utils/Local.cpp index b6ca52e..46f2903 100644 --- a/llvm/lib/Transforms/Utils/Local.cpp +++ b/llvm/lib/Transforms/Utils/Local.cpp @@ -3246,6 +3246,13 @@ unsigned llvm::replaceDominatedUsesWith(Value *From, Value *To, return ::replaceDominatedUsesWith(From, To, Dominates); } +unsigned llvm::replaceDominatedUsesWith(Value *From, Value *To, + DominatorTree &DT, + const Instruction *I) { + auto Dominates = [&](const Use &U) { return DT.dominates(I, U); }; + return ::replaceDominatedUsesWith(From, To, Dominates); +} + unsigned llvm::replaceDominatedUsesWithIf( Value *From, Value *To, DominatorTree &DT, const BasicBlockEdge &Root, function_ref<bool(const Use &U, const Value *To)> ShouldReplace) { @@ -3264,6 +3271,15 @@ unsigned llvm::replaceDominatedUsesWithIf( return ::replaceDominatedUsesWith(From, To, DominatesAndShouldReplace); } +unsigned llvm::replaceDominatedUsesWithIf( + Value *From, Value *To, DominatorTree &DT, const Instruction *I, + function_ref<bool(const Use &U, const Value *To)> ShouldReplace) { + auto DominatesAndShouldReplace = [&](const Use &U) { + return DT.dominates(I, U) && ShouldReplace(U, To); + }; + return ::replaceDominatedUsesWith(From, To, DominatesAndShouldReplace); +} + bool llvm::callsGCLeafFunction(const CallBase *Call, const TargetLibraryInfo &TLI) { // Check if the function is specifically marked as a gc leaf function. diff --git a/llvm/lib/Transforms/Utils/LowerInvoke.cpp b/llvm/lib/Transforms/Utils/LowerInvoke.cpp index ff2ab3c..cecb662 100644 --- a/llvm/lib/Transforms/Utils/LowerInvoke.cpp +++ b/llvm/lib/Transforms/Utils/LowerInvoke.cpp @@ -27,15 +27,15 @@ using namespace llvm; STATISTIC(NumInvokes, "Number of invokes replaced"); namespace { - class LowerInvokeLegacyPass : public FunctionPass { - public: - static char ID; // Pass identification, replacement for typeid - explicit LowerInvokeLegacyPass() : FunctionPass(ID) { - initializeLowerInvokeLegacyPassPass(*PassRegistry::getPassRegistry()); - } - bool runOnFunction(Function &F) override; - }; -} +class LowerInvokeLegacyPass : public FunctionPass { +public: + static char ID; // Pass identification, replacement for typeid + explicit LowerInvokeLegacyPass() : FunctionPass(ID) { + initializeLowerInvokeLegacyPassPass(*PassRegistry::getPassRegistry()); + } + bool runOnFunction(Function &F) override; +}; +} // namespace char LowerInvokeLegacyPass::ID = 0; INITIALIZE_PASS(LowerInvokeLegacyPass, "lowerinvoke", @@ -78,11 +78,12 @@ bool LowerInvokeLegacyPass::runOnFunction(Function &F) { return runImpl(F); } -namespace llvm { -char &LowerInvokePassID = LowerInvokeLegacyPass::ID; +char &llvm::LowerInvokePassID = LowerInvokeLegacyPass::ID; // Public Interface To the LowerInvoke pass. -FunctionPass *createLowerInvokePass() { return new LowerInvokeLegacyPass(); } +FunctionPass *llvm::createLowerInvokePass() { + return new LowerInvokeLegacyPass(); +} PreservedAnalyses LowerInvokePass::run(Function &F, FunctionAnalysisManager &AM) { @@ -92,4 +93,3 @@ PreservedAnalyses LowerInvokePass::run(Function &F, return PreservedAnalyses::none(); } -} diff --git a/llvm/lib/Transforms/Utils/MisExpect.cpp b/llvm/lib/Transforms/Utils/MisExpect.cpp index ca7e09d..1585e9e 100644 --- a/llvm/lib/Transforms/Utils/MisExpect.cpp +++ b/llvm/lib/Transforms/Utils/MisExpect.cpp @@ -48,8 +48,6 @@ using namespace llvm; using namespace misexpect; -namespace llvm { - // Command line option to enable/disable the warning when profile data suggests // a mismatch with the use of the llvm.expect intrinsic static cl::opt<bool> PGOWarnMisExpect( @@ -63,22 +61,18 @@ static cl::opt<uint32_t> MisExpectTolerance( cl::desc("Prevents emitting diagnostics when profile counts are " "within N% of the threshold..")); -} // namespace llvm - -namespace { - -bool isMisExpectDiagEnabled(LLVMContext &Ctx) { +static bool isMisExpectDiagEnabled(const LLVMContext &Ctx) { return PGOWarnMisExpect || Ctx.getMisExpectWarningRequested(); } -uint32_t getMisExpectTolerance(LLVMContext &Ctx) { +static uint32_t getMisExpectTolerance(const LLVMContext &Ctx) { return std::max(static_cast<uint32_t>(MisExpectTolerance), Ctx.getDiagnosticsMisExpectTolerance()); } -Instruction *getInstCondition(Instruction *I) { +static const Instruction *getInstCondition(const Instruction *I) { assert(I != nullptr && "MisExpect target Instruction cannot be nullptr"); - Instruction *Ret = nullptr; + const Instruction *Ret = nullptr; if (auto *B = dyn_cast<BranchInst>(I)) { Ret = dyn_cast<Instruction>(B->getCondition()); } @@ -97,8 +91,8 @@ Instruction *getInstCondition(Instruction *I) { return Ret ? Ret : I; } -void emitMisexpectDiagnostic(Instruction *I, LLVMContext &Ctx, - uint64_t ProfCount, uint64_t TotalCount) { +static void emitMisexpectDiagnostic(const Instruction *I, LLVMContext &Ctx, + uint64_t ProfCount, uint64_t TotalCount) { double PercentageCorrect = (double)ProfCount / TotalCount; auto PerString = formatv("{0:P} ({1} / {2})", PercentageCorrect, ProfCount, TotalCount); @@ -106,20 +100,16 @@ void emitMisexpectDiagnostic(Instruction *I, LLVMContext &Ctx, "Potential performance regression from use of the llvm.expect intrinsic: " "Annotation was correct on {0} of profiled executions.", PerString); - Instruction *Cond = getInstCondition(I); + const Instruction *Cond = getInstCondition(I); if (isMisExpectDiagEnabled(Ctx)) Ctx.diagnose(DiagnosticInfoMisExpect(Cond, Twine(PerString))); OptimizationRemarkEmitter ORE(I->getParent()->getParent()); ORE.emit(OptimizationRemark(DEBUG_TYPE, "misexpect", Cond) << RemStr.str()); } -} // namespace - -namespace llvm { -namespace misexpect { - -void verifyMisExpect(Instruction &I, ArrayRef<uint32_t> RealWeights, - ArrayRef<uint32_t> ExpectedWeights) { +void misexpect::verifyMisExpect(const Instruction &I, + ArrayRef<uint32_t> RealWeights, + ArrayRef<uint32_t> ExpectedWeights) { // To determine if we emit a diagnostic, we need to compare the branch weights // from the profile to those added by the llvm.expect intrinsic. // So first, we extract the "likely" and "unlikely" weights from @@ -128,15 +118,13 @@ void verifyMisExpect(Instruction &I, ArrayRef<uint32_t> RealWeights, uint64_t LikelyBranchWeight = 0, UnlikelyBranchWeight = std::numeric_limits<uint32_t>::max(); size_t MaxIndex = 0; - for (size_t Idx = 0, End = ExpectedWeights.size(); Idx < End; Idx++) { - uint32_t V = ExpectedWeights[Idx]; + for (const auto &[Idx, V] : enumerate(ExpectedWeights)) { if (LikelyBranchWeight < V) { LikelyBranchWeight = V; MaxIndex = Idx; } - if (UnlikelyBranchWeight > V) { + if (UnlikelyBranchWeight > V) UnlikelyBranchWeight = V; - } } const uint64_t ProfiledWeight = RealWeights[MaxIndex]; @@ -161,7 +149,7 @@ void verifyMisExpect(Instruction &I, ArrayRef<uint32_t> RealWeights, uint64_t ScaledThreshold = LikelyProbablilty.scale(RealWeightsTotal); // clamp tolerance range to [0, 100) - auto Tolerance = getMisExpectTolerance(I.getContext()); + uint32_t Tolerance = getMisExpectTolerance(I.getContext()); Tolerance = std::clamp(Tolerance, 0u, 99u); // Allow users to relax checking by N% i.e., if they use a 5% tolerance, @@ -175,8 +163,8 @@ void verifyMisExpect(Instruction &I, ArrayRef<uint32_t> RealWeights, RealWeightsTotal); } -void checkBackendInstrumentation(Instruction &I, - const ArrayRef<uint32_t> RealWeights) { +void misexpect::checkBackendInstrumentation(const Instruction &I, + ArrayRef<uint32_t> RealWeights) { // Backend checking assumes any existing weight comes from an `llvm.expect` // intrinsic. However, SampleProfiling + ThinLTO add branch weights multiple // times, leading to an invalid assumption in our checking. Backend checks @@ -190,24 +178,19 @@ void checkBackendInstrumentation(Instruction &I, verifyMisExpect(I, RealWeights, ExpectedWeights); } -void checkFrontendInstrumentation(Instruction &I, - const ArrayRef<uint32_t> ExpectedWeights) { +void misexpect::checkFrontendInstrumentation( + const Instruction &I, ArrayRef<uint32_t> ExpectedWeights) { SmallVector<uint32_t> RealWeights; if (!extractBranchWeights(I, RealWeights)) return; verifyMisExpect(I, RealWeights, ExpectedWeights); } -void checkExpectAnnotations(Instruction &I, - const ArrayRef<uint32_t> ExistingWeights, - bool IsFrontend) { - if (IsFrontend) { +void misexpect::checkExpectAnnotations(const Instruction &I, + ArrayRef<uint32_t> ExistingWeights, + bool IsFrontend) { + if (IsFrontend) checkFrontendInstrumentation(I, ExistingWeights); - } else { + else checkBackendInstrumentation(I, ExistingWeights); - } } - -} // namespace misexpect -} // namespace llvm -#undef DEBUG_TYPE diff --git a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp index 155fcc5..d831c27 100644 --- a/llvm/lib/Transforms/Utils/SimplifyCFG.cpp +++ b/llvm/lib/Transforms/Utils/SimplifyCFG.cpp @@ -5959,7 +5959,11 @@ bool SimplifyCFGOpt::turnSwitchRangeIntoICmp(SwitchInst *SI, unsigned PreviousEdges = OtherCases->size(); if (OtherDest == SI->getDefaultDest()) ++PreviousEdges; - for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) + unsigned E = PreviousEdges - 1; + // Remove all incoming values from OtherDest if OtherDest is unreachable. + if (NewBI->isUnconditional()) + ++E; + for (unsigned I = 0; I != E; ++I) cast<PHINode>(BBI)->removeIncomingValue(SI->getParent()); } @@ -7736,8 +7740,7 @@ struct SwitchSuccWrapper { DenseMap<PHINode *, SmallDenseMap<BasicBlock *, Value *, 8>> *PhiPredIVs; }; -namespace llvm { -template <> struct DenseMapInfo<const SwitchSuccWrapper *> { +template <> struct llvm::DenseMapInfo<const SwitchSuccWrapper *> { static const SwitchSuccWrapper *getEmptyKey() { return static_cast<SwitchSuccWrapper *>( DenseMapInfo<void *>::getEmptyKey()); @@ -7805,7 +7808,6 @@ template <> struct DenseMapInfo<const SwitchSuccWrapper *> { return true; } }; -} // namespace llvm bool SimplifyCFGOpt::simplifyDuplicateSwitchArms(SwitchInst *SI, DomTreeUpdater *DTU) { diff --git a/llvm/test/Analysis/DependenceAnalysis/PreliminaryNoValidityCheckFixedSize.ll b/llvm/test/Analysis/DependenceAnalysis/PreliminaryNoValidityCheckFixedSize.ll index e67cae7d..8cd29e6 100644 --- a/llvm/test/Analysis/DependenceAnalysis/PreliminaryNoValidityCheckFixedSize.ll +++ b/llvm/test/Analysis/DependenceAnalysis/PreliminaryNoValidityCheckFixedSize.ll @@ -20,7 +20,7 @@ define void @p2(i64 %n, ptr %A, ptr %B) nounwind uwtable ssp { ; CHECK-NEXT: Src: store i64 %i.011, ptr %arrayidx8, align 8 --> Dst: store i64 %i.011, ptr %arrayidx8, align 8 ; CHECK-NEXT: da analyze - none! ; CHECK-NEXT: Src: store i64 %i.011, ptr %arrayidx8, align 8 --> Dst: %0 = load i64, ptr %arrayidx17, align 8 -; CHECK-NEXT: da analyze - flow [-3 -2]! +; CHECK-NEXT: da analyze - flow [-3 -2] / assuming 1 loop level(s) fused: [-3 -2 -1]! ; CHECK-NEXT: Src: store i64 %i.011, ptr %arrayidx8, align 8 --> Dst: store i64 %0, ptr %B.addr.24, align 8 ; CHECK-NEXT: da analyze - confused! ; CHECK-NEXT: Src: %0 = load i64, ptr %arrayidx17, align 8 --> Dst: %0 = load i64, ptr %arrayidx17, align 8 diff --git a/llvm/test/Analysis/DependenceAnalysis/SameSDLoops.ll b/llvm/test/Analysis/DependenceAnalysis/SameSDLoops.ll index f80e2cf..57962e0 100644 --- a/llvm/test/Analysis/DependenceAnalysis/SameSDLoops.ll +++ b/llvm/test/Analysis/DependenceAnalysis/SameSDLoops.ll @@ -18,7 +18,7 @@ define void @samebd0(ptr %A) nounwind uwtable ssp { ; CHECK-NEXT: Src: store i64 %i.013, ptr %arrayidx12, align 8 --> Dst: store i64 %i.013, ptr %arrayidx12, align 8 ; CHECK-NEXT: da analyze - none! ; CHECK-NEXT: Src: store i64 %i.013, ptr %arrayidx12, align 8 --> Dst: store i64 %l17.04, ptr %arrayidx24, align 8 -; CHECK-NEXT: da analyze - output [-4 -3]! / assuming 2 loop level(s) fused: [-4 -3 -3 -1]! +; CHECK-NEXT: da analyze - output [-4 -3] / assuming 2 loop level(s) fused: [-4 -3 -3 -1]! ; CHECK-NEXT: Src: store i64 %l17.04, ptr %arrayidx24, align 8 --> Dst: store i64 %l17.04, ptr %arrayidx24, align 8 ; CHECK-NEXT: da analyze - none! ; @@ -96,7 +96,7 @@ define void @samebd1(ptr %A) nounwind uwtable ssp { ; CHECK-NEXT: Src: store i64 %i.03, ptr %arrayidx, align 4 --> Dst: store i64 %i.03, ptr %arrayidx, align 4 ; CHECK-NEXT: da analyze - none! ; CHECK-NEXT: Src: store i64 %i.03, ptr %arrayidx, align 4 --> Dst: %0 = load i64, ptr %arrayidx7, align 4 -; CHECK-NEXT: da analyze - flow [|<]! / assuming 1 loop level(s) fused: [<=|<]! +; CHECK-NEXT: da analyze - flow [|<] / assuming 1 loop level(s) fused: [<=|<]! ; CHECK-NEXT: Src: %0 = load i64, ptr %arrayidx7, align 4 --> Dst: %0 = load i64, ptr %arrayidx7, align 4 ; CHECK-NEXT: da analyze - none! ; @@ -148,7 +148,7 @@ define void @non_samebd0(ptr %A) nounwind uwtable ssp { ; CHECK-NEXT: Src: store i64 %i.013, ptr %arrayidx12, align 8 --> Dst: store i64 %i.013, ptr %arrayidx12, align 8 ; CHECK-NEXT: da analyze - none! ; CHECK-NEXT: Src: store i64 %i.013, ptr %arrayidx12, align 8 --> Dst: store i64 %l17.04, ptr %arrayidx24, align 8 -; CHECK-NEXT: da analyze - output [-4 -3]!{{$}} +; CHECK-NEXT: da analyze - output [-4 -3]! ; CHECK-NEXT: Src: store i64 %l17.04, ptr %arrayidx24, align 8 --> Dst: store i64 %l17.04, ptr %arrayidx24, align 8 ; CHECK-NEXT: da analyze - none! ; @@ -227,7 +227,7 @@ define void @non_samebd1(ptr %A) nounwind uwtable ssp { ; CHECK-NEXT: Src: store i64 %i.03, ptr %arrayidx, align 4 --> Dst: store i64 %i.03, ptr %arrayidx, align 4 ; CHECK-NEXT: da analyze - none! ; CHECK-NEXT: Src: store i64 %i.03, ptr %arrayidx, align 4 --> Dst: %0 = load i64, ptr %arrayidx7, align 4 -; CHECK-NEXT: da analyze - flow [|<]!{{$}} +; CHECK-NEXT: da analyze - flow [|<]! ; CHECK-NEXT: Src: %0 = load i64, ptr %arrayidx7, align 4 --> Dst: %0 = load i64, ptr %arrayidx7, align 4 ; CHECK-NEXT: da analyze - none! ; diff --git a/llvm/test/Analysis/IR2Vec/unreachable.ll b/llvm/test/Analysis/IR2Vec/unreachable.ll index 9be0ee1..627e2c9 100644 --- a/llvm/test/Analysis/IR2Vec/unreachable.ll +++ b/llvm/test/Analysis/IR2Vec/unreachable.ll @@ -30,13 +30,17 @@ return: ; preds = %if.else, %if.then %4 = load i32, ptr %retval, align 4 ret i32 %4 } - -; CHECK: Basic block vectors: +; We'll get individual basic block embeddings for all blocks in the function. +; But unreachable blocks are not counted for computing the function embedding. +; CHECK: Function vector: [ 1301.20 1318.20 1335.20 ] +; CHECK-NEXT: Basic block vectors: ; CHECK-NEXT: Basic block: entry: ; CHECK-NEXT: [ 816.20 825.20 834.20 ] ; CHECK-NEXT: Basic block: if.then: ; CHECK-NEXT: [ 195.00 198.00 201.00 ] ; CHECK-NEXT: Basic block: if.else: ; CHECK-NEXT: [ 195.00 198.00 201.00 ] +; CHECK-NEXT: Basic block: unreachable: +; CHECK-NEXT: [ 101.00 103.00 105.00 ] ; CHECK-NEXT: Basic block: return: ; CHECK-NEXT: [ 95.00 97.00 99.00 ] diff --git a/llvm/test/Assembler/dicompileunit-conflicting-language-fields.ll b/llvm/test/Assembler/dicompileunit-conflicting-language-fields.ll new file mode 100644 index 0000000..3aad27b --- /dev/null +++ b/llvm/test/Assembler/dicompileunit-conflicting-language-fields.ll @@ -0,0 +1,4 @@ +; RUN: not llvm-as < %s -disable-output 2>&1 | FileCheck %s + +; CHECK: <stdin>:[[@LINE+1]]:15: error: can only specify one of 'language' and 'sourceLanguageName' on !DICompileUnit +!0 = distinct !DICompileUnit(language: DW_LANG_C, sourceLanguageName: DW_LNAME_C, file: !DIFile(filename: "a", directory: "b")) diff --git a/llvm/test/Assembler/dicompileunit-invalid-language.ll b/llvm/test/Assembler/dicompileunit-invalid-language.ll new file mode 100644 index 0000000..da93c4f --- /dev/null +++ b/llvm/test/Assembler/dicompileunit-invalid-language.ll @@ -0,0 +1,22 @@ +; RUN: split-file %s %t +; RUN: not llvm-as < %t/invalid_dw_lang.ll -disable-output 2>&1 | FileCheck %s --check-prefix=INVALID_DW_LANG +; RUN: not llvm-as < %t/invalid_dw_lang_2.ll -disable-output 2>&1 | FileCheck %s --check-prefix=INVALID_DW_LANG_2 +; RUN: not llvm-as < %t/invalid_dw_lname.ll -disable-output 2>&1 | FileCheck %s --check-prefix=INVALID_DW_LNAME +; RUN: not llvm-as < %t/invalid_dw_lname_2.ll -disable-output 2>&1 | FileCheck %s --check-prefix=INVALID_DW_LNAME_2 + +; INVALID_DW_LANG: invalid DWARF language 'DW_LANG_blah' +; INVALID_DW_LANG_2: expected DWARF language +; INVALID_DW_LNAME: invalid DWARF source language name 'DW_LNAME_blah' +; INVALID_DW_LNAME_2: expected DWARF source language name + +;--- invalid_dw_lang.ll +!0 = distinct !DICompileUnit(language: DW_LANG_blah) + +;--- invalid_dw_lang_2.ll +!0 = distinct !DICompileUnit(language: DW_LNAME_C) + +;--- invalid_dw_lname.ll +!0 = distinct !DICompileUnit(sourceLanguageName: DW_LNAME_blah) + +;--- invalid_dw_lname_2.ll +!0 = distinct !DICompileUnit(sourceLanguageName: DW_LANG_C) diff --git a/llvm/test/Assembler/invalid-dicompileunit-missing-language.ll b/llvm/test/Assembler/invalid-dicompileunit-missing-language.ll index 8e4cb02..ebc86e3 100644 --- a/llvm/test/Assembler/invalid-dicompileunit-missing-language.ll +++ b/llvm/test/Assembler/invalid-dicompileunit-missing-language.ll @@ -1,4 +1,4 @@ ; RUN: not llvm-as < %s -disable-output 2>&1 | FileCheck %s -; CHECK: <stdin>:[[@LINE+1]]:74: error: missing required field 'language' +; CHECK: <stdin>:[[@LINE+1]]:15: error: missing one of 'language' or 'sourceLanguageName', required for !DICompileUnit !0 = distinct !DICompileUnit(file: !DIFile(filename: "a", directory: "b")) diff --git a/llvm/test/Bitcode/dwarf-source-language-name.ll b/llvm/test/Bitcode/dwarf-source-language-name.ll new file mode 100644 index 0000000..e989363 --- /dev/null +++ b/llvm/test/Bitcode/dwarf-source-language-name.ll @@ -0,0 +1,15 @@ +; RUN: llvm-as < %s | llvm-dis | llvm-as | llvm-dis | FileCheck %s + +; CHECK: sourceLanguageName: DW_LNAME_ObjC_plus_plus + +source_filename = "cu.cpp" +target triple = "arm64-apple-macosx" + +!llvm.dbg.cu = !{!0} +!llvm.module.flags = !{!3, !4} + +!0 = distinct !DICompileUnit(sourceLanguageName: DW_LNAME_ObjC_plus_plus, file: !1, producer: "handwritten", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, globals: !2, splitDebugInlining: false, nameTableKind: Apple, sysroot: "/") +!1 = !DIFile(filename: "cu.cpp", directory: "/tmp") +!2 = !{} +!3 = !{i32 7, !"Dwarf Version", i32 5} +!4 = !{i32 2, !"Debug Info Version", i32 3} diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir new file mode 100644 index 0000000..b9cde95 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-smulh.mir @@ -0,0 +1,137 @@ +# NOTE: Assertions have been autogenerated by utils/update_givaluetracking_test_checks.py UTC_ARGS: --version 6 +# RUN: llc -mtriple=aarch64 -passes='print<gisel-value-tracking>' -filetype=null %s 2>&1 | FileCheck %s + +--- +name: Cst +body: | + bb.0: + ; CHECK-LABEL: name: @Cst + ; CHECK-NEXT: %0:_ KnownBits:00010011 SignBits:3 + ; CHECK-NEXT: %1:_ KnownBits:11100000 SignBits:3 + ; CHECK-NEXT: %2:_ KnownBits:11111101 SignBits:6 + %0:_(s8) = G_CONSTANT i8 19 + %1:_(s8) = G_CONSTANT i8 224 + %2:_(s8) = G_SMULH %0, %1 +... +--- +name: CstZero +body: | + bb.0: + ; CHECK-LABEL: name: @CstZero + ; CHECK-NEXT: %0:_ KnownBits:11111111 SignBits:8 + ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8 + ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8 + %0:_(s8) = G_CONSTANT i8 255 + %1:_(s8) = G_CONSTANT i8 0 + %2:_(s8) = G_SMULH %0, %1 +... +--- +name: ScalarVar +body: | + bb.0: + ; CHECK-LABEL: name: @ScalarVar + ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1 + %0:_(s8) = COPY $b0 + %1:_(s8) = COPY $b1 + %2:_(s8) = G_SMULH %0, %1 +... +--- +name: ScalarZero +body: | + bb.0: + ; CHECK-LABEL: name: @ScalarZero + ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8 + ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8 + %0:_(s8) = COPY $b0 + %1:_(s8) = G_CONSTANT i8 0 + %2:_(s8) = G_SMULH %0, %1 +... +--- +name: ScalarVarAbs +body: | + bb.0: + ; CHECK-LABEL: name: @ScalarVarAbs + ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %2:_ KnownBits:???????????????? SignBits:9 + ; CHECK-NEXT: %3:_ KnownBits:0000000000000001 SignBits:15 + ; CHECK-NEXT: %4:_ KnownBits:???????????????? SignBits:1 + %0:_(s8) = COPY $b0 + %1:_(s8) = G_ABS %0 + %2:_(s16) = G_SEXT %1 + %3:_(s16) = G_CONSTANT i16 1 + %4:_(s16) = G_SMULH %2, %3 +... +--- +name: SplatVecCst +body: | + bb.0: + ; CHECK-LABEL: name: @SplatVecCst + ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %2:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %3:_ KnownBits:00000000 SignBits:8 + %0:_(s8) = G_CONSTANT i8 250 + %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8) + %2:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8) + %3:_(<vscale x 16 x s8>) = G_SMULH %1, %2 +... +--- +name: SplatVecPartScalar +body: | + bb.0: + ; CHECK-LABEL: name: @SplatVecPartScalar + ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4 + ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %6:_ KnownBits:???????? SignBits:1 + %0:_(s8) = G_CONSTANT i8 250 + %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8) + %2:_(s8) = G_IMPLICIT_DEF + %3:_(s8) = G_CONSTANT i8 15 + %4:_(s8) = G_AND %2, %3 + %5:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %4(s8) + %6:_(<vscale x 16 x s8>) = G_SMULH %1, %5 +... +--- +name: VecCst +body: | + bb.0: + ; CHECK-LABEL: name: @VecCst + ; CHECK-NEXT: %0:_ KnownBits:00011001 SignBits:3 + ; CHECK-NEXT: %1:_ KnownBits:11100001 SignBits:3 + ; CHECK-NEXT: %2:_ KnownBits:?????001 SignBits:3 + ; CHECK-NEXT: %3:_ KnownBits:?????001 SignBits:3 + ; CHECK-NEXT: %4:_ KnownBits:???????? SignBits:1 + %0:_(s8) = G_CONSTANT i8 25 + %1:_(s8) = G_CONSTANT i8 225 + %2:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8) + %3:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8) + %4:_(<2 x s8>) = G_SMULH %2, %3 +... +--- +name: VecPartScalar +body: | + bb.0: + ; CHECK-LABEL: name: @VecPartScalar + ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4 + ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %6:_ KnownBits:???????? SignBits:1 + %0:_(s8) = G_CONSTANT i8 250 + %1:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %0:_(s8) + %2:_(s8) = G_IMPLICIT_DEF + %3:_(s8) = G_CONSTANT i8 15 + %4:_(s8) = G_AND %2, %3 + %5:_(<2 x s8>) = G_BUILD_VECTOR %4:_(s8), %4:_(s8) + %6:_(<2 x s8>) = G_SMULH %1, %5 +... diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir new file mode 100644 index 0000000..debdbaa --- /dev/null +++ b/llvm/test/CodeGen/AArch64/GlobalISel/knownbits-umulh.mir @@ -0,0 +1,137 @@ +# NOTE: Assertions have been autogenerated by utils/update_givaluetracking_test_checks.py UTC_ARGS: --version 6 +# RUN: llc -mtriple=aarch64 -passes='print<gisel-value-tracking>' -filetype=null %s 2>&1 | FileCheck %s + +--- +name: Cst +body: | + bb.0: + ; CHECK-LABEL: name: @Cst + ; CHECK-NEXT: %0:_ KnownBits:00010011 SignBits:3 + ; CHECK-NEXT: %1:_ KnownBits:11100000 SignBits:3 + ; CHECK-NEXT: %2:_ KnownBits:00010000 SignBits:3 + %0:_(s8) = G_CONSTANT i8 19 + %1:_(s8) = G_CONSTANT i8 224 + %2:_(s8) = G_UMULH %0, %1 +... +--- +name: CstZero +body: | + bb.0: + ; CHECK-LABEL: name: @CstZero + ; CHECK-NEXT: %0:_ KnownBits:11111111 SignBits:8 + ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8 + ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8 + %0:_(s8) = G_CONSTANT i8 255 + %1:_(s8) = G_CONSTANT i8 0 + %2:_(s8) = G_UMULH %0, %1 +... +--- +name: ScalarVar +body: | + bb.0: + ; CHECK-LABEL: name: @ScalarVar + ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1 + %0:_(s8) = COPY $b0 + %1:_(s8) = COPY $b1 + %2:_(s8) = G_UMULH %0, %1 +... +--- +name: ScalarZero +body: | + bb.0: + ; CHECK-LABEL: name: @ScalarZero + ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %1:_ KnownBits:00000000 SignBits:8 + ; CHECK-NEXT: %2:_ KnownBits:00000000 SignBits:8 + %0:_(s8) = COPY $b0 + %1:_(s8) = G_CONSTANT i8 0 + %2:_(s8) = G_UMULH %0, %1 +... +--- +name: ScalarVarAbs +body: | + bb.0: + ; CHECK-LABEL: name: @ScalarVarAbs + ; CHECK-NEXT: %0:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %1:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %2:_ KnownBits:???????????????? SignBits:9 + ; CHECK-NEXT: %3:_ KnownBits:0000000000000001 SignBits:15 + ; CHECK-NEXT: %4:_ KnownBits:0000000000000000 SignBits:16 + %0:_(s8) = COPY $b0 + %1:_(s8) = G_ABS %0 + %2:_(s16) = G_SEXT %1 + %3:_(s16) = G_CONSTANT i16 1 + %4:_(s16) = G_UMULH %2, %3 +... +--- +name: SplatVecCst +body: | + bb.0: + ; CHECK-LABEL: name: @SplatVecCst + ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %2:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %3:_ KnownBits:11110100 SignBits:4 + %0:_(s8) = G_CONSTANT i8 250 + %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8) + %2:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8) + %3:_(<vscale x 16 x s8>) = G_UMULH %1, %2 +... +--- +name: SplatVecPartScalar +body: | + bb.0: + ; CHECK-LABEL: name: @SplatVecPartScalar + ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4 + ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %6:_ KnownBits:0000???? SignBits:4 + %0:_(s8) = G_CONSTANT i8 250 + %1:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %0(s8) + %2:_(s8) = G_IMPLICIT_DEF + %3:_(s8) = G_CONSTANT i8 15 + %4:_(s8) = G_AND %2, %3 + %5:_(<vscale x 16 x s8>) = G_SPLAT_VECTOR %4(s8) + %6:_(<vscale x 16 x s8>) = G_UMULH %1, %5 +... +--- +name: VecCst +body: | + bb.0: + ; CHECK-LABEL: name: @VecCst + ; CHECK-NEXT: %0:_ KnownBits:00011001 SignBits:3 + ; CHECK-NEXT: %1:_ KnownBits:11100001 SignBits:3 + ; CHECK-NEXT: %2:_ KnownBits:?????001 SignBits:3 + ; CHECK-NEXT: %3:_ KnownBits:?????001 SignBits:3 + ; CHECK-NEXT: %4:_ KnownBits:???????? SignBits:1 + %0:_(s8) = G_CONSTANT i8 25 + %1:_(s8) = G_CONSTANT i8 225 + %2:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8) + %3:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %1:_(s8) + %4:_(<2 x s8>) = G_UMULH %2, %3 +... +--- +name: VecPartScalar +body: | + bb.0: + ; CHECK-LABEL: name: @VecPartScalar + ; CHECK-NEXT: %0:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %1:_ KnownBits:11111010 SignBits:5 + ; CHECK-NEXT: %2:_ KnownBits:???????? SignBits:1 + ; CHECK-NEXT: %3:_ KnownBits:00001111 SignBits:4 + ; CHECK-NEXT: %4:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %5:_ KnownBits:0000???? SignBits:4 + ; CHECK-NEXT: %6:_ KnownBits:0000???? SignBits:4 + %0:_(s8) = G_CONSTANT i8 250 + %1:_(<2 x s8>) = G_BUILD_VECTOR %0:_(s8), %0:_(s8) + %2:_(s8) = G_IMPLICIT_DEF + %3:_(s8) = G_CONSTANT i8 15 + %4:_(s8) = G_AND %2, %3 + %5:_(<2 x s8>) = G_BUILD_VECTOR %4:_(s8), %4:_(s8) + %6:_(<2 x s8>) = G_UMULH %1, %5 +... diff --git a/llvm/test/CodeGen/AArch64/arm64-indexed-memory.ll b/llvm/test/CodeGen/AArch64/arm64-indexed-memory.ll index 322a96a..e8e5631 100644 --- a/llvm/test/CodeGen/AArch64/arm64-indexed-memory.ll +++ b/llvm/test/CodeGen/AArch64/arm64-indexed-memory.ll @@ -739,14 +739,12 @@ define ptr @postidx32_shalf(ptr %src, ptr %out, half %a) { ; ; GISEL-LABEL: postidx32_shalf: ; GISEL: ; %bb.0: -; GISEL-NEXT: movi d1, #0000000000000000 -; GISEL-NEXT: ldr h2, [x0], #4 +; GISEL-NEXT: ldr h1, [x0], #4 ; GISEL-NEXT: ; kill: def $h0 killed $h0 def $s0 ; GISEL-NEXT: fmov w9, s0 -; GISEL-NEXT: fcvt s3, h2 -; GISEL-NEXT: fmov w8, s2 -; GISEL-NEXT: fcvt s1, h1 -; GISEL-NEXT: fcmp s3, s1 +; GISEL-NEXT: fcvt s2, h1 +; GISEL-NEXT: fmov w8, s1 +; GISEL-NEXT: fcmp s2, #0.0 ; GISEL-NEXT: csel w8, w8, w9, mi ; GISEL-NEXT: strh w8, [x1] ; GISEL-NEXT: ret diff --git a/llvm/test/CodeGen/AArch64/f16-instructions.ll b/llvm/test/CodeGen/AArch64/f16-instructions.ll index b234ef7..085170c 100644 --- a/llvm/test/CodeGen/AArch64/f16-instructions.ll +++ b/llvm/test/CodeGen/AArch64/f16-instructions.ll @@ -782,18 +782,16 @@ define void @test_fccmp(half %in, ptr %out) { ; ; CHECK-CVT-GI-LABEL: test_fccmp: ; CHECK-CVT-GI: // %bb.0: -; CHECK-CVT-GI-NEXT: adrp x8, .LCPI29_0 ; CHECK-CVT-GI-NEXT: // kill: def $h0 killed $h0 def $s0 -; CHECK-CVT-GI-NEXT: fcvt s2, h0 -; CHECK-CVT-GI-NEXT: ldr h1, [x8, :lo12:.LCPI29_0] -; CHECK-CVT-GI-NEXT: adrp x8, .LCPI29_1 -; CHECK-CVT-GI-NEXT: ldr h4, [x8, :lo12:.LCPI29_1] +; CHECK-CVT-GI-NEXT: fcvt s1, h0 +; CHECK-CVT-GI-NEXT: fmov s2, #5.00000000 +; CHECK-CVT-GI-NEXT: adrp x8, .LCPI29_0 +; CHECK-CVT-GI-NEXT: fmov s3, #8.00000000 +; CHECK-CVT-GI-NEXT: fcmp s1, s2 +; CHECK-CVT-GI-NEXT: ldr h2, [x8, :lo12:.LCPI29_0] ; CHECK-CVT-GI-NEXT: fmov w8, s0 -; CHECK-CVT-GI-NEXT: fcvt s3, h1 -; CHECK-CVT-GI-NEXT: fmov w9, s1 -; CHECK-CVT-GI-NEXT: fcvt s4, h4 -; CHECK-CVT-GI-NEXT: fcmp s2, s3 -; CHECK-CVT-GI-NEXT: fccmp s2, s4, #4, mi +; CHECK-CVT-GI-NEXT: fmov w9, s2 +; CHECK-CVT-GI-NEXT: fccmp s1, s3, #4, mi ; CHECK-CVT-GI-NEXT: csel w8, w8, w9, gt ; CHECK-CVT-GI-NEXT: strh w8, [x0] ; CHECK-CVT-GI-NEXT: ret diff --git a/llvm/test/CodeGen/AArch64/fcvt-fixed.ll b/llvm/test/CodeGen/AArch64/fcvt-fixed.ll index 7409bfb..743d160 100644 --- a/llvm/test/CodeGen/AArch64/fcvt-fixed.ll +++ b/llvm/test/CodeGen/AArch64/fcvt-fixed.ll @@ -149,33 +149,21 @@ define i64 @fcvtzs_f64_i64_64(double %dbl) { } define i32 @fcvtzs_f16_i32_7(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzs_f16_i32_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_f16_i32_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_f16_i32_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs w0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_f16_i32_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI8_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI8_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_f16_i32_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI8_0 @@ -189,33 +177,21 @@ define i32 @fcvtzs_f16_i32_7(half %flt) { } define i32 @fcvtzs_f16_i32_15(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzs_f16_i32_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_f16_i32_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_f16_i32_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs w0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_f16_i32_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI9_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI9_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_f16_i32_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI9_0 @@ -229,33 +205,21 @@ define i32 @fcvtzs_f16_i32_15(half %flt) { } define i64 @fcvtzs_f16_i64_7(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzs_f16_i64_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_f16_i64_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_f16_i64_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs x0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_f16_i64_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI10_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI10_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_f16_i64_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI10_0 @@ -269,33 +233,21 @@ define i64 @fcvtzs_f16_i64_7(half %flt) { } define i64 @fcvtzs_f16_i64_15(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzs_f16_i64_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_f16_i64_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_f16_i64_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs x0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_f16_i64_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI11_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI11_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_f16_i64_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI11_0 @@ -453,33 +405,21 @@ define i64 @fcvtzu_f64_i64_64(double %dbl) { } define i32 @fcvtzu_f16_i32_7(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzu_f16_i32_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_f16_i32_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_f16_i32_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu w0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_f16_i32_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI20_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI20_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_f16_i32_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI20_0 @@ -493,33 +433,21 @@ define i32 @fcvtzu_f16_i32_7(half %flt) { } define i32 @fcvtzu_f16_i32_15(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzu_f16_i32_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_f16_i32_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_f16_i32_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu w0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_f16_i32_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI21_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI21_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_f16_i32_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI21_0 @@ -533,33 +461,21 @@ define i32 @fcvtzu_f16_i32_15(half %flt) { } define i64 @fcvtzu_f16_i64_7(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzu_f16_i64_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_f16_i64_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_f16_i64_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu x0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_f16_i64_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI22_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI22_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_f16_i64_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI22_0 @@ -573,33 +489,21 @@ define i64 @fcvtzu_f16_i64_7(half %flt) { } define i64 @fcvtzu_f16_i64_15(half %flt) { -; CHECK-SD-NO16-LABEL: fcvtzu_f16_i64_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_f16_i64_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_f16_i64_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu x0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_f16_i64_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI23_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI23_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_f16_i64_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI23_0 @@ -774,13 +678,11 @@ define half @scvtf_f16_i32_7(i32 %int) { ; ; CHECK-GI-NO16-LABEL: scvtf_f16_i32_7: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: scvtf s0, w0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI32_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI32_0] +; CHECK-GI-NO16-NEXT: scvtf s1, w0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #67, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -814,13 +716,11 @@ define half @scvtf_f16_i32_15(i32 %int) { ; ; CHECK-GI-NO16-LABEL: scvtf_f16_i32_15: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: scvtf s0, w0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI33_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI33_0] +; CHECK-GI-NO16-NEXT: scvtf s1, w0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #71, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -854,13 +754,11 @@ define half @scvtf_f16_i64_7(i64 %long) { ; ; CHECK-GI-NO16-LABEL: scvtf_f16_i64_7: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: scvtf s0, x0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI34_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI34_0] +; CHECK-GI-NO16-NEXT: scvtf s1, x0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #67, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -894,13 +792,11 @@ define half @scvtf_f16_i64_15(i64 %long) { ; ; CHECK-GI-NO16-LABEL: scvtf_f16_i64_15: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: scvtf s0, x0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI35_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI35_0] +; CHECK-GI-NO16-NEXT: scvtf s1, x0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #71, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -1078,13 +974,11 @@ define half @ucvtf_f16_i32_7(i32 %int) { ; ; CHECK-GI-NO16-LABEL: ucvtf_f16_i32_7: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: ucvtf s0, w0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI44_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI44_0] +; CHECK-GI-NO16-NEXT: ucvtf s1, w0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #67, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -1118,13 +1012,11 @@ define half @ucvtf_f16_i32_15(i32 %int) { ; ; CHECK-GI-NO16-LABEL: ucvtf_f16_i32_15: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: ucvtf s0, w0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI45_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI45_0] +; CHECK-GI-NO16-NEXT: ucvtf s1, w0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #71, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -1158,13 +1050,11 @@ define half @ucvtf_f16_i64_7(i64 %long) { ; ; CHECK-GI-NO16-LABEL: ucvtf_f16_i64_7: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: ucvtf s0, x0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI46_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI46_0] +; CHECK-GI-NO16-NEXT: ucvtf s1, x0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #67, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -1198,13 +1088,11 @@ define half @ucvtf_f16_i64_15(i64 %long) { ; ; CHECK-GI-NO16-LABEL: ucvtf_f16_i64_15: ; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: ucvtf s0, x0 -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI47_0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI47_0] +; CHECK-GI-NO16-NEXT: ucvtf s1, x0 +; CHECK-GI-NO16-NEXT: movi v0.2s, #71, lsl #24 +; CHECK-GI-NO16-NEXT: fcvt h1, s1 ; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fdiv s0, s0, s1 +; CHECK-GI-NO16-NEXT: fdiv s0, s1, s0 ; CHECK-GI-NO16-NEXT: fcvt h0, s0 ; CHECK-GI-NO16-NEXT: ret ; @@ -1356,33 +1244,21 @@ define i64 @fcvtzs_sat_f64_i64_64(double %dbl) { } define i32 @fcvtzs_sat_f16_i32_7(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzs_sat_f16_i32_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_sat_f16_i32_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_sat_f16_i32_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs w0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_sat_f16_i32_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI55_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI55_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_sat_f16_i32_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI55_0 @@ -1396,33 +1272,21 @@ define i32 @fcvtzs_sat_f16_i32_7(half %dbl) { } define i32 @fcvtzs_sat_f16_i32_15(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzs_sat_f16_i32_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_sat_f16_i32_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_sat_f16_i32_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs w0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_sat_f16_i32_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI56_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI56_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_sat_f16_i32_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI56_0 @@ -1436,33 +1300,21 @@ define i32 @fcvtzs_sat_f16_i32_15(half %dbl) { } define i64 @fcvtzs_sat_f16_i64_7(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzs_sat_f16_i64_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_sat_f16_i64_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_sat_f16_i64_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs x0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_sat_f16_i64_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI57_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI57_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_sat_f16_i64_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI57_0 @@ -1476,33 +1328,21 @@ define i64 @fcvtzs_sat_f16_i64_7(half %dbl) { } define i64 @fcvtzs_sat_f16_i64_15(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzs_sat_f16_i64_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzs x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzs_sat_f16_i64_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzs x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzs_sat_f16_i64_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzs x0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzs_sat_f16_i64_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI58_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI58_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzs x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzs_sat_f16_i64_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI58_0 @@ -1650,33 +1490,21 @@ define i64 @fcvtzu_sat_f64_i64_64(double %dbl) { } define i32 @fcvtzu_sat_f16_i32_7(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzu_sat_f16_i32_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_sat_f16_i32_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_sat_f16_i32_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu w0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_sat_f16_i32_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI66_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI66_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_sat_f16_i32_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI66_0 @@ -1690,33 +1518,21 @@ define i32 @fcvtzu_sat_f16_i32_7(half %dbl) { } define i32 @fcvtzu_sat_f16_i32_15(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzu_sat_f16_i32_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu w0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_sat_f16_i32_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu w0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_sat_f16_i32_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu w0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_sat_f16_i32_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI67_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI67_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu w0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_sat_f16_i32_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI67_0 @@ -1730,33 +1546,21 @@ define i32 @fcvtzu_sat_f16_i32_15(half %dbl) { } define i64 @fcvtzu_sat_f16_i64_7(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzu_sat_f16_i64_7: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #67, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_sat_f16_i64_7: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #67, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_sat_f16_i64_7: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu x0, h0, #7 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_sat_f16_i64_7: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI68_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI68_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_sat_f16_i64_7: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI68_0 @@ -1770,33 +1574,21 @@ define i64 @fcvtzu_sat_f16_i64_7(half %dbl) { } define i64 @fcvtzu_sat_f16_i64_15(half %dbl) { -; CHECK-SD-NO16-LABEL: fcvtzu_sat_f16_i64_15: -; CHECK-SD-NO16: // %bb.0: -; CHECK-SD-NO16-NEXT: movi v1.2s, #71, lsl #24 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fmul s0, s0, s1 -; CHECK-SD-NO16-NEXT: fcvt h0, s0 -; CHECK-SD-NO16-NEXT: fcvt s0, h0 -; CHECK-SD-NO16-NEXT: fcvtzu x0, s0 -; CHECK-SD-NO16-NEXT: ret +; CHECK-NO16-LABEL: fcvtzu_sat_f16_i64_15: +; CHECK-NO16: // %bb.0: +; CHECK-NO16-NEXT: movi v1.2s, #71, lsl #24 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fmul s0, s0, s1 +; CHECK-NO16-NEXT: fcvt h0, s0 +; CHECK-NO16-NEXT: fcvt s0, h0 +; CHECK-NO16-NEXT: fcvtzu x0, s0 +; CHECK-NO16-NEXT: ret ; ; CHECK-SD-FP16-LABEL: fcvtzu_sat_f16_i64_15: ; CHECK-SD-FP16: // %bb.0: ; CHECK-SD-FP16-NEXT: fcvtzu x0, h0, #15 ; CHECK-SD-FP16-NEXT: ret ; -; CHECK-GI-NO16-LABEL: fcvtzu_sat_f16_i64_15: -; CHECK-GI-NO16: // %bb.0: -; CHECK-GI-NO16-NEXT: adrp x8, .LCPI69_0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: ldr h1, [x8, :lo12:.LCPI69_0] -; CHECK-GI-NO16-NEXT: fcvt s1, h1 -; CHECK-GI-NO16-NEXT: fmul s0, s0, s1 -; CHECK-GI-NO16-NEXT: fcvt h0, s0 -; CHECK-GI-NO16-NEXT: fcvt s0, h0 -; CHECK-GI-NO16-NEXT: fcvtzu x0, s0 -; CHECK-GI-NO16-NEXT: ret -; ; CHECK-GI-FP16-LABEL: fcvtzu_sat_f16_i64_15: ; CHECK-GI-FP16: // %bb.0: ; CHECK-GI-FP16-NEXT: adrp x8, .LCPI69_0 @@ -1811,4 +1603,3 @@ define i64 @fcvtzu_sat_f16_i64_15(half %dbl) { ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: ; CHECK: {{.*}} ; CHECK-FP16: {{.*}} -; CHECK-NO16: {{.