diff options
Diffstat (limited to 'llvm')
113 files changed, 6001 insertions, 2043 deletions
diff --git a/llvm/docs/HowToReleaseLLVM.rst b/llvm/docs/HowToReleaseLLVM.rst index 1795d3a..171bf88 100644 --- a/llvm/docs/HowToReleaseLLVM.rst +++ b/llvm/docs/HowToReleaseLLVM.rst @@ -18,11 +18,11 @@ create the binary packages, please refer to the :doc:`ReleaseProcess` instead. Release Timeline ================ -LLVM is released on a time based schedule --- with major releases roughly +LLVM is released on a time-based schedule --- with major releases roughly every 6 months. In between major releases there may be dot releases. The release manager will determine if and when to make a dot release based on feedback from the community. Typically, dot releases should be made if -there are large number of bug-fixes in the stable branch or a critical bug +there are a large number of bug fixes in the stable branch or a critical bug has been discovered that affects a large number of users. Unless otherwise stated, dot releases will follow the same procedure as @@ -73,7 +73,7 @@ Release Process Summary * Generate and send out the second release candidate sources. Only *critical* bugs found during this testing phase will be fixed. Any bugs introduced by - merged patches will be fixed. If so a third round of testing is needed. + merged patches will be fixed. If so, a third round of testing is needed. * The release notes are updated. @@ -107,15 +107,15 @@ Create Release Branch and Update LLVM Version Branch the Git trunk using the following procedure: #. Remind developers that the release branching is imminent and to refrain from - committing patches that might break the build. E.g., new features, large + committing patches that might break the build, e.g., new features, large patches for works in progress, an overhaul of the type system, an exciting new TableGen feature, etc. #. Verify that the current git trunk is in decent shape by examining nightly tester and buildbot results. -#. Bump the version in trunk to N.0.0git with the script in - ``llvm/utils/release/bump-version.py``, and tag the commit with llvmorg-N-init. +#. Bump the version in trunk to ``N.0.0git`` with the script in + ``llvm/utils/release/bump-version.py``, and tag the commit with ``llvmorg-N-init``. If ``X`` is the version to be released, then ``N`` is ``X + 1``. :: $ git tag -sa llvmorg-N-init @@ -124,14 +124,14 @@ Branch the Git trunk using the following procedure: ``llvm/utils/release/clear-release-notes.py``. #. Create the release branch from the last known good revision from before the - version bump. The branch's name is release/X.x where ``X`` is the major version + version bump. The branch's name is ``release/X.x`` where ``X`` is the major version number and ``x`` is just the letter ``x``. #. On the newly-created release branch, immediately bump the version - to X.1.0git (where ``X`` is the major version of the branch.) + to ``X.1.0git`` (where ``X`` is the major version of the branch.) -#. All tags and branches need to be created in both the llvm/llvm-project and - llvm/llvm-test-suite repos. +#. All tags and branches need to be created in both the ``llvm/llvm-project`` and + ``llvm/llvm-test-suite`` repos. Tagging the LLVM Release Candidates ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ @@ -157,7 +157,7 @@ the release page. $ for f in *.xz; do gh attestation verify --owner llvm $f && gpg -b $f; done Tarballs, release binaries, or any other release artifacts must be uploaded to -GitHub. This can be done using the github-upload-release.py script in utils/release. +GitHub. This can be done using the ``github-upload-release.py`` script in ``utils/release``. :: @@ -170,10 +170,10 @@ Build The Binary Distribution Creating the binary distribution requires following the instructions :doc:`here <ReleaseProcess>`. -That process will perform both Release+Asserts and Release builds but only -pack the Release build for upload. You should use the Release+Asserts sysroot, +That process performs both Release+Asserts and Release builds but only packs +the Release build for upload. You should use the Release+Asserts sysroot, normally under ``final/Phase3/Release+Asserts/llvmCore-3.8.1-RCn.install/``, -for test-suite and run-time benchmarks, to make sure nothing serious has +for test-suite and run-time benchmarks, to ensure nothing serious has passed through the net. For compile-time benchmarks, use the Release version. The minimum required version of the tools you'll need are :doc:`here <GettingStarted>` @@ -181,14 +181,14 @@ The minimum required version of the tools you'll need are :doc:`here <GettingSta Release Qualification Criteria ------------------------------ -There are no official release qualification criteria. It is up to the -the release manager to determine when a release is ready. The release manager +There are no official release qualification criteria. +The release manager determines when a release is ready. The release manager should pay attention to the results of community testing, the number of outstanding -bugs, and then number of regressions when determining whether or not to make a +bugs, and the number of regressions when determining whether or not to make a release. The community values time based releases, so releases should not be delayed for -too long unless there are critical issues remaining. In most cases, the only +too long unless critical issues remain. In most cases, the only kind of bugs that are critical enough to block a release would be a major regression from a previous release. @@ -199,33 +199,33 @@ A few developers in the community have dedicated time to validate the release candidates and volunteered to be the official release testers for each architecture. -These will be the ones testing, generating and uploading the official binaries +These will be the ones testing, generating, and uploading the official binaries to the server, and will be the minimum tests *necessary* for the release to proceed. This will obviously not cover all OSs and distributions, so additional community -validation is important. However, if community input is not reached before the -release is out, all bugs reported will have to go on the next stable release. +validation is important. However, if community input is not received before the +release, all reported bugs will be deferred to the next stable release. The official release managers are: * Even releases: Tom Stellard (tstellar@redhat.com) * Odd releases: Tobias Hieta (tobias@hieta.se) -The official release testers are volunteered from the community and have +The official release testers are volunteers from the community who have consistently validated and released binaries for their targets/OSs. To contact them, you should post on the `Discourse forums (Project Infrastructure - Release Testers). <https://discourse.llvm.org/c/infrastructure/release-testers/66>`_ -The official testers list is in the file `RELEASE_TESTERS.TXT +The official testers list is in the file ``RELEASE_TESTERS.TXT`` <https://github.com/llvm/llvm-project/blob/main/llvm/RELEASE_TESTERS.TXT>`_, in the LLVM repository. Community Testing ----------------- -Once all testing has been completed and appropriate bugs filed, the release -candidate tarballs are put on the website and the LLVM community is notified. +Once all testing is complete and appropriate bugs are filed, the release +candidate tarballs are put on the website, and the LLVM community is notified. We ask that all LLVM developers test the release in any the following ways: @@ -251,7 +251,7 @@ We ask that all LLVM developers test the release in any the following ways: architecture. We also ask that the OS distribution release managers test their packages with -the first candidate of every release, and report any *new* errors in GitHub. +the first candidate of every release and report any *new* errors in GitHub. If the bug can be reproduced with an unpatched upstream version of the release candidate (as opposed to the distribution's own build), the priority should be release blocker. @@ -268,10 +268,10 @@ next stage. Reporting Regressions --------------------- -Every regression that is found during the tests (as per the criteria above), +Every regression found during the tests (as per the criteria above) should be filled in a bug in GitHub and added to the release milestone. -If a bug can't be reproduced, or stops being a blocker, it should be removed +If a bug can't be reproduced or stops being a blocker, it should be removed from the Milestone. Debugging can continue, but on trunk. Backport Requests @@ -299,15 +299,15 @@ This section describes how to triage bug reports: to see the list of bugs that are being considered for the release. #. Review each bug and first check if it has been fixed in main. If it has, update - its status to "Needs Pull Request", and create a pull request for the fix - using the /cherry-pick or /branch comments if this has not been done already. + its status to "Needs Pull Request" and create a pull request for the fix + using the ``/cherry-pick`` or ``/branch`` comments if this has not been done already. #. If a bug has been fixed and has a pull request created for backporting it, then update its status to "Needs Review" and notify a knowledgeable reviewer. Usually you will want to notify the person who approved the patch, but you may use your best judgement on who a good reviewer would be. Once you have identified the reviewer(s), assign the issue to them and - mention them (i.e @username) in a comment and ask them if the patch is safe + mention them (i.e., ``@username``) in a comment and ask them if the patch is safe to backport. You should also review the bug yourself to ensure that it meets the requirements for committing to the release branch. @@ -323,11 +323,11 @@ Release Patch Rules Below are the rules regarding patching the release branch: #. Patches applied to the release branch may only be applied by the release - manager, the official release testers or the maintainers with approval from + manager, the official release testers, or the maintainers with approval from the release manager. #. Release managers are encouraged, but not required, to get approval from a - maintainer before approving patches. If there are no reachable maintainers + maintainer before approving patches. If there are no reachable maintainers, then release managers can ask approval from patch reviewers or other developers active in that area. @@ -336,7 +336,7 @@ Below are the rules regarding patching the release branch: was created. As with all phases, release managers and maintainers can reject patches that are deemed too invasive. -#. *Before RC2/RC3* Patches should be limited to bug fixes or backend specific +#. *Before RC2/RC3* Patches should be limited to bug fixes or backend-specific improvements that are determined to be very safe. #. *Before Final Major Release* Patches should be limited to critical @@ -349,7 +349,7 @@ Below are the rules regarding patching the release branch: Release Final Tasks ------------------- -The final stages of the release process involves tagging the "final" release +The final stages of the release process involve tagging the "final" release branch, updating documentation that refers to the release, and updating the demo page. @@ -394,11 +394,11 @@ is what to do: #. Update the ``releases/index.html`` with the new release and link to release documentation. -#. After you push the changes to the www-releases repo, someone with admin - access must login to prereleases-origin.llvm.org and manually pull the new - changes into /data/www-releases/. This is where the website is served from. +#. After you push the changes to the ``www-releases`` repo, someone with admin + access must log in to ``prereleases-origin.llvm.org`` and manually pull the new + changes into ``/data/www-releases/``. This is where the website is served from. -#. Finally checkout the llvm-www repo and update the main page +#. Finally, check out the ``llvm-www`` repo and update the main page (``index.html`` and sidebar) to point to the new release and release announcement. @@ -414,5 +414,5 @@ using this command and add it to the post. $ git log --format="- %aN: [%s (%h)](https://github.com/llvm/llvm-project/commit/%H)" llvmorg-X.1.N-1..llvmorg-X.1.N -Once the release has been announced add a link to the announcement on the llvm -homepage (from the llvm-www repo) in the "Release Emails" section. +Once the release has been announced, add a link to the announcement on the llvm +homepage (from the ``llvm-www`` repo) in the "Release Emails" section. diff --git a/llvm/docs/ReleaseNotes.md b/llvm/docs/ReleaseNotes.md index c352cd6..9cdd983 100644 --- a/llvm/docs/ReleaseNotes.md +++ b/llvm/docs/ReleaseNotes.md @@ -140,6 +140,7 @@ Changes to the X86 Backend -------------------------- * `-mcpu=wildcatlake` is now supported. +* `-mcpu=novalake` is now supported. Changes to the OCaml bindings ----------------------------- diff --git a/llvm/include/llvm/Analysis/DXILResource.h b/llvm/include/llvm/Analysis/DXILResource.h index 88ac0a1..c7aff16 100644 --- a/llvm/include/llvm/Analysis/DXILResource.h +++ b/llvm/include/llvm/Analysis/DXILResource.h @@ -243,6 +243,25 @@ public: } }; +/// The dx.Padding target extension type +/// +/// `target("dx.Padding", NumBytes)` +class PaddingExtType : public TargetExtType { +public: + PaddingExtType() = delete; + PaddingExtType(const PaddingExtType &) = delete; + PaddingExtType &operator=(const PaddingExtType &) = delete; + + unsigned getNumBytes() const { return getIntParameter(0); } + + static bool classof(const TargetExtType *T) { + return T->getName() == "dx.Padding"; + } + static bool classof(const Type *T) { + return isa<TargetExtType>(T) && classof(cast<TargetExtType>(T)); + } +}; + //===----------------------------------------------------------------------===// class ResourceTypeInfo { diff --git a/llvm/include/llvm/Analysis/LoopInfo.h b/llvm/include/llvm/Analysis/LoopInfo.h index a7a6a27..0ecb114 100644 --- a/llvm/include/llvm/Analysis/LoopInfo.h +++ b/llvm/include/llvm/Analysis/LoopInfo.h @@ -617,7 +617,7 @@ public: }; /// Function to print a loop's contents as LLVM's text IR assembly. -LLVM_ABI void printLoop(Loop &L, raw_ostream &OS, +LLVM_ABI void printLoop(const Loop &L, raw_ostream &OS, const std::string &Banner = ""); /// Find and return the loop attribute node for the attribute @p Name in diff --git a/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h b/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h index 871028d..68198ec 100644 --- a/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h +++ b/llvm/include/llvm/Analysis/ScalarEvolutionPatternMatch.h @@ -95,6 +95,10 @@ inline bind_ty<const SCEVAddExpr> m_scev_Add(const SCEVAddExpr *&V) { return V; } +inline bind_ty<const SCEVMulExpr> m_scev_Mul(const SCEVMulExpr *&V) { + return V; +} + /// Match a specified const SCEV *. struct specificscev_ty { const SCEV *Expr; @@ -284,14 +288,10 @@ template <typename Op0_t, typename Op1_t> struct SCEVURem_match { << SE.getTypeSizeInBits(TruncTy)); return Op0.match(LHS) && Op1.match(RHS); } - const auto *Add = dyn_cast<SCEVAddExpr>(Expr); - if (Add == nullptr || Add->getNumOperands() != 2) - return false; - - const SCEV *A = Add->getOperand(1); - const auto *Mul = dyn_cast<SCEVMulExpr>(Add->getOperand(0)); - if (Mul == nullptr) + const SCEV *A; + const SCEVMulExpr *Mul; + if (!SCEVPatternMatch::match(Expr, m_scev_Add(m_scev_Mul(Mul), m_SCEV(A)))) return false; const auto MatchURemWithDivisor = [&](const SCEV *B) { diff --git a/llvm/include/llvm/BinaryFormat/ELF.h b/llvm/include/llvm/BinaryFormat/ELF.h index e619b18..8d0dc64 100644 --- a/llvm/include/llvm/BinaryFormat/ELF.h +++ b/llvm/include/llvm/BinaryFormat/ELF.h @@ -647,6 +647,7 @@ enum { EF_HEXAGON_ISA_V85 = 0x00000085, // Hexagon V85 ISA EF_HEXAGON_ISA_V87 = 0x00000087, // Hexagon V87 ISA EF_HEXAGON_ISA_V89 = 0x00000089, // Hexagon V89 ISA + EF_HEXAGON_ISA_V91 = 0x00000091, // Hexagon V91 ISA EF_HEXAGON_ISA = 0x000003ff, // Hexagon V.. ISA // Tiny core flag, bit[15] @@ -680,6 +681,7 @@ enum { EF_HEXAGON_MACH_V85 = EF_HEXAGON_ISA_V85, // Hexagon V85 EF_HEXAGON_MACH_V87 = EF_HEXAGON_ISA_V87, // Hexagon V87 EF_HEXAGON_MACH_V89 = EF_HEXAGON_ISA_V89, // Hexagon V89 + EF_HEXAGON_MACH_V91 = EF_HEXAGON_ISA_V91, // Hexagon V91 EF_HEXAGON_MACH = 0x0000ffff, // Hexagon V.. }; diff --git a/llvm/include/llvm/Frontend/OpenMP/ClauseT.h b/llvm/include/llvm/Frontend/OpenMP/ClauseT.h index d49bade..1a01fa6 100644 --- a/llvm/include/llvm/Frontend/OpenMP/ClauseT.h +++ b/llvm/include/llvm/Frontend/OpenMP/ClauseT.h @@ -571,7 +571,9 @@ struct DoacrossT { // V5.2: [8.2.1] `requirement` clauses template <typename T, typename I, typename E> // struct DynamicAllocatorsT { - using EmptyTrait = std::true_type; + using Requires = E; + using WrapperTrait = std::true_type; + OPT(Requires) v; }; template <typename T, typename I, typename E> // @@ -1056,7 +1058,9 @@ struct ReplayableT { // V5.2: [8.2.1] `requirement` clauses template <typename T, typename I, typename E> // struct ReverseOffloadT { - using EmptyTrait = std::true_type; + using Requires = E; + using WrapperTrait = std::true_type; + OPT(Requires) v; }; // V5.2: [10.4.2] `safelen` clause @@ -1078,6 +1082,14 @@ struct ScheduleT { std::tuple<Kind, OPT(OrderingModifier), OPT(ChunkModifier), OPT(ChunkSize)> t; }; +// [6.0:361] +template <typename T, typename I, typename E> // +struct SelfMapsT { + using Requires = E; + using WrapperTrait = std::true_type; + OPT(Requires) v; +}; + // V5.2: [15.8.1] Memory-order clauses template <typename T, typename I, typename E> // struct SeqCstT { @@ -1169,18 +1181,17 @@ struct TransparentT { // V5.2: [8.2.1] `requirement` clauses template <typename T, typename I, typename E> // struct UnifiedAddressT { - using EmptyTrait = std::true_type; + using Requires = E; + using WrapperTrait = std::true_type; + OPT(Requires) v; }; // V5.2: [8.2.1] `requirement` clauses template <typename T, typename I, typename E> // struct UnifiedSharedMemoryT { - using EmptyTrait = std::true_type; -}; - -template <typename T, typename I, typename E> // -struct SelfMapsT { - using EmptyTrait = std::true_type; + using Requires = E; + using WrapperTrait = std::true_type; + OPT(Requires) v; }; // V5.2: [5.10] `uniform` clause @@ -1288,14 +1299,12 @@ using ExtensionClausesT = template <typename T, typename I, typename E> using EmptyClausesT = std::variant< AcqRelT<T, I, E>, AcquireT<T, I, E>, CaptureT<T, I, E>, CompareT<T, I, E>, - DynamicAllocatorsT<T, I, E>, FullT<T, I, E>, InbranchT<T, I, E>, - MergeableT<T, I, E>, NogroupT<T, I, E>, NoOpenmpRoutinesT<T, I, E>, + FullT<T, I, E>, InbranchT<T, I, E>, MergeableT<T, I, E>, NogroupT<T, I, E>, + NoOpenmpConstructsT<T, I, E>, NoOpenmpRoutinesT<T, I, E>, NoOpenmpT<T, I, E>, NoParallelismT<T, I, E>, NotinbranchT<T, I, E>, NowaitT<T, I, E>, ReadT<T, I, E>, RelaxedT<T, I, E>, ReleaseT<T, I, E>, - ReverseOffloadT<T, I, E>, SeqCstT<T, I, E>, SimdT<T, I, E>, - ThreadsT<T, I, E>, UnifiedAddressT<T, I, E>, UnifiedSharedMemoryT<T, I, E>, - UnknownT<T, I, E>, UntiedT<T, I, E>, UseT<T, I, E>, WeakT<T, I, E>, - WriteT<T, I, E>, NoOpenmpConstructsT<T, I, E>, SelfMapsT<T, I, E>>; + SeqCstT<T, I, E>, SimdT<T, I, E>, ThreadsT<T, I, E>, UnknownT<T, I, E>, + UntiedT<T, I, E>, UseT<T, I, E>, WeakT<T, I, E>, WriteT<T, I, E>>; template <typename T, typename I, typename E> using IncompleteClausesT = @@ -1323,18 +1332,20 @@ using WrapperClausesT = std::variant< AtomicDefaultMemOrderT<T, I, E>, AtT<T, I, E>, BindT<T, I, E>, CollapseT<T, I, E>, ContainsT<T, I, E>, CopyinT<T, I, E>, CopyprivateT<T, I, E>, DefaultT<T, I, E>, DestroyT<T, I, E>, - DetachT<T, I, E>, DeviceTypeT<T, I, E>, EnterT<T, I, E>, - ExclusiveT<T, I, E>, FailT<T, I, E>, FilterT<T, I, E>, FinalT<T, I, E>, - FirstprivateT<T, I, E>, HasDeviceAddrT<T, I, E>, HintT<T, I, E>, - HoldsT<T, I, E>, InclusiveT<T, I, E>, IndirectT<T, I, E>, + DetachT<T, I, E>, DeviceTypeT<T, I, E>, DynamicAllocatorsT<T, I, E>, + EnterT<T, I, E>, ExclusiveT<T, I, E>, FailT<T, I, E>, FilterT<T, I, E>, + FinalT<T, I, E>, FirstprivateT<T, I, E>, HasDeviceAddrT<T, I, E>, + HintT<T, I, E>, HoldsT<T, I, E>, InclusiveT<T, I, E>, IndirectT<T, I, E>, InitializerT<T, I, E>, IsDevicePtrT<T, I, E>, LinkT<T, I, E>, MessageT<T, I, E>, NocontextT<T, I, E>, NontemporalT<T, I, E>, NovariantsT<T, I, E>, NumTeamsT<T, I, E>, NumThreadsT<T, I, E>, OrderedT<T, I, E>, PartialT<T, I, E>, PriorityT<T, I, E>, PrivateT<T, I, E>, - ProcBindT<T, I, E>, SafelenT<T, I, E>, SeverityT<T, I, E>, SharedT<T, I, E>, - SimdlenT<T, I, E>, SizesT<T, I, E>, PermutationT<T, I, E>, - ThreadLimitT<T, I, E>, UniformT<T, I, E>, UpdateT<T, I, E>, - UseDeviceAddrT<T, I, E>, UseDevicePtrT<T, I, E>, UsesAllocatorsT<T, I, E>>; + ProcBindT<T, I, E>, ReverseOffloadT<T, I, E>, SafelenT<T, I, E>, + SelfMapsT<T, I, E>, SeverityT<T, I, E>, SharedT<T, I, E>, SimdlenT<T, I, E>, + SizesT<T, I, E>, PermutationT<T, I, E>, ThreadLimitT<T, I, E>, + UnifiedAddressT<T, I, E>, UnifiedSharedMemoryT<T, I, E>, UniformT<T, I, E>, + UpdateT<T, I, E>, UseDeviceAddrT<T, I, E>, UseDevicePtrT<T, I, E>, + UsesAllocatorsT<T, I, E>>; template <typename T, typename I, typename E> using UnionOfAllClausesT = typename type::Union< // diff --git a/llvm/include/llvm/Frontend/OpenMP/OMP.td b/llvm/include/llvm/Frontend/OpenMP/OMP.td index 86a9e24..edcf7a9 100644 --- a/llvm/include/llvm/Frontend/OpenMP/OMP.td +++ b/llvm/include/llvm/Frontend/OpenMP/OMP.td @@ -177,6 +177,8 @@ def OMPC_Doacross : Clause<[Spelling<"doacross">]> { } def OMPC_DynamicAllocators : Clause<[Spelling<"dynamic_allocators">]> { let clangClass = "OMPDynamicAllocatorsClause"; + let flangClass = "OmpDynamicAllocatorsClause"; + let isValueOptional = true; } def OMPC_DynGroupprivate : Clause<[Spelling<"dyn_groupprivate">]> { let flangClass = "OmpDynGroupprivateClause"; @@ -467,6 +469,8 @@ def OMPC_Replayable : Clause<[Spelling<"replayable">]> { } def OMPC_ReverseOffload : Clause<[Spelling<"reverse_offload">]> { let clangClass = "OMPReverseOffloadClause"; + let flangClass = "OmpReverseOffloadClause"; + let isValueOptional = true; } def OMPC_SafeLen : Clause<[Spelling<"safelen">]> { let clangClass = "OMPSafelenClause"; @@ -541,12 +545,18 @@ def OMPC_Transparent : Clause<[Spelling<"transparent">]> { } def OMPC_UnifiedAddress : Clause<[Spelling<"unified_address">]> { let clangClass = "OMPUnifiedAddressClause"; + let flangClass = "OmpUnifiedAddressClause"; + let isValueOptional = true; } def OMPC_UnifiedSharedMemory : Clause<[Spelling<"unified_shared_memory">]> { let clangClass = "OMPUnifiedSharedMemoryClause"; + let flangClass = "OmpUnifiedSharedMemoryClause"; + let isValueOptional = true; } def OMPC_SelfMaps : Clause<[Spelling<"self_maps">]> { let clangClass = "OMPSelfMapsClause"; + let flangClass = "OmpSelfMapsClause"; + let isValueOptional = true; } def OMPC_Uniform : Clause<[Spelling<"uniform">]> { let flangClass = "Name"; diff --git a/llvm/include/llvm/LTO/LTO.h b/llvm/include/llvm/LTO/LTO.h index 3a9a7f7..000472f 100644 --- a/llvm/include/llvm/LTO/LTO.h +++ b/llvm/include/llvm/LTO/LTO.h @@ -105,12 +105,6 @@ setupStatsFile(StringRef StatsFilename); /// ordered indices to elements in the input array. LLVM_ABI std::vector<int> generateModulesOrdering(ArrayRef<BitcodeModule *> R); -/// Updates MemProf attributes (and metadata) based on whether the index -/// has recorded that we are linking with allocation libraries containing -/// the necessary APIs for downstream transformations. -LLVM_ABI void updateMemProfAttributes(Module &Mod, - const ModuleSummaryIndex &Index); - class LTO; struct SymbolResolution; diff --git a/llvm/include/llvm/Support/X86DisassemblerDecoderCommon.h b/llvm/include/llvm/Support/X86DisassemblerDecoderCommon.h index faaff4a..4aa6c01 100644 --- a/llvm/include/llvm/Support/X86DisassemblerDecoderCommon.h +++ b/llvm/include/llvm/Support/X86DisassemblerDecoderCommon.h @@ -121,6 +121,7 @@ enum attributeBits { "The Dynamic Duo! Prefer over all else because this changes " \ "most operands' meaning") \ ENUM_ENTRY(IC_64BIT_REX2, 2, "requires a REX2 prefix") \ + ENUM_ENTRY(IC_64BIT_REX2_REXW, 3, "requires a REX2 and the W prefix") \ ENUM_ENTRY(IC_VEX, 1, "requires a VEX prefix") \ ENUM_ENTRY(IC_VEX_XS, 2, "requires VEX and the XS prefix") \ ENUM_ENTRY(IC_VEX_XD, 2, "requires VEX and the XD prefix") \ diff --git a/llvm/include/llvm/TargetParser/X86TargetParser.def b/llvm/include/llvm/TargetParser/X86TargetParser.def index 254587b..a94eab1 100644 --- a/llvm/include/llvm/TargetParser/X86TargetParser.def +++ b/llvm/include/llvm/TargetParser/X86TargetParser.def @@ -108,6 +108,7 @@ X86_CPU_SUBTYPE(INTEL_COREI7_ARROWLAKE_S, "arrowlake-s") X86_CPU_SUBTYPE(INTEL_COREI7_PANTHERLAKE, "pantherlake") X86_CPU_SUBTYPE(AMDFAM1AH_ZNVER5, "znver5") X86_CPU_SUBTYPE(INTEL_COREI7_DIAMONDRAPIDS, "diamondrapids") +X86_CPU_SUBTYPE(INTEL_COREI7_NOVALAKE, "novalake") // Alternate names supported by __builtin_cpu_is and target multiversioning. X86_CPU_SUBTYPE_ALIAS(INTEL_COREI7_ALDERLAKE, "raptorlake") diff --git a/llvm/include/llvm/TargetParser/X86TargetParser.h b/llvm/include/llvm/TargetParser/X86TargetParser.h index e4c43cd..80f3d35 100644 --- a/llvm/include/llvm/TargetParser/X86TargetParser.h +++ b/llvm/include/llvm/TargetParser/X86TargetParser.h @@ -117,6 +117,7 @@ enum CPUKind { CK_Lunarlake, CK_Pantherlake, CK_Wildcatlake, + CK_Novalake, CK_Sierraforest, CK_Grandridge, CK_Graniterapids, diff --git a/llvm/include/llvm/Transforms/IPO/MemProfContextDisambiguation.h b/llvm/include/llvm/Transforms/IPO/MemProfContextDisambiguation.h index f2de083..576f1eb 100644 --- a/llvm/include/llvm/Transforms/IPO/MemProfContextDisambiguation.h +++ b/llvm/include/llvm/Transforms/IPO/MemProfContextDisambiguation.h @@ -95,6 +95,16 @@ public: function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> isPrevailing); }; + +/// Strips MemProf attributes and metadata. Can be invoked by the pass pipeline +/// when we don't have an index that has recorded that we are linking with +/// allocation libraries containing the necessary APIs for downstream +/// transformations. +class MemProfRemoveInfo : public PassInfoMixin<MemProfRemoveInfo> { +public: + PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM); +}; + } // end namespace llvm #endif // LLVM_TRANSFORMS_IPO_MEMPROF_CONTEXT_DISAMBIGUATION_H diff --git a/llvm/lib/Analysis/DXILResource.cpp b/llvm/lib/Analysis/DXILResource.cpp index b78cc03e..f9bf092 100644 --- a/llvm/lib/Analysis/DXILResource.cpp +++ b/llvm/lib/Analysis/DXILResource.cpp @@ -281,6 +281,38 @@ static StructType *getOrCreateElementStruct(Type *ElemType, StringRef Name) { return StructType::create(ElemType, Name); } +static Type *getTypeWithoutPadding(Type *Ty) { + // Recursively remove padding from structures. + if (auto *ST = dyn_cast<StructType>(Ty)) { + LLVMContext &Ctx = Ty->getContext(); + SmallVector<Type *> ElementTypes; + ElementTypes.reserve(ST->getNumElements()); + for (Type *ElTy : ST->elements()) { + if (isa<PaddingExtType>(ElTy)) + continue; + ElementTypes.push_back(getTypeWithoutPadding(ElTy)); + } + + // Handle explicitly padded cbuffer arrays like { [ n x paddedty ], ty } + if (ElementTypes.size() == 2) + if (auto *AT = dyn_cast<ArrayType>(ElementTypes[0])) + if (ElementTypes[1] == AT->getElementType()) + return ArrayType::get(ElementTypes[1], AT->getNumElements() + 1); + + // If we only have a single element, don't wrap it in a struct. + if (ElementTypes.size() == 1) + return ElementTypes[0]; + + return StructType::get(Ctx, ElementTypes, /*IsPacked=*/false); + } + // Arrays just need to have their element type adjusted. + if (auto *AT = dyn_cast<ArrayType>(Ty)) + return ArrayType::get(getTypeWithoutPadding(AT->getElementType()), + AT->getNumElements()); + // Anything else should be good as is. + return Ty; +} + StructType *ResourceTypeInfo::createElementStruct(StringRef CBufferName) { SmallString<64> TypeName; @@ -334,14 +366,21 @@ StructType *ResourceTypeInfo::createElementStruct(StringRef CBufferName) { } case ResourceKind::CBuffer: { auto *RTy = cast<CBufferExtType>(HandleTy); - LayoutExtType *LayoutType = cast<LayoutExtType>(RTy->getResourceType()); - StructType *Ty = cast<StructType>(LayoutType->getWrappedType()); SmallString<64> Name = getResourceKindName(Kind); if (!CBufferName.empty()) { Name.append("."); Name.append(CBufferName); } - return StructType::create(Ty->elements(), Name); + + // TODO: Remove this when we update the frontend to use explicit padding. + if (LayoutExtType *LayoutType = + dyn_cast<LayoutExtType>(RTy->getResourceType())) { + StructType *Ty = cast<StructType>(LayoutType->getWrappedType()); + return StructType::create(Ty->elements(), Name); + } + + return getOrCreateElementStruct( + getTypeWithoutPadding(RTy->getResourceType()), Name); } case ResourceKind::Sampler: { auto *RTy = cast<SamplerExtType>(HandleTy); @@ -454,10 +493,10 @@ uint32_t ResourceTypeInfo::getCBufferSize(const DataLayout &DL) const { Type *ElTy = cast<CBufferExtType>(HandleTy)->getResourceType(); + // TODO: Remove this when we update the frontend to use explicit padding. if (auto *LayoutTy = dyn_cast<LayoutExtType>(ElTy)) return LayoutTy->getSize(); - // TODO: What should we do with unannotated arrays? return DL.getTypeAllocSize(ElTy); } diff --git a/llvm/lib/Analysis/LoopInfo.cpp b/llvm/lib/Analysis/LoopInfo.cpp index a8c3173..d84721b 100644 --- a/llvm/lib/Analysis/LoopInfo.cpp +++ b/llvm/lib/Analysis/LoopInfo.cpp @@ -986,8 +986,8 @@ PreservedAnalyses LoopPrinterPass::run(Function &F, return PreservedAnalyses::all(); } -void llvm::printLoop(Loop &L, raw_ostream &OS, const std::string &Banner) { - +void llvm::printLoop(const Loop &L, raw_ostream &OS, + const std::string &Banner) { if (forcePrintModuleIR()) { // handling -print-module-scope OS << Banner << " (loop: "; diff --git a/llvm/lib/Analysis/ScalarEvolution.cpp b/llvm/lib/Analysis/ScalarEvolution.cpp index a64b93d..442b9d1 100644 --- a/llvm/lib/Analysis/ScalarEvolution.cpp +++ b/llvm/lib/Analysis/ScalarEvolution.cpp @@ -4623,17 +4623,11 @@ const SCEV *ScalarEvolution::getNegativeSCEV(const SCEV *V, /// If Expr computes ~A, return A else return nullptr static const SCEV *MatchNotExpr(const SCEV *Expr) { - const SCEVAddExpr *Add = dyn_cast<SCEVAddExpr>(Expr); - if (!Add || Add->getNumOperands() != 2 || - !Add->getOperand(0)->isAllOnesValue()) - return nullptr; - - const SCEVMulExpr *AddRHS = dyn_cast<SCEVMulExpr>(Add->getOperand(1)); - if (!AddRHS || AddRHS->getNumOperands() != 2 || - !AddRHS->getOperand(0)->isAllOnesValue()) - return nullptr; - - return AddRHS->getOperand(1); + const SCEV *MulOp; + if (match(Expr, m_scev_Add(m_scev_AllOnes(), + m_scev_Mul(m_scev_AllOnes(), m_SCEV(MulOp))))) + return MulOp; + return nullptr; } /// Return a SCEV corresponding to ~V = -1-V @@ -12220,12 +12214,11 @@ ScalarEvolution::computeConstantDifference(const SCEV *More, const SCEV *Less) { // Try to match a common constant multiply. auto MatchConstMul = [](const SCEV *S) -> std::optional<std::pair<const SCEV *, APInt>> { - auto *M = dyn_cast<SCEVMulExpr>(S); - if (!M || M->getNumOperands() != 2 || - !isa<SCEVConstant>(M->getOperand(0))) - return std::nullopt; - return { - {M->getOperand(1), cast<SCEVConstant>(M->getOperand(0))->getAPInt()}}; + const APInt *C; + const SCEV *Op; + if (match(S, m_scev_Mul(m_scev_APInt(C), m_SCEV(Op)))) + return {{Op, *C}}; + return std::nullopt; }; if (auto MatchedMore = MatchConstMul(More)) { if (auto MatchedLess = MatchConstMul(Less)) { diff --git a/llvm/lib/IR/Type.cpp b/llvm/lib/IR/Type.cpp index 9db48e8..0e9535d 100644 --- a/llvm/lib/IR/Type.cpp +++ b/llvm/lib/IR/Type.cpp @@ -1034,6 +1034,10 @@ static TargetTypeInfo getTargetTypeInfo(const TargetExtType *Ty) { } // DirectX resources + if (Name == "dx.Padding") + return TargetTypeInfo( + ArrayType::get(Type::getInt8Ty(C), Ty->getIntParameter(0)), + TargetExtType::CanBeGlobal); if (Name.starts_with("dx.")) return TargetTypeInfo(PointerType::get(C, 0), TargetExtType::CanBeGlobal, TargetExtType::CanBeLocal, diff --git a/llvm/lib/LTO/LTO.cpp b/llvm/lib/LTO/LTO.cpp index e6544f3..aec8891 100644 --- a/llvm/lib/LTO/LTO.cpp +++ b/llvm/lib/LTO/LTO.cpp @@ -1257,38 +1257,6 @@ Error LTO::run(AddStreamFn AddStream, FileCache Cache) { return Result; } -void lto::updateMemProfAttributes(Module &Mod, - const ModuleSummaryIndex &Index) { - llvm::TimeTraceScope timeScope("LTO update memprof attributes"); - if (Index.withSupportsHotColdNew()) - return; - - // The profile matcher applies hotness attributes directly for allocations, - // and those will cause us to generate calls to the hot/cold interfaces - // unconditionally. If supports-hot-cold-new was not enabled in the LTO - // link then assume we don't want these calls (e.g. not linking with - // the appropriate library, or otherwise trying to disable this behavior). - for (auto &F : Mod) { - for (auto &BB : F) { - for (auto &I : BB) { - auto *CI = dyn_cast<CallBase>(&I); - if (!CI) - continue; - if (CI->hasFnAttr("memprof")) - CI->removeFnAttr("memprof"); - // Strip off all memprof metadata as it is no longer needed. - // Importantly, this avoids the addition of new memprof attributes - // after inlining propagation. - // TODO: If we support additional types of MemProf metadata beyond hot - // and cold, we will need to update the metadata based on the allocator - // APIs supported instead of completely stripping all. - CI->setMetadata(LLVMContext::MD_memprof, nullptr); - CI->setMetadata(LLVMContext::MD_callsite, nullptr); - } - } - } -} - Error LTO::runRegularLTO(AddStreamFn AddStream) { llvm::TimeTraceScope timeScope("Run regular LTO"); LLVMContext &CombinedCtx = RegularLTO.CombinedModule->getContext(); @@ -1346,8 +1314,6 @@ Error LTO::runRegularLTO(AddStreamFn AddStream) { } } - updateMemProfAttributes(*RegularLTO.CombinedModule, ThinLTO.CombinedIndex); - bool WholeProgramVisibilityEnabledInLTO = Conf.HasWholeProgramVisibility && // If validation is enabled, upgrade visibility only when all vtables diff --git a/llvm/lib/LTO/LTOBackend.cpp b/llvm/lib/LTO/LTOBackend.cpp index 11a7b32..280c3d1 100644 --- a/llvm/lib/LTO/LTOBackend.cpp +++ b/llvm/lib/LTO/LTOBackend.cpp @@ -726,7 +726,6 @@ Error lto::thinBackend(const Config &Conf, unsigned Task, AddStreamFn AddStream, } // Do this after any importing so that imported code is updated. - updateMemProfAttributes(Mod, CombinedIndex); updatePublicTypeTestCalls(Mod, CombinedIndex.withWholeProgramVisibility()); if (Conf.PostImportModuleHook && !Conf.PostImportModuleHook(Task, Mod)) diff --git a/llvm/lib/ObjectYAML/ELFYAML.cpp b/llvm/lib/ObjectYAML/ELFYAML.cpp index 421d6603..c3a27c9 100644 --- a/llvm/lib/ObjectYAML/ELFYAML.cpp +++ b/llvm/lib/ObjectYAML/ELFYAML.cpp @@ -488,6 +488,7 @@ void ScalarBitSetTraits<ELFYAML::ELF_EF>::bitset(IO &IO, BCaseMask(EF_HEXAGON_MACH_V5, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V55, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V60, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V61, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V62, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V65, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V66, EF_HEXAGON_MACH); @@ -499,12 +500,21 @@ void ScalarBitSetTraits<ELFYAML::ELF_EF>::bitset(IO &IO, BCaseMask(EF_HEXAGON_MACH_V71T, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V73, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_MACH_V75, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V77, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V79, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V81, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V83, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V85, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V87, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V89, EF_HEXAGON_MACH); + BCaseMask(EF_HEXAGON_MACH_V91, EF_HEXAGON_MACH); BCaseMask(EF_HEXAGON_ISA_V2, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V3, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V4, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V5, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V55, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V60, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V61, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V62, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V65, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V66, EF_HEXAGON_ISA); @@ -514,6 +524,14 @@ void ScalarBitSetTraits<ELFYAML::ELF_EF>::bitset(IO &IO, BCaseMask(EF_HEXAGON_ISA_V71, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V73, EF_HEXAGON_ISA); BCaseMask(EF_HEXAGON_ISA_V75, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V77, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V79, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V81, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V83, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V85, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V87, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V89, EF_HEXAGON_ISA); + BCaseMask(EF_HEXAGON_ISA_V91, EF_HEXAGON_ISA); break; case ELF::EM_AVR: BCaseMask(EF_AVR_ARCH_AVR1, EF_AVR_ARCH_MASK); diff --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp index fea0d25..3f3939eaf 100644 --- a/llvm/lib/Passes/PassBuilderPipelines.cpp +++ b/llvm/lib/Passes/PassBuilderPipelines.cpp @@ -1658,6 +1658,16 @@ PassBuilder::buildPerModuleDefaultPipeline(OptimizationLevel Level, ModulePassManager MPM; + // Currently this pipeline is only invoked in an LTO pre link pass or when we + // are not running LTO. If that changes the below checks may need updating. + assert(isLTOPreLink(Phase) || Phase == ThinOrFullLTOPhase::None); + + // If we are invoking this in non-LTO mode, remove any MemProf related + // attributes and metadata, as we don't know whether we are linking with + // a library containing the necessary interfaces. + if (Phase == ThinOrFullLTOPhase::None) + MPM.addPass(MemProfRemoveInfo()); + // Convert @llvm.global.annotations to !annotation metadata. MPM.addPass(Annotation2MetadataPass()); @@ -1803,6 +1813,12 @@ ModulePassManager PassBuilder::buildThinLTODefaultPipeline( OptimizationLevel Level, const ModuleSummaryIndex *ImportSummary) { ModulePassManager MPM; + // If we are invoking this without a summary index noting that we are linking + // with a library containing the necessary APIs, remove any MemProf related + // attributes and metadata. + if (!ImportSummary || !ImportSummary->withSupportsHotColdNew()) + MPM.addPass(MemProfRemoveInfo()); + if (ImportSummary) { // For ThinLTO we must apply the context disambiguation decisions early, to // ensure we can correctly match the callsites to summary data. @@ -1874,6 +1890,12 @@ PassBuilder::buildLTODefaultPipeline(OptimizationLevel Level, invokeFullLinkTimeOptimizationEarlyEPCallbacks(MPM, Level); + // If we are invoking this without a summary index noting that we are linking + // with a library containing the necessary APIs, remove any MemProf related + // attributes and metadata. + if (!ExportSummary || !ExportSummary->withSupportsHotColdNew()) + MPM.addPass(MemProfRemoveInfo()); + // Create a function that performs CFI checks for cross-DSO calls with targets // in the current module. MPM.addPass(CrossDSOCFIPass()); diff --git a/llvm/lib/Passes/PassRegistry.def b/llvm/lib/Passes/PassRegistry.def index 1b16525..884d8da 100644 --- a/llvm/lib/Passes/PassRegistry.def +++ b/llvm/lib/Passes/PassRegistry.def @@ -113,6 +113,7 @@ MODULE_PASS("pgo-force-function-attrs", ? PGOOpt->ColdOptType : PGOOptions::ColdFuncOpt::Default)) MODULE_PASS("memprof-context-disambiguation", MemProfContextDisambiguation()) +MODULE_PASS("memprof-remove-attributes", MemProfRemoveInfo()) MODULE_PASS("memprof-module", ModuleMemProfilerPass()) MODULE_PASS("mergefunc", MergeFunctionsPass()) MODULE_PASS("metarenamer", MetaRenamerPass()) diff --git a/llvm/lib/Support/Windows/Signals.inc b/llvm/lib/Support/Windows/Signals.inc index dad0fa3..648d6a5 100644 --- a/llvm/lib/Support/Windows/Signals.inc +++ b/llvm/lib/Support/Windows/Signals.inc @@ -354,8 +354,8 @@ namespace llvm { /// Emulates hitting "retry" from an "abort, retry, ignore" CRT debug report /// dialog. "retry" raises an exception which ultimately triggers our stack /// dumper. -static LLVM_ATTRIBUTE_UNUSED int -AvoidMessageBoxHook(int ReportType, char *Message, int *Return) { +[[maybe_unused]] static int AvoidMessageBoxHook(int ReportType, char *Message, + int *Return) { // Set *Return to the retry code for the return value of _CrtDbgReport: // http://msdn.microsoft.com/en-us/library/8hyw4sy7(v=vs.71).aspx // This may also trigger just-in-time debugging via DebugBreak(). diff --git a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp index 91c1f59..662d84b 100644 --- a/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp +++ b/llvm/lib/Target/AArch64/AArch64ISelLowering.cpp @@ -18176,8 +18176,8 @@ bool AArch64TargetLowering::lowerInterleavedStore(Instruction *Store, bool AArch64TargetLowering::lowerDeinterleaveIntrinsicToLoad( Instruction *Load, Value *Mask, IntrinsicInst *DI) const { const unsigned Factor = getDeinterleaveIntrinsicFactor(DI->getIntrinsicID()); - if (Factor != 2 && Factor != 4) { - LLVM_DEBUG(dbgs() << "Matching ld2 and ld4 patterns failed\n"); + if (Factor != 2 && Factor != 3 && Factor != 4) { + LLVM_DEBUG(dbgs() << "Matching ld2, ld3 and ld4 patterns failed\n"); return false; } auto *LI = dyn_cast<LoadInst>(Load); @@ -18255,8 +18255,8 @@ bool AArch64TargetLowering::lowerInterleaveIntrinsicToStore( Instruction *Store, Value *Mask, ArrayRef<Value *> InterleavedValues) const { unsigned Factor = InterleavedValues.size(); - if (Factor != 2 && Factor != 4) { - LLVM_DEBUG(dbgs() << "Matching st2 and st4 patterns failed\n"); + if (Factor != 2 && Factor != 3 && Factor != 4) { + LLVM_DEBUG(dbgs() << "Matching st2, st3 and st4 patterns failed\n"); return false; } StoreInst *SI = dyn_cast<StoreInst>(Store); diff --git a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp index 30dfcf2b..12c600f 100644 --- a/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp +++ b/llvm/lib/Target/AArch64/AArch64InstrInfo.cpp @@ -10600,6 +10600,9 @@ describeORRLoadedValue(const MachineInstr &MI, Register DescribedReg, Register DestReg = DestSrc->Destination->getReg(); Register SrcReg = DestSrc->Source->getReg(); + if (!DestReg.isValid() || !SrcReg.isValid()) + return std::nullopt; + auto Expr = DIExpression::get(MI.getMF()->getFunction().getContext(), {}); // If the described register is the destination, just return the source. diff --git a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp index bfe2c80..a67b12a 100644 --- a/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp +++ b/llvm/lib/Target/AMDGPU/AMDGPURegBankLegalizeRules.cpp @@ -901,6 +901,8 @@ RegBankLegalizeRules::RegBankLegalizeRules(const GCNSubtarget &_ST, addRulesForGOpcs({G_ABS}, Standard).Uni(S16, {{Sgpr32Trunc}, {Sgpr32SExt}}); + addRulesForGOpcs({G_READSTEADYCOUNTER}, Standard).Uni(S64, {{Sgpr64}, {}}); + bool hasSALUFloat = ST->hasSALUFloatInsts(); addRulesForGOpcs({G_FADD}, Standard) diff --git a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp index 58482ea..9fbf9e5 100644 --- a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp +++ b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp @@ -69,6 +69,12 @@ static cl::opt<bool> GCNTrackers( cl::desc("Use the AMDGPU specific RPTrackers during scheduling"), cl::init(false)); +static cl::opt<unsigned> PendingQueueLimit( + "amdgpu-scheduler-pending-queue-limit", cl::Hidden, + cl::desc( + "Max (Available+Pending) size to inspect pending queue (0 disables)"), + cl::init(256)); + #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) #define DUMP_MAX_REG_PRESSURE static cl::opt<bool> PrintMaxRPRegUsageBeforeScheduler( @@ -335,17 +341,52 @@ void GCNSchedStrategy::initCandidate(SchedCandidate &Cand, SUnit *SU, } } +static bool shouldCheckPending(SchedBoundary &Zone, + const TargetSchedModel *SchedModel) { + bool HasBufferedModel = + SchedModel->hasInstrSchedModel() && SchedModel->getMicroOpBufferSize(); + unsigned Combined = Zone.Available.size() + Zone.Pending.size(); + return Combined <= PendingQueueLimit && HasBufferedModel; +} + +static SUnit *pickOnlyChoice(SchedBoundary &Zone, + const TargetSchedModel *SchedModel) { + // pickOnlyChoice() releases pending instructions and checks for new hazards. + SUnit *OnlyChoice = Zone.pickOnlyChoice(); + if (!shouldCheckPending(Zone, SchedModel) || Zone.Pending.empty()) + return OnlyChoice; + + return nullptr; +} + +void GCNSchedStrategy::printCandidateDecision(const SchedCandidate &Current, + const SchedCandidate &Preferred) { + LLVM_DEBUG({ + dbgs() << "Prefer:\t\t"; + DAG->dumpNode(*Preferred.SU); + + if (Current.SU) { + dbgs() << "Not:\t"; + DAG->dumpNode(*Current.SU); + } + + dbgs() << "Reason:\t\t"; + traceCandidate(Preferred); + }); +} + // This function is mostly cut and pasted from // GenericScheduler::pickNodeFromQueue() void GCNSchedStrategy::pickNodeFromQueue(SchedBoundary &Zone, const CandPolicy &ZonePolicy, const RegPressureTracker &RPTracker, - SchedCandidate &Cand, + SchedCandidate &Cand, bool &IsPending, bool IsBottomUp) { const SIRegisterInfo *SRI = static_cast<const SIRegisterInfo *>(TRI); ArrayRef<unsigned> Pressure = RPTracker.getRegSetPressureAtPos(); unsigned SGPRPressure = 0; unsigned VGPRPressure = 0; + IsPending = false; if (DAG->isTrackingPressure()) { if (!GCNTrackers) { SGPRPressure = Pressure[AMDGPU::RegisterPressureSets::SReg_32]; @@ -358,8 +399,9 @@ void GCNSchedStrategy::pickNodeFromQueue(SchedBoundary &Zone, VGPRPressure = T->getPressure().getArchVGPRNum(); } } - ReadyQueue &Q = Zone.Available; - for (SUnit *SU : Q) { + LLVM_DEBUG(dbgs() << "Available Q:\n"); + ReadyQueue &AQ = Zone.Available; + for (SUnit *SU : AQ) { SchedCandidate TryCand(ZonePolicy); initCandidate(TryCand, SU, Zone.isTop(), RPTracker, SRI, SGPRPressure, @@ -371,27 +413,55 @@ void GCNSchedStrategy::pickNodeFromQueue(SchedBoundary &Zone, // Initialize resource delta if needed in case future heuristics query it. if (TryCand.ResDelta == SchedResourceDelta()) TryCand.initResourceDelta(Zone.DAG, SchedModel); + LLVM_DEBUG(printCandidateDecision(Cand, TryCand)); Cand.setBest(TryCand); - LLVM_DEBUG(traceCandidate(Cand)); + } else { + printCandidateDecision(TryCand, Cand); + } + } + + if (!shouldCheckPending(Zone, SchedModel)) + return; + + LLVM_DEBUG(dbgs() << "Pending Q:\n"); + ReadyQueue &PQ = Zone.Pending; + for (SUnit *SU : PQ) { + + SchedCandidate TryCand(ZonePolicy); + initCandidate(TryCand, SU, Zone.isTop(), RPTracker, SRI, SGPRPressure, + VGPRPressure, IsBottomUp); + // Pass SchedBoundary only when comparing nodes from the same boundary. + SchedBoundary *ZoneArg = Cand.AtTop == TryCand.AtTop ? &Zone : nullptr; + tryPendingCandidate(Cand, TryCand, ZoneArg); + if (TryCand.Reason != NoCand) { + // Initialize resource delta if needed in case future heuristics query it. + if (TryCand.ResDelta == SchedResourceDelta()) + TryCand.initResourceDelta(Zone.DAG, SchedModel); + LLVM_DEBUG(printCandidateDecision(Cand, TryCand)); + IsPending = true; + Cand.setBest(TryCand); + } else { + printCandidateDecision(TryCand, Cand); } } } // This function is mostly cut and pasted from // GenericScheduler::pickNodeBidirectional() -SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode) { +SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode, + bool &PickedPending) { // Schedule as far as possible in the direction of no choice. This is most // efficient, but also provides the best heuristics for CriticalPSets. - if (SUnit *SU = Bot.pickOnlyChoice()) { + if (SUnit *SU = pickOnlyChoice(Bot, SchedModel)) { IsTopNode = false; return SU; } - if (SUnit *SU = Top.pickOnlyChoice()) { + if (SUnit *SU = pickOnlyChoice(Top, SchedModel)) { IsTopNode = true; return SU; } - // Set the bottom-up policy based on the state of the current bottom zone and - // the instructions outside the zone, including the top zone. + // Set the bottom-up policy based on the state of the current bottom zone + // and the instructions outside the zone, including the top zone. CandPolicy BotPolicy; setPolicy(BotPolicy, /*IsPostRA=*/false, Bot, &Top); // Set the top-down policy based on the state of the current top zone and @@ -399,12 +469,14 @@ SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode) { CandPolicy TopPolicy; setPolicy(TopPolicy, /*IsPostRA=*/false, Top, &Bot); + bool BotPending = false; // See if BotCand is still valid (because we previously scheduled from Top). LLVM_DEBUG(dbgs() << "Picking from Bot:\n"); if (!BotCand.isValid() || BotCand.SU->isScheduled || BotCand.Policy != BotPolicy) { BotCand.reset(CandPolicy()); pickNodeFromQueue(Bot, BotPolicy, DAG->getBotRPTracker(), BotCand, + BotPending, /*IsBottomUp=*/true); assert(BotCand.Reason != NoCand && "failed to find the first candidate"); } else { @@ -414,6 +486,7 @@ SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode) { SchedCandidate TCand; TCand.reset(CandPolicy()); pickNodeFromQueue(Bot, BotPolicy, DAG->getBotRPTracker(), TCand, + BotPending, /*IsBottomUp=*/true); assert(TCand.SU == BotCand.SU && "Last pick result should correspond to re-picking right now"); @@ -421,12 +494,14 @@ SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode) { #endif } + bool TopPending = false; // Check if the top Q has a better candidate. LLVM_DEBUG(dbgs() << "Picking from Top:\n"); if (!TopCand.isValid() || TopCand.SU->isScheduled || TopCand.Policy != TopPolicy) { TopCand.reset(CandPolicy()); pickNodeFromQueue(Top, TopPolicy, DAG->getTopRPTracker(), TopCand, + TopPending, /*IsBottomUp=*/false); assert(TopCand.Reason != NoCand && "failed to find the first candidate"); } else { @@ -436,6 +511,7 @@ SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode) { SchedCandidate TCand; TCand.reset(CandPolicy()); pickNodeFromQueue(Top, TopPolicy, DAG->getTopRPTracker(), TCand, + TopPending, /*IsBottomUp=*/false); assert(TCand.SU == TopCand.SU && "Last pick result should correspond to re-picking right now"); @@ -446,12 +522,21 @@ SUnit *GCNSchedStrategy::pickNodeBidirectional(bool &IsTopNode) { // Pick best from BotCand and TopCand. LLVM_DEBUG(dbgs() << "Top Cand: "; traceCandidate(TopCand); dbgs() << "Bot Cand: "; traceCandidate(BotCand);); - SchedCandidate Cand = BotCand; - TopCand.Reason = NoCand; - tryCandidate(Cand, TopCand, nullptr); - if (TopCand.Reason != NoCand) { - Cand.setBest(TopCand); + SchedCandidate Cand = BotPending ? TopCand : BotCand; + SchedCandidate TryCand = BotPending ? BotCand : TopCand; + PickedPending = BotPending && TopPending; + + TryCand.Reason = NoCand; + if (BotPending || TopPending) { + PickedPending |= tryPendingCandidate(Cand, TopCand, nullptr); + } else { + tryCandidate(Cand, TryCand, nullptr); + } + + if (TryCand.Reason != NoCand) { + Cand.setBest(TryCand); } + LLVM_DEBUG(dbgs() << "Picking: "; traceCandidate(Cand);); IsTopNode = Cand.AtTop; @@ -466,35 +551,55 @@ SUnit *GCNSchedStrategy::pickNode(bool &IsTopNode) { Bot.Available.empty() && Bot.Pending.empty() && "ReadyQ garbage"); return nullptr; } + bool PickedPending; SUnit *SU; do { + PickedPending = false; if (RegionPolicy.OnlyTopDown) { - SU = Top.pickOnlyChoice(); + SU = pickOnlyChoice(Top, SchedModel); if (!SU) { CandPolicy NoPolicy; TopCand.reset(NoPolicy); pickNodeFromQueue(Top, NoPolicy, DAG->getTopRPTracker(), TopCand, + PickedPending, /*IsBottomUp=*/false); assert(TopCand.Reason != NoCand && "failed to find a candidate"); SU = TopCand.SU; } IsTopNode = true; } else if (RegionPolicy.OnlyBottomUp) { - SU = Bot.pickOnlyChoice(); + SU = pickOnlyChoice(Bot, SchedModel); if (!SU) { CandPolicy NoPolicy; BotCand.reset(NoPolicy); pickNodeFromQueue(Bot, NoPolicy, DAG->getBotRPTracker(), BotCand, + PickedPending, /*IsBottomUp=*/true); assert(BotCand.Reason != NoCand && "failed to find a candidate"); SU = BotCand.SU; } IsTopNode = false; } else { - SU = pickNodeBidirectional(IsTopNode); + SU = pickNodeBidirectional(IsTopNode, PickedPending); } } while (SU->isScheduled); + if (PickedPending) { + unsigned ReadyCycle = IsTopNode ? SU->TopReadyCycle : SU->BotReadyCycle; + SchedBoundary &Zone = IsTopNode ? Top : Bot; + unsigned CurrentCycle = Zone.getCurrCycle(); + if (ReadyCycle > CurrentCycle) + Zone.bumpCycle(ReadyCycle); + + // FIXME: checkHazard() doesn't give information about which cycle the + // hazard will resolve so just keep bumping the cycle by 1. This could be + // made more efficient if checkHazard() returned more details. + while (Zone.checkHazard(SU)) + Zone.bumpCycle(Zone.getCurrCycle() + 1); + + Zone.releasePending(); + } + if (SU->isTopReady()) Top.removeReady(SU); if (SU->isBottomReady()) @@ -540,6 +645,47 @@ GCNSchedStageID GCNSchedStrategy::getNextStage() const { return *std::next(CurrentStage); } +bool GCNSchedStrategy::tryPendingCandidate(SchedCandidate &Cand, + SchedCandidate &TryCand, + SchedBoundary *Zone) const { + // Initialize the candidate if needed. + if (!Cand.isValid()) { + TryCand.Reason = NodeOrder; + return true; + } + + // Bias PhysReg Defs and copies to their uses and defined respectively. + if (tryGreater(biasPhysReg(TryCand.SU, TryCand.AtTop), + biasPhysReg(Cand.SU, Cand.AtTop), TryCand, Cand, PhysReg)) + return TryCand.Reason != NoCand; + + // Avoid exceeding the target's limit. + if (DAG->isTrackingPressure() && + tryPressure(TryCand.RPDelta.Excess, Cand.RPDelta.Excess, TryCand, Cand, + RegExcess, TRI, DAG->MF)) + return TryCand.Reason != NoCand; + + // Avoid increasing the max critical pressure in the scheduled region. + if (DAG->isTrackingPressure() && + tryPressure(TryCand.RPDelta.CriticalMax, Cand.RPDelta.CriticalMax, + TryCand, Cand, RegCritical, TRI, DAG->MF)) + return TryCand.Reason != NoCand; + + bool SameBoundary = Zone != nullptr; + if (SameBoundary) { + TryCand.initResourceDelta(DAG, SchedModel); + if (tryLess(TryCand.ResDelta.CritResources, Cand.ResDelta.CritResources, + TryCand, Cand, ResourceReduce)) + return TryCand.Reason != NoCand; + if (tryGreater(TryCand.ResDelta.DemandedResources, + Cand.ResDelta.DemandedResources, TryCand, Cand, + ResourceDemand)) + return TryCand.Reason != NoCand; + } + + return false; +} + GCNMaxOccupancySchedStrategy::GCNMaxOccupancySchedStrategy( const MachineSchedContext *C, bool IsLegacyScheduler) : GCNSchedStrategy(C) { diff --git a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h index 8ea4267..975781f 100644 --- a/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h +++ b/llvm/lib/Target/AMDGPU/GCNSchedStrategy.h @@ -44,17 +44,32 @@ raw_ostream &operator<<(raw_ostream &OS, const GCNSchedStageID &StageID); /// heuristics to determine excess/critical pressure sets. class GCNSchedStrategy : public GenericScheduler { protected: - SUnit *pickNodeBidirectional(bool &IsTopNode); + SUnit *pickNodeBidirectional(bool &IsTopNode, bool &PickedPending); void pickNodeFromQueue(SchedBoundary &Zone, const CandPolicy &ZonePolicy, const RegPressureTracker &RPTracker, - SchedCandidate &Cand, bool IsBottomUp); + SchedCandidate &Cand, bool &IsPending, + bool IsBottomUp); void initCandidate(SchedCandidate &Cand, SUnit *SU, bool AtTop, const RegPressureTracker &RPTracker, const SIRegisterInfo *SRI, unsigned SGPRPressure, unsigned VGPRPressure, bool IsBottomUp); + /// Evaluates instructions in the pending queue using a subset of scheduling + /// heuristics. + /// + /// Instructions that cannot be issued due to hardware constraints are placed + /// in the pending queue rather than the available queue, making them normally + /// invisible to scheduling heuristics. However, in certain scenarios (such as + /// avoiding register spilling), it may be beneficial to consider scheduling + /// these not-yet-ready instructions. + bool tryPendingCandidate(SchedCandidate &Cand, SchedCandidate &TryCand, + SchedBoundary *Zone) const; + + void printCandidateDecision(const SchedCandidate &Current, + const SchedCandidate &Preferred); + std::vector<unsigned> Pressure; std::vector<unsigned> MaxPressure; diff --git a/llvm/lib/Target/AMDGPU/SIInstrInfo.h b/llvm/lib/Target/AMDGPU/SIInstrInfo.h index e979eeb..df27ec1 100644 --- a/llvm/lib/Target/AMDGPU/SIInstrInfo.h +++ b/llvm/lib/Target/AMDGPU/SIInstrInfo.h @@ -879,6 +879,11 @@ public: MI.getOpcode() != AMDGPU::V_ACCVGPR_READ_B32_e64; } + bool isMFMA(uint16_t Opcode) const { + return isMAI(Opcode) && Opcode != AMDGPU::V_ACCVGPR_WRITE_B32_e64 && + Opcode != AMDGPU::V_ACCVGPR_READ_B32_e64; + } + static bool isDOT(const MachineInstr &MI) { return MI.getDesc().TSFlags & SIInstrFlags::IsDOT; } @@ -895,6 +900,10 @@ public: return isMFMA(MI) || isWMMA(MI) || isSWMMAC(MI); } + bool isMFMAorWMMA(uint16_t Opcode) const { + return isMFMA(Opcode) || isWMMA(Opcode) || isSWMMAC(Opcode); + } + static bool isSWMMAC(const MachineInstr &MI) { return MI.getDesc().TSFlags & SIInstrFlags::IsSWMMAC; } diff --git a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h index a01a5fd..5e3195b 100644 --- a/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h +++ b/llvm/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h @@ -1697,9 +1697,6 @@ LLVM_READNONE bool isInlinableLiteralFP16(int16_t Literal, bool HasInv2Pi); LLVM_READNONE -bool isInlinableLiteralBF16(int16_t Literal, bool HasInv2Pi); - -LLVM_READNONE bool isInlinableLiteralI16(int32_t Literal, bool HasInv2Pi); LLVM_READNONE diff --git a/llvm/lib/Target/Mips/MipsFastISel.cpp b/llvm/lib/Target/Mips/MipsFastISel.cpp index 1ce8d7e3..df0c8c1 100644 --- a/llvm/lib/Target/Mips/MipsFastISel.cpp +++ b/llvm/lib/Target/Mips/MipsFastISel.cpp @@ -264,9 +264,10 @@ public: } // end anonymous namespace -static bool CC_Mips(unsigned ValNo, MVT ValVT, MVT LocVT, - CCValAssign::LocInfo LocInfo, ISD::ArgFlagsTy ArgFlags, - Type *OrigTy, CCState &State) LLVM_ATTRIBUTE_UNUSED; +[[maybe_unused]] static bool CC_Mips(unsigned ValNo, MVT ValVT, MVT LocVT, + CCValAssign::LocInfo LocInfo, + ISD::ArgFlagsTy ArgFlags, Type *OrigTy, + CCState &State); static bool CC_MipsO32_FP32(unsigned ValNo, MVT ValVT, MVT LocVT, CCValAssign::LocInfo LocInfo, diff --git a/llvm/lib/Target/Mips/MipsISelLowering.cpp b/llvm/lib/Target/Mips/MipsISelLowering.cpp index 7f1ff45..2fd7327 100644 --- a/llvm/lib/Target/Mips/MipsISelLowering.cpp +++ b/llvm/lib/Target/Mips/MipsISelLowering.cpp @@ -3176,9 +3176,10 @@ static bool CC_MipsO32_FP64(unsigned ValNo, MVT ValVT, MVT LocVT, F64Regs); } -static bool CC_MipsO32(unsigned ValNo, MVT ValVT, MVT LocVT, - CCValAssign::LocInfo LocInfo, ISD::ArgFlagsTy ArgFlags, - Type *OrigTy, CCState &State) LLVM_ATTRIBUTE_UNUSED; +[[maybe_unused]] static bool CC_MipsO32(unsigned ValNo, MVT ValVT, MVT LocVT, + CCValAssign::LocInfo LocInfo, + ISD::ArgFlagsTy ArgFlags, Type *OrigTy, + CCState &State); #include "MipsGenCallingConv.inc" diff --git a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp index 272c21f..2f1a7ad 100644 --- a/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp +++ b/llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp @@ -749,7 +749,7 @@ NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM, setTruncStoreAction(VT, MVT::i1, Expand); } - // Disable generations of extload/truncstore for v2i16/v2i8. The generic + // Disable generations of extload/truncstore for v2i32/v2i16/v2i8. The generic // expansion for these nodes when they are unaligned is incorrect if the // type is a vector. // @@ -757,7 +757,11 @@ NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM, // TargetLowering::expandUnalignedLoad/Store. setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, MVT::v2i16, MVT::v2i8, Expand); + setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, MVT::v2i32, + {MVT::v2i8, MVT::v2i16}, Expand); setTruncStoreAction(MVT::v2i16, MVT::v2i8, Expand); + setTruncStoreAction(MVT::v2i32, MVT::v2i16, Expand); + setTruncStoreAction(MVT::v2i32, MVT::v2i8, Expand); // Register custom handling for illegal type loads/stores. We'll try to custom // lower almost all illegal types and logic in the lowering will discard cases diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp index 7123a2d..eb87558 100644 --- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp +++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp @@ -1672,6 +1672,8 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM, if (Subtarget.useRVVForFixedLengthVectors()) setTargetDAGCombine(ISD::BITCAST); + setMaxDivRemBitWidthSupported(Subtarget.is64Bit() ? 128 : 64); + // Disable strict node mutation. IsStrictFPEnabled = true; EnableExtLdPromotion = true; diff --git a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td index eedfdb3..ed54404d 100644 --- a/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td +++ b/llvm/lib/Target/WebAssembly/WebAssemblyInstrSIMD.td @@ -1763,6 +1763,26 @@ defm RELAXED_DOT : "i16x8.relaxed_dot_i8x16_i7x16_s\t$dst, $lhs, $rhs", "i16x8.relaxed_dot_i8x16_i7x16_s", 0x112>; +def : Pat< + (v8i16 (add + (wasm_shuffle + (v8i16 (extmul_low_s v16i8:$lhs, v16i8:$rhs)), + (v8i16 (extmul_high_s v16i8:$lhs, v16i8:$rhs)), + (i32 0), (i32 1), (i32 4), (i32 5), + (i32 8), (i32 9), (i32 12), (i32 13), + (i32 16), (i32 17), (i32 20), (i32 21), + (i32 24), (i32 25), (i32 28), (i32 29)), + (wasm_shuffle + (v8i16 (extmul_low_s v16i8:$lhs, v16i8:$rhs)), + (v8i16 (extmul_high_s v16i8:$lhs, v16i8:$rhs)), + (i32 2), (i32 3), (i32 6), (i32 7), + (i32 10), (i32 11), (i32 14), (i32 15), + (i32 18), (i32 19), (i32 22), (i32 23), + (i32 26), (i32 27), (i32 30), (i32 31))) + ), + (v8i16 (RELAXED_DOT v16i8:$lhs, v16i8:$rhs)) +>; + defm RELAXED_DOT_ADD : RELAXED_I<(outs V128:$dst), (ins V128:$lhs, V128:$rhs, V128:$acc), (outs), (ins), @@ -1771,6 +1791,14 @@ defm RELAXED_DOT_ADD : "i32x4.relaxed_dot_i8x16_i7x16_add_s\t$dst, $lhs, $rhs, $acc", "i32x4.relaxed_dot_i8x16_i7x16_add_s", 0x113>; +def : Pat< + (v4i32 (add + (v4i32 (int_wasm_extadd_pairwise_signed + (v8i16 (int_wasm_relaxed_dot_i8x16_i7x16_signed v16i8:$lhs, v16i8:$rhs)))), + (v4i32 V128:$acc))), + (v4i32 (RELAXED_DOT_ADD v16i8:$lhs, v16i8:$rhs, (v4i32 V128:$acc))) + >; + def : Pat<(v4i32 (partial_reduce_smla (v4i32 V128:$acc), (v16i8 V128:$lhs), (v16i8 V128:$rhs))), (RELAXED_DOT_ADD $lhs, $rhs, $acc)>, Requires<[HasRelaxedSIMD]>; diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp index b81641f..28fa2cd 100644 --- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp +++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp @@ -414,8 +414,6 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI, getActionDefinitionsBuilder(G_SEXT_INREG).lower(); - getActionDefinitionsBuilder(G_IS_FPCLASS).lower(); - // fp constants getActionDefinitionsBuilder(G_FCONSTANT) .legalFor({s32, s64}) diff --git a/llvm/lib/Target/X86/X86.td b/llvm/lib/Target/X86/X86.td index 6db780f..8e08d16 100644 --- a/llvm/lib/Target/X86/X86.td +++ b/llvm/lib/Target/X86/X86.td @@ -1338,6 +1338,10 @@ def ProcessorFeatures { list<SubtargetFeature> PTLFeatures = !listremove(ARLSFeatures, [FeatureWIDEKL]); + // Novalake + list<SubtargetFeature> NVLFeatures = + !listconcat(PTLFeatures, [FeaturePREFETCHI]); + // Clearwaterforest list<SubtargetFeature> CWFAdditionalFeatures = [FeaturePREFETCHI, FeatureAVXVNNIINT16, @@ -1883,6 +1887,9 @@ foreach P = ["pantherlake", "wildcatlake"] in { def : ProcModel<P, AlderlakePModel, ProcessorFeatures.PTLFeatures, ProcessorFeatures.ADLTuning>; } +def : ProcModel<"novalake", AlderlakePModel, ProcessorFeatures.NVLFeatures, + ProcessorFeatures.ADLTuning>; + def : ProcModel<"clearwaterforest", AlderlakePModel, ProcessorFeatures.CWFFeatures, ProcessorFeatures.ADLTuning>; def : ProcModel<"emeraldrapids", SapphireRapidsModel, diff --git a/llvm/lib/TargetParser/Host.cpp b/llvm/lib/TargetParser/Host.cpp index 3479106..6065575 100644 --- a/llvm/lib/TargetParser/Host.cpp +++ b/llvm/lib/TargetParser/Host.cpp @@ -1152,6 +1152,20 @@ static StringRef getIntelProcessorTypeAndSubtype(unsigned Family, break; } break; + case 0x12: + switch (Model) { + // Novalake: + case 0x1: + case 0x3: + CPU = "novalake"; + *Type = X86::INTEL_COREI7; + *Subtype = X86::INTEL_COREI7_NOVALAKE; + break; + default: // Unknown family 0x12 CPU. + break; + } + break; + default: break; // Unknown. } diff --git a/llvm/lib/TargetParser/RISCVISAInfo.cpp b/llvm/lib/TargetParser/RISCVISAInfo.cpp index 9268df2..31126cc 100644 --- a/llvm/lib/TargetParser/RISCVISAInfo.cpp +++ b/llvm/lib/TargetParser/RISCVISAInfo.cpp @@ -887,7 +887,7 @@ void RISCVISAInfo::updateImplication() { } static constexpr StringLiteral CombineIntoExts[] = { - {"b"}, {"zk"}, {"zkn"}, {"zks"}, {"zvkn"}, + {"a"}, {"b"}, {"zk"}, {"zkn"}, {"zks"}, {"zvkn"}, {"zvknc"}, {"zvkng"}, {"zvks"}, {"zvksc"}, {"zvksg"}, }; diff --git a/llvm/lib/TargetParser/Unix/Host.inc b/llvm/lib/TargetParser/Unix/Host.inc index aeb2f59..38b942d 100644 --- a/llvm/lib/TargetParser/Unix/Host.inc +++ b/llvm/lib/TargetParser/Unix/Host.inc @@ -59,10 +59,30 @@ static std::string updateTripleOSVersion(std::string TargetTripleString) { if (TT.getOS() == Triple::AIX && !TT.getOSMajorVersion()) { struct utsname name; if (uname(&name) != -1) { + std::string release = name.release; + + if (strcmp(name.sysname, "OS400") == 0) { + /* + PASE uses different versioning system than AIX. + The following table shows the currently supported PASE + releases and the corresponding AIX release: + -------------------------- + PASE | AIX + -------------------------- + V7R4 | 7.2 (TL2) + -------------------------- + V7R5 | 7.2 (TL5) + -------------------------- + V7R6 | 7.3 (TL1) + -------------------------- + */ + release = (release == "4" || release == "5") ? "2" : "3"; + } + std::string NewOSName = std::string(Triple::getOSTypeName(Triple::AIX)); NewOSName += name.version; NewOSName += '.'; - NewOSName += name.release; + NewOSName += release; NewOSName += ".0.0"; TT.setOSName(NewOSName); return TT.str(); diff --git a/llvm/lib/TargetParser/X86TargetParser.cpp b/llvm/lib/TargetParser/X86TargetParser.cpp index e382cfe..dd13ce3 100644 --- a/llvm/lib/TargetParser/X86TargetParser.cpp +++ b/llvm/lib/TargetParser/X86TargetParser.cpp @@ -176,6 +176,8 @@ constexpr FeatureBitset FeaturesArrowlakeS = FeatureSM4; constexpr FeatureBitset FeaturesPantherlake = (FeaturesArrowlakeS ^ FeatureWIDEKL); +constexpr FeatureBitset FeaturesNovalake = + FeaturesPantherlake | FeaturePREFETCHI; constexpr FeatureBitset FeaturesClearwaterforest = (FeaturesSierraforest ^ FeatureWIDEKL) | FeatureAVXVNNIINT16 | FeatureSHA512 | FeatureSM3 | FeatureSM4 | FeaturePREFETCHI | FeatureUSERMSR; @@ -379,6 +381,8 @@ constexpr ProcInfo Processors[] = { // Pantherlake microarchitecture based processors. { {"pantherlake"}, CK_Lunarlake, FEATURE_AVX2, FeaturesPantherlake, 'p', false }, { {"wildcatlake"}, CK_Lunarlake, FEATURE_AVX2, FeaturesPantherlake, 'p', false }, + // Novalake microarchitecture based processors. + { {"novalake"}, CK_Novalake, FEATURE_AVX2, FeaturesNovalake, 'r', false }, // Sierraforest microarchitecture based processors. { {"sierraforest"}, CK_Sierraforest, FEATURE_AVX2, FeaturesSierraforest, 'p', false }, // Grandridge microarchitecture based processors. diff --git a/llvm/lib/Transforms/Coroutines/CoroAnnotationElide.cpp b/llvm/lib/Transforms/Coroutines/CoroAnnotationElide.cpp index f166fef..cf7e450 100644 --- a/llvm/lib/Transforms/Coroutines/CoroAnnotationElide.cpp +++ b/llvm/lib/Transforms/Coroutines/CoroAnnotationElide.cpp @@ -153,26 +153,23 @@ PreservedAnalyses CoroAnnotationElidePass::run(LazyCallGraph::SCC &C, bool IsCallerPresplitCoroutine = Caller->isPresplitCoroutine(); bool HasAttr = CB->hasFnAttr(llvm::Attribute::CoroElideSafe); if (IsCallerPresplitCoroutine && HasAttr) { - BranchProbability MinBranchProbability( - static_cast<int>(CoroElideBranchRatio * MinBlockCounterExecution), - MinBlockCounterExecution); - auto &BFI = FAM.getResult<BlockFrequencyAnalysis>(*Caller); - auto Prob = BranchProbability::getBranchProbability( - BFI.getBlockFreq(CB->getParent()).getFrequency(), - BFI.getEntryFreq().getFrequency()); + auto BlockFreq = BFI.getBlockFreq(CB->getParent()).getFrequency(); + auto EntryFreq = BFI.getEntryFreq().getFrequency(); + uint64_t MinFreq = + static_cast<uint64_t>(EntryFreq * CoroElideBranchRatio); - if (Prob < MinBranchProbability) { + if (BlockFreq < MinFreq) { ORE.emit([&]() { return OptimizationRemarkMissed( DEBUG_TYPE, "CoroAnnotationElideUnlikely", Caller) << "'" << ore::NV("callee", Callee->getName()) << "' not elided in '" << ore::NV("caller", Caller->getName()) - << "' because of low probability: " - << ore::NV("probability", Prob) << " (threshold: " - << ore::NV("threshold", MinBranchProbability) << ")"; + << "' because of low frequency: " + << ore::NV("block_freq", BlockFreq) + << " (threshold: " << ore::NV("min_freq", MinFreq) << ")"; }); continue; } @@ -188,7 +185,8 @@ PreservedAnalyses CoroAnnotationElidePass::run(LazyCallGraph::SCC &C, return OptimizationRemark(DEBUG_TYPE, "CoroAnnotationElide", Caller) << "'" << ore::NV("callee", Callee->getName()) << "' elided in '" << ore::NV("caller", Caller->getName()) - << "' (probability: " << ore::NV("probability", Prob) << ")"; + << "' (block_freq: " << ore::NV("block_freq", BlockFreq) + << ")"; }); FAM.invalidate(*Caller, PreservedAnalyses::none()); diff --git a/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp b/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp index 5066a99..894d83f 100644 --- a/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp +++ b/llvm/lib/Transforms/IPO/MemProfContextDisambiguation.cpp @@ -6150,3 +6150,42 @@ void MemProfContextDisambiguation::run( IndexCallsiteContextGraph CCG(Index, isPrevailing); CCG.process(); } + +// Strips MemProf attributes and metadata. Can be invoked by the pass pipeline +// when we don't have an index that has recorded that we are linking with +// allocation libraries containing the necessary APIs for downstream +// transformations. +PreservedAnalyses MemProfRemoveInfo::run(Module &M, ModuleAnalysisManager &AM) { + // The profile matcher applies hotness attributes directly for allocations, + // and those will cause us to generate calls to the hot/cold interfaces + // unconditionally. If supports-hot-cold-new was not enabled in the LTO + // link then assume we don't want these calls (e.g. not linking with + // the appropriate library, or otherwise trying to disable this behavior). + bool Changed = false; + for (auto &F : M) { + for (auto &BB : F) { + for (auto &I : BB) { + auto *CI = dyn_cast<CallBase>(&I); + if (!CI) + continue; + if (CI->hasFnAttr("memprof")) { + CI->removeFnAttr("memprof"); + Changed = true; + } + if (!CI->hasMetadata(LLVMContext::MD_callsite)) { + assert(!CI->hasMetadata(LLVMContext::MD_memprof)); + continue; + } + // Strip off all memprof metadata as it is no longer needed. + // Importantly, this avoids the addition of new memprof attributes + // after inlining propagation. + CI->setMetadata(LLVMContext::MD_memprof, nullptr); + CI->setMetadata(LLVMContext::MD_callsite, nullptr); + Changed = true; + } + } + } + if (!Changed) + return PreservedAnalyses::all(); + return PreservedAnalyses::none(); +} diff --git a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp index 511bca4..6e17801 100644 --- a/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp +++ b/llvm/lib/Transforms/Instrumentation/AddressSanitizer.cpp @@ -605,17 +605,16 @@ static ShadowMapping getShadowMapping(const Triple &TargetTriple, int LongSize, return Mapping; } -namespace llvm { -void getAddressSanitizerParams(const Triple &TargetTriple, int LongSize, - bool IsKasan, uint64_t *ShadowBase, - int *MappingScale, bool *OrShadowOffset) { +void llvm::getAddressSanitizerParams(const Triple &TargetTriple, int LongSize, + bool IsKasan, uint64_t *ShadowBase, + int *MappingScale, bool *OrShadowOffset) { auto Mapping = getShadowMapping(TargetTriple, LongSize, IsKasan); *ShadowBase = Mapping.Offset; *MappingScale = Mapping.Scale; *OrShadowOffset = Mapping.OrShadowOffset; } -void removeASanIncompatibleFnAttributes(Function &F, bool ReadsArgMem) { +void llvm::removeASanIncompatibleFnAttributes(Function &F, bool ReadsArgMem) { // Sanitizer checks read from shadow, which invalidates memory(argmem: *). // // This is not only true for sanitized functions, because AttrInfer can @@ -668,8 +667,6 @@ ASanAccessInfo::ASanAccessInfo(bool IsWrite, bool CompileKernel, AccessSizeIndex(AccessSizeIndex), IsWrite(IsWrite), CompileKernel(CompileKernel) {} -} // namespace llvm - static uint64_t getRedzoneSizeForScale(int MappingScale) { // Redzone used for stack and globals is at least 32 bytes. // For scales 6 and 7, the redzone has to be 64 and 128 bytes respectively. @@ -677,11 +674,10 @@ static uint64_t getRedzoneSizeForScale(int MappingScale) { } static uint64_t GetCtorAndDtorPriority(Triple &TargetTriple) { - if (TargetTriple.isOSEmscripten()) { + if (TargetTriple.isOSEmscripten()) return kAsanEmscriptenCtorAndDtorPriority; - } else { + else return kAsanCtorAndDtorPriority; - } } static Twine genName(StringRef suffix) { diff --git a/llvm/lib/Transforms/Instrumentation/ControlHeightReduction.cpp b/llvm/lib/Transforms/Instrumentation/ControlHeightReduction.cpp index 444b390..72e8e50 100644 --- a/llvm/lib/Transforms/Instrumentation/ControlHeightReduction.cpp +++ b/llvm/lib/Transforms/Instrumentation/ControlHeightReduction.cpp @@ -2092,8 +2092,6 @@ bool CHR::run() { return Changed; } -namespace llvm { - ControlHeightReductionPass::ControlHeightReductionPass() { parseCHRFilterFiles(); } @@ -2116,5 +2114,3 @@ PreservedAnalyses ControlHeightReductionPass::run( return PreservedAnalyses::all(); return PreservedAnalyses::none(); } - -} // namespace llvm diff --git a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp index c327311..7ebcc21 100644 --- a/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp +++ b/llvm/lib/Transforms/Scalar/IndVarSimplify.cpp @@ -53,6 +53,7 @@ #include "llvm/IR/InstrTypes.h" #include "llvm/IR/Instruction.h" #include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/Intrinsics.h" #include "llvm/IR/PassManager.h" #include "llvm/IR/PatternMatch.h" @@ -117,6 +118,10 @@ static cl::opt<bool> LoopPredication("indvars-predicate-loops", cl::Hidden, cl::init(true), cl::desc("Predicate conditions in read only loops")); +static cl::opt<bool> LoopPredicationTraps( + "indvars-predicate-loop-traps", cl::Hidden, cl::init(true), + cl::desc("Predicate conditions that trap in loops with only local writes")); + static cl::opt<bool> AllowIVWidening("indvars-widen-indvars", cl::Hidden, cl::init(true), cl::desc("Allow widening of indvars to eliminate s/zext")); @@ -1704,6 +1709,24 @@ bool IndVarSimplify::optimizeLoopExits(Loop *L, SCEVExpander &Rewriter) { return Changed; } +static bool crashingBBWithoutEffect(const BasicBlock &BB) { + return llvm::all_of(BB, [](const Instruction &I) { + // TODO: for now this is overly restrictive, to make sure nothing in this + // BB can depend on the loop body. + // It's not enough to check for !I.mayHaveSideEffects(), because e.g. a + // load does not have a side effect, but we could have + // %a = load ptr, ptr %ptr + // %b = load i32, ptr %a + // Now if the loop stored a non-nullptr to %a, we could cause a nullptr + // dereference by skipping over loop iterations. + if (const auto *CB = dyn_cast<CallBase>(&I)) { + if (CB->onlyAccessesInaccessibleMemory()) + return true; + } + return isa<UnreachableInst>(I); + }); +} + bool IndVarSimplify::predicateLoopExits(Loop *L, SCEVExpander &Rewriter) { SmallVector<BasicBlock*, 16> ExitingBlocks; L->getExitingBlocks(ExitingBlocks); @@ -1816,11 +1839,25 @@ bool IndVarSimplify::predicateLoopExits(Loop *L, SCEVExpander &Rewriter) { // suggestions on how to improve this? I can obviously bail out for outer // loops, but that seems less than ideal. MemorySSA can find memory writes, // is that enough for *all* side effects? + bool HasThreadLocalSideEffects = false; for (BasicBlock *BB : L->blocks()) for (auto &I : *BB) // TODO:isGuaranteedToTransfer - if (I.mayHaveSideEffects()) - return false; + if (I.mayHaveSideEffects()) { + if (!LoopPredicationTraps) + return false; + HasThreadLocalSideEffects = true; + if (StoreInst *SI = dyn_cast<StoreInst>(&I)) { + // Simple stores cannot be observed by other threads. + // If HasThreadLocalSideEffects is set, we check + // crashingBBWithoutEffect to make sure that the crashing BB cannot + // observe them either. + if (!SI->isSimple()) + return false; + } else { + return false; + } + } bool Changed = false; // Finally, do the actual predication for all predicatable blocks. A couple @@ -1840,6 +1877,19 @@ bool IndVarSimplify::predicateLoopExits(Loop *L, SCEVExpander &Rewriter) { const SCEV *ExitCount = SE->getExitCount(L, ExitingBB); auto *BI = cast<BranchInst>(ExitingBB->getTerminator()); + if (HasThreadLocalSideEffects) { + const BasicBlock *Unreachable = nullptr; + for (const BasicBlock *Succ : BI->successors()) { + if (isa<UnreachableInst>(Succ->getTerminator())) + Unreachable = Succ; + } + // Exit BB which have one branch back into the loop and another one to + // a trap can still be optimized, because local side effects cannot + // be observed in the exit case (the trap). We could be smarter about + // this, but for now lets pattern match common cases that directly trap. + if (Unreachable == nullptr || !crashingBBWithoutEffect(*Unreachable)) + return Changed; + } Value *NewCond; if (ExitCount == ExactBTC) { NewCond = L->contains(BI->getSuccessor(0)) ? diff --git a/llvm/lib/Transforms/Scalar/LoopFuse.cpp b/llvm/lib/Transforms/Scalar/LoopFuse.cpp index 20733032..19eccb9 100644 --- a/llvm/lib/Transforms/Scalar/LoopFuse.cpp +++ b/llvm/lib/Transforms/Scalar/LoopFuse.cpp @@ -368,7 +368,7 @@ private: Valid = false; } - bool reportInvalidCandidate(llvm::Statistic &Stat) const { + bool reportInvalidCandidate(Statistic &Stat) const { using namespace ore; assert(L && Preheader && "Fusion candidate not initialized properly!"); #if LLVM_ENABLE_STATS @@ -445,6 +445,7 @@ struct FusionCandidateCompare { "No dominance relationship between these fusion candidates!"); } }; +} // namespace using LoopVector = SmallVector<Loop *, 4>; @@ -461,9 +462,15 @@ using LoopVector = SmallVector<Loop *, 4>; using FusionCandidateSet = std::set<FusionCandidate, FusionCandidateCompare>; using FusionCandidateCollection = SmallVector<FusionCandidateSet, 4>; -#if !defined(NDEBUG) -static llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, - const FusionCandidate &FC) { +#ifndef NDEBUG +static void printLoopVector(const LoopVector &LV) { + dbgs() << "****************************\n"; + for (const Loop *L : LV) + printLoop(*L, dbgs()); + dbgs() << "****************************\n"; +} + +static raw_ostream &operator<<(raw_ostream &OS, const FusionCandidate &FC) { if (FC.isValid()) OS << FC.Preheader->getName(); else @@ -472,8 +479,8 @@ static llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, return OS; } -static llvm::raw_ostream &operator<<(llvm::raw_ostream &OS, - const FusionCandidateSet &CandSet) { +static raw_ostream &operator<<(raw_ostream &OS, + const FusionCandidateSet &CandSet) { for (const FusionCandidate &FC : CandSet) OS << FC << '\n'; @@ -489,7 +496,9 @@ printFusionCandidates(const FusionCandidateCollection &FusionCandidates) { dbgs() << "****************************\n"; } } -#endif +#endif // NDEBUG + +namespace { /// Collect all loops in function at the same nest level, starting at the /// outermost level. @@ -550,15 +559,6 @@ private: LoopsOnLevelTy LoopsOnLevel; }; -#ifndef NDEBUG -static void printLoopVector(const LoopVector &LV) { - dbgs() << "****************************\n"; - for (auto *L : LV) - printLoop(*L, dbgs()); - dbgs() << "****************************\n"; -} -#endif - struct LoopFuser { private: // Sets of control flow equivalent fusion candidates for a given nest level. @@ -1850,7 +1850,7 @@ private: /// <Cand1 Preheader> and <Cand2 Preheader>: <Stat Description> template <typename RemarkKind> void reportLoopFusion(const FusionCandidate &FC0, const FusionCandidate &FC1, - llvm::Statistic &Stat) { + Statistic &Stat) { assert(FC0.Preheader && FC1.Preheader && "Expecting valid fusion candidates"); using namespace ore; diff --git a/llvm/lib/Transforms/Scalar/LoopPassManager.cpp b/llvm/lib/Transforms/Scalar/LoopPassManager.cpp index 32078b1..7da8586 100644 --- a/llvm/lib/Transforms/Scalar/LoopPassManager.cpp +++ b/llvm/lib/Transforms/Scalar/LoopPassManager.cpp @@ -16,8 +16,6 @@ using namespace llvm; -namespace llvm { - /// Explicitly specialize the pass manager's run method to handle loop nest /// structure updates. PreservedAnalyses @@ -185,7 +183,6 @@ LoopPassManager::runWithoutLoopNestPasses(Loop &L, LoopAnalysisManager &AM, } return PA; } -} // namespace llvm void FunctionToLoopPassAdaptor::printPipeline( raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) { @@ -193,6 +190,7 @@ void FunctionToLoopPassAdaptor::printPipeline( Pass->printPipeline(OS, MapClassName2PassName); OS << ')'; } + PreservedAnalyses FunctionToLoopPassAdaptor::run(Function &F, FunctionAnalysisManager &AM) { // Before we even compute any loop analyses, first run a miniature function diff --git a/llvm/lib/Transforms/Scalar/LoopVersioningLICM.cpp b/llvm/lib/Transforms/Scalar/LoopVersioningLICM.cpp index 448dc2b..f3e6cbf 100644 --- a/llvm/lib/Transforms/Scalar/LoopVersioningLICM.cpp +++ b/llvm/lib/Transforms/Scalar/LoopVersioningLICM.cpp @@ -540,8 +540,6 @@ bool LoopVersioningLICM::run(DominatorTree *DT) { return Changed; } -namespace llvm { - PreservedAnalyses LoopVersioningLICMPass::run(Loop &L, LoopAnalysisManager &AM, LoopStandardAnalysisResults &LAR, LPMUpdater &U) { @@ -556,4 +554,3 @@ PreservedAnalyses LoopVersioningLICMPass::run(Loop &L, LoopAnalysisManager &AM, return PreservedAnalyses::all(); return getLoopPassPreservedAnalyses(); } -} // namespace llvm diff --git a/llvm/lib/Transforms/Scalar/NewGVN.cpp b/llvm/lib/Transforms/Scalar/NewGVN.cpp index 80aa98d..5a8f18a 100644 --- a/llvm/lib/Transforms/Scalar/NewGVN.cpp +++ b/llvm/lib/Transforms/Scalar/NewGVN.cpp @@ -160,9 +160,6 @@ static cl::opt<bool> EnablePhiOfOps("enable-phi-of-ops", cl::init(true), //===----------------------------------------------------------------------===// // Anchor methods. -namespace llvm { -namespace GVNExpression { - Expression::~Expression() = default; BasicExpression::~BasicExpression() = default; CallExpression::~CallExpression() = default; @@ -171,9 +168,6 @@ StoreExpression::~StoreExpression() = default; AggregateValueExpression::~AggregateValueExpression() = default; PHIExpression::~PHIExpression() = default; -} // end namespace GVNExpression -} // end namespace llvm - namespace { // Tarjan's SCC finding algorithm with Nuutila's improvements diff --git a/llvm/lib/Transforms/Scalar/Reassociate.cpp b/llvm/lib/Transforms/Scalar/Reassociate.cpp index ba58b8e..6d7ce36 100644 --- a/llvm/lib/Transforms/Scalar/Reassociate.cpp +++ b/llvm/lib/Transforms/Scalar/Reassociate.cpp @@ -2623,32 +2623,32 @@ PreservedAnalyses ReassociatePass::run(Function &F, FunctionAnalysisManager &) { namespace { - class ReassociateLegacyPass : public FunctionPass { - ReassociatePass Impl; +class ReassociateLegacyPass : public FunctionPass { + ReassociatePass Impl; - public: - static char ID; // Pass identification, replacement for typeid +public: + static char ID; // Pass identification, replacement for typeid - ReassociateLegacyPass() : FunctionPass(ID) { - initializeReassociateLegacyPassPass(*PassRegistry::getPassRegistry()); - } + ReassociateLegacyPass() : FunctionPass(ID) { + initializeReassociateLegacyPassPass(*PassRegistry::getPassRegistry()); + } - bool runOnFunction(Function &F) override { - if (skipFunction(F)) - return false; + bool runOnFunction(Function &F) override { + if (skipFunction(F)) + return false; - FunctionAnalysisManager DummyFAM; - auto PA = Impl.run(F, DummyFAM); - return !PA.areAllPreserved(); - } + FunctionAnalysisManager DummyFAM; + auto PA = Impl.run(F, DummyFAM); + return !PA.areAllPreserved(); + } - void getAnalysisUsage(AnalysisUsage &AU) const override { - AU.setPreservesCFG(); - AU.addPreserved<AAResultsWrapperPass>(); - AU.addPreserved<BasicAAWrapperPass>(); - AU.addPreserved<GlobalsAAWrapperPass>(); - } - }; + void getAnalysisUsage(AnalysisUsage &AU) const override { + AU.setPreservesCFG(); + AU.addPreserved<AAResultsWrapperPass>(); + AU.addPreserved<BasicAAWrapperPass>(); + AU.addPreserved<GlobalsAAWrapperPass>(); + } +}; } // end anonymous namespace diff --git a/llvm/lib/Transforms/Scalar/Scalarizer.cpp b/llvm/lib/Transforms/Scalar/Scalarizer.cpp index aae5d60..25a531c 100644 --- a/llvm/lib/Transforms/Scalar/Scalarizer.cpp +++ b/llvm/lib/Transforms/Scalar/Scalarizer.cpp @@ -50,9 +50,7 @@ using namespace llvm; #define DEBUG_TYPE "scalarizer" -namespace { - -BasicBlock::iterator skipPastPhiNodesAndDbg(BasicBlock::iterator Itr) { +static BasicBlock::iterator skipPastPhiNodesAndDbg(BasicBlock::iterator Itr) { BasicBlock *BB = Itr->getParent(); if (isa<PHINode>(Itr)) Itr = BB->getFirstInsertionPt(); @@ -76,6 +74,8 @@ using ScatterMap = std::map<std::pair<Value *, Type *>, ValueVector>; // along with a pointer to their scattered forms. using GatherList = SmallVector<std::pair<Instruction *, ValueVector *>, 16>; +namespace { + struct VectorSplit { // The type of the vector. FixedVectorType *VecTy = nullptr; @@ -196,6 +196,7 @@ struct VectorLayout { // The size of each (non-remainder) fragment in bytes. uint64_t SplitSize = 0; }; +} // namespace static bool isStructOfMatchingFixedVectors(Type *Ty) { if (!isa<StructType>(Ty)) @@ -268,6 +269,7 @@ static Value *concatenate(IRBuilder<> &Builder, ArrayRef<Value *> Fragments, return Res; } +namespace { class ScalarizerVisitor : public InstVisitor<ScalarizerVisitor, bool> { public: ScalarizerVisitor(DominatorTree *DT, const TargetTransformInfo *TTI, diff --git a/llvm/lib/Transforms/Scalar/SpeculativeExecution.cpp b/llvm/lib/Transforms/Scalar/SpeculativeExecution.cpp index ebcbd2b..fa66a03 100644 --- a/llvm/lib/Transforms/Scalar/SpeculativeExecution.cpp +++ b/llvm/lib/Transforms/Scalar/SpeculativeExecution.cpp @@ -149,8 +149,6 @@ bool SpeculativeExecutionLegacyPass::runOnFunction(Function &F) { return Impl.runImpl(F, TTI); } -namespace llvm { - bool SpeculativeExecutionPass::runImpl(Function &F, TargetTransformInfo *TTI) { if (OnlyIfDivergentTarget && !TTI->hasBranchDivergence(&F)) { LLVM_DEBUG(dbgs() << "Not running SpeculativeExecution because " @@ -328,11 +326,11 @@ bool SpeculativeExecutionPass::considerHoistingFromTo( return true; } -FunctionPass *createSpeculativeExecutionPass() { +FunctionPass *llvm::createSpeculativeExecutionPass() { return new SpeculativeExecutionLegacyPass(); } -FunctionPass *createSpeculativeExecutionIfHasBranchDivergencePass() { +FunctionPass *llvm::createSpeculativeExecutionIfHasBranchDivergencePass() { return new SpeculativeExecutionLegacyPass(/* OnlyIfDivergentTarget = */ true); } @@ -362,4 +360,3 @@ void SpeculativeExecutionPass::printPipeline( OS << "only-if-divergent-target"; OS << '>'; } -} // namespace llvm diff --git a/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp b/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp index 7d01709..e94ad19 100644 --- a/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp +++ b/llvm/lib/Transforms/Scalar/StraightLineStrengthReduce.cpp @@ -716,8 +716,6 @@ bool StraightLineStrengthReduce::runOnFunction(Function &F) { return Ret; } -namespace llvm { - PreservedAnalyses StraightLineStrengthReducePass::run(Function &F, FunctionAnalysisManager &AM) { const DataLayout *DL = &F.getDataLayout(); @@ -735,5 +733,3 @@ StraightLineStrengthReducePass::run(Function &F, FunctionAnalysisManager &AM) { PA.preserve<TargetIRAnalysis>(); return PA; } - -} // namespace llvm diff --git a/llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp b/llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp index 1d83ddc..89d41f3e 100644 --- a/llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp +++ b/llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp @@ -192,7 +192,7 @@ struct AllocaDerivedValueTracker { SmallPtrSet<Instruction *, 32> AllocaUsers; SmallPtrSet<Instruction *, 32> EscapePoints; }; -} +} // namespace static bool markTails(Function &F, OptimizationRemarkEmitter *ORE) { if (F.callsFunctionThatReturnsTwice()) @@ -967,7 +967,7 @@ struct TailCallElim : public FunctionPass { /*BFI=*/nullptr); } }; -} +} // namespace char TailCallElim::ID = 0; INITIALIZE_PASS_BEGIN(TailCallElim, "tailcallelim", "Tail Call Elimination", diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp index 7f5a41c..cae9aee8 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp @@ -2124,6 +2124,11 @@ static void licm(VPlan &Plan) { // Return true if we do not know how to (mechanically) hoist a given recipe // out of a loop region. auto CannotHoistRecipe = [](VPRecipeBase &R) { + // Assumes don't alias anything or throw; as long as they're guaranteed to + // execute, they're safe to hoist. + if (match(&R, m_Intrinsic<Intrinsic::assume>())) + return false; + // TODO: Relax checks in the future, e.g. we could also hoist reads, if // their memory location is not modified in the vector loop. if (R.mayHaveSideEffects() || R.mayReadFromMemory() || R.isPhi()) diff --git a/llvm/test/Analysis/DXILResource/buffer-frombinding.ll b/llvm/test/Analysis/DXILResource/buffer-frombinding.ll index aeeb21e..ab1945d 100644 --- a/llvm/test/Analysis/DXILResource/buffer-frombinding.ll +++ b/llvm/test/Analysis/DXILResource/buffer-frombinding.ll @@ -150,7 +150,7 @@ define void @test_typedbuffer() { ; CHECK: Kind: CBuffer ; CHECK: CBuffer size: 4 - %cb1 = call target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) + %cb1 = call target("dx.CBuffer", <{ [2 x <{ float, target("dx.Padding", 12) }>], float }>) @llvm.dx.resource.handlefrombinding(i32 1, i32 8, i32 1, i32 0, ptr @Constants.str) ; CHECK: Resource [[CB1:[0-9]+]]: ; CHECK: Name: Constants @@ -161,7 +161,7 @@ define void @test_typedbuffer() { ; CHECK: Size: 1 ; CHECK: Class: CBV ; CHECK: Kind: CBuffer - ; CHECK: CBuffer size: 4 + ; CHECK: CBuffer size: 36 ; CHECK-NOT: Resource {{[0-9]+}}: diff --git a/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-fpr.ll b/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmove-fpr.ll index a0f1b71..bb362d2 100644 --- a/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-fpr.ll +++ b/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmove-fpr.ll @@ -4,7 +4,7 @@ ; RUN: llc < %s -mtriple=arm64-apple-macosx -mcpu=apple-m1 -mattr=-zcm-fpr128 | FileCheck %s -check-prefixes=NOZCM-FPR128-ATTR --match-full-lines ; RUN: llc < %s -mtriple=arm64-apple-macosx -mattr=+zcm-fpr128 | FileCheck %s -check-prefixes=ZCM-FPR128-ATTR --match-full-lines -define void @zero_cycle_regmov_FPR64(double %a, double %b, double %c, double %d) { +define void @zero_cycle_regmove_FPR64(double %a, double %b, double %c, double %d) { entry: ; CHECK-LABEL: t: ; NOZCM-FPR128-CPU: fmov d0, d2 @@ -45,7 +45,7 @@ entry: declare float @foo_double(double, double) -define void @zero_cycle_regmov_FPR32(float %a, float %b, float %c, float %d) { +define void @zero_cycle_regmove_FPR32(float %a, float %b, float %c, float %d) { entry: ; CHECK-LABEL: t: ; NOZCM-FPR128-CPU: fmov s0, s2 @@ -86,7 +86,7 @@ entry: declare float @foo_float(float, float) -define void @zero_cycle_regmov_FPR16(half %a, half %b, half %c, half %d) { +define void @zero_cycle_regmove_FPR16(half %a, half %b, half %c, half %d) { entry: ; CHECK-LABEL: t: ; NOZCM-FPR128-CPU: fmov s0, s2 diff --git a/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-gpr.ll b/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmove-gpr.ll index e14e69b..d6d3f15 100644 --- a/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmov-gpr.ll +++ b/llvm/test/CodeGen/AArch64/arm64-zero-cycle-regmove-gpr.ll @@ -4,7 +4,7 @@ ; RUN: llc < %s -mtriple=arm64-apple-macosx -mcpu=apple-m1 -mattr=-zcm-gpr64 | FileCheck %s -check-prefixes=NOTATTR --match-full-lines ; RUN: llc < %s -mtriple=arm64-apple-macosx -mattr=+zcm-gpr64 | FileCheck %s -check-prefixes=ATTR --match-full-lines -define void @zero_cycle_regmov_GPR32(i32 %a, i32 %b, i32 %c, i32 %d) { +define void @zero_cycle_regmove_GPR32(i32 %a, i32 %b, i32 %c, i32 %d) { entry: ; CHECK-LABEL: t: ; NOTCPU-LINUX: mov w0, w2 diff --git a/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll b/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll index 6b09424..eee232a 100644 --- a/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll +++ b/llvm/test/CodeGen/AMDGPU/insert-delay-alu-bug.ll @@ -49,7 +49,6 @@ bb: ret void } -; FIXME: This generates "instid1(/* invalid instid value */)". define amdgpu_kernel void @f2(i32 %arg, i32 %arg1, i32 %arg2, i1 %arg3, i32 %arg4, i1 %arg5, ptr %arg6, i32 %arg7, i32 %arg8, i32 %arg9, i32 %arg10, i1 %arg11) { ; GFX11-LABEL: f2: ; GFX11: ; %bb.0: ; %bb diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.large.mir b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.large.mir index b07dec3..689d147 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.large.mir +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.exp.large.mir @@ -6,1153 +6,1147 @@ define amdgpu_kernel void @largeInterleave() #0 { ret void } ; GCN-LABEL: largeInterleave: ; GCN: ; %bb.0: - ; GCN-NEXT: ; implicit-def: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 - ; GCN-NEXT: ; implicit-def: $vgpr0 - ; GCN-NEXT: ; implicit-def: $vgpr2 - ; GCN-NEXT: ; implicit-def: $vgpr1 - ; GCN-NEXT: ; implicit-def: $vgpr8 - ; GCN-NEXT: ; implicit-def: $vgpr94 - ; GCN-NEXT: ; implicit-def: $vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: ; implicit-def: $vgpr106 - ; GCN-NEXT: ; implicit-def: $vgpr132 - ; GCN-NEXT: ; implicit-def: $vgpr133 - ; GCN-NEXT: ; implicit-def: $vgpr139 - ; GCN-NEXT: ; implicit-def: $vgpr112_vgpr113_vgpr114_vgpr115_vgpr116_vgpr117_vgpr118_vgpr119_vgpr120_vgpr121_vgpr122_vgpr123_vgpr124_vgpr125_vgpr126_vgpr127 - ; GCN-NEXT: ; iglp_opt mask(0x00000002) - ; GCN-NEXT: ; implicit-def: $sgpr0 + ; GCN-NEXT: ; implicit-def: $vgpr16 + ; GCN-NEXT: ; implicit-def: $vgpr25 ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) - ; GCN-NEXT: v_readfirstlane_b32 s7, v0 + ; GCN-NEXT: v_readfirstlane_b32 s17, v16 + ; GCN-NEXT: ; implicit-def: $sgpr0_sgpr1_sgpr2_sgpr3_sgpr4_sgpr5_sgpr6_sgpr7_sgpr8_sgpr9_sgpr10_sgpr11_sgpr12_sgpr13_sgpr14_sgpr15 + ; GCN-NEXT: ; implicit-def: $vgpr17 + ; GCN-NEXT: ; implicit-def: $sgpr15 ; GCN-NEXT: ; implicit-def: $sgpr8_sgpr9_sgpr10_sgpr11 - ; GCN-NEXT: ; kill: killed $sgpr8_sgpr9_sgpr10_sgpr11 - ; GCN-NEXT: ; implicit-def: $sgpr5 - ; GCN-NEXT: s_nop 1 - ; GCN-NEXT: v_lshl_add_u32 v0, s7, 4, v2 - ; GCN-NEXT: v_mul_lo_u32 v0, v0, s6 - ; GCN-NEXT: v_add_lshl_u32 v92, v0, v1, 1 - ; GCN-NEXT: v_add_u32_e32 v93, s0, v92 - ; GCN-NEXT: buffer_load_dwordx4 v[0:3], v92, s[8:11], 0 offen sc0 sc1 + ; GCN-NEXT: s_lshl_b32 s18, s17, 7 + ; GCN-NEXT: ; implicit-def: $vgpr18 + ; GCN-NEXT: v_add_lshl_u32 v230, v18, s18, 1 + ; GCN-NEXT: v_lshl_add_u32 v25, s17, 4, v25 + ; GCN-NEXT: v_mul_lo_u32 v25, v25, s6 + ; GCN-NEXT: v_add_lshl_u32 v226, v25, v17, 1 + ; GCN-NEXT: v_add_u32_e32 v17, s15, v226 + ; GCN-NEXT: buffer_load_dwordx4 v[64:67], v226, s[8:11], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx4 v[4:7], v93, s[8:11], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx4 v[68:71], v17, s[8:11], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: s_lshl_b32 s0, s7, 7 - ; GCN-NEXT: v_add_lshl_u32 v95, v8, s0, 1 - ; GCN-NEXT: v_add_u32_e32 v8, 64, v93 - ; GCN-NEXT: ; kill: killed $vgpr8 + ; GCN-NEXT: v_add_u32_e32 v72, 64, v17 + ; GCN-NEXT: ; implicit-def: $vgpr213 + ; GCN-NEXT: ; implicit-def: $vgpr152_vgpr153_vgpr154_vgpr155 + ; GCN-NEXT: ; implicit-def: $vgpr246 + ; GCN-NEXT: v_add_u32_e32 v188, 0x80, v17 + ; GCN-NEXT: ; implicit-def: $vgpr156_vgpr157_vgpr158_vgpr159 + ; GCN-NEXT: ; implicit-def: $vgpr144_vgpr145_vgpr146_vgpr147 + ; GCN-NEXT: ; implicit-def: $vgpr19 + ; GCN-NEXT: ; implicit-def: $vgpr26 + ; GCN-NEXT: ; implicit-def: $vgpr27 + ; GCN-NEXT: v_add_u32_e32 v227, 0xc0, v17 + ; GCN-NEXT: v_add_u32_e32 v231, v19, v26 + ; GCN-NEXT: v_add_u32_e32 v232, v19, v27 ; GCN-NEXT: ; implicit-def: $sgpr0_sgpr1_sgpr2_sgpr3 - ; GCN-NEXT: ; kill: killed $vgpr92 - ; GCN-NEXT: ; implicit-def: $sgpr6 + ; GCN-NEXT: ; implicit-def: $vgpr28 + ; GCN-NEXT: ; implicit-def: $vgpr29 + ; GCN-NEXT: v_add_u32_e32 v233, v19, v28 + ; GCN-NEXT: v_add_u32_e32 v234, v19, v29 + ; GCN-NEXT: ; implicit-def: $vgpr140_vgpr141_vgpr142_vgpr143 + ; GCN-NEXT: ; implicit-def: $sgpr5 + ; GCN-NEXT: ; implicit-def: $sgpr7 + ; GCN-NEXT: ; implicit-def: $vgpr148_vgpr149_vgpr150_vgpr151 + ; GCN-NEXT: ; implicit-def: $vgpr136_vgpr137_vgpr138_vgpr139 + ; GCN-NEXT: ; implicit-def: $vgpr132_vgpr133_vgpr134_vgpr135 + ; GCN-NEXT: ; implicit-def: $vgpr20 + ; GCN-NEXT: v_add_u32_e32 v18, s17, v20 + ; GCN-NEXT: v_and_b32_e32 v18, 0x1fffffff, v18 + ; GCN-NEXT: ; implicit-def: $sgpr16 + ; GCN-NEXT: v_mul_lo_u32 v18, v18, s16 + ; GCN-NEXT: ; implicit-def: $vgpr21 + ; GCN-NEXT: v_add_lshl_u32 v199, v21, v18, 1 + ; GCN-NEXT: ; implicit-def: $vgpr22 + ; GCN-NEXT: v_lshl_add_u32 v200, v22, 1, v199 + ; GCN-NEXT: ; implicit-def: $vgpr23 + ; GCN-NEXT: v_lshl_add_u32 v201, v23, 1, v200 + ; GCN-NEXT: ; implicit-def: $vgpr24 + ; GCN-NEXT: v_lshl_add_u32 v202, v24, 1, v201 + ; GCN-NEXT: ; implicit-def: $vgpr16 + ; GCN-NEXT: ; implicit-def: $vgpr18 + ; GCN-NEXT: ; implicit-def: $vgpr20 + ; GCN-NEXT: ; implicit-def: $vgpr24 + ; GCN-NEXT: v_add_u32_e32 v247, v19, v24 + ; GCN-NEXT: v_add_u32_e32 v248, v19, v16 + ; GCN-NEXT: v_add_u32_e32 v249, v19, v18 + ; GCN-NEXT: v_add_u32_e32 v250, v19, v20 + ; GCN-NEXT: ; implicit-def: $vgpr128_vgpr129_vgpr130_vgpr131 + ; GCN-NEXT: ; implicit-def: $sgpr14 + ; GCN-NEXT: ; implicit-def: $vgpr196 + ; GCN-NEXT: ; implicit-def: $sgpr12_sgpr13 + ; GCN-NEXT: ; implicit-def: $vgpr211 + ; GCN-NEXT: v_max_f32_e32 v212, v211, v211 + ; GCN-NEXT: ; implicit-def: $vgpr198 + ; GCN-NEXT: ; implicit-def: $vgpr0_vgpr1_vgpr2_vgpr3_vgpr4_vgpr5_vgpr6_vgpr7_vgpr8_vgpr9_vgpr10_vgpr11_vgpr12_vgpr13_vgpr14_vgpr15 + ; GCN-NEXT: ; implicit-def: $vgpr32 + ; GCN-NEXT: ; implicit-def: $vgpr33 + ; GCN-NEXT: ; implicit-def: $vgpr34 + ; GCN-NEXT: v_add_u32_e32 v210, v19, v34 + ; GCN-NEXT: v_add_u32_e32 v206, v19, v33 + ; GCN-NEXT: v_add_u32_e32 v205, v19, v32 + ; GCN-NEXT: ; implicit-def: $vgpr32_vgpr33_vgpr34_vgpr35_vgpr36_vgpr37_vgpr38_vgpr39_vgpr40_vgpr41_vgpr42_vgpr43_vgpr44_vgpr45_vgpr46_vgpr47 + ; GCN-NEXT: ; implicit-def: $vgpr21 + ; GCN-NEXT: ; implicit-def: $vgpr22 + ; GCN-NEXT: ; implicit-def: $vgpr23 + ; GCN-NEXT: ; implicit-def: $vgpr30 + ; GCN-NEXT: ; implicit-def: $vgpr31 + ; GCN-NEXT: v_add_u32_e32 v207, v19, v21 + ; GCN-NEXT: v_add_u32_e32 v208, v19, v22 + ; GCN-NEXT: v_add_u32_e32 v209, v19, v23 + ; GCN-NEXT: v_add_u32_e32 v203, v19, v30 + ; GCN-NEXT: v_add_u32_e32 v204, v19, v31 + ; GCN-NEXT: ; kill: killed $vgpr17 + ; GCN-NEXT: ; implicit-def: $vgpr16_vgpr17_vgpr18_vgpr19_vgpr20_vgpr21_vgpr22_vgpr23_vgpr24_vgpr25_vgpr26_vgpr27_vgpr28_vgpr29_vgpr30_vgpr31 + ; GCN-NEXT: ; implicit-def: $vgpr48_vgpr49_vgpr50_vgpr51_vgpr52_vgpr53_vgpr54_vgpr55_vgpr56_vgpr57_vgpr58_vgpr59_vgpr60_vgpr61_vgpr62_vgpr63 + ; GCN-NEXT: ; implicit-def: $vgpr197 + ; GCN-NEXT: ; iglp_opt mask(0x00000002) ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b128 v95, v[0:3] + ; GCN-NEXT: ds_write_b128 v230, v[64:67] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b128 v95, v[4:7] offset:1024 + ; GCN-NEXT: ds_write_b128 v230, v[68:71] offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_load_dwordx4 v[64:67], v92, s[8:11], 0 offen offset:64 sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx4 v[160:163], v226, s[8:11], 0 offen offset:64 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx4 v[68:71], v8, s[8:11], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx4 v[164:167], v72, s[8:11], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: ds_read_b128 v[72:75], v94 + ; GCN-NEXT: ds_read_b128 v[64:67], v213 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[80:83], v94 offset:512 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[64:65], v[152:153], 0 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[66:67], v[154:155], v[112:127] + ; GCN-NEXT: ds_read_b128 v[64:67], v213 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[84:87], v94 offset:1024 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[64:65], v[152:153], 0 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[66:67], v[154:155], v[96:111] + ; GCN-NEXT: ds_read_b128 v[64:67], v213 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[72:73], v[76:77], 0 - ; GCN-NEXT: ds_read_b128 v[88:91], v94 offset:1536 + ; GCN-NEXT: ds_read_b128 v[168:171], v213 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[74:75], v[78:79], v[48:63] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 + ; GCN-NEXT: ds_read_b128 v[172:175], v246 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[80:81], v[76:77], 0 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[84:85], v[76:77], 0 