*}} diff --git a/llvm/test/CodeGen/AArch64/frem-power2.ll b/llvm/test/CodeGen/AArch64/frem-power2.ll index 98276b6..e1bc742 100644 --- a/llvm/test/CodeGen/AArch64/frem-power2.ll +++ b/llvm/test/CodeGen/AArch64/frem-power2.ll @@ -100,9 +100,8 @@ define half @hrem2_nsz(half %x) { ; CHECK-GI-NEXT: str x30, [sp, #-16]! // 8-byte Folded Spill ; CHECK-GI-NEXT: .cfi_def_cfa_offset 16 ; CHECK-GI-NEXT: .cfi_offset w30, -16 -; CHECK-GI-NEXT: fmov h1, #2.00000000 ; CHECK-GI-NEXT: fcvt s0, h0 -; CHECK-GI-NEXT: fcvt s1, h1 +; CHECK-GI-NEXT: fmov s1, #2.00000000 ; CHECK-GI-NEXT: bl fmodf ; CHECK-GI-NEXT: fcvt h0, s0 ; CHECK-GI-NEXT: ldr x30, [sp], #16 // 8-byte Folded Reload diff --git a/llvm/test/CodeGen/AArch64/pr58431.ll b/llvm/test/CodeGen/AArch64/pr58431.ll index 467ceb0..a373004 100644 --- a/llvm/test/CodeGen/AArch64/pr58431.ll +++ b/llvm/test/CodeGen/AArch64/pr58431.ll @@ -9,7 +9,7 @@ define i32 @f(i64 %0) { ; CHECK-NEXT: mov w10, #10 // =0xa ; CHECK-NEXT: eor x8, x8, #0x8000000000000003 ; CHECK-NEXT: umulh x8, x9, x8 -; CHECK-NEXT: msub x0, x8, x10, x9 +; CHECK-NEXT: umsubl x0, w8, w10, x9 ; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0 ; CHECK-NEXT: ret %2 = trunc i64 %0 to i32 diff --git a/llvm/test/CodeGen/AArch64/sve-asrd.ll b/llvm/test/CodeGen/AArch64/sve-asrd.ll new file mode 100644 index 0000000..66db1a5 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/sve-asrd.ll @@ -0,0 +1,53 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mattr=+sve -combiner-disabled < %s | FileCheck %s + +target triple = "aarch64-unknown-linux-gnu" + +; Ensure we don't try to represent sdiv-by-one using ARSD. +define <16 x i16> @sdiv_by_one_v16i16(<16 x i16> %a) vscale_range(2,2) { +; CHECK-LABEL: sdiv_by_one_v16i16: +; CHECK: // %bb.0: +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: adrp x8, .LCPI0_0 +; CHECK-NEXT: add x8, x8, :lo12:.LCPI0_0 +; CHECK-NEXT: // kill: def $q1 killed $q1 def $z1 +; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0 +; CHECK-NEXT: ld1h { z2.h }, p0/z, [x8] +; CHECK-NEXT: sunpklo z0.s, z0.h +; CHECK-NEXT: sunpklo z1.s, z1.h +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: sunpklo z3.s, z2.h +; CHECK-NEXT: ext z2.b, z2.b, z2.b, #16 +; CHECK-NEXT: sunpklo z2.s, z2.h +; CHECK-NEXT: sdiv z0.s, p0/m, z0.s, z3.s +; CHECK-NEXT: sdiv z1.s, p0/m, z1.s, z2.s +; CHECK-NEXT: ptrue p0.h, vl8 +; CHECK-NEXT: uzp1 z0.h, z0.h, z0.h +; CHECK-NEXT: uzp1 z1.h, z1.h, z1.h +; CHECK-NEXT: splice z0.h, p0, z0.h, z1.h +; CHECK-NEXT: movprfx z1, z0 +; CHECK-NEXT: ext z1.b, z1.b, z0.b, #16 +; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z0 +; CHECK-NEXT: // kill: def $q1 killed $q1 killed $z1 +; CHECK-NEXT: ret + %res = sdiv <16 x i16> %a, splat(i16 1) + ret <16 x i16> %res +} + +; Ensure we don't try to represent sdiv-by-one using ARSD. +define <vscale x 8 x i16> @sdiv_by_one_nxv8i16(<vscale x 8 x i16> %a) { +; CHECK-LABEL: sdiv_by_one_nxv8i16: +; CHECK: // %bb.0: +; CHECK-NEXT: mov z1.h, #1 // =0x1 +; CHECK-NEXT: sunpkhi z2.s, z0.h +; CHECK-NEXT: sunpklo z0.s, z0.h +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: sunpkhi z3.s, z1.h +; CHECK-NEXT: sunpklo z1.s, z1.h +; CHECK-NEXT: sdiv z2.s, p0/m, z2.s, z3.s +; CHECK-NEXT: sdiv z0.s, p0/m, z0.s, z1.s +; CHECK-NEXT: uzp1 z0.h, z0.h, z2.h +; CHECK-NEXT: ret + %res = sdiv <vscale x 8 x i16> %a, splat(i16 1) + ret <vscale x 8 x i16> %res +} diff --git a/llvm/test/CodeGen/AArch64/sve-vector-deinterleave.ll b/llvm/test/CodeGen/AArch64/sve-vector-deinterleave.ll index 139ecaf..67197b3fe 100644 --- a/llvm/test/CodeGen/AArch64/sve-vector-deinterleave.ll +++ b/llvm/test/CodeGen/AArch64/sve-vector-deinterleave.ll @@ -231,6 +231,274 @@ define {<vscale x 2 x i64>, <vscale x 2 x i64>} @vector_deinterleave_nxv2i64_nxv ret {<vscale x 2 x i64>, <vscale x 2 x i64>} %retval } +define {<vscale x 2 x half>, <vscale x 2 x half>, <vscale x 2 x half>} @vector_deinterleave_nxv2f16_nxv6f16(<vscale x 6 x half> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv2f16_nxv6f16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: uunpkhi z1.s, z0.h +; CHECK-NEXT: uunpklo z0.s, z0.h +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: uunpklo z1.d, z1.s +; CHECK-NEXT: uunpkhi z2.d, z0.s +; CHECK-NEXT: uunpklo z0.d, z0.s +; CHECK-NEXT: str z1, [sp, #2, mul vl] +; CHECK-NEXT: str z2, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3d { z0.d - z2.d }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 2 x half>, <vscale x 2 x half>, <vscale x 2 x half>} @llvm.vector.deinterleave3.nxv6f16(<vscale x 6 x half> %vec) + ret {<vscale x 2 x half>, <vscale x 2 x half>, <vscale x 2 x half>} %retval +} + +define {<vscale x 4 x half>, <vscale x 4 x half>, <vscale x 4 x half>} @vector_deinterleave_nxv4f16_nxv12f16(<vscale x 12 x half> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv4f16_nxv12f16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: uunpklo z1.s, z1.h +; CHECK-NEXT: uunpkhi z2.s, z0.h +; CHECK-NEXT: uunpklo z0.s, z0.h +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: str z1, [sp, #2, mul vl] +; CHECK-NEXT: str z2, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3w { z0.s - z2.s }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 4 x half>, <vscale x 4 x half>, <vscale x 4 x half>} @llvm.vector.deinterleave3.nxv12f16(<vscale x 12 x half> %vec) + ret {<vscale x 4 x half>, <vscale x 4 x half>, <vscale x 4 x half>} %retval +} + +define {<vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>} @vector_deinterleave_nxv8f16_nxv24f16(<vscale x 24 x half> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv8f16_nxv24f16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3h { z0.h - z2.h }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>} @llvm.vector.deinterleave3.nxv24f16(<vscale x 24 x half> %vec) + ret {<vscale x 8 x half>, <vscale x 8 x half>, <vscale x 8 x half>} %retval +} + +define {<vscale x 2 x float>, <vscale x 2 x float>, <vscale x 2 x float>} @vector_deinterleave_nxv2f32_nxv6f32(<vscale x 6 x float> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv2f32_nxv6f32: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: uunpklo z1.d, z1.s +; CHECK-NEXT: uunpkhi z2.d, z0.s +; CHECK-NEXT: uunpklo z0.d, z0.s +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: str z1, [sp, #2, mul vl] +; CHECK-NEXT: str z2, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3d { z0.d - z2.d }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 2 x float>, <vscale x 2 x float>, <vscale x 2 x float>} @llvm.vector.deinterleave3.nxv6f32(<vscale x 6 x float> %vec) + ret {<vscale x 2 x float>, <vscale x 2 x float>, <vscale x 2 x float>} %retval +} + +define {<vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>} @vector_deinterleave_nxv4f32_nxv12f32(<vscale x 12 x float> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv4f32_nxv12f32: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3w { z0.s - z2.s }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>} @llvm.vector.deinterleave3.nxv12f32(<vscale x 12 x float> %vec) + ret {<vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>} %retval +} + +define {<vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>} @vector_deinterleave_nxv2f64_nxv6f64(<vscale x 6 x double> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv2f64_nxv6f64: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3d { z0.d - z2.d }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>} @llvm.vector.deinterleave3.nxv6f64(<vscale x 6 x double> %vec) + ret {<vscale x 2 x double>, <vscale x 2 x double>, <vscale x 2 x double>} %retval +} + +define {<vscale x 2 x bfloat>, <vscale x 2 x bfloat>, <vscale x 2 x bfloat>} @vector_deinterleave_nxv2bf16_nxv6bf16(<vscale x 6 x bfloat> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv2bf16_nxv6bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: uunpkhi z1.s, z0.h +; CHECK-NEXT: uunpklo z0.s, z0.h +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: uunpklo z1.d, z1.s +; CHECK-NEXT: uunpkhi z2.d, z0.s +; CHECK-NEXT: uunpklo z0.d, z0.s +; CHECK-NEXT: str z1, [sp, #2, mul vl] +; CHECK-NEXT: str z2, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3d { z0.d - z2.d }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 2 x bfloat>, <vscale x 2 x bfloat>, <vscale x 2 x bfloat>} @llvm.vector.deinterleave3.nxv6bf16(<vscale x 6 x bfloat> %vec) + ret {<vscale x 2 x bfloat>, <vscale x 2 x bfloat>, <vscale x 2 x bfloat>} %retval +} + +define {<vscale x 4 x bfloat>, <vscale x 4 x bfloat>, <vscale x 4 x bfloat>} @vector_deinterleave_nxv4bf16_nxv12bf16(<vscale x 12 x bfloat> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv4bf16_nxv12bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: uunpklo z1.s, z1.h +; CHECK-NEXT: uunpkhi z2.s, z0.h +; CHECK-NEXT: uunpklo z0.s, z0.h +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: str z1, [sp, #2, mul vl] +; CHECK-NEXT: str z2, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3w { z0.s - z2.s }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 4 x bfloat>, <vscale x 4 x bfloat>, <vscale x 4 x bfloat>} @llvm.vector.deinterleave3.nxv12bf16(<vscale x 12 x bfloat> %vec) + ret {<vscale x 4 x bfloat>, <vscale x 4 x bfloat>, <vscale x 4 x bfloat>} %retval +} + +define {<vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>} @vector_deinterleave_nxv8bf16_nxv24bf16(<vscale x 24 x bfloat> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv8bf16_nxv24bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3h { z0.h - z2.h }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>} @llvm.vector.deinterleave3.nxv24bf16(<vscale x 24 x bfloat> %vec) + ret {<vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>} %retval +} + +; Integers + +define {<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>} @vector_deinterleave_nxv16i8_nxv48i8(<vscale x 48 x i8> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv16i8_nxv48i8: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.b +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3b { z0.b - z2.b }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>} @llvm.vector.deinterleave3.nxv48i8(<vscale x 48 x i8> %vec) + ret {<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>} %retval +} + +define {<vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>} @vector_deinterleave_nxv8i16_nxv24i16(<vscale x 24 x i16> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv8i16_nxv24i16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3h { z0.h - z2.h }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>} @llvm.vector.deinterleave3.nxv24i16(<vscale x 24 x i16> %vec) + ret {<vscale x 8 x i16>, <vscale x 8 x i16>, <vscale x 8 x i16>} %retval +} + +define {<vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>} @vector_deinterleave_nxv4i32_nxvv12i32(<vscale x 12 x i32> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv4i32_nxvv12i32: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3w { z0.s - z2.s }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>} @llvm.vector.deinterleave3.nxv12i32(<vscale x 12 x i32> %vec) + ret {<vscale x 4 x i32>, <vscale x 4 x i32>, <vscale x 4 x i32>} %retval +} + +define {<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>} @vector_deinterleave_nxv2i64_nxv6i64(<vscale x 6 x i64> %vec) { +; CHECK-LABEL: vector_deinterleave_nxv2i64_nxv6i64: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: str z2, [sp, #2, mul vl] +; CHECK-NEXT: str z1, [sp, #1, mul vl] +; CHECK-NEXT: str z0, [sp] +; CHECK-NEXT: ld3d { z0.d - z2.d }, p0/z, [sp] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call {<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>} @llvm.vector.deinterleave3.nxv6i64(<vscale x 6 x i64> %vec) + ret {<vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64>} %retval +} + define {<vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>, <vscale x 16 x i8>} @vector_deinterleave_nxv16i8_nxv64i8(<vscale x 64 x i8> %vec) { ; SVE-LABEL: vector_deinterleave_nxv16i8_nxv64i8: ; SVE: // %bb.0: @@ -599,31 +867,3 @@ define {<vscale x 2 x i32>, <vscale x 2 x i32>} @vector_deinterleave_nxv2i32_nxv %retval = call {<vscale x 2 x i32>,<vscale x 2 x i32>} @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32> %vec) ret {<vscale x 2 x i32>, <vscale x 2 x i32>} %retval } - -; Floating declarations -declare {<vscale x 2 x half>,<vscale x 2 x half>} @llvm.vector.deinterleave2.nxv4f16(<vscale x 4 x half>) -declare {<vscale x 4 x half>, <vscale x 4 x half>} @llvm.vector.deinterleave2.nxv8f16(<vscale x 8 x half>) -declare {<vscale x 2 x float>, <vscale x 2 x float>} @llvm.vector.deinterleave2.nxv4f32(<vscale x 4 x float>) -declare {<vscale x 8 x half>, <vscale x 8 x half>} @llvm.vector.deinterleave2.nxv16f16(<vscale x 16 x half>) -declare {<vscale x 4 x float>, <vscale x 4 x float>} @llvm.vector.deinterleave2.nxv8f32(<vscale x 8 x float>) -declare {<vscale x 2 x double>, <vscale x 2 x double>} @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>) - -; Integer declarations -declare {<vscale x 16 x i8>, <vscale x 16 x i8>} @llvm.vector.deinterleave2.nxv32i8(<vscale x 32 x i8>) -declare {<vscale x 8 x i16>, <vscale x 8 x i16>} @llvm.vector.deinterleave2.nxv16i16(<vscale x 16 x i16>) -declare {<vscale x 4 x i32>, <vscale x 4 x i32>} @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32>) -declare {<vscale x 2 x i64>, <vscale x 2 x i64>} @llvm.vector.deinterleave2.nxv4i64(<vscale x 4 x i64>) - -; Predicated declarations -declare {<vscale x 16 x i1>, <vscale x 16 x i1>} @llvm.vector.deinterleave2.nxv32i1(<vscale x 32 x i1>) -declare {<vscale x 8 x i1>, <vscale x 8 x i1>} @llvm.vector.deinterleave2.nxv16i1(<vscale x 16 x i1>) -declare {<vscale x 4 x i1>, <vscale x 4 x i1>} @llvm.vector.deinterleave2.nxv8i1(<vscale x 8 x i1>) -declare {<vscale x 2 x i1>, <vscale x 2 x i1>} @llvm.vector.deinterleave2.nxv4i1(<vscale x 4 x i1>) - -; Illegal size type -declare {<vscale x 4 x i64>, <vscale x 4 x i64>} @llvm.vector.deinterleave2.nxv8i64(<vscale x 8 x i64>) -declare {<vscale x 8 x i64>, <vscale x 8 x i64>} @llvm.vector.deinterleave2.nxv16i64(<vscale x 16 x i64>) - -declare {<vscale x 8 x i8>, <vscale x 8 x i8>} @llvm.vector.deinterleave2.nxv16i8(<vscale x 16 x i8>) -declare {<vscale x 4 x i16>, <vscale x 4 x i16>} @llvm.vector.deinterleave2.nxv8i16(<vscale x 8 x i16>) -declare {<vscale x 2 x i32>, <vscale x 2 x i32>} @llvm.vector.deinterleave2.nxv4i32(<vscale x 4 x i32>) diff --git a/llvm/test/CodeGen/AArch64/sve-vector-interleave.ll b/llvm/test/CodeGen/AArch64/sve-vector-interleave.ll index c7fb2db..49f185c 100644 --- a/llvm/test/CodeGen/AArch64/sve-vector-interleave.ll +++ b/llvm/test/CodeGen/AArch64/sve-vector-interleave.ll @@ -221,6 +221,318 @@ define <vscale x 4 x i64> @interleave2_nxv4i64(<vscale x 2 x i64> %vec0, <vscale ret <vscale x 4 x i64> %retval } +define <vscale x 6 x half> @interleave3_nxv6f16(<vscale x 2 x half> %vec0, <vscale x 2 x half> %vec1, <vscale x 2 x half> %vec2) { +; CHECK-LABEL: interleave3_nxv6f16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3d { z0.d - z2.d }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp, #2, mul vl] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp] +; CHECK-NEXT: uzp1 z0.s, z0.s, z0.s +; CHECK-NEXT: uzp1 z1.s, z2.s, z1.s +; CHECK-NEXT: uzp1 z0.h, z1.h, z0.h +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 6 x half> @llvm.vector.interleave3.nxv6f16(<vscale x 2 x half> %vec0, <vscale x 2 x half> %vec1, <vscale x 2 x half> %vec2) + ret <vscale x 6 x half> %retval +} + +define <vscale x 12 x half> @interleave3_nxv12f16(<vscale x 4 x half> %vec0, <vscale x 4 x half> %vec1, <vscale x 4 x half> %vec2) { +; CHECK-LABEL: interleave3_nxv12f16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-5 +; CHECK-NEXT: .cfi_escape 0x0f, 0x09, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x11, 0x28, 0x1e, 0x22 // sp + 16 + 40 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: addpl x8, sp, #4 +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3w { z0.s - z2.s }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp, #1, mul vl] +; CHECK-NEXT: ldr z1, [sp] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: uzp1 z0.h, z1.h, z0.h +; CHECK-NEXT: st1h { z2.s }, p0, [x8, #7, mul vl] +; CHECK-NEXT: str z0, [sp, #3, mul vl] +; CHECK-NEXT: ldr z1, [sp, #4, mul vl] +; CHECK-NEXT: ldr z0, [sp, #3, mul vl] +; CHECK-NEXT: addvl sp, sp, #5 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 12 x half> @llvm.vector.interleave3.nxv12f16(<vscale x 4 x half> %vec0, <vscale x 4 x half> %vec1, <vscale x 4 x half> %vec2) + ret <vscale x 12 x half> %retval +} + +define <vscale x 24 x half> @interleave3_nxv24f16(<vscale x 8 x half> %vec0, <vscale x 8 x half> %vec1, <vscale x 8 x half> %vec2) { +; CHECK-LABEL: interleave3_nxv24f16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3h { z0.h - z2.h }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 24 x half> @llvm.vector.interleave3.nxv24f16(<vscale x 8 x half> %vec0, <vscale x 8 x half> %vec1, <vscale x 8 x half> %vec2) + ret <vscale x 24 x half> %retval +} + +define <vscale x 6 x float> @interleave3_nxv6f32(<vscale x 2 x float> %vec0, <vscale x 2 x float> %vec1, <vscale x 2 x float> %vec2) { +; CHECK-LABEL: interleave3_nxv6f32: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-5 +; CHECK-NEXT: .cfi_escape 0x0f, 0x09, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x11, 0x28, 0x1e, 0x22 // sp + 16 + 40 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: addpl x8, sp, #4 +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3d { z0.d - z2.d }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp, #1, mul vl] +; CHECK-NEXT: ldr z1, [sp] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: uzp1 z0.s, z1.s, z0.s +; CHECK-NEXT: st1w { z2.d }, p0, [x8, #7, mul vl] +; CHECK-NEXT: str z0, [sp, #3, mul vl] +; CHECK-NEXT: ldr z1, [sp, #4, mul vl] +; CHECK-NEXT: ldr z0, [sp, #3, mul vl] +; CHECK-NEXT: addvl sp, sp, #5 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 6 x float> @llvm.vector.interleave3.nxv6f32(<vscale x 2 x float> %vec0, <vscale x 2 x float> %vec1, <vscale x 2 x float> %vec2) + ret <vscale x 6 x float> %retval +} + +define <vscale x 12 x float> @interleave3_nxv12f32(<vscale x 4 x float> %vec0, <vscale x 4 x float> %vec1, <vscale x 4 x float> %vec2) { +; CHECK-LABEL: interleave3_nxv12f32: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3w { z0.s - z2.s }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 12 x float> @llvm.vector.interleave3.nxv12f32(<vscale x 4 x float> %vec0, <vscale x 4 x float> %vec1, <vscale x 4 x float> %vec2) + ret <vscale x 12 x float> %retval +} + +define <vscale x 6 x double> @interleave3_nxv6f64(<vscale x 2 x double> %vec0, <vscale x 2 x double> %vec1, <vscale x 2 x double> %vec2) { +; CHECK-LABEL: interleave3_nxv6f64: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3d { z0.d - z2.d }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 6 x double>@llvm.vector.interleave3.nxv6f64(<vscale x 2 x double> %vec0, <vscale x 2 x double> %vec1, <vscale x 2 x double> %vec2) + ret <vscale x 6 x double> %retval +} + +define <vscale x 6 x bfloat> @interleave3_nxv6bf16(<vscale x 2 x bfloat> %vec0, <vscale x 2 x bfloat> %vec1, <vscale x 2 x bfloat> %vec2) { +; CHECK-LABEL: interleave3_nxv6bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3d { z0.d - z2.d }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp, #2, mul vl] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp] +; CHECK-NEXT: uzp1 z0.s, z0.s, z0.s +; CHECK-NEXT: uzp1 z1.s, z2.s, z1.s +; CHECK-NEXT: uzp1 z0.h, z1.h, z0.h +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 6 x bfloat> @llvm.vector.interleave3.nxv6bf16(<vscale x 2 x bfloat> %vec0, <vscale x 2 x bfloat> %vec1, <vscale x 2 x bfloat> %vec2) + ret <vscale x 6 x bfloat> %retval +} + +define <vscale x 12 x bfloat> @interleave3_nxv12bf16(<vscale x 4 x bfloat> %vec0, <vscale x 4 x bfloat> %vec1, <vscale x 4 x bfloat> %vec2) { +; CHECK-LABEL: interleave3_nxv12bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-5 +; CHECK-NEXT: .cfi_escape 0x0f, 0x09, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x11, 0x28, 0x1e, 0x22 // sp + 16 + 40 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: addpl x8, sp, #4 +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3w { z0.s - z2.s }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp, #1, mul vl] +; CHECK-NEXT: ldr z1, [sp] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: uzp1 z0.h, z1.h, z0.h +; CHECK-NEXT: st1h { z2.s }, p0, [x8, #7, mul vl] +; CHECK-NEXT: str z0, [sp, #3, mul vl] +; CHECK-NEXT: ldr z1, [sp, #4, mul vl] +; CHECK-NEXT: ldr z0, [sp, #3, mul vl] +; CHECK-NEXT: addvl sp, sp, #5 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 12 x bfloat> @llvm.vector.interleave3.nxv12bf16(<vscale x 4 x bfloat> %vec0, <vscale x 4 x bfloat> %vec1, <vscale x 4 x bfloat> %vec2) + ret <vscale x 12 x bfloat> %retval +} + +define <vscale x 24 x bfloat> @interleave3_nxv24bf16(<vscale x 8 x bfloat> %vec0, <vscale x 8 x bfloat> %vec1, <vscale x 8 x bfloat> %vec2) { +; CHECK-LABEL: interleave3_nxv24bf16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3h { z0.h - z2.h }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 24 x bfloat> @llvm.vector.interleave3.nxv24bf16(<vscale x 8 x bfloat> %vec0, <vscale x 8 x bfloat> %vec1, <vscale x 8 x bfloat> %vec2) + ret <vscale x 24 x bfloat> %retval +} + +; Integers + +define <vscale x 48 x i8> @interleave3_nxv48i8(<vscale x 16 x i8> %vec0, <vscale x 16 x i8> %vec1, <vscale x 16 x i8> %vec2) { +; CHECK-LABEL: interleave3_nxv48i8: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.b +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3b { z0.b - z2.b }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 48 x i8> @llvm.vector.interleave3.nxv48i8(<vscale x 16 x i8> %vec0, <vscale x 16 x i8> %vec1, <vscale x 16 x i8> %vec2) + ret <vscale x 48 x i8> %retval +} + +define <vscale x 24 x i16> @interleave3_nxv24i16(<vscale x 8 x i16> %vec0, <vscale x 8 x i16> %vec1, <vscale x 8 x i16> %vec2) { +; CHECK-LABEL: interleave3_nxv24i16: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.h +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3h { z0.h - z2.h }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 24 x i16> @llvm.vector.interleave3.nxv24i16(<vscale x 8 x i16> %vec0, <vscale x 8 x i16> %vec1, <vscale x 8 x i16> %vec2) + ret <vscale x 24 x i16> %retval +} + +define <vscale x 12 x i32> @interleave3_nxv12i32(<vscale x 4 x i32> %vec0, <vscale x 4 x i32> %vec1, <vscale x 4 x i32> %vec2) { +; CHECK-LABEL: interleave3_nxv12i32: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.s +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3w { z0.s - z2.s }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 12 x i32> @llvm.vector.interleave3.nxv12i32(<vscale x 4 x i32> %vec0, <vscale x 4 x i32> %vec1, <vscale x 4 x i32> %vec2) + ret <vscale x 12 x i32> %retval +} + +define <vscale x 6 x i64> @interleave3_nxv6i64(<vscale x 2 x i64> %vec0, <vscale x 2 x i64> %vec1, <vscale x 2 x i64> %vec2) { +; CHECK-LABEL: interleave3_nxv6i64: +; CHECK: // %bb.0: +; CHECK-NEXT: str x29, [sp, #-16]! // 8-byte Folded Spill +; CHECK-NEXT: addvl sp, sp, #-3 +; CHECK-NEXT: .cfi_escape 0x0f, 0x08, 0x8f, 0x10, 0x92, 0x2e, 0x00, 0x48, 0x1e, 0x22 // sp + 16 + 24 * VG +; CHECK-NEXT: .cfi_offset w29, -16 +; CHECK-NEXT: // kill: def $z2 killed $z2 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: ptrue p0.d +; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1_z2 def $z0_z1_z2 +; CHECK-NEXT: st3d { z0.d - z2.d }, p0, [sp] +; CHECK-NEXT: ldr z0, [sp] +; CHECK-NEXT: ldr z1, [sp, #1, mul vl] +; CHECK-NEXT: ldr z2, [sp, #2, mul vl] +; CHECK-NEXT: addvl sp, sp, #3 +; CHECK-NEXT: ldr x29, [sp], #16 // 8-byte Folded Reload +; CHECK-NEXT: ret + %retval = call <vscale x 6 x i64> @llvm.vector.interleave3.nxv6i64(<vscale x 2 x i64> %vec0, <vscale x 2 x i64> %vec1, <vscale x 2 x i64> %vec2) + ret <vscale x 6 x i64> %retval +} + define <vscale x 64 x i8> @interleave4_nxv16i8(<vscale x 16 x i8> %vec0, <vscale x 16 x i8> %vec1, <vscale x 16 x i8> %vec2, <vscale x 16 x i8> %vec3) { ; SVE-LABEL: interleave4_nxv16i8: ; SVE: // %bb.0: diff --git a/llvm/test/CodeGen/AArch64/vecreduce-fadd-strict.ll b/llvm/test/CodeGen/AArch64/vecreduce-fadd-strict.ll index be07978..8e0328e 100644 --- a/llvm/test/CodeGen/AArch64/vecreduce-fadd-strict.ll +++ b/llvm/test/CodeGen/AArch64/vecreduce-fadd-strict.ll @@ -38,17 +38,11 @@ define half @add_v2HalfH(<2 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: add_v2HalfH: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI1_0 ; CHECK-GI-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h0 -; CHECK-GI-NOFP16-NEXT: ldr h1, [x8, :lo12:.LCPI1_0] -; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 +; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 -; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s0, s1, s0 +; CHECK-GI-NOFP16-NEXT: fadd s0, s0, s1 ; CHECK-GI-NOFP16-NEXT: fcvt h0, s0 ; CHECK-GI-NOFP16-NEXT: ret ; @@ -88,19 +82,13 @@ define half @add_v3HalfH(<3 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: add_v3HalfH: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI2_0 ; CHECK-GI-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s2, h0 -; CHECK-GI-NOFP16-NEXT: ldr h1, [x8, :lo12:.LCPI2_0] -; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 -; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] ; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[2] -; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 -; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 +; CHECK-GI-NOFP16-NEXT: fcvt s0, h0 +; CHECK-GI-NOFP16-NEXT: fadd s1, s2, s1 ; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 ; CHECK-GI-NOFP16-NEXT: fadd s0, s1, s0 @@ -152,17 +140,11 @@ define half @add_HalfH(<4 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: add_HalfH: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI3_0 ; CHECK-GI-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s2, h0 -; CHECK-GI-NOFP16-NEXT: ldr h1, [x8, :lo12:.LCPI3_0] -; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 -; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 +; CHECK-GI-NOFP16-NEXT: fadd s1, s2, s1 ; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[2] ; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[3] ; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 @@ -250,16 +232,10 @@ define half @add_H(<8 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: add_H: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI4_0 +; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s2, h0 -; CHECK-GI-NOFP16-NEXT: ldr h1, [x8, :lo12:.LCPI4_0] -; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 -; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fadd s1, s1, s2 +; CHECK-GI-NOFP16-NEXT: fadd s1, s2, s1 ; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[2] ; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 ; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 @@ -448,16 +424,10 @@ define half @add_2H(<16 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: add_2H: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI7_0 +; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s3, h0 -; CHECK-GI-NOFP16-NEXT: ldr h2, [x8, :lo12:.LCPI7_0] -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 -; CHECK-GI-NOFP16-NEXT: fadd s2, s2, s3 -; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt h2, s2 -; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 ; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 -; CHECK-GI-NOFP16-NEXT: fadd s2, s2, s3 +; CHECK-GI-NOFP16-NEXT: fadd s2, s3, s2 ; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2] ; CHECK-GI-NOFP16-NEXT: fcvt h2, s2 ; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 diff --git a/llvm/test/CodeGen/AArch64/vecreduce-fmul-strict.ll b/llvm/test/CodeGen/AArch64/vecreduce-fmul-strict.ll index c10d6e9..716401e 100644 --- a/llvm/test/CodeGen/AArch64/vecreduce-fmul-strict.ll +++ b/llvm/test/CodeGen/AArch64/vecreduce-fmul-strict.ll @@ -52,17 +52,11 @@ define half @mul_HalfH(<4 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: mul_HalfH: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI1_0 ; CHECK-GI-NOFP16-NEXT: // kill: def $d0 killed $d0 def $q0 +; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s2, h0 -; CHECK-GI-NOFP16-NEXT: ldr h1, [x8, :lo12:.LCPI1_0] ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fmul s1, s1, s2 -; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 -; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fmul s1, s1, s2 +; CHECK-GI-NOFP16-NEXT: fmul s1, s2, s1 ; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[2] ; CHECK-GI-NOFP16-NEXT: mov h0, v0.h[3] ; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 @@ -144,16 +138,10 @@ define half @mul_H(<8 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: mul_H: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI2_0 +; CHECK-GI-NOFP16-NEXT: mov h1, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s2, h0 -; CHECK-GI-NOFP16-NEXT: ldr h1, [x8, :lo12:.LCPI2_0] -; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fmul s1, s1, s2 -; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 ; CHECK-GI-NOFP16-NEXT: fcvt s1, h1 -; CHECK-GI-NOFP16-NEXT: fmul s1, s1, s2 +; CHECK-GI-NOFP16-NEXT: fmul s1, s2, s1 ; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[2] ; CHECK-GI-NOFP16-NEXT: fcvt h1, s1 ; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 @@ -321,16 +309,10 @@ define half @mul_2H(<16 x half> %bin.rdx) { ; ; CHECK-GI-NOFP16-LABEL: mul_2H: ; CHECK-GI-NOFP16: // %bb.0: -; CHECK-GI-NOFP16-NEXT: adrp x8, .LCPI5_0 +; CHECK-GI-NOFP16-NEXT: mov h2, v0.h[1] ; CHECK-GI-NOFP16-NEXT: fcvt s3, h0 -; CHECK-GI-NOFP16-NEXT: ldr h2, [x8, :lo12:.LCPI5_0] ; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 -; CHECK-GI-NOFP16-NEXT: fmul s2, s2, s3 -; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[1] -; CHECK-GI-NOFP16-NEXT: fcvt h2, s2 -; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 -; CHECK-GI-NOFP16-NEXT: fcvt s2, h2 -; CHECK-GI-NOFP16-NEXT: fmul s2, s2, s3 +; CHECK-GI-NOFP16-NEXT: fmul s2, s3, s2 ; CHECK-GI-NOFP16-NEXT: mov h3, v0.h[2] ; CHECK-GI-NOFP16-NEXT: fcvt h2, s2 ; CHECK-GI-NOFP16-NEXT: fcvt s3, h3 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll index 1aee6ab..1b879a6 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f16.ll @@ -403,40 +403,38 @@ define half @v_neg_rcp_f16(half %x) { ; GFX6-IEEE-LABEL: v_neg_rcp_f16: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_neg_rcp_f16: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] @@ -460,40 +458,38 @@ define half @v_rcp_f16(half %x) { ; GFX6-IEEE-LABEL: v_rcp_f16: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_rcp_f16: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] @@ -517,40 +513,38 @@ define half @v_rcp_f16_arcp(half %x) { ; GFX6-IEEE-LABEL: v_rcp_f16_arcp: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_rcp_f16_arcp: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] @@ -575,9 +569,7 @@ define half @v_rcp_f16_arcp_afn(half %x) { ; GFX6: ; %bb.0: ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-NEXT: v_rcp_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_f32_e32 v0, v1, v0 ; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-NEXT: s_setpc_b64 s[30:31] ; @@ -600,40 +592,38 @@ define half @v_rcp_f16_ulp25(half %x) { ; GFX6-IEEE-LABEL: v_rcp_f16_ulp25: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_rcp_f16_ulp25: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] @@ -1454,70 +1444,67 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX6-IEEE-LABEL: v_rcp_v2f16: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v1, v1, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v1, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_rcp_v2f16: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -1526,30 +1513,27 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX8-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v6, -v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, 1.0, v1 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v4, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -1561,26 +1545,23 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX8-FLUSH-LABEL: v_rcp_v2f16: ; GFX8-FLUSH: ; %bb.0: ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v8, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v9, v8, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, -v1, v4, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v6, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -1594,30 +1575,27 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX9-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v6, -v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, 1.0, v1 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v4, v7, v5 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -1628,26 +1606,24 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX9-FLUSH-LABEL: v_rcp_v2f16: ; GFX9-FLUSH: ; %bb.0: ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v1 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v1 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v1, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v3, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v5, v5, v1, v1 +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v3, v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v7, v1 +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -1660,30 +1636,27 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v6, -v2, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, 1.0, v2 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 @@ -1696,24 +1669,21 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, -v2, v4, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 @@ -1726,27 +1696,25 @@ define <2 x half> @v_rcp_v2f16(<2 x half> %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX11-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX11-NEXT: v_rcp_f32_e32 v2, v2 ; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 -; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 +; GFX11-NEXT: v_fma_mix_f32 v5, -v0, v2, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v3, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v5, v5, v2, v2 +; GFX11-NEXT: v_fma_f32 v6, v6, v3, v3 +; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v4, -v0, v6, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_dual_mul_f32 v2, v7, v2 :: v_dual_mul_f32 v3, v4, v3 +; GFX11-NEXT: v_and_b32_e32 v2, 0xff800000, v2 +; GFX11-NEXT: v_dual_add_f32 v2, v2, v5 :: v_dual_and_b32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 +; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] %fdiv = fdiv <2 x half> <half 1.0, half 1.0>, %x @@ -1757,70 +1725,67 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX6-IEEE-LABEL: v_neg_rcp_v2f16: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v1, v1, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v1, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, -1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, -1.0, v1, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_neg_rcp_v2f16: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, -1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v1, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -1829,30 +1794,27 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX8-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, -1.0, v1 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v4, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -1864,26 +1826,23 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX8-FLUSH-LABEL: v_neg_rcp_v2f16: ; GFX8-FLUSH: ; %bb.0: ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v8, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v9, v8, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v1, v4, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v6, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -1897,30 +1856,27 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX9-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX9-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, -1.0, v1 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX9-IEEE-NEXT: v_sub_f32_e32 v4, v7, v5 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -1931,26 +1887,24 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX9-FLUSH-LABEL: v_neg_rcp_v2f16: ; GFX9-FLUSH: ; %bb.0: ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v1 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v1 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, v0, v1, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, v0, v3, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v5, v5, v1, -v1 +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v3, -v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v7, v1 +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -1963,30 +1917,27 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, -1.0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v2, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_sub_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, -1.0, v2 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -1999,24 +1950,21 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, -1.0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v2, v4, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -2029,27 +1977,25 @@ define <2 x half> @v_neg_rcp_v2f16(<2 x half> %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX11-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX11-NEXT: v_rcp_f32_e32 v2, v2 ; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 -; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 +; GFX11-NEXT: v_fma_mix_f32 v5, v0, v2, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v6, v0, v3, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v5, v5, v2, -v2 +; GFX11-NEXT: v_fma_f32 v6, v6, v3, -v3 +; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v4, -v0, v6, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_dual_mul_f32 v2, v7, v2 :: v_dual_mul_f32 v3, v4, v3 +; GFX11-NEXT: v_and_b32_e32 v2, 0xff800000, v2 +; GFX11-NEXT: v_dual_add_f32 v2, v2, v5 :: v_dual_and_b32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 +; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] %fdiv = fdiv <2 x half> <half -1.0, half -1.0>, %x @@ -2064,33 +2010,32 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX6-IEEE-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; GFX6-IEEE-NEXT: v_or_b32_e32 v0, v1, v0 ; GFX6-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX6-IEEE-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v2, v2, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v1, v2, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v5, v0 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v0, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v0, v2, v1 -; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v5, v5, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 ; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v5, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v2, v6, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v6, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v5, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v0 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v0, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v0, v1, 1.0 +; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v4, v4, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v4, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v5, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v1, v5, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v2, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v5, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v1, v4, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -2101,39 +2046,37 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX6-FLUSH-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; GFX6-FLUSH-NEXT: v_or_b32_e32 v0, v1, v0 ; GFX6-FLUSH-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX6-FLUSH-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v2, v2, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v1, v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, v0 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v2, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v5, v5, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v3, v3, 1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, v4, v5, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v3, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v1, v2, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, v6, v2, v2 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v3, v2 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v1, v6, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v2, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v6, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v5, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v1, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v2, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v1, v5, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v1, v3, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -2143,30 +2086,27 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX8-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v6, -v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, 1.0, v1 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v4, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -2179,26 +2119,23 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX8-FLUSH: ; %bb.0: ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-FLUSH-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v8, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v9, v8, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, -v1, v4, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v6, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -2213,30 +2150,27 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX9-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v6, -v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, 1.0, v1 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v4, v7, v5 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -2248,26 +2182,24 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX9-FLUSH: ; %bb.0: ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v1 +; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v3 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, 1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v5, 1.0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v4, v4 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v2 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v1, v6, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v5, v4 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v7, v2 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v7, v4 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v6, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v2, v5 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -|v0|, v4, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v2, v2 +; GFX9-FLUSH-NEXT: v_mad_f32 v7, v7, v4, v4 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v6, v5 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v7, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v8, v2 ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v0, v0, v4 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0 ; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v5 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v7 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0 @@ -2279,32 +2211,29 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX10-IEEE: ; %bb.0: ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v6, -v2, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, 1.0, v2 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 @@ -2316,26 +2245,23 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX10-FLUSH: ; %bb.0: ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, -v2, v4, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 @@ -2346,30 +2272,30 @@ define <2 x half> @v_rcp_v2f16_fabs(<2 x half> %x) { ; GFX11-LABEL: v_rcp_v2f16_fabs: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_cvt_f32_f16_e32 v5, 1.0 ; GFX11-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 -; GFX11-NEXT: v_rcp_f32_e32 v3, v3 -; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v6, v5, v3 ; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, 1.0 op_sel_hi:[1,0,1] ; GFX11-NEXT: v_cvt_f32_f16_e32 v4, v2 -; GFX11-NEXT: v_fmac_f32_e32 v6, v7, v3 ; GFX11-NEXT: v_rcp_f32_e32 v4, v4 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, 1.0 op_sel_hi:[1,0,1] +; GFX11-NEXT: v_mov_b32_e32 v5, 1.0 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v5, v4 -; GFX11-NEXT: v_fma_mix_f32 v8, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v8, v4 -; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v5, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v3, v7, v3 :: v_dual_mul_f32 v0, v0, v4 +; GFX11-NEXT: v_fma_mix_f32 v7, -|v0|, v4, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v7, v7, v4, v4 +; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v7, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_mul_f32_e32 v0, v0, v4 ; GFX11-NEXT: v_and_b32_e32 v0, 0xff800000, v0 -; GFX11-NEXT: v_dual_add_f32 v0, v0, v5 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX11-NEXT: v_add_f32_e32 v0, v0, v7 +; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v0, v2, 1.0 +; GFX11-NEXT: s_waitcnt_depctr 0xfff +; GFX11-NEXT: v_fma_mix_f32 v6, -v1, v3, v5 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v6, v6, v3, v3 +; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v5 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_mul_f32_e32 v3, v8, v3 +; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v1, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] @@ -2386,33 +2312,32 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX6-IEEE-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; GFX6-IEEE-NEXT: v_or_b32_e32 v0, v1, v0 ; GFX6-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX6-IEEE-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v2, v2, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v1, v2, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v5, v0 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v0, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v0, v2, v1 -; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v5, v5, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, -1.0 ; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v5, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v2, v6, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v6, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v5, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, -1.0, v1, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v4, v0 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v0, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v0, v1, -1.0 +; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v4, v4, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v4, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v5, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v1, v5, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v2, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v5, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v1, v4, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -2423,39 +2348,37 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX6-FLUSH-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; GFX6-FLUSH-NEXT: v_or_b32_e32 v0, v1, v0 ; GFX6-FLUSH-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX6-FLUSH-NEXT: v_lshrrev_b32_e32 v0, 16, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v2, v2, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v1, v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v1, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, v0 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v2, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v1, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v5, v5, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v3, v3, -1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, v4, v5, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v3, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v1, v2, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, v6, v2, v2 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v3, v2 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v1, v6, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v2, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v6, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v5, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v1, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v2, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v1, v5, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v1, v3, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -2465,30 +2388,27 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX8-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, -1.0, v1 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v4, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -2501,26 +2421,23 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX8-FLUSH: ; %bb.0: ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-FLUSH-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v8, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v9, v8, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v1, v4, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v6, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -2535,30 +2452,27 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX9-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX9-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, -1.0, v1 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX9-IEEE-NEXT: v_sub_f32_e32 v4, v7, v5 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, -1.0 @@ -2570,26 +2484,24 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX9-FLUSH: ; %bb.0: ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v3, 16, v1 +; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, v3 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v5, -1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v5, -1.0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v4, v4 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v2 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v1, v6, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v5, v4 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v6, v7, v2 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v7, v4 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v6, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, v1, v2, v5 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, |v0|, v4, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v2, -v2 +; GFX9-FLUSH-NEXT: v_mad_f32 v7, v7, v4, -v4 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v8, -v1, v6, v5 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v0, -|v0|, v7, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v8, v2 ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v0, v0, v4 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v0, 0xff800000, v0 ; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v5 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v0, v0, v7 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0 @@ -2601,32 +2513,29 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX10-IEEE: ; %bb.0: ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, -1.0 ; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v2, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_sub_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, -1.0, v2 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -2638,26 +2547,23 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX10-FLUSH: ; %bb.0: ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_and_b32_e32 v0, 0x7fff7fff, v0 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, -1.0 ; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v2, v4, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -2668,30 +2574,30 @@ define <2 x half> @v_neg_rcp_v2f16_fabs(<2 x half> %x) { ; GFX11-LABEL: v_neg_rcp_v2f16_fabs: ; GFX11: ; %bb.0: ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NEXT: v_cvt_f32_f16_e32 v5, -1.0 ; GFX11-NEXT: v_and_b32_e32 v1, 0x7fff7fff, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 -; GFX11-NEXT: v_rcp_f32_e32 v3, v3 -; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v6, v5, v3 ; GFX11-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, -1.0 op_sel_hi:[1,0,1] ; GFX11-NEXT: v_cvt_f32_f16_e32 v4, v2 -; GFX11-NEXT: v_fmac_f32_e32 v6, v7, v3 ; GFX11-NEXT: v_rcp_f32_e32 v4, v4 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v6, -1.0 op_sel_hi:[1,0,1] +; GFX11-NEXT: v_mov_b32_e32 v5, -1.0 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v5, v4 -; GFX11-NEXT: v_fma_mix_f32 v8, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v8, v4 -; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v5, -1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v3, v7, v3 :: v_dual_mul_f32 v0, v0, v4 +; GFX11-NEXT: v_fma_mix_f32 v7, |v0|, v4, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v7, v7, v4, -v4 +; GFX11-NEXT: v_fma_mix_f32 v0, -|v0|, v7, v5 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_mul_f32_e32 v0, v0, v4 ; GFX11-NEXT: v_and_b32_e32 v0, 0xff800000, v0 -; GFX11-NEXT: v_dual_add_f32 v0, v0, v5 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX11-NEXT: v_add_f32_e32 v0, v0, v7 +; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX11-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 +; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v0, v2, -1.0 +; GFX11-NEXT: s_waitcnt_depctr 0xfff +; GFX11-NEXT: v_fma_mix_f32 v6, v1, v3, v5 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v6, v6, v3, -v3 +; GFX11-NEXT: v_fma_mix_f32 v8, -v1, v6, v5 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_mul_f32_e32 v3, v8, v3 +; GFX11-NEXT: v_and_b32_e32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v1, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] @@ -2704,70 +2610,67 @@ define <2 x half> @v_rcp_v2f16_arcp(<2 x half> %x) { ; GFX6-IEEE-LABEL: v_rcp_v2f16_arcp: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v1, v1, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v1, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_rcp_v2f16_arcp: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -2814,11 +2717,8 @@ define <2 x half> @v_rcp_v2f16_arcp_afn(<2 x half> %x) { ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-NEXT: v_rcp_f32_e32 v0, v0 ; GFX6-NEXT: v_rcp_f32_e32 v1, v1 -; GFX6-NEXT: v_mul_f32_e32 v0, v2, v0 -; GFX6-NEXT: v_mul_f32_e32 v1, v2, v1 ; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-NEXT: s_setpc_b64 s[30:31] @@ -2864,70 +2764,67 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX6-IEEE-LABEL: v_rcp_v2f16_ulp25: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v1, v1, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v2, v1, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; ; GFX6-FLUSH-LABEL: v_rcp_v2f16_ulp25: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -2936,30 +2833,27 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX8-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v6, -v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, 1.0, v1 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v4, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -2971,26 +2865,23 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX8-FLUSH-LABEL: v_rcp_v2f16_ulp25: ; GFX8-FLUSH: ; %bb.0: ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v1 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v8, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v9, v8, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v7, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v1 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, -v1, v4, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v1, -v1, v6, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -3004,30 +2895,27 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX9-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_lshrrev_b32_e32 v2, 16, v0 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v1 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v8, -v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v7 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v10, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v5, v5, v6 -; GFX9-IEEE-NEXT: v_add_f32_e32 v5, v5, v9 -; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v1 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v6, -v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v1, -v1, v6 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, 1.0, v1 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v1, v1, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v4, v7, v5 +; GFX9-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v4 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v1, 0xff800000, v1 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 +; GFX9-IEEE-NEXT: v_add_f32_e32 v1, v1, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -3038,26 +2926,24 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX9-FLUSH-LABEL: v_rcp_v2f16_ulp25: ; GFX9-FLUSH: ; %bb.0: ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_lshrrev_b32_e32 v2, 16, v0 +; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v2 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v1, v1 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v1 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v1 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v1, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v3, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v5, v5, v1, v1 +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v3, v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v1, v7, v1 +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v1, 0xff800000, v1 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v5, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v1, v1, v5 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v1, v0, 1.0 @@ -3070,30 +2956,27 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v9, v9, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v10, v10, v6 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v9, v9, v4 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v10, v10, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v6, -v2, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX10-IEEE-NEXT: v_add_f32_e32 v7, v7, v5 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX10-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, 1.0, v2 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 @@ -3106,24 +2989,21 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, -v2, v4, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 @@ -3136,27 +3016,25 @@ define <2 x half> @v_rcp_v2f16_ulp25(<2 x half> %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX11-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX11-NEXT: v_rcp_f32_e32 v2, v2 ; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel:[1,0,0] op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 -; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 +; GFX11-NEXT: v_fma_mix_f32 v5, -v0, v2, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v3, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v5, v5, v2, v2 +; GFX11-NEXT: v_fma_f32 v6, v6, v3, v3 +; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v4, -v0, v6, v4 op_sel:[1,0,0] op_sel_hi:[1,0,0] +; GFX11-NEXT: v_dual_mul_f32 v2, v7, v2 :: v_dual_mul_f32 v3, v4, v3 +; GFX11-NEXT: v_and_b32_e32 v2, 0xff800000, v2 +; GFX11-NEXT: v_dual_add_f32 v2, v2, v5 :: v_dual_and_b32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 +; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] %fdiv = fdiv <2 x half> <half 1.0, half 1.0>, %x @@ -4033,40 +3911,38 @@ define amdgpu_ps i32 @s_fdiv_v2f16(i32 inreg %a.arg, i32 inreg %b.arg) { define amdgpu_ps i16 @s_rcp_f16(i16 inreg %a.arg) { ; GFX6-IEEE-LABEL: s_rcp_f16: ; GFX6-IEEE: ; %bb.0: -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, 1.0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, s0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[0:1], v1, v1, v0 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v0, v1, v0 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v1, v0 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, s0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[0:1], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_readfirstlane_b32 s0, v0 ; GFX6-IEEE-NEXT: ; return to shader part epilog ; ; GFX6-FLUSH-LABEL: s_rcp_f16: ; GFX6-FLUSH: ; %bb.0: -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, 1.0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[0:1], v1, v1, v0 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v0, v1, v0 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s0 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[0:1], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v1, v0 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_readfirstlane_b32 s0, v0 @@ -4099,40 +3975,38 @@ define amdgpu_ps i16 @s_rcp_f16(i16 inreg %a.arg) { define amdgpu_ps i16 @s_neg_rcp_f16(i16 inreg %a.arg) { ; GFX6-IEEE-LABEL: s_neg_rcp_f16: ; GFX6-IEEE: ; %bb.0: -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, -1.0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, s0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[0:1], v1, v1, v0 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v0, v1, v0 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v1, v0 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, s0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[0:1], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_readfirstlane_b32 s0, v0 ; GFX6-IEEE-NEXT: ; return to shader part epilog ; ; GFX6-FLUSH-LABEL: s_neg_rcp_f16: ; GFX6-FLUSH: ; %bb.0: -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, -1.0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[0:1], v1, v1, v0 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v0, v1, v0 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s0 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[0:1], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v1, v0 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_readfirstlane_b32 s0, v0 @@ -4166,21 +4040,20 @@ define amdgpu_ps i16 @s_rsq_f16(i16 inreg %a.arg) { ; GFX6-IEEE-LABEL: s_rsq_f16: ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, s0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[0:1], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[0:1], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_readfirstlane_b32 s0, v0 ; GFX6-IEEE-NEXT: ; return to shader part epilog @@ -4188,24 +4061,23 @@ define amdgpu_ps i16 @s_rsq_f16(i16 inreg %a.arg) { ; GFX6-FLUSH-LABEL: s_rsq_f16: ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[0:1], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[0:1], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_readfirstlane_b32 s0, v0 ; GFX6-FLUSH-NEXT: ; return to shader part epilog @@ -4241,36 +4113,35 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, s0 ; GFX6-IEEE-NEXT: s_lshr_b32 s0, s0, 16 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, s0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[0:1], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v5, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v6, s[0:1], v1, v1, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v9, -v3, v5, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v9, v5, v5 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v9, v4, v5 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v8, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v10, -v3, v9, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v9, v10, v5, v9 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v9, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v5, v9 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v6, v8, 1.0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v7, s[0:1], v2, v1, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v3, v8, v8 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v7, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v6, v4, v7 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v3, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v6, v4, v7 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[0:1], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v5, s[0:1], v1, v1, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v8, -v2, v4, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v8, v4, v4 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v6, v5 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v9, v3, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v10, -v2, v9, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v9, v10, v4, v9 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v9, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v8, -v5, v6, 1.0 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v4, v9 +; GFX6-IEEE-NEXT: v_div_scale_f32 v7, s[0:1], -1.0, v1, -1.0 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v8, v6, v6 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v3, v7, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v5, v3, v7 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v4, v2, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v5, v3, v7 ; GFX6-IEEE-NEXT: s_mov_b64 vcc, s[0:1] -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v5, v3, v4 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v4, v2, v3 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_lshlrev_b32_e32 v1, 16, v1 @@ -4283,42 +4154,40 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, s0 ; GFX6-FLUSH-NEXT: s_lshr_b32 s0, s0, 16 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, s0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[0:1], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[0:1], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[0:1], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[0:1], v1, v1, -1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v1, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: v_lshlrev_b32_e32 v1, 16, v1 ; GFX6-FLUSH-NEXT: v_or_b32_e32 v0, v0, v1 @@ -4330,31 +4199,28 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX8-IEEE-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX8-IEEE-NEXT: s_lshr_b32 s0, s0, 16 ; GFX8-IEEE-NEXT: v_sqrt_f16_e32 v1, s0 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v8, -v2, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v8, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v8, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v9 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v8, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v10, v10, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v10, v6 -; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v2, v2, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v5, v8, v9 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v2, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 +; GFX8-IEEE-NEXT: v_add_f32_e32 v2, -1.0, v2 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX8-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 @@ -4369,25 +4235,22 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX8-FLUSH-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX8-FLUSH-NEXT: s_lshr_b32 s0, s0, 16 ; GFX8-FLUSH-NEXT: v_sqrt_f16_e32 v1, s0 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v2 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v2, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v8, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v8, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v9, v8, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v9, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v5 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v2, v4, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, -1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v9 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 @@ -4402,25 +4265,22 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX9-IEEE-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX9-IEEE-NEXT: s_lshr_b32 s0, s0, 16 ; GFX9-IEEE-NEXT: v_sqrt_f16_e32 v1, s0 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_fma_f32 v8, -v2, v7, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX9-IEEE-NEXT: v_fma_f32 v7, v8, v5, v7 -; GFX9-IEEE-NEXT: v_fma_f32 v8, -v3, v9, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v8, v8, v6, v9 -; GFX9-IEEE-NEXT: v_fma_f32 v2, -v2, v7, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v3, -v3, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v2, v2, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_fma_f32 v6, v2, v4, -1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v6, v6, v4, -v4 +; GFX9-IEEE-NEXT: v_fma_f32 v7, v3, v5, -1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v7, v7, v5, -v5 +; GFX9-IEEE-NEXT: v_fma_f32 v2, -v2, v6, -1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v3, -v3, v7, -1.0 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX9-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -4434,25 +4294,23 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX9-FLUSH-NEXT: s_lshr_b32 s0, s0, 16 ; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v1, s0 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v6, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v4, -1.0 op_sel_hi:[1,0,1] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, v0, v2, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v5, v5, v2, -v2 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, v1, v3, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v3, -v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v1, v6, v4 op_sel_hi:[1,0,0] ; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v7, v2 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v6, v3 +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -4466,25 +4324,23 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX10-IEEE-NEXT: s_lshr_b32 s1, s0, 16 ; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v1, s1 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX10-IEEE-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3 +; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, v0, v2, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, v1, v3, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_f32 v5, v5, v2, -v2 +; GFX10-IEEE-NEXT: v_fma_f32 v6, v6, v3, -v3 +; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v1, v6, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v7, v2 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -4498,25 +4354,22 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX10-FLUSH-NEXT: s_lshr_b32 s1, s0, 16 ; GFX10-FLUSH-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX10-FLUSH-NEXT: v_sqrt_f16_e32 v1, s1 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, -1.0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v2, v4, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 @@ -4530,29 +4383,27 @@ define amdgpu_ps i32 @s_rsq_v2f16(i32 inreg %a.arg) { ; GFX11-NEXT: s_lshr_b32 s1, s0, 16 ; GFX11-NEXT: v_sqrt_f16_e32 v0, s0 ; GFX11-NEXT: v_sqrt_f16_e32 v1, s1 -; GFX11-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX11-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX11-NEXT: s_waitcnt_depctr 0xfff ; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0 ; GFX11-NEXT: v_cvt_f32_f16_e32 v3, v1 ; GFX11-NEXT: v_rcp_f32_e32 v2, v2 ; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 -; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 +; GFX11-NEXT: v_fma_mix_f32 v5, v0, v2, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v6, v1, v3, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v5, v5, v2, -v2 +; GFX11-NEXT: v_fma_f32 v6, v6, v3, -v3 +; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v6, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_dual_mul_f32 v2, v7, v2 :: v_dual_mul_f32 v3, v4, v3 +; GFX11-NEXT: v_and_b32_e32 v2, 0xff800000, v2 +; GFX11-NEXT: v_dual_add_f32 v2, v2, v5 :: v_dual_and_b32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 +; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: v_readfirstlane_b32 s0, v0 ; GFX11-NEXT: ; return to shader part epilog @@ -4568,21 +4419,20 @@ define half @v_rsq_f16(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -4590,24 +4440,23 @@ define half @v_rsq_f16(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -4632,21 +4481,20 @@ define half @v_neg_rsq_f16(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -4654,24 +4502,23 @@ define half @v_neg_rsq_f16(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -4706,21 +4553,20 @@ define { half, half } @v_rsq_f16_multi_use(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v2, v2, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, vcc, v1, v2, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v6, v5 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v2, v1 +; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -4728,24 +4574,23 @@ define { half, half } @v_rsq_f16_multi_use(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v2, v2, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v1, v2, v1 +; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v0 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v3, v2, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -4785,21 +4630,20 @@ define half @v_rsq_f16_missing_contract0(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -4807,24 +4651,23 @@ define half @v_rsq_f16_missing_contract0(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -4859,21 +4702,20 @@ define half @v_rsq_f16_missing_contract1(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -4881,24 +4723,23 @@ define half @v_rsq_f16_missing_contract1(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -4933,21 +4774,20 @@ define half @v_neg_rsq_f16_missing_contract0(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -4955,24 +4795,23 @@ define half @v_neg_rsq_f16_missing_contract0(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5007,21 +4846,20 @@ define half @v_neg_rsq_f16_missing_contract1(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -5029,24 +4867,23 @@ define half @v_neg_rsq_f16_missing_contract1(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5081,21 +4918,20 @@ define half @v_neg_rsq_f16_fabs(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e64 v0, |v0| -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -5103,24 +4939,23 @@ define half @v_neg_rsq_f16_fabs(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e64 v0, |v0| -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5156,21 +4991,20 @@ define half @v_rsq_f16_arcp(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -5178,24 +5012,23 @@ define half @v_rsq_f16_arcp(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5220,21 +5053,20 @@ define half @v_neg_rsq_f16_arcp(half %a) { ; GFX6-IEEE: ; %bb.0: ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v5, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-IEEE-NEXT: v_fma_f32 v1, -v1, v4, v3 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] ; @@ -5242,24 +5074,23 @@ define half @v_neg_rsq_f16_arcp(half %a) { ; GFX6-FLUSH: ; %bb.0: ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, v1 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, v1, v0, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v1 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, v4, v2, v2 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v4, v3, v2 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v1, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v4, v5, v2, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v1, -v1, v4, v3 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, v1 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v1, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5294,12 +5125,10 @@ define half @v_rsq_f16_afn(half %a) { ; GFX6: ; %bb.0: ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-NEXT: v_rcp_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_f32_e32 v0, v1, v0 ; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-NEXT: s_setpc_b64 s[30:31] ; @@ -5324,12 +5153,10 @@ define half @v_rsq_f16_afn_nocontract(half %a) { ; GFX6: ; %bb.0: ; GFX6-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GFX6-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-NEXT: v_rcp_f32_e32 v0, v0 -; GFX6-NEXT: v_mul_f32_e32 v0, v1, v0 ; GFX6-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-NEXT: s_setpc_b64 s[30:31] ; @@ -5365,36 +5192,35 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, s[4:5], v1, v1, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v9, -v3, v6, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v9, v6, v6 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v8, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v10, -v3, v9, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v9, v10, v6, v9 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v9, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v6, v9 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v5, v8, 1.0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v7, s[4:5], v2, v1, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v3, v8, v8 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v7, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v5, v4, v7 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v3, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v5, v4, v7 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v5, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, s[4:5], v1, v1, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v8, -v2, v5, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v8, v5, v5 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v6, v4 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v8, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v10, -v2, v8, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v8, v10, v5, v8 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v8, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v9, -v4, v6, 1.0 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v5, v8 +; GFX6-IEEE-NEXT: v_div_scale_f32 v7, s[4:5], 1.0, v1, 1.0 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v9, v6, v6 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v3, v7, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v4, v3, v7 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v2, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v4, v3, v7 ; GFX6-IEEE-NEXT: s_mov_b64 vcc, s[4:5] -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v5, v3, v4 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v4, v2, v3 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] @@ -5404,42 +5230,40 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, 1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v0, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, 1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, 1.0, v1, 1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, 1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5448,31 +5272,28 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX8-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-IEEE-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v10, v10, v4 -; GFX8-IEEE-NEXT: v_add_f32_e32 v9, v9, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v10, v10, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v9, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v2, v2, v5 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v6, -v2, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v7, -v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, 1.0, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, 1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, 1.0, v3 +; GFX8-IEEE-NEXT: v_add_f32_e32 v2, 1.0, v2 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX8-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 -; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v3, v0, 1.0 @@ -5486,25 +5307,22 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v2 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v8, v4, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v5 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, -v2, v4, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, 1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v3, v0, 1.0 @@ -5518,25 +5336,22 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX9-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-IEEE-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v4, v6 -; GFX9-IEEE-NEXT: v_fma_f32 v9, -v2, v7, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v10, -v3, v8, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v7, v9, v5, v7 -; GFX9-IEEE-NEXT: v_fma_f32 v8, v10, v6, v8 -; GFX9-IEEE-NEXT: v_fma_f32 v2, -v2, v7, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v3, -v3, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v2, v2, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_fma_f32 v6, -v2, v4, 1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v7, -v3, v5, 1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v6, v6, v4, v4 +; GFX9-IEEE-NEXT: v_fma_f32 v7, v7, v5, v5 +; GFX9-IEEE-NEXT: v_fma_f32 v2, -v2, v6, 1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v3, -v3, v7, 1.0 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX9-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0 @@ -5549,25 +5364,23 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v7, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v6, v2 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v7, v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, -v1, v2, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v0, v3, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v5, v5, v2, v2 +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v3, v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v1, v5, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v7, v2 +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0 @@ -5580,25 +5393,23 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX10-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX10-IEEE-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, 1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, 1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3 +; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, -v1, v2, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v0, v3, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_f32 v5, v5, v2, v2 +; GFX10-IEEE-NEXT: v_fma_f32 v6, v6, v3, v3 +; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v0, v6, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v7, v2 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0 @@ -5611,25 +5422,22 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX10-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, 1.0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, -v2, v4, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, -v3, v5, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, 1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, 1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, 1.0 @@ -5642,7 +5450,7 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX11-NEXT: v_sqrt_f16_e32 v0, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v4, 1.0 +; GFX11-NEXT: v_mov_b32_e32 v4, 1.0 ; GFX11-NEXT: v_sqrt_f16_e32 v1, v1 ; GFX11-NEXT: s_waitcnt_depctr 0xfff ; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0 @@ -5650,22 +5458,20 @@ define <2 x half> @v_rsq_v2f16(<2 x half> %a) { ; GFX11-NEXT: v_rcp_f32_e32 v2, v2 ; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, 1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 -; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 +; GFX11-NEXT: v_fma_mix_f32 v5, -v0, v2, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v6, -v1, v3, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v5, v5, v2, v2 +; GFX11-NEXT: v_fma_f32 v6, v6, v3, v3 +; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v6, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_dual_mul_f32 v2, v7, v2 :: v_dual_mul_f32 v3, v4, v3 +; GFX11-NEXT: v_and_b32_e32 v2, 0xff800000, v2 +; GFX11-NEXT: v_dual_add_f32 v2, v2, v5 :: v_dual_and_b32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, 1.0 +; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, 1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] %sqrt = call contract <2 x half> @llvm.sqrt.v2f16(<2 x half> %a) @@ -5679,36 +5485,35 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v2, -1.0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_sqrt_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-IEEE-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX6-IEEE-NEXT: v_div_scale_f32 v4, vcc, v2, v0, v2 -; GFX6-IEEE-NEXT: v_div_scale_f32 v5, s[4:5], v1, v1, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v9, -v3, v6, 1.0 -; GFX6-IEEE-NEXT: v_fma_f32 v6, v9, v6, v6 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v9, v4, v6 -; GFX6-IEEE-NEXT: v_rcp_f32_e32 v8, v5 -; GFX6-IEEE-NEXT: v_fma_f32 v10, -v3, v9, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v9, v10, v6, v9 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v3, v9, v4 -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v3, v6, v9 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v3, v0, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v3, -v5, v8, 1.0 -; GFX6-IEEE-NEXT: v_div_scale_f32 v7, s[4:5], v2, v1, v2 -; GFX6-IEEE-NEXT: v_fma_f32 v3, v3, v8, v8 -; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v7, v3 -; GFX6-IEEE-NEXT: v_fma_f32 v6, -v5, v4, v7 -; GFX6-IEEE-NEXT: v_fma_f32 v4, v6, v3, v4 -; GFX6-IEEE-NEXT: v_fma_f32 v5, -v5, v4, v7 +; GFX6-IEEE-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, -1.0 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v5, v2 +; GFX6-IEEE-NEXT: v_div_scale_f32 v3, vcc, -1.0, v0, -1.0 +; GFX6-IEEE-NEXT: v_div_scale_f32 v4, s[4:5], v1, v1, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v8, -v2, v5, 1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v5, v8, v5, v5 +; GFX6-IEEE-NEXT: v_rcp_f32_e32 v6, v4 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v8, v3, v5 +; GFX6-IEEE-NEXT: v_fma_f32 v10, -v2, v8, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v8, v10, v5, v8 +; GFX6-IEEE-NEXT: v_fma_f32 v2, -v2, v8, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v9, -v4, v6, 1.0 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v2, v5, v8 +; GFX6-IEEE-NEXT: v_div_scale_f32 v7, s[4:5], -1.0, v1, -1.0 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v0, v2, v0, -1.0 +; GFX6-IEEE-NEXT: v_fma_f32 v2, v9, v6, v6 +; GFX6-IEEE-NEXT: v_mul_f32_e32 v3, v7, v2 +; GFX6-IEEE-NEXT: v_fma_f32 v5, -v4, v3, v7 +; GFX6-IEEE-NEXT: v_fma_f32 v3, v5, v2, v3 +; GFX6-IEEE-NEXT: v_fma_f32 v4, -v4, v3, v7 ; GFX6-IEEE-NEXT: s_mov_b64 vcc, s[4:5] -; GFX6-IEEE-NEXT: v_div_fmas_f32 v3, v5, v3, v4 -; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v3, v1, v2 +; GFX6-IEEE-NEXT: v_div_fmas_f32 v2, v4, v2, v3 +; GFX6-IEEE-NEXT: v_div_fixup_f32 v1, v2, v1, -1.0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-IEEE-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-IEEE-NEXT: s_setpc_b64 s[30:31] @@ -5718,42 +5523,40 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, -1.0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_sqrt_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 2, 2), 0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v3, s[4:5], v0, v0, v2 -; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v4, v3 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v2, v0, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v0, v0, -1.0 +; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v0, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v3, v4, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v4, v6, v4, v4 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v4 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v3, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v4, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, -v3, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v3, v3, v4, v6 -; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v3, v0, v2 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v0, v2, v0, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, -1.0 ; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v2 -; GFX6-FLUSH-NEXT: v_div_scale_f32 v5, vcc, v4, v1, v4 +; GFX6-FLUSH-NEXT: v_div_scale_f32 v4, vcc, -1.0, v1, -1.0 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v3, 1.0 -; GFX6-FLUSH-NEXT: v_fma_f32 v3, v6, v3, v3 -; GFX6-FLUSH-NEXT: v_mul_f32_e32 v6, v5, v3 -; GFX6-FLUSH-NEXT: v_fma_f32 v7, -v2, v6, v5 -; GFX6-FLUSH-NEXT: v_fma_f32 v6, v7, v3, v6 -; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v6, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, -v2, v3, 1.0 +; GFX6-FLUSH-NEXT: v_fma_f32 v3, v5, v3, v3 +; GFX6-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v3 +; GFX6-FLUSH-NEXT: v_fma_f32 v6, -v2, v5, v4 +; GFX6-FLUSH-NEXT: v_fma_f32 v5, v6, v3, v5 +; GFX6-FLUSH-NEXT: v_fma_f32 v2, -v2, v5, v4 ; GFX6-FLUSH-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v6 -; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, v4 +; GFX6-FLUSH-NEXT: v_div_fmas_f32 v2, v2, v3, v5 +; GFX6-FLUSH-NEXT: v_div_fixup_f32 v1, v2, v1, -1.0 ; GFX6-FLUSH-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GFX6-FLUSH-NEXT: s_setpc_b64 s[30:31] ; @@ -5762,31 +5565,28 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX8-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-IEEE-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX8-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX8-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v2 -; GFX8-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v8, v4, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v10, -v3, v8 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v9, -v2, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v10, v10, v4 -; GFX8-IEEE-NEXT: v_add_f32_e32 v9, v9, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v10, v10, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v9, v9, v5 -; GFX8-IEEE-NEXT: v_add_f32_e32 v8, v10, v8 -; GFX8-IEEE-NEXT: v_add_f32_e32 v7, v9, v7 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v8 -; GFX8-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v7 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 -; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v4 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 -; GFX8-IEEE-NEXT: v_mul_f32_e32 v2, v2, v5 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX8-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v2, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v3, v5 +; GFX8-IEEE-NEXT: v_add_f32_e32 v7, -1.0, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v6, -1.0, v6 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v7, v7, v5 +; GFX8-IEEE-NEXT: v_sub_f32_e32 v6, v6, v4 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v3, -v3, v7 +; GFX8-IEEE-NEXT: v_mul_f32_e64 v2, -v2, v6 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, -1.0, v3 +; GFX8-IEEE-NEXT: v_add_f32_e32 v2, -1.0, v2 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 +; GFX8-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX8-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX8-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 -; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 -; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 +; GFX8-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX8-IEEE-NEXT: v_div_fixup_f16 v0, v3, v0, -1.0 @@ -5800,25 +5600,22 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX8-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX8-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX8-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX8-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v2 -; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v6, v3 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v8, v4, v6 -; GFX8-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v4 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v6 -; GFX8-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v5 -; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v4 -; GFX8-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v4 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v6 -; GFX8-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v5 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 +; GFX8-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v2, v4, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX8-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX8-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 +; GFX8-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, -1.0 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 +; GFX8-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX8-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 -; GFX8-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 +; GFX8-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX8-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX8-FLUSH-NEXT: v_div_fixup_f16 v0, v3, v0, -1.0 @@ -5832,25 +5629,22 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX9-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-IEEE-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX9-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v2 -; GFX9-IEEE-NEXT: v_rcp_f32_e32 v6, v3 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v7, v4, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v8, v4, v6 -; GFX9-IEEE-NEXT: v_fma_f32 v9, -v2, v7, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v10, -v3, v8, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v7, v9, v5, v7 -; GFX9-IEEE-NEXT: v_fma_f32 v8, v10, v6, v8 -; GFX9-IEEE-NEXT: v_fma_f32 v2, -v2, v7, v4 -; GFX9-IEEE-NEXT: v_fma_f32 v3, -v3, v8, v4 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v2, v2, v5 -; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v6 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v4, v2 +; GFX9-IEEE-NEXT: v_rcp_f32_e32 v5, v3 +; GFX9-IEEE-NEXT: v_fma_f32 v6, v2, v4, -1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v7, v3, v5, -1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v6, v6, v4, -v4 +; GFX9-IEEE-NEXT: v_fma_f32 v7, v7, v5, -v5 +; GFX9-IEEE-NEXT: v_fma_f32 v2, -v2, v6, -1.0 +; GFX9-IEEE-NEXT: v_fma_f32 v3, -v3, v7, -1.0 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v2, v2, v4 +; GFX9-IEEE-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX9-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX9-IEEE-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX9-IEEE-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX9-IEEE-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-IEEE-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0 @@ -5863,25 +5657,23 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX9-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX9-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX9-FLUSH-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX9-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_rcp_f32_e32 v3, v3 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v5, v6, v2 -; GFX9-FLUSH-NEXT: v_mac_f32_e32 v4, v7, v3 -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1] -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v6, v2 -; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v7, v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v5, v1, v2, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v6, v0, v3, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_f32 v5, v5, v2, -v2 +; GFX9-FLUSH-NEXT: v_mad_f32 v6, v6, v3, -v3 +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v7, -v1, v5, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mad_mix_f32 v4, -v0, v6, v4 op_sel_hi:[1,0,0] +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v2, v7, v2 +; GFX9-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX9-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX9-FLUSH-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX9-FLUSH-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX9-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX9-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0 @@ -5894,25 +5686,23 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX10-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-IEEE-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX10-IEEE-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX10-IEEE-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX10-IEEE-NEXT: v_cvt_f32_f16_e32 v3, v0 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_rcp_f32_e32 v3, v3 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX10-IEEE-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, -v1, v5, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v0, v4, -1.0 op_sel_hi:[1,0,1] -; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v6, v2 -; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v7, v3 +; GFX10-IEEE-NEXT: v_fma_mix_f32 v5, v1, v2, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_mix_f32 v6, v0, v3, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_f32 v5, v5, v2, -v2 +; GFX10-IEEE-NEXT: v_fma_f32 v6, v6, v3, -v3 +; GFX10-IEEE-NEXT: v_fma_mix_f32 v7, -v1, v5, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_fma_mix_f32 v4, -v0, v6, v4 op_sel_hi:[1,0,0] +; GFX10-IEEE-NEXT: v_mul_f32_e32 v2, v7, v2 +; GFX10-IEEE-NEXT: v_mul_f32_e32 v3, v4, v3 ; GFX10-IEEE-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-IEEE-NEXT: v_and_b32_e32 v3, 0xff800000, v3 ; GFX10-IEEE-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v4 +; GFX10-IEEE-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-IEEE-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-IEEE-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0 @@ -5925,25 +5715,22 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX10-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-FLUSH-NEXT: v_sqrt_f16_e32 v1, v0 ; GFX10-FLUSH-NEXT: v_sqrt_f16_sdwa v0, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_1 -; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v6, -1.0 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v2, v1 ; GFX10-FLUSH-NEXT: v_cvt_f32_f16_e32 v3, v0 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v4, v2 ; GFX10-FLUSH-NEXT: v_rcp_f32_e32 v5, v3 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v7, v6, v4 -; GFX10-FLUSH-NEXT: v_mul_f32_e32 v8, v6, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v9, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v10, -v3, v8, v6 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v7, v9, v4 -; GFX10-FLUSH-NEXT: v_mac_f32_e32 v8, v10, v5 -; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v7, v6 -; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v8, v6 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v2, v4, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v3, v5, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v6, v6, v4, -v4 +; GFX10-FLUSH-NEXT: v_mad_f32 v7, v7, v5, -v5 +; GFX10-FLUSH-NEXT: v_mad_f32 v2, -v2, v6, -1.0 +; GFX10-FLUSH-NEXT: v_mad_f32 v3, -v3, v7, -1.0 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v4 ; GFX10-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v5 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v2, 0xff800000, v2 ; GFX10-FLUSH-NEXT: v_and_b32_e32 v3, 0xff800000, v3 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v7 -; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v8 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v2, v2, v6 +; GFX10-FLUSH-NEXT: v_add_f32_e32 v3, v3, v7 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GFX10-FLUSH-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX10-FLUSH-NEXT: v_div_fixup_f16 v1, v2, v1, -1.0 @@ -5956,7 +5743,7 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GFX11-NEXT: v_sqrt_f16_e32 v0, v0 -; GFX11-NEXT: v_cvt_f32_f16_e32 v4, -1.0 +; GFX11-NEXT: v_mov_b32_e32 v4, -1.0 ; GFX11-NEXT: v_sqrt_f16_e32 v1, v1 ; GFX11-NEXT: s_waitcnt_depctr 0xfff ; GFX11-NEXT: v_cvt_f32_f16_e32 v2, v0 @@ -5964,22 +5751,20 @@ define <2 x half> @v_neg_rsq_v2f16(<2 x half> %a) { ; GFX11-NEXT: v_rcp_f32_e32 v2, v2 ; GFX11-NEXT: v_rcp_f32_e32 v3, v3 ; GFX11-NEXT: s_waitcnt_depctr 0xfff -; GFX11-NEXT: v_mul_f32_e32 v5, v4, v2 -; GFX11-NEXT: v_mul_f32_e32 v4, v4, v3 -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v4, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fma_mix_f32 v7, -v1, v4, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_fmac_f32_e32 v5, v6, v2 -; GFX11-NEXT: v_mul_f32_e32 v3, v7, v3 -; GFX11-NEXT: v_fma_mix_f32 v6, -v0, v5, -1.0 op_sel_hi:[1,0,1] -; GFX11-NEXT: v_dual_mul_f32 v2, v6, v2 :: v_dual_and_b32 v3, 0xff800000, v3 -; GFX11-NEXT: v_dual_add_f32 v3, v3, v4 :: v_dual_and_b32 v2, 0xff800000, v2 -; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 -; GFX11-NEXT: v_add_f32_e32 v2, v2, v5 -; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 +; GFX11-NEXT: v_fma_mix_f32 v5, v0, v2, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v6, v1, v3, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_f32 v5, v5, v2, -v2 +; GFX11-NEXT: v_fma_f32 v6, v6, v3, -v3 +; GFX11-NEXT: v_fma_mix_f32 v7, -v0, v5, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_fma_mix_f32 v4, -v1, v6, v4 op_sel_hi:[1,0,0] +; GFX11-NEXT: v_dual_mul_f32 v2, v7, v2 :: v_dual_mul_f32 v3, v4, v3 +; GFX11-NEXT: v_and_b32_e32 v2, 0xff800000, v2 +; GFX11-NEXT: v_dual_add_f32 v2, v2, v5 :: v_dual_and_b32 v3, 0xff800000, v3 +; GFX11-NEXT: v_add_f32_e32 v3, v3, v6 ; GFX11-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GFX11-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GFX11-NEXT: v_div_fixup_f16 v0, v2, v0, -1.0 +; GFX11-NEXT: v_div_fixup_f16 v1, v3, v1, -1.0 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] %sqrt = call contract <2 x half> @llvm.sqrt.v2f16(<2 x half> %a) diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/frem.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/frem.ll index 302b239..549af87 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/frem.