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[88:89], v[76:77], 0 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[82:83], v[78:79], v[32:47] - ; GCN-NEXT: ds_read_b128 v[80:83], v106 offset:512 + ; GCN-NEXT: ds_read_b128 v[176:179], v246 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[86:87], v[78:79], v[16:31] - ; GCN-NEXT: ds_read_b128 v[84:87], v106 offset:1024 + ; GCN-NEXT: ds_read_b128 v[180:183], v246 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[90:91], v[78:79], v[0:15] - ; GCN-NEXT: ; implicit-def: $vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: ds_read_b128 v[88:91], v106 offset:1536 + ; GCN-NEXT: ds_read_b128 v[184:187], v246 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[64:65], v[152:153], 0 ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b128 v95, v[64:67] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[72:73], v[76:77], v[48:63] - ; GCN-NEXT: v_add_u32_e32 v72, 0x80, v93 + ; GCN-NEXT: ds_write_b128 v230, v[160:163] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[66:67], v[154:155], v[80:95] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b128 v95, v[68:71] offset:1024 + ; GCN-NEXT: ds_write_b128 v230, v[164:167] offset:1024 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[168:169], v[152:153], 0 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[170:171], v[154:155], v[64:79] ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_load_dwordx4 v[64:67], v92, s[8:11], 0 offen offset:128 sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx4 v[152:155], v226, s[8:11], 0 offen offset:128 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx4 v[68:71], v72, s[8:11], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx4 v[160:163], v188, s[8:11], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: ; kill: killed $vgpr72 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[74:75], v[78:79], v[48:63] - ; GCN-NEXT: ds_read_b128 v[72:75], v94 + ; GCN-NEXT: ds_read_b128 v[188:191], v213 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[80:81], v[76:77], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[84:85], v[76:77], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[88:89], v[76:77], v[0:15] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[82:83], v[78:79], v[32:47] - ; GCN-NEXT: ds_read_b128 v[80:83], v94 offset:512 + ; GCN-NEXT: ds_read_b128 v[192:195], v213 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[86:87], v[78:79], v[16:31] - ; GCN-NEXT: ds_read_b128 v[84:87], v94 offset:1024 + ; GCN-NEXT: ds_read_b128 v[164:167], v213 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[90:91], v[78:79], v[0:15] - ; GCN-NEXT: ; implicit-def: $vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: ds_read_b128 v[88:91], v94 offset:1536 + ; GCN-NEXT: ds_read_b128 v[214:217], v213 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[72:73], v[76:77], v[48:63] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[74:75], v[78:79], v[48:63] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[172:173], v[156:157], v[112:127] + ; GCN-NEXT: ds_read_b128 v[218:221], v246 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[80:81], v[76:77], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[84:85], v[76:77], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[88:89], v[76:77], v[0:15] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[82:83], v[78:79], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[86:87], v[78:79], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[90:91], v[78:79], v[0:15] - ; GCN-NEXT: ; implicit-def: $vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[72:73], v[76:77], v[48:63] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[74:75], v[78:79], v[48:63] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:512 + ; GCN-NEXT: ds_read_b128 v[222:225], v246 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[72:73], v[76:77], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[74:75], v[78:79], v[32:47] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:1024 + ; GCN-NEXT: ds_read_b128 v[168:171], v246 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[72:73], v[76:77], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[74:75], v[78:79], v[16:31] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:1536 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[174:175], v[158:159], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[188:189], v[144:145], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[190:191], v[146:147], v[112:127] + ; GCN-NEXT: ds_read_b128 v[188:191], v246 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b128 v95, v[64:67] + ; GCN-NEXT: ds_write_b128 v230, v[152:155] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b128 v95, v[68:71] offset:1024 - ; GCN-NEXT: ; implicit-def: $vgpr64 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[72:73], v[76:77], v[0:15] - ; GCN-NEXT: v_add_u32_e32 v72, 0xc0, v93 - ; GCN-NEXT: ; implicit-def: $vgpr73 - ; GCN-NEXT: v_add_u32_e32 v76, v132, v64 + ; GCN-NEXT: ds_write_b128 v230, v[160:163] offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_load_dwordx4 v[64:67], v92, s[8:11], 0 offen offset:192 sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx4 v[152:155], v226, s[8:11], 0 offen offset:192 sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx4 v[68:71], v72, s[8:11], 0 offen sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[184:185], v[156:157], v[64:79] + ; GCN-NEXT: buffer_load_dwordx4 v[226:229], v227, s[8:11], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ; kill: killed $vgpr72 - ; GCN-NEXT: v_add_u32_e32 v72, v132, v73 - ; GCN-NEXT: buffer_load_dwordx2 v[98:99], v76, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[160:161], v231, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx2 v[102:103], v72, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[162:163], v232, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[74:75], v[78:79], v[0:15] - ; GCN-NEXT: ; implicit-def: $vgpr74 - ; GCN-NEXT: v_add_u32_e32 v72, v132, v74 - ; GCN-NEXT: ; implicit-def: $vgpr75 - ; GCN-NEXT: buffer_load_dwordx2 v[100:101], v72, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[172:173], v233, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_add_u32_e32 v72, v132, v75 - ; GCN-NEXT: buffer_load_dwordx2 v[104:105], v72, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[174:175], v234, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: ds_read_b128 v[72:75], v94 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[186:187], v[158:159], v[64:79] + ; GCN-NEXT: v_perm_b32 v238, v162, v160, s5 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[218:219], v[140:141], v[112:127] + ; GCN-NEXT: v_perm_b32 v240, v162, v160, s7 + ; GCN-NEXT: v_perm_b32 v242, v163, v161, s5 + ; GCN-NEXT: v_perm_b32 v244, v163, v161, s7 + ; GCN-NEXT: ds_read_b128 v[160:163], v213 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ; kill: killed $vgpr76 - ; GCN-NEXT: ; implicit-def: $vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: ; implicit-def: $sgpr8 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[72:73], v[76:77], v[48:63] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[74:75], v[78:79], v[48:63] - ; GCN-NEXT: ds_read_b128 v[72:75], v94 offset:512 + ; GCN-NEXT: v_perm_b32 v239, v174, v172, s5 + ; GCN-NEXT: v_perm_b32 v241, v174, v172, s7 + ; GCN-NEXT: v_perm_b32 v243, v175, v173, s5 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[214:215], v[144:145], v[64:79] + ; GCN-NEXT: v_perm_b32 v245, v175, v173, s7 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[176:177], v[156:157], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[220:221], v[142:143], v[112:127] + ; GCN-NEXT: ds_read_b128 v[218:221], v213 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[72:73], v[76:77], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[74:75], v[78:79], v[32:47] - ; GCN-NEXT: ds_read_b128 v[72:75], v94 offset:1024 + ; GCN-NEXT: ds_read_b128 v[172:175], v213 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[72:73], v[76:77], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[74:75], v[78:79], v[16:31] - ; GCN-NEXT: ds_read_b128 v[72:75], v94 offset:1536 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[216:217], v[146:147], v[64:79] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[178:179], v[158:159], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[160:161], v[148:149], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[188:189], v[140:141], v[64:79] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[192:193], v[144:145], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[162:163], v[150:151], v[112:127] + ; GCN-NEXT: ds_read_b128 v[160:163], v213 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[72:73], v[76:77], v[0:15] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[74:75], v[78:79], v[0:15] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 + ; GCN-NEXT: ds_read_b128 v[184:187], v246 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ; implicit-def: $vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[72:73], v[76:77], v[48:63] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[74:75], v[78:79], v[48:63] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:512 + ; GCN-NEXT: ds_read_b128 v[214:217], v246 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[72:73], v[76:77], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[74:75], v[78:79], v[32:47] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:1024 + ; GCN-NEXT: ds_read_b128 v[176:179], v246 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[72:73], v[76:77], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[74:75], v[78:79], v[16:31] - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:1536 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[190:191], v[142:143], v[64:79] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[194:195], v[146:147], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[160:161], v[148:149], v[64:79] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[180:181], v[156:157], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[184:185], v[136:137], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[222:223], v[140:141], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[162:163], v[150:151], v[64:79] + ; GCN-NEXT: ds_read_b128 v[160:163], v246 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b128 v95, v[64:67] + ; GCN-NEXT: ds_write_b128 v230, v[152:155] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b128 v95, v[68:71] offset:1024 + ; GCN-NEXT: ds_write_b128 v230, v[226:229] offset:1024 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[182:183], v[158:159], v[80:95] ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_read_b128 v[64:67], v94 + ; GCN-NEXT: ds_read_b128 v[156:159], v213 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[90:93], v94 offset:512 + ; GCN-NEXT: ds_read_b128 v[226:229], v213 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[72:73], v[76:77], v[0:15] - ; GCN-NEXT: ; implicit-def: $vgpr68_vgpr69_vgpr70_vgpr71 - ; GCN-NEXT: ds_read_b128 v[84:87], v94 offset:1024 + ; GCN-NEXT: ds_read_b128 v[180:183], v213 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[64:65], v[68:69], v[48:63] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[74:75], v[78:79], v[0:15] - ; GCN-NEXT: ds_read_b128 v[76:79], v94 offset:1536 + ; GCN-NEXT: ds_read_b128 v[152:155], v213 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[94:97], v106 + ; GCN-NEXT: ds_read_b128 v[230:233], v246 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[66:67], v[70:71], v[48:63] - ; GCN-NEXT: ; implicit-def: $vgpr64_vgpr65_vgpr66_vgpr67 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[90:91], v[68:69], v[32:47] - ; GCN-NEXT: ds_read_b128 v[88:91], v106 offset:512 + ; GCN-NEXT: ds_read_b128 v[234:237], v246 offset:512 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[80:83], v106 offset:1024 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[186:187], v[138:139], v[112:127] + ; GCN-NEXT: ds_read_b128 v[184:187], v246 offset:1024 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[72:75], v106 offset:1536 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[224:225], v[142:143], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[156:157], v[132:133], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[218:219], v[148:149], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[158:159], v[134:135], v[112:127] + ; GCN-NEXT: ds_read_b128 v[156:159], v246 offset:1536 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[94:95], v[64:65], v[48:63] - ; GCN-NEXT: v_perm_b32 v94, v102, v98, s5 - ; GCN-NEXT: v_perm_b32 v98, v102, v98, s8 - ; GCN-NEXT: v_perm_b32 v102, v103, v99, s5 - ; GCN-NEXT: v_perm_b32 v95, v104, v100, s5 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[92:93], v[70:71], v[32:47] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[84:85], v[68:69], v[16:31] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[96:97], v[66:67], v[48:63] - ; GCN-NEXT: v_perm_b32 v96, v103, v99, s8 - ; GCN-NEXT: v_perm_b32 v99, v104, v100, s8 - ; GCN-NEXT: v_perm_b32 v103, v105, v101, s5 - ; GCN-NEXT: v_perm_b32 v97, v105, v101, s8 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[88:89], v[64:65], v[32:47] - ; GCN-NEXT: s_nop 5 - ; GCN-NEXT: v_mul_f32_e32 v100, s4, v48 - ; GCN-NEXT: v_mul_f32_e32 v101, s4, v49 - ; GCN-NEXT: v_max3_f32 v92, v100, s6, v101 - ; GCN-NEXT: v_mul_f32_e32 v93, s4, v50 - ; GCN-NEXT: v_mul_f32_e32 v100, s4, v51 - ; GCN-NEXT: v_max3_f32 v92, v92, v93, v100 - ; GCN-NEXT: v_mul_f32_e32 v93, s4, v52 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[86:87], v[70:71], v[16:31] - ; GCN-NEXT: v_mul_f32_e32 v100, s4, v53 - ; GCN-NEXT: v_max3_f32 v92, v92, v93, v100 - ; GCN-NEXT: v_mul_f32_e32 v84, s4, v54 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v55 - ; GCN-NEXT: v_max3_f32 v84, v92, v84, v85 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v56 - ; GCN-NEXT: v_mul_f32_e32 v92, s4, v57 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[76:77], v[68:69], v[0:15] - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v92 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v58 - ; GCN-NEXT: v_mul_f32_e32 v88, s4, v59 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v88 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v60 - ; GCN-NEXT: v_mul_f32_e32 v88, s4, v61 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v88 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[90:91], v[66:67], v[32:47] - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v62 - ; GCN-NEXT: v_mul_f32_e32 v88, s4, v63 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v88 - ; GCN-NEXT: ; implicit-def: $sgpr6 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[80:81], v[64:65], v[16:31] - ; GCN-NEXT: s_nop 6 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v32 - ; GCN-NEXT: v_mul_f32_e32 v88, s4, v33 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v88 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v34 - ; GCN-NEXT: v_mul_f32_e32 v88, s4, v35 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v88 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v36 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[78:79], v[70:71], v[0:15] - ; GCN-NEXT: v_mul_f32_e32 v86, s4, v37 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v86 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v38 - ; GCN-NEXT: v_mul_f32_e32 v86, s4, v39 - ; GCN-NEXT: v_max3_f32 v84, v84, v85, v86 - ; GCN-NEXT: v_mul_f32_e32 v85, s4, v40 - ; GCN-NEXT: v_mul_f32_e32 v80, s4, v41 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[82:83], v[66:67], v[16:31] - ; GCN-NEXT: v_max3_f32 v80, v84, v85, v80 - ; GCN-NEXT: v_mul_f32_e32 v81, s4, v42 - ; GCN-NEXT: v_mul_f32_e32 v84, s4, v43 - ; GCN-NEXT: v_max3_f32 v80, v80, v81, v84 - ; GCN-NEXT: v_mul_f32_e32 v81, s4, v44 - ; GCN-NEXT: v_mul_f32_e32 v84, s4, v45 - ; GCN-NEXT: v_max3_f32 v80, v80, v81, v84 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[72:73], v[64:65], v[0:15] - ; GCN-NEXT: v_mul_f32_e32 v81, s4, v46 - ; GCN-NEXT: v_mul_f32_e32 v82, s4, v47 - ; GCN-NEXT: v_max3_f32 v80, v80, v81, v82 - ; GCN-NEXT: v_mul_f32_e32 v81, s4, v16 - ; GCN-NEXT: v_mul_f32_e32 v82, s4, v17 - ; GCN-NEXT: v_max3_f32 v80, v80, v81, v82 - ; GCN-NEXT: v_mul_f32_e32 v68, s4, v18 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[74:75], v[66:67], v[0:15] - ; GCN-NEXT: v_mul_f32_e32 v69, s4, v19 - ; GCN-NEXT: v_max3_f32 v68, v80, v68, v69 - ; GCN-NEXT: v_mul_f32_e32 v69, s4, v20 - ; GCN-NEXT: v_mul_f32_e32 v76, s4, v21 - ; GCN-NEXT: v_max3_f32 v68, v68, v69, v76 - ; GCN-NEXT: v_mul_f32_e32 v69, s4, v22 - ; GCN-NEXT: v_mul_f32_e32 v70, s4, v23 - ; GCN-NEXT: v_max3_f32 v68, v68, v69, v70 - ; GCN-NEXT: v_mul_f32_e32 v69, s4, v24 - ; GCN-NEXT: v_mul_f32_e32 v70, s4, v25 - ; GCN-NEXT: v_max3_f32 v68, v68, v69, v70 - ; GCN-NEXT: v_mul_f32_e32 v69, s4, v26 - ; GCN-NEXT: v_mul_f32_e32 v70, s4, v27 - ; GCN-NEXT: v_max3_f32 v64, v68, v69, v70 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v28 - ; GCN-NEXT: v_mul_f32_e32 v68, s4, v29 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v68 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v30 - ; GCN-NEXT: v_mul_f32_e32 v68, s4, v31 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v68 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v0 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v1 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v2 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v3 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v4 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v5 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v6 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v7 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v8 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v9 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v10 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v11 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v12 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v13 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: v_mul_f32_e32 v65, s4, v14 - ; GCN-NEXT: v_mul_f32_e32 v66, s4, v15 - ; GCN-NEXT: v_max3_f32 v64, v64, v65, v66 - ; GCN-NEXT: ; implicit-def: $vgpr65 - ; GCN-NEXT: ; implicit-def: $vgpr66 - ; GCN-NEXT: ; implicit-def: $vgpr68 - ; GCN-NEXT: ; implicit-def: $vgpr67 - ; GCN-NEXT: v_add_u32_e32 v65, s7, v65 - ; GCN-NEXT: v_and_b32_e32 v65, 0x1fffffff, v65 - ; GCN-NEXT: v_mul_lo_u32 v65, v65, s6 - ; GCN-NEXT: v_add_lshl_u32 v135, v66, v65, 1 - ; GCN-NEXT: ds_bpermute_b32 v65, v133, v64 - ; GCN-NEXT: ; implicit-def: $vgpr66 - ; GCN-NEXT: v_lshl_add_u32 v136, v66, 1, v135 - ; GCN-NEXT: ; implicit-def: $vgpr66 - ; GCN-NEXT: v_lshl_add_u32 v137, v66, 1, v136 - ; GCN-NEXT: ; implicit-def: $vgpr66 - ; GCN-NEXT: ; implicit-def: $sgpr6_sgpr7 - ; GCN-NEXT: v_lshl_add_u32 v138, v66, 1, v137 ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v135, v[94:95] - ; GCN-NEXT: v_max_f32_e32 v65, v65, v65 - ; GCN-NEXT: v_max_f32_e32 v64, v64, v65 - ; GCN-NEXT: ds_bpermute_b32 v65, v133, v64 + ; GCN-NEXT: ds_write_b64 v199, v[238:239] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v136, v[98:99] + ; GCN-NEXT: ds_write_b64 v200, v[240:241] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v137, v[102:103] + ; GCN-NEXT: ds_write_b64 v201, v[242:243] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v138, v[96:97] - ; GCN-NEXT: v_add_u32_e32 v68, v132, v68 - ; GCN-NEXT: v_cndmask_b32_e64 v64, v65, v64, s[6:7] - ; GCN-NEXT: v_max_f32_e32 v64, v64, v64 - ; GCN-NEXT: ; implicit-def: $vgpr65 - ; GCN-NEXT: v_max_f32_e32 v66, v65, v65 - ; GCN-NEXT: v_max_f32_e32 v134, v66, v64 - ; GCN-NEXT: ; implicit-def: $vgpr64 + ; GCN-NEXT: ds_write_b64 v202, v[244:245] ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_load_dwordx2 v[156:157], v68, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[192:193], v247, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_add_u32_e32 v64, v132, v64 - ; GCN-NEXT: buffer_load_dwordx2 v[158:159], v64, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[220:221], v[150:151], v[96:111] + ; GCN-NEXT: buffer_load_dwordx2 v[194:195], v248, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ; implicit-def: $vgpr66 - ; GCN-NEXT: v_add_u32_e32 v64, v132, v66 - ; GCN-NEXT: buffer_load_dwordx2 v[128:129], v64, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[218:219], v249, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_add_u32_e32 v64, v132, v67 - ; GCN-NEXT: buffer_load_dwordx2 v[130:131], v64, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[220:221], v250, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v57, s4, v57, -v134 - ; GCN-NEXT: v_fma_f32 v48, s4, v48, -v134 - ; GCN-NEXT: v_fma_f32 v96, s4, v58, -v134 - ; GCN-NEXT: v_mul_f32_e32 v57, 0x3fb8aa3b, v57 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v48 - ; GCN-NEXT: v_fma_f32 v64, s4, v49, -v134 - ; GCN-NEXT: v_exp_f32_e32 v163, v57 - ; GCN-NEXT: v_mul_f32_e32 v57, 0x3fb8aa3b, v96 - ; GCN-NEXT: v_fma_f32 v66, s4, v50, -v134 - ; GCN-NEXT: v_exp_f32_e32 v164, v57 - ; GCN-NEXT: v_exp_f32_e32 v49, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v64 - ; GCN-NEXT: v_fma_f32 v67, s4, v51, -v134 - ; GCN-NEXT: v_exp_f32_e32 v50, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v66 - ; GCN-NEXT: v_fma_f32 v68, s4, v52, -v134 - ; GCN-NEXT: v_exp_f32_e32 v51, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v67 - ; GCN-NEXT: v_fma_f32 v69, s4, v53, -v134 - ; GCN-NEXT: v_exp_f32_e32 v52, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v68 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: v_fma_f32 v70, s4, v54, -v134 - ; GCN-NEXT: v_exp_f32_e32 v53, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v69 - ; GCN-NEXT: v_fma_f32 v71, s4, v55, -v134 - ; GCN-NEXT: ds_read_b128 v[140:143], v139 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v54, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v70 - ; GCN-NEXT: v_exp_f32_e32 v55, v48 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v71 - ; GCN-NEXT: ds_read_b128 v[144:147], v139 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v66, s4, v56, -v134 - ; GCN-NEXT: v_exp_f32_e32 v56, v48 - ; GCN-NEXT: v_sub_f32_e32 v48, v65, v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v64, v49 - ; GCN-NEXT: v_cvt_f16_f32_e32 v67, v50 - ; GCN-NEXT: v_cvt_f16_f32_e32 v68, v51 - ; GCN-NEXT: v_cvt_f16_f32_e32 v58, v52 - ; GCN-NEXT: v_mul_f32_e32 v48, 0x3fb8aa3b, v48 - ; GCN-NEXT: ds_read_b128 v[148:151], v139 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v48, v48 - ; GCN-NEXT: v_pack_b32_f16 v161, v68, v58 - ; GCN-NEXT: v_pack_b32_f16 v160, v64, v67 - ; GCN-NEXT: v_mul_f32_e32 v58, 0x3fb8aa3b, v66 - ; GCN-NEXT: ; implicit-def: $vgpr64_vgpr65_vgpr66_vgpr67_vgpr68_vgpr69_vgpr70_vgpr71_vgpr72_vgpr73_vgpr74_vgpr75_vgpr76_vgpr77_vgpr78_vgpr79 - ; GCN-NEXT: ds_read_b128 v[152:155], v139 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v162, s4, v61, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v61, v55 - ; GCN-NEXT: v_cvt_f16_f32_e32 v57, v56 - ; GCN-NEXT: v_pk_mul_f32 v[64:65], v[64:65], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[66:67], v[66:67], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[68:69], v[68:69], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[70:71], v[70:71], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[72:73], v[72:73], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[74:75], v[74:75], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[76:77], v[76:77], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[78:79], v[78:79], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: ; implicit-def: $vgpr80_vgpr81_vgpr82_vgpr83_vgpr84_vgpr85_vgpr86_vgpr87_vgpr88_vgpr89_vgpr90_vgpr91_vgpr92_vgpr93_vgpr94_vgpr95 - ; GCN-NEXT: v_fma_f32 v59, s4, v59, -v134 - ; GCN-NEXT: v_pk_mul_f32 v[80:81], v[80:81], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[140:141], v[160:161], v[64:79] - ; GCN-NEXT: v_mul_f32_e64 v82, v82, v48 - ; GCN-NEXT: v_mul_f32_e64 v83, v83, v48 - ; GCN-NEXT: v_mul_f32_e64 v84, v84, v48 - ; GCN-NEXT: v_mul_f32_e64 v85, v85, v48 - ; GCN-NEXT: v_mul_f32_e64 v86, v86, v48 - ; GCN-NEXT: v_mul_f32_e64 v87, v87, v48 - ; GCN-NEXT: v_pk_mul_f32 v[88:89], v[88:89], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[90:91], v[90:91], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[92:93], v[92:93], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[94:95], v[94:95], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: ; implicit-def: $vgpr96_vgpr97_vgpr98_vgpr99_vgpr100_vgpr101_vgpr102_vgpr103_vgpr104_vgpr105_vgpr106_vgpr107_vgpr108_vgpr109_vgpr110_vgpr111 - ; GCN-NEXT: v_exp_f32_e32 v58, v58 - ; GCN-NEXT: v_pk_mul_f32 v[96:97], v[96:97], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[144:145], v[160:161], v[80:95] - ; GCN-NEXT: v_mul_f32_e64 v98, v98, v48 - ; GCN-NEXT: v_mul_f32_e64 v99, v99, v48 - ; GCN-NEXT: v_mul_f32_e64 v100, v100, v48 - ; GCN-NEXT: v_mul_f32_e64 v101, v101, v48 - ; GCN-NEXT: v_mul_f32_e64 v102, v102, v48 - ; GCN-NEXT: v_mul_f32_e64 v103, v103, v48 - ; GCN-NEXT: v_pk_mul_f32 v[104:105], v[104:105], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[106:107], v[106:107], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[108:109], v[108:109], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[110:111], v[110:111], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pack_b32_f16 v145, v61, v57 - ; GCN-NEXT: v_mul_f32_e32 v57, 0x3fb8aa3b, v59 - ; GCN-NEXT: v_cvt_f16_f32_e32 v140, v53 - ; GCN-NEXT: v_cvt_f16_f32_e32 v141, v54 - ; GCN-NEXT: v_exp_f32_e32 v59, v57 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[148:149], v[160:161], v[96:111] - ; GCN-NEXT: v_fma_f32 v60, s4, v60, -v134 - ; GCN-NEXT: v_mul_f32_e64 v112, v112, v48 - ; GCN-NEXT: v_mul_f32_e64 v113, v113, v48 - ; GCN-NEXT: v_mul_f32_e64 v114, v114, v48 - ; GCN-NEXT: v_mul_f32_e64 v115, v115, v48 - ; GCN-NEXT: v_pk_mul_f32 v[116:117], v[116:117], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[118:119], v[118:119], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[120:121], v[120:121], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[122:123], v[122:123], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[124:125], v[124:125], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_pk_mul_f32 v[126:127], v[126:127], v[48:49] op_sel_hi:[1,0] - ; GCN-NEXT: v_fma_f32 v148, s4, v62, -v134 - ; GCN-NEXT: v_pack_b32_f16 v144, v140, v141 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[152:153], v[160:161], v[112:127] - ; GCN-NEXT: v_fma_f32 v152, s4, v63, -v134 - ; GCN-NEXT: v_mul_f32_e32 v149, 0x3fb8aa3b, v60 - ; GCN-NEXT: ; implicit-def: $vgpr57 - ; GCN-NEXT: ds_read_b128 v[60:63], v57 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v160, v149 - ; GCN-NEXT: v_fma_f32 v161, s4, v33, -v134 - ; GCN-NEXT: v_mul_f32_e32 v33, 0x3fb8aa3b, v148 - ; GCN-NEXT: v_cvt_f16_f32_e32 v153, v58 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[142:143], v[144:145], v[64:79] - ; GCN-NEXT: v_fma_f32 v32, s4, v32, -v134 - ; GCN-NEXT: ds_read_b128 v[140:143], v57 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v40, s4, v40, -v134 - ; GCN-NEXT: v_fma_f32 v44, s4, v44, -v134 - ; GCN-NEXT: v_fma_f32 v16, s4, v16, -v134 - ; GCN-NEXT: v_fma_f32 v166, s4, v20, -v134 - ; GCN-NEXT: v_fma_f32 v24, s4, v24, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[146:147], v[144:145], v[80:95] - ; GCN-NEXT: v_mul_f32_e32 v146, 0x3fb8aa3b, v162 - ; GCN-NEXT: v_cvt_f16_f32_e32 v147, v163 - ; GCN-NEXT: v_exp_f32_e32 v162, v146 - ; GCN-NEXT: v_cvt_f16_f32_e32 v146, v164 - ; GCN-NEXT: v_fma_f32 v28, s4, v28, -v134 - ; GCN-NEXT: v_pack_b32_f16 v148, v153, v147 - ; GCN-NEXT: v_fma_f32 v0, s4, v0, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[150:151], v[144:145], v[96:111] - ; GCN-NEXT: v_exp_f32_e32 v151, v33 - ; GCN-NEXT: v_cvt_f16_f32_e32 v33, v59 - ; GCN-NEXT: v_fma_f32 v150, s4, v34, -v134 - ; GCN-NEXT: v_fma_f32 v8, s4, v8, -v134 - ; GCN-NEXT: v_fma_f32 v12, s4, v12, -v134 - ; GCN-NEXT: v_pack_b32_f16 v149, v146, v33 - ; GCN-NEXT: v_mul_f32_e32 v33, 0x3fb8aa3b, v152 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[154:155], v[144:145], v[112:127] - ; GCN-NEXT: v_fma_f32 v152, s4, v35, -v134 - ; GCN-NEXT: v_exp_f32_e32 v153, v33 - ; GCN-NEXT: v_fma_f32 v155, s4, v36, -v134 - ; GCN-NEXT: v_perm_b32 v36, v158, v156, s5 - ; GCN-NEXT: v_cvt_f16_f32_e32 v154, v160 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[60:61], v[148:149], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v60, 0x3fb8aa3b, v32 - ; GCN-NEXT: ds_read_b128 v[32:35], v57 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[144:147], v57 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mul_f32_e32 v61, 0x3fb8aa3b, v161 - ; GCN-NEXT: v_exp_f32_e32 v165, v60 - ; GCN-NEXT: v_perm_b32 v60, v158, v156, s8 - ; GCN-NEXT: v_fma_f32 v158, s4, v37, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[140:141], v[148:149], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v161, v61 - ; GCN-NEXT: v_perm_b32 v140, v159, v157, s8 - ; GCN-NEXT: v_perm_b32 v37, v130, v128, s5 - ; GCN-NEXT: v_perm_b32 v61, v130, v128, s8 - ; GCN-NEXT: v_perm_b32 v141, v131, v129, s8 + ; GCN-NEXT: v_perm_b32 v188, v194, v192, s5 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[164:165], v[144:145], v[80:95] + ; GCN-NEXT: v_perm_b32 v189, v220, v218, s5 + ; GCN-NEXT: v_perm_b32 v191, v220, v218, s7 + ; GCN-NEXT: v_perm_b32 v190, v194, v192, s7 + ; GCN-NEXT: v_perm_b32 v192, v195, v193, s5 + ; GCN-NEXT: v_perm_b32 v194, v195, v193, s7 + ; GCN-NEXT: v_perm_b32 v193, v221, v219, s5 + ; GCN-NEXT: v_perm_b32 v195, v221, v219, s7 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[166:167], v[146:147], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[168:169], v[140:141], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[170:171], v[142:143], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[172:173], v[148:149], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[214:215], v[136:137], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[174:175], v[150:151], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[216:217], v[138:139], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[176:177], v[136:137], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[226:227], v[132:133], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[178:179], v[138:139], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[160:161], v[136:137], v[64:79] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[230:231], v[128:129], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[228:229], v[134:135], v[96:111] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[180:181], v[132:133], v[80:95] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[162:163], v[138:139], v[64:79] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[232:233], v[130:131], v[112:127] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[234:235], v[128:129], v[96:111] + ; GCN-NEXT: s_nop 9 + ; GCN-NEXT: v_mul_f32_e32 v213, s4, v112 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v113 + ; GCN-NEXT: v_max3_f32 v213, v213, s14, v218 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v114 + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v115 + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v116 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[182:183], v[134:135], v[80:95] + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v117 + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v118 + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v119 + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v120 + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v121 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[152:153], v[132:133], v[64:79] + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v122 + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v123 + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v124 + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v125 + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[236:237], v[130:131], v[96:111] + ; GCN-NEXT: v_mul_f32_e32 v218, s4, v126 + ; GCN-NEXT: v_mul_f32_e32 v219, s4, v127 + ; GCN-NEXT: v_max3_f32 v213, v213, v218, v219 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[184:185], v[128:129], v[80:95] + ; GCN-NEXT: s_nop 6 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v96 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v97 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v98 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v99 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v100 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[154:155], v[134:135], v[64:79] + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v101 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v102 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v103 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v104 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v105 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[186:187], v[130:131], v[80:95] + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v106 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v107 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v108 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v109 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[156:157], v[128:129], v[64:79] + ; GCN-NEXT: v_mul_f32_e32 v214, s4, v110 + ; GCN-NEXT: v_mul_f32_e32 v215, s4, v111 + ; GCN-NEXT: v_max3_f32 v213, v213, v214, v215 + ; GCN-NEXT: v_mul_f32_e32 v140, s4, v80 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v81 + ; GCN-NEXT: v_max3_f32 v140, v213, v140, v141 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v82 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[158:159], v[130:131], v[64:79] + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v83 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v84 + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v85 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v86 + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v87 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v88 + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v89 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v90 + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v91 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v92 + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v93 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v141, s4, v94 + ; GCN-NEXT: v_mul_f32_e32 v142, s4, v95 + ; GCN-NEXT: v_max3_f32 v140, v140, v141, v142 + ; GCN-NEXT: v_mul_f32_e32 v128, s4, v64 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v65 + ; GCN-NEXT: v_max3_f32 v128, v140, v128, v129 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v66 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v67 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v68 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v69 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v70 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v71 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v72 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v73 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v74 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v75 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v76 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v77 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: v_mul_f32_e32 v129, s4, v78 + ; GCN-NEXT: v_mul_f32_e32 v130, s4, v79 + ; GCN-NEXT: v_max3_f32 v128, v128, v129, v130 + ; GCN-NEXT: ds_bpermute_b32 v129, v196, v128 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: ds_read_b128 v[130:133], v198 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[134:137], v198 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_max_f32_e32 v129, v129, v129 + ; GCN-NEXT: v_max_f32_e32 v128, v128, v129 + ; GCN-NEXT: ds_bpermute_b32 v129, v196, v128 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: v_cndmask_b32_e64 v128, v129, v128, s[12:13] + ; GCN-NEXT: v_max_f32_e32 v128, v128, v128 + ; GCN-NEXT: v_max_f32_e32 v128, v212, v128 + ; GCN-NEXT: v_fma_f32 v113, s4, v113, -v128 + ; GCN-NEXT: v_mul_f32_e32 v138, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_fma_f32 v113, s4, v114, -v128 + ; GCN-NEXT: v_mul_f32_e32 v139, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_fma_f32 v113, s4, v115, -v128 + ; GCN-NEXT: v_mul_f32_e32 v140, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_fma_f32 v113, s4, v116, -v128 + ; GCN-NEXT: v_mul_f32_e32 v141, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_fma_f32 v113, s4, v117, -v128 + ; GCN-NEXT: v_mul_f32_e32 v142, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_fma_f32 v113, s4, v118, -v128 + ; GCN-NEXT: v_fma_f32 v112, s4, v112, -v128 + ; GCN-NEXT: v_mul_f32_e32 v143, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_fma_f32 v113, s4, v119, -v128 + ; GCN-NEXT: v_fma_f32 v118, s4, v120, -v128 + ; GCN-NEXT: v_fma_f32 v120, s4, v121, -v128 + ; GCN-NEXT: v_mul_f32_e32 v112, 0x3fb8aa3b, v112 + ; GCN-NEXT: v_mul_f32_e32 v144, 0x3fb8aa3b, v113 + ; GCN-NEXT: v_mul_f32_e32 v149, 0x3fb8aa3b, v120 + ; GCN-NEXT: v_fma_f32 v120, s4, v122, -v128 + ; GCN-NEXT: v_exp_f32_e32 v114, v138 + ; GCN-NEXT: v_exp_f32_e32 v115, v139 + ; GCN-NEXT: v_exp_f32_e32 v116, v140 + ; GCN-NEXT: v_exp_f32_e32 v117, v141 + ; GCN-NEXT: v_mul_f32_e32 v148, 0x3fb8aa3b, v118 + ; GCN-NEXT: v_exp_f32_e32 v118, v142 + ; GCN-NEXT: v_mul_f32_e32 v150, 0x3fb8aa3b, v120 + ; GCN-NEXT: v_exp_f32_e32 v120, v144 + ; GCN-NEXT: v_exp_f32_e32 v113, v112 + ; GCN-NEXT: v_cvt_f16_f32_e32 v119, v114 + ; GCN-NEXT: v_cvt_f16_f32_e32 v121, v116 + ; GCN-NEXT: v_sub_f32_e32 v129, v211, v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v112, v113 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v129 + ; GCN-NEXT: ds_read_b128 v[138:141], v198 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_fma_f32 v122, s4, v123, -v128 + ; GCN-NEXT: v_pack_b32_f16 v146, v112, v119 + ; GCN-NEXT: v_cvt_f16_f32_e32 v112, v115 + ; GCN-NEXT: v_mul_f32_e32 v151, 0x3fb8aa3b, v122 + ; GCN-NEXT: v_cvt_f16_f32_e32 v123, v117 + ; GCN-NEXT: v_fma_f32 v122, s4, v124, -v128 + ; GCN-NEXT: v_pack_b32_f16 v147, v112, v121 + ; GCN-NEXT: v_exp_f32_e32 v112, v129 + ; GCN-NEXT: v_cvt_f16_f32_e32 v124, v118 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v122 + ; GCN-NEXT: v_fma_f32 v125, s4, v125, -v128 + ; GCN-NEXT: v_pk_mul_f32 v[0:1], v[0:1], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[2:3], v[2:3], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[4:5], v[4:5], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[6:7], v[6:7], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[8:9], v[8:9], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[10:11], v[10:11], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[12:13], v[12:13], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[14:15], v[14:15], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[32:33], v[32:33], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[34:35], v[34:35], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[130:131], v[146:147], v[0:15] + ; GCN-NEXT: v_exp_f32_e32 v119, v143 + ; GCN-NEXT: ds_read_b128 v[142:145], v198 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_pk_mul_f32 v[36:37], v[36:37], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[38:39], v[38:39], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[40:41], v[40:41], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[42:43], v[42:43], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[44:45], v[44:45], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[46:47], v[46:47], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[16:17], v[16:17], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[18:19], v[18:19], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[134:135], v[146:147], v[32:47] + ; GCN-NEXT: v_mul_f32_e64 v20, v20, v112 + ; GCN-NEXT: v_mul_f32_e64 v21, v21, v112 + ; GCN-NEXT: v_mul_f32_e64 v22, v22, v112 + ; GCN-NEXT: v_mul_f32_e64 v23, v23, v112 + ; GCN-NEXT: v_mul_f32_e64 v24, v24, v112 + ; GCN-NEXT: v_mul_f32_e64 v25, v25, v112 + ; GCN-NEXT: v_pk_mul_f32 v[26:27], v[26:27], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[28:29], v[28:29], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[30:31], v[30:31], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[48:49], v[48:49], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[50:51], v[50:51], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[52:53], v[52:53], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[54:55], v[54:55], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[56:57], v[56:57], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[58:59], v[58:59], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[60:61], v[60:61], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pk_mul_f32 v[62:63], v[62:63], v[112:113] op_sel_hi:[1,0] + ; GCN-NEXT: v_pack_b32_f16 v134, v123, v124 + ; GCN-NEXT: v_cvt_f16_f32_e32 v130, v119 + ; GCN-NEXT: v_fma_f32 v124, s4, v126, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v120 + ; GCN-NEXT: v_exp_f32_e32 v121, v148 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[138:139], v[146:147], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v122, v149 + ; GCN-NEXT: v_pack_b32_f16 v135, v130, v126 + ; GCN-NEXT: v_mul_f32_e32 v138, 0x3fb8aa3b, v124 + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v121 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v125 + ; GCN-NEXT: v_fma_f32 v139, s4, v96, -v128 + ; GCN-NEXT: v_fma_f32 v127, s4, v127, -v128 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[142:143], v[146:147], v[48:63] + ; GCN-NEXT: v_exp_f32_e32 v123, v150 + ; GCN-NEXT: v_mul_f32_e32 v127, 0x3fb8aa3b, v127 + ; GCN-NEXT: v_fma_f32 v143, s4, v101, -v128 + ; GCN-NEXT: v_fma_f32 v64, s4, v64, -v128 + ; GCN-NEXT: v_fma_f32 v65, s4, v65, -v128 + ; GCN-NEXT: v_fma_f32 v68, s4, v68, -v128 + ; GCN-NEXT: v_fma_f32 v69, s4, v69, -v128 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[132:133], v[134:135], v[0:15] + ; GCN-NEXT: v_exp_f32_e32 v124, v151 + ; GCN-NEXT: ds_read_b128 v[130:133], v197 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[146:149], v197 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[136:137], v[134:135], v[32:47] + ; GCN-NEXT: v_cvt_f16_f32_e32 v136, v122 + ; GCN-NEXT: v_exp_f32_e32 v96, v129 + ; GCN-NEXT: v_fma_f32 v137, s4, v97, -v128 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v139 + ; GCN-NEXT: v_pack_b32_f16 v126, v126, v136 + ; GCN-NEXT: v_cvt_f16_f32_e32 v136, v123 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[140:141], v[134:135], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v97, v125 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v137 + ; GCN-NEXT: v_fma_f32 v137, s4, v98, -v128 + ; GCN-NEXT: v_mul_f32_e32 v142, 0x3fb8aa3b, v137 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[144:145], v[134:135], v[48:63] + ; GCN-NEXT: v_cvt_f16_f32_e32 v134, v124 + ; GCN-NEXT: v_fma_f32 v135, s4, v99, -v128 + ; GCN-NEXT: v_exp_f32_e32 v98, v138 + ; GCN-NEXT: v_exp_f32_e32 v99, v127 + ; GCN-NEXT: v_mul_f32_e32 v150, 0x3fb8aa3b, v135 + ; GCN-NEXT: v_pack_b32_f16 v127, v136, v134 + ; GCN-NEXT: ds_read_b128 v[134:137], v197 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[138:141], v197 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[130:131], v[126:127], v[0:15] + ; GCN-NEXT: v_fma_f32 v131, s4, v100, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v130, v96 + ; GCN-NEXT: v_exp_f32_e32 v100, v129 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v131 + ; GCN-NEXT: v_cvt_f16_f32_e32 v131, v97 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b64 v135, v[36:37] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[32:33], v[148:149], v[96:111] - ; GCN-NEXT: v_perm_b32 v32, v159, v157, s5 - ; GCN-NEXT: v_mul_f32_e32 v33, 0x3fb8aa3b, v150 - ; GCN-NEXT: v_cvt_f16_f32_e32 v150, v151 - ; GCN-NEXT: v_fma_f32 v157, s4, v38, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v38, v153 - ; GCN-NEXT: v_exp_f32_e32 v159, v33 - ; GCN-NEXT: v_perm_b32 v33, v131, v129, s5 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[144:145], v[148:149], v[112:127] - ; GCN-NEXT: v_pack_b32_f16 v129, v150, v38 - ; GCN-NEXT: v_mul_f32_e32 v38, 0x3fb8aa3b, v152 - ; GCN-NEXT: v_exp_f32_e32 v152, v38 + ; GCN-NEXT: ds_write_b64 v199, v[188:189] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v136, v[60:61] + ; GCN-NEXT: ds_write_b64 v200, v[190:191] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v137, v[32:33] - ; GCN-NEXT: ; implicit-def: $vgpr33 - ; GCN-NEXT: ; implicit-def: $vgpr38 + ; GCN-NEXT: ds_write_b64 v201, v[192:193] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v138, v[140:141] - ; GCN-NEXT: v_add_u32_e32 v38, v132, v38 - ; GCN-NEXT: v_add_u32_e32 v33, v132, v33 + ; GCN-NEXT: ds_write_b64 v202, v[194:195] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[146:147], v[126:127], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v101, v125 + ; GCN-NEXT: v_pack_b32_f16 v146, v130, v131 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_load_dwordx2 v[130:131], v38, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[130:131], v210, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx2 v[140:141], v33, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v143 + ; GCN-NEXT: v_cvt_f16_f32_e32 v147, v98 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[134:135], v[126:127], v[16:31] + ; GCN-NEXT: v_fma_f32 v134, s4, v102, -v128 + ; GCN-NEXT: v_mul_f32_e32 v156, 0x3fb8aa3b, v134 + ; GCN-NEXT: buffer_load_dwordx2 v[134:135], v207, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ; implicit-def: $vgpr36 - ; GCN-NEXT: v_add_u32_e32 v33, v132, v36 - ; GCN-NEXT: ; implicit-def: $vgpr37 - ; GCN-NEXT: buffer_load_dwordx2 v[144:145], v33, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: v_exp_f32_e32 v102, v142 + ; GCN-NEXT: buffer_load_dwordx2 v[142:143], v208, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_add_u32_e32 v33, v132, v37 - ; GCN-NEXT: buffer_load_dwordx2 v[148:149], v33, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[144:145], v209, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_cvt_f16_f32_e32 v156, v162 - ; GCN-NEXT: v_mul_f32_e32 v32, 0x3fb8aa3b, v155 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: v_cvt_f16_f32_e32 v33, v165 - ; GCN-NEXT: v_pack_b32_f16 v128, v154, v156 - ; GCN-NEXT: v_fma_f32 v150, s4, v39, -v134 - ; GCN-NEXT: ds_read_b128 v[36:39], v139 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[62:63], v[128:129], v[64:79] - ; GCN-NEXT: v_exp_f32_e32 v154, v32 - ; GCN-NEXT: v_mul_f32_e32 v32, 0x3fb8aa3b, v158 - ; GCN-NEXT: ds_read_b128 v[60:63], v139 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v156, s4, v42, -v134 - ; GCN-NEXT: v_perm_b32 v20, v140, v130, s5 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[142:143], v[128:129], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v155, v32 - ; GCN-NEXT: v_mul_f32_e32 v32, 0x3fb8aa3b, v157 - ; GCN-NEXT: v_cvt_f16_f32_e32 v142, v161 - ; GCN-NEXT: v_fma_f32 v143, s4, v41, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[34:35], v[128:129], v[96:111] - ; GCN-NEXT: v_cvt_f16_f32_e32 v34, v159 - ; GCN-NEXT: v_exp_f32_e32 v157, v32 - ; GCN-NEXT: v_cvt_f16_f32_e32 v32, v152 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[146:147], v[128:129], v[112:127] - ; GCN-NEXT: v_pack_b32_f16 v129, v34, v32 - ; GCN-NEXT: v_mul_f32_e32 v32, 0x3fb8aa3b, v150 - ; GCN-NEXT: v_pack_b32_f16 v128, v33, v142 - ; GCN-NEXT: v_exp_f32_e32 v146, v32 - ; GCN-NEXT: ds_read_b128 v[32:35], v139 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v142, s4, v43, -v134 - ; GCN-NEXT: v_fma_f32 v150, s4, v46, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[36:37], v[128:129], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v36, 0x3fb8aa3b, v40 - ; GCN-NEXT: ds_read_b128 v[40:43], v139 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v147, v36 - ; GCN-NEXT: v_mul_f32_e32 v36, 0x3fb8aa3b, v143 - ; GCN-NEXT: v_cvt_f16_f32_e32 v37, v154 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[60:61], v[128:129], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v143, v36 - ; GCN-NEXT: v_cvt_f16_f32_e32 v60, v155 - ; GCN-NEXT: v_mul_f32_e32 v36, 0x3fb8aa3b, v142 - ; GCN-NEXT: v_fma_f32 v61, s4, v45, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[32:33], v[128:129], v[96:111] - ; GCN-NEXT: v_mul_f32_e32 v32, 0x3fb8aa3b, v156 - ; GCN-NEXT: v_cvt_f16_f32_e32 v33, v157 - ; GCN-NEXT: v_exp_f32_e32 v156, v32 - ; GCN-NEXT: v_cvt_f16_f32_e32 v32, v146 - ; GCN-NEXT: v_pack_b32_f16 v33, v33, v32 - ; GCN-NEXT: v_pack_b32_f16 v32, v37, v60 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[40:41], v[128:129], v[112:127] - ; GCN-NEXT: v_exp_f32_e32 v129, v36 - ; GCN-NEXT: v_mul_f32_e32 v40, 0x3fb8aa3b, v44 - ; GCN-NEXT: v_cvt_f16_f32_e32 v60, v147 - ; GCN-NEXT: v_fma_f32 v128, s4, v47, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[38:39], v[32:33], v[64:79] - ; GCN-NEXT: ds_read_b128 v[36:39], v57 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v142, v40 - ; GCN-NEXT: v_mul_f32_e32 v40, 0x3fb8aa3b, v61 - ; GCN-NEXT: v_cvt_f16_f32_e32 v61, v143 - ; GCN-NEXT: ds_read_b128 v[44:47], v57 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[62:63], v[32:33], v[80:95] - ; GCN-NEXT: v_fma_f32 v62, s4, v17, -v134 - ; GCN-NEXT: v_mul_f32_e32 v17, 0x3fb8aa3b, v150 - ; GCN-NEXT: v_exp_f32_e32 v63, v40 - ; GCN-NEXT: v_pack_b32_f16 v40, v60, v61 - ; GCN-NEXT: v_fma_f32 v150, s4, v18, -v134 - ; GCN-NEXT: v_fma_f32 v60, s4, v19, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v61, v142 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[34:35], v[32:33], v[96:111] - ; GCN-NEXT: v_cvt_f16_f32_e32 v34, v156 - ; GCN-NEXT: v_exp_f32_e32 v158, v17 - ; GCN-NEXT: v_cvt_f16_f32_e32 v17, v129 - ; GCN-NEXT: v_pack_b32_f16 v41, v34, v17 - ; GCN-NEXT: v_mul_f32_e32 v17, 0x3fb8aa3b, v128 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[42:43], v[32:33], v[112:127] - ; GCN-NEXT: v_exp_f32_e32 v128, v17 - ; GCN-NEXT: v_perm_b32 v42, v141, v131, s8 - ; GCN-NEXT: v_perm_b32 v43, v149, v145, s8 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[36:37], v[40:41], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v36, 0x3fb8aa3b, v16 - ; GCN-NEXT: ds_read_b128 v[16:19], v57 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[32:35], v57 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mul_f32_e32 v37, 0x3fb8aa3b, v62 - ; GCN-NEXT: v_exp_f32_e32 v167, v36 - ; GCN-NEXT: v_perm_b32 v36, v140, v130, s8 - ; GCN-NEXT: v_fma_f32 v62, s4, v21, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[44:45], v[40:41], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v130, v37 - ; GCN-NEXT: v_cvt_f16_f32_e32 v45, v158 - ; GCN-NEXT: v_perm_b32 v21, v148, v144, s5 - ; GCN-NEXT: v_perm_b32 v37, v148, v144, s8 - ; GCN-NEXT: v_cvt_f16_f32_e32 v44, v63 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[138:139], v[126:127], v[48:63] + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v99 + ; GCN-NEXT: v_fma_f32 v127, s4, v103, -v128 + ; GCN-NEXT: v_exp_f32_e32 v103, v150 + ; GCN-NEXT: v_fma_f32 v139, s4, v105, -v128 + ; GCN-NEXT: v_pack_b32_f16 v147, v147, v126 + ; GCN-NEXT: v_mul_f32_e32 v138, 0x3fb8aa3b, v127 + ; GCN-NEXT: v_perm_b32 v152, v135, v131, s5 + ; GCN-NEXT: v_perm_b32 v154, v135, v131, s7 + ; GCN-NEXT: v_fma_f32 v135, s4, v104, -v128 + ; GCN-NEXT: v_perm_b32 v126, v134, v130, s5 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[132:133], v[146:147], v[0:15] + ; GCN-NEXT: v_perm_b32 v150, v134, v130, s7 + ; GCN-NEXT: v_cvt_f16_f32_e32 v134, v100 + ; GCN-NEXT: v_exp_f32_e32 v104, v129 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v135 + ; GCN-NEXT: v_cvt_f16_f32_e32 v135, v101 + ; GCN-NEXT: ds_read_b128 v[130:133], v198 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_perm_b32 v127, v144, v142, s5 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[148:149], v[146:147], v[32:47] + ; GCN-NEXT: v_pack_b32_f16 v148, v134, v135 + ; GCN-NEXT: v_fma_f32 v135, s4, v106, -v128 + ; GCN-NEXT: v_exp_f32_e32 v105, v125 + ; GCN-NEXT: v_cvt_f16_f32_e32 v134, v102 + ; GCN-NEXT: v_perm_b32 v151, v144, v142, s7 + ; GCN-NEXT: v_perm_b32 v153, v145, v143, s5 + ; GCN-NEXT: v_perm_b32 v155, v145, v143, s7 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[136:137], v[146:147], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v106, v156 + ; GCN-NEXT: v_mul_f32_e32 v156, 0x3fb8aa3b, v135 + ; GCN-NEXT: v_cvt_f16_f32_e32 v135, v103 + ; GCN-NEXT: v_fma_f32 v136, s4, v107, -v128 + ; GCN-NEXT: ds_read_b128 v[142:145], v198 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v139 + ; GCN-NEXT: v_pack_b32_f16 v149, v134, v135 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[140:141], v[146:147], v[48:63] + ; GCN-NEXT: v_mul_f32_e32 v146, 0x3fb8aa3b, v136 + ; GCN-NEXT: ds_read_b128 v[134:137], v198 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_exp_f32_e32 v107, v138 + ; GCN-NEXT: ds_read_b128 v[138:141], v198 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[130:131], v[148:149], v[0:15] + ; GCN-NEXT: v_fma_f32 v131, s4, v108, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v130, v104 + ; GCN-NEXT: v_exp_f32_e32 v108, v129 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v131 + ; GCN-NEXT: v_cvt_f16_f32_e32 v131, v105 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[142:143], v[148:149], v[32:47] + ; GCN-NEXT: v_fma_f32 v142, s4, v109, -v128 + ; GCN-NEXT: v_exp_f32_e32 v109, v125 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v142 + ; GCN-NEXT: v_pack_b32_f16 v142, v130, v131 + ; GCN-NEXT: v_fma_f32 v131, s4, v110, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v130, v106 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[134:135], v[148:149], v[16:31] + ; GCN-NEXT: v_mul_f32_e32 v134, 0x3fb8aa3b, v131 + ; GCN-NEXT: v_cvt_f16_f32_e32 v131, v107 + ; GCN-NEXT: v_exp_f32_e32 v110, v156 + ; GCN-NEXT: v_fma_f32 v135, s4, v111, -v128 + ; GCN-NEXT: v_mul_f32_e32 v135, 0x3fb8aa3b, v135 + ; GCN-NEXT: v_pack_b32_f16 v143, v130, v131 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[138:139], v[148:149], v[48:63] + ; GCN-NEXT: v_exp_f32_e32 v111, v146 + ; GCN-NEXT: v_fma_f32 v139, s4, v80, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v138, v108 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[132:133], v[142:143], v[0:15] + ; GCN-NEXT: v_exp_f32_e32 v80, v129 + ; GCN-NEXT: ds_read_b128 v[130:133], v197 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[146:149], v197 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v139 + ; GCN-NEXT: v_cvt_f16_f32_e32 v139, v109 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[144:145], v[142:143], v[32:47] + ; GCN-NEXT: v_fma_f32 v144, s4, v81, -v128 + ; GCN-NEXT: v_exp_f32_e32 v81, v125 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v144 + ; GCN-NEXT: v_pack_b32_f16 v144, v138, v139 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[136:137], v[142:143], v[16:31] + ; GCN-NEXT: v_cvt_f16_f32_e32 v136, v110 + ; GCN-NEXT: v_fma_f32 v137, s4, v82, -v128 + ; GCN-NEXT: v_exp_f32_e32 v82, v134 + ; GCN-NEXT: v_cvt_f16_f32_e32 v134, v111 + ; GCN-NEXT: v_mul_f32_e32 v156, 0x3fb8aa3b, v137 + ; GCN-NEXT: v_fma_f32 v137, s4, v83, -v128 + ; GCN-NEXT: v_mul_f32_e32 v157, 0x3fb8aa3b, v137 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[140:141], v[142:143], v[48:63] + ; GCN-NEXT: v_exp_f32_e32 v83, v135 + ; GCN-NEXT: v_pack_b32_f16 v145, v136, v134 + ; GCN-NEXT: ds_read_b128 v[134:137], v197 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[138:141], v197 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b64 v135, v[20:21] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[16:17], v[40:41], v[96:111] - ; GCN-NEXT: v_perm_b32 v16, v141, v131, s5 - ; GCN-NEXT: v_fma_f32 v131, s4, v22, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v22, v128 - ; GCN-NEXT: v_mul_f32_e32 v17, 0x3fb8aa3b, v150 - ; GCN-NEXT: v_exp_f32_e32 v140, v17 - ; GCN-NEXT: v_perm_b32 v17, v149, v145, s5 + ; GCN-NEXT: ds_write_b64 v199, v[126:127] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v136, v[36:37] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[32:33], v[40:41], v[112:127] - ; GCN-NEXT: v_pack_b32_f16 v33, v45, v22 - ; GCN-NEXT: v_mul_f32_e32 v22, 0x3fb8aa3b, v60 - ; GCN-NEXT: v_exp_f32_e32 v144, v22 + ; GCN-NEXT: ds_write_b64 v200, v[150:151] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[130:131], v[144:145], v[0:15] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v137, v[16:17] - ; GCN-NEXT: ; implicit-def: $vgpr17 - ; GCN-NEXT: ; implicit-def: $vgpr22 + ; GCN-NEXT: ds_write_b64 v201, v[152:153] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v138, v[42:43] - ; GCN-NEXT: v_add_u32_e32 v22, v132, v22 - ; GCN-NEXT: v_add_u32_e32 v17, v132, v17 - ; GCN-NEXT: ; implicit-def: $vgpr20 - ; GCN-NEXT: ; implicit-def: $vgpr21 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_load_dwordx2 v[40:41], v22, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: ds_write_b64 v202, v[154:155] + ; GCN-NEXT: v_fma_f32 v127, s4, v84, -v128 + ; GCN-NEXT: v_exp_f32_e32 v84, v129 + ; GCN-NEXT: v_fma_f32 v130, s4, v85, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v80 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v127 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[146:147], v[144:145], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v85, v125 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v130 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_load_dwordx2 v[130:131], v206, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx2 v[42:43], v17, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: v_cvt_f16_f32_e32 v127, v81 + ; GCN-NEXT: v_pack_b32_f16 v126, v126, v127 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[134:135], v[144:145], v[16:31] + ; GCN-NEXT: v_fma_f32 v134, s4, v86, -v128 + ; GCN-NEXT: v_mul_f32_e32 v158, 0x3fb8aa3b, v134 + ; GCN-NEXT: buffer_load_dwordx2 v[134:135], v203, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_add_u32_e32 v20, v132, v20 - ; GCN-NEXT: v_add_u32_e32 v21, v132, v21 - ; GCN-NEXT: v_pack_b32_f16 v32, v61, v44 - ; GCN-NEXT: buffer_load_dwordx2 v[44:45], v20, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[142:143], v204, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: buffer_load_dwordx2 v[60:61], v21, s[0:3], 0 offen sc0 sc1 + ; GCN-NEXT: buffer_load_dwordx2 v[146:147], v205, s[0:3], 0 offen sc0 sc1 ; GCN-NEXT: s_waitcnt vmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mul_f32_e32 v16, 0x3fb8aa3b, v166 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[38:39], v[32:33], v[64:79] - ; GCN-NEXT: v_exp_f32_e32 v132, v16 - ; GCN-NEXT: v_mul_f32_e32 v16, 0x3fb8aa3b, v62 + ; GCN-NEXT: v_cvt_f16_f32_e32 v127, v82 + ; GCN-NEXT: v_exp_f32_e32 v86, v156 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[138:139], v[144:145], v[48:63] + ; GCN-NEXT: v_cvt_f16_f32_e32 v138, v83 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: v_cvt_f16_f32_e32 v17, v167 - ; GCN-NEXT: v_fma_f32 v141, s4, v23, -v134 - ; GCN-NEXT: ds_read_b128 v[20:23], v139 + ; GCN-NEXT: v_fma_f32 v139, s4, v87, -v128 + ; GCN-NEXT: v_exp_f32_e32 v87, v157 + ; GCN-NEXT: v_pack_b32_f16 v127, v127, v138 + ; GCN-NEXT: v_fma_f32 v138, s4, v89, -v128 + ; GCN-NEXT: v_mul_f32_e32 v139, 0x3fb8aa3b, v139 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[132:133], v[126:127], v[0:15] + ; GCN-NEXT: ; implicit-def: $sgpr0 + ; GCN-NEXT: v_perm_b32 v154, v135, v131, s5 + ; GCN-NEXT: v_perm_b32 v156, v135, v131, s7 + ; GCN-NEXT: v_fma_f32 v135, s4, v88, -v128 + ; GCN-NEXT: v_perm_b32 v150, v134, v130, s5 + ; GCN-NEXT: v_perm_b32 v152, v134, v130, s7 + ; GCN-NEXT: ds_read_b128 v[130:133], v198 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_cvt_f16_f32_e32 v134, v84 + ; GCN-NEXT: v_exp_f32_e32 v88, v129 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v135 + ; GCN-NEXT: v_cvt_f16_f32_e32 v135, v85 + ; GCN-NEXT: v_perm_b32 v151, v146, v142, s5 + ; GCN-NEXT: v_perm_b32 v153, v146, v142, s7 + ; GCN-NEXT: v_perm_b32 v155, v147, v143, s5 + ; GCN-NEXT: v_perm_b32 v157, v147, v143, s7 + ; GCN-NEXT: ds_read_b128 v[142:145], v198 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[148:149], v[126:127], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v89, v125 + ; GCN-NEXT: v_pack_b32_f16 v146, v134, v135 + ; GCN-NEXT: v_cvt_f16_f32_e32 v134, v86 + ; GCN-NEXT: v_fma_f32 v135, s4, v90, -v128 + ; GCN-NEXT: v_mul_f32_e32 v125, 0x3fb8aa3b, v138 + ; GCN-NEXT: v_mul_f32_e32 v148, 0x3fb8aa3b, v135 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[136:137], v[126:127], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v90, v158 + ; GCN-NEXT: v_mul_f32_e32 v158, 0x3fb8aa3b, v64 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[140:141], v[126:127], v[48:63] + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v87 + ; GCN-NEXT: v_fma_f32 v127, s4, v91, -v128 + ; GCN-NEXT: v_exp_f32_e32 v91, v139 + ; GCN-NEXT: v_mul_f32_e32 v127, 0x3fb8aa3b, v127 + ; GCN-NEXT: v_pack_b32_f16 v147, v134, v126 + ; GCN-NEXT: ds_read_b128 v[134:137], v198 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[138:141], v198 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[130:131], v[146:147], v[0:15] + ; GCN-NEXT: v_fma_f32 v130, s4, v92, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v88 + ; GCN-NEXT: v_exp_f32_e32 v92, v129 + ; GCN-NEXT: v_mul_f32_e32 v129, 0x3fb8aa3b, v130 + ; GCN-NEXT: v_cvt_f16_f32_e32 v130, v89 + ; GCN-NEXT: v_fma_f32 v131, s4, v93, -v128 + ; GCN-NEXT: v_pack_b32_f16 v130, v126, v130 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[142:143], v[146:147], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v93, v125 + ; GCN-NEXT: v_fma_f32 v126, s4, v94, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v125, v90 + ; GCN-NEXT: v_mul_f32_e32 v143, 0x3fb8aa3b, v126 + ; GCN-NEXT: v_cvt_f16_f32_e32 v126, v91 + ; GCN-NEXT: v_mul_f32_e32 v142, 0x3fb8aa3b, v131 + ; GCN-NEXT: v_fma_f32 v131, s4, v95, -v128 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[134:135], v[146:147], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v94, v148 + ; GCN-NEXT: v_cvt_f16_f32_e32 v64, v93 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[138:139], v[146:147], v[48:63] + ; GCN-NEXT: v_exp_f32_e32 v95, v127 + ; GCN-NEXT: v_cvt_f16_f32_e32 v127, v92 + ; GCN-NEXT: v_mul_f32_e32 v138, 0x3fb8aa3b, v131 + ; GCN-NEXT: v_pack_b32_f16 v131, v125, v126 + ; GCN-NEXT: s_nop 1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[132:133], v[130:131], v[0:15] + ; GCN-NEXT: v_exp_f32_e32 v125, v129 + ; GCN-NEXT: ds_read_b128 v[132:135], v197 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[36:39], v139 offset:576 + ; GCN-NEXT: ds_read_b128 v[146:149], v197 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[144:145], v[130:131], v[32:47] + ; GCN-NEXT: v_mul_f32_e32 v144, 0x3fb8aa3b, v65 + ; GCN-NEXT: v_fma_f32 v65, s4, v66, -v128 + ; GCN-NEXT: v_exp_f32_e32 v126, v142 + ; GCN-NEXT: v_pack_b32_f16 v142, v127, v64 + ; GCN-NEXT: v_cvt_f16_f32_e32 v64, v94 + ; GCN-NEXT: v_mul_f32_e32 v145, 0x3fb8aa3b, v65 + ; GCN-NEXT: v_cvt_f16_f32_e32 v65, v95 + ; GCN-NEXT: v_fma_f32 v66, s4, v67, -v128 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[136:137], v[130:131], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v127, v143 + ; GCN-NEXT: v_pack_b32_f16 v143, v64, v65 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[140:141], v[130:131], v[48:63] + ; GCN-NEXT: v_exp_f32_e32 v129, v138 + ; GCN-NEXT: v_mul_f32_e32 v141, 0x3fb8aa3b, v66 + ; GCN-NEXT: ds_read_b128 v[64:67], v197 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[136:139], v197 offset:1728 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[46:47], v[32:33], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v62, v16 - ; GCN-NEXT: v_mul_f32_e32 v16, 0x3fb8aa3b, v131 - ; GCN-NEXT: v_cvt_f16_f32_e32 v46, v130 - ; GCN-NEXT: v_fma_f32 v47, s4, v25, -v134 - ; GCN-NEXT: v_fma_f32 v131, s4, v26, -v134 - ; GCN-NEXT: v_fma_f32 v149, s4, v4, -v134 - ; GCN-NEXT: ; implicit-def: $sgpr0 - ; GCN-NEXT: v_perm_b32 v4, v42, v40, s5 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[18:19], v[32:33], v[96:111] - ; GCN-NEXT: v_cvt_f16_f32_e32 v18, v140 - ; GCN-NEXT: v_exp_f32_e32 v145, v16 - ; GCN-NEXT: v_cvt_f16_f32_e32 v16, v144 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[34:35], v[32:33], v[112:127] - ; GCN-NEXT: v_pack_b32_f16 v33, v18, v16 - ; GCN-NEXT: v_mul_f32_e32 v16, 0x3fb8aa3b, v141 - ; GCN-NEXT: v_pack_b32_f16 v32, v17, v46 - ; GCN-NEXT: v_exp_f32_e32 v35, v16 - ; GCN-NEXT: ds_read_b128 v[16:19], v139 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v34, s4, v27, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[20:21], v[32:33], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v20, 0x3fb8aa3b, v24 - ; GCN-NEXT: ds_read_b128 v[24:27], v139 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v46, v20 - ; GCN-NEXT: v_mul_f32_e32 v20, 0x3fb8aa3b, v47 - ; GCN-NEXT: v_cvt_f16_f32_e32 v21, v132 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[36:37], v[32:33], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v47, v20 - ; GCN-NEXT: v_cvt_f16_f32_e32 v36, v62 - ; GCN-NEXT: v_mul_f32_e32 v20, 0x3fb8aa3b, v34 - ; GCN-NEXT: v_fma_f32 v37, s4, v29, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v34, v46 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[16:17], v[32:33], v[96:111] - ; GCN-NEXT: v_mul_f32_e32 v16, 0x3fb8aa3b, v131 - ; GCN-NEXT: v_cvt_f16_f32_e32 v17, v145 - ; GCN-NEXT: v_exp_f32_e32 v141, v16 - ; GCN-NEXT: v_cvt_f16_f32_e32 v16, v35 - ; GCN-NEXT: v_fma_f32 v131, s4, v30, -v134 - ; GCN-NEXT: v_pack_b32_f16 v17, v17, v16 - ; GCN-NEXT: v_pack_b32_f16 v16, v21, v36 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[24:25], v[32:33], v[112:127] - ; GCN-NEXT: v_exp_f32_e32 v33, v20 - ; GCN-NEXT: v_mul_f32_e32 v24, 0x3fb8aa3b, v28 - ; GCN-NEXT: v_fma_f32 v32, s4, v31, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[22:23], v[16:17], v[64:79] - ; GCN-NEXT: ds_read_b128 v[20:23], v57 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v36, v24 - ; GCN-NEXT: v_mul_f32_e32 v24, 0x3fb8aa3b, v37 - ; GCN-NEXT: v_cvt_f16_f32_e32 v37, v47 - ; GCN-NEXT: ds_read_b128 v[28:31], v57 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[38:39], v[16:17], v[80:95] - ; GCN-NEXT: v_fma_f32 v38, s4, v1, -v134 - ; GCN-NEXT: v_mul_f32_e32 v1, 0x3fb8aa3b, v131 - ; GCN-NEXT: v_exp_f32_e32 v39, v24 - ; GCN-NEXT: v_pack_b32_f16 v24, v34, v37 - ; GCN-NEXT: v_fma_f32 v131, s4, v2, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v37, v36 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[18:19], v[16:17], v[96:111] - ; GCN-NEXT: v_cvt_f16_f32_e32 v18, v141 - ; GCN-NEXT: v_exp_f32_e32 v148, v1 - ; GCN-NEXT: v_cvt_f16_f32_e32 v1, v33 - ; GCN-NEXT: v_pack_b32_f16 v25, v18, v1 - ; GCN-NEXT: v_mul_f32_e32 v1, 0x3fb8aa3b, v32 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[26:27], v[16:17], v[112:127] - ; GCN-NEXT: v_fma_f32 v32, s4, v3, -v134 - ; GCN-NEXT: v_exp_f32_e32 v34, v1 - ; GCN-NEXT: v_perm_b32 v26, v43, v41, s8 - ; GCN-NEXT: v_perm_b32 v27, v61, v45, s8 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[20:21], v[24:25], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v20, 0x3fb8aa3b, v0 - ; GCN-NEXT: ds_read_b128 v[0:3], v57 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[16:19], v57 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mul_f32_e32 v21, 0x3fb8aa3b, v38 - ; GCN-NEXT: v_exp_f32_e32 v150, v20 - ; GCN-NEXT: v_perm_b32 v20, v42, v40, s8 - ; GCN-NEXT: v_cvt_f16_f32_e32 v40, v148 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[28:29], v[24:25], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v38, v21 - ; GCN-NEXT: v_cvt_f16_f32_e32 v28, v39 - ; GCN-NEXT: v_fma_f32 v29, s4, v5, -v134 - ; GCN-NEXT: v_perm_b32 v5, v60, v44, s5 - ; GCN-NEXT: v_perm_b32 v21, v60, v44, s8 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND ; GCN-NEXT: buffer_wbl2 sc0 sc1 - ; GCN-NEXT: ds_write_b64 v135, v[4:5] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[0:1], v[24:25], v[96:111] - ; GCN-NEXT: v_perm_b32 v0, v43, v41, s5 - ; GCN-NEXT: v_fma_f32 v41, s4, v6, -v134 - ; GCN-NEXT: v_cvt_f16_f32_e32 v6, v34 - ; GCN-NEXT: v_mul_f32_e32 v1, 0x3fb8aa3b, v131 - ; GCN-NEXT: v_exp_f32_e32 v42, v1 - ; GCN-NEXT: v_perm_b32 v1, v61, v45, s5 + ; GCN-NEXT: ds_write_b64 v199, v[150:151] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v136, v[20:21] + ; GCN-NEXT: ds_write_b64 v200, v[152:153] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[132:133], v[142:143], v[0:15] + ; GCN-NEXT: v_cvt_f16_f32_e32 v132, v125 + ; GCN-NEXT: v_exp_f32_e32 v130, v158 ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v137, v[0:1] + ; GCN-NEXT: ds_write_b64 v201, v[154:155] ; GCN-NEXT: buffer_wbl2 sc0 sc1 ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_write_b64 v138, v[26:27] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[16:17], v[24:25], v[112:127] - ; GCN-NEXT: v_pack_b32_f16 v17, v40, v6 - ; GCN-NEXT: v_mul_f32_e32 v6, 0x3fb8aa3b, v32 + ; GCN-NEXT: ds_write_b64 v202, v[156:157] ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND - ; GCN-NEXT: v_pack_b32_f16 v16, v37, v28 - ; GCN-NEXT: v_fma_f32 v24, s4, v7, -v134 - ; GCN-NEXT: v_exp_f32_e32 v25, v6 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_read_b128 v[4:7], v139 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[22:23], v[16:17], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v0, 0x3fb8aa3b, v149 - ; GCN-NEXT: v_exp_f32_e32 v26, v0 - ; GCN-NEXT: v_mul_f32_e32 v0, 0x3fb8aa3b, v29 - ; GCN-NEXT: v_cvt_f16_f32_e32 v1, v150 - ; GCN-NEXT: v_cvt_f16_f32_e32 v27, v38 - ; GCN-NEXT: ds_read_b128 v[20:23], v139 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_fma_f32 v28, s4, v9, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[30:31], v[16:17], v[80:95] - ; GCN-NEXT: v_exp_f32_e32 v29, v0 - ; GCN-NEXT: v_mul_f32_e32 v0, 0x3fb8aa3b, v41 - ; GCN-NEXT: v_fma_f32 v30, s4, v10, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[2:3], v[16:17], v[96:111] - ; GCN-NEXT: v_cvt_f16_f32_e32 v2, v42 - ; GCN-NEXT: v_exp_f32_e32 v31, v0 - ; GCN-NEXT: v_cvt_f16_f32_e32 v0, v25 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[18:19], v[16:17], v[112:127] - ; GCN-NEXT: v_pack_b32_f16 v17, v2, v0 - ; GCN-NEXT: v_pack_b32_f16 v16, v1, v27 - ; GCN-NEXT: v_mul_f32_e32 v0, 0x3fb8aa3b, v24 - ; GCN-NEXT: v_fma_f32 v18, s4, v11, -v134 - ; GCN-NEXT: v_exp_f32_e32 v19, v0 - ; GCN-NEXT: ds_read_b128 v[0:3], v139 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[4:5], v[16:17], v[64:79] - ; GCN-NEXT: v_mul_f32_e32 v4, 0x3fb8aa3b, v8 - ; GCN-NEXT: ds_read_b128 v[8:11], v139 offset:1728 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_exp_f32_e32 v24, v4 - ; GCN-NEXT: v_mul_f32_e32 v4, 0x3fb8aa3b, v28 - ; GCN-NEXT: v_cvt_f16_f32_e32 v5, v26 - ; GCN-NEXT: v_exp_f32_e32 v27, v4 - ; GCN-NEXT: v_mul_f32_e32 v4, 0x3fb8aa3b, v18 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[20:21], v[16:17], v[80:95] - ; GCN-NEXT: v_cvt_f16_f32_e32 v20, v29 - ; GCN-NEXT: v_fma_f32 v21, s4, v13, -v134 - ; GCN-NEXT: v_fma_f32 v28, s4, v14, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[0:1], v[16:17], v[96:111] - ; GCN-NEXT: v_mul_f32_e32 v0, 0x3fb8aa3b, v30 - ; GCN-NEXT: v_cvt_f16_f32_e32 v1, v31 - ; GCN-NEXT: v_exp_f32_e32 v30, v0 - ; GCN-NEXT: v_cvt_f16_f32_e32 v0, v19 - ; GCN-NEXT: v_pack_b32_f16 v1, v1, v0 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[8:9], v[16:17], v[112:127] - ; GCN-NEXT: v_exp_f32_e32 v16, v4 - ; GCN-NEXT: v_pack_b32_f16 v0, v5, v20 - ; GCN-NEXT: v_mul_f32_e32 v9, 0x3fb8aa3b, v12 - ; GCN-NEXT: v_exp_f32_e32 v18, v9 - ; GCN-NEXT: v_mul_f32_e32 v9, 0x3fb8aa3b, v21 - ; GCN-NEXT: v_exp_f32_e32 v21, v9 - ; GCN-NEXT: v_fma_f32 v8, s4, v15, -v134 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[6:7], v[0:1], v[64:79] - ; GCN-NEXT: ds_read_b128 v[4:7], v57 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: ds_read_b128 v[12:15], v57 offset:576 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 - ; GCN-NEXT: v_cvt_f16_f32_e32 v17, v24 - ; GCN-NEXT: v_cvt_f16_f32_e32 v20, v27 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[22:23], v[0:1], v[80:95] - ; GCN-NEXT: v_cvt_f16_f32_e32 v22, v21 - ; GCN-NEXT: v_cvt_f16_f32_e32 v23, v18 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[2:3], v[0:1], v[96:111] - ; GCN-NEXT: v_cvt_f16_f32_e32 v3, v30 - ; GCN-NEXT: v_mul_f32_e32 v2, 0x3fb8aa3b, v28 - ; GCN-NEXT: v_exp_f32_e32 v2, v2 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[112:127], v[10:11], v[0:1], v[112:127] - ; GCN-NEXT: v_cvt_f16_f32_e32 v0, v16 - ; GCN-NEXT: v_mul_f32_e32 v1, 0x3fb8aa3b, v8 - ; GCN-NEXT: v_exp_f32_e32 v10, v1 - ; GCN-NEXT: v_pack_b32_f16 v8, v17, v20 - ; GCN-NEXT: v_pack_b32_f16 v9, v3, v0 - ; GCN-NEXT: v_add_f32_e32 v3, 0, v49 - ; GCN-NEXT: v_add_f32_e32 v3, v50, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v51, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v52, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v53, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v54, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v55, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v56, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v58, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v163, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v164, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v59, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v160, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v162, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v151, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v153, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v165, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v161, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v159, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v152, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v154, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v155, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v157, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v146, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v147, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v143, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v156, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v129, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v142, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v63, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v158, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v128, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v167, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v130, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v140, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v144, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v132, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v62, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v145, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v35, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v46, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v47, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v141, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v33, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v36, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v39, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v148, v3 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[12:13], v[8:9], v[80:95] - ; GCN-NEXT: v_add_f32_e32 v3, v34, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v150, v3 - ; GCN-NEXT: v_cvt_f16_f32_e32 v1, v10 - ; GCN-NEXT: v_cvt_f16_f32_e32 v11, v2 - ; GCN-NEXT: v_add_f32_e32 v3, v38, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v42, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v25, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v26, v3 - ; GCN-NEXT: v_pack_b32_f16 v1, v11, v1 - ; GCN-NEXT: v_pack_b32_f16 v0, v23, v22 - ; GCN-NEXT: v_add_f32_e32 v3, v29, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v31, v3 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[80:95], v[14:15], v[0:1], v[80:95] - ; GCN-NEXT: v_add_f32_e32 v3, v19, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v24, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v27, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v30, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v16, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v18, v3 - ; GCN-NEXT: v_add_f32_e32 v3, v21, v3 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[4:5], v[8:9], v[64:79] - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[64:79], v[6:7], v[0:1], v[64:79] - ; GCN-NEXT: v_add_f32_e32 v0, v2, v3 - ; GCN-NEXT: v_add_f32_e32 v4, v10, v0 - ; GCN-NEXT: ds_bpermute_b32 v5, v133, v4 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: ds_read_b128 v[0:3], v57 offset:1152 - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[146:147], v[142:143], v[32:47] + ; GCN-NEXT: v_mul_f32_e32 v146, 0x3fb8aa3b, v68 + ; GCN-NEXT: v_cvt_f16_f32_e32 v68, v126 + ; GCN-NEXT: v_exp_f32_e32 v131, v144 + ; GCN-NEXT: v_mul_f32_e32 v144, 0x3fb8aa3b, v69 + ; GCN-NEXT: v_fma_f32 v69, s4, v71, -v128 + ; GCN-NEXT: v_pack_b32_f16 v140, v132, v68 + ; GCN-NEXT: v_cvt_f16_f32_e32 v68, v129 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[64:65], v[142:143], v[16:31] + ; GCN-NEXT: v_cvt_f16_f32_e32 v64, v127 + ; GCN-NEXT: v_exp_f32_e32 v132, v145 + ; GCN-NEXT: v_fma_f32 v65, s4, v70, -v128 + ; GCN-NEXT: v_mul_f32_e32 v65, 0x3fb8aa3b, v65 + ; GCN-NEXT: v_fma_f32 v145, s4, v73, -v128 + ; GCN-NEXT: v_mul_f32_e32 v147, 0x3fb8aa3b, v145 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[136:137], v[142:143], v[48:63] + ; GCN-NEXT: v_exp_f32_e32 v133, v141 + ; GCN-NEXT: v_mul_f32_e32 v142, 0x3fb8aa3b, v69 + ; GCN-NEXT: v_pack_b32_f16 v141, v64, v68 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: ds_read_b128 v[68:71], v198 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_fma_f32 v143, s4, v72, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v64, v130 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[134:135], v[140:141], v[0:15] + ; GCN-NEXT: v_exp_f32_e32 v72, v146 + ; GCN-NEXT: v_mul_f32_e32 v146, 0x3fb8aa3b, v143 + ; GCN-NEXT: v_cvt_f16_f32_e32 v143, v131 + ; GCN-NEXT: ds_read_b128 v[134:137], v198 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_pack_b32_f16 v64, v64, v143 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[148:149], v[140:141], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v73, v144 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[66:67], v[140:141], v[16:31] + ; GCN-NEXT: v_cvt_f16_f32_e32 v66, v132 + ; GCN-NEXT: v_fma_f32 v67, s4, v74, -v128 + ; GCN-NEXT: v_exp_f32_e32 v74, v65 + ; GCN-NEXT: v_cvt_f16_f32_e32 v65, v133 + ; GCN-NEXT: v_mul_f32_e32 v67, 0x3fb8aa3b, v67 + ; GCN-NEXT: v_pack_b32_f16 v65, v66, v65 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[138:139], v[140:141], v[48:63] + ; GCN-NEXT: v_fma_f32 v138, s4, v75, -v128 + ; GCN-NEXT: v_exp_f32_e32 v75, v142 + ; GCN-NEXT: v_mul_f32_e32 v148, 0x3fb8aa3b, v138 + ; GCN-NEXT: ds_read_b128 v[138:141], v198 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[142:145], v198 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_cvt_f16_f32_e32 v66, v72 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[68:69], v[64:65], v[0:15] + ; GCN-NEXT: v_fma_f32 v68, s4, v76, -v128 + ; GCN-NEXT: v_exp_f32_e32 v76, v146 + ; GCN-NEXT: v_mul_f32_e32 v146, 0x3fb8aa3b, v68 + ; GCN-NEXT: v_cvt_f16_f32_e32 v68, v73 + ; GCN-NEXT: v_fma_f32 v69, s4, v77, -v128 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[134:135], v[64:65], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v77, v147 + ; GCN-NEXT: v_pack_b32_f16 v134, v66, v68 + ; GCN-NEXT: v_fma_f32 v68, s4, v78, -v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v66, v74 + ; GCN-NEXT: v_mul_f32_e32 v147, 0x3fb8aa3b, v69 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[138:139], v[64:65], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v78, v67 + ; GCN-NEXT: v_mul_f32_e32 v138, 0x3fb8aa3b, v68 + ; GCN-NEXT: v_cvt_f16_f32_e32 v139, v76 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[142:143], v[64:65], v[48:63] + ; GCN-NEXT: v_cvt_f16_f32_e32 v64, v75 + ; GCN-NEXT: v_fma_f32 v65, s4, v79, -v128 + ; GCN-NEXT: v_exp_f32_e32 v79, v148 + ; GCN-NEXT: v_mul_f32_e32 v128, 0x3fb8aa3b, v65 + ; GCN-NEXT: v_pack_b32_f16 v135, v66, v64 + ; GCN-NEXT: s_nop 1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[70:71], v[134:135], v[0:15] + ; GCN-NEXT: v_exp_f32_e32 v142, v146 + ; GCN-NEXT: ds_read_b128 v[68:71], v197 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: ds_read_b128 v[64:67], v197 offset:576 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[136:137], v[134:135], v[32:47] + ; GCN-NEXT: v_exp_f32_e32 v137, v147 + ; GCN-NEXT: v_cvt_f16_f32_e32 v136, v77 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[140:141], v[134:135], v[16:31] + ; GCN-NEXT: v_exp_f32_e32 v138, v138 + ; GCN-NEXT: v_cvt_f16_f32_e32 v140, v78 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[48:63], v[144:145], v[134:135], v[48:63] + ; GCN-NEXT: s_nop 10 + ; GCN-NEXT: v_exp_f32_e32 v52, v128 + ; GCN-NEXT: v_cvt_f16_f32_e32 v50, v137 + ; GCN-NEXT: v_cvt_f16_f32_e32 v51, v142 + ; GCN-NEXT: v_cvt_f16_f32_e32 v54, v138 + ; GCN-NEXT: v_cvt_f16_f32_e32 v53, v52 + ; GCN-NEXT: v_cvt_f16_f32_e32 v49, v79 + ; GCN-NEXT: v_pack_b32_f16 v50, v51, v50 + ; GCN-NEXT: v_pack_b32_f16 v48, v139, v136 + ; GCN-NEXT: v_pack_b32_f16 v51, v54, v53 + ; GCN-NEXT: v_add_f32_e32 v53, 0, v113 + ; GCN-NEXT: v_add_f32_e32 v53, v114, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v115, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v116, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v117, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v118, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v119, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v120, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v121, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v122, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v123, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v124, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v96, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v97, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v98, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v99, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v100, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v101, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v102, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v103, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v104, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v105, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v106, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v107, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v108, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v109, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v110, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v111, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v80, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v81, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v82, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v83, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v84, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v85, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v86, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v87, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v88, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v89, v53 + ; GCN-NEXT: v_pack_b32_f16 v49, v140, v49 + ; GCN-NEXT: v_add_f32_e32 v53, v90, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v91, v53 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[68:69], v[48:49], v[0:15] + ; GCN-NEXT: v_add_f32_e32 v53, v92, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v93, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v94, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v95, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v125, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v126, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v127, v53 + ; GCN-NEXT: v_add_f32_e32 v53, v129, v53 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[0:15], v[70:71], v[50:51], v[0:15] + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[64:65], v[48:49], v[32:47] + ; GCN-NEXT: s_nop 9 + ; GCN-NEXT: v_add_f32_e32 v0, v130, v53 + ; GCN-NEXT: v_add_f32_e32 v0, v131, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v132, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v133, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v72, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v73, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v74, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v75, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v76, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v77, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v78, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v79, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v142, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v137, v0 + ; GCN-NEXT: v_add_f32_e32 v0, v138, v0 + ; GCN-NEXT: v_add_f32_e32 v4, v52, v0 + ; GCN-NEXT: ds_bpermute_b32 v5, v196, v4 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: ds_read_b128 v[0:3], v197 offset:1152 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: buffer_inv sc0 sc1 + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[16:31], v[0:1], v[48:49], v[16:31] ; GCN-NEXT: v_add_f32_e32 v2, v4, v5 - ; GCN-NEXT: ds_bpermute_b32 v3, v133, v2 - ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[96:111], v[0:1], v[8:9], v[96:111] - ; GCN-NEXT: s_waitcnt lgkmcnt(0) - ; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[6:7] + ; GCN-NEXT: ds_bpermute_b32 v3, v196, v2 ; GCN-NEXT: ; implicit-def: $vgpr4 - ; GCN-NEXT: v_fmac_f32_e32 v0, v4, v48 - ; GCN-NEXT: ds_read_b128 v[0:3], v57 offset:1728 + ; GCN-NEXT: s_waitcnt lgkmcnt(0) + ; GCN-NEXT: v_cndmask_b32_e64 v0, v3, v2, s[12:13] + ; GCN-NEXT: v_fmac_f32_e32 v0, v4, v112 + ; GCN-NEXT: ds_read_b128 v[0:3], v197 offset:1728 ; GCN-NEXT: s_waitcnt lgkmcnt(0) ; GCN-NEXT: buffer_inv sc0 sc1 ; GCN-NEXT: ;;#ASMSTART ; GCN-NEXT: s_waitcnt vmcnt(8) ; GCN-NEXT: ;;#ASMEND + ; GCN-NEXT: v_mfma_f32_32x32x8_f16 v[32:47], v[66:67], v[50:51], v[32:47] ; GCN-NEXT: s_endpgm attributes #0 = {"amdgpu-flat-work-group-size"="256,256"} diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll index 7959cee..e174fc1 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.iglp.opt.ll @@ -156,62 +156,62 @@ define amdgpu_kernel void @test_iglp_opt_rev_mfma_gemm(ptr addrspace(3) noalias ; GCN-NEXT: v_lshlrev_b32_e32 v0, 7, v0 ; GCN-NEXT: v_and_b32_e32 v0, 0x1ff80, v0 ; GCN-NEXT: v_mov_b32_e32 v2, 1.0 -; GCN-NEXT: v_mov_b32_e32 v3, 2.0 +; GCN-NEXT: v_mov_b32_e32 v1, 2.0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_add_u32_e32 v1, s0, v0 -; GCN-NEXT: ds_read_b128 a[28:31], v1 offset:112 -; GCN-NEXT: ds_read_b128 a[24:27], v1 offset:96 -; GCN-NEXT: ds_read_b128 a[20:23], v1 offset:80 -; GCN-NEXT: ds_read_b128 a[16:19], v1 offset:64 -; GCN-NEXT: ds_read_b128 a[0:3], v1 -; GCN-NEXT: ds_read_b128 a[4:7], v1 offset:16 -; GCN-NEXT: ds_read_b128 a[8:11], v1 offset:32 -; GCN-NEXT: ds_read_b128 a[12:15], v1 offset:48 +; GCN-NEXT: v_add_u32_e32 v3, s0, v0 +; GCN-NEXT: ds_read_b128 a[28:31], v3 offset:112 +; GCN-NEXT: ds_read_b128 a[24:27], v3 offset:96 +; GCN-NEXT: ds_read_b128 a[20:23], v3 offset:80 +; GCN-NEXT: ds_read_b128 a[16:19], v3 offset:64 +; GCN-NEXT: ds_read_b128 a[0:3], v3 +; GCN-NEXT: ds_read_b128 a[4:7], v3 offset:16 +; GCN-NEXT: ds_read_b128 a[8:11], v3 offset:32 +; GCN-NEXT: ds_read_b128 a[12:15], v3 offset:48 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v2, v3, a[0:31] -; GCN-NEXT: ds_read_b128 a[156:159], v1 offset:8304 -; GCN-NEXT: ds_read_b128 a[152:155], v1 offset:8288 -; GCN-NEXT: ds_read_b128 a[148:151], v1 offset:8272 -; GCN-NEXT: ds_read_b128 a[144:147], v1 offset:8256 -; GCN-NEXT: ds_read_b128 a[140:143], v1 offset:8240 -; GCN-NEXT: ds_read_b128 a[136:139], v1 offset:8224 -; GCN-NEXT: ds_read_b128 a[132:135], v1 offset:8208 -; GCN-NEXT: ds_read_b128 a[128:131], v1 offset:8192 +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v2, v1, a[0:31] +; GCN-NEXT: ds_read_b128 a[156:159], v3 offset:8304 +; GCN-NEXT: ds_read_b128 a[152:155], v3 offset:8288 +; GCN-NEXT: ds_read_b128 a[148:151], v3 offset:8272 +; GCN-NEXT: ds_read_b128 a[144:147], v3 offset:8256 +; GCN-NEXT: ds_read_b128 a[140:143], v3 offset:8240 +; GCN-NEXT: ds_read_b128 a[136:139], v3 offset:8224 +; GCN-NEXT: ds_read_b128 a[132:135], v3 offset:8208 +; GCN-NEXT: ds_read_b128 a[128:131], v3 offset:8192 +; GCN-NEXT: v_add_u32_e32 v4, 0x6000, v3 ; GCN-NEXT: v_add_u32_e32 v0, s1, v0 ; GCN-NEXT: ; iglp_opt mask(0x00000001) ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[128:159], v2, v3, a[128:159] -; GCN-NEXT: ds_read_b128 a[124:127], v1 offset:24688 -; GCN-NEXT: ds_read_b128 a[120:123], v1 offset:24672 -; GCN-NEXT: ds_read_b128 a[116:119], v1 offset:24656 -; GCN-NEXT: ds_read_b128 a[112:115], v1 offset:24640 -; GCN-NEXT: ds_read_b128 a[108:111], v1 offset:24624 -; GCN-NEXT: ds_read_b128 a[104:107], v1 offset:24608 -; GCN-NEXT: ds_read_b128 a[100:103], v1 offset:24592 -; GCN-NEXT: ds_read_b128 a[96:99], v1 offset:24576 +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[128:159], v2, v1, a[128:159] +; GCN-NEXT: ds_read_b128 a[124:127], v3 offset:24688 +; GCN-NEXT: ds_read_b128 a[120:123], v3 offset:24672 +; GCN-NEXT: ds_read_b128 a[116:119], v3 offset:24656 +; GCN-NEXT: ds_read_b128 a[112:115], v3 offset:24640 +; GCN-NEXT: ds_read_b128 a[108:111], v3 offset:24624 +; GCN-NEXT: ds_read_b128 a[104:107], v3 offset:24608 +; GCN-NEXT: ds_read_b128 a[100:103], v3 offset:24592 +; GCN-NEXT: ds_read_b128 a[96:99], v3 offset:24576 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[96:127], v2, v3, a[96:127] -; GCN-NEXT: ds_read_b128 a[92:95], v1 offset:49264 -; GCN-NEXT: ds_read_b128 a[88:91], v1 offset:49248 -; GCN-NEXT: ds_read_b128 a[84:87], v1 offset:49232 -; GCN-NEXT: ds_read_b128 a[80:83], v1 offset:49216 -; GCN-NEXT: ds_read_b128 a[76:79], v1 offset:49200 -; GCN-NEXT: ds_read_b128 a[72:75], v1 offset:49184 -; GCN-NEXT: ds_read_b128 a[68:71], v1 offset:49168 -; GCN-NEXT: ds_read_b128 a[64:67], v1 offset:49152 -; GCN-NEXT: v_add_u32_e32 v1, 0x6000, v1 +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[96:127], v2, v1, a[96:127] +; GCN-NEXT: ds_read_b128 a[92:95], v3 offset:49264 +; GCN-NEXT: ds_read_b128 a[88:91], v3 offset:49248 +; GCN-NEXT: ds_read_b128 a[84:87], v3 offset:49232 +; GCN-NEXT: ds_read_b128 a[80:83], v3 offset:49216 +; GCN-NEXT: ds_read_b128 a[76:79], v3 offset:49200 +; GCN-NEXT: ds_read_b128 a[72:75], v3 offset:49184 +; GCN-NEXT: ds_read_b128 a[68:71], v3 offset:49168 +; GCN-NEXT: ds_read_b128 a[64:67], v3 offset:49152 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[64:95], v2, v3, a[64:95] -; GCN-NEXT: ds_read_b128 a[60:63], v1 offset:57456 -; GCN-NEXT: ds_read_b128 a[56:59], v1 offset:57440 -; GCN-NEXT: ds_read_b128 a[52:55], v1 offset:57424 -; GCN-NEXT: ds_read_b128 a[48:51], v1 offset:57408 -; GCN-NEXT: ds_read_b128 a[32:35], v1 offset:57344 -; GCN-NEXT: ds_read_b128 a[36:39], v1 offset:57360 -; GCN-NEXT: ds_read_b128 a[40:43], v1 offset:57376 -; GCN-NEXT: ds_read_b128 a[44:47], v1 offset:57392 +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[64:95], v2, v1, a[64:95] +; GCN-NEXT: ds_read_b128 a[60:63], v4 offset:57456 +; GCN-NEXT: ds_read_b128 a[56:59], v4 offset:57440 +; GCN-NEXT: ds_read_b128 a[52:55], v4 offset:57424 +; GCN-NEXT: ds_read_b128 a[48:51], v4 offset:57408 +; GCN-NEXT: ds_read_b128 a[32:35], v4 offset:57344 +; GCN-NEXT: ds_read_b128 a[36:39], v4 offset:57360 +; GCN-NEXT: ds_read_b128 a[40:43], v4 offset:57376 +; GCN-NEXT: ds_read_b128 a[44:47], v4 offset:57392 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v2, v3, a[32:63] +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v2, v1, a[32:63] ; GCN-NEXT: ds_write_b128 v0, a[28:31] offset:112 ; GCN-NEXT: ds_write_b128 v0, a[24:27] offset:96 ; GCN-NEXT: ds_write_b128 v0, a[20:23] offset:80 diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.ll index aa099b6..b65a1a8 100644 --- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.ll +++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.sched.group.barrier.ll @@ -623,62 +623,62 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_cluster(ptr ad ; GCN-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 ; GCN-NEXT: v_lshlrev_b32_e32 v0, 7, v0 ; GCN-NEXT: v_and_b32_e32 v0, 0x1ff80, v0 +; GCN-NEXT: v_mov_b32_e32 v2, 1.0 +; GCN-NEXT: v_mov_b32_e32 v1, 2.0 ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_add_u32_e32 v1, s0, v0 -; GCN-NEXT: ds_read_b128 a[156:159], v1 offset:112 -; GCN-NEXT: ds_read_b128 a[152:155], v1 offset:96 -; GCN-NEXT: ds_read_b128 a[148:151], v1 offset:80 -; GCN-NEXT: ds_read_b128 a[144:147], v1 offset:64 -; GCN-NEXT: ds_read_b128 a[128:131], v1 -; GCN-NEXT: ds_read_b128 a[132:135], v1 offset:16 -; GCN-NEXT: ds_read_b128 a[136:139], v1 offset:32 -; GCN-NEXT: ds_read_b128 a[140:143], v1 offset:48 -; GCN-NEXT: ds_read_b128 a[28:31], v1 offset:8304 -; GCN-NEXT: ds_read_b128 a[24:27], v1 offset:8288 -; GCN-NEXT: ds_read_b128 a[20:23], v1 offset:8272 -; GCN-NEXT: ds_read_b128 a[16:19], v1 offset:8256 -; GCN-NEXT: ds_read_b128 a[12:15], v1 offset:8240 -; GCN-NEXT: ds_read_b128 a[8:11], v1 offset:8224 -; GCN-NEXT: ds_read_b128 a[4:7], v1 offset:8208 -; GCN-NEXT: ds_read_b128 a[0:3], v1 offset:8192 -; GCN-NEXT: v_add_u32_e32 v2, 0x6000, v1 -; GCN-NEXT: ds_read_b128 a[124:127], v1 offset:24688 -; GCN-NEXT: ds_read_b128 a[120:123], v1 offset:24672 -; GCN-NEXT: ds_read_b128 a[116:119], v1 offset:24656 -; GCN-NEXT: ds_read_b128 a[112:115], v1 offset:24640 -; GCN-NEXT: ds_read_b128 a[108:111], v1 offset:24624 -; GCN-NEXT: ds_read_b128 a[104:107], v1 offset:24608 -; GCN-NEXT: ds_read_b128 a[100:103], v1 offset:24592 -; GCN-NEXT: ds_read_b128 a[96:99], v1 offset:24576 -; GCN-NEXT: ds_read_b128 a[92:95], v1 offset:49264 -; GCN-NEXT: ds_read_b128 a[88:91], v1 offset:49248 -; GCN-NEXT: ds_read_b128 a[84:87], v1 offset:49232 -; GCN-NEXT: ds_read_b128 a[80:83], v1 offset:49216 -; GCN-NEXT: ds_read_b128 a[76:79], v1 offset:49200 -; GCN-NEXT: ds_read_b128 a[72:75], v1 offset:49184 -; GCN-NEXT: ds_read_b128 a[68:71], v1 offset:49168 -; GCN-NEXT: ds_read_b128 a[64:67], v1 offset:49152 -; GCN-NEXT: v_mov_b32_e32 v1, 1.0 -; GCN-NEXT: ds_read_b128 a[60:63], v2 offset:57456 -; GCN-NEXT: ds_read_b128 a[56:59], v2 offset:57440 -; GCN-NEXT: ds_read_b128 a[52:55], v2 offset:57424 -; GCN-NEXT: ds_read_b128 a[48:51], v2 offset:57408 -; GCN-NEXT: ds_read_b128 a[32:35], v2 offset:57344 -; GCN-NEXT: ds_read_b128 a[36:39], v2 offset:57360 -; GCN-NEXT: ds_read_b128 a[40:43], v2 offset:57376 -; GCN-NEXT: ds_read_b128 a[44:47], v2 offset:57392 -; GCN-NEXT: v_mov_b32_e32 v2, 2.0 +; GCN-NEXT: v_add_u32_e32 v3, s0, v0 +; GCN-NEXT: ds_read_b128 a[156:159], v3 offset:112 +; GCN-NEXT: ds_read_b128 a[152:155], v3 offset:96 +; GCN-NEXT: ds_read_b128 a[148:151], v3 offset:80 +; GCN-NEXT: ds_read_b128 a[144:147], v3 offset:64 +; GCN-NEXT: ds_read_b128 a[128:131], v3 +; GCN-NEXT: ds_read_b128 a[132:135], v3 offset:16 +; GCN-NEXT: ds_read_b128 a[136:139], v3 offset:32 +; GCN-NEXT: ds_read_b128 a[140:143], v3 offset:48 +; GCN-NEXT: v_add_u32_e32 v4, 0x6000, v3 +; GCN-NEXT: ds_read_b128 a[28:31], v3 offset:8304 +; GCN-NEXT: ds_read_b128 a[24:27], v3 offset:8288 +; GCN-NEXT: ds_read_b128 a[20:23], v3 offset:8272 +; GCN-NEXT: ds_read_b128 a[16:19], v3 offset:8256 +; GCN-NEXT: ds_read_b128 a[12:15], v3 offset:8240 +; GCN-NEXT: ds_read_b128 a[8:11], v3 offset:8224 +; GCN-NEXT: ds_read_b128 a[4:7], v3 offset:8208 +; GCN-NEXT: ds_read_b128 a[0:3], v3 offset:8192 +; GCN-NEXT: ds_read_b128 a[124:127], v3 offset:24688 +; GCN-NEXT: ds_read_b128 a[120:123], v3 offset:24672 +; GCN-NEXT: ds_read_b128 a[116:119], v3 offset:24656 +; GCN-NEXT: ds_read_b128 a[112:115], v3 offset:24640 +; GCN-NEXT: ds_read_b128 a[108:111], v3 offset:24624 +; GCN-NEXT: ds_read_b128 a[104:107], v3 offset:24608 +; GCN-NEXT: ds_read_b128 a[100:103], v3 offset:24592 +; GCN-NEXT: ds_read_b128 a[96:99], v3 offset:24576 +; GCN-NEXT: ds_read_b128 a[92:95], v3 offset:49264 +; GCN-NEXT: ds_read_b128 a[88:91], v3 offset:49248 +; GCN-NEXT: ds_read_b128 a[84:87], v3 offset:49232 +; GCN-NEXT: ds_read_b128 a[80:83], v3 offset:49216 +; GCN-NEXT: ds_read_b128 a[76:79], v3 offset:49200 +; GCN-NEXT: ds_read_b128 a[72:75], v3 offset:49184 +; GCN-NEXT: ds_read_b128 a[68:71], v3 offset:49168 +; GCN-NEXT: ds_read_b128 a[64:67], v3 offset:49152 +; GCN-NEXT: ds_read_b128 a[60:63], v4 offset:57456 +; GCN-NEXT: ds_read_b128 a[56:59], v4 offset:57440 +; GCN-NEXT: ds_read_b128 a[52:55], v4 offset:57424 +; GCN-NEXT: ds_read_b128 a[48:51], v4 offset:57408 +; GCN-NEXT: ds_read_b128 a[32:35], v4 offset:57344 +; GCN-NEXT: ds_read_b128 a[36:39], v4 offset:57360 +; GCN-NEXT: ds_read_b128 a[40:43], v4 offset:57376 +; GCN-NEXT: ds_read_b128 a[44:47], v4 offset:57392 +; GCN-NEXT: s_waitcnt lgkmcnt(14) +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[128:159], v2, v1, a[128:159] ; GCN-NEXT: v_add_u32_e32 v0, s1, v0 ; GCN-NEXT: ; sched_group_barrier mask(0x00000100) size(40) SyncID(0) -; GCN-NEXT: s_waitcnt lgkmcnt(14) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[128:159], v1, v2, a[128:159] -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v1, v2, a[0:31] -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[96:127], v1, v2, a[96:127] ; GCN-NEXT: s_waitcnt lgkmcnt(8) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[64:95], v1, v2, a[64:95] +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[64:95], v2, v1, a[64:95] +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[96:127], v2, v1, a[96:127] +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v2, v1, a[0:31] ; GCN-NEXT: s_waitcnt lgkmcnt(0) -; GCN-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v1, v2, a[32:63] -; GCN-NEXT: s_nop 12 +; GCN-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v2, v1, a[32:63] +; GCN-NEXT: s_nop 11 ; GCN-NEXT: ds_write_b128 v0, a[156:159] offset:112 ; GCN-NEXT: ds_write_b128 v0, a[152:155] offset:96 ; GCN-NEXT: ds_write_b128 v0, a[148:151] offset:80 @@ -729,62 +729,62 @@ define amdgpu_kernel void @test_sched_group_barrier_pipeline_MFMA_cluster(ptr ad ; EXACTCUTOFF-NEXT: s_load_dwordx2 s[0:1], s[4:5], 0x24 ; EXACTCUTOFF-NEXT: v_lshlrev_b32_e32 v0, 7, v0 ; EXACTCUTOFF-NEXT: v_and_b32_e32 v0, 0x1ff80, v0 +; EXACTCUTOFF-NEXT: v_mov_b32_e32 v2, 1.0 +; EXACTCUTOFF-NEXT: v_mov_b32_e32 v1, 2.0 ; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0) -; EXACTCUTOFF-NEXT: v_add_u32_e32 v1, s0, v0 -; EXACTCUTOFF-NEXT: ds_read_b128 a[156:159], v1 offset:112 -; EXACTCUTOFF-NEXT: ds_read_b128 a[152:155], v1 offset:96 -; EXACTCUTOFF-NEXT: ds_read_b128 a[148:151], v1 offset:80 -; EXACTCUTOFF-NEXT: ds_read_b128 a[144:147], v1 offset:64 -; EXACTCUTOFF-NEXT: ds_read_b128 a[128:131], v1 -; EXACTCUTOFF-NEXT: ds_read_b128 a[132:135], v1 offset:16 -; EXACTCUTOFF-NEXT: ds_read_b128 a[136:139], v1 offset:32 -; EXACTCUTOFF-NEXT: ds_read_b128 a[140:143], v1 offset:48 -; EXACTCUTOFF-NEXT: ds_read_b128 a[28:31], v1 offset:8304 -; EXACTCUTOFF-NEXT: ds_read_b128 a[24:27], v1 offset:8288 -; EXACTCUTOFF-NEXT: ds_read_b128 a[20:23], v1 offset:8272 -; EXACTCUTOFF-NEXT: ds_read_b128 a[16:19], v1 offset:8256 -; EXACTCUTOFF-NEXT: ds_read_b128 a[12:15], v1 offset:8240 -; EXACTCUTOFF-NEXT: ds_read_b128 a[8:11], v1 offset:8224 -; EXACTCUTOFF-NEXT: ds_read_b128 a[4:7], v1 offset:8208 -; EXACTCUTOFF-NEXT: ds_read_b128 a[0:3], v1 offset:8192 -; EXACTCUTOFF-NEXT: v_add_u32_e32 v2, 0x6000, v1 -; EXACTCUTOFF-NEXT: ds_read_b128 a[124:127], v1 offset:24688 -; EXACTCUTOFF-NEXT: ds_read_b128 a[120:123], v1 offset:24672 -; EXACTCUTOFF-NEXT: ds_read_b128 a[116:119], v1 offset:24656 -; EXACTCUTOFF-NEXT: ds_read_b128 a[112:115], v1 offset:24640 -; EXACTCUTOFF-NEXT: ds_read_b128 a[108:111], v1 offset:24624 -; EXACTCUTOFF-NEXT: ds_read_b128 a[104:107], v1 offset:24608 -; EXACTCUTOFF-NEXT: ds_read_b128 a[100:103], v1 offset:24592 -; EXACTCUTOFF-NEXT: ds_read_b128 a[96:99], v1 offset:24576 -; EXACTCUTOFF-NEXT: ds_read_b128 a[92:95], v1 offset:49264 -; EXACTCUTOFF-NEXT: ds_read_b128 a[88:91], v1 