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/frem.ll @@ -88,11 +88,10 @@ define amdgpu_kernel void @frem_f16(ptr addrspace(1) %out, ptr addrspace(1) %in1 ; CI-NEXT: v_or_b32_e32 v1, s4, v0 ; CI-NEXT: .LBB0_8: ; %Flow19 ; CI-NEXT: v_cvt_f32_f16_e32 v0, s3 -; CI-NEXT: v_cvt_f32_f16_e32 v2, 0 ; CI-NEXT: s_and_b32 s2, s2, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s2, 0x7c00 ; CI-NEXT: s_cselect_b32 s2, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e32 vcc, v0, v2 +; CI-NEXT: v_cmp_nlg_f32_e32 vcc, 0, v0 ; CI-NEXT: v_mov_b32_e32 v0, 0x7e00 ; CI-NEXT: s_and_b32 s2, 1, s2 ; CI-NEXT: v_cndmask_b32_e32 v1, v1, v0, vcc @@ -1197,16 +1196,15 @@ define amdgpu_kernel void @frem_v2f16(ptr addrspace(1) %out, ptr addrspace(1) %i ; CI-NEXT: v_or_b32_e32 v1, s4, v1 ; CI-NEXT: .LBB9_16: ; %Flow54 ; CI-NEXT: v_cvt_f32_f16_e32 v2, s1 -; CI-NEXT: v_cvt_f32_f16_e32 v3, 0 ; CI-NEXT: s_and_b32 s0, s0, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s0, 0x7c00 ; CI-NEXT: s_cselect_b32 s4, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e32 vcc, v2, v3 +; CI-NEXT: v_cmp_nlg_f32_e32 vcc, 0, v2 ; CI-NEXT: v_cvt_f32_f16_e32 v2, s3 ; CI-NEXT: s_and_b32 s2, s2, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s2, 0x7c00 ; CI-NEXT: s_cselect_b32 s2, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e64 s[0:1], v2, v3 +; CI-NEXT: v_cmp_nlg_f32_e64 s[0:1], 0, v2 ; CI-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; CI-NEXT: v_mov_b32_e32 v2, 0x7e00 ; CI-NEXT: s_and_b32 s3, 1, s4 @@ -1730,26 +1728,25 @@ define amdgpu_kernel void @frem_v4f16(ptr addrspace(1) %out, ptr addrspace(1) %i ; CI-NEXT: v_or_b32_e32 v3, s1, v3 ; CI-NEXT: .LBB10_32: ; %Flow124 ; CI-NEXT: v_cvt_f32_f16_e32 v4, s2 -; CI-NEXT: v_cvt_f32_f16_e32 v5, 0 ; CI-NEXT: s_and_b32 s1, s4, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s1, 0x7c00 ; CI-NEXT: s_cselect_b32 s11, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e32 vcc, v4, v5 +; CI-NEXT: v_cmp_nlg_f32_e32 vcc, 0, v4 ; CI-NEXT: v_cvt_f32_f16_e32 v4, s0 ; CI-NEXT: s_and_b32 s2, s6, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s2, 0x7c00 ; CI-NEXT: s_cselect_b32 s6, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e64 s[0:1], v4, v5 +; CI-NEXT: v_cmp_nlg_f32_e64 s[0:1], 0, v4 ; CI-NEXT: v_cvt_f32_f16_e32 v4, s3 ; CI-NEXT: s_and_b32 s4, s5, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s4, 0x7c00 ; CI-NEXT: s_cselect_b32 s12, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e64 s[2:3], v4, v5 +; CI-NEXT: v_cmp_nlg_f32_e64 s[2:3], 0, v4 ; CI-NEXT: v_cvt_f32_f16_e32 v4, s10 ; CI-NEXT: s_and_b32 s7, s7, 0x7fff ; CI-NEXT: s_cmpk_lg_i32 s7, 0x7c00 ; CI-NEXT: s_cselect_b32 s7, 1, 0 -; CI-NEXT: v_cmp_nlg_f32_e64 s[4:5], v4, v5 +; CI-NEXT: v_cmp_nlg_f32_e64 s[4:5], 0, v4 ; CI-NEXT: v_and_b32_e32 v0, 0xffff, v0 ; CI-NEXT: v_mov_b32_e32 v4, 0x7e00 ; CI-NEXT: s_and_b32 s10, 1, s11 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smax.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smax.mir index eee553e..a7023d0 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smax.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smax.mir @@ -1,6 +1,5 @@ # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s | FileCheck %s -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s | FileCheck %s +# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s --- name: smax_s32_ss @@ -188,8 +187,7 @@ body: | ; CHECK-NEXT: [[ASHR:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST]], [[C]](s32) ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>) ; CHECK-NEXT: [[SEXT_INREG1:%[0-9]+]]:sgpr(s32) = G_SEXT_INREG [[BITCAST1]], 16 - ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 - ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C1]](s32) + ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C]](s32) ; CHECK-NEXT: [[SMAX:%[0-9]+]]:sgpr(s32) = G_SMAX [[SEXT_INREG]], [[SEXT_INREG1]] ; CHECK-NEXT: [[SMAX1:%[0-9]+]]:sgpr(s32) = G_SMAX [[ASHR]], [[ASHR1]] ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[SMAX]](s32), [[SMAX1]](s32) diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smin.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smin.mir index ef60aa8..9dd5f45 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smin.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-smin.mir @@ -1,6 +1,5 @@ # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s | FileCheck %s -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s | FileCheck %s +# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s --- name: smin_s32_ss @@ -191,8 +190,7 @@ body: | ; CHECK-NEXT: [[ASHR:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST]], [[C]](s32) ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>) ; CHECK-NEXT: [[SEXT_INREG1:%[0-9]+]]:sgpr(s32) = G_SEXT_INREG [[BITCAST1]], 16 - ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 - ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C1]](s32) + ; CHECK-NEXT: [[ASHR1:%[0-9]+]]:sgpr(s32) = G_ASHR [[BITCAST1]], [[C]](s32) ; CHECK-NEXT: [[SMIN:%[0-9]+]]:sgpr(s32) = G_SMIN [[SEXT_INREG]], [[SEXT_INREG1]] ; CHECK-NEXT: [[SMIN1:%[0-9]+]]:sgpr(s32) = G_SMIN [[ASHR]], [[ASHR1]] ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[SMIN]](s32), [[SMIN1]](s32) diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umax.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umax.mir index 36a38aa..59d7dce 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umax.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umax.mir @@ -1,6 +1,5 @@ # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s | FileCheck %s -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s | FileCheck %s +# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s --- name: umax_s32_ss @@ -186,15 +185,13 @@ body: | ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1 ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 - ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535 - ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C1]] + ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535 + ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C]] + ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 + ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C1]](s32) ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>) - ; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 - ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C2]](s32) - ; CHECK-NEXT: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535 - ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C3]] + ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C]] + ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C1]](s32) ; CHECK-NEXT: [[UMAX:%[0-9]+]]:sgpr(s32) = G_UMAX [[AND]], [[AND1]] ; CHECK-NEXT: [[UMAX1:%[0-9]+]]:sgpr(s32) = G_UMAX [[LSHR]], [[LSHR1]] ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[UMAX]](s32), [[UMAX1]](s32) diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umin.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umin.mir index bb232b5e..fdb05f6 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umin.mir +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/regbankselect-umin.mir @@ -1,6 +1,5 @@ # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-fast -o - %s | FileCheck %s -# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass=regbankselect -verify-machineinstrs -regbankselect-greedy -o - %s | FileCheck %s +# RUN: llc -mtriple=amdgcn -mcpu=gfx900 -run-pass='amdgpu-regbankselect,amdgpu-regbanklegalize' -o - %s | FileCheck %s --- name: umin_s32_ss @@ -190,15 +189,13 @@ body: | ; CHECK-NEXT: [[COPY:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr0 ; CHECK-NEXT: [[COPY1:%[0-9]+]]:sgpr(<2 x s16>) = COPY $sgpr1 ; CHECK-NEXT: [[BITCAST:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY]](<2 x s16>) - ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 - ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C]](s32) - ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535 - ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C1]] + ; CHECK-NEXT: [[C:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535 + ; CHECK-NEXT: [[AND:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST]], [[C]] + ; CHECK-NEXT: [[C1:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 + ; CHECK-NEXT: [[LSHR:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST]], [[C1]](s32) ; CHECK-NEXT: [[BITCAST1:%[0-9]+]]:sgpr(s32) = G_BITCAST [[COPY1]](<2 x s16>) - ; CHECK-NEXT: [[C2:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 16 - ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C2]](s32) - ; CHECK-NEXT: [[C3:%[0-9]+]]:sgpr(s32) = G_CONSTANT i32 65535 - ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C3]] + ; CHECK-NEXT: [[AND1:%[0-9]+]]:sgpr(s32) = G_AND [[BITCAST1]], [[C]] + ; CHECK-NEXT: [[LSHR1:%[0-9]+]]:sgpr(s32) = G_LSHR [[BITCAST1]], [[C1]](s32) ; CHECK-NEXT: [[UMIN:%[0-9]+]]:sgpr(s32) = G_UMIN [[AND]], [[AND1]] ; CHECK-NEXT: [[UMIN1:%[0-9]+]]:sgpr(s32) = G_UMIN [[LSHR]], [[LSHR1]] ; CHECK-NEXT: [[BUILD_VECTOR_TRUNC:%[0-9]+]]:sgpr(<2 x s16>) = G_BUILD_VECTOR_TRUNC [[UMIN]](s32), [[UMIN1]](s32) diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll index 9ffc565..4f2c454 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/sdiv.i64.ll @@ -2537,202 +2537,195 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) { ; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4 ; GISEL-NEXT: v_cvt_f32_u32_e32 v3, v1 -; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v5, 0 -; GISEL-NEXT: v_sub_i32_e32 v10, vcc, 0, v1 -; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v5 +; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v9, 0 +; GISEL-NEXT: v_sub_i32_e32 v8, vcc, 0, v1 +; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v9 ; GISEL-NEXT: v_rcp_iflag_f32_e32 v3, v3 -; GISEL-NEXT: v_subb_u32_e64 v11, s[4:5], 0, 0, vcc +; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, 0, vcc ; GISEL-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3 ; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3 -; GISEL-NEXT: v_trunc_f32_e32 v7, v4 -; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v7 -; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v3 -; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v7 -; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v9, 0 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[4:5] -; GISEL-NEXT: v_mul_lo_u32 v4, v12, v3 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8] -; GISEL-NEXT: v_mul_hi_u32 v8, v9, v3 -; GISEL-NEXT: v_mul_hi_u32 v3, v12, v3 -; GISEL-NEXT: v_mul_lo_u32 v13, v9, v7 -; GISEL-NEXT: v_mul_lo_u32 v14, v12, v7 -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13 +; GISEL-NEXT: v_trunc_f32_e32 v5, v4 +; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5 +; GISEL-NEXT: v_cvt_u32_f32_e32 v7, v3 +; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v5 +; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v7, 0 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v11, v[4:5] +; GISEL-NEXT: v_mul_hi_u32 v12, v7, v3 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v7, v[4:5] +; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3 +; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3 +; GISEL-NEXT: v_mul_lo_u32 v13, v7, v4 +; GISEL-NEXT: v_mul_lo_u32 v14, v11, v4 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13 ; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8 -; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7 -; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v13, v4 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12 +; GISEL-NEXT: v_mul_hi_u32 v12, v7, v4 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5 ; GISEL-NEXT: v_add_i32_e32 v3, vcc, v14, v3 ; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v8 -; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8 -; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4 -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4 -; GISEL-NEXT: v_add_i32_e32 v9, vcc, v9, v3 -; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v12, v4, vcc -; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v10, v9, 0 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v10, v12, v[4:5] -; GISEL-NEXT: v_mul_lo_u32 v4, v12, v3 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v12 +; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12 +; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5 +; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v3 +; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v4, vcc +; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v8, v7, 0 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v8, v11, v[4:5] +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v7, v[4:5] +; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3 ; GISEL-NEXT: v_and_b32_e32 v10, 0xffffff, v0 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v9, v[7:8] -; GISEL-NEXT: v_mul_hi_u32 v0, v9, v3 -; GISEL-NEXT: v_mul_hi_u32 v3, v12, v3 -; GISEL-NEXT: v_mul_lo_u32 v8, v9, v7 -; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v2 -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8 +; GISEL-NEXT: v_mul_lo_u32 v8, v7, v4 +; GISEL-NEXT: v_mul_hi_u32 v0, v7, v3 +; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8 ; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0 ; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GISEL-NEXT: v_mul_lo_u32 v4, v12, v7 +; GISEL-NEXT: v_mul_lo_u32 v5, v11, v4 ; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0 -; GISEL-NEXT: v_mul_hi_u32 v8, v9, v7 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3 -; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc +; GISEL-NEXT: v_mul_hi_u32 v8, v7, v4 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v8 ; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8 -; GISEL-NEXT: v_mul_hi_u32 v7, v12, v7 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8 +; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4 ; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0 ; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3 ; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0 -; GISEL-NEXT: v_addc_u32_e32 v4, vcc, v12, v3, vcc -; GISEL-NEXT: v_mul_lo_u32 v7, 0, v0 -; GISEL-NEXT: v_mul_lo_u32 v8, v10, v4 -; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6 -; GISEL-NEXT: v_mul_hi_u32 v6, v10, v0 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0 +; GISEL-NEXT: v_addc_u32_e32 v3, vcc, v11, v3, vcc +; GISEL-NEXT: v_mul_lo_u32 v4, 0, v0 +; GISEL-NEXT: v_mul_lo_u32 v5, v10, v3 +; GISEL-NEXT: v_mul_hi_u32 v7, v10, v0 ; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v8 -; GISEL-NEXT: v_mul_lo_u32 v8, 0, v4 -; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6 -; GISEL-NEXT: v_mul_hi_u32 v7, v10, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0 +; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v2 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5 +; GISEL-NEXT: v_mul_lo_u32 v5, 0, v3 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v7 +; GISEL-NEXT: v_mul_hi_u32 v7, v10, v3 +; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0 ; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7 -; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6 -; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v1, v0, 0 -; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9 -; GISEL-NEXT: v_cvt_f32_u32_e32 v9, v3 -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v8 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v1, v4, v[7:8] -; GISEL-NEXT: v_mac_f32_e32 v9, 0x4f800000, v5 -; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v9 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], 0, v0, v[7:8] -; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v10, v6 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v4 +; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v1, v0, 0 +; GISEL-NEXT: v_mul_hi_u32 v4, 0, v3 +; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6 +; GISEL-NEXT: v_mov_b32_e32 v5, v8 +; GISEL-NEXT: v_cvt_f32_u32_e32 v8, v3 +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v4, v[5:6] +; GISEL-NEXT: v_mac_f32_e32 v8, 0x4f800000, v9 +; GISEL-NEXT: v_rcp_iflag_f32_e32 v2, v8 +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v0, v[5:6] +; GISEL-NEXT: v_sub_i32_e32 v9, vcc, v10, v7 ; GISEL-NEXT: v_mul_f32_e32 v2, 0x5f7ffffc, v2 -; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v2 -; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], 0, v7, vcc -; GISEL-NEXT: v_sub_i32_e64 v10, s[4:5], 0, v7 -; GISEL-NEXT: v_trunc_f32_e32 v7, v5 -; GISEL-NEXT: v_mac_f32_e32 v2, 0xcf800000, v7 +; GISEL-NEXT: v_mul_f32_e32 v6, 0x2f800000, v2 +; GISEL-NEXT: v_trunc_f32_e32 v8, v6 +; GISEL-NEXT: v_mac_f32_e32 v2, 0xcf800000, v8 ; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v2 +; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, v5, vcc ; GISEL-NEXT: v_sub_i32_e64 v13, s[4:5], 0, v3 ; GISEL-NEXT: v_subb_u32_e64 v14, s[4:5], 0, 0, s[4:5] -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v12, 0 -; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v7 -; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v8, v1 -; GISEL-NEXT: v_mov_b32_e32 v2, v6 -; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, -1, s[4:5] -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v15, v[2:3] -; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v9 -; GISEL-NEXT: v_cndmask_b32_e64 v9, -1, v16, s[4:5] -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v14, v12, v[6:7] -; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v10, vcc -; GISEL-NEXT: v_mul_lo_u32 v7, v15, v5 -; GISEL-NEXT: v_mul_lo_u32 v10, v12, v6 -; GISEL-NEXT: v_sub_i32_e32 v8, vcc, v8, v1 -; GISEL-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v2, vcc -; GISEL-NEXT: v_mul_hi_u32 v2, v12, v5 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10 +; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v12, 0 +; GISEL-NEXT: v_cvt_u32_f32_e32 v15, v8 +; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v5 +; GISEL-NEXT: v_mov_b32_e32 v2, v7 +; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v15, v[2:3] +; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v9, v1 +; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, s[4:5] +; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v14, v12, v[7:8] +; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v10 +; GISEL-NEXT: v_cndmask_b32_e64 v8, -1, v2, s[4:5] +; GISEL-NEXT: v_mul_lo_u32 v2, v15, v6 +; GISEL-NEXT: v_mul_lo_u32 v10, v12, v7 +; GISEL-NEXT: v_subbrev_u32_e32 v16, vcc, 0, v5, vcc +; GISEL-NEXT: v_mul_hi_u32 v5, v12, v6 +; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v10 ; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2 +; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v5 ; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc -; GISEL-NEXT: v_mul_lo_u32 v7, v15, v6 -; GISEL-NEXT: v_mul_hi_u32 v5, v15, v5 +; GISEL-NEXT: v_mul_lo_u32 v5, v15, v7 +; GISEL-NEXT: v_mul_hi_u32 v6, v15, v6 ; GISEL-NEXT: v_add_i32_e32 v2, vcc, v10, v2 -; GISEL-NEXT: v_mul_hi_u32 v10, v12, v6 -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5 -; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc +; GISEL-NEXT: v_mul_hi_u32 v10, v12, v7 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6 +; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10 ; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10 -; GISEL-NEXT: v_mul_hi_u32 v6, v15, v6 +; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10 +; GISEL-NEXT: v_mul_hi_u32 v7, v15, v7 ; GISEL-NEXT: v_add_i32_e32 v2, vcc, v5, v2 ; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5 ; GISEL-NEXT: v_add_i32_e32 v5, vcc, v6, v5 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v12, v2 -; GISEL-NEXT: v_addc_u32_e32 v10, vcc, v15, v5, vcc -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v7, 0 -; GISEL-NEXT: v_add_i32_e32 v12, vcc, 1, v0 -; GISEL-NEXT: v_addc_u32_e32 v15, vcc, 0, v4, vcc -; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v8, v1 -; GISEL-NEXT: v_mov_b32_e32 v1, v6 -; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v13, v10, v[1:2] -; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, -1, vcc -; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v16 -; GISEL-NEXT: v_mad_u64_u32 v[1:2], s[4:5], v14, v7, v[1:2] -; GISEL-NEXT: v_cndmask_b32_e32 v6, -1, v8, vcc -; GISEL-NEXT: v_add_i32_e32 v2, vcc, 1, v12 -; GISEL-NEXT: v_addc_u32_e32 v8, vcc, 0, v15, vcc -; GISEL-NEXT: v_mul_lo_u32 v13, v10, v5 -; GISEL-NEXT: v_mul_lo_u32 v14, v7, v1 -; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6 -; GISEL-NEXT: v_mul_hi_u32 v6, v7, v5 -; GISEL-NEXT: v_cndmask_b32_e32 v2, v12, v2, vcc -; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v13, v14 -; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v12, v6 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5] -; GISEL-NEXT: v_mul_lo_u32 v12, v10, v1 -; GISEL-NEXT: v_mul_hi_u32 v5, v10, v5 -; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v13, v6 -; GISEL-NEXT: v_mul_hi_u32 v13, v7, v1 -; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v12, v5 -; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v13 -; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v12, s[4:5], v12, v13 -; GISEL-NEXT: v_mul_hi_u32 v1, v10, v1 -; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v5, v6 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v6, s[4:5], v12, v6 -; GISEL-NEXT: v_add_i32_e64 v1, s[4:5], v1, v6 -; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5 -; GISEL-NEXT: v_addc_u32_e64 v1, s[4:5], v10, v1, s[4:5] +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5 +; GISEL-NEXT: v_add_i32_e32 v10, vcc, v12, v2 +; GISEL-NEXT: v_addc_u32_e32 v12, vcc, v15, v5, vcc +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v10, 0 +; GISEL-NEXT: v_sub_i32_e32 v9, vcc, v9, v1 +; GISEL-NEXT: v_mov_b32_e32 v2, v6 +; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v13, v12, v[2:3] +; GISEL-NEXT: v_subbrev_u32_e32 v15, vcc, 0, v16, vcc +; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v14, v10, v[6:7] +; GISEL-NEXT: v_add_i32_e32 v2, vcc, 1, v0 +; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v4, vcc +; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v9, v1 +; GISEL-NEXT: v_mul_lo_u32 v7, v12, v5 +; GISEL-NEXT: v_mul_lo_u32 v9, v10, v6 +; GISEL-NEXT: v_mul_hi_u32 v14, v10, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v1, 0, -1, vcc +; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v15 +; GISEL-NEXT: v_cndmask_b32_e32 v1, -1, v1, vcc +; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9 +; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v14 +; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc +; GISEL-NEXT: v_mul_lo_u32 v14, v12, v6 +; GISEL-NEXT: v_mul_hi_u32 v5, v12, v5 +; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7 +; GISEL-NEXT: v_mul_hi_u32 v9, v10, v6 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v14, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v9 +; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v9, vcc, v14, v9 +; GISEL-NEXT: v_mul_hi_u32 v6, v12, v6 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v7 +; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7 +; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v7 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v10, v5 +; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v12, v6, vcc ; GISEL-NEXT: v_mul_lo_u32 v6, 0, v5 -; GISEL-NEXT: v_mul_lo_u32 v7, v11, v1 -; GISEL-NEXT: v_mul_hi_u32 v10, v11, v5 -; GISEL-NEXT: v_cndmask_b32_e32 v8, v15, v8, vcc +; GISEL-NEXT: v_mul_lo_u32 v9, v11, v7 +; GISEL-NEXT: v_mul_hi_u32 v14, v11, v5 +; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v2 +; GISEL-NEXT: v_addc_u32_e32 v12, vcc, 0, v13, vcc +; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v9 +; GISEL-NEXT: v_mul_lo_u32 v9, 0, v7 ; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5 -; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v7 -; GISEL-NEXT: v_mul_lo_u32 v7, 0, v1 -; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v10 -; GISEL-NEXT: v_mul_hi_u32 v10, v11, v1 +; GISEL-NEXT: v_add_i32_e32 v6, vcc, v6, v14 +; GISEL-NEXT: v_mul_hi_u32 v14, v11, v7 ; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v7, v5 -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10 -; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v5, v6 -; GISEL-NEXT: v_mul_hi_u32 v1, 0, v1 -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v10, 0 -; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v12 -; GISEL-NEXT: v_add_i32_e32 v12, vcc, v1, v7 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v14 +; GISEL-NEXT: v_add_i32_e32 v9, vcc, v5, v6 +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v9, 0 +; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1 +; GISEL-NEXT: v_cndmask_b32_e32 v2, v2, v10, vcc +; GISEL-NEXT: v_mul_hi_u32 v10, 0, v7 ; GISEL-NEXT: v_mov_b32_e32 v1, v6 -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v3, v12, v[1:2] -; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9 +; GISEL-NEXT: v_cndmask_b32_e32 v12, v13, v12, vcc +; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v8 +; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v3, v10, v[1:2] ; GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v10, v[6:7] -; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v8, vcc +; GISEL-NEXT: v_cndmask_b32_e32 v1, v4, v12, vcc +; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v9, v[6:7] ; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v11, v5 ; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v6 ; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], 0, v6, vcc @@ -2743,8 +2736,8 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) { ; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v4 ; GISEL-NEXT: v_subbrev_u32_e32 v5, vcc, 0, v5, vcc ; GISEL-NEXT: v_cndmask_b32_e64 v4, -1, v6, s[4:5] -; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v10 -; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v12, vcc +; GISEL-NEXT: v_add_i32_e32 v6, vcc, 1, v9 +; GISEL-NEXT: v_addc_u32_e32 v7, vcc, 0, v10, vcc ; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v2, v3 ; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, -1, vcc ; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v5 @@ -2755,8 +2748,8 @@ define <2 x i64> @v_sdiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) { ; GISEL-NEXT: v_cndmask_b32_e32 v2, v6, v3, vcc ; GISEL-NEXT: v_cndmask_b32_e32 v3, v7, v5, vcc ; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v4 -; GISEL-NEXT: v_cndmask_b32_e32 v2, v10, v2, vcc -; GISEL-NEXT: v_cndmask_b32_e32 v3, v12, v3, vcc +; GISEL-NEXT: v_cndmask_b32_e32 v2, v9, v2, vcc +; GISEL-NEXT: v_cndmask_b32_e32 v3, v10, v3, vcc ; GISEL-NEXT: s_setpc_b64 s[30:31] ; ; CGP-LABEL: v_sdiv_v2i64_24bit: diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll index ac1e11b..dfa613c 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/smed3.ll @@ -1,7 +1,7 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s -; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s -; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s define i32 @test_min_max_ValK0_K1_i32(i32 %a) { ; GFX89-LABEL: test_min_max_ValK0_K1_i32: diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll index 82279e6..40b5db0 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/srem.i64.ll @@ -3035,203 +3035,193 @@ define <2 x i64> @v_srem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) { ; GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GISEL-NEXT: v_and_b32_e32 v1, 0xffffff, v4 ; GISEL-NEXT: v_cvt_f32_u32_e32 v3, v1 -; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v4, 0 -; GISEL-NEXT: v_sub_i32_e32 v11, vcc, 0, v1 -; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v4 +; GISEL-NEXT: v_cvt_f32_ubyte0_e32 v7, 0 +; GISEL-NEXT: v_sub_i32_e32 v9, vcc, 0, v1 +; GISEL-NEXT: v_mac_f32_e32 v3, 0x4f800000, v7 ; GISEL-NEXT: v_rcp_iflag_f32_e32 v3, v3 -; GISEL-NEXT: v_subb_u32_e64 v12, s[4:5], 0, 0, vcc +; GISEL-NEXT: v_subb_u32_e64 v10, s[4:5], 0, 0, vcc ; GISEL-NEXT: v_and_b32_e32 v2, 0xffffff, v2 ; GISEL-NEXT: v_mul_f32_e32 v3, 0x5f7ffffc, v3 -; GISEL-NEXT: v_mul_f32_e32 v5, 0x2f800000, v3 -; GISEL-NEXT: v_trunc_f32_e32 v5, v5 +; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v3 +; GISEL-NEXT: v_trunc_f32_e32 v5, v4 ; GISEL-NEXT: v_mac_f32_e32 v3, 0xcf800000, v5 -; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v3 -; GISEL-NEXT: v_cvt_u32_f32_e32 v5, v5 -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v10, 0 -; GISEL-NEXT: v_mov_b32_e32 v3, v8 -; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v11, v5, v[3:4] -; GISEL-NEXT: v_mul_lo_u32 v3, v5, v7 -; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v10, v[8:9] -; GISEL-NEXT: v_mul_hi_u32 v9, v10, v7 -; GISEL-NEXT: v_mul_hi_u32 v7, v5, v7 -; GISEL-NEXT: v_mul_lo_u32 v13, v10, v8 -; GISEL-NEXT: v_mul_lo_u32 v14, v5, v8 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v13 +; GISEL-NEXT: v_cvt_u32_f32_e32 v8, v3 +; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v5 +; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v9, v8, 0 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v11, v[4:5] +; GISEL-NEXT: v_mul_hi_u32 v12, v8, v3 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v8, v[4:5] +; GISEL-NEXT: v_mul_lo_u32 v5, v11, v3 +; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3 +; GISEL-NEXT: v_mul_lo_u32 v13, v8, v4 +; GISEL-NEXT: v_mul_lo_u32 v14, v11, v4 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13 ; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9 -; GISEL-NEXT: v_mul_hi_u32 v9, v10, v8 -; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v13, v3 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v14, v7 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12 +; GISEL-NEXT: v_mul_hi_u32 v12, v8, v4 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v14, v3 ; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9 -; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v9, vcc, v13, v9 -; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3 -; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v7 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v8, v7 -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v3 -; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v5, v7, vcc -; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v11, v10, 0 -; GISEL-NEXT: v_mov_b32_e32 v3, v8 -; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v11, v5, v[3:4] -; GISEL-NEXT: v_mul_lo_u32 v3, v5, v7 -; GISEL-NEXT: v_and_b32_e32 v11, 0xffffff, v0 -; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v12, v10, v[8:9] -; GISEL-NEXT: v_mul_hi_u32 v0, v10, v7 -; GISEL-NEXT: v_mul_hi_u32 v7, v5, v7 -; GISEL-NEXT: v_mul_lo_u32 v9, v10, v8 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9 -; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v12 +; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12 +; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v12, v5 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5 +; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v3 +; GISEL-NEXT: v_addc_u32_e32 v11, vcc, v11, v4, vcc +; GISEL-NEXT: v_mad_u64_u32 v[3:4], s[4:5], v9, v8, 0 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v9, v11, v[4:5] +; GISEL-NEXT: v_mul_hi_u32 v9, v8, v3 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v10, v8, v[4:5] +; GISEL-NEXT: v_and_b32_e32 v10, 0xffffff, v0 +; GISEL-NEXT: v_mul_lo_u32 v0, v11, v3 +; GISEL-NEXT: v_mul_lo_u32 v5, v8, v4 +; GISEL-NEXT: v_mul_hi_u32 v3, v11, v3 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v9 ; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GISEL-NEXT: v_mul_lo_u32 v3, v5, v8 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0 -; GISEL-NEXT: v_mul_hi_u32 v9, v10, v8 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v7 -; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v9 +; GISEL-NEXT: v_mul_lo_u32 v9, v11, v4 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0 +; GISEL-NEXT: v_mul_hi_u32 v5, v8, v4 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v9, v3 ; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9 -; GISEL-NEXT: v_mul_hi_u32 v8, v5, v8 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v3, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v9, v5 +; GISEL-NEXT: v_mul_hi_u32 v4, v11, v4 ; GISEL-NEXT: v_add_i32_e32 v0, vcc, v3, v0 ; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v7, v3 -; GISEL-NEXT: v_add_i32_e32 v3, vcc, v8, v3 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v10, v0 -; GISEL-NEXT: v_addc_u32_e32 v5, vcc, v5, v3, vcc -; GISEL-NEXT: v_mul_lo_u32 v7, 0, v0 -; GISEL-NEXT: v_mul_lo_u32 v8, v11, v5 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v5, v3 +; GISEL-NEXT: v_add_i32_e32 v3, vcc, v4, v3 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0 +; GISEL-NEXT: v_addc_u32_e32 v8, vcc, v11, v3, vcc +; GISEL-NEXT: v_mul_lo_u32 v4, 0, v0 +; GISEL-NEXT: v_mul_lo_u32 v5, v10, v8 ; GISEL-NEXT: v_and_b32_e32 v3, 0xffffff, v6 -; GISEL-NEXT: v_mul_hi_u32 v6, v11, v0 +; GISEL-NEXT: v_mul_hi_u32 v6, v10, v0 ; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v8 -; GISEL-NEXT: v_mul_lo_u32 v8, 0, v5 -; GISEL-NEXT: v_add_i32_e32 v6, vcc, v7, v6 -; GISEL-NEXT: v_mul_hi_u32 v7, v11, v5 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v8, v0 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v7 -; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v8, vcc, v0, v6 -; GISEL-NEXT: v_mul_hi_u32 v9, 0, v5 -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v8, 0 -; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v7, v0 -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v9, v0 -; GISEL-NEXT: v_mov_b32_e32 v0, v6 -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], v1, v7, v[0:1] -; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v3 -; GISEL-NEXT: v_mad_u64_u32 v[6:7], s[4:5], 0, v8, v[6:7] -; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v4 -; GISEL-NEXT: v_rcp_iflag_f32_e32 v0, v0 -; GISEL-NEXT: v_sub_i32_e32 v7, vcc, v11, v5 -; GISEL-NEXT: v_subb_u32_e64 v8, s[4:5], 0, v6, vcc -; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v0 -; GISEL-NEXT: v_mul_f32_e32 v4, 0x2f800000, v0 -; GISEL-NEXT: v_trunc_f32_e32 v9, v4 -; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v9 -; GISEL-NEXT: v_cvt_u32_f32_e32 v10, v0 -; GISEL-NEXT: v_sub_i32_e64 v11, s[4:5], 0, v3 -; GISEL-NEXT: v_subb_u32_e64 v12, s[4:5], 0, 0, s[4:5] -; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, 0 -; GISEL-NEXT: v_cvt_u32_f32_e32 v9, v9 -; GISEL-NEXT: v_sub_i32_e64 v13, s[4:5], 0, v6 -; GISEL-NEXT: v_mov_b32_e32 v0, v5 -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[0:1] -; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v7, v1 -; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, s[4:5] -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6] -; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v8 -; GISEL-NEXT: v_cndmask_b32_e64 v14, -1, v0, s[4:5] -; GISEL-NEXT: v_mul_lo_u32 v0, v9, v4 -; GISEL-NEXT: v_mul_lo_u32 v6, v10, v5 -; GISEL-NEXT: v_mul_hi_u32 v15, v10, v4 -; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v13, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v15 -; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GISEL-NEXT: v_mul_lo_u32 v15, v9, v5 -; GISEL-NEXT: v_mul_hi_u32 v4, v9, v4 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0 -; GISEL-NEXT: v_mul_hi_u32 v6, v10, v5 -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v15, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v5 +; GISEL-NEXT: v_mul_lo_u32 v5, 0, v8 ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v6 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v6, vcc, v15, v6 -; GISEL-NEXT: v_mul_hi_u32 v5, v9, v5 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v4, v0 +; GISEL-NEXT: v_mul_hi_u32 v6, v10, v8 ; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v6, v4 -; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v4 -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v0 -; GISEL-NEXT: v_addc_u32_e32 v9, vcc, v9, v4, vcc -; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v11, v10, 0 -; GISEL-NEXT: v_sub_i32_e32 v15, vcc, v7, v1 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v5, v0 +; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6 +; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v4 +; GISEL-NEXT: v_cvt_f32_u32_e32 v0, v3 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v1, v9, 0 +; GISEL-NEXT: v_mul_hi_u32 v6, 0, v8 +; GISEL-NEXT: v_mac_f32_e32 v0, 0x4f800000, v7 +; GISEL-NEXT: v_rcp_iflag_f32_e32 v7, v0 ; GISEL-NEXT: v_mov_b32_e32 v0, v5 -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v11, v9, v[0:1] -; GISEL-NEXT: v_subbrev_u32_e32 v13, vcc, 0, v13, vcc -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v12, v10, v[5:6] -; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v15, v1 -; GISEL-NEXT: v_cndmask_b32_e64 v16, 0, -1, vcc -; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v13 -; GISEL-NEXT: v_cndmask_b32_e32 v0, -1, v16, vcc -; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v15, v1 -; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v13, vcc -; GISEL-NEXT: v_mul_lo_u32 v11, v9, v4 -; GISEL-NEXT: v_mul_lo_u32 v12, v10, v5 -; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v0 -; GISEL-NEXT: v_mul_hi_u32 v0, v10, v4 -; GISEL-NEXT: v_mul_hi_u32 v4, v9, v4 -; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v11, v12 -; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v11, v0 -; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, s[4:5] -; GISEL-NEXT: v_mul_lo_u32 v11, v9, v5 -; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v12, v0 -; GISEL-NEXT: v_mul_hi_u32 v12, v10, v5 -; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v11, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v12 -; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v11, s[4:5], v11, v12 -; GISEL-NEXT: v_mul_hi_u32 v5, v9, v5 -; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v4, v0 +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v1, v6, v[0:1] +; GISEL-NEXT: v_mul_f32_e32 v0, 0x5f7ffffc, v7 +; GISEL-NEXT: v_mul_f32_e32 v7, 0x2f800000, v0 +; GISEL-NEXT: v_trunc_f32_e32 v11, v7 +; GISEL-NEXT: v_mac_f32_e32 v0, 0xcf800000, v11 +; GISEL-NEXT: v_cvt_u32_f32_e32 v12, v0 +; GISEL-NEXT: v_sub_i32_e32 v13, vcc, 0, v3 +; GISEL-NEXT: v_cvt_u32_f32_e32 v11, v11 +; GISEL-NEXT: v_mad_u64_u32 v[7:8], s[4:5], v13, v12, 0 +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v9, v[5:6] +; GISEL-NEXT: v_mov_b32_e32 v0, v8 +; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v13, v11, v[0:1] +; GISEL-NEXT: v_subb_u32_e64 v14, s[4:5], 0, 0, vcc +; GISEL-NEXT: v_mad_u64_u32 v[8:9], s[4:5], v14, v12, v[8:9] +; GISEL-NEXT: v_sub_i32_e32 v10, vcc, v10, v4 +; GISEL-NEXT: v_subb_u32_e64 v15, s[4:5], 0, v5, vcc +; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], 0, v5 +; GISEL-NEXT: v_mul_lo_u32 v4, v11, v7 +; GISEL-NEXT: v_mul_lo_u32 v5, v12, v8 +; GISEL-NEXT: v_mul_hi_u32 v9, v12, v7 +; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v10, v1 +; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, s[4:5] +; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5] +; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v4, v9 ; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, s[4:5] -; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v11, v4 +; GISEL-NEXT: v_mul_lo_u32 v9, v11, v8 +; GISEL-NEXT: v_mul_hi_u32 v7, v11, v7 ; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v5, v4 -; GISEL-NEXT: v_add_i32_e64 v0, s[4:5], v10, v0 -; GISEL-NEXT: v_addc_u32_e64 v4, s[4:5], v9, v4, s[4:5] -; GISEL-NEXT: v_mul_lo_u32 v5, 0, v0 -; GISEL-NEXT: v_mul_lo_u32 v9, v2, v4 -; GISEL-NEXT: v_cndmask_b32_e32 v10, v13, v6, vcc -; GISEL-NEXT: v_mul_hi_u32 v6, v2, v0 -; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v1, vcc -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v9 -; GISEL-NEXT: v_mul_lo_u32 v9, 0, v4 -; GISEL-NEXT: v_mul_hi_u32 v0, 0, v0 -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v6 -; GISEL-NEXT: v_mul_hi_u32 v6, v2, v4 +; GISEL-NEXT: v_mul_hi_u32 v5, v12, v8 +; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v9, v7 +; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, s[4:5] +; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, 1, s[4:5] +; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v9, v7 +; GISEL-NEXT: v_mul_hi_u32 v8, v11, v8 +; GISEL-NEXT: v_add_i32_e64 v4, s[4:5], v5, v4 +; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, s[4:5] +; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v7, v5 +; GISEL-NEXT: v_add_i32_e64 v5, s[4:5], v8, v5 +; GISEL-NEXT: v_add_i32_e64 v7, s[4:5], v12, v4 +; GISEL-NEXT: v_addc_u32_e64 v8, s[4:5], v11, v5, s[4:5] +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v13, v7, 0 +; GISEL-NEXT: v_cmp_eq_u32_e64 s[4:5], 0, v15 +; GISEL-NEXT: v_subbrev_u32_e32 v11, vcc, 0, v0, vcc +; GISEL-NEXT: v_mov_b32_e32 v0, v5 +; GISEL-NEXT: v_cndmask_b32_e64 v9, -1, v6, s[4:5] +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v13, v8, v[0:1] +; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v10, v1 +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v14, v7, v[5:6] +; GISEL-NEXT: v_mul_lo_u32 v12, v8, v4 +; GISEL-NEXT: v_subbrev_u32_e32 v11, vcc, 0, v11, vcc +; GISEL-NEXT: v_mul_lo_u32 v13, v7, v5 +; GISEL-NEXT: v_mul_hi_u32 v14, v7, v4 +; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1 +; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, -1, vcc +; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v11 +; GISEL-NEXT: v_cndmask_b32_e32 v6, -1, v6, vcc +; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v13 +; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v14 +; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc +; GISEL-NEXT: v_mul_lo_u32 v14, v8, v5 +; GISEL-NEXT: v_mul_hi_u32 v4, v8, v4 +; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12 +; GISEL-NEXT: v_mul_hi_u32 v13, v7, v5 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v14, v4 +; GISEL-NEXT: v_cndmask_b32_e64 v14, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13 +; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v13, vcc, v14, v13 +; GISEL-NEXT: v_mul_hi_u32 v5, v8, v5 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12 +; GISEL-NEXT: v_cndmask_b32_e64 v12, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v12, vcc, v13, v12 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v12 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v7, v4 +; GISEL-NEXT: v_addc_u32_e32 v7, vcc, v8, v5, vcc +; GISEL-NEXT: v_mul_lo_u32 v5, 0, v4 +; GISEL-NEXT: v_mul_lo_u32 v8, v2, v7 +; GISEL-NEXT: v_mul_hi_u32 v13, v2, v4 +; GISEL-NEXT: v_sub_i32_e32 v1, vcc, v0, v1 +; GISEL-NEXT: v_subbrev_u32_e32 v12, vcc, 0, v11, vcc +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v8 +; GISEL-NEXT: v_mul_lo_u32 v8, 0, v7 +; GISEL-NEXT: v_mul_hi_u32 v4, 0, v4 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v13 +; GISEL-NEXT: v_mul_hi_u32 v13, v2, v7 ; GISEL-NEXT: v_cndmask_b32_e64 v5, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v9, v0 -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v0, v6 -; GISEL-NEXT: v_cndmask_b32_e64 v6, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v9, vcc, v0, v5 -; GISEL-NEXT: v_mul_hi_u32 v11, 0, v4 -; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v9, 0 -; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v0, vcc, v6, v0 -; GISEL-NEXT: v_add_i32_e32 v6, vcc, v11, v0 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4 +; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v13 +; GISEL-NEXT: v_add_i32_e32 v8, vcc, v4, v5 +; GISEL-NEXT: v_mad_u64_u32 v[4:5], s[4:5], v3, v8, 0 +; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v6 +; GISEL-NEXT: v_mul_hi_u32 v6, 0, v7 +; GISEL-NEXT: v_cndmask_b32_e32 v1, v0, v1, vcc ; GISEL-NEXT: v_mov_b32_e32 v0, v5 +; GISEL-NEXT: v_cndmask_b32_e32 v7, v11, v12, vcc ; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], v3, v6, v[0:1] -; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v14 -; GISEL-NEXT: v_cndmask_b32_e32 v0, v7, v1, vcc -; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v9, v[5:6] -; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v10, vcc +; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v9 +; GISEL-NEXT: v_cndmask_b32_e32 v0, v10, v1, vcc +; GISEL-NEXT: v_mad_u64_u32 v[5:6], s[4:5], 0, v8, v[5:6] +; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v7, vcc ; GISEL-NEXT: v_sub_i32_e32 v2, vcc, v2, v4 ; GISEL-NEXT: v_subb_u32_e64 v4, s[4:5], 0, v5, vcc ; GISEL-NEXT: v_sub_i32_e64 v5, s[4:5], 0, v5 diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll index 4de1078..ded985e 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/udiv.i64.ll @@ -2053,90 +2053,82 @@ define <2 x i64> @v_udiv_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) { ; GISEL-NEXT: v_mul_lo_u32 v12, v6, v2 ; GISEL-NEXT: v_mul_lo_u32 v13, 0, v2 ; GISEL-NEXT: v_mul_hi_u32 v14, v6, v2 -; GISEL-NEXT: v_mul_hi_u32 v2, 0, v2 -; GISEL-NEXT: v_mul_lo_u32 v15, v0, v5 +; GISEL-NEXT: v_mul_hi_u32 v15, 0, v2 +; GISEL-NEXT: v_mul_lo_u32 v2, v0, v5 ; GISEL-NEXT: v_mul_lo_u32 v16, 0, v5 ; GISEL-NEXT: v_mul_hi_u32 v17, v0, v5 ; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5 ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12 ; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8 -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15 +; GISEL-NEXT: v_mul_lo_u32 v12, v3, v15 +; GISEL-NEXT: v_add_i32_e32 v2, vcc, v10, v2 ; GISEL-NEXT: v_add_i32_e32 v7, vcc, v16, v7 +; GISEL-NEXT: v_mul_lo_u32 v10, v1, v5 ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9 ; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v14 -; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11 -; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc +; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v11 +; GISEL-NEXT: v_cndmask_b32_e64 v2, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17 -; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10 -; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8 -; GISEL-NEXT: v_mul_lo_u32 v9, v3, v4 -; GISEL-NEXT: v_mul_lo_u32 v12, 0, v4 -; GISEL-NEXT: v_mul_hi_u32 v13, v3, v4 -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10 -; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7 -; GISEL-NEXT: v_mul_lo_u32 v14, 0, v7 -; GISEL-NEXT: v_mul_hi_u32 v15, v1, v7 -; GISEL-NEXT: v_add_i32_e32 v8, vcc, v2, v8 -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10 -; GISEL-NEXT: v_mul_lo_u32 v2, v3, v8 -; GISEL-NEXT: v_add_i32_e32 v10, vcc, 1, v4 -; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v8, vcc -; GISEL-NEXT: v_mul_lo_u32 v17, v1, v5 -; GISEL-NEXT: v_add_i32_e32 v18, vcc, 1, v7 +; GISEL-NEXT: v_add_i32_e32 v2, vcc, v7, v2 +; GISEL-NEXT: v_mul_lo_u32 v7, v3, v4 +; GISEL-NEXT: v_mul_lo_u32 v8, 0, v4 +; GISEL-NEXT: v_mul_hi_u32 v9, v3, v4 +; GISEL-NEXT: v_add_i32_e32 v11, vcc, 1, v4 +; GISEL-NEXT: v_addc_u32_e32 v13, vcc, 0, v15, vcc +; GISEL-NEXT: v_mul_lo_u32 v14, v1, v2 +; GISEL-NEXT: v_mul_lo_u32 v16, 0, v2 +; GISEL-NEXT: v_mul_hi_u32 v17, v1, v2 +; GISEL-NEXT: v_add_i32_e32 v18, vcc, 1, v2 ; GISEL-NEXT: v_addc_u32_e32 v19, vcc, 0, v5, vcc -; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2 -; GISEL-NEXT: v_add_i32_e32 v12, vcc, v14, v17 -; GISEL-NEXT: v_add_i32_e32 v14, vcc, 1, v10 -; GISEL-NEXT: v_addc_u32_e32 v17, vcc, 0, v16, vcc -; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v13 -; GISEL-NEXT: v_add_i32_e32 v12, vcc, v12, v15 -; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v9 -; GISEL-NEXT: v_subb_u32_e64 v9, s[4:5], 0, v2, vcc +; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v12 +; GISEL-NEXT: v_add_i32_e32 v10, vcc, v16, v10 +; GISEL-NEXT: v_add_i32_e32 v12, vcc, 1, v11 +; GISEL-NEXT: v_addc_u32_e32 v16, vcc, 0, v13, vcc +; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v9 +; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v17 +; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v7 +; GISEL-NEXT: v_subb_u32_e64 v7, s[4:5], 0, v8, vcc ; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v6, v3 -; GISEL-NEXT: v_cndmask_b32_e64 v13, 0, -1, s[4:5] -; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v11 -; GISEL-NEXT: v_subb_u32_e64 v11, s[6:7], 0, v12, s[4:5] +; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, -1, s[4:5] +; GISEL-NEXT: v_sub_i32_e64 v0, s[4:5], v0, v14 +; GISEL-NEXT: v_subb_u32_e64 v14, s[6:7], 0, v9, s[4:5] ; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v0, v1 -; GISEL-NEXT: v_cndmask_b32_e64 v15, 0, -1, s[6:7] -; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v9 -; GISEL-NEXT: v_cmp_eq_u32_e64 s[8:9], 0, v11 -; GISEL-NEXT: v_add_i32_e64 v9, s[10:11], 1, v18 -; GISEL-NEXT: v_addc_u32_e64 v11, s[10:11], 0, v19, s[10:11] -; GISEL-NEXT: v_sub_i32_e64 v2, s[10:11], 0, v2 -; GISEL-NEXT: v_sub_i32_e64 v12, s[10:11], 0, v12 -; GISEL-NEXT: v_cndmask_b32_e64 v13, -1, v13, s[6:7] -; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc -; GISEL-NEXT: v_cndmask_b32_e64 v15, -1, v15, s[8:9] -; GISEL-NEXT: v_subbrev_u32_e64 v12, vcc, 0, v12, s[4:5] +; GISEL-NEXT: v_cndmask_b32_e64 v17, 0, -1, s[6:7] +; GISEL-NEXT: v_cmp_eq_u32_e64 s[6:7], 0, v7 +; GISEL-NEXT: v_cmp_eq_u32_e64 s[8:9], 0, v14 +; GISEL-NEXT: v_add_i32_e64 v7, s[10:11], 1, v18 +; GISEL-NEXT: v_addc_u32_e64 v14, s[10:11], 0, v19, s[10:11] +; GISEL-NEXT: v_sub_i32_e64 v8, s[10:11], 0, v8 +; GISEL-NEXT: v_sub_i32_e64 v9, s[10:11], 0, v9 +; GISEL-NEXT: v_cndmask_b32_e64 v10, -1, v10, s[6:7] +; GISEL-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v8, vcc +; GISEL-NEXT: v_cndmask_b32_e64 v17, -1, v17, s[8:9] +; GISEL-NEXT: v_subbrev_u32_e64 v9, vcc, 0, v9, s[4:5] ; GISEL-NEXT: v_sub_i32_e32 v6, vcc, v6, v3 -; GISEL-NEXT: v_subbrev_u32_e32 v2, vcc, 0, v2, vcc +; GISEL-NEXT: v_subbrev_u32_e32 v8, vcc, 0, v8, vcc ; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v6, v3 ; GISEL-NEXT: v_cndmask_b32_e64 v3, 0, -1, vcc ; GISEL-NEXT: v_sub_i32_e32 v0, vcc, v0, v1 -; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v12, vcc +; GISEL-NEXT: v_subbrev_u32_e32 v6, vcc, 0, v9, vcc ; GISEL-NEXT: v_cmp_ge_u32_e32 vcc, v0, v1 ; GISEL-NEXT: v_cndmask_b32_e64 v0, 0, -1, vcc -; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v2 +; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v8 ; GISEL-NEXT: v_cndmask_b32_e32 v1, -1, v3, vcc ; GISEL-NEXT: v_cmp_eq_u32_e32 vcc, 0, v6 ; GISEL-NEXT: v_cndmask_b32_e32 v0, -1, v0, vcc ; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v1 -; GISEL-NEXT: v_cndmask_b32_e32 v1, v10, v14, vcc +; GISEL-NEXT: v_cndmask_b32_e32 v1, v11, v12, vcc ; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v0 -; GISEL-NEXT: v_cndmask_b32_e64 v2, v18, v9, s[4:5] -; GISEL-NEXT: v_cndmask_b32_e32 v3, v16, v17, vcc -; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v13 +; GISEL-NEXT: v_cndmask_b32_e64 v3, v18, v7, s[4:5] +; GISEL-NEXT: v_cndmask_b32_e32 v6, v13, v16, vcc +; GISEL-NEXT: v_cmp_ne_u32_e32 vcc, 0, v10 ; GISEL-NEXT: v_cndmask_b32_e32 v0, v4, v1, vcc -; GISEL-NEXT: v_cndmask_b32_e64 v4, v19, v11, s[4:5] -; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v15 -; GISEL-NEXT: v_cndmask_b32_e64 v2, v7, v2, s[4:5] -; GISEL-NEXT: v_cndmask_b32_e32 v1, v8, v3, vcc +; GISEL-NEXT: v_cndmask_b32_e64 v4, v19, v14, s[4:5] +; GISEL-NEXT: v_cmp_ne_u32_e64 s[4:5], 0, v17 +; GISEL-NEXT: v_cndmask_b32_e64 v2, v2, v3, s[4:5] +; GISEL-NEXT: v_cndmask_b32_e32 v1, v15, v6, vcc ; GISEL-NEXT: v_cndmask_b32_e64 v3, v5, v4, s[4:5] ; GISEL-NEXT: s_setpc_b64 s[30:31] ; diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll index 2b54123..f5068f5 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/umed3.ll @@ -1,7 +1,7 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s -; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s -; RUN: llc -global-isel -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx802 < %s | FileCheck -check-prefixes=GFX89,GFX8 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX89,GFX9 %s +; RUN: llc -global-isel -new-reg-bank-select -mtriple=amdgcn-amd-mesa3d -mcpu=gfx1010 < %s | FileCheck -check-prefix=GFX10 %s define i32 @test_min_max_ValK0_K1_u32(i32 %a) { ; GFX89-LABEL: test_min_max_ValK0_K1_u32: diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll index a41ec8e..be5543b 100644 --- a/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll +++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/urem.i64.ll @@ -2058,42 +2058,34 @@ define <2 x i64> @v_urem_v2i64_24bit(<2 x i64> %num, <2 x i64> %den) { ; GISEL-NEXT: v_mul_hi_u32 v5, 0, v5 ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v12 ; GISEL-NEXT: v_add_i32_e32 v8, vcc, v13, v8 +; GISEL-NEXT: v_mul_lo_u32 v2, v3, v2 ; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v15 ; GISEL-NEXT: v_add_i32_e32 v7, vcc, v16, v7 +; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5 ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v4, v9 ; GISEL-NEXT: v_cndmask_b32_e64 v4, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v8, vcc, v8, v14 +; GISEL-NEXT: v_add_i32_e32 v9, vcc, v10, v11 ; GISEL-NEXT: v_cndmask_b32_e64 v9, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v10, v11 -; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v17 -; GISEL-NEXT: v_cndmask_b32_e64 v11, 0, 1, vcc ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v8, v4 -; GISEL-NEXT: v_cndmask_b32_e64 v8, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v10 -; GISEL-NEXT: v_cndmask_b32_e64 v10, 0, 1, vcc -; GISEL-NEXT: v_add_i32_e32 v8, vcc, v9, v8 -; GISEL-NEXT: v_mul_lo_u32 v9, v3, v4 -; GISEL-NEXT: v_mul_lo_u32 v12, 0, v4 +; GISEL-NEXT: v_add_i32_e32 v7, vcc, v7, v9 +; GISEL-NEXT: v_mul_lo_u32 v8, v3, v4 +; GISEL-NEXT: v_mul_lo_u32 v9, 0, v4 ; GISEL-NEXT: v_mul_hi_u32 v4, v3, v4 -; GISEL-NEXT: v_add_i32_e32 v10, vcc, v11, v10 -; GISEL-NEXT: v_mul_lo_u32 v11, v1, v7 -; GISEL-NEXT: v_mul_lo_u32 v13, 0, v7 +; GISEL-NEXT: v_mul_lo_u32 v10, v1, v7 +; GISEL-NEXT: v_mul_lo_u32 v11, 0, v7 ; GISEL-NEXT: v_mul_hi_u32 v7, v1, v7 -; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v8 -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v5, v10 -; GISEL-NEXT: v_mul_lo_u32 v2, v3, v2 -; GISEL-NEXT: v_mul_lo_u32 v5, v1, v5 -; GISEL-NEXT: v_add_i32_e32 v2, vcc, v12, v2 -; GISEL-NEXT: v_add_i32_e32 v5, vcc, v13, v5 +; GISEL-NEXT: v_add_i32_e32 v2, vcc, v9, v2 +; GISEL-NEXT: v_add_i32_e32 v5, vcc, v11, v5 ; GISEL-NEXT: v_add_i32_e32 v2, vcc, v2, v4 ; GISEL-NEXT: v_add_i32_e32 v4, vcc, v5, v7 -; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v9 +; GISEL-NEXT: v_sub_i32_e32 v5, vcc, v6, v8 ; GISEL-NEXT: v_subb_u32_e64 v6, s[4:5], 0, v2, vcc ; GISEL-NEXT: v_sub_i32_e64 v2, s[4:5], 0, v2 ; GISEL-NEXT: v_cmp_ge_u32_e64 s[4:5], v5, v3 ; GISEL-NEXT: v_cndmask_b32_e64 v7, 0, -1, s[4:5] -; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v11 +; GISEL-NEXT: v_sub_i32_e64 v8, s[4:5], v0, v10 ; GISEL-NEXT: v_subb_u32_e64 v9, s[6:7], 0, v4, s[4:5] ; GISEL-NEXT: v_sub_i32_e64 v0, s[6:7], 0, v4 ; GISEL-NEXT: v_cmp_ge_u32_e64 s[6:7], v8, v1 diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-break-large-phis-heuristics.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-break-large-phis-heuristics.ll index c94b333..1f36902 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-break-large-phis-heuristics.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-break-large-phis-heuristics.ll @@ -726,16 +726,16 @@ define amdgpu_kernel void @used_by_unbreakable_and_breakable_phi(<5 x double> %i ; CHECK-NEXT: [[LARGEPHI_EXTRACTSLICE815:%.*]] = extractelement <5 x double> [[LARGEPHI_INSERTSLICE4]], i64 4 ; CHECK-NEXT: br label [[END]] ; CHECK: end: -; CHECK-NEXT: [[TMP5:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE01]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE1]], [[FINALLY]] ] -; CHECK-NEXT: [[TMP6:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE22]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE3]], [[FINALLY]] ] -; CHECK-NEXT: [[TMP7:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE43]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE5]], [[FINALLY]] ] -; CHECK-NEXT: [[TMP8:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE64]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE7]], [[FINALLY]] ] -; CHECK-NEXT: [[TMP9:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE85]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE9]], [[FINALLY]] ] -; CHECK-NEXT: [[TMP10:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE011]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] -; CHECK-NEXT: [[TMP11:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE212]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] -; CHECK-NEXT: [[TMP12:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE413]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] -; CHECK-NEXT: [[TMP13:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE614]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] -; CHECK-NEXT: [[TMP14:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE815]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] +; CHECK-NEXT: [[TMP5:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE01]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] +; CHECK-NEXT: [[TMP6:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE22]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] +; CHECK-NEXT: [[TMP7:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE43]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] +; CHECK-NEXT: [[TMP8:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE64]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] +; CHECK-NEXT: [[TMP9:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE85]], [[THEN1]] ], [ 0.000000e+00, [[FINALLY]] ] +; CHECK-NEXT: [[TMP10:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE011]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE1]], [[FINALLY]] ] +; CHECK-NEXT: [[TMP11:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE212]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE3]], [[FINALLY]] ] +; CHECK-NEXT: [[TMP12:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE413]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE5]], [[FINALLY]] ] +; CHECK-NEXT: [[TMP13:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE614]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE7]], [[FINALLY]] ] +; CHECK-NEXT: [[TMP14:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE815]], [[THEN1]] ], [ [[LARGEPHI_EXTRACTSLICE9]], [[FINALLY]] ] ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE016:%.*]] = insertelement <5 x double> poison, double [[TMP10]], i64 0 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE117:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE016]], double [[TMP11]], i64 1 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE218:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE117]], double [[TMP12]], i64 2 @@ -746,8 +746,8 @@ define amdgpu_kernel void @used_by_unbreakable_and_breakable_phi(<5 x double> %i ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE28:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE17]], double [[TMP7]], i64 2 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE39:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE28]], double [[TMP8]], i64 3 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE410:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE39]], double [[TMP9]], i64 4 -; CHECK-NEXT: store <5 x double> [[LARGEPHI_INSERTSLICE410]], ptr [[OUT]], align 1 ; CHECK-NEXT: store <5 x double> [[LARGEPHI_INSERTSLICE420]], ptr [[OUT]], align 1 +; CHECK-NEXT: store <5 x double> [[LARGEPHI_INSERTSLICE410]], ptr [[OUT]], align 1 ; CHECK-NEXT: ret void ; entry: @@ -1187,11 +1187,11 @@ define amdgpu_kernel void @test_breakable_chain_5_out_of_7(<5 x double> %in, ptr ; CHECK-NEXT: [[LARGEPHI_EXTRACTSLICE960:%.*]] = extractelement <5 x double> [[IN]], i64 4 ; CHECK-NEXT: br i1 [[COND]], label [[END:%.*]], label [[COND5_END]] ; CHECK: cond5.end: -; CHECK-NEXT: [[TMP25:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE041]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE1]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP26:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE242]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE3]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP27:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE443]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE5]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP28:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE644]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE7]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP29:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE845]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE9]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP25:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE041]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE152]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP26:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE242]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE354]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP27:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE443]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE556]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP28:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE644]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE758]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP29:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE845]], [[COND4_END]] ], [ [[LARGEPHI_EXTRACTSLICE960]], [[COND5_TRUE]] ] ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE046:%.*]] = insertelement <5 x double> poison, double [[TMP25]], i64 0 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE147:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE046]], double [[TMP26]], i64 1 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE248:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE147]], double [[TMP27]], i64 2 @@ -1204,11 +1204,11 @@ define amdgpu_kernel void @test_breakable_chain_5_out_of_7(<5 x double> %in, ptr ; CHECK-NEXT: [[LARGEPHI_EXTRACTSLICE859:%.*]] = extractelement <5 x double> [[LARGEPHI_INSERTSLICE450]], i64 4 ; CHECK-NEXT: br label [[END]] ; CHECK: end: -; CHECK-NEXT: [[TMP30:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE051]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE152]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP31:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE253]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE354]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP32:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE455]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE556]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP33:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE657]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE758]], [[COND5_TRUE]] ] -; CHECK-NEXT: [[TMP34:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE859]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE960]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP30:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE051]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE1]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP31:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE253]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE3]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP32:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE455]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE5]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP33:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE657]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE7]], [[COND5_TRUE]] ] +; CHECK-NEXT: [[TMP34:%.*]] = phi double [ [[LARGEPHI_EXTRACTSLICE859]], [[COND5_END]] ], [ [[LARGEPHI_EXTRACTSLICE9]], [[COND5_TRUE]] ] ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE061:%.*]] = insertelement <5 x double> poison, double [[TMP30]], i64 0 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE162:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE061]], double [[TMP31]], i64 1 ; CHECK-NEXT: [[LARGEPHI_INSERTSLICE263:%.*]] = insertelement <5 x double> [[LARGEPHI_INSERTSLICE162]], double [[TMP32]], i64 2 diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll index a4f9ce3..7ff86ac 100644 --- a/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll +++ b/llvm/test/CodeGen/AMDGPU/amdgpu-codegenprepare-fdiv.ll @@ -2160,7 +2160,22 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-GOODFREXP-NEXT: [[TMP38:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i64 0 ; IEEE-GOODFREXP-NEXT: [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP38]], float [[TMP37]], i64 1 ; IEEE-GOODFREXP-NEXT: store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3:![0-9]+]] +; IEEE-GOODFREXP-NEXT: [[TMP56:%.*]] = extractelement <2 x float> [[X]], i64 0 +; IEEE-GOODFREXP-NEXT: [[TMP57:%.*]] = extractelement <2 x float> [[X]], i64 1 +; IEEE-GOODFREXP-NEXT: [[TMP58:%.*]] = fcmp olt float [[TMP56]], 0x3810000000000000 +; IEEE-GOODFREXP-NEXT: [[TMP59:%.*]] = select i1 [[TMP58]], i32 32, i32 0 +; IEEE-GOODFREXP-NEXT: [[TMP60:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP56]], i32 [[TMP59]]) +; IEEE-GOODFREXP-NEXT: [[TMP61:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP60]]) +; IEEE-GOODFREXP-NEXT: [[TMP62:%.*]] = select i1 [[TMP58]], i32 -16, i32 0 +; IEEE-GOODFREXP-NEXT: [[TMP63:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP61]], i32 [[TMP62]]) +; IEEE-GOODFREXP-NEXT: [[TMP64:%.*]] = fcmp olt float [[TMP57]], 0x3810000000000000 +; IEEE-GOODFREXP-NEXT: [[TMP65:%.*]] = select i1 [[TMP64]], i32 32, i32 0 +; IEEE-GOODFREXP-NEXT: [[TMP66:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP57]], i32 [[TMP65]]) +; IEEE-GOODFREXP-NEXT: [[TMP67:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP66]]) +; IEEE-GOODFREXP-NEXT: [[TMP68:%.*]] = select i1 [[TMP64]], i32 -16, i32 0 +; IEEE-GOODFREXP-NEXT: [[TMP69:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP67]], i32 [[TMP68]]) +; IEEE-GOODFREXP-NEXT: [[TMP70:%.*]] = insertelement <2 x float> poison, float [[TMP63]], i64 0 +; IEEE-GOODFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = insertelement <2 x float> [[TMP70]], float [[TMP69]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP39:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0 ; IEEE-GOODFREXP-NEXT: [[TMP40:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP41:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2231,7 +2246,22 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; IEEE-BADFREXP-NEXT: [[TMP38:%.*]] = insertelement <2 x float> poison, float [[TMP27]], i64 0 ; IEEE-BADFREXP-NEXT: [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP38]], float [[TMP37]], i64 1 ; IEEE-BADFREXP-NEXT: store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4 -; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3:![0-9]+]] +; IEEE-BADFREXP-NEXT: [[TMP56:%.*]] = extractelement <2 x float> [[X]], i64 0 +; IEEE-BADFREXP-NEXT: [[TMP57:%.*]] = extractelement <2 x float> [[X]], i64 1 +; IEEE-BADFREXP-NEXT: [[TMP58:%.*]] = fcmp olt float [[TMP56]], 0x3810000000000000 +; IEEE-BADFREXP-NEXT: [[TMP59:%.*]] = select i1 [[TMP58]], i32 32, i32 0 +; IEEE-BADFREXP-NEXT: [[TMP60:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP56]], i32 [[TMP59]]) +; IEEE-BADFREXP-NEXT: [[TMP61:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP60]]) +; IEEE-BADFREXP-NEXT: [[TMP62:%.*]] = select i1 [[TMP58]], i32 -16, i32 0 +; IEEE-BADFREXP-NEXT: [[TMP63:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP61]], i32 [[TMP62]]) +; IEEE-BADFREXP-NEXT: [[TMP64:%.*]] = fcmp olt float [[TMP57]], 0x3810000000000000 +; IEEE-BADFREXP-NEXT: [[TMP65:%.*]] = select i1 [[TMP64]], i32 32, i32 0 +; IEEE-BADFREXP-NEXT: [[TMP66:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP57]], i32 [[TMP65]]) +; IEEE-BADFREXP-NEXT: [[TMP67:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP66]]) +; IEEE-BADFREXP-NEXT: [[TMP68:%.*]] = select i1 [[TMP64]], i32 -16, i32 0 +; IEEE-BADFREXP-NEXT: [[TMP69:%.*]] = call float @llvm.ldexp.f32.i32(float [[TMP67]], i32 [[TMP68]]) +; IEEE-BADFREXP-NEXT: [[TMP70:%.*]] = insertelement <2 x float> poison, float [[TMP63]], i64 0 +; IEEE-BADFREXP-NEXT: [[SQRT_X_3ULP:%.*]] = insertelement <2 x float> [[TMP70]], float [[TMP69]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP39:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0 ; IEEE-BADFREXP-NEXT: [[TMP40:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP41:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2258,7 +2288,12 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; DAZ-NEXT: [[SQRT_X_NO_MD:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]) ; DAZ-NEXT: [[NO_MD:%.*]] = fdiv contract <2 x float> splat (float 1.000000e+00), [[SQRT_X_NO_MD]] ; DAZ-NEXT: store volatile <2 x float> [[NO_MD]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[SQRT_MD_1ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META2:![0-9]+]] +; DAZ-NEXT: [[TMP39:%.*]] = extractelement <2 x float> [[X]], i64 0 +; DAZ-NEXT: [[TMP40:%.*]] = extractelement <2 x float> [[X]], i64 1 +; DAZ-NEXT: [[TMP41:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP39]]) +; DAZ-NEXT: [[TMP42:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP40]]) +; DAZ-NEXT: [[TMP43:%.*]] = insertelement <2 x float> poison, float [[TMP41]], i64 0 +; DAZ-NEXT: [[SQRT_MD_1ULP:%.*]] = insertelement <2 x float> [[TMP43]], float [[TMP42]], i64 1 ; DAZ-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 0 ; DAZ-NEXT: [[TMP2:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP]], i64 1 ; DAZ-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -2276,7 +2311,9 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; DAZ-NEXT: [[SQRT_MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP12]], float [[TMP11]], i64 1 ; DAZ-NEXT: [[TMP13:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 0 ; DAZ-NEXT: [[TMP14:%.*]] = extractelement <2 x float> [[SQRT_MD_1ULP_UNDEF]], i64 1 -; DAZ-NEXT: [[TMP15:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP13]]) +; DAZ-NEXT: [[TMP44:%.*]] = extractelement <2 x float> [[X]], i64 0 +; DAZ-NEXT: [[TMP45:%.*]] = extractelement <2 x float> [[X]], i64 1 +; DAZ-NEXT: [[TMP15:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float [[TMP44]]) ; DAZ-NEXT: [[TMP16:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP14]]) ; DAZ-NEXT: [[TMP17:%.*]] = extractvalue { float, i32 } [[TMP16]], 0 ; DAZ-NEXT: [[TMP18:%.*]] = extractvalue { float, i32 } [[TMP16]], 1 @@ -2290,7 +2327,12 @@ define amdgpu_kernel void @rsq_f32_vector_fpmath(ptr addrspace(1) %out, <2 x flo ; DAZ-NEXT: [[TMP26:%.*]] = insertelement <2 x float> poison, float [[TMP15]], i64 0 ; DAZ-NEXT: [[MD_1ULP_UNDEF:%.*]] = insertelement <2 x float> [[TMP26]], float [[TMP25]], i64 1 ; DAZ-NEXT: store volatile <2 x float> [[MD_1ULP_UNDEF]], ptr addrspace(1) [[OUT]], align 4 -; DAZ-NEXT: [[SQRT_X_3ULP:%.*]] = call contract <2 x float> @llvm.sqrt.v2f32(<2 x float> [[X]]), !fpmath [[META3:![0-9]+]] +; DAZ-NEXT: [[TMP34:%.*]] = extractelement <2 x float> [[X]], i64 0 +; DAZ-NEXT: [[TMP35:%.*]] = extractelement <2 x float> [[X]], i64 1 +; DAZ-NEXT: [[TMP36:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP34]]) +; DAZ-NEXT: [[TMP37:%.*]] = call float @llvm.amdgcn.sqrt.f32(float [[TMP35]]) +; DAZ-NEXT: [[TMP38:%.*]] = insertelement <2 x float> poison, float [[TMP36]], i64 0 +; DAZ-NEXT: [[SQRT_X_3ULP:%.*]] = insertelement <2 x float> [[TMP38]], float [[TMP37]], i64 1 ; DAZ-NEXT: [[TMP27:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 0 ; DAZ-NEXT: [[TMP28:%.*]] = extractelement <2 x float> [[SQRT_X_3ULP]], i64 1 ; DAZ-NEXT: [[TMP29:%.*]] = extractelement <2 x float> [[X]], i64 0 @@ -3200,9 +3242,13 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator(<4 x float> %arg) { ; DAZ-NEXT: [[TMP13:%.*]] = extractelement <4 x float> [[DENOM]], i64 1 ; DAZ-NEXT: [[TMP14:%.*]] = extractelement <4 x float> [[DENOM]], i64 2 ; DAZ-NEXT: [[TMP15:%.*]] = extractelement <4 x float> [[DENOM]], i64 3 -; DAZ-NEXT: [[TMP16:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP12]]) -; DAZ-NEXT: [[TMP17:%.*]] = fneg contract float [[TMP13]] -; DAZ-NEXT: [[TMP18:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP17]]) +; DAZ-NEXT: [[TMP42:%.*]] = extractelement <4 x float> [[ARG]], i64 0 +; DAZ-NEXT: [[TMP17:%.*]] = extractelement <4 x float> [[ARG]], i64 1 +; DAZ-NEXT: [[TMP43:%.*]] = extractelement <4 x float> [[ARG]], i64 2 +; DAZ-NEXT: [[TMP44:%.*]] = extractelement <4 x float> [[ARG]], i64 3 +; DAZ-NEXT: [[TMP16:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float [[TMP42]]) +; DAZ-NEXT: [[TMP45:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float [[TMP17]]) +; DAZ-NEXT: [[TMP18:%.*]] = fneg contract float [[TMP45]] ; DAZ-NEXT: [[TMP19:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP14]]) ; DAZ-NEXT: [[TMP20:%.*]] = extractvalue { float, i32 } [[TMP19]], 0 ; DAZ-NEXT: [[TMP21:%.*]] = extractvalue { float, i32 } [[TMP19]], 1 @@ -3675,9 +3721,13 @@ define <4 x float> @rsq_f32_vector_mixed_constant_numerator_arcp(<4 x float> %ar ; DAZ-NEXT: [[TMP13:%.*]] = extractelement <4 x float> [[DENOM]], i64 1 ; DAZ-NEXT: [[TMP14:%.*]] = extractelement <4 x float> [[DENOM]], i64 2 ; DAZ-NEXT: [[TMP15:%.*]] = extractelement <4 x float> [[DENOM]], i64 3 -; DAZ-NEXT: [[TMP16:%.*]] = call arcp contract float @llvm.amdgcn.rcp.f32(float [[TMP12]]) -; DAZ-NEXT: [[TMP17:%.*]] = fneg arcp contract float [[TMP13]] -; DAZ-NEXT: [[TMP18:%.*]] = call arcp contract float @llvm.amdgcn.rcp.f32(float [[TMP17]]) +; DAZ-NEXT: [[TMP26:%.*]] = extractelement <4 x float> [[ARG]], i64 0 +; DAZ-NEXT: [[TMP17:%.*]] = extractelement <4 x float> [[ARG]], i64 1 +; DAZ-NEXT: [[TMP27:%.*]] = extractelement <4 x float> [[ARG]], i64 2 +; DAZ-NEXT: [[TMP28:%.*]] = extractelement <4 x float> [[ARG]], i64 3 +; DAZ-NEXT: [[TMP16:%.*]] = call arcp contract float @llvm.amdgcn.rsq.f32(float [[TMP26]]) +; DAZ-NEXT: [[TMP29:%.*]] = call arcp contract float @llvm.amdgcn.rsq.f32(float [[TMP17]]) +; DAZ-NEXT: [[TMP18:%.*]] = fneg arcp contract float [[TMP29]] ; DAZ-NEXT: [[TMP19:%.*]] = call arcp contract float @llvm.amdgcn.rcp.f32(float [[TMP14]]) ; DAZ-NEXT: [[TMP20:%.*]] = fmul arcp contract float 4.000000e+00, [[TMP19]] ; DAZ-NEXT: [[TMP21:%.*]] = call arcp contract float @llvm.amdgcn.rcp.f32(float [[TMP15]]) @@ -3850,19 +3900,9 @@ define <4 x float> @rsq_f32_vector_const_denom(ptr addrspace(1) %out, <2 x float ; IEEE-GOODFREXP-NEXT: [[TMP9:%.*]] = extractelement <4 x float> [[SQRT]], i64 1 ; IEEE-GOODFREXP-NEXT: [[TMP10:%.*]] = extractelement <4 x float> [[SQRT]], i64 2 ; IEEE-GOODFREXP-NEXT: [[TMP11:%.*]] = extractelement <4 x float> [[SQRT]], i64 3 -; IEEE-GOODFREXP-NEXT: [[TMP12:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP8]]) -; IEEE-GOODFREXP-NEXT: [[TMP13:%.*]] = extractvalue { float, i32 } [[TMP12]], 0 -; IEEE-GOODFREXP-NEXT: [[TMP14:%.*]] = extractvalue { float, i32 } [[TMP12]], 1 -; IEEE-GOODFREXP-NEXT: [[TMP15:%.*]] = sub i32 0, [[TMP14]] -; IEEE-GOODFREXP-NEXT: [[TMP16:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP13]]) -; IEEE-GOODFREXP-NEXT: [[TMP17:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP16]], i32 [[TMP15]]) -; IEEE-GOODFREXP-NEXT: [[TMP18:%.*]] = fneg contract float [[TMP9]] -; IEEE-GOODFREXP-NEXT: [[TMP48:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP18]]) -; IEEE-GOODFREXP-NEXT: [[TMP49:%.*]] = extractvalue { float, i32 } [[TMP48]], 0 -; IEEE-GOODFREXP-NEXT: [[TMP50:%.*]] = extractvalue { float, i32 } [[TMP48]], 1 -; IEEE-GOODFREXP-NEXT: [[TMP22:%.*]] = sub i32 0, [[TMP50]] -; IEEE-GOODFREXP-NEXT: [[TMP51:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP49]]) -; IEEE-GOODFREXP-NEXT: [[TMP24:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP51]], i32 [[TMP22]]) +; IEEE-GOODFREXP-NEXT: [[TMP17:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float 4.000000e+00) +; IEEE-GOODFREXP-NEXT: [[TMP13:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float 2.000000e+00) +; IEEE-GOODFREXP-NEXT: [[TMP24:%.*]] = fneg contract float [[TMP13]] ; IEEE-GOODFREXP-NEXT: [[TMP29:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) ; IEEE-GOODFREXP-NEXT: [[TMP30:%.*]] = extractvalue { float, i32 } [[TMP29]], 0 ; IEEE-GOODFREXP-NEXT: [[TMP31:%.*]] = extractvalue { float, i32 } [[TMP29]], 1 @@ -3903,19 +3943,9 @@ define <4 x float> @rsq_f32_vector_const_denom(ptr addrspace(1) %out, <2 x float ; IEEE-BADFREXP-NEXT: [[TMP9:%.*]] = extractelement <4 x float> [[SQRT]], i64 1 ; IEEE-BADFREXP-NEXT: [[TMP10:%.*]] = extractelement <4 x float> [[SQRT]], i64 2 ; IEEE-BADFREXP-NEXT: [[TMP11:%.*]] = extractelement <4 x float> [[SQRT]], i64 3 -; IEEE-BADFREXP-NEXT: [[TMP12:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP8]]) -; IEEE-BADFREXP-NEXT: [[TMP13:%.*]] = extractvalue { float, i32 } [[TMP12]], 0 -; IEEE-BADFREXP-NEXT: [[TMP14:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[TMP8]]) -; IEEE-BADFREXP-NEXT: [[TMP15:%.*]] = sub i32 0, [[TMP14]] -; IEEE-BADFREXP-NEXT: [[TMP16:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP13]]) -; IEEE-BADFREXP-NEXT: [[TMP17:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP16]], i32 [[TMP15]]) -; IEEE-BADFREXP-NEXT: [[TMP18:%.*]] = fneg contract float [[TMP9]] -; IEEE-BADFREXP-NEXT: [[TMP48:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP18]]) -; IEEE-BADFREXP-NEXT: [[TMP49:%.*]] = extractvalue { float, i32 } [[TMP48]], 0 -; IEEE-BADFREXP-NEXT: [[TMP21:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[TMP18]]) -; IEEE-BADFREXP-NEXT: [[TMP22:%.*]] = sub i32 0, [[TMP21]] -; IEEE-BADFREXP-NEXT: [[TMP50:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP49]]) -; IEEE-BADFREXP-NEXT: [[TMP24:%.*]] = call contract float @llvm.ldexp.f32.i32(float [[TMP50]], i32 [[TMP22]]) +; IEEE-BADFREXP-NEXT: [[TMP17:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float 4.000000e+00) +; IEEE-BADFREXP-NEXT: [[TMP13:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float 2.000000e+00) +; IEEE-BADFREXP-NEXT: [[TMP24:%.*]] = fneg contract float [[TMP13]] ; IEEE-BADFREXP-NEXT: [[TMP29:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) ; IEEE-BADFREXP-NEXT: [[TMP30:%.*]] = extractvalue { float, i32 } [[TMP29]], 0 ; IEEE-BADFREXP-NEXT: [[TMP27:%.*]] = call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[TMP10]]) @@ -3956,9 +3986,9 @@ define <4 x float> @rsq_f32_vector_const_denom(ptr addrspace(1) %out, <2 x float ; DAZ-NEXT: [[TMP9:%.*]] = extractelement <4 x float> [[SQRT]], i64 1 ; DAZ-NEXT: [[TMP10:%.*]] = extractelement <4 x float> [[SQRT]], i64 2 ; DAZ-NEXT: [[TMP11:%.*]] = extractelement <4 x float> [[SQRT]], i64 3 -; DAZ-NEXT: [[TMP12:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP8]]) -; DAZ-NEXT: [[TMP13:%.*]] = fneg contract float [[TMP9]] -; DAZ-NEXT: [[TMP14:%.*]] = call contract float @llvm.amdgcn.rcp.f32(float [[TMP13]]) +; DAZ-NEXT: [[TMP12:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float 4.000000e+00) +; DAZ-NEXT: [[TMP13:%.*]] = call contract float @llvm.amdgcn.rsq.f32(float 2.000000e+00) +; DAZ-NEXT: [[TMP14:%.*]] = fneg contract float [[TMP13]] ; DAZ-NEXT: [[TMP15:%.*]] = call { float, i32 } @llvm.frexp.f32.i32(float [[TMP10]]) ; DAZ-NEXT: [[TMP16:%.*]] = extractvalue { float, i32 } [[TMP15]], 0 ; DAZ-NEXT: [[TMP17:%.*]] = extractvalue { float, i32 } [[TMP15]], 1 diff --git a/llvm/test/CodeGen/AMDGPU/coalescer-avoid-coalesce-class-with-no-registers.ll b/llvm/test/CodeGen/AMDGPU/coalescer-avoid-coalesce-class-with-no-registers.ll new file mode 100644 index 0000000..f466513 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/coalescer-avoid-coalesce-class-with-no-registers.ll @@ -0,0 +1,27 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 < %s | FileCheck %s + +; Make sure the coalescer doesn't introduce any uses of +; vreg_1024. None are available to allocate with the register budget +; of this function. + +define void @no_introduce_vreg_1024() #0 { +; CHECK-LABEL: no_introduce_vreg_1024: +; CHECK: ; %bb.0: +; CHECK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; def v[0:7] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: v_mov_b32_e32 v9, v0 +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; use v[0:15] +; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: s_setpc_b64 s[30:31] + %tuple = call <8 x i32> asm sideeffect "; def $0","=v"() + %sub0 = extractelement <8 x i32> %tuple, i32 0 + %insert = insertelement <16 x i32> poison, i32 %sub0, i32 9 + call void asm sideeffect "; use $0","v"(<16 x i32> %insert) + ret void +} + +attributes #0 = { nounwind "amdgpu-waves-per-eu"="10,10" } diff --git a/llvm/test/CodeGen/AMDGPU/coalescer-avoid-coalesce-class-with-no-registers.mir b/llvm/test/CodeGen/AMDGPU/coalescer-avoid-coalesce-class-with-no-registers.mir new file mode 100644 index 0000000..1f414eb --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/coalescer-avoid-coalesce-class-with-no-registers.mir @@ -0,0 +1,34 @@ +# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 6 +# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=register-coalescer -o - %s | FileCheck %s + +# The register budget for this function does not permit using 1024-bit +# registers. The coalescer should not introduce a 1024-bit virtual +# register which will fail to allocate. + +--- | + define void @no_introduce_vreg_1024() #0 { + ret void + } + + attributes #0 = { "amdgpu-waves-per-eu"="10,10" } +... +--- +name: no_introduce_vreg_1024 +tracksRegLiveness: true +machineFunctionInfo: + occupancy: 10 +body: | + bb.0: + liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 + + ; CHECK-LABEL: name: no_introduce_vreg_1024 + ; CHECK: liveins: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 + ; CHECK-NEXT: {{ $}} + ; CHECK-NEXT: [[COPY:%[0-9]+]]:vreg_256 = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 + ; CHECK-NEXT: undef [[COPY1:%[0-9]+]].sub9:vreg_512 = COPY [[COPY]].sub0 + ; CHECK-NEXT: SI_RETURN implicit [[COPY1]] + %0:vreg_256 = COPY $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7 + undef %1.sub9:vreg_512 = COPY %0.sub0 + SI_RETURN implicit %1 + +... diff --git a/llvm/test/CodeGen/AMDGPU/fdiv_flags.f32.ll b/llvm/test/CodeGen/AMDGPU/fdiv_flags.f32.ll index 3983655..38239c5 100644 --- a/llvm/test/CodeGen/AMDGPU/fdiv_flags.f32.ll +++ b/llvm/test/CodeGen/AMDGPU/fdiv_flags.f32.ll @@ -1634,29 +1634,18 @@ define float @v_recip_sqrt_f32_ulp25_contract(float %x) { ; IR-IEEE-SDAG-LABEL: v_recip_sqrt_f32_ulp25_contract: ; IR-IEEE-SDAG: ; %bb.0: ; IR-IEEE-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; IR-IEEE-SDAG-NEXT: s_mov_b32 s4, 0xf800000 -; IR-IEEE-SDAG-NEXT: v_mul_f32_e32 v1, 0x4f800000, v0 +; IR-IEEE-SDAG-NEXT: s_mov_b32 s4, 0x800000 ; IR-IEEE-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0 -; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; IR-IEEE-SDAG-NEXT: v_sqrt_f32_e32 v1, v0 -; IR-IEEE-SDAG-NEXT: v_add_i32_e64 v2, s[4:5], -1, v1 -; IR-IEEE-SDAG-NEXT: v_fma_f32 v3, -v2, v1, v0 -; IR-IEEE-SDAG-NEXT: v_cmp_ge_f32_e64 s[4:5], 0, v3 -; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e64 v2, v1, v2, s[4:5] -; IR-IEEE-SDAG-NEXT: v_add_i32_e64 v3, s[4:5], 1, v1 -; IR-IEEE-SDAG-NEXT: v_fma_f32 v1, -v3, v1, v0 -; IR-IEEE-SDAG-NEXT: v_cmp_lt_f32_e64 s[4:5], 0, v1 -; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e64 v1, v2, v3, s[4:5] -; IR-IEEE-SDAG-NEXT: v_mul_f32_e32 v2, 0x37800000, v1 -; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc -; IR-IEEE-SDAG-NEXT: v_mov_b32_e32 v2, 0x260 -; IR-IEEE-SDAG-NEXT: v_cmp_class_f32_e32 vcc, v0, v2 -; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e64 v1, 0, 32, vcc +; IR-IEEE-SDAG-NEXT: v_ldexp_f32_e32 v0, v0, v1 +; IR-IEEE-SDAG-NEXT: v_sqrt_f32_e32 v0, v0 +; IR-IEEE-SDAG-NEXT: v_cndmask_b32_e64 v1, 0, -16, vcc +; IR-IEEE-SDAG-NEXT: v_ldexp_f32_e32 v0, v0, v1 ; IR-IEEE-SDAG-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 ; IR-IEEE-SDAG-NEXT: v_rcp_f32_e32 v2, v1 -; IR-IEEE-SDAG-NEXT: v_fma_f32 v3, -v1, v2, 1.0 -; IR-IEEE-SDAG-NEXT: v_fma_f32 v2, v3, v2, v2 ; IR-IEEE-SDAG-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; IR-IEEE-SDAG-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; IR-IEEE-SDAG-NEXT: v_fma_f32 v2, v4, v2, v2 ; IR-IEEE-SDAG-NEXT: v_mul_f32_e32 v4, v3, v2 ; IR-IEEE-SDAG-NEXT: v_fma_f32 v5, -v1, v4, v3 ; IR-IEEE-SDAG-NEXT: v_fma_f32 v4, v5, v2, v4 @@ -1668,24 +1657,14 @@ define float @v_recip_sqrt_f32_ulp25_contract(float %x) { ; IR-IEEE-GISEL-LABEL: v_recip_sqrt_f32_ulp25_contract: ; IR-IEEE-GISEL: ; %bb.0: ; IR-IEEE-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; IR-IEEE-GISEL-NEXT: v_mov_b32_e32 v1, 0xf800000 -; IR-IEEE-GISEL-NEXT: v_mul_f32_e32 v2, 0x4f800000, v0 +; IR-IEEE-GISEL-NEXT: v_mov_b32_e32 v1, 0x800000 ; IR-IEEE-GISEL-NEXT: v_cmp_lt_f32_e32 vcc, v0, v1 -; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; IR-IEEE-GISEL-NEXT: v_sqrt_f32_e32 v1, v0 -; IR-IEEE-GISEL-NEXT: v_add_i32_e64 v2, s[4:5], -1, v1 -; IR-IEEE-GISEL-NEXT: v_fma_f32 v3, -v2, v1, v0 -; IR-IEEE-GISEL-NEXT: v_add_i32_e64 v4, s[4:5], 1, v1 -; IR-IEEE-GISEL-NEXT: v_fma_f32 v5, -v4, v1, v0 -; IR-IEEE-GISEL-NEXT: v_cmp_ge_f32_e64 s[4:5], 0, v3 -; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e64 v1, v1, v2, s[4:5] -; IR-IEEE-GISEL-NEXT: v_cmp_lt_f32_e64 s[4:5], 0, v5 -; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e64 v1, v1, v4, s[4:5] -; IR-IEEE-GISEL-NEXT: v_mul_f32_e32 v2, 0x37800000, v1 -; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc -; IR-IEEE-GISEL-NEXT: v_mov_b32_e32 v2, 0x260 -; IR-IEEE-GISEL-NEXT: v_cmp_class_f32_e32 vcc, v0, v2 -; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e64 v1, 0, 1, vcc +; IR-IEEE-GISEL-NEXT: v_lshlrev_b32_e32 v1, 5, v1 +; IR-IEEE-GISEL-NEXT: v_ldexp_f32_e32 v0, v0, v1 +; IR-IEEE-GISEL-NEXT: v_sqrt_f32_e32 v0, v0 +; IR-IEEE-GISEL-NEXT: v_cndmask_b32_e64 v1, 0, -16, vcc +; IR-IEEE-GISEL-NEXT: v_ldexp_f32_e32 v0, v0, v1 ; IR-IEEE-GISEL-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 ; IR-IEEE-GISEL-NEXT: v_rcp_f32_e32 v2, v1 ; IR-IEEE-GISEL-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 @@ -1705,75 +1684,24 @@ define float @v_recip_sqrt_f32_ulp25_contract(float %x) { ; CODEGEN-DAZ-NEXT: v_rsq_f32_e32 v0, v0 ; CODEGEN-DAZ-NEXT: s_setpc_b64 s[30:31] ; -; IR-DAZ-SDAG-LABEL: v_recip_sqrt_f32_ulp25_contract: -; IR-DAZ-SDAG: ; %bb.0: -; IR-DAZ-SDAG-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; IR-DAZ-SDAG-NEXT: s_mov_b32 s4, 0xf800000 -; IR-DAZ-SDAG-NEXT: v_mul_f32_e32 v1, 0x4f800000, v0 -; IR-DAZ-SDAG-NEXT: v_cmp_gt_f32_e32 vcc, s4, v0 -; IR-DAZ-SDAG-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc -; IR-DAZ-SDAG-NEXT: v_rsq_f32_e32 v1, v0 -; IR-DAZ-SDAG-NEXT: v_mul_f32_e32 v2, v0, v1 -; IR-DAZ-SDAG-NEXT: v_mul_f32_e32 v1, 0.5, v1 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v3, -v1, v2, 0.5 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v2, v2, v3, v2 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v4, -v2, v2, v0 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v1, v1, v3, v1 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v1, v4, v1, v2 -; IR-DAZ-SDAG-NEXT: v_mul_f32_e32 v2, 0x37800000, v1 -; IR-DAZ-SDAG-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc -; IR-DAZ-SDAG-NEXT: v_mov_b32_e32 v2, 0x260 -; IR-DAZ-SDAG-NEXT: v_cmp_class_f32_e32 vcc, v0, v2 -; IR-DAZ-SDAG-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; IR-DAZ-SDAG-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 -; IR-DAZ-SDAG-NEXT: v_rcp_f32_e32 v2, v1 -; IR-DAZ-SDAG-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 -; IR-DAZ-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v4, -v1, v2, 1.0 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v2, v4, v2, v2 -; IR-DAZ-SDAG-NEXT: v_mul_f32_e32 v4, v3, v2 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v5, -v1, v4, v3 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v4, v5, v2, v4 -; IR-DAZ-SDAG-NEXT: v_fma_f32 v1, -v1, v4, v3 -; IR-DAZ-SDAG-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; IR-DAZ-SDAG-NEXT: v_div_fmas_f32 v1, v1, v2, v4 -; IR-DAZ-SDAG-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 -; IR-DAZ-SDAG-NEXT: s_setpc_b64 s[30:31] -; -; IR-DAZ-GISEL-LABEL: v_recip_sqrt_f32_ulp25_contract: -; IR-DAZ-GISEL: ; %bb.0: -; IR-DAZ-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; IR-DAZ-GISEL-NEXT: v_mov_b32_e32 v1, 0xf800000 -; IR-DAZ-GISEL-NEXT: v_mul_f32_e32 v2, 0x4f800000, v0 -; IR-DAZ-GISEL-NEXT: v_cmp_lt_f32_e32 vcc, v0, v1 -; IR-DAZ-GISEL-NEXT: v_cndmask_b32_e32 v0, v0, v2, vcc -; IR-DAZ-GISEL-NEXT: v_rsq_f32_e32 v1, v0 -; IR-DAZ-GISEL-NEXT: v_mul_f32_e32 v2, v0, v1 -; IR-DAZ-GISEL-NEXT: v_mul_f32_e32 v1, 0.5, v1 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v3, -v1, v2, 0.5 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v2, v2, v3, v2 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v1, v1, v3, v1 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v3, -v2, v2, v0 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v1, v3, v1, v2 -; IR-DAZ-GISEL-NEXT: v_mul_f32_e32 v2, 0x37800000, v1 -; IR-DAZ-GISEL-NEXT: v_cndmask_b32_e32 v1, v1, v2, vcc -; IR-DAZ-GISEL-NEXT: v_mov_b32_e32 v2, 0x260 -; IR-DAZ-GISEL-NEXT: v_cmp_class_f32_e32 vcc, v0, v2 -; IR-DAZ-GISEL-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; IR-DAZ-GISEL-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 -; IR-DAZ-GISEL-NEXT: v_rcp_f32_e32 v2, v1 -; IR-DAZ-GISEL-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 -; IR-DAZ-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v4, -v1, v2, 1.0 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v2, v4, v2, v2 -; IR-DAZ-GISEL-NEXT: v_mul_f32_e32 v4, v3, v2 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v5, -v1, v4, v3 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v4, v5, v2, v4 -; IR-DAZ-GISEL-NEXT: v_fma_f32 v1, -v1, v4, v3 -; IR-DAZ-GISEL-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 -; IR-DAZ-GISEL-NEXT: v_div_fmas_f32 v1, v1, v2, v4 -; IR-DAZ-GISEL-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 -; IR-DAZ-GISEL-NEXT: s_setpc_b64 s[30:31] +; IR-DAZ-LABEL: v_recip_sqrt_f32_ulp25_contract: +; IR-DAZ: ; %bb.0: +; IR-DAZ-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; IR-DAZ-NEXT: v_sqrt_f32_e32 v0, v0 +; IR-DAZ-NEXT: v_div_scale_f32 v1, s[4:5], v0, v0, 1.0 +; IR-DAZ-NEXT: v_rcp_f32_e32 v2, v1 +; IR-DAZ-NEXT: v_div_scale_f32 v3, vcc, 1.0, v0, 1.0 +; IR-DAZ-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3 +; IR-DAZ-NEXT: v_fma_f32 v4, -v1, v2, 1.0 +; IR-DAZ-NEXT: v_fma_f32 v2, v4, v2, v2 +; IR-DAZ-NEXT: v_mul_f32_e32 v4, v3, v2 +; IR-DAZ-NEXT: v_fma_f32 v5, -v1, v4, v3 +; IR-DAZ-NEXT: v_fma_f32 v4, v5, v2, v4 +; IR-DAZ-NEXT: v_fma_f32 v1, -v1, v4, v3 +; IR-DAZ-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 0 +; IR-DAZ-NEXT: v_div_fmas_f32 v1, v1, v2, v4 +; IR-DAZ-NEXT: v_div_fixup_f32 v0, v1, v0, 1.0 +; IR-DAZ-NEXT: s_setpc_b64 s[30:31] %sqrt = call contract float @llvm.sqrt.f32(float %x), !fpmath !0 %fdiv = fdiv contract float 1.0, %sqrt, !fpmath !0 ret float %fdiv diff --git a/llvm/test/CodeGen/AMDGPU/fmed3.ll b/llvm/test/CodeGen/AMDGPU/fmed3.ll index 9233f80..9e15225 100644 --- a/llvm/test/CodeGen/AMDGPU/fmed3.ll +++ b/llvm/test/CodeGen/AMDGPU/fmed3.ll @@ -7464,18 +7464,15 @@ define amdgpu_kernel void @v_test_nnan_input_fmed3_r_i_i_f16(ptr addrspace(1) %o ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) ; SI-GISEL-NEXT: s_mov_b64 s[4:5], s[2:3] ; SI-GISEL-NEXT: buffer_load_ushort v2, v[0:1], s[4:7], 0 addr64 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 1.0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, 2.0 ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-GISEL-NEXT: v_add_f32_e32 v2, v2, v3 +; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-GISEL-NEXT: v_max_f32_e32 v2, v2, v4 +; SI-GISEL-NEXT: v_max_f32_e32 v2, 2.0, v2 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 4.0 -; SI-GISEL-NEXT: v_min_f32_e32 v2, v2, v3 +; SI-GISEL-NEXT: v_min_f32_e32 v2, 4.0, v2 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 ; SI-GISEL-NEXT: s_mov_b64 s[2:3], s[6:7] ; SI-GISEL-NEXT: buffer_store_short v2, v[0:1], s[0:3], 0 addr64 @@ -7639,27 +7636,24 @@ define amdgpu_kernel void @v_nnan_inputs_med3_f16_pat0(ptr addrspace(1) %out, pt ; SI-GISEL-NEXT: v_mov_b32_e32 v1, 0 ; SI-GISEL-NEXT: s_mov_b32 s10, 0 ; SI-GISEL-NEXT: s_mov_b32 s11, 0xf000 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, 1.0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 2.0 ; SI-GISEL-NEXT: s_waitcnt lgkmcnt(0) ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[2:3] -; SI-GISEL-NEXT: buffer_load_ushort v4, v[0:1], s[8:11], 0 addr64 glc +; SI-GISEL-NEXT: buffer_load_ushort v2, v[0:1], s[8:11], 0 addr64 glc ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v5, 4.0 ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[4:5] -; SI-GISEL-NEXT: buffer_load_ushort v6, v[0:1], s[8:11], 0 addr64 glc +; SI-GISEL-NEXT: buffer_load_ushort v3, v[0:1], s[8:11], 0 addr64 glc ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) ; SI-GISEL-NEXT: s_mov_b64 s[8:9], s[6:7] -; SI-GISEL-NEXT: buffer_load_ushort v7, v[0:1], s[8:11], 0 addr64 glc +; SI-GISEL-NEXT: buffer_load_ushort v4, v[0:1], s[8:11], 0 addr64 glc ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-GISEL-NEXT: v_add_f32_e32 v2, v4, v2 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v6 +; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-GISEL-NEXT: v_add_f32_e32 v2, 1.0, v2 +; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, v3 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-GISEL-NEXT: v_add_f32_e32 v3, v4, v3 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v7 +; SI-GISEL-NEXT: v_add_f32_e32 v3, 2.0, v3 +; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v4, v4 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-GISEL-NEXT: v_add_f32_e32 v4, v4, v5 +; SI-GISEL-NEXT: v_add_f32_e32 v4, 4.0, v4 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v4, v4 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, v3 @@ -8712,12 +8706,10 @@ define half @v_test_fmed3_r_i_i_f16_minimumnum_maximumnum(half %a) #1 { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, 2.0 -; SI-GISEL-NEXT: v_max_f32_e32 v0, v0, v1 +; SI-GISEL-NEXT: v_max_f32_e32 v0, 2.0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, 4.0 -; SI-GISEL-NEXT: v_min_f32_e32 v0, v0, v1 +; SI-GISEL-NEXT: v_min_f32_e32 v0, 4.0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -8796,17 +8788,15 @@ define <2 x half> @v_test_fmed3_r_i_i_v2f16_minimumnum_maximumnum(<2 x half> %a) ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, 2.0 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 4.0 -; SI-GISEL-NEXT: v_max_f32_e32 v0, v0, v2 -; SI-GISEL-NEXT: v_max_f32_e32 v1, v1, v2 +; SI-GISEL-NEXT: v_max_f32_e32 v0, 2.0, v0 +; SI-GISEL-NEXT: v_max_f32_e32 v1, 2.0, v1 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-GISEL-NEXT: v_min_f32_e32 v0, v0, v3 -; SI-GISEL-NEXT: v_min_f32_e32 v1, v1, v3 +; SI-GISEL-NEXT: v_min_f32_e32 v0, 4.0, v0 +; SI-GISEL-NEXT: v_min_f32_e32 v1, 4.0, v1 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v1, v1 ; SI-GISEL-NEXT: s_setpc_b64 s[30:31] diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll index ee11b92..0c1448a 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.smfmac.gfx950.ll @@ -44,23 +44,23 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x64_f16__vgpr(ptr addrspace(1) % ; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: global_load_dwordx4 v[14:17], v0, s[6:7] +; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[6:7] ; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x44 ; GISEL-NEXT: s_load_dword s16, s[4:5], 0x64 -; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[2:3] -; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[0:1] +; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[2:3] +; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[0:1] ; GISEL-NEXT: s_waitcnt lgkmcnt(0) ; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[8:9] ; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[10:11] ; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[12:13] ; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[14:15] -; GISEL-NEXT: v_mov_b32_e32 v12, s16 +; GISEL-NEXT: v_mov_b32_e32 v16, s16 ; GISEL-NEXT: s_waitcnt vmcnt(0) ; GISEL-NEXT: s_nop 0 -; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[14:17], v[8:11], v[0:7], v12 cbsz:1 abid:2 +; GISEL-NEXT: v_smfmac_f32_16x16x64_f16 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2 ; GISEL-NEXT: v_mov_b32_e32 v0, 0 ; GISEL-NEXT: s_nop 6 -; GISEL-NEXT: global_store_dwordx4 v0, v[14:17], s[6:7] +; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[6:7] ; GISEL-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() @@ -834,24 +834,24 @@ define amdgpu_kernel void @test_smfmac_i32_16x16x128_i8__vgpr(ptr addrspace(1) % ; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: global_load_dwordx4 v[14:17], v0, s[0:1] +; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[0:1] ; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GISEL-NEXT: s_load_dwordx4 s[16:19], s[4:5], 0x54 ; GISEL-NEXT: s_load_dword s2, s[4:5], 0x64 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[8:9] -; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[8:9] ; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[12:13] ; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[14:15] ; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17] ; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19] -; GISEL-NEXT: v_mov_b32_e32 v12, s2 +; GISEL-NEXT: v_mov_b32_e32 v16, s2 ; GISEL-NEXT: s_waitcnt vmcnt(0) ; GISEL-NEXT: s_nop 0 -; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[14:17], v[8:11], v[0:7], v12 cbsz:1 abid:2 +; GISEL-NEXT: v_smfmac_i32_16x16x128_i8 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2 ; GISEL-NEXT: v_mov_b32_e32 v0, 0 ; GISEL-NEXT: s_nop 6 -; GISEL-NEXT: global_store_dwordx4 v0, v[14:17], s[0:1] +; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[0:1] ; GISEL-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() @@ -1349,24 +1349,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_bf8_bf8__vgpr(ptr addrspace ; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: global_load_dwordx4 v[14:17], v0, s[0:1] +; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[0:1] ; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GISEL-NEXT: s_load_dwordx4 s[16:19], s[4:5], 0x54 ; GISEL-NEXT: s_load_dword s2, s[4:5], 0x64 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[8:9] -; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[8:9] ; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[12:13] ; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[14:15] ; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17] ; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19] -; GISEL-NEXT: v_mov_b32_e32 v12, s2 +; GISEL-NEXT: v_mov_b32_e32 v16, s2 ; GISEL-NEXT: s_waitcnt vmcnt(0) ; GISEL-NEXT: s_nop 0 -; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[14:17], v[8:11], v[0:7], v12 cbsz:1 abid:2 +; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_bf8 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2 ; GISEL-NEXT: v_mov_b32_e32 v0, 0 ; GISEL-NEXT: s_nop 6 -; GISEL-NEXT: global_store_dwordx4 v0, v[14:17], s[0:1] +; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[0:1] ; GISEL-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() @@ -1513,24 +1513,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_bf8_fp8__vgpr(ptr addrspace ; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: global_load_dwordx4 v[14:17], v0, s[0:1] +; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[0:1] ; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GISEL-NEXT: s_load_dwordx4 s[16:19], s[4:5], 0x54 ; GISEL-NEXT: s_load_dword s2, s[4:5], 0x64 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[8:9] -; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[8:9] ; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[12:13] ; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[14:15] ; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17] ; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19] -; GISEL-NEXT: v_mov_b32_e32 v12, s2 +; GISEL-NEXT: v_mov_b32_e32 v16, s2 ; GISEL-NEXT: s_waitcnt vmcnt(0) ; GISEL-NEXT: s_nop 0 -; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[14:17], v[8:11], v[0:7], v12 cbsz:1 abid:2 +; GISEL-NEXT: v_smfmac_f32_16x16x128_bf8_fp8 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2 ; GISEL-NEXT: v_mov_b32_e32 v0, 0 ; GISEL-NEXT: s_nop 6 -; GISEL-NEXT: global_store_dwordx4 v0, v[14:17], s[0:1] +; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[0:1] ; GISEL-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() @@ -1677,24 +1677,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_fp8_bf8__vgpr(ptr addrspace ; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: global_load_dwordx4 v[14:17], v0, s[0:1] +; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[0:1] ; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GISEL-NEXT: s_load_dwordx4 s[16:19], s[4:5], 0x54 ; GISEL-NEXT: s_load_dword s2, s[4:5], 0x64 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[8:9] -; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[8:9] ; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[12:13] ; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[14:15] ; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17] ; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19] -; GISEL-NEXT: v_mov_b32_e32 v12, s2 +; GISEL-NEXT: v_mov_b32_e32 v16, s2 ; GISEL-NEXT: s_waitcnt vmcnt(0) ; GISEL-NEXT: s_nop 0 -; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[14:17], v[8:11], v[0:7], v12 cbsz:1 abid:2 +; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_bf8 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2 ; GISEL-NEXT: v_mov_b32_e32 v0, 0 ; GISEL-NEXT: s_nop 6 -; GISEL-NEXT: global_store_dwordx4 v0, v[14:17], s[0:1] +; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[0:1] ; GISEL-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() @@ -1841,24 +1841,24 @@ define amdgpu_kernel void @test_smfmac_f32_16x16x128_fp8_fp8__vgpr(ptr addrspace ; GISEL-NEXT: v_and_b32_e32 v0, 0x3ff, v0 ; GISEL-NEXT: v_lshlrev_b32_e32 v0, 4, v0 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: global_load_dwordx4 v[14:17], v0, s[0:1] +; GISEL-NEXT: global_load_dwordx4 v[8:11], v0, s[0:1] ; GISEL-NEXT: s_load_dwordx8 s[8:15], s[4:5], 0x34 ; GISEL-NEXT: s_load_dwordx4 s[16:19], s[4:5], 0x54 ; GISEL-NEXT: s_load_dword s2, s[4:5], 0x64 ; GISEL-NEXT: s_waitcnt lgkmcnt(0) -; GISEL-NEXT: v_mov_b64_e32 v[8:9], s[8:9] -; GISEL-NEXT: v_mov_b64_e32 v[10:11], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[14:15], s[10:11] +; GISEL-NEXT: v_mov_b64_e32 v[12:13], s[8:9] ; GISEL-NEXT: v_mov_b64_e32 v[0:1], s[12:13] ; GISEL-NEXT: v_mov_b64_e32 v[2:3], s[14:15] ; GISEL-NEXT: v_mov_b64_e32 v[4:5], s[16:17] ; GISEL-NEXT: v_mov_b64_e32 v[6:7], s[18:19] -; GISEL-NEXT: v_mov_b32_e32 v12, s2 +; GISEL-NEXT: v_mov_b32_e32 v16, s2 ; GISEL-NEXT: s_waitcnt vmcnt(0) ; GISEL-NEXT: s_nop 0 -; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[14:17], v[8:11], v[0:7], v12 cbsz:1 abid:2 +; GISEL-NEXT: v_smfmac_f32_16x16x128_fp8_fp8 v[8:11], v[12:15], v[0:7], v16 cbsz:1 abid:2 ; GISEL-NEXT: v_mov_b32_e32 v0, 0 ; GISEL-NEXT: s_nop 6 -; GISEL-NEXT: global_store_dwordx4 v0, v[14:17], s[0:1] +; GISEL-NEXT: global_store_dwordx4 v0, v[8:11], s[0:1] ; GISEL-NEXT: s_endpgm bb: %id = call i32 @llvm.amdgcn.workitem.id.x() diff --git a/llvm/test/CodeGen/AMDGPU/llvm.exp.ll b/llvm/test/CodeGen/AMDGPU/llvm.exp.ll index af79c91..ac356fa 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.exp.