offset:49248 -; EXACTCUTOFF-NEXT: ds_read_b128 a[84:87], v1 offset:49232 -; EXACTCUTOFF-NEXT: ds_read_b128 a[80:83], v1 offset:49216 -; EXACTCUTOFF-NEXT: ds_read_b128 a[76:79], v1 offset:49200 -; EXACTCUTOFF-NEXT: ds_read_b128 a[72:75], v1 offset:49184 -; EXACTCUTOFF-NEXT: ds_read_b128 a[68:71], v1 offset:49168 -; EXACTCUTOFF-NEXT: ds_read_b128 a[64:67], v1 offset:49152 -; EXACTCUTOFF-NEXT: v_mov_b32_e32 v1, 1.0 -; EXACTCUTOFF-NEXT: ds_read_b128 a[60:63], v2 offset:57456 -; EXACTCUTOFF-NEXT: ds_read_b128 a[56:59], v2 offset:57440 -; EXACTCUTOFF-NEXT: ds_read_b128 a[52:55], v2 offset:57424 -; EXACTCUTOFF-NEXT: ds_read_b128 a[48:51], v2 offset:57408 -; EXACTCUTOFF-NEXT: ds_read_b128 a[32:35], v2 offset:57344 -; EXACTCUTOFF-NEXT: ds_read_b128 a[36:39], v2 offset:57360 -; EXACTCUTOFF-NEXT: ds_read_b128 a[40:43], v2 offset:57376 -; EXACTCUTOFF-NEXT: ds_read_b128 a[44:47], v2 offset:57392 -; EXACTCUTOFF-NEXT: v_mov_b32_e32 v2, 2.0 +; EXACTCUTOFF-NEXT: v_add_u32_e32 v3, s0, v0 +; EXACTCUTOFF-NEXT: ds_read_b128 a[156:159], v3 offset:112 +; EXACTCUTOFF-NEXT: ds_read_b128 a[152:155], v3 offset:96 +; EXACTCUTOFF-NEXT: ds_read_b128 a[148:151], v3 offset:80 +; EXACTCUTOFF-NEXT: ds_read_b128 a[144:147], v3 offset:64 +; EXACTCUTOFF-NEXT: ds_read_b128 a[128:131], v3 +; EXACTCUTOFF-NEXT: ds_read_b128 a[132:135], v3 offset:16 +; EXACTCUTOFF-NEXT: ds_read_b128 a[136:139], v3 offset:32 +; EXACTCUTOFF-NEXT: ds_read_b128 a[140:143], v3 offset:48 +; EXACTCUTOFF-NEXT: v_add_u32_e32 v4, 0x6000, v3 +; EXACTCUTOFF-NEXT: ds_read_b128 a[28:31], v3 offset:8304 +; EXACTCUTOFF-NEXT: ds_read_b128 a[24:27], v3 offset:8288 +; EXACTCUTOFF-NEXT: ds_read_b128 a[20:23], v3 offset:8272 +; EXACTCUTOFF-NEXT: ds_read_b128 a[16:19], v3 offset:8256 +; EXACTCUTOFF-NEXT: ds_read_b128 a[12:15], v3 offset:8240 +; EXACTCUTOFF-NEXT: ds_read_b128 a[8:11], v3 offset:8224 +; EXACTCUTOFF-NEXT: ds_read_b128 a[4:7], v3 offset:8208 +; EXACTCUTOFF-NEXT: ds_read_b128 a[0:3], v3 offset:8192 +; EXACTCUTOFF-NEXT: ds_read_b128 a[124:127], v3 offset:24688 +; EXACTCUTOFF-NEXT: ds_read_b128 a[120:123], v3 offset:24672 +; EXACTCUTOFF-NEXT: ds_read_b128 a[116:119], v3 offset:24656 +; EXACTCUTOFF-NEXT: ds_read_b128 a[112:115], v3 offset:24640 +; EXACTCUTOFF-NEXT: ds_read_b128 a[108:111], v3 offset:24624 +; EXACTCUTOFF-NEXT: ds_read_b128 a[104:107], v3 offset:24608 +; EXACTCUTOFF-NEXT: ds_read_b128 a[100:103], v3 offset:24592 +; EXACTCUTOFF-NEXT: ds_read_b128 a[96:99], v3 offset:24576 +; EXACTCUTOFF-NEXT: ds_read_b128 a[92:95], v3 offset:49264 +; EXACTCUTOFF-NEXT: ds_read_b128 a[88:91], v3 offset:49248 +; EXACTCUTOFF-NEXT: ds_read_b128 a[84:87], v3 offset:49232 +; EXACTCUTOFF-NEXT: ds_read_b128 a[80:83], v3 offset:49216 +; EXACTCUTOFF-NEXT: ds_read_b128 a[76:79], v3 offset:49200 +; EXACTCUTOFF-NEXT: ds_read_b128 a[72:75], v3 offset:49184 +; EXACTCUTOFF-NEXT: ds_read_b128 a[68:71], v3 offset:49168 +; EXACTCUTOFF-NEXT: ds_read_b128 a[64:67], v3 offset:49152 +; EXACTCUTOFF-NEXT: ds_read_b128 a[60:63], v4 offset:57456 +; EXACTCUTOFF-NEXT: ds_read_b128 a[56:59], v4 offset:57440 +; EXACTCUTOFF-NEXT: ds_read_b128 a[52:55], v4 offset:57424 +; EXACTCUTOFF-NEXT: ds_read_b128 a[48:51], v4 offset:57408 +; EXACTCUTOFF-NEXT: ds_read_b128 a[32:35], v4 offset:57344 +; EXACTCUTOFF-NEXT: ds_read_b128 a[36:39], v4 offset:57360 +; EXACTCUTOFF-NEXT: ds_read_b128 a[40:43], v4 offset:57376 +; EXACTCUTOFF-NEXT: ds_read_b128 a[44:47], v4 offset:57392 +; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(14) +; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[128:159], v2, v1, a[128:159] ; EXACTCUTOFF-NEXT: v_add_u32_e32 v0, s1, v0 ; EXACTCUTOFF-NEXT: ; sched_group_barrier mask(0x00000100) size(40) SyncID(0) -; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(14) -; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[128:159], v1, v2, a[128:159] -; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v1, v2, a[0:31] -; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[96:127], v1, v2, a[96:127] ; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(8) -; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[64:95], v1, v2, a[64:95] +; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[64:95], v2, v1, a[64:95] +; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[96:127], v2, v1, a[96:127] +; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[0:31], v2, v1, a[0:31] ; EXACTCUTOFF-NEXT: s_waitcnt lgkmcnt(0) -; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v1, v2, a[32:63] -; EXACTCUTOFF-NEXT: s_nop 12 +; EXACTCUTOFF-NEXT: v_mfma_f32_32x32x1f32 a[32:63], v2, v1, a[32:63] +; EXACTCUTOFF-NEXT: s_nop 11 ; EXACTCUTOFF-NEXT: ds_write_b128 v0, a[156:159] offset:112 ; EXACTCUTOFF-NEXT: ds_write_b128 v0, a[152:155] offset:96 ; EXACTCUTOFF-NEXT: ds_write_b128 v0, a[148:151] offset:80 diff --git a/llvm/test/CodeGen/AMDGPU/readsteadycounter.ll b/llvm/test/CodeGen/AMDGPU/readsteadycounter.ll index ddbae64..a95d8c7 100644 --- a/llvm/test/CodeGen/AMDGPU/readsteadycounter.ll +++ b/llvm/test/CodeGen/AMDGPU/readsteadycounter.ll @@ -1,8 +1,8 @@ ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx700 < %s | FileCheck %s -check-prefixes=GCN,GFX700 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=GCN,GFX900 -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=GCN,GFX900 +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx900 < %s | FileCheck %s -check-prefixes=GCN,GFX900 ; RUN: llc -global-isel=0 -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck %s -check-prefixes=GCN,GFX1100 -; RUN: llc -global-isel=1 -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck %s -check-prefixes=GCN,GFX1100 +; RUN: llc -global-isel=1 -new-reg-bank-select -mtriple=amdgcn -mcpu=gfx1100 < %s | FileCheck %s -check-prefixes=GCN,GFX1100 declare i64 @llvm.readsteadycounter() #0 diff --git a/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll b/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll index 9a23788..8803f3a 100644 --- a/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll +++ b/llvm/test/CodeGen/AMDGPU/rewrite-vgpr-mfma-to-agpr.ll @@ -367,77 +367,76 @@ bb: define amdgpu_kernel void @illegal_mfma_after_rewrite() #1 { ; CHECK-LABEL: illegal_mfma_after_rewrite: ; CHECK: ; %bb.0: ; %entry -; CHECK-NEXT: s_mov_b32 s0, 0 -; CHECK-NEXT: s_mov_b32 s1, s0 -; CHECK-NEXT: v_mov_b64_e32 v[28:29], s[0:1] +; CHECK-NEXT: s_mov_b32 s4, 0 +; CHECK-NEXT: s_mov_b32 s5, s4 +; CHECK-NEXT: v_mov_b64_e32 v[26:27], s[4:5] ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; def s[0:3] ; CHECK-NEXT: ;;#ASMEND +; CHECK-NEXT: ;;#ASMSTART +; CHECK-NEXT: ; def v[16:19] +; CHECK-NEXT: ;;#ASMEND ; CHECK-NEXT: s_nop 0 -; CHECK-NEXT: v_mov_b64_e32 v[6:7], s[2:3] -; CHECK-NEXT: v_mov_b64_e32 v[4:5], s[0:1] +; CHECK-NEXT: v_mov_b64_e32 v[0:1], s[0:1] +; CHECK-NEXT: v_mov_b64_e32 v[2:3], s[2:3] ; CHECK-NEXT: s_mov_b32 s0, 0x3c003c00 ; CHECK-NEXT: s_mov_b32 s1, s0 -; CHECK-NEXT: v_mov_b64_e32 v[30:31], s[0:1] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[4:7], v[26:27], v[26:27], v[0:3] +; CHECK-NEXT: v_mov_b64_e32 v[28:29], s[0:1] ; CHECK-NEXT: s_mov_b32 s0, 0x7e007e00 ; CHECK-NEXT: s_mov_b32 s1, s0 -; CHECK-NEXT: v_accvgpr_write_b32 a0, s0 -; CHECK-NEXT: v_accvgpr_write_b32 a1, s1 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[28:29], v[28:29], v[4:7] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[30:31], v[4:7] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[12:15], v[28:29], a[0:1], v[4:7] -; CHECK-NEXT: s_nop 2 -; CHECK-NEXT: v_mov_b32_e32 v4, 0x7fc00000 -; CHECK-NEXT: v_mov_b32_e32 v5, v4 -; CHECK-NEXT: v_mov_b32_e32 v6, v4 -; CHECK-NEXT: v_mov_b32_e32 v7, v4 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[28:29], v[8:11] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[4:7], v[26:27], v[26:27], v[4:7] +; CHECK-NEXT: v_mov_b64_e32 v[30:31], s[0:1] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[6:9], v[26:27], v[28:29], v[0:3] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[6:9], v[26:27], v[26:27], v[6:9] +; CHECK-NEXT: s_nop 3 +; CHECK-NEXT: v_cvt_f16_f32_e32 v24, v4 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[12:15], v[26:27], v[30:31], v[0:3] ; CHECK-NEXT: s_nop 0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[16:19], v[28:29], v[28:29], v[4:7] -; CHECK-NEXT: ;;#ASMSTART -; CHECK-NEXT: ; def v[4:7] -; CHECK-NEXT: ;;#ASMEND -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[16:19], v[28:29], v[28:29], v[16:19] -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[24:27], v[28:29], v[30:31], v[4:7] -; CHECK-NEXT: s_nop 5 -; CHECK-NEXT: v_cvt_f16_f32_e32 v17, v8 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[28:29], v[12:15] -; CHECK-NEXT: s_nop 2 -; CHECK-NEXT: v_mov_b64_e32 v[12:13], 0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[28:29], v[28:29], v[0:3] -; CHECK-NEXT: global_store_short v[12:13], v17, off +; CHECK-NEXT: v_mov_b32_e32 v8, 0x7fc00000 +; CHECK-NEXT: v_mov_b32_e32 v9, v8 +; CHECK-NEXT: v_mov_b32_e32 v10, v8 +; CHECK-NEXT: v_mov_b32_e32 v11, v8 +; CHECK-NEXT: v_cvt_f16_f32_e32 v2, v6 +; CHECK-NEXT: v_mov_b64_e32 v[0:1], 0 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[26:27], v[26:27], v[8:11] +; CHECK-NEXT: global_store_short v[0:1], v2, off ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: v_cvt_f16_f32_e32 v9, v16 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[20:23], v[28:29], v[28:29], v[4:7] -; CHECK-NEXT: global_store_short v[12:13], v9, off -; CHECK-NEXT: v_cvt_f16_f32_e32 v1, v8 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[8:11], v[28:29], v[28:29], v[24:27] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[2:5], v[26:27], v[28:29], v[16:19] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[6:9], v[26:27], v[26:27], v[8:11] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[20:23], v[26:27], v[26:27], v[16:19] +; CHECK-NEXT: s_nop 5 +; CHECK-NEXT: v_cvt_f16_f32_e32 v10, v6 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[6:9], v[26:27], v[26:27], v[12:15] +; CHECK-NEXT: global_store_short v[0:1], v10, off +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[2:5], v[26:27], v[26:27], v[2:5] ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: v_cvt_f16_f32_e32 v14, v0 -; CHECK-NEXT: global_store_short v[12:13], v1, off -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[4:7], v[28:29], v[28:29], v[20:23] +; CHECK-NEXT: s_nop 1 +; CHECK-NEXT: v_cvt_f16_f32_e32 v6, v6 +; CHECK-NEXT: global_store_short v[0:1], v6, off +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[16:19], v[26:27], v[26:27], v[20:23] ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: global_store_short v[12:13], v14, off +; CHECK-NEXT: global_store_short v[0:1], v24, off ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], v[30:31], v[28:29], v[8:11] +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[2:5], v[28:29], v[26:27], v[2:5] ; CHECK-NEXT: s_nop 6 -; CHECK-NEXT: v_cvt_f16_f32_e32 v8, v0 -; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[0:3], a[0:1], v[28:29], v[4:7] -; CHECK-NEXT: global_store_short v[12:13], v8, off +; CHECK-NEXT: v_cvt_f16_f32_e32 v6, v2 +; CHECK-NEXT: v_mfma_f32_16x16x16_f16 v[2:5], v[30:31], v[26:27], v[16:19] +; CHECK-NEXT: global_store_short v[0:1], v6, off ; CHECK-NEXT: buffer_wbl2 sc0 sc1 ; CHECK-NEXT: s_waitcnt vmcnt(0) ; CHECK-NEXT: buffer_inv sc0 sc1 ; CHECK-NEXT: s_nop 2 -; CHECK-NEXT: v_cvt_f16_f32_e32 v0, v0 -; CHECK-NEXT: global_store_short v[12:13], v0, off +; CHECK-NEXT: v_cvt_f16_f32_e32 v2, v2 +; CHECK-NEXT: global_store_short v[0:1], v2, off ; CHECK-NEXT: s_endpgm entry: %k0 = call <4 x float> asm sideeffect "; def $0", "=s"() @@ -546,100 +545,14 @@ define void @test_rewrite_mfma_subreg_insert2(double %arg0, double %arg1, ptr ad define amdgpu_kernel void @test_rewrite_mfma_direct_copy_from_agpr_class(ptr addrspace(1) %arg0, ptr addrspace(1) %arg1) #0 { ; CHECK-LABEL: test_rewrite_mfma_direct_copy_from_agpr_class: ; CHECK: ; %bb.0: +; CHECK-NEXT: v_accvgpr_write_b32 a34, 2.0 +; CHECK-NEXT: v_and_b32_e32 v0, 0x3ff, v0 +; CHECK-NEXT: v_lshlrev_b32_e32 v0, 7, v0 ; CHECK-NEXT: ;;#ASMSTART ; CHECK-NEXT: ; def a[0:31] ; CHECK-NEXT: ;;#ASMEND ; CHECK-NEXT: v_accvgpr_write_b32 a32, v0 -; CHECK-NEXT: v_accvgpr_read_b32 v63, a31 -; CHECK-NEXT: v_accvgpr_read_b32 v62, a30 -; CHECK-NEXT: v_accvgpr_read_b32 v61, a29 -; CHECK-NEXT: v_accvgpr_read_b32 v60, a28 -; CHECK-NEXT: v_accvgpr_read_b32 v59, a27 -; CHECK-NEXT: v_accvgpr_read_b32 v58, a26 -; CHECK-NEXT: v_accvgpr_read_b32 v57, a25 -; CHECK-NEXT: v_accvgpr_read_b32 v56, a24 -; CHECK-NEXT: v_accvgpr_read_b32 v55, a23 -; CHECK-NEXT: v_accvgpr_read_b32 v54, a22 -; CHECK-NEXT: v_accvgpr_read_b32 v53, a21 -; CHECK-NEXT: v_accvgpr_read_b32 v52, a20 -; CHECK-NEXT: v_accvgpr_read_b32 v51, a19 -; CHECK-NEXT: v_accvgpr_read_b32 v50, a18 -; CHECK-NEXT: v_accvgpr_read_b32 v49, a17 -; CHECK-NEXT: v_accvgpr_read_b32 v48, a16 -; CHECK-NEXT: v_accvgpr_read_b32 v47, a15 -; CHECK-NEXT: v_accvgpr_read_b32 v46, a14 -; CHECK-NEXT: v_accvgpr_read_b32 v45, a13 -; CHECK-NEXT: v_accvgpr_read_b32 v44, a12 -; CHECK-NEXT: v_accvgpr_read_b32 v43, a11 -; CHECK-NEXT: v_accvgpr_read_b32 v42, a10 -; CHECK-NEXT: v_accvgpr_read_b32 v41, a9 -; CHECK-NEXT: v_accvgpr_read_b32 v40, a8 -; CHECK-NEXT: v_accvgpr_read_b32 v39, a7 -; CHECK-NEXT: v_accvgpr_read_b32 v38, a6 -; CHECK-NEXT: v_accvgpr_read_b32 v37, a5 -; CHECK-NEXT: v_accvgpr_read_b32 v36, a4 -; CHECK-NEXT: v_accvgpr_read_b32 v35, a3 -; CHECK-NEXT: v_accvgpr_read_b32 v34, a2 -; CHECK-NEXT: v_accvgpr_read_b32 v33, a1 -; CHECK-NEXT: v_accvgpr_read_b32 v32, a0 -; CHECK-NEXT: v_accvgpr_write_b32 a0, 2.0 -; CHECK-NEXT: v_accvgpr_write_b32 a1, 4.0 -; CHECK-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 -; CHECK-NEXT: s_nop 0 -; CHECK-NEXT: v_mfma_f32_32x32x1_2b_f32 v[0:31], a0, a1, v[32:63] -; CHECK-NEXT: v_accvgpr_write_b32 a0, v32 -; CHECK-NEXT: v_accvgpr_write_b32 a1, v33 -; CHECK-NEXT: v_accvgpr_write_b32 a2, v34 -; CHECK-NEXT: v_accvgpr_write_b32 a3, v35 -; CHECK-NEXT: v_accvgpr_write_b32 a4, v36 -; CHECK-NEXT: v_accvgpr_write_b32 a5, v37 -; CHECK-NEXT: v_accvgpr_write_b32 a6, v38 -; CHECK-NEXT: v_accvgpr_write_b32 a7, v39 -; CHECK-NEXT: v_accvgpr_write_b32 a8, v40 -; CHECK-NEXT: v_accvgpr_write_b32 a9, v41 -; CHECK-NEXT: v_accvgpr_write_b32 a10, v42 -; CHECK-NEXT: v_accvgpr_write_b32 a11, v43 -; CHECK-NEXT: v_accvgpr_write_b32 a12, v44 -; CHECK-NEXT: v_accvgpr_write_b32 a13, v45 -; CHECK-NEXT: v_accvgpr_write_b32 a14, v46 -; CHECK-NEXT: v_accvgpr_write_b32 a15, v47 -; CHECK-NEXT: v_accvgpr_write_b32 a16, v48 -; CHECK-NEXT: v_accvgpr_write_b32 a17, v49 -; CHECK-NEXT: v_accvgpr_write_b32 a18, v50 -; CHECK-NEXT: v_accvgpr_write_b32 a19, v51 -; CHECK-NEXT: v_accvgpr_write_b32 a20, v52 -; CHECK-NEXT: v_accvgpr_write_b32 a21, v53 -; CHECK-NEXT: v_accvgpr_write_b32 a22, v54 -; CHECK-NEXT: v_accvgpr_write_b32 a23, v55 -; CHECK-NEXT: v_accvgpr_write_b32 a24, v56 -; CHECK-NEXT: v_accvgpr_write_b32 a25, v57 -; CHECK-NEXT: v_accvgpr_write_b32 a26, v58 -; CHECK-NEXT: v_accvgpr_write_b32 a27, v59 -; CHECK-NEXT: v_accvgpr_write_b32 a28, v60 -; CHECK-NEXT: v_accvgpr_write_b32 a29, v61 -; CHECK-NEXT: v_accvgpr_write_b32 a30, v62 -; CHECK-NEXT: v_accvgpr_write_b32 a31, v63 -; CHECK-NEXT: v_mov_b32_e32 v33, 0x41000000 -; CHECK-NEXT: v_mov_b32_e32 v34, 0x41800000 -; CHECK-NEXT: v_accvgpr_read_b32 v32, a32 -; CHECK-NEXT: v_and_b32_e32 v32, 0x3ff, v32 -; CHECK-NEXT: v_mfma_f32_32x32x1_2b_f32 a[0:31], v33, v34, a[0:31] -; CHECK-NEXT: v_lshlrev_b32_e32 v32, 7, v32 -; CHECK-NEXT: s_waitcnt lgkmcnt(0) -; CHECK-NEXT: global_store_dwordx4 v32, v[28:31], s[0:1] offset:112 -; CHECK-NEXT: global_store_dwordx4 v32, v[24:27], s[0:1] offset:96 -; CHECK-NEXT: global_store_dwordx4 v32, v[20:23], s[0:1] offset:80 -; CHECK-NEXT: global_store_dwordx4 v32, v[16:19], s[0:1] offset:64 -; CHECK-NEXT: global_store_dwordx4 v32, v[12:15], s[0:1] offset:48 -; CHECK-NEXT: global_store_dwordx4 v32, v[8:11], s[0:1] offset:32 -; CHECK-NEXT: global_store_dwordx4 v32, v[4:7], s[0:1] offset:16 -; CHECK-NEXT: global_store_dwordx4 v32, v[0:3], s[0:1] -; CHECK-NEXT: s_nop 7 ; CHECK-NEXT: v_accvgpr_read_b32 v0, a0 -; CHECK-NEXT: v_accvgpr_read_b32 v24, a24 -; CHECK-NEXT: v_accvgpr_read_b32 v25, a25 -; CHECK-NEXT: v_accvgpr_read_b32 v26, a26 -; CHECK-NEXT: v_accvgpr_read_b32 v27, a27 ; CHECK-NEXT: v_accvgpr_read_b32 v1, a1 ; CHECK-NEXT: v_accvgpr_read_b32 v2, a2 ; CHECK-NEXT: v_accvgpr_read_b32 v3, a3 @@ -663,18 +576,60 @@ define amdgpu_kernel void @test_rewrite_mfma_direct_copy_from_agpr_class(ptr add ; CHECK-NEXT: v_accvgpr_read_b32 v21, a21 ; CHECK-NEXT: v_accvgpr_read_b32 v22, a22 ; CHECK-NEXT: v_accvgpr_read_b32 v23, a23 +; CHECK-NEXT: v_accvgpr_read_b32 v24, a24 +; CHECK-NEXT: v_accvgpr_read_b32 v25, a25 +; CHECK-NEXT: v_accvgpr_read_b32 v26, a26 +; CHECK-NEXT: v_accvgpr_read_b32 v27, a27 ; CHECK-NEXT: v_accvgpr_read_b32 v28, a28 ; CHECK-NEXT: v_accvgpr_read_b32 v29, a29 ; CHECK-NEXT: v_accvgpr_read_b32 v30, a30 ; CHECK-NEXT: v_accvgpr_read_b32 v31, a31 -; CHECK-NEXT: global_store_dwordx4 v32, v[24:27], s[2:3] offset:96 -; CHECK-NEXT: global_store_dwordx4 v32, v[28:31], s[2:3] offset:112 -; CHECK-NEXT: global_store_dwordx4 v32, v[16:19], s[2:3] offset:64 -; CHECK-NEXT: global_store_dwordx4 v32, v[20:23], s[2:3] offset:80 -; CHECK-NEXT: global_store_dwordx4 v32, v[8:11], s[2:3] offset:32 -; CHECK-NEXT: global_store_dwordx4 v32, v[12:15], s[2:3] offset:48 -; CHECK-NEXT: global_store_dwordx4 v32, v[0:3], s[2:3] -; CHECK-NEXT: global_store_dwordx4 v32, v[4:7], s[2:3] offset:16 +; CHECK-NEXT: v_accvgpr_write_b32 a33, 4.0 +; CHECK-NEXT: s_load_dwordx4 s[0:3], s[4:5], 0x0 +; CHECK-NEXT: s_nop 0 +; CHECK-NEXT: v_mfma_f32_32x32x1_2b_f32 v[32:63], a34, a33, v[0:31] +; CHECK-NEXT: v_mov_b32_e32 v1, 0x41000000 +; CHECK-NEXT: v_accvgpr_read_b32 v0, a32 +; CHECK-NEXT: s_nop 15 +; CHECK-NEXT: v_mov_b64_e32 v[2:3], v[32:33] +; CHECK-NEXT: v_mov_b64_e32 v[4:5], v[34:35] +; CHECK-NEXT: v_mov_b64_e32 v[6:7], v[36:37] +; CHECK-NEXT: v_mov_b64_e32 v[8:9], v[38:39] +; CHECK-NEXT: v_mov_b64_e32 v[10:11], v[40:41] +; CHECK-NEXT: v_mov_b64_e32 v[12:13], v[42:43] +; CHECK-NEXT: v_mov_b64_e32 v[14:15], v[44:45] +; CHECK-NEXT: v_mov_b64_e32 v[16:17], v[46:47] +; CHECK-NEXT: v_mov_b64_e32 v[18:19], v[48:49] +; CHECK-NEXT: v_mov_b64_e32 v[20:21], v[50:51] +; CHECK-NEXT: v_mov_b64_e32 v[22:23], v[52:53] +; CHECK-NEXT: v_mov_b64_e32 v[24:25], v[54:55] +; CHECK-NEXT: v_mov_b64_e32 v[26:27], v[56:57] +; CHECK-NEXT: v_mov_b64_e32 v[28:29], v[58:59] +; CHECK-NEXT: v_mov_b64_e32 v[30:31], v[60:61] +; CHECK-NEXT: v_mov_b64_e32 v[32:33], v[62:63] +; CHECK-NEXT: s_waitcnt lgkmcnt(0) +; CHECK-NEXT: global_store_dwordx4 v0, v[30:33], s[0:1] offset:112 +; CHECK-NEXT: global_store_dwordx4 v0, v[26:29], s[0:1] offset:96 +; CHECK-NEXT: global_store_dwordx4 v0, v[22:25], s[0:1] offset:80 +; CHECK-NEXT: global_store_dwordx4 v0, v[18:21], s[0:1] offset:64 +; CHECK-NEXT: global_store_dwordx4 v0, v[14:17], s[0:1] offset:48 +; CHECK-NEXT: global_store_dwordx4 v0, v[10:13], s[0:1] offset:32 +; CHECK-NEXT: global_store_dwordx4 v0, v[6:9], s[0:1] offset:16 +; CHECK-NEXT: global_store_dwordx4 v0, v[2:5], s[0:1] +; CHECK-NEXT: s_nop 1 +; CHECK-NEXT: v_mov_b32_e32 v2, 0x41800000 +; CHECK-NEXT: s_nop 1 +; CHECK-NEXT: v_mfma_f32_32x32x1_2b_f32 a[0:31], v1, v2, a[0:31] +; CHECK-NEXT: s_nop 15 +; CHECK-NEXT: s_nop 1 +; CHECK-NEXT: global_store_dwordx4 v0, a[24:27], s[2:3] offset:96 +; CHECK-NEXT: global_store_dwordx4 v0, a[28:31], s[2:3] offset:112 +; CHECK-NEXT: global_store_dwordx4 v0, a[16:19], s[2:3] offset:64 +; CHECK-NEXT: global_store_dwordx4 v0, a[20:23], s[2:3] offset:80 +; CHECK-NEXT: global_store_dwordx4 v0, a[8:11], s[2:3] offset:32 +; CHECK-NEXT: global_store_dwordx4 v0, a[12:15], s[2:3] offset:48 +; CHECK-NEXT: global_store_dwordx4 v0, a[0:3], s[2:3] +; CHECK-NEXT: global_store_dwordx4 v0, a[4:7], s[2:3] offset:16 ; CHECK-NEXT: s_endpgm %src2 = call <32 x float> asm sideeffect "; def $0", "=a"() %mai0 = call <32 x float> @llvm.amdgcn.mfma.f32.32x32x1f32(float 2.0, float 4.0, <32 x float> %src2, i32 0, i32 0, i32 0) diff --git a/llvm/test/CodeGen/AMDGPU/schedule-pending-queue.mir b/llvm/test/CodeGen/AMDGPU/schedule-pending-queue.mir new file mode 100644 index 0000000..33b2f69 --- /dev/null +++ b/llvm/test/CodeGen/AMDGPU/schedule-pending-queue.mir @@ -0,0 +1,32 @@ +# RUN: llc -march=amdgcn -mcpu=gfx908 -run-pass machine-scheduler --misched-prera-direction=topdown -verify-machineinstrs %s -o - -debug-only=machine-scheduler 2>&1 | FileCheck %s +# REQUIRES: asserts + +# Check that cycle counts are consistent with hazards. + +# CHECK: Cycle: 3 TopQ.A +# CHECK: hazard: SU(6) HWXDL[0]=9c, is later than CurrCycle = 3c +# CHECK-NOT: Cycle: 9 TopQ.A +# CHECK: Cycle: 83 TopQ.A +# CHECK: Checking pending node SU(6) +# CHECK: Move SU(6) into Available Q + +--- +name: pending_queue_ready_cycle +tracksRegLiveness: true +body: | + bb.0: + liveins: $sgpr4_sgpr5 + + %2:sgpr_128 = IMPLICIT_DEF + %14:vgpr_32 = IMPLICIT_DEF + %15:vgpr_32 = IMPLICIT_DEF + %18:areg_512 = IMPLICIT_DEF + %18:areg_512 = V_MFMA_F32_16X16X1F32_mac_e64 %15, %14, %18, 0, 0, 0, implicit $mode, implicit $exec + %5:vreg_128 = BUFFER_LOAD_DWORDX4_OFFSET %2, 0, 0, 0, 0, implicit $exec + %18:areg_512 = V_MFMA_F32_16X16X1F32_mac_e64 %15, %14, %18, 0, 0, 0, implicit $mode, implicit $exec + undef %84.sub0:vreg_128_align2 = V_ADD_U32_e32 %5.sub0, %14, implicit $exec + %7:vreg_512 = COPY %18 + SCHED_BARRIER 0 + S_NOP 0, implicit %18, implicit %7, implicit %84 + S_ENDPGM 0 +... diff --git a/llvm/test/CodeGen/DirectX/CBufferLoadLegacy-errors.ll b/llvm/test/CodeGen/DirectX/CBufferLoadLegacy-errors.ll index 71dcf11..196560f 100644 --- a/llvm/test/CodeGen/DirectX/CBufferLoadLegacy-errors.ll +++ b/llvm/test/CodeGen/DirectX/CBufferLoadLegacy-errors.ll @@ -11,11 +11,11 @@ declare void @f16_user(half) ; CHECK-SAME: in function four64 ; CHECK-SAME: Type mismatch between intrinsic and DXIL op define void @four64() "hlsl.export" { - %buffer = call target("dx.CBuffer", target("dx.Layout", {double}, 8, 0)) + %buffer = call target("dx.CBuffer", <{ double }>) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) %load = call {double, double, double, double} @llvm.dx.resource.load.cbufferrow.4( - target("dx.CBuffer", target("dx.Layout", {double}, 8, 0)) %buffer, + target("dx.CBuffer", <{ double }>) %buffer, i32 0) %data = extractvalue {double, double, double, double} %load, 0 @@ -28,11 +28,11 @@ define void @four64() "hlsl.export" { ; CHECK-SAME: in function two32 ; CHECK-SAME: Type mismatch between intrinsic and DXIL op define void @two32() "hlsl.export" { - %buffer = call target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) + %buffer = call target("dx.CBuffer", <{ float }>) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) %load = call {float, float} @llvm.dx.resource.load.cbufferrow.2( - target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) %buffer, + target("dx.CBuffer", <{ float }>) %buffer, i32 0) %data = extractvalue {float, float} %load, 0 @@ -41,5 +41,5 @@ define void @two32() "hlsl.export" { ret void } -declare { double, double, double, double } @llvm.dx.resource.load.cbufferrow.4.f64.f64.f64.f64.tdx.CBuffer_tdx.Layout_sl_f64s_8_0tt(target("dx.CBuffer", target("dx.Layout", { double }, 8, 0)), i32) -declare { float, float } @llvm.dx.resource.load.cbufferrow.2.f32.f32.tdx.CBuffer_tdx.Layout_sl_f32s_4_0tt(target("dx.CBuffer", target("dx.Layout", { float }, 4, 0)), i32) +declare { double, double, double, double } @llvm.dx.resource.load.cbufferrow.4.f64.f64.f64.f64.tdx.CBuffer_sl_f64st(target("dx.CBuffer", <{ double }>), i32) +declare { float, float } @llvm.dx.resource.load.cbufferrow.2.f32.f32.tdx.CBuffer_sl_f32st(target("dx.CBuffer", <{ float }>), i32) diff --git a/llvm/test/CodeGen/DirectX/CBufferLoadLegacy.ll b/llvm/test/CodeGen/DirectX/CBufferLoadLegacy.ll index d690651..dd40aa8 100644 --- a/llvm/test/CodeGen/DirectX/CBufferLoadLegacy.ll +++ b/llvm/test/CodeGen/DirectX/CBufferLoadLegacy.ll @@ -8,12 +8,12 @@ declare void @f16_user(half) ; CHECK-LABEL: define void @loadf32 define void @loadf32() { - %buffer = call target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) + %buffer = call target("dx.CBuffer", <{ float }>) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) ; CHECK: [[DATA:%.*]] = call %dx.types.CBufRet.f32 @dx.op.cbufferLoadLegacy.f32(i32 59, %dx.types.Handle %{{.*}}, i32 0) %load = call {float, float, float, float} @llvm.dx.resource.load.cbufferrow.4( - target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) %buffer, + target("dx.CBuffer", <{ float }>) %buffer, i32 0) %data = extractvalue {float, float, float, float} %load, 0 @@ -27,12 +27,12 @@ define void @loadf32() { ; CHECK-LABEL: define void @loadf64 define void @loadf64() { %buffer = call - target("dx.CBuffer", target("dx.Layout", {double, double, double, double}, 64, 0, 8, 16, 24)) + target("dx.CBuffer", <{ <4 x double> }>) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) ; CHECK: [[DATA:%.*]] = call %dx.types.CBufRet.f64 @dx.op.cbufferLoadLegacy.f64(i32 59, %dx.types.Handle %{{.*}}, i32 1) %load = call {double, double} @llvm.dx.resource.load.cbufferrow.2( - target("dx.CBuffer", target("dx.Layout", {double, double, double, double}, 64, 0, 8, 16, 24)) %buffer, + target("dx.CBuffer", <{ <4 x double> }>) %buffer, i32 1) %data = extractvalue {double, double} %load, 1 @@ -46,12 +46,12 @@ define void @loadf64() { ; CHECK-LABEL: define void @loadf16 define void @loadf16() { %buffer = call - target("dx.CBuffer", target("dx.Layout", {half}, 2, 0)) + target("dx.CBuffer", <{ half }>) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) ; CHECK: [[DATA:%.*]] = call %dx.types.CBufRet.f16.8 @dx.op.cbufferLoadLegacy.f16(i32 59, %dx.types.Handle %{{.*}}, i32 0) %load = call {half, half, half, half, half, half, half, half} @llvm.dx.resource.load.cbufferrow.8( - target("dx.CBuffer", target("dx.Layout", {half}, 2, 0)) %buffer, + target("dx.CBuffer", <{ half }>) %buffer, i32 0) %data = extractvalue {half, half, half, half, half, half, half, half} %load, 0 diff --git a/llvm/test/CodeGen/DirectX/ContainerData/PSVResources-order.ll b/llvm/test/CodeGen/DirectX/ContainerData/PSVResources-order.ll index bcf82a6..5cd67be 100644 --- a/llvm/test/CodeGen/DirectX/ContainerData/PSVResources-order.ll +++ b/llvm/test/CodeGen/DirectX/ContainerData/PSVResources-order.ll @@ -18,7 +18,7 @@ define void @main() #0 { %srv0 = call target("dx.RawBuffer", i8, 0, 0) @llvm.dx.resource.handlefrombinding.tdx.RawBuffer_i8_0_0t( i32 1, i32 8, i32 1, i32 0, ptr null) - %cbuf = call target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) + %cbuf = call target("dx.CBuffer", <{ float }>) @llvm.dx.resource.handlefrombinding(i32 3, i32 2, i32 1, i32 0, ptr null) ret void } diff --git a/llvm/test/CodeGen/DirectX/ContainerData/PSVResources.ll b/llvm/test/CodeGen/DirectX/ContainerData/PSVResources.ll index 70224fc..d792078 100644 --- a/llvm/test/CodeGen/DirectX/ContainerData/PSVResources.ll +++ b/llvm/test/CodeGen/DirectX/ContainerData/PSVResources.ll @@ -14,7 +14,7 @@ define void @main() #0 { ; CHECK: Kind: CBuffer ; CHECK: Flags: ; CHECK: UsedByAtomic64: false - %cbuf = call target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) + %cbuf = call target("dx.CBuffer", <{ float }>) @llvm.dx.resource.handlefrombinding(i32 3, i32 2, i32 1, i32 0, ptr null) ; ByteAddressBuffer Buf : register(t8, space1) diff --git a/llvm/test/CodeGen/DirectX/CreateHandleFromBinding.ll b/llvm/test/CodeGen/DirectX/CreateHandleFromBinding.ll index 38f2de2..671fcef 100644 --- a/llvm/test/CodeGen/DirectX/CreateHandleFromBinding.ll +++ b/llvm/test/CodeGen/DirectX/CreateHandleFromBinding.ll @@ -72,7 +72,7 @@ define void @test_bindings() { ; CHECK: call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[BUF5]], %dx.types.ResourceProperties { i32 10, i32 1033 }) #[[#ATTR]] ; cbuffer cb0 : register(b0) { int4 i; float4 f; } - %cb0 = call target("dx.CBuffer", target("dx.Layout", {<4 x i32>, <4 x float>}, 32, 0, 16)) + %cb0 = call target("dx.CBuffer", <{ <4 x i32>, <4 x float> }>) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) ; CHECK: [[BUF6:%.*]] = call %dx.types.Handle @dx.op.createHandleFromBinding(i32 217, %dx.types.ResBind { i32 0, i32 0, i32 0, i8 2 }, i32 0, i1 false) #[[#ATTR]] ; CHECK: call %dx.types.Handle @dx.op.annotateHandle(i32 216, %dx.types.Handle [[BUF6]], %dx.types.ResourceProperties { i32 13, i32 32 }) #[[#ATTR]] diff --git a/llvm/test/CodeGen/DirectX/ForwardHandleAccesses/cbuffer-access.ll b/llvm/test/CodeGen/DirectX/ForwardHandleAccesses/cbuffer-access.ll index 26b157f..d674863 100644 --- a/llvm/test/CodeGen/DirectX/ForwardHandleAccesses/cbuffer-access.ll +++ b/llvm/test/CodeGen/DirectX/ForwardHandleAccesses/cbuffer-access.ll @@ -4,27 +4,27 @@ %__cblayout_CB2 = type <{ float }> %struct.Scalars = type { float, i32, i32 } -@CB.cb = local_unnamed_addr global target("dx.CBuffer", target("dx.Layout", %__cblayout_CB, 12, 0, 4, 8)) poison -@CB2.cb = local_unnamed_addr global target("dx.CBuffer", target("dx.Layout", %__cblayout_CB2, 4, 0)) poison +@CB.cb = local_unnamed_addr global target("dx.CBuffer", %__cblayout_CB) poison +@CB2.cb = local_unnamed_addr global target("dx.CBuffer", %__cblayout_CB2) poison define void @main() local_unnamed_addr #1 { entry: ; CHECK: [[CB:%.*]] = tail call target({{.*}}) @llvm.dx.resource.handlefrombinding - %h = tail call target("dx.CBuffer", target("dx.Layout", %__cblayout_CB, 12, 0, 4, 8)) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) - store target("dx.CBuffer", target("dx.Layout", %__cblayout_CB, 12, 0, 4, 8)) %h, ptr @CB.cb, align 4 + %h = tail call target("dx.CBuffer", %__cblayout_CB) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) + store target("dx.CBuffer", %__cblayout_CB) %h, ptr @CB.cb, align 4 %_ZL3Out_h.i.i = tail call target("dx.RawBuffer", %struct.Scalars, 1, 0) @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr null) ; CHECK-NOT: load target({{.*}}), ptr @CB.cb - %cb = load target("dx.CBuffer", target("dx.Layout", %__cblayout_CB, 12, 0, 4, 8)), ptr @CB.cb, align 4 + %cb = load target("dx.CBuffer", %__cblayout_CB), ptr @CB.cb, align 4 ; CHECK: call { float, float, float, float } @llvm.dx.resource.load.cbufferrow.4.{{.*}}(target({{.*}}) [[CB]], i32 0) - %0 = call { float, float, float, float } @llvm.dx.resource.load.cbufferrow.4(target("dx.CBuffer", target("dx.Layout", %__cblayout_CB, 12, 0, 4, 8)) %cb, i32 0) + %0 = call { float, float, float, float } @llvm.dx.resource.load.cbufferrow.4(target("dx.CBuffer", %__cblayout_CB) %cb, i32 0) %1 = extractvalue { float, float, float, float } %0, 0 call void @llvm.dx.resource.store.rawbuffer(target("dx.RawBuffer", %struct.Scalars, 1, 0) %_ZL3Out_h.i.i, i32 0, i32 0, float %1) - + ; CHECK: [[CB2:%.*]] = tail call target({{.*}}) @llvm.dx.resource.handlefromimplicitbinding - %h2 = tail call target("dx.CBuffer", target("dx.Layout", %__cblayout_CB2, 4, 0)) @llvm.dx.resource.handlefromimplicitbinding(i32 100, i32 0, i32 1, i32 0, ptr null) - store target("dx.CBuffer", target("dx.Layout", %__cblayout_CB2, 4, 0)) %h2, ptr @CB2.cb, align 4 + %h2 = tail call target("dx.CBuffer", %__cblayout_CB2) @llvm.dx.resource.handlefromimplicitbinding(i32 100, i32 0, i32 1, i32 0, ptr null) + store target("dx.CBuffer", %__cblayout_CB2) %h2, ptr @CB2.cb, align 4 ; CHECK-NOT: load target({{.*}}), ptr @CB2.cb - %cb2 = load target("dx.CBuffer", target("dx.Layout", %__cblayout_CB2, 4, 0)), ptr @CB2.cb, align 4 + %cb2 = load target("dx.CBuffer", %__cblayout_CB2), ptr @CB2.cb, align 4 ret void } diff --git a/llvm/test/CodeGen/DirectX/Metadata/cbuffer_metadata.ll b/llvm/test/CodeGen/DirectX/Metadata/cbuffer-layouttype.ll index f1d28e2..85952c9 100644 --- a/llvm/test/CodeGen/DirectX/Metadata/cbuffer_metadata.ll +++ b/llvm/test/CodeGen/DirectX/Metadata/cbuffer-layouttype.ll @@ -1,3 +1,6 @@ +; TODO: Remove this test once we've updated the frontend to use explicit +; padding. The cbuffer-metadata.ll test covers the newer logic. + ; RUN: opt -S -dxil-translate-metadata < %s | FileCheck %s ; RUN: opt -S --passes="dxil-pretty-printer" < %s 2>&1 | FileCheck %s --check-prefix=PRINT ; RUN: llc %s --filetype=asm -o - < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,PRINT diff --git a/llvm/test/CodeGen/DirectX/Metadata/cbuffer-metadata.ll b/llvm/test/CodeGen/DirectX/Metadata/cbuffer-metadata.ll new file mode 100644 index 0000000..6b90e17 --- /dev/null +++ b/llvm/test/CodeGen/DirectX/Metadata/cbuffer-metadata.ll @@ -0,0 +1,89 @@ +; RUN: opt -S -dxil-translate-metadata < %s | FileCheck %s +; RUN: opt -S --passes="dxil-pretty-printer" < %s 2>&1 | FileCheck %s --check-prefix=PRINT +; RUN: llc %s --filetype=asm -o - < %s 2>&1 | FileCheck %s --check-prefixes=CHECK,PRINT + +target triple = "dxil-pc-shadermodel6.6-compute" + +%__cblayout_CB1 = type <{ float, i32, double, <2 x i32> }> +@CB1.cb = global target("dx.CBuffer", %__cblayout_CB1) poison +@CB1.str = private unnamed_addr constant [4 x i8] c"CB1\00", align 1 + +%__cblayout_CB2 = type <{ float, target("dx.Padding", 4), double, float, half, i16, i64, i32 }> +@CB2.cb = global target("dx.CBuffer", %__cblayout_CB2) poison +@CB2.str = private unnamed_addr constant [4 x i8] c"CB2\00", align 1 + +%__cblayout_MyConstants = type <{ + double, target("dx.Padding", 8), + <3 x float>, float, + <3 x double>, half, target("dx.Padding", 6), + <2 x double>, + float, <3 x half>, <3 x half> +}> +@MyConstants.cb = global target("dx.CBuffer", %__cblayout_MyConstants) poison +@MyConstants.str = private unnamed_addr constant [12 x i8] c"MyConstants\00", align 1 + +; PRINT:; Resource Bindings: +; PRINT-NEXT:; +; PRINT-NEXT:; Name Type Format Dim ID HLSL Bind Count +; PRINT-NEXT:; ---- +; PRINT-NEXT:; CB1 cbuffer NA NA CB0 cb0 1 +; PRINT-NEXT:; CB2 cbuffer NA NA CB1 cb1 1 +; PRINT-NEXT:; MyConstants cbuffer NA NA CB2 cb5,space15 1 + +define void @test() #0 { + + ; cbuffer CB1 : register(b0) { + ; float a; + ; int b; + ; double c; + ; int2 d; + ; } + %CB1.cb_h = call target("dx.CBuffer", %__cblayout_CB1) + @llvm.dx.resource.handlefrombinding(i32 0, i32 0, i32 1, i32 0, ptr @CB1.str) + + ; cbuffer CB2 : register(b0) { + ; float a; + ; double b; + ; float c; + ; half d; + ; uint16_t e; + ; int64_t f; + ; int g; + ;} + %CB2.cb_h = call target("dx.CBuffer", %__cblayout_CB2) + @llvm.dx.resource.handlefrombinding(i32 0, i32 1, i32 1, i32 0, ptr @CB2.str) + + ; cbuffer CB3 : register(b5) { + ; double B0; + ; float3 B1; + ; float B2; + ; double3 B3; + ; half B4; + ; double2 B5; + ; float B6; + ; half3 B7; + ; half3 B8; + ; } + %CB3.cb_h = call target("dx.CBuffer", %__cblayout_MyConstants) + @llvm.dx.resource.handlefrombinding(i32 15, i32 5, i32 1, i32 0, ptr @MyConstants.str) + + ret void +} + +attributes #0 = { noinline nounwind "hlsl.shader"="compute" } + +; CHECK: %CBuffer.CB1 = type { { float, i32, double, <2 x i32> } } +; CHECK: %CBuffer.CB2 = type { { float, double, float, half, i16, i64, i32 } } +; CHECK: %CBuffer.MyConstants = type { { double, <3 x float>, float, <3 x double>, half, <2 x double>, float, <3 x half>, <3 x half> } } + +; CHECK: @CB1 = external constant %CBuffer.CB1 +; CHECK: @CB2 = external constant %CBuffer.CB2 +; CHECK: @MyConstants = external constant %CBuffer.MyConstants + +; CHECK: !dx.resources = !