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.exp.ll @@ -6011,8 +6011,7 @@ define half @v_exp_f16_fast(half %in) { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, 0x3dc5 -; SI-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1 +; SI-GISEL-NEXT: v_mul_f32_e32 v0, 0x3fb8a000, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 ; SI-GISEL-NEXT: v_exp_f32_e32 v0, v0 @@ -6512,10 +6511,9 @@ define <2 x half> @v_exp_v2f16_fast(<2 x half> %in) { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, 0x3dc5 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-GISEL-NEXT: v_mul_f32_e32 v0, v0, v2 -; SI-GISEL-NEXT: v_mul_f32_e32 v1, v1, v2 +; SI-GISEL-NEXT: v_mul_f32_e32 v0, 0x3fb8a000, v0 +; SI-GISEL-NEXT: v_mul_f32_e32 v1, 0x3fb8a000, v1 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 @@ -6709,12 +6707,11 @@ define <3 x half> @v_exp_v3f16_afn(<3 x half> %in) { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 0x3dc5 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-GISEL-NEXT: v_mul_f32_e32 v0, v0, v3 -; SI-GISEL-NEXT: v_mul_f32_e32 v1, v1, v3 -; SI-GISEL-NEXT: v_mul_f32_e32 v2, v2, v3 +; SI-GISEL-NEXT: v_mul_f32_e32 v0, 0x3fb8a000, v0 +; SI-GISEL-NEXT: v_mul_f32_e32 v1, 0x3fb8a000, v1 +; SI-GISEL-NEXT: v_mul_f32_e32 v2, 0x3fb8a000, v2 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 diff --git a/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll b/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll index a99c199..d12ebe4 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.exp10.ll @@ -6092,8 +6092,7 @@ define half @v_exp10_f16_fast(half %in) { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, 0x3dc5 -; SI-GISEL-NEXT: v_mul_f32_e32 v0, v0, v1 +; SI-GISEL-NEXT: v_mul_f32_e32 v0, 0x3fb8a000, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 ; SI-GISEL-NEXT: v_exp_f32_e32 v0, v0 @@ -6594,10 +6593,9 @@ define <2 x half> @v_exp10_v2f16_fast(<2 x half> %in) { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, 0x3dc5 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-GISEL-NEXT: v_mul_f32_e32 v0, v0, v2 -; SI-GISEL-NEXT: v_mul_f32_e32 v1, v1, v2 +; SI-GISEL-NEXT: v_mul_f32_e32 v0, 0x3fb8a000, v0 +; SI-GISEL-NEXT: v_mul_f32_e32 v1, 0x3fb8a000, v1 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 @@ -6791,12 +6789,11 @@ define <3 x half> @v_exp10_v3f16_afn(<3 x half> %in) { ; SI-GISEL: ; %bb.0: ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v3, 0x3dc5 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-GISEL-NEXT: v_mul_f32_e32 v0, v0, v3 -; SI-GISEL-NEXT: v_mul_f32_e32 v1, v1, v3 -; SI-GISEL-NEXT: v_mul_f32_e32 v2, v2, v3 +; SI-GISEL-NEXT: v_mul_f32_e32 v0, 0x3fb8a000, v0 +; SI-GISEL-NEXT: v_mul_f32_e32 v1, 0x3fb8a000, v1 +; SI-GISEL-NEXT: v_mul_f32_e32 v2, 0x3fb8a000, v2 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v1, v1 ; SI-GISEL-NEXT: v_cvt_f16_f32_e32 v2, v2 diff --git a/llvm/test/CodeGen/AMDGPU/mad-mix-hi.ll b/llvm/test/CodeGen/AMDGPU/mad-mix-hi.ll index 3f66c23..259ee0b 100644 --- a/llvm/test/CodeGen/AMDGPU/mad-mix-hi.ll +++ b/llvm/test/CodeGen/AMDGPU/mad-mix-hi.ll @@ -488,13 +488,11 @@ define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_postcvt(half ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2 ; GISEL-CI-NEXT: v_mac_f32_e32 v2, v0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v2 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, 0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_max_f32_e32 v0, v0, v1 +; GISEL-CI-NEXT: v_max_f32_e32 v0, 0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_min_f32_e32 v0, v0, v1 +; GISEL-CI-NEXT: v_min_f32_e32 v0, 1.0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v0 ; GISEL-CI-NEXT: s_setpc_b64 s[30:31] %src0.ext = fpext half %src0 to float @@ -582,15 +580,13 @@ define <2 x half> @v_mad_mixhi_f16_f16lo_f16lo_f16lo_undeflo_clamp_postcvt_multi ; GISEL-CI-NEXT: s_mov_b32 s7, 0xf000 ; GISEL-CI-NEXT: v_mac_f32_e32 v2, v0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v2 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, 0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v0 +; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GISEL-CI-NEXT: buffer_store_short v0, off, s[4:7], 0 ; GISEL-CI-NEXT: s_waitcnt vmcnt(0) -; GISEL-CI-NEXT: v_max_f32_e32 v1, v2, v1 +; GISEL-CI-NEXT: v_max_f32_e32 v1, 0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GISEL-CI-NEXT: v_min_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_min_f32_e32 v1, 1.0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: s_setpc_b64 s[30:31] %src0.ext = fpext half %src0 to float diff --git a/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll b/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll index 21e6faf4..ba77552 100644 --- a/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll +++ b/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll @@ -313,13 +313,11 @@ define half @v_mad_mixlo_f16_f16lo_f16lo_f32_clamp_post_cvt(half %src0, half %sr ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 ; GISEL-CI-NEXT: v_mac_f32_e32 v2, v0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v2 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, 0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_max_f32_e32 v0, v0, v1 +; GISEL-CI-NEXT: v_max_f32_e32 v0, 0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_min_f32_e32 v0, v0, v1 +; GISEL-CI-NEXT: v_min_f32_e32 v0, 1.0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GISEL-CI-NEXT: s_setpc_b64 s[30:31] %src0.ext = fpext half %src0 to float @@ -1009,28 +1007,26 @@ define <2 x half> @v_mad_mix_v2f32_clamp_postcvt(<2 x half> %src0, <2 x half> %s ; GISEL-CI: ; %bb.0: ; GISEL-CI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, v3 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v5, v5 ; GISEL-CI-NEXT: v_mac_f32_e32 v4, v0, v2 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v4 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 0 ; GISEL-CI-NEXT: v_mac_f32_e32 v5, v1, v3 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v5 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GISEL-CI-NEXT: v_max_f32_e32 v0, v0, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v0, 0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GISEL-CI-NEXT: v_max_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v1, 0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GISEL-CI-NEXT: v_min_f32_e32 v0, v0, v2 +; GISEL-CI-NEXT: v_min_f32_e32 v0, 1.0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 -; GISEL-CI-NEXT: v_min_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_min_f32_e32 v1, 1.0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: s_setpc_b64 s[30:31] %src0.ext = fpext <2 x half> %src0 to <2 x float> @@ -1225,25 +1221,23 @@ define <3 x half> @v_mad_mix_v3f32_clamp_postcvt(<3 x half> %src0, <3 x half> %s ; GISEL-CI-NEXT: v_mac_f32_e32 v7, v1, v4 ; GISEL-CI-NEXT: v_mac_f32_e32 v8, v2, v5 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v7 -; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v3, v8 +; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v2, v8 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, v3 -; GISEL-CI-NEXT: v_max_f32_e32 v0, v0, v2 -; GISEL-CI-NEXT: v_max_f32_e32 v1, v1, v2 -; GISEL-CI-NEXT: v_max_f32_e32 v2, v3, v2 +; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v0, 0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GISEL-CI-NEXT: v_max_f32_e32 v1, 0, v1 +; GISEL-CI-NEXT: v_max_f32_e32 v2, 0, v2 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v2, v2 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2 -; GISEL-CI-NEXT: v_min_f32_e32 v0, v0, v3 -; GISEL-CI-NEXT: v_min_f32_e32 v1, v1, v3 -; GISEL-CI-NEXT: v_min_f32_e32 v2, v2, v3 +; GISEL-CI-NEXT: v_min_f32_e32 v0, 1.0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; GISEL-CI-NEXT: v_min_f32_e32 v1, 1.0, v1 +; GISEL-CI-NEXT: v_min_f32_e32 v2, 1.0, v2 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v2, v2 ; GISEL-CI-NEXT: s_setpc_b64 s[30:31] @@ -1441,30 +1435,28 @@ define <4 x half> @v_mad_mix_v4f32_clamp_postcvt(<4 x half> %src0, <4 x half> %s ; GISEL-CI-NEXT: v_mac_f32_e32 v11, v3, v7 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v8 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v9 -; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v3, v10 -; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v4, v11 +; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v2, v10 +; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v3, v11 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, v3 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v4, v4 -; GISEL-CI-NEXT: v_max_f32_e32 v0, v0, v2 -; GISEL-CI-NEXT: v_max_f32_e32 v1, v1, v2 -; GISEL-CI-NEXT: v_max_f32_e32 v3, v3, v2 -; GISEL-CI-NEXT: v_max_f32_e32 v2, v4, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v0, 0, v0 +; GISEL-CI-NEXT: v_max_f32_e32 v1, 0, v1 +; GISEL-CI-NEXT: v_max_f32_e32 v2, 0, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v3, 0, v3 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 -; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v3, v3 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v5, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, v2 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v3, v3 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v4, v2 -; GISEL-CI-NEXT: v_min_f32_e32 v0, v0, v5 -; GISEL-CI-NEXT: v_min_f32_e32 v1, v1, v5 -; GISEL-CI-NEXT: v_min_f32_e32 v2, v3, v5 -; GISEL-CI-NEXT: v_min_f32_e32 v3, v4, v5 +; GISEL-CI-NEXT: v_min_f32_e32 v0, 1.0, v0 +; GISEL-CI-NEXT: v_min_f32_e32 v1, 1.0, v1 +; GISEL-CI-NEXT: v_min_f32_e32 v2, 1.0, v2 +; GISEL-CI-NEXT: v_min_f32_e32 v3, 1.0, v3 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v2, v2 @@ -1622,16 +1614,14 @@ define <2 x half> @v_mad_mix_v2f32_clamp_postcvt_lo(<2 x half> %src0, <2 x half> ; GISEL-CI-NEXT: v_mac_f32_e32 v4, v0, v2 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v5 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v4 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 0 ; GISEL-CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0 ; GISEL-CI-NEXT: v_or_b32_e32 v0, v1, v0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v0 ; GISEL-CI-NEXT: v_and_b32_e32 v0, 0xffff0000, v0 -; GISEL-CI-NEXT: v_max_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v1, 0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GISEL-CI-NEXT: v_min_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_min_f32_e32 v1, 1.0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: v_or_b32_e32 v0, v0, v1 ; GISEL-CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0 @@ -1790,17 +1780,15 @@ define <2 x half> @v_mad_mix_v2f32_clamp_postcvt_hi(<2 x half> %src0, <2 x half> ; GISEL-CI-NEXT: v_mac_f32_e32 v4, v0, v2 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v0, v5 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v4 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 0 ; GISEL-CI-NEXT: v_lshlrev_b32_e32 v0, 16, v0 ; GISEL-CI-NEXT: v_or_b32_e32 v0, v1, v0 ; GISEL-CI-NEXT: v_lshrrev_b32_e32 v1, 16, v0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 ; GISEL-CI-NEXT: v_and_b32_e32 v0, 0xffff, v0 -; GISEL-CI-NEXT: v_max_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_max_f32_e32 v1, 0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 -; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v2, 1.0 ; GISEL-CI-NEXT: v_cvt_f32_f16_e32 v1, v1 -; GISEL-CI-NEXT: v_min_f32_e32 v1, v1, v2 +; GISEL-CI-NEXT: v_min_f32_e32 v1, 1.0, v1 ; GISEL-CI-NEXT: v_cvt_f16_f32_e32 v1, v1 ; GISEL-CI-NEXT: v_lshlrev_b32_e32 v1, 16, v1 ; GISEL-CI-NEXT: v_or_b32_e32 v0, v0, v1 diff --git a/llvm/test/CodeGen/AMDGPU/maximumnum.ll b/llvm/test/CodeGen/AMDGPU/maximumnum.ll index 4f73e8e..c90b2c9 100644 --- a/llvm/test/CodeGen/AMDGPU/maximumnum.ll +++ b/llvm/test/CodeGen/AMDGPU/maximumnum.ll @@ -271,8 +271,7 @@ define half @v_maximumnum_f16_1.0(half %x) { ; GFX7-GISEL: ; %bb.0: ; GFX7-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX7-GISEL-NEXT: v_cvt_f32_f16_e32 v1, 1.0 -; GFX7-GISEL-NEXT: v_max_f32_e32 v0, v0, v1 +; GFX7-GISEL-NEXT: v_max_f32_e32 v0, 1.0, v0 ; GFX7-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX7-GISEL-NEXT: s_setpc_b64 s[30:31] ; diff --git a/llvm/test/CodeGen/AMDGPU/mfma-loop.ll b/llvm/test/CodeGen/AMDGPU/mfma-loop.ll index 0af655df..4bb6538 100644 --- a/llvm/test/CodeGen/AMDGPU/mfma-loop.ll +++ b/llvm/test/CodeGen/AMDGPU/mfma-loop.ll @@ -2399,8 +2399,9 @@ define amdgpu_kernel void @test_mfma_nested_loop_zeroinit(ptr addrspace(1) %arg) ; GFX90A-NEXT: v_accvgpr_mov_b32 a29, a0 ; GFX90A-NEXT: v_accvgpr_mov_b32 a30, a0 ; GFX90A-NEXT: v_accvgpr_mov_b32 a31, a0 -; GFX90A-NEXT: v_mov_b32_e32 v0, 2.0 -; GFX90A-NEXT: v_mov_b32_e32 v1, 1.0 +; GFX90A-NEXT: v_mov_b32_e32 v0, 1.0 +; GFX90A-NEXT: v_mov_b32_e32 v1, 2.0 +; GFX90A-NEXT: ; kill: def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 killed $exec ; GFX90A-NEXT: .LBB9_1: ; %for.cond.preheader ; GFX90A-NEXT: ; =>This Loop Header: Depth=1 ; GFX90A-NEXT: ; Child Loop BB9_2 Depth 2 @@ -2409,7 +2410,7 @@ define amdgpu_kernel void @test_mfma_nested_loop_zeroinit(ptr addrspace(1) %arg) ; GFX90A-NEXT: ; Parent Loop BB9_1 Depth=1 ; GFX90A-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX90A-NEXT: s_nop 0 -; GFX90A-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v1, v0, a[0:31] +; GFX90A-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v0, v1, a[0:31] ; GFX90A-NEXT: s_add_i32 s1, s1, -1 ; GFX90A-NEXT: s_cmp_lg_u32 s1, 0 ; GFX90A-NEXT: s_cbranch_scc1 .LBB9_2 @@ -2468,8 +2469,9 @@ define amdgpu_kernel void @test_mfma_nested_loop_zeroinit(ptr addrspace(1) %arg) ; GFX942-NEXT: v_accvgpr_mov_b32 a29, a0 ; GFX942-NEXT: v_accvgpr_mov_b32 a30, a0 ; GFX942-NEXT: v_accvgpr_mov_b32 a31, a0 -; GFX942-NEXT: v_mov_b32_e32 v0, 2.0 -; GFX942-NEXT: v_mov_b32_e32 v1, 1.0 +; GFX942-NEXT: v_mov_b32_e32 v0, 1.0 +; GFX942-NEXT: v_mov_b32_e32 v1, 2.0 +; GFX942-NEXT: ; kill: def $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 killed $agpr0_agpr1_agpr2_agpr3_agpr4_agpr5_agpr6_agpr7_agpr8_agpr9_agpr10_agpr11_agpr12_agpr13_agpr14_agpr15_agpr16_agpr17_agpr18_agpr19_agpr20_agpr21_agpr22_agpr23_agpr24_agpr25_agpr26_agpr27_agpr28_agpr29_agpr30_agpr31 killed $exec ; GFX942-NEXT: .LBB9_1: ; %for.cond.preheader ; GFX942-NEXT: ; =>This Loop Header: Depth=1 ; GFX942-NEXT: ; Child Loop BB9_2 Depth 2 @@ -2478,7 +2480,7 @@ define amdgpu_kernel void @test_mfma_nested_loop_zeroinit(ptr addrspace(1) %arg) ; GFX942-NEXT: ; Parent Loop BB9_1 Depth=1 ; GFX942-NEXT: ; => This Inner Loop Header: Depth=2 ; GFX942-NEXT: s_nop 0 -; GFX942-NEXT: v_mfma_f32_32x32x1_2b_f32 a[0:31], v1, v0, a[0:31] +; GFX942-NEXT: v_mfma_f32_32x32x1_2b_f32 a[0:31], v0, v1, a[0:31] ; GFX942-NEXT: s_add_i32 s1, s1, -1 ; GFX942-NEXT: s_cmp_lg_u32 s1, 0 ; GFX942-NEXT: s_cbranch_scc1 .LBB9_2 diff --git a/llvm/test/CodeGen/AMDGPU/minimumnum.ll b/llvm/test/CodeGen/AMDGPU/minimumnum.ll index 558006d..64e8b7b 100644 --- a/llvm/test/CodeGen/AMDGPU/minimumnum.ll +++ b/llvm/test/CodeGen/AMDGPU/minimumnum.ll @@ -271,8 +271,7 @@ define half @v_minimumnum_f16_1.0(half %x) { ; GFX7-GISEL: ; %bb.0: ; GFX7-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX7-GISEL-NEXT: v_cvt_f32_f16_e32 v0, v0 -; GFX7-GISEL-NEXT: v_cvt_f32_f16_e32 v1, 1.0 -; GFX7-GISEL-NEXT: v_min_f32_e32 v0, v0, v1 +; GFX7-GISEL-NEXT: v_min_f32_e32 v0, 1.0, v0 ; GFX7-GISEL-NEXT: v_cvt_f16_f32_e32 v0, v0 ; GFX7-GISEL-NEXT: s_setpc_b64 s[30:31] ; diff --git a/llvm/test/CodeGen/AMDGPU/regpressure_printer.mir b/llvm/test/CodeGen/AMDGPU/regpressure_printer.mir index 8d5b5e4..b41aa08 100644 --- a/llvm/test/CodeGen/AMDGPU/regpressure_printer.mir +++ b/llvm/test/CodeGen/AMDGPU/regpressure_printer.mir @@ -510,14 +510,14 @@ body: | ; RPU-NEXT: 0 0 $sgpr0 = S_BUFFER_LOAD_DWORD_IMM $sgpr0_sgpr1_sgpr2_sgpr3, 0, 0 ; RPU-NEXT: 0 0 ; RPU-NEXT: 0 1 undef %0.sub5:vreg_512 = V_MOV_B32_e32 5, implicit $exec - ; RPU-NEXT: 0 0 - ; RPU-NEXT: 0 0 S_CMP_GT_U32 $sgpr0, 15, implicit-def $scc - ; RPU-NEXT: 0 0 - ; RPU-NEXT: 0 0 S_CBRANCH_SCC1 %bb.2, implicit $scc - ; RPU-NEXT: 0 0 - ; RPU-NEXT: 0 0 S_BRANCH %bb.1 - ; RPU-NEXT: 0 0 - ; RPU-NEXT: Live-out: + ; RPU-NEXT: 0 1 + ; RPU-NEXT: 0 1 S_CMP_GT_U32 $sgpr0, 15, implicit-def $scc + ; RPU-NEXT: 0 1 + ; RPU-NEXT: 0 1 S_CBRANCH_SCC1 %bb.2, implicit $scc + ; RPU-NEXT: 0 1 + ; RPU-NEXT: 0 1 S_BRANCH %bb.1 + ; RPU-NEXT: 0 1 + ; RPU-NEXT: Live-out: %0:0000000000000C00 ; RPU-NEXT: Live-thr: ; RPU-NEXT: 0 0 ; RPU-NEXT: bb.1: @@ -571,8 +571,6 @@ body: | ; RPD-NEXT: 0 1 S_BRANCH %bb.1 ; RPD-NEXT: 0 1 ; RPD-NEXT: Live-out: %0:0000000000000C00 - ; RPD-NEXT: mis LIS: - ; RPD-NEXT: %0:L0000000000000C00 isn't found in LIS reported set ; RPD-NEXT: Live-thr: ; RPD-NEXT: 0 0 ; RPD-NEXT: bb.1: diff --git a/llvm/test/CodeGen/AMDGPU/uniform-select.ll b/llvm/test/CodeGen/AMDGPU/uniform-select.ll index f001bf0..b52913f 100644 --- a/llvm/test/CodeGen/AMDGPU/uniform-select.ll +++ b/llvm/test/CodeGen/AMDGPU/uniform-select.ll @@ -20,34 +20,34 @@ define amdgpu_kernel void @test_insert_extract(i32 %p, i32 %q) { ; GFX90A-NEXT: s_cmp_eq_u32 s1, 1 ; GFX90A-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX90A-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX90A-NEXT: s_cselect_b32 s7, s4, s3 +; GFX90A-NEXT: s_cselect_b32 s7, s3, s2 ; GFX90A-NEXT: s_cmp_eq_u32 s1, 2 ; GFX90A-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX90A-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX90A-NEXT: s_cselect_b32 s7, s5, s7 +; GFX90A-NEXT: s_cselect_b32 s7, s4, s7 ; GFX90A-NEXT: s_cmp_eq_u32 s1, 3 ; GFX90A-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX90A-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX90A-NEXT: s_cselect_b32 s7, s6, s7 +; GFX90A-NEXT: s_cselect_b32 s7, s5, s7 ; GFX90A-NEXT: s_or_b32 s7, s7, s0 ; GFX90A-NEXT: s_cmp_eq_u32 s1, 1 ; GFX90A-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX90A-NEXT: s_and_b64 s[10:11], s[8:9], exec -; GFX90A-NEXT: s_cselect_b32 s4, s7, s4 +; GFX90A-NEXT: s_cselect_b32 s3, s7, s3 ; GFX90A-NEXT: s_cmp_eq_u32 s1, 3 ; GFX90A-NEXT: s_cselect_b64 s[10:11], -1, 0 ; GFX90A-NEXT: s_and_b64 s[12:13], s[10:11], exec -; GFX90A-NEXT: s_cselect_b32 s6, s7, s6 +; GFX90A-NEXT: s_cselect_b32 s5, s7, s5 ; GFX90A-NEXT: s_cmp_eq_u32 s1, 2 ; GFX90A-NEXT: s_cselect_b64 s[12:13], -1, 0 ; GFX90A-NEXT: s_and_b64 s[14:15], s[12:13], exec -; GFX90A-NEXT: s_cselect_b32 s5, s7, s5 +; GFX90A-NEXT: s_cselect_b32 s4, s7, s4 ; GFX90A-NEXT: s_cmp_eq_u32 s1, 0 -; GFX90A-NEXT: s_cselect_b32 s3, s7, s3 +; GFX90A-NEXT: s_cselect_b32 s2, s7, s2 ; GFX90A-NEXT: s_or_b64 s[8:9], s[12:13], s[8:9] ; GFX90A-NEXT: s_or_b64 s[8:9], s[10:11], s[8:9] ; GFX90A-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX90A-NEXT: s_cselect_b32 s2, 0, s2 +; GFX90A-NEXT: s_cselect_b32 s6, 0, s6 ; GFX90A-NEXT: s_mov_b64 vcc, vcc ; GFX90A-NEXT: s_cbranch_vccnz .LBB0_1 ; GFX90A-NEXT: ; %bb.2: ; %DummyReturnBlock @@ -68,34 +68,34 @@ define amdgpu_kernel void @test_insert_extract(i32 %p, i32 %q) { ; GFX942-NEXT: s_cmp_eq_u32 s1, 1 ; GFX942-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX942-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX942-NEXT: s_cselect_b32 s7, s4, s3 +; GFX942-NEXT: s_cselect_b32 s7, s3, s2 ; GFX942-NEXT: s_cmp_eq_u32 s1, 2 ; GFX942-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX942-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX942-NEXT: s_cselect_b32 s7, s5, s7 +; GFX942-NEXT: s_cselect_b32 s7, s4, s7 ; GFX942-NEXT: s_cmp_eq_u32 s1, 3 ; GFX942-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX942-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX942-NEXT: s_cselect_b32 s7, s6, s7 +; GFX942-NEXT: s_cselect_b32 s7, s5, s7 ; GFX942-NEXT: s_or_b32 s7, s7, s0 ; GFX942-NEXT: s_cmp_eq_u32 s1, 1 ; GFX942-NEXT: s_cselect_b64 s[8:9], -1, 0 ; GFX942-NEXT: s_and_b64 s[10:11], s[8:9], exec -; GFX942-NEXT: s_cselect_b32 s4, s7, s4 +; GFX942-NEXT: s_cselect_b32 s3, s7, s3 ; GFX942-NEXT: s_cmp_eq_u32 s1, 3 ; GFX942-NEXT: s_cselect_b64 s[10:11], -1, 0 ; GFX942-NEXT: s_and_b64 s[12:13], s[10:11], exec -; GFX942-NEXT: s_cselect_b32 s6, s7, s6 +; GFX942-NEXT: s_cselect_b32 s5, s7, s5 ; GFX942-NEXT: s_cmp_eq_u32 s1, 2 ; GFX942-NEXT: s_cselect_b64 s[12:13], -1, 0 ; GFX942-NEXT: s_and_b64 s[14:15], s[12:13], exec -; GFX942-NEXT: s_cselect_b32 s5, s7, s5 +; GFX942-NEXT: s_cselect_b32 s4, s7, s4 ; GFX942-NEXT: s_cmp_eq_u32 s1, 0 -; GFX942-NEXT: s_cselect_b32 s3, s7, s3 +; GFX942-NEXT: s_cselect_b32 s2, s7, s2 ; GFX942-NEXT: s_or_b64 s[8:9], s[12:13], s[8:9] ; GFX942-NEXT: s_or_b64 s[8:9], s[10:11], s[8:9] ; GFX942-NEXT: s_and_b64 s[8:9], s[8:9], exec -; GFX942-NEXT: s_cselect_b32 s2, 0, s2 +; GFX942-NEXT: s_cselect_b32 s6, 0, s6 ; GFX942-NEXT: s_mov_b64 vcc, vcc ; GFX942-NEXT: s_cbranch_vccnz .LBB0_1 ; GFX942-NEXT: ; %bb.2: ; %DummyReturnBlock @@ -117,34 +117,34 @@ define amdgpu_kernel void @test_insert_extract(i32 %p, i32 %q) { ; GFX1030-NEXT: s_cmp_eq_u32 s1, 1 ; GFX1030-NEXT: s_cselect_b32 s7, -1, 0 ; GFX1030-NEXT: s_and_b32 s7, s7, exec_lo -; GFX1030-NEXT: s_cselect_b32 s7, s4, s3 +; GFX1030-NEXT: s_cselect_b32 s7, s3, s2 ; GFX1030-NEXT: s_cmp_eq_u32 s1, 2 ; GFX1030-NEXT: s_cselect_b32 s8, -1, 0 ; GFX1030-NEXT: s_and_b32 s8, s8, exec_lo -; GFX1030-NEXT: s_cselect_b32 s7, s5, s7 +; GFX1030-NEXT: s_cselect_b32 s7, s4, s7 ; GFX1030-NEXT: s_cmp_eq_u32 s1, 3 ; GFX1030-NEXT: s_cselect_b32 s8, -1, 0 ; GFX1030-NEXT: s_and_b32 s8, s8, exec_lo -; GFX1030-NEXT: s_cselect_b32 s7, s6, s7 +; GFX1030-NEXT: s_cselect_b32 s7, s5, s7 ; GFX1030-NEXT: s_or_b32 s7, s7, s0 ; GFX1030-NEXT: s_cmp_eq_u32 s1, 1 ; GFX1030-NEXT: s_cselect_b32 s8, -1, 0 ; GFX1030-NEXT: s_and_b32 s9, s8, exec_lo -; GFX1030-NEXT: s_cselect_b32 s4, s7, s4 +; GFX1030-NEXT: s_cselect_b32 s3, s7, s3 ; GFX1030-NEXT: s_cmp_eq_u32 s1, 3 ; GFX1030-NEXT: s_cselect_b32 s9, -1, 0 ; GFX1030-NEXT: s_and_b32 s10, s9, exec_lo -; GFX1030-NEXT: s_cselect_b32 s6, s7, s6 +; GFX1030-NEXT: s_cselect_b32 s5, s7, s5 ; GFX1030-NEXT: s_cmp_eq_u32 s1, 2 ; GFX1030-NEXT: s_cselect_b32 s10, -1, 0 ; GFX1030-NEXT: s_and_b32 s11, s10, exec_lo -; GFX1030-NEXT: s_cselect_b32 s5, s7, s5 +; GFX1030-NEXT: s_cselect_b32 s4, s7, s4 ; GFX1030-NEXT: s_cmp_eq_u32 s1, 0 -; GFX1030-NEXT: s_cselect_b32 s3, s7, s3 +; GFX1030-NEXT: s_cselect_b32 s2, s7, s2 ; GFX1030-NEXT: s_or_b32 s7, s10, s8 ; GFX1030-NEXT: s_or_b32 s7, s9, s7 ; GFX1030-NEXT: s_and_b32 s7, s7, exec_lo -; GFX1030-NEXT: s_cselect_b32 s2, 0, s2 +; GFX1030-NEXT: s_cselect_b32 s6, 0, s6 ; GFX1030-NEXT: s_cbranch_vccnz .LBB0_1 ; GFX1030-NEXT: ; %bb.2: ; %DummyReturnBlock ; GFX1030-NEXT: s_endpgm @@ -166,38 +166,38 @@ define amdgpu_kernel void @test_insert_extract(i32 %p, i32 %q) { ; GFX1100-NEXT: s_cselect_b32 s7, -1, 0 ; GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1) ; GFX1100-NEXT: s_and_b32 s7, s7, exec_lo -; GFX1100-NEXT: s_cselect_b32 s7, s4, s3 +; GFX1100-NEXT: s_cselect_b32 s7, s3, s2 ; GFX1100-NEXT: s_cmp_eq_u32 s1, 2 ; GFX1100-NEXT: s_cselect_b32 s8, -1, 0 ; GFX1100-NEXT: s_and_b32 s8, s8, exec_lo -; GFX1100-NEXT: s_cselect_b32 s7, s5, s7 +; GFX1100-NEXT: s_cselect_b32 s7, s4, s7 ; GFX1100-NEXT: s_cmp_eq_u32 s1, 3 ; GFX1100-NEXT: s_cselect_b32 s8, -1, 0 ; GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_1) | instid1(SALU_CYCLE_1) ; GFX1100-NEXT: s_and_b32 s8, s8, exec_lo -; GFX1100-NEXT: s_cselect_b32 s7, s6, s7 +; GFX1100-NEXT: s_cselect_b32 s7, s5, s7 ; GFX1100-NEXT: s_or_b32 s7, s7, s0 ; GFX1100-NEXT: s_cmp_eq_u32 s1, 1 ; GFX1100-NEXT: s_cselect_b32 s8, -1, 0 ; GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_3) | instid1(SALU_CYCLE_1) ; GFX1100-NEXT: s_and_b32 s9, s8, exec_lo -; GFX1100-NEXT: s_cselect_b32 s4, s7, s4 +; GFX1100-NEXT: s_cselect_b32 s3, s7, s3 ; GFX1100-NEXT: s_cmp_eq_u32 s1, 3 ; GFX1100-NEXT: s_cselect_b32 s9, -1, 0 ; GFX1100-NEXT: s_and_b32 s10, s9, exec_lo -; GFX1100-NEXT: s_cselect_b32 s6, s7, s6 +; GFX1100-NEXT: s_cselect_b32 s5, s7, s5 ; GFX1100-NEXT: s_cmp_eq_u32 s1, 2 ; GFX1100-NEXT: s_cselect_b32 s10, -1, 0 ; GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) | instskip(SKIP_4) | instid1(SALU_CYCLE_1) ; GFX1100-NEXT: s_and_b32 s11, s10, exec_lo -; GFX1100-NEXT: s_cselect_b32 s5, s7, s5 +; GFX1100-NEXT: s_cselect_b32 s4, s7, s4 ; GFX1100-NEXT: s_cmp_eq_u32 s1, 0 -; GFX1100-NEXT: s_cselect_b32 s3, s7, s3 +; GFX1100-NEXT: s_cselect_b32 s2, s7, s2 ; GFX1100-NEXT: s_or_b32 s7, s10, s8 ; GFX1100-NEXT: s_or_b32 s7, s9, s7 ; GFX1100-NEXT: s_delay_alu instid0(SALU_CYCLE_1) ; GFX1100-NEXT: s_and_b32 s7, s7, exec_lo -; GFX1100-NEXT: s_cselect_b32 s2, 0, s2 +; GFX1100-NEXT: s_cselect_b32 s6, 0, s6 ; GFX1100-NEXT: s_cbranch_vccnz .LBB0_1 ; GFX1100-NEXT: ; %bb.2: ; %DummyReturnBlock ; GFX1100-NEXT: s_endpgm diff --git a/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll b/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll index 6b5bae0..c9b94e0 100644 --- a/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll +++ b/llvm/test/CodeGen/AMDGPU/v_sat_pk_u8_i16.ll @@ -6,12 +6,12 @@ ; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=SDAG-GFX12,SDAG-GFX12-TRUE16 %s ; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=SDAG-GFX12,SDAG-GFX12-FAKE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=fiji -global-isel < %s | FileCheck -check-prefixes=GISEL-VI %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -global-isel < %s | FileCheck -check-prefixes=GISEL-GFX9 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=+real-true16 -global-isel < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-TRUE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=-real-true16 -global-isel < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-FAKE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -global-isel < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-TRUE16 %s -; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 -global-isel < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-FAKE16 %s +; RUN: llc -mtriple=amdgcn -mcpu=fiji -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-VI %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx900 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-GFX9 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=+real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-TRUE16 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx1101 -mattr=-real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GFX11,GISEL-GFX11,GISEL-GFX11-FAKE16 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-TRUE16 %s +; RUN: llc -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 -global-isel -new-reg-bank-select < %s | FileCheck -check-prefixes=GISEL-GFX12,GISEL-GFX12-FAKE16 %s ; <GFX9 has no V_SAT_PK, GFX9+ has V_SAT_PK, GFX11 has V_SAT_PK with t16 diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-smax.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-smax.ll index b5d9d00..8d0e003 100644 --- a/llvm/test/CodeGen/AMDGPU/vector-reduce-smax.ll +++ b/llvm/test/CodeGen/AMDGPU/vector-reduce-smax.ll @@ -1,21 +1,21 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s ; FIXME-TRUE16. enable gisel -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s define i8 @test_vector_reduce_smax_v2i8(<2 x i8> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smax_v2i8: @@ -1632,6 +1632,7 @@ entry: ret i8 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smax_v2i16(<2 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smax_v2i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1678,7 +1679,7 @@ define i16 @test_vector_reduce_smax_v2i16(<2 x i16> %v) { ; GFX9-GISEL-LABEL: test_vector_reduce_smax_v2i16: ; GFX9-GISEL: ; %bb.0: ; %entry ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1692,7 +1693,7 @@ define i16 @test_vector_reduce_smax_v2i16(<2 x i16> %v) { ; GFX10-GISEL-LABEL: test_vector_reduce_smax_v2i16: ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1713,7 +1714,7 @@ define i16 @test_vector_reduce_smax_v2i16(<2 x i16> %v) { ; GFX11-GISEL-LABEL: test_vector_reduce_smax_v2i16: ; GFX11-GISEL: ; %bb.0: ; %entry ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1747,7 +1748,7 @@ define i16 @test_vector_reduce_smax_v2i16(<2 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 ; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1900,6 +1901,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smax_v4i16(<4 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smax_v4i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1961,7 +1963,7 @@ define i16 @test_vector_reduce_smax_v4i16(<4 x i16> %v) { ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1977,7 +1979,7 @@ define i16 @test_vector_reduce_smax_v4i16(<4 x i16> %v) { ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2003,7 +2005,7 @@ define i16 @test_vector_reduce_smax_v4i16(<4 x i16> %v) { ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2041,7 +2043,7 @@ define i16 @test_vector_reduce_smax_v4i16(<4 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] entry: @@ -2049,6 +2051,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smax_v8i16(<8 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smax_v8i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -2139,7 +2142,7 @@ define i16 @test_vector_reduce_smax_v8i16(<8 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2159,7 +2162,7 @@ define i16 @test_vector_reduce_smax_v8i16(<8 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_max_i16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2192,7 +2195,7 @@ define i16 @test_vector_reduce_smax_v8i16(<8 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_max_i16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2238,7 +2241,7 @@ define i16 @test_vector_reduce_smax_v8i16(<8 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_max_i16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2247,6 +2250,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smax_v16i16(<16 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smax_v16i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -2391,7 +2395,7 @@ define i16 @test_vector_reduce_smax_v16i16(<16 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2419,7 +2423,7 @@ define i16 @test_vector_reduce_smax_v16i16(<16 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_max_i16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2467,7 +2471,7 @@ define i16 @test_vector_reduce_smax_v16i16(<16 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_max_i16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2528,7 +2532,7 @@ define i16 @test_vector_reduce_smax_v16i16(<16 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_max_i16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-smin.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-smin.ll index 2a989ec..f15ecf0 100644 --- a/llvm/test/CodeGen/AMDGPU/vector-reduce-smin.ll +++ b/llvm/test/CodeGen/AMDGPU/vector-reduce-smin.ll @@ -1,21 +1,21 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s ; FIXME-TRUE16. enable gisel -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s define i8 @test_vector_reduce_smin_v2i8(<2 x i8> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smin_v2i8: @@ -1632,6 +1632,7 @@ entry: ret i8 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smin_v2i16(<2 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smin_v2i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1678,7 +1679,7 @@ define i16 @test_vector_reduce_smin_v2i16(<2 x i16> %v) { ; GFX9-GISEL-LABEL: test_vector_reduce_smin_v2i16: ; GFX9-GISEL: ; %bb.0: ; %entry ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1692,7 +1693,7 @@ define i16 @test_vector_reduce_smin_v2i16(<2 x i16> %v) { ; GFX10-GISEL-LABEL: test_vector_reduce_smin_v2i16: ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1713,7 +1714,7 @@ define i16 @test_vector_reduce_smin_v2i16(<2 x i16> %v) { ; GFX11-GISEL-LABEL: test_vector_reduce_smin_v2i16: ; GFX11-GISEL: ; %bb.0: ; %entry ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1747,7 +1748,7 @@ define i16 @test_vector_reduce_smin_v2i16(<2 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 ; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1900,6 +1901,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smin_v4i16(<4 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smin_v4i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1961,7 +1963,7 @@ define i16 @test_vector_reduce_smin_v4i16(<4 x i16> %v) { ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1977,7 +1979,7 @@ define i16 @test_vector_reduce_smin_v4i16(<4 x i16> %v) { ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2003,7 +2005,7 @@ define i16 @test_vector_reduce_smin_v4i16(<4 x i16> %v) { ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2041,7 +2043,7 @@ define i16 @test_vector_reduce_smin_v4i16(<4 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] entry: @@ -2049,6 +2051,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_smin_v8i16(<8 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_smin_v8i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -2139,7 +2142,7 @@ define i16 @test_vector_reduce_smin_v8i16(<8 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2159,7 +2162,7 @@ define i16 @test_vector_reduce_smin_v8i16(<8 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_min_i16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2192,7 +2195,7 @@ define i16 @test_vector_reduce_smin_v8i16(<8 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_min_i16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2238,7 +2241,7 @@ define i16 @test_vector_reduce_smin_v8i16(<8 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_min_i16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2391,7 +2394,7 @@ define i16 @test_vector_reduce_smin_v16i16(<16 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2419,7 +2422,7 @@ define i16 @test_vector_reduce_smin_v16i16(<16 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_min_i16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2467,7 +2470,7 @@ define i16 @test_vector_reduce_smin_v16i16(<16 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_min_i16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2528,7 +2531,7 @@ define i16 @test_vector_reduce_smin_v16i16(<16 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_min_i16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_i16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-umax.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-umax.ll index 69fd58a..e62165c 100644 --- a/llvm/test/CodeGen/AMDGPU/vector-reduce-umax.ll +++ b/llvm/test/CodeGen/AMDGPU/vector-reduce-umax.ll @@ -1,21 +1,21 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s ; FIXME-TRUE16. enable gisel -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s define i8 @test_vector_reduce_umax_v2i8(<2 x i8> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umax_v2i8: @@ -1525,6 +1525,7 @@ entry: ret i8 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umax_v2i16(<2 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umax_v2i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1569,7 +1570,7 @@ define i16 @test_vector_reduce_umax_v2i16(<2 x i16> %v) { ; GFX9-GISEL-LABEL: test_vector_reduce_umax_v2i16: ; GFX9-GISEL: ; %bb.0: ; %entry ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1583,7 +1584,7 @@ define i16 @test_vector_reduce_umax_v2i16(<2 x i16> %v) { ; GFX10-GISEL-LABEL: test_vector_reduce_umax_v2i16: ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1604,7 +1605,7 @@ define i16 @test_vector_reduce_umax_v2i16(<2 x i16> %v) { ; GFX11-GISEL-LABEL: test_vector_reduce_umax_v2i16: ; GFX11-GISEL: ; %bb.0: ; %entry ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1638,7 +1639,7 @@ define i16 @test_vector_reduce_umax_v2i16(<2 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 ; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1782,6 +1783,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umax_v4i16(<4 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umax_v4i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1841,7 +1843,7 @@ define i16 @test_vector_reduce_umax_v4i16(<4 x i16> %v) { ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1857,7 +1859,7 @@ define i16 @test_vector_reduce_umax_v4i16(<4 x i16> %v) { ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1883,7 +1885,7 @@ define i16 @test_vector_reduce_umax_v4i16(<4 x i16> %v) { ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1921,7 +1923,7 @@ define i16 @test_vector_reduce_umax_v4i16(<4 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] entry: @@ -1929,6 +1931,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umax_v8i16(<8 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umax_v8i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -2017,7 +2020,7 @@ define i16 @test_vector_reduce_umax_v8i16(<8 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2037,7 +2040,7 @@ define i16 @test_vector_reduce_umax_v8i16(<8 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_max_u16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2070,7 +2073,7 @@ define i16 @test_vector_reduce_umax_v8i16(<8 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_max_u16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2116,7 +2119,7 @@ define i16 @test_vector_reduce_umax_v8i16(<8 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_max_u16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2125,6 +2128,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umax_v16i16(<16 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umax_v16i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -2267,7 +2271,7 @@ define i16 @test_vector_reduce_umax_v16i16(<16 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2295,7 +2299,7 @@ define i16 @test_vector_reduce_umax_v16i16(<16 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_max_u16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2343,7 +2347,7 @@ define i16 @test_vector_reduce_umax_v16i16(<16 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_max_u16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2404,7 +2408,7 @@ define i16 @test_vector_reduce_umax_v16i16(<16 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_max_u16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_max_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] diff --git a/llvm/test/CodeGen/AMDGPU/vector-reduce-umin.ll b/llvm/test/CodeGen/AMDGPU/vector-reduce-umin.ll index 1d3b42e..83ecaaa 100644 --- a/llvm/test/CodeGen/AMDGPU/vector-reduce-umin.ll +++ b/llvm/test/CodeGen/AMDGPU/vector-reduce-umin.ll @@ -1,21 +1,21 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck -check-prefixes=GFX7,GFX7-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx801 < %s | FileCheck -check-prefixes=GFX8,GFX8-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx942 < %s | FileCheck -check-prefixes=GFX9,GFX9-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-SDAG %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG,GFX11-SDAG-FAKE16 %s ; FIXME-TRUE16. enable gisel -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL,GFX11-GISEL-FAKE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-TRUE16 %s ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG,GFX12-SDAG-FAKE16 %s -; XUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s +; XUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=+real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-TRUE16 %s +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1200 -mattr=-real-true16 < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL,GFX12-GISEL-FAKE16 %s define i8 @test_vector_reduce_umin_v2i8(<2 x i8> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umin_v2i8: @@ -1271,6 +1271,7 @@ entry: ret i8 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umin_v2i16(<2 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umin_v2i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1312,7 +1313,7 @@ define i16 @test_vector_reduce_umin_v2i16(<2 x i16> %v) { ; GFX9-GISEL-LABEL: test_vector_reduce_umin_v2i16: ; GFX9-GISEL: ; %bb.0: ; %entry ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1326,7 +1327,7 @@ define i16 @test_vector_reduce_umin_v2i16(<2 x i16> %v) { ; GFX10-GISEL-LABEL: test_vector_reduce_umin_v2i16: ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1347,7 +1348,7 @@ define i16 @test_vector_reduce_umin_v2i16(<2 x i16> %v) { ; GFX11-GISEL-LABEL: test_vector_reduce_umin_v2i16: ; GFX11-GISEL: ; %bb.0: ; %entry ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1381,7 +1382,7 @@ define i16 @test_vector_reduce_umin_v2i16(<2 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_samplecnt 0x0 ; GFX12-GISEL-NEXT: s_wait_bvhcnt 0x0 ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1527,6 +1528,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umin_v4i16(<4 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umin_v4i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1583,7 +1585,7 @@ define i16 @test_vector_reduce_umin_v4i16(<4 x i16> %v) { ; GFX9-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1599,7 +1601,7 @@ define i16 @test_vector_reduce_umin_v4i16(<4 x i16> %v) { ; GFX10-GISEL: ; %bb.0: ; %entry ; GFX10-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1625,7 +1627,7 @@ define i16 @test_vector_reduce_umin_v4i16(<4 x i16> %v) { ; GFX11-GISEL-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1663,7 +1665,7 @@ define i16 @test_vector_reduce_umin_v4i16(<4 x i16> %v) { ; GFX12-GISEL-NEXT: s_wait_kmcnt 0x0 ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] entry: @@ -1671,6 +1673,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umin_v8i16(<8 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umin_v8i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -1756,7 +1759,7 @@ define i16 @test_vector_reduce_umin_v8i16(<8 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1776,7 +1779,7 @@ define i16 @test_vector_reduce_umin_v8i16(<8 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_min_u16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -1809,7 +1812,7 @@ define i16 @test_vector_reduce_umin_v8i16(<8 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_min_u16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1855,7 +1858,7 @@ define i16 @test_vector_reduce_umin_v8i16(<8 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_min_u16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -1864,6 +1867,7 @@ entry: ret i16 %res } +; FIXME: With -new-reg-bank-select, v_alignbit_b32 is regression. Need pattern to look through COPY. define i16 @test_vector_reduce_umin_v16i16(<16 x i16> %v) { ; GFX7-SDAG-LABEL: test_vector_reduce_umin_v16i16: ; GFX7-SDAG: ; %bb.0: ; %entry @@ -2003,7 +2007,7 @@ define i16 @test_vector_reduce_umin_v16i16(<16 x i16> %v) { ; GFX9-GISEL-NEXT: s_nop 0 ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_nop 0 -; GFX9-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX9-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX9-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX9-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2031,7 +2035,7 @@ define i16 @test_vector_reduce_umin_v16i16(<16 x i16> %v) { ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v2 ; GFX10-GISEL-NEXT: v_pk_min_u16 v1, v1, v3 ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX10-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX10-GISEL-NEXT: v_alignbit_b32 v1, s4, v0, 16 ; GFX10-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX10-GISEL-NEXT: s_setpc_b64 s[30:31] ; @@ -2079,7 +2083,7 @@ define i16 @test_vector_reduce_umin_v16i16(<16 x i16> %v) { ; GFX11-GISEL-NEXT: v_pk_min_u16 v1, v1, v3 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX11-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX11-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX11-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX11-GISEL-NEXT: s_setpc_b64 s[30:31] @@ -2140,7 +2144,7 @@ define i16 @test_vector_reduce_umin_v16i16(<16 x i16> %v) { ; GFX12-GISEL-NEXT: v_pk_min_u16 v1, v1, v3 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 -; GFX12-GISEL-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-GISEL-NEXT: v_alignbit_b32 v1, s0, v0, 16 ; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX12-GISEL-NEXT: v_pk_min_u16 v0, v0, v1 ; GFX12-GISEL-NEXT: s_setpc_b64 s[30:31] diff --git a/llvm/test/CodeGen/ARM/sincos.ll b/llvm/test/CodeGen/ARM/sincos.ll index e1b683a..1a4313e 100644 --- a/llvm/test/CodeGen/ARM/sincos.ll +++ b/llvm/test/CodeGen/ARM/sincos.ll @@ -2,8 +2,7 @@ ; RUN: llc < %s -mtriple=armv7-apple-ios7 -mcpu=cortex-a8 | FileCheck %s --check-prefix=SINCOS ; RUN: llc < %s -mtriple=armv7-linux-gnu -mcpu=cortex-a8 | FileCheck %s --check-prefix=SINCOS-GNU ; RUN: llc < %s -mtriple=armv7-linux-gnueabi -mcpu=cortex-a8 | FileCheck %s --check-prefix=SINCOS-GNU -; RUN: llc < %s -mtriple=armv7-linux-android -mcpu=cortex-a8 | FileCheck %s --check-prefix=NOOPT-ANDROID -; RUN: llc < %s -mtriple=armv7-linux-android9 -mcpu=cortex-a8 | FileCheck %s --check-prefix=SINCOS-GNU +; RUN: llc < %s -mtriple=armv7-linux-android -mcpu=cortex-a8 | FileCheck %s --check-prefix=SINCOS-GNU ; Combine sin / cos into a single call unless they may write errno (as ; captured by readnone attrbiute, controlled by clang -fmath-errno @@ -22,10 +21,6 @@ entry: ; NOOPT: bl _sinf ; NOOPT: bl _cosf -; NOOPT-ANDROID-LABEL: test1: -; NOOPT-ANDROID: bl sinf -; NOOPT-ANDROID: bl cosf - %call = tail call float @sinf(float %x) readnone %call1 = tail call float @cosf(float %x) readnone %add = fadd float %call, %call1 @@ -44,10 +39,6 @@ entry: ; NOOPT: bl _sinf ; NOOPT: bl _cosf -; NOOPT-ANDROID-LABEL: test1_fast: -; NOOPT-ANDROID: bl sinf -; NOOPT-ANDROID: bl cosf - %call = tail call fast float @sinf(float %x) readnone %call1 = tail call fast float @cosf(float %x) readnone %add = fadd float %call, %call1 @@ -68,10 +59,6 @@ entry: ; NOOPT: bl _sinf ; NOOPT: bl _cosf -; NOOPT-ANDROID-LABEL: test1_errno: -; NOOPT-ANDROID: bl sinf -; NOOPT-ANDROID: bl cosf - %call = tail call float @sinf(float %x) %call1 = tail call float @cosf(float %x) %add = fadd float %call, %call1 @@ -90,10 +77,6 @@ entry: ; NOOPT: bl _sin ; NOOPT: bl _cos -; NOOPT-ANDROID-LABEL: test2: -; NOOPT-ANDROID: bl sin -; NOOPT-ANDROID: bl cos - %call = tail call double @sin(double %x) readnone %call1 = tail call double @cos(double %x) readnone %add = fadd double %call, %call1 @@ -112,10 +95,6 @@ entry: ; NOOPT: bl _sin ; NOOPT: bl _cos -; NOOPT-ANDROID-LABEL: test2_fast: -; NOOPT-ANDROID: bl sin -; NOOPT-ANDROID: bl cos - %call = tail call fast double @sin(double %x) readnone %call1 = tail call fast double @cos(double %x) readnone %add = fadd double %call, %call1 @@ -136,10 +115,6 @@ entry: ; NOOPT: bl _sin ; NOOPT: bl _cos -; NOOPT-ANDROID-LABEL: test2_errno: -; NOOPT-ANDROID: bl sin -; NOOPT-ANDROID: bl cos - %call = tail call double @sin(double %x) %call1 = tail call double @cos(double %x) %add = fadd double %call, %call1 diff --git a/llvm/test/CodeGen/RISCV/cmov-branch-opt.ll b/llvm/test/CodeGen/RISCV/cmov-branch-opt.ll index edec1d0..1957019 100644 --- a/llvm/test/CodeGen/RISCV/cmov-branch-opt.ll +++ b/llvm/test/CodeGen/RISCV/cmov-branch-opt.ll @@ -201,8 +201,9 @@ define signext i32 @test4(i32 signext %x, i32 signext %y, i32 signext %z) { ; ; RV32IXQCI-LABEL: test4: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: li a0, 0 -; RV32IXQCI-NEXT: qc.lieqi a0, a2, 0, 3 +; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: li a1, 3 +; RV32IXQCI-NEXT: qc.selectieqi a0, 0, a1, 0 ; RV32IXQCI-NEXT: ret %c = icmp eq i32 %z, 0 %a = select i1 %c, i32 3, i32 0 diff --git a/llvm/test/CodeGen/RISCV/features-info.ll b/llvm/test/CodeGen/RISCV/features-info.ll index 1a7a72d..693a40d 100644 --- a/llvm/test/CodeGen/RISCV/features-info.ll +++ b/llvm/test/CodeGen/RISCV/features-info.ll @@ -142,6 +142,7 @@ ; CHECK-NEXT: shvstvecd - 'Shvstvecd' (vstvec supports Direct mode). ; CHECK-NEXT: shxadd-load-fusion - Enable SH(1|2|3)ADD(.UW) + load macrofusion. ; CHECK-NEXT: sifive7 - SiFive 7-Series processors. +; CHECK-NEXT: single-element-vec-fp64 - Certain vector FP64 operations produce a single result element per cycle. ; CHECK-NEXT: smaia - 'Smaia' (Advanced Interrupt Architecture Machine Level). ; CHECK-NEXT: smcdeleg - 'Smcdeleg' (Counter Delegation Machine Level). ; CHECK-NEXT: smcntrpmf - 'Smcntrpmf' (Cycle and Instret Privilege Mode Filtering). diff --git a/llvm/test/CodeGen/RISCV/xqcicli.ll b/llvm/test/CodeGen/RISCV/xqcicli.ll index 8d4caa1..cdb1947 100644 --- a/llvm/test/CodeGen/RISCV/xqcicli.ll +++ b/llvm/test/CodeGen/RISCV/xqcicli.ll @@ -23,7 +23,8 @@ define i32 @select_cc_example_eq(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eq: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieq a0, a1, a2, 11 +; RV32IXQCI-NEXT: qc.selectine a1, a2, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %b, %x @@ -47,7 +48,8 @@ define i32 @select_cc_example_ne(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ne: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.line a0, a1, a2, 11 +; RV32IXQCI-NEXT: qc.selectieq a1, a2, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %b, %x @@ -167,7 +169,8 @@ define i32 @select_cc_example_eq_c(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eq_c: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.line a0, a1, a2, 11 +; RV32IXQCI-NEXT: qc.selectieq a1, a2, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %b, %x @@ -191,7 +194,8 @@ define i32 @select_cc_example_ne_c(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ne_c: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieq a0, a1, a2, 11 +; RV32IXQCI-NEXT: qc.selectine a1, a2, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %b, %x @@ -312,7 +316,8 @@ define i32 @select_cc_example_eqi(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eqi: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectinei a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %b, 12 @@ -337,7 +342,8 @@ define i32 @select_cc_example_nei(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_nei: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectieqi a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %b, 12 @@ -462,7 +468,8 @@ define i32 @select_cc_example_eqi_c1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eqi_c1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectinei a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 12, %b @@ -487,7 +494,8 @@ define i32 @select_cc_example_nei_c1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_nei_c1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectieqi a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 12, %b @@ -612,7 +620,8 @@ define i32 @select_cc_example_eqi_c2(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eqi_c2: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectieqi a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 12, %b @@ -637,7 +646,8 @@ define i32 @select_cc_example_nei_c2(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_nei_c2: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectinei a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 12, %b @@ -762,7 +772,8 @@ define i32 @select_cc_example_eqi_c3(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eqi_c3: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectieqi a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %b, 12 @@ -787,7 +798,8 @@ define i32 @select_cc_example_nei_c3(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_nei_c3: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a0, a1, 12, 11 +; RV32IXQCI-NEXT: qc.selectinei a1, 12, a0, 11 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %b, 12 diff --git a/llvm/test/CodeGen/RISCV/xqcics.ll b/llvm/test/CodeGen/RISCV/xqcics.ll index c0839c9..7656a0c 100644 --- a/llvm/test/CodeGen/RISCV/xqcics.ll +++ b/llvm/test/CodeGen/RISCV/xqcics.ll @@ -270,8 +270,7 @@ define i32 @select_cc_example_eqi(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eqi: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.line a2, a0, a1, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieq a0, a1, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, %b @@ -301,8 +300,7 @@ define i32 @select_cc_example_eqi_c(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eqi_c: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieq a2, a0, a1, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectine a0, a1, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, %b @@ -332,8 +330,7 @@ define i32 @select_cc_example_nei(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_nei: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieq a2, a0, a1, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectine a0, a1, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, %b @@ -363,8 +360,7 @@ define i32 @select_cc_example_nei_c(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_nei_c: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.line a2, a0, a1, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieq a0, a1, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, %b @@ -395,8 +391,7 @@ define i32 @select_cc_example_ieqi(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ieqi: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieqi a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, 12 @@ -427,8 +422,7 @@ define i32 @select_cc_example_ieqi_c1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ieqi_c1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieqi a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 12, %a @@ -459,8 +453,7 @@ define i32 @select_cc_example_ieqi_c2(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ieqi_c2: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectinei a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, 12 @@ -491,8 +484,7 @@ define i32 @select_cc_example_ieqi_c3(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ieqi_c3: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectinei a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 12, %a @@ -523,8 +515,7 @@ define i32 @select_cc_example_inei(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_inei: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectinei a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, 12 @@ -555,8 +546,7 @@ define i32 @select_cc_example_inei_c1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_inei_c1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieqi a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectinei a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 12, %a @@ -587,8 +577,7 @@ define i32 @select_cc_example_inei_c2(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_inei_c2: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieqi a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, 12 @@ -619,8 +608,7 @@ define i32 @select_cc_example_inei_c3(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_inei_c3: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.linei a2, a0, 12, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieqi a0, 12, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 12, %a @@ -712,8 +700,7 @@ define i32 @select_cc_example_eq1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_eq1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.line a2, a1, a0, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectieq a0, a1, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %b, %a @@ -743,8 +730,7 @@ define i32 @select_cc_example_ne1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCI-LABEL: select_cc_example_ne1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.lieq a2, a1, a0, 11 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.selectine a0, a1, a2, 11 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %b, %a diff --git a/llvm/test/CodeGen/X86/fast-isel-fneg.ll b/llvm/test/CodeGen/X86/fast-isel-fneg.ll deleted file mode 100644 index 128f5ee..0000000 --- a/llvm/test/CodeGen/X86/fast-isel-fneg.ll +++ /dev/null @@ -1,101 +0,0 @@ -; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -fast-isel -fast-isel-abort=3 -mtriple=x86_64-apple-darwin10 | FileCheck %s -; RUN: llc < %s -fast-isel -mtriple=i686-- -mattr=+sse2 | FileCheck --check-prefix=SSE2 %s - -define double @fneg_f64(double %x) nounwind { -; CHECK-LABEL: fneg_f64: -; CHECK: ## %bb.0: -; CHECK-NEXT: movq %xmm0, %rax -; CHECK-NEXT: movabsq $-9223372036854775808, %rcx ## imm = 0x8000000000000000 -; CHECK-NEXT: xorq %rax, %rcx -; CHECK-NEXT: movq %rcx, %xmm0 -; CHECK-NEXT: retq -; -; SSE2-LABEL: fneg_f64: -; SSE2: # %bb.0: -; SSE2-NEXT: pushl %ebp -; SSE2-NEXT: movl %esp, %ebp -; SSE2-NEXT: andl $-8, %esp -; SSE2-NEXT: subl $8, %esp -; SSE2-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero -; SSE2-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 -; SSE2-NEXT: movlps %xmm0, (%esp) -; SSE2-NEXT: fldl (%esp) -; SSE2-NEXT: movl %ebp, %esp -; SSE2-NEXT: popl %ebp -; SSE2-NEXT: retl - %y = fneg double %x - ret double %y -} - -define float @fneg_f32(float %x) nounwind { -; CHECK-LABEL: fneg_f32: -; CHECK: ## %bb.0: -; CHECK-NEXT: movd %xmm0, %eax -; CHECK-NEXT: xorl $2147483648, %eax ## imm = 0x80000000 -; CHECK-NEXT: movd %eax, %xmm0 -; CHECK-NEXT: retq -; -; SSE2-LABEL: fneg_f32: -; SSE2: # %bb.0: -; SSE2-NEXT: pushl %eax -; SSE2-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero -; SSE2-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 -; SSE2-NEXT: movss %xmm0, (%esp) -; SSE2-NEXT: flds (%esp) -; SSE2-NEXT: popl %eax -; SSE2-NEXT: retl - %y = fneg float %x - ret float %y -} - -define void @fneg_f64_mem(ptr %x, ptr %y) nounwind { -; CHECK-LABEL: fneg_f64_mem: -; CHECK: ## %bb.0: -; CHECK-NEXT: movq {{.*#+}} xmm0 = mem[0],zero -; CHECK-NEXT: movq %xmm0, %rax -; CHECK-NEXT: movabsq $-9223372036854775808, %rcx ## imm = 0x8000000000000000 -; CHECK-NEXT: xorq %rax, %rcx -; CHECK-NEXT: movq %rcx, %xmm0 -; CHECK-NEXT: movq %xmm0, (%rsi) -; CHECK-NEXT: retq -; -; SSE2-LABEL: fneg_f64_mem: -; SSE2: # %bb.0: -; SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax -; SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx -; SSE2-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero -; SSE2-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 -; SSE2-NEXT: movsd %xmm0, (%eax) -; SSE2-NEXT: retl - %a = load double, ptr %x - %b = fneg double %a - store double %b, ptr %y - ret void -} - -define void @fneg_f32_mem(ptr %x, ptr %y) nounwind { -; CHECK-LABEL: fneg_f32_mem: -; CHECK: ## %bb.0: -; CHECK-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero -; CHECK-NEXT: movd %xmm0, %eax -; CHECK-NEXT: xorl $2147483648, %eax ## imm = 0x80000000 -; CHECK-NEXT: movd %eax, %xmm0 -; CHECK-NEXT: movd %xmm0, (%rsi) -; CHECK-NEXT: retq -; -; SSE2-LABEL: fneg_f32_mem: -; SSE2: # %bb.0: -; SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax -; SSE2-NEXT: movl {{[0-9]+}}(%esp), %ecx -; SSE2-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero -; SSE2-NEXT: movd %xmm0, %ecx -; SSE2-NEXT: xorl $2147483648, %ecx # imm = 0x80000000 -; SSE2-NEXT: movd %ecx, %xmm0 -; SSE2-NEXT: movd %xmm0, (%eax) -; SSE2-NEXT: retl - %a = load float, ptr %x - %b = fneg float %a - store float %b, ptr %y - ret void -} diff --git a/llvm/test/CodeGen/X86/fp-int-fp-cvt.ll b/llvm/test/CodeGen/X86/fp-int-fp-cvt.ll new file mode 100644 index 0000000..b6c17ce --- /dev/null +++ b/llvm/test/CodeGen/X86/fp-int-fp-cvt.ll @@ -0,0 +1,240 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py +; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64 | FileCheck %s --check-prefixes=SSE +; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE +; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX,AVX2 +; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX,AVX512,AVX512-VL +; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=x86-64-v4 -mattr=-avx512vl | FileCheck %s --check-prefixes=AVX,AVX512,AVX512-NOVL + +; +; fptosi -> sitofp +; + +define double @scvtf64_i32(double %a0) { +; SSE-LABEL: scvtf64_i32: +; SSE: # %bb.0: +; SSE-NEXT: cvttpd2dq %xmm0, %xmm0 +; SSE-NEXT: cvtdq2pd %xmm0, %xmm0 +; SSE-NEXT: retq +; +; AVX-LABEL: scvtf64_i32: +; AVX: # %bb.0: +; AVX-NEXT: vcvttpd2dq %xmm0, %xmm0 +; AVX-NEXT: vcvtdq2pd %xmm0, %xmm0 +; AVX-NEXT: retq + %ii = fptosi double %a0 to i32 + %ff = sitofp i32 %ii to double + ret double %ff +} + +define double @scvtf64_i64(double %a0) { +; SSE-LABEL: scvtf64_i64: +; SSE: # %bb.0: +; SSE-NEXT: cvttsd2si %xmm0, %rax +; SSE-NEXT: xorps %xmm0, %xmm0 +; SSE-NEXT: cvtsi2sd %rax, %xmm0 +; SSE-NEXT: retq +; +; AVX-LABEL: scvtf64_i64: +; AVX: # %bb.0: +; AVX-NEXT: vcvttsd2si %xmm0, %rax +; AVX-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0 +; AVX-NEXT: retq + %ii = fptosi double %a0 to i64 + %ff = sitofp i64 %ii to double + ret double %ff +} + +define float @scvtf32_i32(float %a0) { +; SSE-LABEL: scvtf32_i32: +; SSE: # %bb.0: +; SSE-NEXT: cvttps2dq %xmm0, %xmm0 +; SSE-NEXT: cvtdq2ps %xmm0, %xmm0 +; SSE-NEXT: retq +; +; AVX-LABEL: scvtf32_i32: +; AVX: # %bb.0: +; AVX-NEXT: vcvttps2dq %xmm0, %xmm0 +; AVX-NEXT: vcvtdq2ps %xmm0, %xmm0 +; AVX-NEXT: retq + %ii = fptosi float %a0 to i32 + %ff = sitofp i32 %ii to float + ret float %ff +} + +define float @scvtf32_i64(float %a0) { +; SSE-LABEL: scvtf32_i64: +; SSE: # %bb.0: +; SSE-NEXT: cvttss2si %xmm0, %rax +; SSE-NEXT: xorps %xmm0, %xmm0 +; SSE-NEXT: cvtsi2ss %rax, %xmm0 +; SSE-NEXT: retq +; +; AVX-LABEL: scvtf32_i64: +; AVX: # %bb.0: +; AVX-NEXT: vcvttss2si %xmm0, %rax +; AVX-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0 +; AVX-NEXT: retq + %ii = fptosi float %a0 to i64 + %ff = sitofp i64 %ii to float + ret float %ff +} + +; +; fptoui -> uitofp +; + +define double @ucvtf64_i32(double %a0) { +; SSE-LABEL: ucvtf64_i32: +; SSE: # %bb.0: +; SSE-NEXT: cvttsd2si %xmm0, %rax +; SSE-NEXT: movl %eax, %eax +; SSE-NEXT: xorps %xmm0, %xmm0 +; SSE-NEXT: cvtsi2sd %rax, %xmm0 +; SSE-NEXT: retq +; +; AVX2-LABEL: ucvtf64_i32: +; AVX2: # %bb.0: +; AVX2-NEXT: vcvttsd2si %xmm0, %rax +; AVX2-NEXT: movl %eax, %eax +; AVX2-NEXT: vcvtsi2sd %rax, %xmm15, %xmm0 +; AVX2-NEXT: retq +; +; AVX512-LABEL: ucvtf64_i32: +; AVX512: # %bb.0: +; AVX512-NEXT: vcvttsd2usi %xmm0, %eax +; AVX512-NEXT: vcvtusi2sd %eax, %xmm15, %xmm0 +; AVX512-NEXT: retq + %ii = fptoui double %a0 to i32 + %ff = uitofp i32 %ii to double + ret double %ff +} + +define double @ucvtf64_i64(double %a0) { +; SSE-LABEL: ucvtf64_i64: +; SSE: # %bb.0: +; SSE-NEXT: cvttsd2si %xmm0, %rax +; SSE-NEXT: movq %rax, %rcx +; SSE-NEXT: subsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 +; SSE-NEXT: cvttsd2si %xmm0, %rdx +; SSE-NEXT: sarq $63, %rcx +; SSE-NEXT: andq %rcx, %rdx +; SSE-NEXT: orq %rax, %rdx +; SSE-NEXT: movq %rdx, %xmm1 +; SSE-NEXT: punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1] +; SSE-NEXT: subpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1 +; SSE-NEXT: movapd %xmm1, %xmm0 +; SSE-NEXT: unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1] +; SSE-NEXT: addsd %xmm1, %xmm0 +; SSE-NEXT: retq +; +; AVX2-LABEL: ucvtf64_i64: +; AVX2: # %bb.0: +; AVX2-NEXT: vcvttsd2si %xmm0, %rax +; AVX2-NEXT: movq %rax, %rcx +; AVX2-NEXT: vsubsd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: sarq $63, %rcx +; AVX2-NEXT: vcvttsd2si %xmm0, %rdx +; AVX2-NEXT: andq %rcx, %rdx +; AVX2-NEXT: orq %rax, %rdx +; AVX2-NEXT: vmovq %rdx, %xmm0 +; AVX2-NEXT: vpunpckldq {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1] +; AVX2-NEXT: vsubpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: vshufpd {{.*#+}} xmm1 = xmm0[1,0] +; AVX2-NEXT: vaddsd %xmm0, %xmm1, %xmm0 +; AVX2-NEXT: retq +; +; AVX512-LABEL: ucvtf64_i64: +; AVX512: # %bb.0: +; AVX512-NEXT: vcvttsd2usi %xmm0, %rax +; AVX512-NEXT: vcvtusi2sd %rax, %xmm15, %xmm0 +; AVX512-NEXT: retq + %ii = fptoui double %a0 to i64 + %ff = uitofp i64 %ii to double + ret double %ff +} + +define float @ucvtf32_i32(float %a0) { +; SSE-LABEL: ucvtf32_i32: +; SSE: # %bb.0: +; SSE-NEXT: cvttss2si %xmm0, %rax +; SSE-NEXT: movl %eax, %eax +; SSE-NEXT: xorps %xmm0, %xmm0 +; SSE-NEXT: cvtsi2ss %rax, %xmm0 +; SSE-NEXT: retq +; +; AVX2-LABEL: ucvtf32_i32: +; AVX2: # %bb.0: +; AVX2-NEXT: vcvttss2si %xmm0, %rax +; AVX2-NEXT: movl %eax, %eax +; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0 +; AVX2-NEXT: retq +; +; AVX512-LABEL: ucvtf32_i32: +; AVX512: # %bb.0: +; AVX512-NEXT: vcvttss2usi %xmm0, %eax +; AVX512-NEXT: vcvtusi2ss %eax, %xmm15, %xmm0 +; AVX512-NEXT: retq + %ii = fptoui float %a0 to i32 + %ff = uitofp i32 %ii to float + ret float %ff +} + +define float @ucvtf32_i64(float %a0) { +; SSE-LABEL: ucvtf32_i64: +; SSE: # %bb.0: +; SSE-NEXT: cvttss2si %xmm0, %rcx +; SSE-NEXT: movq %rcx, %rdx +; SSE-NEXT: sarq $63, %rdx +; SSE-NEXT: subss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 +; SSE-NEXT: cvttss2si %xmm0, %rax +; SSE-NEXT: andq %rdx, %rax +; SSE-NEXT: orq %rcx, %rax +; SSE-NEXT: js .LBB7_1 +; SSE-NEXT: # %bb.2: +; SSE-NEXT: xorps %xmm0, %xmm0 +; SSE-NEXT: cvtsi2ss %rax, %xmm0 +; SSE-NEXT: retq +; SSE-NEXT: .LBB7_1: +; SSE-NEXT: movq %rax, %rcx +; SSE-NEXT: shrq %rcx +; SSE-NEXT: andl $1, %eax +; SSE-NEXT: orq %rcx, %rax +; SSE-NEXT: xorps %xmm0, %xmm0 +; SSE-NEXT: cvtsi2ss %rax, %xmm0 +; SSE-NEXT: addss %xmm0, %xmm0 +; SSE-NEXT: retq +; +; AVX2-LABEL: ucvtf32_i64: +; AVX2: # %bb.0: +; AVX2-NEXT: vcvttss2si %xmm0, %rcx +; AVX2-NEXT: movq %rcx, %rdx +; AVX2-NEXT: sarq $63, %rdx +; AVX2-NEXT: vsubss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 +; AVX2-NEXT: vcvttss2si %xmm0, %rax +; AVX2-NEXT: andq %rdx, %rax +; AVX2-NEXT: orq %rcx, %rax +; AVX2-NEXT: js .LBB7_1 +; AVX2-NEXT: # %bb.2: +; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0 +; AVX2-NEXT: retq +; AVX2-NEXT: .LBB7_1: +; AVX2-NEXT: movq %rax, %rcx +; AVX2-NEXT: shrq %rcx +; AVX2-NEXT: andl $1, %eax +; AVX2-NEXT: orq %rcx, %rax +; AVX2-NEXT: vcvtsi2ss %rax, %xmm15, %xmm0 +; AVX2-NEXT: vaddss %xmm0, %xmm0, %xmm0 +; AVX2-NEXT: retq +; +; AVX512-LABEL: ucvtf32_i64: +; AVX512: # %bb.0: +; AVX512-NEXT: vcvttss2usi %xmm0, %rax +; AVX512-NEXT: vcvtusi2ss %rax, %xmm15, %xmm0 +; AVX512-NEXT: retq + %ii = fptoui float %a0 to i64 + %ff = uitofp i64 %ii to float + ret float %ff +} +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; AVX512-NOVL: {{.*}} +; AVX512-VL: {{.*}} diff --git a/llvm/test/CodeGen/X86/isel-fneg.ll b/llvm/test/CodeGen/X86/isel-fneg.ll new file mode 100644 index 0000000..77b3f26 --- /dev/null +++ b/llvm/test/CodeGen/X86/isel-fneg.ll @@ -0,0 +1,208 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc < %s -mtriple=i686-linux-gnu -fast-isel | FileCheck %s --check-prefixes=X86,FASTISEL-X86 +; RUN: llc < %s -mtriple=i686-linux-gnu -global-isel=0 -fast-isel=0 | FileCheck %s --check-prefixes=X86,SDAG-X86 +; DISABLED: llc < %s -mtriple=i686-linux-gnu -global-isel=1 -global-isel-abort=2 | FileCheck %s --check-prefixes=X86,GISEL-X86 +; RUN: llc < %s -mtriple=i686-linux-gnu -fast-isel -mattr=+sse | FileCheck %s --check-prefixes=X86,SSE-X86,FASTISEL-SSE-X86 +; RUN: llc < %s -mtriple=i686-linux-gnu -global-isel=0 -fast-isel=0 -mattr=+sse | FileCheck %s --check-prefixes=X86,SSE-X86,SDAG-SSE-X86 +; DISABLED: llc < %s -mtriple=i686-linux-gnu -global-isel=1 -global-isel-abort=2 -mattr=+sse | FileCheck %s --check-prefixes=X86,SSE-X86,GISEL-SSE-X86 +; RUN: llc < %s -mtriple=x86_64-linux-gnu -fast-isel -mattr=+sse | FileCheck %s --check-prefixes=X64,SSE-X64,FASTISEL-SSE-X64 +; RUN: llc < %s -mtriple=x86_64-linux-gnu -global-isel=0 -fast-isel=0 -mattr=+sse | FileCheck %s --check-prefixes=X64,SSE-X64,SDAG-SSE-X64 +; RUN: llc < %s -mtriple=x86_64-linux-gnu -global-isel=1 -global-isel-abort=2 -mattr=+sse | FileCheck %s --check-prefixes=X64,SSE-X64,GISEL-SSE-X64 + +define double @fneg_f64(double %x) nounwind { +; X86-LABEL: fneg_f64: +; X86: # %bb.0: +; X86-NEXT: fldl {{[0-9]+}}(%esp) +; X86-NEXT: fchs +; X86-NEXT: retl +; +; FASTISEL-SSE-X64-LABEL: fneg_f64: +; FASTISEL-SSE-X64: # %bb.0: +; FASTISEL-SSE-X64-NEXT: movq %xmm0, %rax +; FASTISEL-SSE-X64-NEXT: movabsq $-9223372036854775808, %rcx # imm = 0x8000000000000000 +; FASTISEL-SSE-X64-NEXT: xorq %rax, %rcx +; FASTISEL-SSE-X64-NEXT: movq %rcx, %xmm0 +; FASTISEL-SSE-X64-NEXT: retq +; +; SDAG-SSE-X64-LABEL: fneg_f64: +; SDAG-SSE-X64: # %bb.0: +; SDAG-SSE-X64-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 +; SDAG-SSE-X64-NEXT: retq +; +; GISEL-SSE-X64-LABEL: fneg_f64: +; GISEL-SSE-X64: # %bb.0: +; GISEL-SSE-X64-NEXT: movabsq $-9223372036854775808, %rax # imm = 0x8000000000000000 +; GISEL-SSE-X64-NEXT: movq %xmm0, %rcx +; GISEL-SSE-X64-NEXT: xorq %rax, %rcx +; GISEL-SSE-X64-NEXT: movq %rcx, %xmm0 +; GISEL-SSE-X64-NEXT: retq + %y = fneg double %x + ret double %y +} + +define float @fneg_f32(float %x) nounwind { +; FASTISEL-X86-LABEL: fneg_f32: +; FASTISEL-X86: # %bb.0: +; FASTISEL-X86-NEXT: flds {{[0-9]+}}(%esp) +; FASTISEL-X86-NEXT: fchs +; FASTISEL-X86-NEXT: retl +; +; SDAG-X86-LABEL: fneg_f32: +; SDAG-X86: # %bb.0: +; SDAG-X86-NEXT: flds {{[0-9]+}}(%esp) +; SDAG-X86-NEXT: fchs +; SDAG-X86-NEXT: retl +; +; SSE-X86-LABEL: fneg_f32: +; SSE-X86: # %bb.0: +; SSE-X86-NEXT: pushl %eax +; SSE-X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; SSE-X86-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 +; SSE-X86-NEXT: movss %xmm0, (%esp) +; SSE-X86-NEXT: flds (%esp) +; SSE-X86-NEXT: popl %eax +; SSE-X86-NEXT: retl +; +; FASTISEL-SSE-X64-LABEL: fneg_f32: +; FASTISEL-SSE-X64: # %bb.0: +; FASTISEL-SSE-X64-NEXT: movd %xmm0, %eax +; FASTISEL-SSE-X64-NEXT: xorl $2147483648, %eax # imm = 0x80000000 +; FASTISEL-SSE-X64-NEXT: movd %eax, %xmm0 +; FASTISEL-SSE-X64-NEXT: retq +; +; SDAG-SSE-X64-LABEL: fneg_f32: +; SDAG-SSE-X64: # %bb.0: +; SDAG-SSE-X64-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0 +; SDAG-SSE-X64-NEXT: retq +; +; GISEL-SSE-X64-LABEL: fneg_f32: +; GISEL-SSE-X64: # %bb.0: +; GISEL-SSE-X64-NEXT: movd %xmm0, %eax +; GISEL-SSE-X64-NEXT: addl $-2147483648, %eax # imm = 0x80000000 +; GISEL-SSE-X64-NEXT: movd %eax, %xmm0 +; GISEL-SSE-X64-NEXT: retq + %y = fneg float %x + ret float %y +} + +define void @fneg_f64_mem(ptr %x, ptr %y) nounwind { +; X86-LABEL: fneg_f64_mem: +; X86: # %bb.0: +; X86-NEXT: movl {{[0-9]+}}(%esp), %eax +; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-NEXT: fldl (%ecx) +; X86-NEXT: fchs +; X86-NEXT: fstpl (%eax) +; X86-NEXT: retl +; +; FASTISEL-SSE-X64-LABEL: fneg_f64_mem: +; FASTISEL-SSE-X64: # %bb.0: +; FASTISEL-SSE-X64-NEXT: movq {{.*#+}} xmm0 = mem[0],zero +; FASTISEL-SSE-X64-NEXT: movq %xmm0, %rax +; FASTISEL-SSE-X64-NEXT: movabsq $-9223372036854775808, %rcx # imm = 0x8000000000000000 +; FASTISEL-SSE-X64-NEXT: xorq %rax, %rcx +; FASTISEL-SSE-X64-NEXT: movq %rcx, %xmm0 +; FASTISEL-SSE-X64-NEXT: movq %xmm0, (%rsi) +; FASTISEL-SSE-X64-NEXT: retq +; +; SDAG-SSE-X64-LABEL: fneg_f64_mem: +; SDAG-SSE-X64: # %bb.0: +; SDAG-SSE-X64-NEXT: movabsq $-9223372036854775808, %rax # imm = 0x8000000000000000 +; SDAG-SSE-X64-NEXT: xorq (%rdi), %rax +; SDAG-SSE-X64-NEXT: movq %rax, (%rsi) +; SDAG-SSE-X64-NEXT: retq +; +; GISEL-SSE-X64-LABEL: fneg_f64_mem: +; GISEL-SSE-X64: # %bb.0: +; GISEL-SSE-X64-NEXT: movabsq $-9223372036854775808, %rax # imm = 0x8000000000000000 +; GISEL-SSE-X64-NEXT: xorq (%rdi), %rax +; GISEL-SSE-X64-NEXT: movq %rax, (%rsi) +; GISEL-SSE-X64-NEXT: retq + %a = load double, ptr %x + %b = fneg double %a + store double %b, ptr %y + ret void +} + +define void @fneg_f32_mem(ptr %x, ptr %y) nounwind { +; FASTISEL-X86-LABEL: fneg_f32_mem: +; FASTISEL-X86: # %bb.0: +; FASTISEL-X86-NEXT: movl {{[0-9]+}}(%esp), %eax +; FASTISEL-X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; FASTISEL-X86-NEXT: movl $-2147483648, %edx # imm = 0x80000000 +; FASTISEL-X86-NEXT: xorl (%ecx), %edx +; FASTISEL-X86-NEXT: movl %edx, (%eax) +; FASTISEL-X86-NEXT: retl +; +; SDAG-X86-LABEL: fneg_f32_mem: +; SDAG-X86: # %bb.0: +; SDAG-X86-NEXT: movl {{[0-9]+}}(%esp), %eax +; SDAG-X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; SDAG-X86-NEXT: movl $-2147483648, %edx # imm = 0x80000000 +; SDAG-X86-NEXT: xorl (%ecx), %edx +; SDAG-X86-NEXT: movl %edx, (%eax) +; SDAG-X86-NEXT: retl +; +; FASTISEL-SSE-X86-LABEL: fneg_f32_mem: +; FASTISEL-SSE-X86: # %bb.0: +; FASTISEL-SSE-X86-NEXT: movl {{[0-9]+}}(%esp), %eax +; FASTISEL-SSE-X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; FASTISEL-SSE-X86-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; FASTISEL-SSE-X86-NEXT: xorps {{\.?LCPI[0-9]+_[0-9]+}}, %xmm0 +; FASTISEL-SSE-X86-NEXT: movss %xmm0, (%eax) +; FASTISEL-SSE-X86-NEXT: retl +; +; SDAG-SSE-X86-LABEL: fneg_f32_mem: +; SDAG-SSE-X86: # %bb.0: +; SDAG-SSE-X86-NEXT: movl {{[0-9]+}}(%esp), %eax +; SDAG-SSE-X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; SDAG-SSE-X86-NEXT: movl $-2147483648, %edx # imm = 0x80000000 +; SDAG-SSE-X86-NEXT: xorl (%ecx), %edx +; SDAG-SSE-X86-NEXT: movl %edx, (%eax) +; SDAG-SSE-X86-NEXT: retl +; +; FASTISEL-SSE-X64-LABEL: fneg_f32_mem: +; FASTISEL-SSE-X64: # %bb.0: +; FASTISEL-SSE-X64-NEXT: movd {{.*#+}} xmm0 = mem[0],zero,zero,zero +; FASTISEL-SSE-X64-NEXT: movd %xmm0, %eax +; FASTISEL-SSE-X64-NEXT: xorl $2147483648, %eax # imm = 0x80000000 +; FASTISEL-SSE-X64-NEXT: movd %eax, %xmm0 +; FASTISEL-SSE-X64-NEXT: movd %xmm0, (%rsi) +; FASTISEL-SSE-X64-NEXT: retq +; +; SDAG-SSE-X64-LABEL: fneg_f32_mem: +; SDAG-SSE-X64: # %bb.0: +; SDAG-SSE-X64-NEXT: movl $-2147483648, %eax # imm = 0x80000000 +; SDAG-SSE-X64-NEXT: xorl (%rdi), %eax +; SDAG-SSE-X64-NEXT: movl %eax, (%rsi) +; SDAG-SSE-X64-NEXT: retq +; +; GISEL-SSE-X64-LABEL: fneg_f32_mem: +; GISEL-SSE-X64: # %bb.0: +; GISEL-SSE-X64-NEXT: movl $-2147483648, %eax # imm = 0x80000000 +; GISEL-SSE-X64-NEXT: xorl (%rdi), %eax +; GISEL-SSE-X64-NEXT: movl %eax, (%rsi) +; GISEL-SSE-X64-NEXT: retq + %a = load float, ptr %x + %b = fneg float %a + store float %b, ptr %y + ret void +} + +define x86_fp80 @test_fp80(x86_fp80 %a) nounwind { +; X86-LABEL: test_fp80: +; X86: # %bb.0: +; X86-NEXT: fldt {{[0-9]+}}(%esp) +; X86-NEXT: fchs +; X86-NEXT: retl +; +; X64-LABEL: test_fp80: +; X64: # %bb.0: +; X64-NEXT: fldt {{[0-9]+}}(%rsp) +; X64-NEXT: fchs +; X64-NEXT: retq + %1 = fneg x86_fp80 %a + ret x86_fp80 %1 +} +;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line: +; SSE-X64: {{.*}} diff --git a/llvm/test/CodeGen/X86/stack-protector-target.ll b/llvm/test/CodeGen/X86/stack-protector-target.ll index f7c5680..4ba0302 100644 --- a/llvm/test/CodeGen/X86/stack-protector-target.ll +++ b/llvm/test/CodeGen/X86/stack-protector-target.ll @@ -2,13 +2,8 @@ ; RUN: llc -mtriple=i386-linux < %s -o - | FileCheck --check-prefix=I386-TLS %s ; RUN: llc -mtriple=x86_64-linux < %s -o - | FileCheck --check-prefix=X64-TLS %s -; RUN: llc -mtriple=i386-linux-android < %s -o - | FileCheck --check-prefix=I386 %s -; RUN: llc -mtriple=i386-linux-android16 < %s -o - | FileCheck --check-prefix=I386 %s -; RUN: llc -mtriple=i386-linux-android17 < %s -o - | FileCheck --check-prefix=I386-TLS %s -; RUN: llc -mtriple=i386-linux-android24 < %s -o - | FileCheck --check-prefix=I386-TLS %s +; RUN: llc -mtriple=i386-linux-android < %s -o - | FileCheck --check-prefix=I386-TLS %s ; RUN: llc -mtriple=x86_64-linux-android < %s -o - | FileCheck --check-prefix=X64-TLS %s -; RUN: llc -mtriple=x86_64-linux-android17 < %s -o - | FileCheck --check-prefix=X64-TLS %s -; RUN: llc -mtriple=x86_64-linux-android24 < %s -o - | FileCheck --check-prefix=X64-TLS %s ; RUN: llc -mtriple=i386-kfreebsd < %s -o - | FileCheck --check-prefix=I386-TLS %s ; RUN: llc -mtriple=x86_64-kfreebsd < %s -o - | FileCheck --check-prefix=X64-TLS %s @@ -27,11 +22,6 @@ declare void @_Z7CapturePi(ptr) ; X64-TLS: movq %fs:40, %[[C:.*]] ; X64-TLS: cmpq 16(%rsp), %[[C]] -; I386: movl __stack_chk_guard, %[[B:.*]] -; I386: movl %[[B]], 8(%esp) -; I386: movl __stack_chk_guard, %[[C:.*]] -; I386: cmpl 8(%esp), %[[C]] - ; I386-TLS: movl %gs:20, %[[B:.*]] ; I386-TLS: movl %[[B]], 8(%esp) ; I386-TLS: movl %gs:20, %[[C:.*]] diff --git a/llvm/test/DebugInfo/X86/DW_OP_LLVM_extract_bits.ll b/llvm/test/DebugInfo/X86/DW_OP_LLVM_extract_bits.ll index 8342d42..5928127 100644 --- a/llvm/test/DebugInfo/X86/DW_OP_LLVM_extract_bits.ll +++ b/llvm/test/DebugInfo/X86/DW_OP_LLVM_extract_bits.ll @@ -67,6 +67,8 @@ entry: ; CHECK: DW_TAG_variable ; CHECK: DW_AT_location (DW_OP_fbreg -4, DW_OP_plus_uconst 0x3, DW_OP_deref_size 0x1, DW_OP_constu 0x38, DW_OP_shl, DW_OP_constu 0x39, DW_OP_shr, DW_OP_stack_value) ; CHECK: DW_AT_name ("z") +; CHECK: DW_AT_location (DW_OP_fbreg -4, DW_OP_deref_size 0x8, DW_OP_stack_value) +; CHECK: DW_AT_name ("q") define i32 @test4() !dbg !28 { entry: @@ -74,6 +76,7 @@ entry: tail call void @llvm.dbg.declare(metadata ptr %0, metadata !29, metadata !DIExpression(DW_OP_LLVM_extract_bits_zext, 31, 1)), !dbg !30 tail call void @llvm.dbg.declare(metadata ptr %0, metadata !31, metadata !DIExpression(DW_OP_LLVM_extract_bits_sext, 1, 31)), !dbg !30 tail call void @llvm.dbg.declare(metadata ptr %0, metadata !32, metadata !DIExpression(DW_OP_plus_uconst, 3, DW_OP_LLVM_extract_bits_zext, 1, 7)), !dbg !30 + tail call void @llvm.dbg.declare(metadata ptr %0, metadata !33, metadata !DIExpression(DW_OP_LLVM_extract_bits_zext, 0, 64)), !dbg !30 ret i32 0, !dbg !30 } @@ -116,3 +119,5 @@ declare void @llvm.dbg.value(metadata, metadata, metadata) !30 = !DILocation(line: 0, scope: !28) !31 = !DILocalVariable(name: "y", scope: !28, file: !3, type: !19) !32 = !DILocalVariable(name: "z", scope: !28, file: !3, type: !9) +!33 = !DILocalVariable(name: "q", scope: !28, file: !3, type: !34) +!34 = !DIBasicType(name: "uint64_t", size: 64, encoding: DW_ATE_unsigned) diff --git a/llvm/test/Instrumentation/AddressSanitizer/with-ifunc.ll b/llvm/test/Instrumentation/AddressSanitizer/with-ifunc.ll index f198fc0..659d6aa 100644 --- a/llvm/test/Instrumentation/AddressSanitizer/with-ifunc.ll +++ b/llvm/test/Instrumentation/AddressSanitizer/with-ifunc.ll @@ -4,13 +4,10 @@ ; RUN: opt -passes=asan -S -asan-with-ifunc=1 -asan-with-ifunc-suppress-remat=0 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-IFUNC ; RUN: opt -passes=asan -S -asan-with-ifunc=1 -asan-with-ifunc-suppress-remat=1 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-IFUNC-NOREMAT -; Pre-Lollipop Android does not support ifunc. -; RUN: opt -passes=asan -S -asan-with-ifunc=1 -asan-with-ifunc-suppress-remat=0 -mtriple=armv7-linux-android20 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-NOIFUNC -; RUN: opt -passes=asan -S -asan-with-ifunc=1 -asan-with-ifunc-suppress-remat=0 -mtriple=armv7-linux-android < %s | FileCheck %s --check-prefixes=CHECK,CHECK-NOIFUNC -; RUN: opt -passes=asan -S -asan-with-ifunc=1 -asan-with-ifunc-suppress-remat=0 -mtriple=armv7-linux-android21 < %s | FileCheck %s --check-prefixes=CHECK,CHECK-IFUNC +; RUN: opt -passes=asan -S -asan-with-ifunc=1 -asan-with-ifunc-suppress-remat=0 -mtriple=armv7-linux-android < %s | FileCheck %s --check-prefixes=CHECK,CHECK-IFUNC target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64" -target triple = "armv7--linux-android22" +target triple = "armv7--linux-android" ; CHECK-IFUNC: @__asan_shadow = external global [0 x i8] ; CHECK-NOIFUNC: @__asan_shadow_memory_dynamic_address = external global i32 diff --git a/llvm/test/MC/Mips/branch-pseudos-bad.s b/llvm/test/MC/Mips/branch-pseudos-bad.s index c23164d..9633414 100644 --- a/llvm/test/MC/Mips/branch-pseudos-bad.s +++ b/llvm/test/MC/Mips/branch-pseudos-bad.s @@ -1,5 +1,13 @@ # RUN: not llvm-mc %s -triple=mips -mcpu=mips32 2>&1 | FileCheck %s +# CHECK: error: invalid operand for instruction + beql $t0, ($t0), 1 +# CHECK: error: invalid operand for instruction + bne $t0, ($t0), 1 +# CHECK: error: invalid operand for instruction + beq $t0, ($t0), 1 + + # Check for errors when using conditional branch pseudos after .set noat. .set noat local_label: diff --git a/llvm/test/TableGen/RuntimeLibcallEmitter-calling-conv.td b/llvm/test/TableGen/RuntimeLibcallEmitter-calling-conv.td index 98a376b..2904474 100644 --- a/llvm/test/TableGen/RuntimeLibcallEmitter-calling-conv.td +++ b/llvm/test/TableGen/RuntimeLibcallEmitter-calling-conv.td @@ -48,6 +48,11 @@ def MSP430LibraryWithCondCC : SystemRuntimeLibrary<isMSP430, // CHECK-NEXT: Entry = DefaultCC; // CHECK-NEXT: } // CHECK-EMPTY: +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x0000000000001a +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: // CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { // CHECK-NEXT: {RTLIB::MALLOC, RTLIB::impl_malloc}, // malloc // CHECK-NEXT: }; @@ -70,9 +75,14 @@ def MSP430LibraryWithCondCC : SystemRuntimeLibrary<isMSP430, // CHECK-NEXT: } // CHECK-EMPTY: // CHECK-NEXT: if (TT.getArch() == Triple::avr) { -// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { -// CHECK-NEXT: {RTLIB::MALLOC, RTLIB::impl_malloc}, // malloc -// CHECK-NEXT: }; +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x0000000000001a +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: +// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { +// CHECK-NEXT: {RTLIB::MALLOC, RTLIB::impl_malloc}, // malloc +// CHECK-NEXT: }; // CHECK-EMPTY: // CHECK-NEXT: for (const auto [Func, Impl] : LibraryCalls) { // CHECK-NEXT: setLibcallImpl(Func, Impl); @@ -92,6 +102,11 @@ def MSP430LibraryWithCondCC : SystemRuntimeLibrary<isMSP430, // CHECK-NEXT: } // CHECK-EMPTY: // CHECK-NEXT: if (TT.getArch() == Triple::msp430) { +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x00000000000010 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: // CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { // CHECK-NEXT: {RTLIB::MALLOC, RTLIB::impl_malloc}, // malloc // CHECK-NEXT: }; diff --git a/llvm/test/TableGen/RuntimeLibcallEmitter-conflict-warning.td b/llvm/test/TableGen/RuntimeLibcallEmitter-conflict-warning.td index 136c81b..f9a148a 100644 --- a/llvm/test/TableGen/RuntimeLibcallEmitter-conflict-warning.td +++ b/llvm/test/TableGen/RuntimeLibcallEmitter-conflict-warning.td @@ -25,9 +25,22 @@ def dup1 : RuntimeLibcallImpl<ANOTHER_DUP>; // func_a and func_b both provide SOME_FUNC. // CHECK: if (isTargetArchA()) { -// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { -// CHECK-NEXT: {RTLIB::SOME_FUNC, RTLIB::impl_func_b}, // func_b -// CHECK-NEXT: }; +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x00000000000018 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: + +// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { +// CHECK-NEXT: {RTLIB::SOME_FUNC, RTLIB::impl_func_b}, // func_b +// CHECK-NEXT: }; +// CHECK-EMPTY: +// CHECK-NEXT: for (const auto [Func, Impl] : LibraryCalls) { +// CHECK-NEXT: setLibcallImpl(Func, Impl); +// CHECK-NEXT: } +// CHECK-EMPTY: +// CHECK-NEXT: return; +// CHECK-NEXT: } // ERR: :[[@LINE+1]]:5: warning: conflicting implementations for libcall SOME_FUNC: func_b, func_a def TheSystemLibraryA : SystemRuntimeLibrary<isTargetArchA, @@ -35,10 +48,22 @@ def TheSystemLibraryA : SystemRuntimeLibrary<isTargetArchA, >; // CHECK: if (isTargetArchB()) { -// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { -// CHECK-NEXT: {RTLIB::OTHER_FUNC, RTLIB::impl_other_func}, // other_func -// CHECK-NEXT: {RTLIB::SOME_FUNC, RTLIB::impl_func_a}, // func_a -// CHECK-NEXT: }; +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x00000000000058 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: +// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { +// CHECK-NEXT: {RTLIB::OTHER_FUNC, RTLIB::impl_other_func}, // other_func +// CHECK-NEXT: {RTLIB::SOME_FUNC, RTLIB::impl_func_a}, // func_a +// CHECK-NEXT: }; +// CHECK-EMPTY: +// CHECK-NEXT: for (const auto [Func, Impl] : LibraryCalls) { +// CHECK-NEXT: setLibcallImpl(Func, Impl); +// CHECK-NEXT: } +// CHECK-EMPTY: +// CHECK-NEXT: return; +// CHECK-NEXT: } // ERR: :[[@LINE+1]]:5: warning: conflicting implementations for libcall SOME_FUNC: func_a, func_b def TheSystemLibraryB : SystemRuntimeLibrary<isTargetArchB, @@ -46,11 +71,23 @@ def TheSystemLibraryB : SystemRuntimeLibrary<isTargetArchB, >; // CHECK: if (isTargetArchC()) { -// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { -// CHECK-NEXT: {RTLIB::ANOTHER_DUP, RTLIB::impl_dup1}, // dup1 -// CHECK-NEXT: {RTLIB::OTHER_FUNC, RTLIB::impl_other_func}, // other_func -// CHECK-NEXT: {RTLIB::SOME_FUNC, RTLIB::impl_func_a}, // func_a -// CHECK-NEXT: }; +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x0000000000007e +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: +// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { +// CHECK-NEXT: {RTLIB::ANOTHER_DUP, RTLIB::impl_dup1}, // dup1 +// CHECK-NEXT: {RTLIB::OTHER_FUNC, RTLIB::impl_other_func}, // other_func +// CHECK-NEXT: {RTLIB::SOME_FUNC, RTLIB::impl_func_a}, // func_a +// CHECK-NEXT: }; +// CHECK-EMPTY: +// CHECK-NEXT: for (const auto [Func, Impl] : LibraryCalls) { +// CHECK-NEXT: setLibcallImpl(Func, Impl); +// CHECK-NEXT: } +// CHECK-EMPTY: +// CHECK-NEXT: return; +// CHECK-NEXT: } // ERR: :[[@LINE+3]]:5: warning: conflicting implementations for libcall ANOTHER_DUP: dup1, dup0 // ERR: :[[@LINE+2]]:5: warning: conflicting implementations for libcall SOME_FUNC: func_a, func_b diff --git a/llvm/test/TableGen/RuntimeLibcallEmitter.td b/llvm/test/TableGen/RuntimeLibcallEmitter.td index c336fee..0c23e3b 100644 --- a/llvm/test/TableGen/RuntimeLibcallEmitter.td +++ b/llvm/test/TableGen/RuntimeLibcallEmitter.td @@ -196,15 +196,20 @@ def BlahLibrary : SystemRuntimeLibrary<isBlahArch, (add calloc, LibraryWithCondi // CHECK-NEXT: }; // CHECK-EMPTY: // CHECK-NEXT: if (TT.getArch() == Triple::blah) { -// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { -// CHECK-NEXT: {RTLIB::BZERO, RTLIB::impl_bzero}, // bzero -// CHECK-NEXT: {RTLIB::CALLOC, RTLIB::impl_calloc}, // calloc -// CHECK-NEXT: {RTLIB::SQRT_F128, RTLIB::impl_sqrtl_f128}, // sqrtl -// CHECK-NEXT: }; +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x000000000000e0 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; // CHECK-EMPTY: -// CHECK-NEXT: for (const auto [Func, Impl] : LibraryCalls) { -// CHECK-NEXT: setLibcallImpl(Func, Impl); -// CHECK-NEXT: } +// CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { +// CHECK-NEXT: {RTLIB::BZERO, RTLIB::impl_bzero}, // bzero +// CHECK-NEXT: {RTLIB::CALLOC, RTLIB::impl_calloc}, // calloc +// CHECK-NEXT: {RTLIB::SQRT_F128, RTLIB::impl_sqrtl_f128}, // sqrtl +// CHECK-NEXT: }; +// CHECK-EMPTY: +// CHECK-NEXT: for (const auto [Func, Impl] : LibraryCalls) { +// CHECK-NEXT: setLibcallImpl(Func, Impl); +// CHECK-NEXT: } // CHECK-EMPTY: // CHECK-NEXT: if (TT.hasCompilerRT()) { // CHECK-NEXT: static const LibcallImplPair LibraryCalls_hasCompilerRT[] = { @@ -233,6 +238,11 @@ def BlahLibrary : SystemRuntimeLibrary<isBlahArch, (add calloc, LibraryWithCondi // CHECK-NEXT: } // CHECK-EMPTY: // CHECK-NEXT: if (TT.getArch() == Triple::buzz) { +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x00000000000118 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: // CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { // CHECK-NEXT: {RTLIB::SHL_I32, RTLIB::impl___ashlsi3}, // __ashlsi3 // CHECK-NEXT: {RTLIB::SQRT_F80, RTLIB::impl_sqrtl_f80}, // sqrtl @@ -247,6 +257,11 @@ def BlahLibrary : SystemRuntimeLibrary<isBlahArch, (add calloc, LibraryWithCondi // CHECK-NEXT: } // CHECK-EMPTY: // CHECK-NEXT: if (TT.getArch() == Triple::foo) { +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x000000000000a0 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: // CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { // CHECK-NEXT: {RTLIB::BZERO, RTLIB::impl_bzero}, // bzero // CHECK-NEXT: {RTLIB::SQRT_F128, RTLIB::impl_sqrtl_f128}, // sqrtl @@ -271,6 +286,11 @@ def BlahLibrary : SystemRuntimeLibrary<isBlahArch, (add calloc, LibraryWithCondi // CHECK-NEXT: } // CHECK-EMPTY: // CHECK-NEXT: if (TT.getArch() == Triple::simple) { +// CHECK-NEXT: static constexpr LibcallImplBitset SystemAvailableImpls({ +// CHECK-NEXT: 0x00000000000158 +// CHECK-NEXT: }); +// CHECK-NEXT: AvailableLibcallImpls = SystemAvailableImpls; +// CHECK-EMPTY: // CHECK-NEXT: static const LibcallImplPair LibraryCalls[] = { // CHECK-NEXT: {RTLIB::CALLOC, RTLIB::impl_calloc}, // calloc // CHECK-NEXT: {RTLIB::SHL_I32, RTLIB::impl___ashlsi3}, // __ashlsi3 diff --git a/llvm/test/Transforms/CodeGenPrepare/X86/baseoffs-sext-bug.ll b/llvm/test/Transforms/CodeGenPrepare/X86/baseoffs-sext-bug.ll new file mode 100644 index 0000000..51a461e --- /dev/null +++ b/llvm/test/Transforms/CodeGenPrepare/X86/baseoffs-sext-bug.ll @@ -0,0 +1,81 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6 +; RUN: opt -S -passes='require<profile-summary>,function(codegenprepare)' < %s | FileCheck --check-prefix=GEP %s +; RUN: opt -S -passes='require<profile-summary>,function(codegenprepare)' -addr-sink-using-gep=false < %s | FileCheck --check-prefix=NO-GEP %s + +target triple = "x86_64--linux-gnu" +target datalayout = "e-m:e-p0:128:128-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128" +; -p0:128:128 is added to ensure that transformation will be triggered. + +define i128 @test(i128 %arg) { +; GEP-LABEL: define i128 @test( +; GEP-SAME: i128 [[ARG:%.*]]) { +; GEP-NEXT: [[ENTRY:.*]]: +; GEP-NEXT: [[CMP:%.*]] = icmp ugt i128 [[ARG]], 10 +; GEP-NEXT: br i1 [[CMP]], label %[[THEN:.*]], label %[[EXIT:.*]] +; GEP: [[THEN]]: +; GEP-NEXT: [[SUNKADDR:%.*]] = inttoptr i128 [[ARG]] to ptr +; GEP-NEXT: [[SUNKADDR1:%.*]] = getelementptr i8, ptr [[SUNKADDR]], i128 -32 +; GEP-NEXT: [[LOAD:%.*]] = load i128, ptr [[SUNKADDR1]], align 16 +; GEP-NEXT: br label %[[EXIT]] +; GEP: [[EXIT]]: +; GEP-NEXT: [[PHI:%.*]] = phi i128 [ [[LOAD]], %[[THEN]] ], [ 0, %[[ENTRY]] ] +; GEP-NEXT: ret i128 [[PHI]] +; +; NO-GEP-LABEL: define i128 @test( +; NO-GEP-SAME: i128 [[ARG:%.*]]) { +; NO-GEP-NEXT: [[ENTRY:.*]]: +; NO-GEP-NEXT: [[CMP:%.*]] = icmp ugt i128 [[ARG]], 10 +; NO-GEP-NEXT: br i1 [[CMP]], label %[[THEN:.*]], label %[[EXIT:.*]] +; NO-GEP: [[THEN]]: +; NO-GEP-NEXT: [[SUNKADDR:%.*]] = add i128 [[ARG]], -32 +; NO-GEP-NEXT: [[SUNKADDR1:%.*]] = inttoptr i128 [[SUNKADDR]] to ptr +; NO-GEP-NEXT: [[LOAD:%.*]] = load i128, ptr [[SUNKADDR1]], align 16 +; NO-GEP-NEXT: br label %[[EXIT]] +; NO-GEP: [[EXIT]]: +; NO-GEP-NEXT: [[PHI:%.*]] = phi i128 [ [[LOAD]], %[[THEN]] ], [ 0, %[[ENTRY]] ] +; NO-GEP-NEXT: ret i128 [[PHI]] +; +entry: + %add = add i128 %arg, -32 + %cmp = icmp ugt i128 %arg, 10 + br i1 %cmp, label %then, label %exit + +then: + %inttoptr = inttoptr i128 %add to ptr + %load = load i128, ptr %inttoptr, align 16 + br label %exit + +exit: + %phi = phi i128 [ %load, %then ], [ 0, %entry ] + ret i128 %phi +} + +define void @test_combine(ptr %ptr, i128 %arg) { +; GEP-LABEL: define void @test_combine( +; GEP-SAME: ptr [[PTR:%.*]], i128 [[ARG:%.*]]) { +; GEP-NEXT: [[ENTRY:.*:]] +; GEP-NEXT: [[CMP:%.*]] = icmp ugt i128 [[ARG]], 10 +; GEP-NEXT: [[SELECT1:%.*]] = select i1 [[CMP]], i128 -32, i128 0 +; GEP-NEXT: [[SUNKADDR:%.*]] = getelementptr inbounds i8, ptr [[PTR]], i128 [[SELECT1]] +; GEP-NEXT: store i128 1, ptr [[SUNKADDR]], align 16 +; GEP-NEXT: ret void +; +; NO-GEP-LABEL: define void @test_combine( +; NO-GEP-SAME: ptr [[PTR:%.*]], i128 [[ARG:%.*]]) { +; NO-GEP-NEXT: [[ENTRY:.*:]] +; NO-GEP-NEXT: [[CMP:%.*]] = icmp ugt i128 [[ARG]], 10 +; NO-GEP-NEXT: [[SELECT1:%.*]] = select i1 [[CMP]], i128 -32, i128 0 +; NO-GEP-NEXT: [[SUNKADDR:%.*]] = ptrtoint ptr [[PTR]] to i128 +; NO-GEP-NEXT: [[SUNKADDR2:%.*]] = add i128 [[SUNKADDR]], [[SELECT1]] +; NO-GEP-NEXT: [[SUNKADDR3:%.*]] = inttoptr i128 [[SUNKADDR2]] to ptr +; NO-GEP-NEXT: store i128 1, ptr [[SUNKADDR3]], align 16 +; NO-GEP-NEXT: ret void +; +entry: + %cmp = icmp ugt i128 %arg, 10 + %gep = getelementptr inbounds i8, ptr %ptr, i128 -32 + %select = select i1 %cmp, ptr %gep, ptr %ptr + store i128 1, ptr %select, align 16 + ret void +} + diff --git a/llvm/test/Transforms/DFAJumpThreading/dfa-unfold-select.ll b/llvm/test/Transforms/DFAJumpThreading/dfa-unfold-select.ll index 663f459..de38752 100644 --- a/llvm/test/Transforms/DFAJumpThreading/dfa-unfold-select.ll +++ b/llvm/test/Transforms/DFAJumpThreading/dfa-unfold-select.ll @@ -227,10 +227,6 @@ define i32 @test3(i32 %num) { ; CHECK-NEXT: i32 1, label [[CASE1:%.*]] ; CHECK-NEXT: i32 2, label [[CASE2:%.*]] ; CHECK-NEXT: ] -; CHECK: for.body.jt4: -; CHECK-NEXT: [[COUNT_JT4:%.*]] = phi i32 [ [[INC_JT4:%.*]], [[FOR_INC_JT4:%.*]] ] -; CHECK-NEXT: [[STATE_JT4:%.*]] = phi i32 [ [[STATE_NEXT_JT4:%.*]], [[FOR_INC_JT4]] ] -; CHECK-NEXT: br label [[FOR_INC_JT1]] ; CHECK: for.body.jt3: ; CHECK-NEXT: [[COUNT_JT3:%.*]] = phi i32 [ [[INC_JT3:%.*]], [[FOR_INC_JT3:%.*]] ] ; CHECK-NEXT: [[STATE_JT3:%.*]] = phi i32 [ [[STATE_NEXT_JT3:%.*]], [[FOR_INC_JT3]] ] @@ -261,17 +257,14 @@ define i32 @test3(i32 %num) { ; CHECK: sel.2.si.unfold.false: ; CHECK-NEXT: [[DOTSI_UNFOLD_PHI1:%.*]] = phi i32 [ 4, [[SEL_2_SI_UNFOLD_TRUE_JT3]] ] ; CHECK-NEXT: br label [[SEL_3_SI_UNFOLD_FALSE]] -; CHECK: sel.2.si.unfold.false.jt4: +; CHECK: sel.2.si.unfold.false.jt3: ; CHECK-NEXT: [[DOTSI_UNFOLD_PHI1_JT4:%.*]] = phi i32 [ 4, [[SEL_2_SI_UNFOLD_TRUE:%.*]] ] -; CHECK-NEXT: br label [[SEL_3_SI_UNFOLD_FALSE_JT4:%.*]] +; CHECK-NEXT: br label [[SEL_3_SI_UNFOLD_FALSE_JT3]] ; CHECK: sel.3.si.unfold.false: ; CHECK-NEXT: [[SEL_2_SI_UNFOLD_PHI:%.*]] = phi i32 [ poison, [[SEL_2_SI_UNFOLD_TRUE]] ], [ [[DOTSI_UNFOLD_PHI1]], [[SEL_2_SI_UNFOLD_FALSE]] ] ; CHECK-NEXT: br label [[FOR_INC]] -; CHECK: sel.3.si.unfold.false.jt4: -; CHECK-NEXT: [[SEL_2_SI_UNFOLD_PHI_JT4:%.*]] = phi i32 [ [[DOTSI_UNFOLD_PHI1_JT4]], [[SEL_2_SI_UNFOLD_FALSE_JT4]] ] -; CHECK-NEXT: br label [[FOR_INC_JT4]] ; CHECK: sel.3.si.unfold.false.jt3: -; CHECK-NEXT: [[SEL_2_SI_UNFOLD_PHI_JT3:%.*]] = phi i32 [ [[DOTSI_UNFOLD_PHI_JT3]], [[SEL_2_SI_UNFOLD_TRUE_JT3]] ] +; CHECK-NEXT: [[SEL_2_SI_UNFOLD_PHI_JT3:%.*]] = phi i32 [ [[DOTSI_UNFOLD_PHI_JT3]], [[SEL_2_SI_UNFOLD_TRUE_JT3]] ], [ [[DOTSI_UNFOLD_PHI1_JT4]], [[SEL_2_SI_UNFOLD_FALSE_JT4]] ] ; CHECK-NEXT: br label [[FOR_INC_JT3]] ; CHECK: sel.1.si.unfold.true: ; CHECK-NEXT: br i1 [[CMP_1]], label [[FOR_INC]], label [[SEL_1_SI_UNFOLD_FALSE_JT2:%.*]] @@ -289,11 +282,6 @@ define i32 @test3(i32 %num) { ; CHECK-NEXT: [[INC]] = add nsw i32 [[COUNT5]], 1 ; CHECK-NEXT: [[CMP_EXIT:%.*]] = icmp slt i32 [[INC]], [[NUM:%.*]] ; CHECK-NEXT: br i1 [[CMP_EXIT]], label [[FOR_BODY]], label [[FOR_END:%.*]] -; CHECK: for.inc.jt4: -; CHECK-NEXT: [[STATE_NEXT_JT4]] = phi i32 [ [[SEL_2_SI_UNFOLD_PHI_JT4]], [[SEL_3_SI_UNFOLD_FALSE_JT4]] ] -; CHECK-NEXT: [[INC_JT4]] = add nsw i32 undef, 1 -; CHECK-NEXT: [[CMP_EXIT_JT4:%.*]] = icmp slt i32 [[INC_JT4]], [[NUM]] -; CHECK-NEXT: br i1 [[CMP_EXIT_JT4]], label [[FOR_BODY_JT4:%.*]], label [[FOR_END]] ; CHECK: for.inc.jt3: ; CHECK-NEXT: [[STATE_NEXT_JT3]] = phi i32 [ [[SEL_2_SI_UNFOLD_PHI_JT3]], [[SEL_3_SI_UNFOLD_FALSE_JT3]] ] ; CHECK-NEXT: [[INC_JT3]] = add nsw i32 [[COUNT5]], 1 @@ -305,8 +293,8 @@ define i32 @test3(i32 %num) { ; CHECK-NEXT: [[CMP_EXIT_JT2:%.*]] = icmp slt i32 [[INC_JT2]], [[NUM]] ; CHECK-NEXT: br i1 [[CMP_EXIT_JT2]], label [[FOR_BODY_JT2]], label [[FOR_END]] ; CHECK: for.inc.jt1: -; CHECK-NEXT: [[COUNT4:%.*]] = phi i32 [ [[COUNT_JT4]], [[FOR_BODY_JT4]] ], [ [[COUNT_JT3]], [[FOR_BODY_JT3]] ], [ [[COUNT5]], [[SEL_1_SI_UNFOLD_TRUE_JT1]] ], [ [[COUNT]], [[FOR_BODY]] ] -; CHECK-NEXT: [[STATE_NEXT_JT1]] = phi i32 [ 1, [[FOR_BODY]] ], [ 1, [[FOR_BODY_JT3]] ], [ 1, [[FOR_BODY_JT4]] ], [ [[DOTSI_UNFOLD_PHI2_JT1]], [[SEL_1_SI_UNFOLD_TRUE_JT1]] ] +; CHECK-NEXT: [[COUNT4:%.*]] = phi i32 [ [[COUNT_JT3]], [[FOR_BODY_JT3]] ], [ [[COUNT5]], [[SEL_1_SI_UNFOLD_TRUE_JT1]] ], [ [[COUNT]], [[FOR_BODY]] ] +; CHECK-NEXT: [[STATE_NEXT_JT1]] = phi i32 [ 1, [[FOR_BODY]] ], [ 1, [[FOR_BODY_JT3]] ], [ [[DOTSI_UNFOLD_PHI2_JT1]], [[SEL_1_SI_UNFOLD_TRUE_JT1]] ] ; CHECK-NEXT: [[INC_JT1]] = add nsw i32 [[COUNT4]], 1 ; CHECK-NEXT: [[CMP_EXIT_JT1:%.*]] = icmp slt i32 [[INC_JT1]], [[NUM]] ; CHECK-NEXT: br i1 [[CMP_EXIT_JT1]], label [[FOR_BODY_JT1]], label [[FOR_END]] @@ -402,36 +390,28 @@ define void @pr65222(i32 %flags, i1 %cmp, i1 %tobool.not) { ; CHECK-NEXT: br label [[IF_END_JT2:%.*]] ; CHECK: cond1.si.unfold.true: ; CHECK-NEXT: br i1 [[CMP]], label [[IF_END]], label [[COND1_SI_UNFOLD_FALSE_JT1:%.*]] -; CHECK: cond1.si.unfold.true.jt3: +; CHECK: cond1.si.unfold.true.jt2: ; CHECK-NEXT: [[DOTSI_UNFOLD_PHI2:%.*]] = phi i32 [ 3, [[THEN]] ] -; CHECK-NEXT: br i1 [[CMP]], label [[IF_END_JT3:%.*]], label [[COND1_SI_UNFOLD_FALSE:%.*]] +; CHECK-NEXT: br i1 [[CMP]], label [[IF_END_JT2]], label [[COND1_SI_UNFOLD_FALSE:%.*]] ; CHECK: cond1.si.unfold.false: ; CHECK-NEXT: [[DOTSI_UNFOLD_PHI3:%.*]] = phi i32 [ 1, [[COND1_SI_UNFOLD_TRUE]] ] ; CHECK-NEXT: br label [[IF_END]] -; CHECK: cond1.si.unfold.false.jt1: +; CHECK: cond1.si.unfold.false.jt2: ; CHECK-NEXT: [[DOTSI_UNFOLD_PHI3_JT1:%.*]] = phi i32 [ 1, [[COND1_SI_UNFOLD_TRUE1:%.*]] ] -; CHECK-NEXT: br label [[IF_END_JT1:%.*]] +; CHECK-NEXT: br label [[IF_END_JT2]] ; CHECK: if.end: ; CHECK-NEXT: [[UNFOLDED:%.*]] = phi i32 [ [[FLAGS:%.*]], [[WHILE_COND]] ], [ [[COND_SI_UNFOLD_PHI]], [[TOUNFOLD_SI_UNFOLD_FALSE1]] ], [ poison, [[COND1_SI_UNFOLD_TRUE1]] ], [ [[DOTSI_UNFOLD_PHI3]], [[COND1_SI_UNFOLD_FALSE]] ] ; CHECK-NEXT: [[OTHER:%.*]] = phi i32 [ [[FLAGS]], [[WHILE_COND]] ], [ 0, [[TOUNFOLD_SI_UNFOLD_FALSE1]] ], [ 0, [[COND1_SI_UNFOLD_TRUE1]] ], [ 0, [[COND1_SI_UNFOLD_FALSE]] ] ; CHECK-NEXT: switch i32 [[UNFOLDED]], label [[UNREACHABLE:%.*]] [ ; CHECK-NEXT: i32 0, label [[SW_BB:%.*]] ; CHECK-NEXT: ] -; CHECK: if.end.jt1: -; CHECK-NEXT: [[UNFOLDED_JT1:%.*]] = phi i32 [ [[DOTSI_UNFOLD_PHI3_JT1]], [[COND1_SI_UNFOLD_FALSE_JT1]] ] -; CHECK-NEXT: [[OTHER_JT1:%.*]] = phi i32 [ 0, [[COND1_SI_UNFOLD_FALSE_JT1]] ] -; CHECK-NEXT: br label [[UNREACHABLE]] -; CHECK: if.end.jt3: -; CHECK-NEXT: [[UNFOLDED_JT3:%.*]] = phi i32 [ [[DOTSI_UNFOLD_PHI2]], [[COND1_SI_UNFOLD_TRUE]] ] -; CHECK-NEXT: [[OTHER_JT3:%.*]] = phi i32 [ 0, [[COND1_SI_UNFOLD_TRUE]] ] -; CHECK-NEXT: br label [[UNREACHABLE]] ; CHECK: if.end.jt0: ; CHECK-NEXT: [[UNFOLDED_JT0:%.*]] = phi i32 [ [[COND_SI_UNFOLD_PHI_JT0]], [[TOUNFOLD_SI_UNFOLD_FALSE_JT0]] ] ; CHECK-NEXT: [[OTHER_JT0:%.*]] = phi i32 [ 0, [[TOUNFOLD_SI_UNFOLD_FALSE_JT0]] ] ; CHECK-NEXT: br label [[SW_BB]] ; CHECK: if.end.jt2: -; CHECK-NEXT: [[UNFOLDED_JT2:%.*]] = phi i32 [ [[COND_SI_UNFOLD_PHI_JT2]], [[TOUNFOLD_SI_UNFOLD_FALSE]] ] -; CHECK-NEXT: [[OTHER_JT2:%.*]] = phi i32 [ 0, [[TOUNFOLD_SI_UNFOLD_FALSE]] ] +; CHECK-NEXT: [[UNFOLDED_JT2:%.*]] = phi i32 [ [[COND_SI_UNFOLD_PHI_JT2]], [[TOUNFOLD_SI_UNFOLD_FALSE]] ], [ [[DOTSI_UNFOLD_PHI2]], [[COND1_SI_UNFOLD_TRUE]] ], [ [[DOTSI_UNFOLD_PHI3_JT1]], [[COND1_SI_UNFOLD_FALSE_JT1]] ] +; CHECK-NEXT: [[OTHER_JT2:%.*]] = phi i32 [ 0, [[TOUNFOLD_SI_UNFOLD_FALSE]] ], [ 0, [[COND1_SI_UNFOLD_TRUE]] ], [ 0, [[COND1_SI_UNFOLD_FALSE_JT1]] ] ; CHECK-NEXT: br label [[UNREACHABLE]] ; CHECK: unreachable: ; CHECK-NEXT: unreachable diff --git a/llvm/test/Transforms/DFAJumpThreading/equivalent-states.ll b/llvm/test/Transforms/DFAJumpThreading/equivalent-states.ll new file mode 100644 index 0000000..4555dfb --- /dev/null +++ b/llvm/test/Transforms/DFAJumpThreading/equivalent-states.ll @@ -0,0 +1,281 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3 +; RUN: opt -S -passes=dfa-jump-threading %s | FileCheck %s + +declare void @do_something() +declare void @user(i32) + +define void @equivalent_on_default(i1 %c1) { +; CHECK-LABEL: define void @equivalent_on_default( +; CHECK-SAME: i1 [[C1:%.*]]) { +; CHECK-NEXT: entry: +; CHECK-NEXT: br label [[SWITCH_BB:%.*]] +; CHECK: switch_bb: +; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ poison, [[CASE2END:%.*]] ] +; CHECK-NEXT: switch i32 [[PHI]], label [[DEFAULT_DEST:%.*]] [ +; CHECK-NEXT: i32 0, label [[CASE1:%.*]] +; CHECK-NEXT: i32 1, label [[CASE2:%.*]] +; CHECK-NEXT: ] +; CHECK: switch_bb.jt2: +; CHECK-NEXT: [[PHI_JT2:%.*]] = phi i32 [ [[PHI_CASE2_JT2:%.*]], [[CASE2END_JT2:%.*]] ] +; CHECK-NEXT: br label [[DEFAULT_DEST]] +; CHECK: switch_bb.jt1: +; CHECK-NEXT: [[PHI_JT1:%.*]] = phi i32 [ 1, [[CASE1]] ] +; CHECK-NEXT: br label [[CASE2]] +; CHECK: case1: +; CHECK-NEXT: br label [[SWITCH_BB_JT1:%.*]] +; CHECK: case2: +; CHECK-NEXT: br i1 [[C1]], label [[CASE2THEN:%.*]], label [[CASE2END_JT2]] +; CHECK: case2then: +; CHECK-NEXT: br label [[CASE2END_JT2]] +; CHECK: case2end: +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB]] +; CHECK: case2end.jt2: +; CHECK-NEXT: [[PHI_CASE2_JT2]] = phi i32 [ 2, [[CASE2]] ], [ 3, [[CASE2THEN]] ] +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB_JT2:%.*]] +; CHECK: default_dest: +; CHECK-NEXT: ret void +; +entry: + br label %switch_bb + +switch_bb: + %phi = phi i32 [ 0, %entry ], [ 1, %case1 ], [ %phi_case2, %case2end ] + switch i32 %phi, label %default_dest [ + i32 0, label %case1 + i32 1, label %case2 + ] + +case1: + br label %switch_bb + +case2: + br i1 %c1, label %case2then, label %case2end + +case2then: + br label %case2end + +case2end: + %phi_case2 = phi i32 [ 2, %case2 ] , [ 3, %case2then ] + call void @do_something() + br label %switch_bb + +default_dest: + ret void +} + +define void @equivalent_on_default_user(i1 %c1) { +; CHECK-LABEL: define void @equivalent_on_default_user( +; CHECK-SAME: i1 [[C1:%.*]]) { +; CHECK-NEXT: entry: +; CHECK-NEXT: br label [[SWITCH_BB:%.*]] +; CHECK: switch_bb: +; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ poison, [[CASE2END:%.*]] ] +; CHECK-NEXT: switch i32 [[PHI]], label [[DEFAULT_DEST:%.*]] [ +; CHECK-NEXT: i32 0, label [[CASE1:%.*]] +; CHECK-NEXT: i32 1, label [[CASE2:%.*]] +; CHECK-NEXT: ] +; CHECK: switch_bb.jt2: +; CHECK-NEXT: [[PHI_JT2:%.*]] = phi i32 [ [[PHI_CASE2_JT2:%.*]], [[CASE2END_JT2:%.*]] ] +; CHECK-NEXT: br label [[DEFAULT_DEST]] +; CHECK: switch_bb.jt1: +; CHECK-NEXT: [[PHI_JT1:%.*]] = phi i32 [ 1, [[CASE1]] ] +; CHECK-NEXT: br label [[CASE2]] +; CHECK: case1: +; CHECK-NEXT: br label [[SWITCH_BB_JT1:%.*]] +; CHECK: case2: +; CHECK-NEXT: br i1 [[C1]], label [[CASE2THEN:%.*]], label [[CASE2END_JT2]] +; CHECK: case2then: +; CHECK-NEXT: br label [[CASE2END_JT2]] +; CHECK: case2end: +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: call void @user(i32 poison) +; CHECK-NEXT: br label [[SWITCH_BB]] +; CHECK: case2end.jt2: +; CHECK-NEXT: [[PHI_CASE2_JT2]] = phi i32 [ 2, [[CASE2]] ], [ 3, [[CASE2THEN]] ] +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: call void @user(i32 [[PHI_CASE2_JT2]]) +; CHECK-NEXT: br label [[SWITCH_BB_JT2:%.*]] +; CHECK: default_dest: +; CHECK-NEXT: ret void +; +entry: + br label %switch_bb + +switch_bb: + %phi = phi i32 [ 0, %entry ], [ 1, %case1 ], [ %phi_case2, %case2end ] + switch i32 %phi, label %default_dest [ + i32 0, label %case1 + i32 1, label %case2 + ] + +case1: + br label %switch_bb + +case2: + br i1 %c1, label %case2then, label %case2end + +case2then: + br label %case2end + +case2end: + %phi_case2 = phi i32 [ 2, %case2 ] , [ 3, %case2then ] + call void @do_something() + call void @user(i32 %phi_case2) + br label %switch_bb + +default_dest: + ret void +} + +define void @equivalent_only_cases(i1 %c1) { +; CHECK-LABEL: define void @equivalent_only_cases( +; CHECK-SAME: i1 [[C1:%.*]]) { +; CHECK-NEXT: entry: +; CHECK-NEXT: br label [[SWITCH_BB:%.*]] +; CHECK: switch_bb: +; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ poison, [[CASE2END:%.*]] ] +; CHECK-NEXT: switch i32 [[PHI]], label [[DEFAULT_DEST:%.*]] [ +; CHECK-NEXT: i32 0, label [[CASE1:%.*]] +; CHECK-NEXT: i32 1, label [[CASE2:%.*]] +; CHECK-NEXT: i32 2, label [[CASE1]] +; CHECK-NEXT: i32 3, label [[CASE1]] +; CHECK-NEXT: ] +; CHECK: switch_bb.jt2: +; CHECK-NEXT: [[PHI_JT2:%.*]] = phi i32 [ [[PHI_CASE2_JT2:%.*]], [[CASE2END_JT2:%.*]] ] +; CHECK-NEXT: br label [[CASE1]] +; CHECK: switch_bb.jt1: +; CHECK-NEXT: [[PHI_JT1:%.*]] = phi i32 [ 1, [[CASE1]] ] +; CHECK-NEXT: br label [[CASE2]] +; CHECK: case1: +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB_JT1:%.*]] +; CHECK: case2: +; CHECK-NEXT: br i1 [[C1]], label [[CASE2THEN:%.*]], label [[CASE2END_JT2]] +; CHECK: case2then: +; CHECK-NEXT: br label [[CASE2END_JT2]] +; CHECK: case2end: +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB]] +; CHECK: case2end.jt2: +; CHECK-NEXT: [[PHI_CASE2_JT2]] = phi i32 [ 2, [[CASE2]] ], [ 3, [[CASE2THEN]] ] +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB_JT2:%.*]] +; CHECK: default_dest: +; CHECK-NEXT: ret void +; +entry: + br label %switch_bb + +switch_bb: + %phi = phi i32 [ 0, %entry ], [ 1, %case1 ], [ %phi_case2, %case2end ] + switch i32 %phi, label %default_dest [ + i32 0, label %case1 + i32 1, label %case2 + i32 2, label %case1 + i32 3, label %case1 + ] + +case1: + call void @do_something() + br label %switch_bb + +case2: + br i1 %c1, label %case2then, label %case2end + +case2then: + br label %case2end + +case2end: + %phi_case2 = phi i32 [ 2, %case2 ] , [ 3, %case2then ] + call void @do_something() + br label %switch_bb + +default_dest: + ret void +} + +define void @equivalent_both_case_and_default(i1 %c1, i1 %c2) { +; CHECK-LABEL: define void @equivalent_both_case_and_default( +; CHECK-SAME: i1 [[C1:%.*]], i1 [[C2:%.*]]) { +; CHECK-NEXT: entry: +; CHECK-NEXT: br label [[SWITCH_BB:%.*]] +; CHECK: switch_bb: +; CHECK-NEXT: [[PHI:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ poison, [[CASE2END:%.*]] ] +; CHECK-NEXT: switch i32 [[PHI]], label [[DEFAULT_DEST:%.*]] [ +; CHECK-NEXT: i32 0, label [[CASE1:%.*]] +; CHECK-NEXT: i32 1, label [[CASE2:%.*]] +; CHECK-NEXT: i32 2, label [[CASE1]] +; CHECK-NEXT: i32 3, label [[CASE1]] +; CHECK-NEXT: ] +; CHECK: switch_bb.jt4: +; CHECK-NEXT: [[PHI_JT3:%.*]] = phi i32 [ [[PHI_CASE2_JT3:%.*]], [[CASE2END_JT3:%.*]] ] +; CHECK-NEXT: br label [[DEFAULT_DEST]] +; CHECK: switch_bb.jt2: +; CHECK-NEXT: [[PHI_JT2:%.*]] = phi i32 [ [[PHI_CASE2_JT2:%.*]], [[CASE2END_JT2:%.*]] ] +; CHECK-NEXT: br label [[CASE1]] +; CHECK: switch_bb.jt1: +; CHECK-NEXT: [[PHI_JT1:%.*]] = phi i32 [ 1, [[CASE1]] ] +; CHECK-NEXT: br label [[CASE2]] +; CHECK: case1: +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB_JT1:%.*]] +; CHECK: case2: +; CHECK-NEXT: br i1 [[C1]], label [[CASE2THEN:%.*]], label [[CASE2END_JT2]] +; CHECK: case2then: +; CHECK-NEXT: br i1 [[C2]], label [[CASE2THEN2:%.*]], label [[CASE2END_JT2]] +; CHECK: case2then2: +; CHECK-NEXT: br i1 [[C2]], label [[CASE2THEN3:%.*]], label [[CASE2END_JT3]] +; CHECK: case2then3: +; CHECK-NEXT: br label [[CASE2END_JT3]] +; CHECK: case2end: +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB]] +; CHECK: case2end.jt4: +; CHECK-NEXT: [[PHI_CASE2_JT3]] = phi i32 [ 4, [[CASE2THEN2]] ], [ 5, [[CASE2THEN3]] ] +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB_JT3:%.*]] +; CHECK: case2end.jt2: +; CHECK-NEXT: [[PHI_CASE2_JT2]] = phi i32 [ 2, [[CASE2]] ], [ 3, [[CASE2THEN]] ] +; CHECK-NEXT: call void @do_something() +; CHECK-NEXT: br label [[SWITCH_BB_JT2:%.*]] +; CHECK: default_dest: +; CHECK-NEXT: ret void +; +entry: + br label %switch_bb + +switch_bb: + %phi = phi i32 [ 0, %entry ], [ 1, %case1 ], [ %phi_case2, %case2end ] + switch i32 %phi, label %default_dest [ + i32 0, label %case1 + i32 1, label %case2 + i32 2, label %case1 + i32 3, label %case1 + ] + +case1: + call void @do_something() + br label %switch_bb + +case2: + br i1 %c1, label %case2then, label %case2end + +case2then: + br i1 %c2, label %case2then2, label %case2end + +case2then2: + br i1 %c2, label %case2then3, label %case2end + +case2then3: + br label %case2end + +case2end: + %phi_case2 = phi i32 [ 2, %case2 ], [ 3, %case2then ], [ 4, %case2then2 ], [ 5, %case2then3 ] + call void @do_something() + br label %switch_bb + +default_dest: + ret void +} diff --git a/llvm/test/Transforms/GVN/condprop.ll b/llvm/test/Transforms/GVN/condprop.ll index eb2a9f1..57bd2f3 100644 --- a/llvm/test/Transforms/GVN/condprop.ll +++ b/llvm/test/Transforms/GVN/condprop.ll @@ -360,7 +360,7 @@ define i1 @test6_phi2(i1 %c, i32 %x, i32 %y) { ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 [[X]], [[Y]] ; CHECK-NEXT: br i1 [[CMP]], label [[BB2]], label [[BB3:%.*]] ; CHECK: bb2: -; CHECK-NEXT: [[PHI:%.*]] = phi i1 [ [[CMP_NOT]], [[BB1]] ], [ true, [[ENTRY:%.*]] ] +; CHECK-NEXT: [[PHI:%.*]] = phi i1 [ false, [[BB1]] ], [ true, [[ENTRY:%.*]] ] ; CHECK-NEXT: ret i1 [[PHI]] ; CHECK: bb3: ; CHECK-NEXT: ret i1 false diff --git a/llvm/test/Transforms/InstCombine/ptrtoaddr.ll b/llvm/test/Transforms/InstCombine/ptrtoaddr.ll index 5211fbd..7b0b152 100644 --- a/llvm/test/Transforms/InstCombine/ptrtoaddr.ll +++ b/llvm/test/Transforms/InstCombine/ptrtoaddr.ll @@ -10,8 +10,18 @@ target datalayout = "pe1:64:64:64:32" @g.as1 = external addrspace(1) global i8 @g2.as1 = external addrspace(1) global i8 -define i32 @ptrtoaddr_inttoptr_arg(i32 %a) { -; CHECK-LABEL: define i32 @ptrtoaddr_inttoptr_arg( +define i64 @ptrtoaddr_inttoptr_arg(i64 %a) { +; CHECK-LABEL: define i64 @ptrtoaddr_inttoptr_arg( +; CHECK-SAME: i64 [[A:%.*]]) { +; CHECK-NEXT: ret i64 [[A]] +; + %toptr = inttoptr i64 %a to ptr + %toaddr = ptrtoaddr ptr %toptr to i64 + ret i64 %toaddr +} + +define i32 @ptrtoaddr_inttoptr_arg_addrsize(i32 %a) { +; CHECK-LABEL: define i32 @ptrtoaddr_inttoptr_arg_addrsize( ; CHECK-SAME: i32 [[A:%.*]]) { ; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[A]] to i64 ; CHECK-NEXT: [[TOPTR:%.*]] = inttoptr i64 [[TMP1]] to ptr addrspace(1) @@ -130,3 +140,27 @@ define i32 @ptrtoaddr_sub_consts_offset_addrsize() { ; ret i32 sub (i32 ptrtoaddr (ptr addrspace(1) getelementptr (i8, ptr addrspace(1) @g.as1, i32 42) to i32), i32 ptrtoaddr (ptr addrspace(1) @g.as1 to i32)) } + +define i64 @ptrtoaddr_sub_known_offset(ptr %p) { +; CHECK-LABEL: define i64 @ptrtoaddr_sub_known_offset( +; CHECK-SAME: ptr [[P:%.*]]) { +; CHECK-NEXT: ret i64 42 +; + %p2 = getelementptr inbounds i8, ptr %p, i64 42 + %p.addr = ptrtoaddr ptr %p to i64 + %p2.addr = ptrtoaddr ptr %p2 to i64 + %sub = sub i64 %p2.addr, %p.addr + ret i64 %sub +} + +define i32 @ptrtoaddr_sub_known_offset_addrsize(ptr addrspace(1) %p) { +; CHECK-LABEL: define i32 @ptrtoaddr_sub_known_offset_addrsize( +; CHECK-SAME: ptr addrspace(1) [[P:%.*]]) { +; CHECK-NEXT: ret i32 42 +; + %p2 = getelementptr inbounds i8, ptr addrspace(1) %p, i32 42 + %p.addr = ptrtoaddr ptr addrspace(1) %p to i32 + %p2.addr = ptrtoaddr ptr addrspace(1) %p2 to i32 + %sub = sub i32 %p2.addr, %p.addr + ret i32 %sub +} diff --git a/llvm/test/Transforms/InstCombine/select-safe-transforms.ll b/llvm/test/Transforms/InstCombine/select-safe-transforms.ll index ebea5bf..d88eaf8 100644 --- a/llvm/test/Transforms/InstCombine/select-safe-transforms.ll +++ b/llvm/test/Transforms/InstCombine/select-safe-transforms.ll @@ -1,8 +1,11 @@ -; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals ; RUN: opt < %s -passes=instcombine -S | FileCheck %s declare i1 @gen1() +;. +; CHECK: @glb = global i8 0 +;. define i1 @cond_eq_and(i8 %X, i8 %Y, i8 noundef %C) { ; CHECK-LABEL: @cond_eq_and( ; CHECK-NEXT: [[COND:%.*]] = icmp eq i8 [[X:%.*]], [[C:%.*]] @@ -16,16 +19,16 @@ define i1 @cond_eq_and(i8 %X, i8 %Y, i8 noundef %C) { ret i1 %res } -define i1 @cond_eq_and_const(i8 %X, i8 %Y) { +define i1 @cond_eq_and_const(i8 %X, i8 %Y) !prof !0 { ; CHECK-LABEL: @cond_eq_and_const( ; CHECK-NEXT: [[COND:%.*]] = icmp eq i8 [[X:%.*]], 10 ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i8 [[Y:%.*]], 10 -; CHECK-NEXT: [[RES:%.*]] = select i1 [[COND]], i1 [[TMP1]], i1 false +; CHECK-NEXT: [[RES:%.*]] = select i1 [[COND]], i1 [[TMP1]], i1 false, !prof [[PROF1:![0-9]+]] ; CHECK-NEXT: ret i1 [[RES]] ; %cond = icmp eq i8 %X, 10 %lhs = icmp ult i8 %X, %Y - %res = select i1 %cond, i1 %lhs, i1 false + %res = select i1 %cond, i1 %lhs, i1 false, !prof !1 ret i1 %res } @@ -42,16 +45,16 @@ define i1 @cond_eq_or(i8 %X, i8 %Y, i8 noundef %C) { ret i1 %res } -define i1 @cond_eq_or_const(i8 %X, i8 %Y) { +define i1 @cond_eq_or_const(i8 %X, i8 %Y) !prof !0 { ; CHECK-LABEL: @cond_eq_or_const( ; CHECK-NEXT: [[COND:%.*]] = icmp ne i8 [[X:%.*]], 10 ; CHECK-NEXT: [[TMP1:%.*]] = icmp ugt i8 [[Y:%.*]], 10 -; CHECK-NEXT: [[RES:%.*]] = select i1 [[COND]], i1 true, i1 [[TMP1]] +; CHECK-NEXT: [[RES:%.*]] = select i1 [[COND]], i1 true, i1 [[TMP1]], !prof [[PROF1]] ; CHECK-NEXT: ret i1 [[RES]] ; %cond = icmp ne i8 %X, 10 %lhs = icmp ult i8 %X, %Y - %res = select i1 %cond, i1 true, i1 %lhs + %res = select i1 %cond, i1 true, i1 %lhs, !prof !1 ret i1 %res } @@ -793,3 +796,10 @@ define <2 x i1> @not_logical_and2(i1 %b, <2 x i32> %a) { %or = select <2 x i1> %and, <2 x i1> <i1 true, i1 true>, <2 x i1> %implied ret <2 x i1> %or } + +!0 = !{!"function_entry_count", i64 1000} +!1 = !{!"branch_weights", i32 2, i32 3} +;. +; CHECK: [[META0:![0-9]+]] = !{!"function_entry_count", i64 1000} +; CHECK: [[PROF1]] = !{!"branch_weights", i32 2, i32 3} +;. diff --git a/llvm/test/Transforms/InstCombine/sprintf-1.ll b/llvm/test/Transforms/InstCombine/sprintf-1.ll index 1d87758..1798514 100644 --- a/llvm/test/Transforms/InstCombine/sprintf-1.ll +++ b/llvm/test/Transforms/InstCombine/sprintf-1.ll @@ -5,8 +5,7 @@ ; RUN: opt < %s -mtriple xcore-xmos-elf -passes=instcombine -S | FileCheck %s -check-prefixes=CHECK,WITHSTPCPY,CHECK-IPRINTF ; RUN: opt < %s -mtriple=i386-pc-windows-msvc -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,NOSTPCPY ; RUN: opt < %s -mtriple=i386-mingw32 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,NOSTPCPY -; RUN: opt < %s -mtriple=armv7-none-linux-android16 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,NOSTPCPY -; RUN: opt < %s -mtriple=armv7-none-linux-android21 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,WITHSTPCPY +; RUN: opt < %s -mtriple=armv7-none-linux-android -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,WITHSTPCPY ; RUN: opt < %s -mtriple=x86_64-scei-ps4 -passes=instcombine -S | FileCheck %s --check-prefixes=CHECK,NOSTPCPY target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128" diff --git a/llvm/test/Transforms/InstSimplify/ptr_diff.ll b/llvm/test/Transforms/InstSimplify/ptr_diff.ll index fdd9e8e..508dfbc 100644 --- a/llvm/test/Transforms/InstSimplify/ptr_diff.ll +++ b/llvm/test/Transforms/InstSimplify/ptr_diff.ll @@ -14,11 +14,7 @@ define i64 @ptrdiff(ptr %ptr) { define i64 @ptrdiff_no_inbounds(ptr %ptr) { ; CHECK-LABEL: @ptrdiff_no_inbounds( -; CHECK-NEXT: [[LAST:%.*]] = getelementptr i8, ptr [[PTR:%.*]], i32 42 -; CHECK-NEXT: [[FIRST_INT:%.*]] = ptrtoint ptr [[PTR]] to i64 -; CHECK-NEXT: [[LAST_INT:%.*]] = ptrtoint ptr [[LAST]] to i64 -; CHECK-NEXT: [[DIFF:%.*]] = sub i64 [[LAST_INT]], [[FIRST_INT]] -; CHECK-NEXT: ret i64 [[DIFF]] +; CHECK-NEXT: ret i64 42 ; %last = getelementptr i8, ptr %ptr, i32 42 %first.int = ptrtoint ptr %ptr to i64 diff --git a/llvm/test/Transforms/SafeStack/X86/abi_ssp.ll b/llvm/test/Transforms/SafeStack/X86/abi_ssp.ll index e66127533..3b6fb3a 100644 --- a/llvm/test/Transforms/SafeStack/X86/abi_ssp.ll +++ b/llvm/test/Transforms/SafeStack/X86/abi_ssp.ll @@ -1,9 +1,7 @@ ; RUN: opt -safe-stack -S -mtriple=i686-pc-linux-gnu < %s -o - | FileCheck --check-prefixes=COMMON,TLS32 %s ; RUN: opt -safe-stack -S -mtriple=x86_64-pc-linux-gnu < %s -o - | FileCheck --check-prefixes=COMMON,TLS64 %s -; RUN: opt -safe-stack -S -mtriple=i686-linux-android < %s -o - | FileCheck --check-prefixes=COMMON,GLOBAL32 %s -; RUN: opt -safe-stack -S -mtriple=i686-linux-android24 < %s -o - | FileCheck --check-prefixes=COMMON,TLS32 %s - +; RUN: opt -safe-stack -S -mtriple=i686-linux-android < %s -o - | FileCheck --check-prefixes=COMMON,TLS32 %s ; RUN: opt -safe-stack -S -mtriple=x86_64-linux-android < %s -o - | FileCheck --check-prefixes=COMMON,TLS64 %s ; RUN: opt -safe-stack -S -mtriple=x86_64-unknown-fuchsia < %s -o - | FileCheck --check-prefixes=COMMON,FUCHSIA64 %s @@ -11,9 +9,7 @@ ; RUN: opt -passes=safe-stack -S -mtriple=i686-pc-linux-gnu < %s -o - | FileCheck --check-prefixes=COMMON,TLS32 %s ; RUN: opt -passes=safe-stack -S -mtriple=x86_64-pc-linux-gnu < %s -o - | FileCheck --check-prefixes=COMMON,TLS64 %s -; RUN: opt -passes=safe-stack -S -mtriple=i686-linux-android < %s -o - | FileCheck --check-prefixes=COMMON,GLOBAL32 %s -; RUN: opt -passes=safe-stack -S -mtriple=i686-linux-android24 < %s -o - | FileCheck --check-prefixes=COMMON,TLS32 %s - +; RUN: opt -passes=safe-stack -S -mtriple=i686-linux-android < %s -o - | FileCheck --check-prefixes=COMMON,TLS32 %s ; RUN: opt -passes=safe-stack -S -mtriple=x86_64-linux-android < %s -o - | FileCheck --check-prefixes=COMMON,TLS64 %s ; RUN: opt -passes=safe-stack -S -mtriple=x86_64-unknown-fuchsia < %s -o - | FileCheck --check-prefixes=COMMON,FUCHSIA64 %s diff --git a/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll b/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll index 0fc3c19..a43e762 100644 --- a/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll +++ b/llvm/test/Transforms/SimplifyCFG/switch-range-to-icmp.ll @@ -401,4 +401,27 @@ b: ret i32 %1 } +define i32 @else_will_be_unreachable(i1 %arg) { +; CHECK-LABEL: @else_will_be_unreachable( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[I:%.*]] = select i1 [[ARG:%.*]], i32 0, i32 1 +; CHECK-NEXT: ret i32 [[I]] +; +entry: + switch i1 %arg, label %else [ + i1 false, label %if + i1 true, label %if + ] + +if: + br i1 %arg, label %else, label %bb + +bb: + br label %else + +else: + %i = phi i32 [ 0, %entry ], [ 0, %if ], [ 1, %bb ] + ret i32 %i +} + declare void @bar(ptr nonnull dereferenceable(4)) diff --git a/llvm/test/tools/llvm-dwarfdump/X86/DW_AT_language_name.s b/llvm/test/tools/llvm-dwarfdump/X86/DW_AT_language_name.s new file mode 100644 index 0000000..af58175 --- /dev/null +++ b/llvm/test/tools/llvm-dwarfdump/X86/DW_AT_language_name.s @@ -0,0 +1,35 @@ +# Demonstrate dumping DW_AT_language_name. +# RUN: llvm-mc -triple=x86_64--linux -filetype=obj < %s | \ +# RUN: llvm-dwarfdump -v - | FileCheck %s + +# CHECK: .debug_abbrev contents: +# CHECK: DW_AT_language_name DW_FORM_data2 +# CHECK: DW_AT_language_name DW_FORM_data2 +# CHECK: .debug_info contents: +# CHECK: DW_AT_language_name [DW_FORM_data2] (DW_LNAME_C) +# CHECK: DW_AT_language_name [DW_FORM_data2] (0x0000) + + .section .debug_abbrev,"",@progbits + .byte 1 # Abbreviation Code + .byte 17 # DW_TAG_compile_unit + .byte 1 # DW_CHILDREN_no + .ascii "\220\001" # DW_AT_language_name + .byte 5 # DW_FORM_data2 + .ascii "\220\001" # DW_AT_language_name + .byte 5 # DW_FORM_data2 + .byte 0 # EOM(1) + .byte 0 # EOM(2) + .byte 0 # EOM(3) + + .section .debug_info,"",@progbits + .long .Ldebug_info_end0-.Ldebug_info_start0 # Length of Unit +.Ldebug_info_start0: + .short 5 # DWARF version number + .byte 1 # Unit type + .byte 8 # Address Size (in bytes) + .long .debug_abbrev # Offset Into Abbrev. Section + .byte 1 # Abbrev [1] DW_TAG_compile_unit + .short 3 # DW_AT_language_name + .short 0 # DW_AT_language_name + .byte 0 +.Ldebug_info_end0: diff --git a/llvm/test/tools/llvm-dwarfdump/X86/DW_AT_language_version.s b/llvm/test/tools/llvm-dwarfdump/X86/DW_AT_language_version.s new file mode 100644 index 0000000..f1be8fd --- /dev/null +++ b/llvm/test/tools/llvm-dwarfdump/X86/DW_AT_language_version.s @@ -0,0 +1,35 @@ +# Demonstrate dumping DW_AT_language_version. +# RUN: llvm-mc -triple=x86_64--linux -filetype=obj < %s | \ +# RUN: llvm-dwarfdump -v - | FileCheck %s + +# CHECK: .debug_abbrev contents: +# CHECK: DW_AT_language_version DW_FORM_data4 +# CHECK: DW_AT_language_version DW_FORM_data2 +# CHECK: .debug_info contents: +# CHECK: DW_AT_language_version [DW_FORM_data4] (201402) +# CHECK: DW_AT_language_version [DW_FORM_data2] (0) + + .section .debug_abbrev,"",@progbits + .byte 1 # Abbreviation Code + .byte 17 # DW_TAG_compile_unit + .byte 1 # DW_CHILDREN_no + .ascii "\221\001" # DW_AT_language_version + .byte 6 # DW_FORM_data4 + .ascii "\221\001" # DW_AT_language_version + .byte 5 # DW_FORM_data2 + .byte 0 # EOM(1) + .byte 0 # EOM(2) + .byte 0 # EOM(3) + + .section .debug_info,"",@progbits + .long .Ldebug_info_end0-.Ldebug_info_start0 # Length of Unit +.Ldebug_info_start0: + .short 5 # DWARF version number + .byte 1 # Unit type + .byte 8 # Address Size (in bytes) + .long .debug_abbrev # Offset Into Abbrev. Section + .byte 1 # Abbrev [1] DW_TAG_compile_unit + .long 201402 # DW_AT_language_version + .short 0 # DW_AT_language_version + .byte 0 +.Ldebug_info_end0: diff --git a/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vector-fp.s b/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vector-fp.s index e1e9b57..64e3ed9 100644 --- a/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vector-fp.s +++ b/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vector-fp.s @@ -2323,13 +2323,13 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_F_F_W vfncvt.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_ROD_F_F_W vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 3 1.00 U 1 VLEN1024X300SiFive7PipeA,VLEN1024X300SiFive7PipeAB VSETVLI vsetvli zero, zero, e64, m1, tu, mu -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 # CHECK-NEXT: 1 228 228.00 228 VLEN1024X300SiFive7VA1[1,229],VLEN1024X300SiFive7VA1OrVA2[1,229],VLEN1024X300SiFive7VCQ VFDIV_VV vfdiv.vv v8, v16, v24 # CHECK-NEXT: 1 228 228.00 228 VLEN1024X300SiFive7VA1[1,229],VLEN1024X300SiFive7VA1OrVA2[1,229],VLEN1024X300SiFive7VCQ VFDIV_VF vfdiv.vf v8, v16, fs0 # CHECK-NEXT: 1 228 228.00 228 VLEN1024X300SiFive7VA1[1,229],VLEN1024X300SiFive7VA1OrVA2[1,229],VLEN1024X300SiFive7VCQ VFRDIV_VF vfrdiv.vf v8, v16, fs0 @@ -2352,22 +2352,22 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 228 228.00 228 VLEN1024X300SiFive7VA1[1,229],VLEN1024X300SiFive7VA1OrVA2[1,229],VLEN1024X300SiFive7VCQ VFSQRT_V vfsqrt.v v8, v24 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFRSQRT7_V vfrsqrt7.v v8, v24 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFREC7_V vfrec7.v v8, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 +# CHECK-NEXT: 1 19 16.00 19 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFCVT_XU_F_V vfcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFCVT_X_F_V vfcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFCVT_RTZ_XU_F_V vfcvt.rtz.xu.f.v v8, v16 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFCVT_RTZ_X_F_V vfcvt.rtz.x.f.v v8, v16 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 -# CHECK-NEXT: 1 16 16.00 16 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 +# CHECK-NEXT: 1 23 16.00 23 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFWCVT_XU_F_V vfwcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFWCVT_X_F_V vfwcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 2.00 8 VLEN1024X300SiFive7VA1[1,3],VLEN1024X300SiFive7VA1OrVA2[1,3],VLEN1024X300SiFive7VCQ VFWCVT_RTZ_XU_F_V vfwcvt.rtz.xu.f.v v8, v16 @@ -2384,13 +2384,13 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_F_F_W vfncvt.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_ROD_F_F_W vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 3 1.00 U 1 VLEN1024X300SiFive7PipeA,VLEN1024X300SiFive7PipeAB VSETVLI vsetvli zero, zero, e64, m2, tu, mu -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 # CHECK-NEXT: 1 456 456.00 456 VLEN1024X300SiFive7VA1[1,457],VLEN1024X300SiFive7VA1OrVA2[1,457],VLEN1024X300SiFive7VCQ VFDIV_VV vfdiv.vv v8, v16, v24 # CHECK-NEXT: 1 456 456.00 456 VLEN1024X300SiFive7VA1[1,457],VLEN1024X300SiFive7VA1OrVA2[1,457],VLEN1024X300SiFive7VCQ VFDIV_VF vfdiv.vf v8, v16, fs0 # CHECK-NEXT: 1 456 456.00 456 VLEN1024X300SiFive7VA1[1,457],VLEN1024X300SiFive7VA1OrVA2[1,457],VLEN1024X300SiFive7VCQ VFRDIV_VF vfrdiv.vf v8, v16, fs0 @@ -2413,22 +2413,22 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 456 456.00 456 VLEN1024X300SiFive7VA1[1,457],VLEN1024X300SiFive7VA1OrVA2[1,457],VLEN1024X300SiFive7VCQ VFSQRT_V vfsqrt.v v8, v24 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFRSQRT7_V vfrsqrt7.v v8, v24 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFREC7_V vfrec7.v v8, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 +# CHECK-NEXT: 1 35 32.00 35 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFCVT_XU_F_V vfcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFCVT_X_F_V vfcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFCVT_RTZ_XU_F_V vfcvt.rtz.xu.f.v v8, v16 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFCVT_RTZ_X_F_V vfcvt.rtz.x.f.v v8, v16 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 -# CHECK-NEXT: 1 32 32.00 32 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 +# CHECK-NEXT: 1 39 32.00 39 VLEN1024X300SiFive7VA1[1,33],VLEN1024X300SiFive7VA1OrVA2[1,33],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFWCVT_XU_F_V vfwcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFWCVT_X_F_V vfwcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 4.00 8 VLEN1024X300SiFive7VA1[1,5],VLEN1024X300SiFive7VA1OrVA2[1,5],VLEN1024X300SiFive7VCQ VFWCVT_RTZ_XU_F_V vfwcvt.rtz.xu.f.v v8, v16 @@ -2445,13 +2445,13 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_F_F_W vfncvt.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_ROD_F_F_W vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 3 1.00 U 1 VLEN1024X300SiFive7PipeA,VLEN1024X300SiFive7PipeAB VSETVLI vsetvli zero, zero, e64, m4, tu, mu -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 # CHECK-NEXT: 1 912 912.00 912 VLEN1024X300SiFive7VA1[1,913],VLEN1024X300SiFive7VA1OrVA2[1,913],VLEN1024X300SiFive7VCQ VFDIV_VV vfdiv.vv v8, v16, v24 # CHECK-NEXT: 1 912 912.00 912 VLEN1024X300SiFive7VA1[1,913],VLEN1024X300SiFive7VA1OrVA2[1,913],VLEN1024X300SiFive7VCQ VFDIV_VF vfdiv.vf v8, v16, fs0 # CHECK-NEXT: 1 912 912.00 912 VLEN1024X300SiFive7VA1[1,913],VLEN1024X300SiFive7VA1OrVA2[1,913],VLEN1024X300SiFive7VCQ VFRDIV_VF vfrdiv.vf v8, v16, fs0 @@ -2474,22 +2474,22 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 912 912.00 912 VLEN1024X300SiFive7VA1[1,913],VLEN1024X300SiFive7VA1OrVA2[1,913],VLEN1024X300SiFive7VCQ VFSQRT_V vfsqrt.v v8, v24 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFRSQRT7_V vfrsqrt7.v v8, v24 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFREC7_V vfrec7.v v8, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 +# CHECK-NEXT: 1 67 64.00 67 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFCVT_XU_F_V vfcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFCVT_X_F_V vfcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFCVT_RTZ_XU_F_V vfcvt.rtz.xu.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFCVT_RTZ_X_F_V vfcvt.rtz.x.f.v v8, v16 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 -# CHECK-NEXT: 1 64 64.00 64 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 +# CHECK-NEXT: 1 71 64.00 71 VLEN1024X300SiFive7VA1[1,65],VLEN1024X300SiFive7VA1OrVA2[1,65],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFWCVT_XU_F_V vfwcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFWCVT_X_F_V vfwcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFWCVT_RTZ_XU_F_V vfwcvt.rtz.xu.f.v v8, v16 @@ -2506,13 +2506,13 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_F_F_W vfncvt.f.f.w v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFNCVT_ROD_F_F_W vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 3 1.00 U 1 VLEN1024X300SiFive7PipeA,VLEN1024X300SiFive7PipeAB VSETVLI vsetvli zero, zero, e64, m8, tu, mu -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFADD_VV vfadd.vv v8, v16, v24 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFADD_VF vfadd.vf v8, v16, fs0 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSUB_VV vfsub.vv v8, v16, v24 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSUB_VF vfsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFRSUB_VF vfrsub.vf v8, v16, fs0 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMUL_VV vfmul.vv v8, v16, v24 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMUL_VF vfmul.vf v8, v16, fs0 # CHECK-NEXT: 1 1824 1824.00 1824 VLEN1024X300SiFive7VA1[1,1825],VLEN1024X300SiFive7VA1OrVA2[1,1825],VLEN1024X300SiFive7VCQ VFDIV_VV vfdiv.vv v8, v16, v24 # CHECK-NEXT: 1 1824 1824.00 1824 VLEN1024X300SiFive7VA1[1,1825],VLEN1024X300SiFive7VA1OrVA2[1,1825],VLEN1024X300SiFive7VCQ VFDIV_VF vfdiv.vf v8, v16, fs0 # CHECK-NEXT: 1 1824 1824.00 1824 VLEN1024X300SiFive7VA1[1,1825],VLEN1024X300SiFive7VA1OrVA2[1,1825],VLEN1024X300SiFive7VCQ VFRDIV_VF vfrdiv.vf v8, v16, fs0 @@ -2535,22 +2535,22 @@ vfncvt.rod.f.f.w v8, v16 # CHECK-NEXT: 1 1824 1824.00 1824 VLEN1024X300SiFive7VA1[1,1825],VLEN1024X300SiFive7VA1OrVA2[1,1825],VLEN1024X300SiFive7VCQ VFSQRT_V vfsqrt.v v8, v24 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFRSQRT7_V vfrsqrt7.v v8, v24 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFREC7_V vfrec7.v v8, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMIN_VV vfmin.vv v8, v16, v24 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMIN_VF vfmin.vf v8, v16, fs0 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMAX_VV vfmax.vv v8, v16, v24 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFMAX_VF vfmax.vf v8, v16, fs0 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJ_VV vfsgnj.vv v8, v16, v24 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJ_VF vfsgnj.vf v8, v16, fs0 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJN_VV vfsgnjn.vv v8, v16, v24 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJN_VF vfsgnjn.vf v8, v16, fs0 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJX_VV vfsgnjx.vv v8, v16, v24 +# CHECK-NEXT: 1 131 128.00 131 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFSGNJX_VF vfsgnjx.vf v8, v16, fs0 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_XU_F_V vfcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_X_F_V vfcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_RTZ_XU_F_V vfcvt.rtz.xu.f.v v8, v16 # CHECK-NEXT: 1 8 16.00 8 VLEN1024X300SiFive7VA1[1,17],VLEN1024X300SiFive7VA1OrVA2[1,17],VLEN1024X300SiFive7VCQ VFCVT_RTZ_X_F_V vfcvt.rtz.x.f.v v8, v16 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 -# CHECK-NEXT: 1 128 128.00 128 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFCVT_F_XU_V vfcvt.f.xu.v v8, v16 +# CHECK-NEXT: 1 135 128.00 135 VLEN1024X300SiFive7VA1[1,129],VLEN1024X300SiFive7VA1OrVA2[1,129],VLEN1024X300SiFive7VCQ VFCVT_F_X_V vfcvt.f.x.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFWCVT_XU_F_V vfwcvt.xu.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFWCVT_X_F_V vfwcvt.x.f.v v8, v16 # CHECK-NEXT: 1 8 8.00 8 VLEN1024X300SiFive7VA1[1,9],VLEN1024X300SiFive7VA1OrVA2[1,9],VLEN1024X300SiFive7VCQ VFWCVT_RTZ_XU_F_V vfwcvt.rtz.xu.f.v v8, v16 diff --git a/llvm/tools/llvm-ir2vec/llvm-ir2vec.cpp b/llvm/tools/llvm-ir2vec/llvm-ir2vec.cpp index 434449c..1031932 100644 --- a/llvm/tools/llvm-ir2vec/llvm-ir2vec.cpp +++ b/llvm/tools/llvm-ir2vec/llvm-ir2vec.cpp @@ -253,25 +253,17 @@ public: break; } case BasicBlockLevel: { - const auto &BBVecMap = Emb->getBBVecMap(); for (const BasicBlock &BB : F) { - auto It = BBVecMap.find(&BB); - if (It != BBVecMap.end()) { - OS << BB.getName() << ":"; - It->second.print(OS); - } + OS << BB.getName() << ":"; + Emb->getBBVector(BB).print(OS); } break; } case InstructionLevel: { - const auto &InstMap = Emb->getInstVecMap(); for (const BasicBlock &BB : F) { for (const Instruction &I : BB) { - auto It = InstMap.find(&I); - if (It != InstMap.end()) { - I.print(OS); - It->second.print(OS); - } + I.print(OS); + Emb->getInstVector(I).print(OS); } } break; diff --git a/llvm/unittests/ADT/BitsetTest.cpp b/llvm/unittests/ADT/BitsetTest.cpp new file mode 100644 index 0000000..0ecd213 --- /dev/null +++ b/llvm/unittests/ADT/BitsetTest.cpp @@ -0,0 +1,71 @@ +//===- llvm/unittest/Support/BitsetTest.cpp -------------------------------===// +// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception +// +//===----------------------------------------------------------------------===// + +#include "llvm/ADT/Bitset.h" +#include "gtest/gtest.h" + +using namespace llvm; + +namespace { + +template <unsigned NumBits> +class TestBitsetUInt64Array : public Bitset<NumBits> { + static constexpr unsigned NumElts = (NumBits + 63) / 64; + +public: + TestBitsetUInt64Array(const std::array<uint64_t, NumElts> &B) + : Bitset<NumBits>(B) {} + + bool verifyValue(const std::array<uint64_t, NumElts> &B) const { + for (unsigned I = 0; I != NumBits; ++I) { + bool ReferenceVal = + (B[(I / 64)] & (static_cast<uint64_t>(1) << (I % 64))) != 0; + if (ReferenceVal != this->test(I)) + return false; + } + + return true; + } + + void verifyStorageSize(size_t elements_64_bit, size_t elements_32_bit) { + if constexpr (sizeof(uintptr_t) == sizeof(uint64_t)) + EXPECT_EQ(sizeof(*this), elements_64_bit * sizeof(uintptr_t)); + else + EXPECT_EQ(sizeof(*this), elements_32_bit * sizeof(uintptr_t)); + } +}; + +TEST(BitsetTest, Construction) { + std::array<uint64_t, 2> TestVals = {0x123456789abcdef3, 0x1337d3a0b22c24}; + TestBitsetUInt64Array<96> Test(TestVals); + EXPECT_TRUE(Test.verifyValue(TestVals)); + Test.verifyStorageSize(2, 3); + + TestBitsetUInt64Array<65> Test1(TestVals); + EXPECT_TRUE(Test1.verifyValue(TestVals)); + Test1.verifyStorageSize(2, 3); + + std::array<uint64_t, 1> TestSingleVal = {0x12345678abcdef99}; + + TestBitsetUInt64Array<64> Test64(TestSingleVal); + EXPECT_TRUE(Test64.verifyValue(TestSingleVal)); + Test64.verifyStorageSize(1, 2); + + TestBitsetUInt64Array<30> Test30(TestSingleVal); + EXPECT_TRUE(Test30.verifyValue(TestSingleVal)); + Test30.verifyStorageSize(1, 1); + + TestBitsetUInt64Array<32> Test32(TestSingleVal); + EXPECT_TRUE(Test32.verifyValue(TestSingleVal)); + Test32.verifyStorageSize(1, 1); + + TestBitsetUInt64Array<33> Test33(TestSingleVal); + EXPECT_TRUE(Test33.verifyValue(TestSingleVal)); + Test33.verifyStorageSize(1, 2); +} +} // namespace diff --git a/llvm/unittests/ADT/CMakeLists.txt b/llvm/unittests/ADT/CMakeLists.txt index dafd735..848ccba 100644 --- a/llvm/unittests/ADT/CMakeLists.txt +++ b/llvm/unittests/ADT/CMakeLists.txt @@ -12,6 +12,7 @@ add_llvm_unittest(ADTTests BitFieldsTest.cpp BitmaskEnumTest.cpp BitTest.cpp + BitsetTest.cpp BitVectorTest.cpp BreadthFirstIteratorTest.cpp BumpPtrListTest.cpp diff --git a/llvm/unittests/Analysis/IR2VecTest.cpp b/llvm/unittests/Analysis/IR2VecTest.cpp index 40b4aa2..8ffc5f6 100644 --- a/llvm/unittests/Analysis/IR2VecTest.cpp +++ b/llvm/unittests/Analysis/IR2VecTest.cpp @@ -30,7 +30,9 @@ namespace { class TestableEmbedder : public Embedder { public: TestableEmbedder(const Function &F, const Vocabulary &V) : Embedder(F, V) {} - void computeEmbeddings(const BasicBlock &BB) const override {} + Embedding computeEmbeddings(const Instruction &I) const override { + return Embedding(); + } }; TEST(EmbeddingTest, ConstructorsAndAccessors) { @@ -321,18 +323,12 @@ protected: } }; -TEST_F(IR2VecTestFixture, GetInstVecMap_Symbolic) { +TEST_F(IR2VecTestFixture, GetInstVec_Symbolic) { auto Emb = Embedder::create(IR2VecKind::Symbolic, *F, *V); ASSERT_TRUE(static_cast<bool>(Emb)); - const auto &InstMap = Emb->getInstVecMap(); - - EXPECT_EQ(InstMap.size(), 2u); - EXPECT_TRUE(InstMap.count(AddInst)); - EXPECT_TRUE(InstMap.count(RetInst)); - - const auto &AddEmb = InstMap.at(AddInst); - const auto &RetEmb = InstMap.at(RetInst); + const auto &AddEmb = Emb->getInstVector(*AddInst); + const auto &RetEmb = Emb->getInstVector(*RetInst); EXPECT_EQ(AddEmb.size(), 2u); EXPECT_EQ(RetEmb.size(), 2u); @@ -340,51 +336,17 @@ TEST_F(IR2VecTestFixture, GetInstVecMap_Symbolic) { EXPECT_TRUE(RetEmb.approximatelyEquals(Embedding(2, 15.5))); } -TEST_F(IR2VecTestFixture, GetInstVecMap_FlowAware) { - auto Emb = Embedder::create(IR2VecKind::FlowAware, *F, *V); - ASSERT_TRUE(static_cast<bool>(Emb)); - - const auto &InstMap = Emb->getInstVecMap(); - - EXPECT_EQ(InstMap.size(), 2u); - EXPECT_TRUE(InstMap.count(AddInst)); - EXPECT_TRUE(InstMap.count(RetInst)); - - EXPECT_EQ(InstMap.at(AddInst).size(), 2u); - EXPECT_EQ(InstMap.at(RetInst).size(), 2u); - - EXPECT_TRUE(InstMap.at(AddInst).approximatelyEquals(Embedding(2, 25.5))); - EXPECT_TRUE(InstMap.at(RetInst).approximatelyEquals(Embedding(2, 32.6))); -} - -TEST_F(IR2VecTestFixture, GetBBVecMap_Symbolic) { - auto Emb = Embedder::create(IR2VecKind::Symbolic, *F, *V); - ASSERT_TRUE(static_cast<bool>(Emb)); - - const auto &BBMap = Emb->getBBVecMap(); - - EXPECT_EQ(BBMap.size(), 1u); - EXPECT_TRUE(BBMap.count(BB)); - EXPECT_EQ(BBMap.at(BB).size(), 2u); - - // BB vector should be sum of add and ret: {25.5, 25.5} + {15.5, 15.5} = - // {41.0, 41.0} - EXPECT_TRUE(BBMap.at(BB).approximatelyEquals(Embedding(2, 41.0))); -} - -TEST_F(IR2VecTestFixture, GetBBVecMap_FlowAware) { +TEST_F(IR2VecTestFixture, GetInstVec_FlowAware) { auto Emb = Embedder::create(IR2VecKind::FlowAware, *F, *V); ASSERT_TRUE(static_cast<bool>(Emb)); - const auto &BBMap = Emb->getBBVecMap(); - - EXPECT_EQ(BBMap.size(), 1u); - EXPECT_TRUE(BBMap.count(BB)); - EXPECT_EQ(BBMap.at(BB).size(), 2u); + const auto &AddEmb = Emb->getInstVector(*AddInst); + const auto &RetEmb = Emb->getInstVector(*RetInst); + EXPECT_EQ(AddEmb.size(), 2u); + EXPECT_EQ(RetEmb.size(), 2u); - // BB vector should be sum of add and ret: {25.5, 25.5} + {32.6, 32.6} = - // {58.1, 58.1} - EXPECT_TRUE(BBMap.at(BB).approximatelyEquals(Embedding(2, 58.1))); + EXPECT_TRUE(AddEmb.approximatelyEquals(Embedding(2, 25.5))); + EXPECT_TRUE(RetEmb.approximatelyEquals(Embedding(2, 32.6))); } TEST_F(IR2VecTestFixture, GetBBVector_Symbolic) { @@ -394,6 +356,8 @@ TEST_F(IR2VecTestFixture, GetBBVector_Symbolic) { const auto &BBVec = Emb->getBBVector(*BB); EXPECT_EQ(BBVec.size(), 2u); + // BB vector should be sum of add and ret: {25.5, 25.5} + {15.5, 15.5} = + // {41.0, 41.0} EXPECT_TRUE(BBVec.approximatelyEquals(Embedding(2, 41.0))); } @@ -404,6 +368,8 @@ TEST_F(IR2VecTestFixture, GetBBVector_FlowAware) { const auto &BBVec = Emb->getBBVector(*BB); EXPECT_EQ(BBVec.size(), 2u); + // BB vector should be sum of add and ret: {25.5, 25.5} + {32.6, 32.6} = + // {58.1, 58.1} EXPECT_TRUE(BBVec.approximatelyEquals(Embedding(2, 58.1))); } @@ -446,15 +412,9 @@ TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_Symbolic) { EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec3)); EXPECT_TRUE(FuncVec2.approximatelyEquals(FuncVec3)); - // Also check that instruction vectors remain consistent - const auto &InstMap1 = Emb->getInstVecMap(); - const auto &InstMap2 = Emb->getInstVecMap(); - - EXPECT_EQ(InstMap1.size(), InstMap2.size()); - for (const auto &[Inst, Vec1] : InstMap1) { - ASSERT_TRUE(InstMap2.count(Inst)); - EXPECT_TRUE(Vec1.approximatelyEquals(InstMap2.at(Inst))); - } + Emb->invalidateEmbeddings(); + const auto &FuncVec4 = Emb->getFunctionVector(); + EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec4)); } TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_FlowAware) { @@ -473,15 +433,9 @@ TEST_F(IR2VecTestFixture, MultipleComputeEmbeddingsConsistency_FlowAware) { EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec3)); EXPECT_TRUE(FuncVec2.approximatelyEquals(FuncVec3)); - // Also check that instruction vectors remain consistent - const auto &InstMap1 = Emb->getInstVecMap(); - const auto &InstMap2 = Emb->getInstVecMap(); - - EXPECT_EQ(InstMap1.size(), InstMap2.size()); - for (const auto &[Inst, Vec1] : InstMap1) { - ASSERT_TRUE(InstMap2.count(Inst)); - EXPECT_TRUE(Vec1.approximatelyEquals(InstMap2.at(Inst))); - } + Emb->invalidateEmbeddings(); + const auto &FuncVec4 = Emb->getFunctionVector(); + EXPECT_TRUE(FuncVec1.approximatelyEquals(FuncVec4)); } static constexpr unsigned MaxOpcodes = Vocabulary::MaxOpcodes; diff --git a/llvm/unittests/IR/ConstantFPRangeTest.cpp b/llvm/unittests/IR/ConstantFPRangeTest.cpp index 255f62d..1436f0f 100644 --- a/llvm/unittests/IR/ConstantFPRangeTest.cpp +++ b/llvm/unittests/IR/ConstantFPRangeTest.cpp @@ -767,4 +767,39 @@ TEST_F(ConstantFPRangeTest, makeExactFCmpRegion) { } } +TEST_F(ConstantFPRangeTest, abs) { + EXPECT_EQ(Full.abs(), + ConstantFPRange(APFloat::getZero(Sem, /*Negative=*/false), + APFloat::getInf(Sem, /*Negative=*/false), + /*MayBeQNaN=*/true, + /*MayBeSNaN=*/true)); + EXPECT_EQ(Empty.abs(), Empty); + EXPECT_EQ(Zero.abs(), PosZero); + EXPECT_EQ(PosInf.abs(), PosInf); + EXPECT_EQ(NegInf.abs(), PosInf); + EXPECT_EQ(Some.abs(), SomePos); + EXPECT_EQ(SomeNeg.abs(), SomePos); + EXPECT_EQ(NaN.abs(), NaN); + EXPECT_EQ(ConstantFPRange::getNonNaN(APFloat(-2.0), APFloat(3.0)).abs(), + ConstantFPRange::getNonNaN(APFloat(0.0), APFloat(3.0))); + EXPECT_EQ(ConstantFPRange::getNonNaN(APFloat(-3.0), APFloat(2.0)).abs(), + ConstantFPRange::getNonNaN(APFloat(0.0), APFloat(3.0))); +} + +TEST_F(ConstantFPRangeTest, negate) { + EXPECT_EQ(Full.negate(), Full); + EXPECT_EQ(Empty.negate(), Empty); + EXPECT_EQ(Zero.negate(), Zero); + EXPECT_EQ(PosInf.negate(), NegInf); + EXPECT_EQ(NegInf.negate(), PosInf); + EXPECT_EQ(Some.negate(), Some); + EXPECT_EQ(SomePos.negate(), SomeNeg); + EXPECT_EQ(SomeNeg.negate(), SomePos); + EXPECT_EQ(NaN.negate(), NaN); + EXPECT_EQ(ConstantFPRange::getNonNaN(APFloat(-2.0), APFloat(3.0)).negate(), + ConstantFPRange::getNonNaN(APFloat(-3.0), APFloat(2.0))); + EXPECT_EQ(ConstantFPRange::getNonNaN(APFloat(-3.0), APFloat(2.0)).negate(), + ConstantFPRange::getNonNaN(APFloat(-2.0), APFloat(3.0))); +} + } // anonymous namespace diff --git a/llvm/utils/TableGen/Basic/RuntimeLibcallsEmitter.cpp b/llvm/utils/TableGen/Basic/RuntimeLibcallsEmitter.cpp index 10f0213..fd8ddb1 100644 --- a/llvm/utils/TableGen/Basic/RuntimeLibcallsEmitter.cpp +++ b/llvm/utils/TableGen/Basic/RuntimeLibcallsEmitter.cpp @@ -588,7 +588,12 @@ void RuntimeLibcallEmitter::emitSystemRuntimeLibrarySetCalls( PredicateSorter.insert( PredicateWithCC()); // No predicate or CC override first. + constexpr unsigned BitsPerStorageElt = 64; DenseMap<PredicateWithCC, LibcallsWithCC> Pred2Funcs; + + SmallVector<uint64_t, 32> BitsetValues( + divideCeil(RuntimeLibcallImplDefList.size(), BitsPerStorageElt)); + for (const Record *Elt : *Elements) { const RuntimeLibcallImpl *LibCallImpl = getRuntimeLibcallImpl(Elt); if (!LibCallImpl) { @@ -597,16 +602,24 @@ void RuntimeLibcallEmitter::emitSystemRuntimeLibrarySetCalls( continue; } + size_t BitIdx = LibCallImpl->getEnumVal(); + uint64_t BitmaskVal = uint64_t(1) << (BitIdx % BitsPerStorageElt); + size_t BitsetIdx = BitIdx / BitsPerStorageElt; + auto It = Func2Preds.find(LibCallImpl); if (It == Func2Preds.end()) { + BitsetValues[BitsetIdx] |= BitmaskVal; Pred2Funcs[PredicateWithCC()].LibcallImpls.push_back(LibCallImpl); continue; } for (const Record *Pred : It->second.first) { const Record *CC = It->second.second; - PredicateWithCC Key(Pred, CC); + AvailabilityPredicate SubsetPredicate(Pred); + if (SubsetPredicate.isAlwaysAvailable()) + BitsetValues[BitsetIdx] |= BitmaskVal; + PredicateWithCC Key(Pred, CC); auto &Entry = Pred2Funcs[Key]; Entry.LibcallImpls.push_back(LibCallImpl); Entry.CallingConv = It->second.second; @@ -614,6 +627,22 @@ void RuntimeLibcallEmitter::emitSystemRuntimeLibrarySetCalls( } } + OS << " static constexpr LibcallImplBitset SystemAvailableImpls({\n" + << indent(6); + + ListSeparator LS; + unsigned EntryCount = 0; + for (uint64_t Bits : BitsetValues) { + if (EntryCount++ == 4) { + EntryCount = 1; + OS << ",\n" << indent(6); + } else + OS << LS; + OS << format_hex(Bits, 16); + } + OS << "\n });\n" + " AvailableLibcallImpls = SystemAvailableImpls;\n\n"; + SmallVector<PredicateWithCC, 0> SortedPredicates = PredicateSorter.takeVector(); diff --git a/llvm/utils/gn/secondary/clang/lib/Analysis/BUILD.gn b/llvm/utils/gn/secondary/clang/lib/Analysis/BUILD.gn index 5f9eb9a..eb8aef2 100644 --- a/llvm/utils/gn/secondary/clang/lib/Analysis/BUILD.gn +++ b/llvm/utils/gn/secondary/clang/lib/Analysis/BUILD.gn @@ -27,8 +27,6 @@ static_library("Analysis") { "FixitUtil.cpp", "IntervalPartition.cpp", "IssueHash.cpp", - "LifetimeAnnotations.cpp", - "LifetimeSafety.cpp", "LiveVariables.cpp", "MacroExpansionContext.cpp", "ObjCNoReturn.cpp", diff --git a/llvm/utils/gn/secondary/clang/lib/Analysis/LifetimeSafety/BUILD.gn b/llvm/utils/gn/secondary/clang/lib/Analysis/LifetimeSafety/BUILD.gn new file mode 100644 index 0000000..a148e78 --- /dev/null +++ b/llvm/utils/gn/secondary/clang/lib/Analysis/LifetimeSafety/BUILD.gn @@ -0,0 +1,16 @@ +static_library("LifetimeSafety") { + output_name = "clangAnalysisLifetimeSafety" + configs += [ "//llvm/utils/gn/build:clang_code" ] + deps = [ "//clang/lib/Basic" ] + sources = [ + "Checker.cpp", + "Facts.cpp", + "FactsGenerator.cpp", + "LifetimeAnnotations.cpp", + "LifetimeSafety.cpp", + "LiveOrigins.cpp", + "Loans.cpp", + "LoanPropagation.cpp", + "Origins.cpp", + ] +} diff --git a/llvm/utils/gn/secondary/clang/lib/Sema/BUILD.gn b/llvm/utils/gn/secondary/clang/lib/Sema/BUILD.gn index 96ff481..14f44c4 100644 --- a/llvm/utils/gn/secondary/clang/lib/Sema/BUILD.gn +++ b/llvm/utils/gn/secondary/clang/lib/Sema/BUILD.gn @@ -29,6 +29,7 @@ static_library("Sema") { "//clang/lib/APINotes", "//clang/lib/AST", "//clang/lib/Analysis", + "//clang/lib/Analysis/LifetimeSafety", "//clang/lib/Basic", "//clang/lib/Edit", "//clang/lib/Lex", diff --git a/llvm/utils/gn/secondary/llvm/unittests/ADT/BUILD.gn b/llvm/utils/gn/secondary/llvm/unittests/ADT/BUILD.gn index 92e596e..8d19d30 100644 --- a/llvm/utils/gn/secondary/llvm/unittests/ADT/BUILD.gn +++ b/llvm/utils/gn/secondary/llvm/unittests/ADT/BUILD.gn @@ -21,6 +21,7 @@ unittest("ADTTests") { "BitTest.cpp", "BitVectorTest.cpp", "BitmaskEnumTest.cpp", + "BitsetTest.cpp", "BreadthFirstIteratorTest.cpp", "BumpPtrListTest.cpp", "CoalescingBitVectorTest.cpp", |