{[[ResList:[!][0-9]+]]} + +; CHECK: [[ResList]] = !{null, null, [[CBList:[!][0-9]+]], null} +; CHECK: [[CBList]] = !{![[CB1:[0-9]+]], ![[CB2:[0-9]+]], ![[MYCONSTANTS:[0-9]+]]} +; CHECK: ![[CB1]] = !{i32 0, ptr @CB1, !"CB1", i32 0, i32 0, i32 1, i32 24, null} +; CHECK: ![[CB2]] = !{i32 1, ptr @CB2, !"CB2", i32 0, i32 1, i32 1, i32 36, null} +; CHECK: ![[MYCONSTANTS]] = !{i32 2, ptr @MyConstants, !"MyConstants", i32 15, i32 5, i32 1, i32 96, null} diff --git a/llvm/test/CodeGen/DirectX/Metadata/cbuffer-only.ll b/llvm/test/CodeGen/DirectX/Metadata/cbuffer-only.ll index e2a1c09..0b454c1 100644 --- a/llvm/test/CodeGen/DirectX/Metadata/cbuffer-only.ll +++ b/llvm/test/CodeGen/DirectX/Metadata/cbuffer-only.ll @@ -7,7 +7,7 @@ target triple = "dxil-pc-shadermodel6.6-compute" define void @cbuffer_is_only_binding() { - %cbuf = call target("dx.CBuffer", target("dx.Layout", {float}, 4, 0)) + %cbuf = call target("dx.CBuffer", <{ float }>) @llvm.dx.resource.handlefrombinding(i32 1, i32 8, i32 1, i32 0, ptr null) ; CHECK: %CBuffer = type { float } diff --git a/llvm/test/CodeGen/NVPTX/i32x2-instructions.ll b/llvm/test/CodeGen/NVPTX/i32x2-instructions.ll index 153ca10..72f10ae 100644 --- a/llvm/test/CodeGen/NVPTX/i32x2-instructions.ll +++ b/llvm/test/CodeGen/NVPTX/i32x2-instructions.ll @@ -1141,29 +1141,88 @@ define <2 x i32> @test_select_cc(<2 x i32> %a, <2 x i32> %b, <2 x i32> %c, <2 x ret <2 x i32> %r } -define <2 x i16> @test_trunc_2xi32(<2 x i32> %a) #0 { -; CHECK-NOI32X2-LABEL: test_trunc_2xi32( +define <2 x i16> @test_trunc_2xi32_to_2xi16(<2 x i32> %a) #0 { +; CHECK-NOI32X2-LABEL: test_trunc_2xi32_to_2xi16( ; CHECK-NOI32X2: { ; CHECK-NOI32X2-NEXT: .reg .b32 %r<4>; ; CHECK-NOI32X2-EMPTY: ; CHECK-NOI32X2-NEXT: // %bb.0: -; CHECK-NOI32X2-NEXT: ld.param.v2.b32 {%r1, %r2}, [test_trunc_2xi32_param_0]; +; CHECK-NOI32X2-NEXT: ld.param.v2.b32 {%r1, %r2}, [test_trunc_2xi32_to_2xi16_param_0]; ; CHECK-NOI32X2-NEXT: prmt.b32 %r3, %r1, %r2, 0x5410U; ; CHECK-NOI32X2-NEXT: st.param.b32 [func_retval0], %r3; ; CHECK-NOI32X2-NEXT: ret; ; -; CHECK-I32X2-LABEL: test_trunc_2xi32( +; CHECK-I32X2-LABEL: test_trunc_2xi32_to_2xi16( ; CHECK-I32X2: { +; CHECK-I32X2-NEXT: .reg .b32 %r<4>; ; CHECK-I32X2-NEXT: .reg .b64 %rd<2>; ; CHECK-I32X2-EMPTY: ; CHECK-I32X2-NEXT: // %bb.0: -; CHECK-I32X2-NEXT: ld.param.b64 %rd1, [test_trunc_2xi32_param_0]; -; CHECK-I32X2-NEXT: st.param.b32 [func_retval0], %rd1; +; CHECK-I32X2-NEXT: ld.param.b64 %rd1, [test_trunc_2xi32_to_2xi16_param_0]; +; CHECK-I32X2-NEXT: mov.b64 {%r1, %r2}, %rd1; +; CHECK-I32X2-NEXT: prmt.b32 %r3, %r1, %r2, 0x5410U; +; CHECK-I32X2-NEXT: st.param.b32 [func_retval0], %r3; ; CHECK-I32X2-NEXT: ret; %r = trunc <2 x i32> %a to <2 x i16> ret <2 x i16> %r } +define <2 x i8> @test_trunc_2xi32_to_2xi8(<2 x i32> %a) #0 { +; CHECK-NOI32X2-LABEL: test_trunc_2xi32_to_2xi8( +; CHECK-NOI32X2: { +; CHECK-NOI32X2-NEXT: .reg .b16 %rs<3>; +; CHECK-NOI32X2-NEXT: .reg .b32 %r<3>; +; CHECK-NOI32X2-EMPTY: +; CHECK-NOI32X2-NEXT: // %bb.0: +; CHECK-NOI32X2-NEXT: ld.param.v2.b32 {%r1, %r2}, [test_trunc_2xi32_to_2xi8_param_0]; +; CHECK-NOI32X2-NEXT: cvt.u16.u32 %rs1, %r2; +; CHECK-NOI32X2-NEXT: cvt.u16.u32 %rs2, %r1; +; CHECK-NOI32X2-NEXT: st.param.v2.b8 [func_retval0], {%rs2, %rs1}; +; CHECK-NOI32X2-NEXT: ret; +; +; CHECK-I32X2-LABEL: test_trunc_2xi32_to_2xi8( +; CHECK-I32X2: { +; CHECK-I32X2-NEXT: .reg .b16 %rs<3>; +; CHECK-I32X2-NEXT: .reg .b32 %r<3>; +; CHECK-I32X2-NEXT: .reg .b64 %rd<2>; +; CHECK-I32X2-EMPTY: +; CHECK-I32X2-NEXT: // %bb.0: +; CHECK-I32X2-NEXT: ld.param.b64 %rd1, [test_trunc_2xi32_to_2xi8_param_0]; +; CHECK-I32X2-NEXT: mov.b64 {%r1, %r2}, %rd1; +; CHECK-I32X2-NEXT: cvt.u16.u32 %rs1, %r2; +; CHECK-I32X2-NEXT: cvt.u16.u32 %rs2, %r1; +; CHECK-I32X2-NEXT: st.param.v2.b8 [func_retval0], {%rs2, %rs1}; +; CHECK-I32X2-NEXT: ret; + %r = trunc <2 x i32> %a to <2 x i8> + ret <2 x i8> %r +} + +define <2 x i1> @test_trunc_2xi32_to_2xi1(<2 x i32> %a) #0 { +; CHECK-NOI32X2-LABEL: test_trunc_2xi32_to_2xi1( +; CHECK-NOI32X2: { +; CHECK-NOI32X2-NEXT: .reg .b32 %r<3>; +; CHECK-NOI32X2-EMPTY: +; CHECK-NOI32X2-NEXT: // %bb.0: +; CHECK-NOI32X2-NEXT: ld.param.v2.b32 {%r1, %r2}, [test_trunc_2xi32_to_2xi1_param_0]; +; CHECK-NOI32X2-NEXT: st.param.b8 [func_retval0], %r1; +; CHECK-NOI32X2-NEXT: st.param.b8 [func_retval0+1], %r2; +; CHECK-NOI32X2-NEXT: ret; +; +; CHECK-I32X2-LABEL: test_trunc_2xi32_to_2xi1( +; CHECK-I32X2: { +; CHECK-I32X2-NEXT: .reg .b32 %r<3>; +; CHECK-I32X2-NEXT: .reg .b64 %rd<2>; +; CHECK-I32X2-EMPTY: +; CHECK-I32X2-NEXT: // %bb.0: +; CHECK-I32X2-NEXT: ld.param.b64 %rd1, [test_trunc_2xi32_to_2xi1_param_0]; +; CHECK-I32X2-NEXT: mov.b64 {%r1, %r2}, %rd1; +; CHECK-I32X2-NEXT: st.param.b8 [func_retval0], %r1; +; CHECK-I32X2-NEXT: st.param.b8 [func_retval0+1], %r2; +; CHECK-I32X2-NEXT: ret; + %r = trunc <2 x i32> %a to <2 x i1> + ret <2 x i1> %r +} + define <2 x i32> @test_trunc_2xi64(<2 x i64> %a) #0 { ; CHECK-LABEL: test_trunc_2xi64( ; CHECK: { @@ -1180,14 +1239,14 @@ define <2 x i32> @test_trunc_2xi64(<2 x i64> %a) #0 { ret <2 x i32> %r } -define <2 x i32> @test_zext_2xi32(<2 x i16> %a) #0 { -; CHECK-LABEL: test_zext_2xi32( +define <2 x i32> @test_zext_2xi16_to_2xi32(<2 x i16> %a) #0 { +; CHECK-LABEL: test_zext_2xi16_to_2xi32( ; CHECK: { ; CHECK-NEXT: .reg .b16 %rs<3>; ; CHECK-NEXT: .reg .b32 %r<4>; ; CHECK-EMPTY: ; CHECK-NEXT: // %bb.0: -; CHECK-NEXT: ld.param.b32 %r1, [test_zext_2xi32_param_0]; +; CHECK-NEXT: ld.param.b32 %r1, [test_zext_2xi16_to_2xi32_param_0]; ; CHECK-NEXT: mov.b32 {%rs1, %rs2}, %r1; ; CHECK-NEXT: cvt.u32.u16 %r2, %rs2; ; CHECK-NEXT: cvt.u32.u16 %r3, %rs1; @@ -1197,6 +1256,47 @@ define <2 x i32> @test_zext_2xi32(<2 x i16> %a) #0 { ret <2 x i32> %r } +define <2 x i32> @test_zext_2xi8_to_2xi32(<2 x i8> %a) #0 { +; CHECK-LABEL: test_zext_2xi8_to_2xi32( +; CHECK: { +; CHECK-NEXT: .reg .b16 %rs<3>; +; CHECK-NEXT: .reg .b32 %r<4>; +; CHECK-EMPTY: +; CHECK-NEXT: // %bb.0: +; CHECK-NEXT: ld.param.v2.b8 {%rs1, %rs2}, [test_zext_2xi8_to_2xi32_param_0]; +; CHECK-NEXT: mov.b32 %r1, {%rs1, %rs2}; +; CHECK-NEXT: cvt.u32.u16 %r2, %rs2; +; CHECK-NEXT: cvt.u32.u16 %r3, %rs1; +; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r3, %r2}; +; CHECK-NEXT: ret; + %r = zext <2 x i8> %a to <2 x i32> + ret <2 x i32> %r +} + +define <2 x i32> @test_zext_2xi1_to_2xi32(<2 x i1> %a) #0 { +; CHECK-LABEL: test_zext_2xi1_to_2xi32( +; CHECK: { +; CHECK-NEXT: .reg .pred %p<3>; +; CHECK-NEXT: .reg .b16 %rs<5>; +; CHECK-NEXT: .reg .b32 %r<5>; +; CHECK-EMPTY: +; CHECK-NEXT: // %bb.0: +; CHECK-NEXT: ld.param.b8 %rs1, [test_zext_2xi1_to_2xi32_param_0+1]; +; CHECK-NEXT: and.b16 %rs2, %rs1, 1; +; CHECK-NEXT: setp.ne.b16 %p2, %rs2, 0; +; CHECK-NEXT: ld.param.b8 %rs3, [test_zext_2xi1_to_2xi32_param_0]; +; CHECK-NEXT: and.b16 %rs4, %rs3, 1; +; CHECK-NEXT: setp.ne.b16 %p1, %rs4, 0; +; CHECK-NEXT: cvt.u32.u16 %r1, %rs1; +; CHECK-NEXT: and.b32 %r2, %r1, 1; +; CHECK-NEXT: cvt.u32.u16 %r3, %rs3; +; CHECK-NEXT: and.b32 %r4, %r3, 1; +; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r4, %r2}; +; CHECK-NEXT: ret; + %r = zext <2 x i1> %a to <2 x i32> + ret <2 x i32> %r +} + define <2 x i64> @test_zext_2xi64(<2 x i32> %a) #0 { ; CHECK-NOI32X2-LABEL: test_zext_2xi64( ; CHECK-NOI32X2: { @@ -1566,6 +1666,55 @@ entry: ret void } +define <2 x i32> @test_sext_v2i8_to_v2i32 (<2 x i8> %a) { +; CHECK-LABEL: test_sext_v2i8_to_v2i32( +; CHECK: { +; CHECK-NEXT: .reg .b16 %rs<3>; +; CHECK-NEXT: .reg .b32 %r<6>; +; CHECK-EMPTY: +; CHECK-NEXT: // %bb.0: +; CHECK-NEXT: ld.param.v2.b8 {%rs1, %rs2}, [test_sext_v2i8_to_v2i32_param_0]; +; CHECK-NEXT: mov.b32 %r1, {%rs1, %rs2}; +; CHECK-NEXT: cvt.u32.u16 %r2, %rs2; +; CHECK-NEXT: cvt.s32.s8 %r3, %r2; +; CHECK-NEXT: cvt.u32.u16 %r4, %rs1; +; CHECK-NEXT: cvt.s32.s8 %r5, %r4; +; CHECK-NEXT: st.param.v2.b32 [func_retval0], {%r5, %r3}; +; CHECK-NEXT: ret; + %r = sext <2 x i8> %a to <2 x i32> + ret <2 x i32> %r +} + +define <2 x i32> @test_sext_v2i16_to_v2i32 (<2 x i16> %a) { +; CHECK-NOI32X2-LABEL: test_sext_v2i16_to_v2i32( +; CHECK-NOI32X2: { +; CHECK-NOI32X2-NEXT: .reg .b16 %rs<2>; +; CHECK-NOI32X2-NEXT: .reg .b32 %r<4>; +; CHECK-NOI32X2-EMPTY: +; CHECK-NOI32X2-NEXT: // %bb.0: +; CHECK-NOI32X2-NEXT: ld.param.b32 %r1, [test_sext_v2i16_to_v2i32_param_0]; +; CHECK-NOI32X2-NEXT: cvt.s32.s16 %r2, %r1; +; CHECK-NOI32X2-NEXT: { .reg .b16 tmp; mov.b32 {tmp, %rs1}, %r1; } +; CHECK-NOI32X2-NEXT: cvt.s32.s16 %r3, %rs1; +; CHECK-NOI32X2-NEXT: st.param.v2.b32 [func_retval0], {%r2, %r3}; +; CHECK-NOI32X2-NEXT: ret; +; +; CHECK-I32X2-LABEL: test_sext_v2i16_to_v2i32( +; CHECK-I32X2: { +; CHECK-I32X2-NEXT: .reg .b16 %rs<2>; +; CHECK-I32X2-NEXT: .reg .b32 %r<4>; +; CHECK-I32X2-EMPTY: +; CHECK-I32X2-NEXT: // %bb.0: +; CHECK-I32X2-NEXT: ld.param.b32 %r1, [test_sext_v2i16_to_v2i32_param_0]; +; CHECK-I32X2-NEXT: cvt.s32.s16 %r2, %r1; +; CHECK-I32X2-NEXT: mov.b32 {_, %rs1}, %r1; +; CHECK-I32X2-NEXT: cvt.s32.s16 %r3, %rs1; +; CHECK-I32X2-NEXT: st.param.v2.b32 [func_retval0], {%r2, %r3}; +; CHECK-I32X2-NEXT: ret; + %r = sext <2 x i16> %a to <2 x i32> + ret <2 x i32> %r +} + define <2 x float> @test_uitofp_v2i32(<2 x i32> %a) { ; CHECK-NOI32X2-LABEL: test_uitofp_v2i32( ; CHECK-NOI32X2: { diff --git a/llvm/test/CodeGen/RISCV/attributes.ll b/llvm/test/CodeGen/RISCV/attributes.ll index f3529b1..22c2d81 100644 --- a/llvm/test/CodeGen/RISCV/attributes.ll +++ b/llvm/test/CodeGen/RISCV/attributes.ll @@ -80,6 +80,7 @@ ; RUN: llc -mtriple=riscv32 -mattr=+xwchc %s -o - | FileCheck --check-prefix=RV32XWCHC %s ; RUN: llc -mtriple=riscv32 -mattr=+zaamo %s -o - | FileCheck --check-prefix=RV32ZAAMO %s ; RUN: llc -mtriple=riscv32 -mattr=+zalrsc %s -o - | FileCheck --check-prefix=RV32ZALRSC %s +; RUN: llc -mtriple=riscv32 -mattr=+zaamo,+zalrsc %s -o - | FileCheck --check-prefixes=CHECK,RV32COMBINEINTOA %s ; RUN: llc -mtriple=riscv32 -mattr=+zca %s -o - | FileCheck --check-prefixes=CHECK,RV32ZCA %s ; RUN: llc -mtriple=riscv32 -mattr=+zcb %s -o - | FileCheck --check-prefixes=CHECK,RV32ZCB %s ; RUN: llc -mtriple=riscv32 -mattr=+zcd %s -o - | FileCheck --check-prefixes=CHECK,RV32ZCD %s @@ -227,6 +228,7 @@ ; RUN: llc -mtriple=riscv64 -mattr=+ztso %s -o - | FileCheck --check-prefixes=CHECK,RV64ZTSO %s ; RUN: llc -mtriple=riscv64 -mattr=+zaamo %s -o - | FileCheck --check-prefix=RV64ZAAMO %s ; RUN: llc -mtriple=riscv64 -mattr=+zalrsc %s -o - | FileCheck --check-prefix=RV64ZALRSC %s +; RUN: llc -mtriple=riscv64 -mattr=+zaamo,+zalrsc %s -o - | FileCheck --check-prefixes=CHECK,RV64COMBINEINTOA %s ; RUN: llc -mtriple=riscv64 -mattr=+zca %s -o - | FileCheck --check-prefixes=CHECK,RV64ZCA %s ; RUN: llc -mtriple=riscv64 -mattr=+zcb %s -o - | FileCheck --check-prefixes=CHECK,RV64ZCB %s ; RUN: llc -mtriple=riscv64 -mattr=+zcd %s -o - | FileCheck --check-prefixes=CHECK,RV64ZCD %s @@ -392,6 +394,7 @@ ; RV32XWCHC: .attribute 5, "rv32i2p1_zca1p0_xwchc2p2" ; RV32ZAAMO: .attribute 5, "rv32i2p1_zaamo1p0" ; RV32ZALRSC: .attribute 5, "rv32i2p1_zalrsc1p0" +; RV32COMBINEINTOA: .attribute 5, "rv32i2p1_a2p1_zaamo1p0_zalrsc1p0" ; RV32ZCA: .attribute 5, "rv32i2p1_zca1p0" ; RV32ZCB: .attribute 5, "rv32i2p1_zca1p0_zcb1p0" ; RV32ZCD: .attribute 5, "rv32i2p1_f2p2_d2p2_zicsr2p0_zca1p0_zcd1p0" @@ -537,6 +540,7 @@ ; RV64ZTSO: .attribute 5, "rv64i2p1_ztso1p0" ; RV64ZAAMO: .attribute 5, "rv64i2p1_zaamo1p0" ; RV64ZALRSC: .attribute 5, "rv64i2p1_zalrsc1p0" +; RV64COMBINEINTOA: .attribute 5, "rv64i2p1_a2p1_zaamo1p0_zalrsc1p0" ; RV64ZCA: .attribute 5, "rv64i2p1_zca1p0" ; RV64ZCB: .attribute 5, "rv64i2p1_zca1p0_zcb1p0" ; RV64ZCD: .attribute 5, "rv64i2p1_f2p2_d2p2_zicsr2p0_zca1p0_zcd1p0" diff --git a/llvm/test/CodeGen/RISCV/idiv_large.ll b/llvm/test/CodeGen/RISCV/idiv_large.ll index 9937627..d7b00f6 100644 --- a/llvm/test/CodeGen/RISCV/idiv_large.ll +++ b/llvm/test/CodeGen/RISCV/idiv_large.ll @@ -1,16 +1,2315 @@ -; RUN: llc -mtriple=riscv32 < %s | FileCheck %s -; RUN: llc -mtriple=riscv64 < %s | FileCheck %s +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc -mtriple=riscv32 < %s | FileCheck %s --check-prefix=RV32 +; RUN: llc -mtriple=riscv64 < %s | FileCheck %s --check-prefix=RV64 + +define i64 @udiv_i64(i64 %x, i64 %y) nounwind { +; RV32-LABEL: udiv_i64: +; RV32: # %bb.0: +; RV32-NEXT: addi sp, sp, -16 +; RV32-NEXT: sw ra, 12(sp) # 4-byte Folded Spill +; RV32-NEXT: call __udivdi3 +; RV32-NEXT: lw ra, 12(sp) # 4-byte Folded Reload +; RV32-NEXT: addi sp, sp, 16 +; RV32-NEXT: ret +; +; RV64-LABEL: udiv_i64: +; RV64: # %bb.0: +; RV64-NEXT: tail __udivdi3 + %res = udiv i64 %x, %y + ret i64 %res +} + +define i65 @udiv_i65(i65 %x, i65 %y) nounwind { +; RV32-LABEL: udiv_i65: +; RV32: # %bb.0: # %_udiv-special-cases +; RV32-NEXT: lw a3, 0(a2) +; RV32-NEXT: lw a4, 4(a2) +; RV32-NEXT: lw t1, 8(a2) +; RV32-NEXT: lui a2, 349525 +; RV32-NEXT: lui a5, 209715 +; RV32-NEXT: lui a6, 61681 +; RV32-NEXT: addi t0, a2, 1365 +; RV32-NEXT: addi a7, a5, 819 +; RV32-NEXT: addi a6, a6, -241 +; RV32-NEXT: srli a2, a4, 1 +; RV32-NEXT: slli a5, t1, 31 +; RV32-NEXT: slli t3, a4, 31 +; RV32-NEXT: or t2, a5, a2 +; RV32-NEXT: srli a2, a3, 1 +; RV32-NEXT: or t4, a2, t3 +; RV32-NEXT: bnez t2, .LBB1_2 +; RV32-NEXT: # %bb.1: # %_udiv-special-cases +; RV32-NEXT: srli a2, t4, 1 +; RV32-NEXT: or a2, t4, a2 +; RV32-NEXT: srli a5, a2, 2 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 4 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 8 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 16 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: not a2, a2 +; RV32-NEXT: srli a5, a2, 1 +; RV32-NEXT: and a5, a5, t0 +; RV32-NEXT: sub a2, a2, a5 +; RV32-NEXT: and a5, a2, a7 +; RV32-NEXT: srli a2, a2, 2 +; RV32-NEXT: and a2, a2, a7 +; RV32-NEXT: add a2, a5, a2 +; RV32-NEXT: srli a5, a2, 4 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: and a2, a2, a6 +; RV32-NEXT: slli a5, a2, 8 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: slli a5, a2, 16 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: srli a2, a2, 24 +; RV32-NEXT: addi t3, a2, 32 +; RV32-NEXT: j .LBB1_3 +; RV32-NEXT: .LBB1_2: +; RV32-NEXT: srli a2, t2, 1 +; RV32-NEXT: or a2, t2, a2 +; RV32-NEXT: srli a5, a2, 2 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 4 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 8 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 16 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: not a2, a2 +; RV32-NEXT: srli a5, a2, 1 +; RV32-NEXT: and a5, a5, t0 +; RV32-NEXT: sub a2, a2, a5 +; RV32-NEXT: and a5, a2, a7 +; RV32-NEXT: srli a2, a2, 2 +; RV32-NEXT: and a2, a2, a7 +; RV32-NEXT: add a2, a5, a2 +; RV32-NEXT: srli a5, a2, 4 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: and a2, a2, a6 +; RV32-NEXT: slli a5, a2, 8 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: slli a5, a2, 16 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: srli t3, a2, 24 +; RV32-NEXT: .LBB1_3: # %_udiv-special-cases +; RV32-NEXT: addi sp, sp, -96 +; RV32-NEXT: sw s0, 92(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s1, 88(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s2, 84(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s3, 80(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s4, 76(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s5, 72(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s6, 68(sp) # 4-byte Folded Spill +; RV32-NEXT: slli a2, a3, 31 +; RV32-NEXT: li t5, 64 +; RV32-NEXT: bnez a2, .LBB1_5 +; RV32-NEXT: # %bb.4: # %_udiv-special-cases +; RV32-NEXT: li s0, 64 +; RV32-NEXT: j .LBB1_6 +; RV32-NEXT: .LBB1_5: +; RV32-NEXT: srli a5, a2, 1 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 2 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 4 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 8 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: srli a5, a2, 16 +; RV32-NEXT: or a2, a2, a5 +; RV32-NEXT: not a2, a2 +; RV32-NEXT: srli a5, a2, 1 +; RV32-NEXT: and a5, a5, t0 +; RV32-NEXT: sub a2, a2, a5 +; RV32-NEXT: and a5, a2, a7 +; RV32-NEXT: srli a2, a2, 2 +; RV32-NEXT: and a2, a2, a7 +; RV32-NEXT: add a2, a5, a2 +; RV32-NEXT: srli a5, a2, 4 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: and a2, a2, a6 +; RV32-NEXT: slli a5, a2, 8 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: slli a5, a2, 16 +; RV32-NEXT: add a2, a2, a5 +; RV32-NEXT: srli s0, a2, 24 +; RV32-NEXT: .LBB1_6: # %_udiv-special-cases +; RV32-NEXT: lw a5, 0(a1) +; RV32-NEXT: lw a2, 4(a1) +; RV32-NEXT: lw s2, 8(a1) +; RV32-NEXT: or a1, t4, t2 +; RV32-NEXT: addi s1, s0, 64 +; RV32-NEXT: bnez a1, .LBB1_8 +; RV32-NEXT: # %bb.7: # %_udiv-special-cases +; RV32-NEXT: mv t3, s1 +; RV32-NEXT: .LBB1_8: # %_udiv-special-cases +; RV32-NEXT: snez s4, a1 +; RV32-NEXT: srli a1, a2, 1 +; RV32-NEXT: slli t2, s2, 31 +; RV32-NEXT: slli t4, a2, 31 +; RV32-NEXT: or a1, t2, a1 +; RV32-NEXT: srli t2, a5, 1 +; RV32-NEXT: or t6, t2, t4 +; RV32-NEXT: bnez a1, .LBB1_10 +; RV32-NEXT: # %bb.9: # %_udiv-special-cases +; RV32-NEXT: srli t2, t6, 1 +; RV32-NEXT: or t2, t6, t2 +; RV32-NEXT: srli t4, t2, 2 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: srli t4, t2, 4 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: srli t4, t2, 8 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: srli t4, t2, 16 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: not t2, t2 +; RV32-NEXT: srli t4, t2, 1 +; RV32-NEXT: and t4, t4, t0 +; RV32-NEXT: sub t2, t2, t4 +; RV32-NEXT: and t4, t2, a7 +; RV32-NEXT: srli t2, t2, 2 +; RV32-NEXT: and t2, t2, a7 +; RV32-NEXT: add t2, t4, t2 +; RV32-NEXT: srli t4, t2, 4 +; RV32-NEXT: add t2, t2, t4 +; RV32-NEXT: and t2, t2, a6 +; RV32-NEXT: slli t4, t2, 8 +; RV32-NEXT: add t2, t2, t4 +; RV32-NEXT: slli t4, t2, 16 +; RV32-NEXT: add t2, t2, t4 +; RV32-NEXT: srli t2, t2, 24 +; RV32-NEXT: addi s3, t2, 32 +; RV32-NEXT: j .LBB1_11 +; RV32-NEXT: .LBB1_10: +; RV32-NEXT: srli t2, a1, 1 +; RV32-NEXT: or t2, a1, t2 +; RV32-NEXT: srli t4, t2, 2 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: srli t4, t2, 4 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: srli t4, t2, 8 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: srli t4, t2, 16 +; RV32-NEXT: or t2, t2, t4 +; RV32-NEXT: not t2, t2 +; RV32-NEXT: srli t4, t2, 1 +; RV32-NEXT: and t4, t4, t0 +; RV32-NEXT: sub t2, t2, t4 +; RV32-NEXT: and t4, t2, a7 +; RV32-NEXT: srli t2, t2, 2 +; RV32-NEXT: and t2, t2, a7 +; RV32-NEXT: add t2, t4, t2 +; RV32-NEXT: srli t4, t2, 4 +; RV32-NEXT: add t2, t2, t4 +; RV32-NEXT: and t2, t2, a6 +; RV32-NEXT: slli t4, t2, 8 +; RV32-NEXT: add t2, t2, t4 +; RV32-NEXT: slli t4, t2, 16 +; RV32-NEXT: add t2, t2, t4 +; RV32-NEXT: srli s3, t2, 24 +; RV32-NEXT: .LBB1_11: # %_udiv-special-cases +; RV32-NEXT: andi t4, s2, 1 +; RV32-NEXT: andi t1, t1, 1 +; RV32-NEXT: or t2, a3, a4 +; RV32-NEXT: or s2, a5, a2 +; RV32-NEXT: sltu s0, s1, s0 +; RV32-NEXT: slli s1, a5, 31 +; RV32-NEXT: addi s4, s4, -1 +; RV32-NEXT: beqz s1, .LBB1_13 +; RV32-NEXT: # %bb.12: +; RV32-NEXT: srli t5, s1, 1 +; RV32-NEXT: or t5, s1, t5 +; RV32-NEXT: srli s1, t5, 2 +; RV32-NEXT: or t5, t5, s1 +; RV32-NEXT: srli s1, t5, 4 +; RV32-NEXT: or t5, t5, s1 +; RV32-NEXT: srli s1, t5, 8 +; RV32-NEXT: or t5, t5, s1 +; RV32-NEXT: srli s1, t5, 16 +; RV32-NEXT: or t5, t5, s1 +; RV32-NEXT: not t5, t5 +; RV32-NEXT: srli s1, t5, 1 +; RV32-NEXT: and t0, s1, t0 +; RV32-NEXT: sub t0, t5, t0 +; RV32-NEXT: and t5, t0, a7 +; RV32-NEXT: srli t0, t0, 2 +; RV32-NEXT: and a7, t0, a7 +; RV32-NEXT: add a7, t5, a7 +; RV32-NEXT: srli t0, a7, 4 +; RV32-NEXT: add a7, a7, t0 +; RV32-NEXT: and a6, a7, a6 +; RV32-NEXT: slli a7, a6, 8 +; RV32-NEXT: add a6, a6, a7 +; RV32-NEXT: slli a7, a6, 16 +; RV32-NEXT: add a6, a6, a7 +; RV32-NEXT: srli t5, a6, 24 +; RV32-NEXT: .LBB1_13: # %_udiv-special-cases +; RV32-NEXT: or t0, t2, t1 +; RV32-NEXT: or a6, s2, t4 +; RV32-NEXT: and a7, s4, s0 +; RV32-NEXT: or t6, t6, a1 +; RV32-NEXT: addi s0, t5, 64 +; RV32-NEXT: bnez t6, .LBB1_15 +; RV32-NEXT: # %bb.14: # %_udiv-special-cases +; RV32-NEXT: mv s3, s0 +; RV32-NEXT: .LBB1_15: # %_udiv-special-cases +; RV32-NEXT: seqz a1, t0 +; RV32-NEXT: sltu t0, s0, t5 +; RV32-NEXT: snez t5, t6 +; RV32-NEXT: addi t5, t5, -1 +; RV32-NEXT: and t0, t5, t0 +; RV32-NEXT: sltu t5, t3, s3 +; RV32-NEXT: seqz a6, a6 +; RV32-NEXT: mv t6, t5 +; RV32-NEXT: beq a7, t0, .LBB1_17 +; RV32-NEXT: # %bb.16: # %_udiv-special-cases +; RV32-NEXT: sltu t6, a7, t0 +; RV32-NEXT: .LBB1_17: # %_udiv-special-cases +; RV32-NEXT: or a1, a1, a6 +; RV32-NEXT: andi a6, t6, 1 +; RV32-NEXT: sub a7, a7, t0 +; RV32-NEXT: sub t5, a7, t5 +; RV32-NEXT: sub a7, t3, s3 +; RV32-NEXT: beqz a6, .LBB1_19 +; RV32-NEXT: # %bb.18: # %_udiv-special-cases +; RV32-NEXT: mv t0, a6 +; RV32-NEXT: j .LBB1_20 +; RV32-NEXT: .LBB1_19: +; RV32-NEXT: sltiu t0, a7, 65 +; RV32-NEXT: xori t0, t0, 1 +; RV32-NEXT: snez t3, t5 +; RV32-NEXT: or t0, t0, t3 +; RV32-NEXT: .LBB1_20: # %_udiv-special-cases +; RV32-NEXT: or t6, a1, t0 +; RV32-NEXT: addi a1, t6, -1 +; RV32-NEXT: and t3, t4, a1 +; RV32-NEXT: and t0, a1, a2 +; RV32-NEXT: and a1, a1, a5 +; RV32-NEXT: bnez t6, .LBB1_30 +; RV32-NEXT: # %bb.21: # %_udiv-special-cases +; RV32-NEXT: xori t6, a7, 64 +; RV32-NEXT: or t6, t6, a6 +; RV32-NEXT: or t6, t6, t5 +; RV32-NEXT: beqz t6, .LBB1_30 +; RV32-NEXT: # %bb.22: # %udiv-bb1 +; RV32-NEXT: addi a1, a7, 1 +; RV32-NEXT: sw zero, 32(sp) +; RV32-NEXT: sw zero, 36(sp) +; RV32-NEXT: sw zero, 40(sp) +; RV32-NEXT: sw zero, 44(sp) +; RV32-NEXT: sw a5, 48(sp) +; RV32-NEXT: sw a2, 52(sp) +; RV32-NEXT: sw t4, 56(sp) +; RV32-NEXT: li t0, 64 +; RV32-NEXT: addi t3, sp, 48 +; RV32-NEXT: neg s1, a7 +; RV32-NEXT: seqz t6, a1 +; RV32-NEXT: sub a7, t0, a7 +; RV32-NEXT: add t5, t5, t6 +; RV32-NEXT: andi t0, a7, 31 +; RV32-NEXT: srli a7, a7, 3 +; RV32-NEXT: or t6, a1, t5 +; RV32-NEXT: xori s2, t0, 31 +; RV32-NEXT: andi a7, a7, 12 +; RV32-NEXT: seqz t0, t6 +; RV32-NEXT: sub s3, t3, a7 +; RV32-NEXT: add a6, a6, t0 +; RV32-NEXT: lw t3, 0(s3) +; RV32-NEXT: lw s4, 4(s3) +; RV32-NEXT: andi a7, a6, 1 +; RV32-NEXT: or t6, t6, a7 +; RV32-NEXT: srli a6, t3, 1 +; RV32-NEXT: sll t0, s4, s1 +; RV32-NEXT: srl a6, a6, s2 +; RV32-NEXT: or t0, t0, a6 +; RV32-NEXT: sll a6, t3, s1 +; RV32-NEXT: li t3, 0 +; RV32-NEXT: beqz t6, .LBB1_28 +; RV32-NEXT: # %bb.23: # %udiv-preheader +; RV32-NEXT: li t6, 0 +; RV32-NEXT: li s0, 0 +; RV32-NEXT: srli s4, s4, 1 +; RV32-NEXT: lw s3, 8(s3) +; RV32-NEXT: sw zero, 16(sp) +; RV32-NEXT: sw zero, 20(sp) +; RV32-NEXT: sw zero, 24(sp) +; RV32-NEXT: sw zero, 28(sp) +; RV32-NEXT: sw a5, 0(sp) +; RV32-NEXT: sw a2, 4(sp) +; RV32-NEXT: sw t4, 8(sp) +; RV32-NEXT: sw zero, 12(sp) +; RV32-NEXT: srli a2, a1, 3 +; RV32-NEXT: srl a5, s4, s2 +; RV32-NEXT: mv t4, sp +; RV32-NEXT: snez t2, t2 +; RV32-NEXT: andi a2, a2, 12 +; RV32-NEXT: add t1, t1, t2 +; RV32-NEXT: add a2, t4, a2 +; RV32-NEXT: lw t2, 0(a2) +; RV32-NEXT: lw t4, 4(a2) +; RV32-NEXT: lw a2, 8(a2) +; RV32-NEXT: sll s1, s3, s1 +; RV32-NEXT: andi s2, a1, 31 +; RV32-NEXT: xori s2, s2, 31 +; RV32-NEXT: or s3, s1, a5 +; RV32-NEXT: slli a2, a2, 1 +; RV32-NEXT: slli a5, t4, 1 +; RV32-NEXT: sll a2, a2, s2 +; RV32-NEXT: sll s2, a5, s2 +; RV32-NEXT: srl s1, t4, a1 +; RV32-NEXT: or s1, s1, a2 +; RV32-NEXT: seqz a2, a3 +; RV32-NEXT: sub a2, a4, a2 +; RV32-NEXT: addi a5, t1, 1 +; RV32-NEXT: andi a5, a5, 1 +; RV32-NEXT: andi s3, s3, 1 +; RV32-NEXT: srl t1, t2, a1 +; RV32-NEXT: or s2, t1, s2 +; RV32-NEXT: addi t1, a3, -1 +; RV32-NEXT: j .LBB1_26 +; RV32-NEXT: .LBB1_24: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB1_26 Depth=1 +; RV32-NEXT: sltu t2, a2, s4 +; RV32-NEXT: .LBB1_25: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB1_26 Depth=1 +; RV32-NEXT: srli s1, s1, 31 +; RV32-NEXT: sub t4, a5, s1 +; RV32-NEXT: sub t2, t4, t2 +; RV32-NEXT: slli t2, t2, 31 +; RV32-NEXT: srai s1, t2, 31 +; RV32-NEXT: and s3, s1, a4 +; RV32-NEXT: li t2, 0 +; RV32-NEXT: li t4, 0 +; RV32-NEXT: srli s5, a6, 31 +; RV32-NEXT: sub s4, s4, s3 +; RV32-NEXT: slli s3, t0, 1 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli t0, t0, 31 +; RV32-NEXT: slli a6, a6, 1 +; RV32-NEXT: or a6, t3, a6 +; RV32-NEXT: seqz t3, a1 +; RV32-NEXT: or s0, s0, t0 +; RV32-NEXT: or s5, a1, t5 +; RV32-NEXT: sub t5, t5, t3 +; RV32-NEXT: and s6, s1, a3 +; RV32-NEXT: addi a1, a1, -1 +; RV32-NEXT: andi t3, s1, 1 +; RV32-NEXT: or t0, t6, s3 +; RV32-NEXT: sltu t6, s2, s6 +; RV32-NEXT: snez s5, s5 +; RV32-NEXT: andi s3, s0, 1 +; RV32-NEXT: sub s1, s4, t6 +; RV32-NEXT: add a7, a7, s5 +; RV32-NEXT: addi a7, a7, 1 +; RV32-NEXT: andi a7, a7, 1 +; RV32-NEXT: or t6, a1, t5 +; RV32-NEXT: or s4, t6, a7 +; RV32-NEXT: sub s2, s2, s6 +; RV32-NEXT: li t6, 0 +; RV32-NEXT: li s0, 0 +; RV32-NEXT: beqz s4, .LBB1_29 +; RV32-NEXT: .LBB1_26: # %udiv-do-while +; RV32-NEXT: # =>This Inner Loop Header: Depth=1 +; RV32-NEXT: srli t2, s2, 31 +; RV32-NEXT: slli t4, s1, 1 +; RV32-NEXT: slli s2, s2, 1 +; RV32-NEXT: or s4, t4, t2 +; RV32-NEXT: andi t2, s3, 1 +; RV32-NEXT: or s2, s2, t2 +; RV32-NEXT: bne a2, s4, .LBB1_24 +; RV32-NEXT: # %bb.27: # in Loop: Header=BB1_26 Depth=1 +; RV32-NEXT: sltu t2, t1, s2 +; RV32-NEXT: j .LBB1_25 +; RV32-NEXT: .LBB1_28: +; RV32-NEXT: li t2, 0 +; RV32-NEXT: li t4, 0 +; RV32-NEXT: .LBB1_29: # %udiv-loop-exit +; RV32-NEXT: srli a2, a6, 31 +; RV32-NEXT: slli a3, t0, 1 +; RV32-NEXT: srli a4, t0, 31 +; RV32-NEXT: slli a6, a6, 1 +; RV32-NEXT: or a1, t3, a6 +; RV32-NEXT: or a2, t2, a2 +; RV32-NEXT: or a4, t4, a4 +; RV32-NEXT: or t0, a2, a3 +; RV32-NEXT: andi t3, a4, 1 +; RV32-NEXT: .LBB1_30: # %udiv-end +; RV32-NEXT: andi a2, t3, 1 +; RV32-NEXT: sw a1, 0(a0) +; RV32-NEXT: sw t0, 4(a0) +; RV32-NEXT: sb a2, 8(a0) +; RV32-NEXT: lw s0, 92(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s1, 88(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s2, 84(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s3, 80(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s4, 76(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s5, 72(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s6, 68(sp) # 4-byte Folded Reload +; RV32-NEXT: addi sp, sp, 96 +; RV32-NEXT: ret +; +; RV64-LABEL: udiv_i65: +; RV64: # %bb.0: +; RV64-NEXT: addi sp, sp, -16 +; RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill +; RV64-NEXT: andi a1, a1, 1 +; RV64-NEXT: andi a3, a3, 1 +; RV64-NEXT: call __udivti3 +; RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload +; RV64-NEXT: addi sp, sp, 16 +; RV64-NEXT: ret + %res = udiv i65 %x, %y + ret i65 %res +} define i128 @udiv_i128(i128 %x, i128 %y) nounwind { -; CHECK-LABEL: udiv_i128: -; CHECK: call __udivti3 +; RV32-LABEL: udiv_i128: +; RV32: # %bb.0: # %_udiv-special-cases +; RV32-NEXT: addi sp, sp, -160 +; RV32-NEXT: sw ra, 156(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s0, 152(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s1, 148(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s2, 144(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s3, 140(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s4, 136(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s5, 132(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s6, 128(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s7, 124(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s8, 120(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s9, 116(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s10, 112(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s11, 108(sp) # 4-byte Folded Spill +; RV32-NEXT: mv s7, a0 +; RV32-NEXT: lw s8, 0(a2) +; RV32-NEXT: lw s9, 4(a2) +; RV32-NEXT: lw s11, 8(a2) +; RV32-NEXT: lw ra, 12(a2) +; RV32-NEXT: lui t4, 349525 +; RV32-NEXT: addi t4, t4, 1365 +; RV32-NEXT: lui t3, 209715 +; RV32-NEXT: addi t3, t3, 819 +; RV32-NEXT: lui t2, 61681 +; RV32-NEXT: addi t2, t2, -241 +; RV32-NEXT: bnez s9, .LBB2_2 +; RV32-NEXT: # %bb.1: # %_udiv-special-cases +; RV32-NEXT: srli a0, s8, 1 +; RV32-NEXT: or a0, s8, a0 +; RV32-NEXT: srli a3, a0, 2 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 8 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 16 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: not a0, a0 +; RV32-NEXT: srli a3, a0, 1 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: sub a0, a0, a3 +; RV32-NEXT: and a3, a0, t3 +; RV32-NEXT: srli a0, a0, 2 +; RV32-NEXT: and a0, a0, t3 +; RV32-NEXT: add a0, a3, a0 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: and a0, a0, t2 +; RV32-NEXT: slli a3, a0, 8 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: slli a3, a0, 16 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: srli a0, a0, 24 +; RV32-NEXT: addi t6, a0, 32 +; RV32-NEXT: j .LBB2_3 +; RV32-NEXT: .LBB2_2: +; RV32-NEXT: srli a0, s9, 1 +; RV32-NEXT: or a0, s9, a0 +; RV32-NEXT: srli a3, a0, 2 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 8 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 16 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: not a0, a0 +; RV32-NEXT: srli a3, a0, 1 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: sub a0, a0, a3 +; RV32-NEXT: and a3, a0, t3 +; RV32-NEXT: srli a0, a0, 2 +; RV32-NEXT: and a0, a0, t3 +; RV32-NEXT: add a0, a3, a0 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: and a0, a0, t2 +; RV32-NEXT: slli a3, a0, 8 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: slli a3, a0, 16 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: srli t6, a0, 24 +; RV32-NEXT: .LBB2_3: # %_udiv-special-cases +; RV32-NEXT: lw a6, 4(a1) +; RV32-NEXT: or s0, s11, ra +; RV32-NEXT: bnez ra, .LBB2_5 +; RV32-NEXT: # %bb.4: # %_udiv-special-cases +; RV32-NEXT: srli a0, s11, 1 +; RV32-NEXT: or a0, s11, a0 +; RV32-NEXT: srli a3, a0, 2 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 8 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 16 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: not a0, a0 +; RV32-NEXT: srli a3, a0, 1 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: sub a0, a0, a3 +; RV32-NEXT: and a3, a0, t3 +; RV32-NEXT: srli a0, a0, 2 +; RV32-NEXT: and a0, a0, t3 +; RV32-NEXT: add a0, a3, a0 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: and a0, a0, t2 +; RV32-NEXT: slli a3, a0, 8 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: slli a3, a0, 16 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: srli a0, a0, 24 +; RV32-NEXT: addi t5, a0, 32 +; RV32-NEXT: j .LBB2_6 +; RV32-NEXT: .LBB2_5: +; RV32-NEXT: srli a0, ra, 1 +; RV32-NEXT: or a0, ra, a0 +; RV32-NEXT: srli a3, a0, 2 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 8 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: srli a3, a0, 16 +; RV32-NEXT: or a0, a0, a3 +; RV32-NEXT: not a0, a0 +; RV32-NEXT: srli a3, a0, 1 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: sub a0, a0, a3 +; RV32-NEXT: and a3, a0, t3 +; RV32-NEXT: srli a0, a0, 2 +; RV32-NEXT: and a0, a0, t3 +; RV32-NEXT: add a0, a3, a0 +; RV32-NEXT: srli a3, a0, 4 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: and a0, a0, t2 +; RV32-NEXT: slli a3, a0, 8 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: slli a3, a0, 16 +; RV32-NEXT: add a0, a0, a3 +; RV32-NEXT: srli t5, a0, 24 +; RV32-NEXT: .LBB2_6: # %_udiv-special-cases +; RV32-NEXT: lw a7, 12(a1) +; RV32-NEXT: addi a0, t6, 64 +; RV32-NEXT: bnez s0, .LBB2_8 +; RV32-NEXT: # %bb.7: # %_udiv-special-cases +; RV32-NEXT: mv t5, a0 +; RV32-NEXT: .LBB2_8: # %_udiv-special-cases +; RV32-NEXT: lw t1, 0(a1) +; RV32-NEXT: lw t0, 8(a1) +; RV32-NEXT: snez s3, s0 +; RV32-NEXT: bnez a6, .LBB2_10 +; RV32-NEXT: # %bb.9: # %_udiv-special-cases +; RV32-NEXT: srli a1, t1, 1 +; RV32-NEXT: or a1, t1, a1 +; RV32-NEXT: srli a3, a1, 2 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: srli a3, a1, 4 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: srli a3, a1, 8 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: srli a3, a1, 16 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: not a1, a1 +; RV32-NEXT: srli a3, a1, 1 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: sub a1, a1, a3 +; RV32-NEXT: and a3, a1, t3 +; RV32-NEXT: srli a1, a1, 2 +; RV32-NEXT: and a1, a1, t3 +; RV32-NEXT: add a1, a3, a1 +; RV32-NEXT: srli a3, a1, 4 +; RV32-NEXT: add a1, a1, a3 +; RV32-NEXT: and a1, a1, t2 +; RV32-NEXT: slli a3, a1, 8 +; RV32-NEXT: add a1, a1, a3 +; RV32-NEXT: slli a3, a1, 16 +; RV32-NEXT: add a1, a1, a3 +; RV32-NEXT: srli a1, a1, 24 +; RV32-NEXT: addi a3, a1, 32 +; RV32-NEXT: j .LBB2_11 +; RV32-NEXT: .LBB2_10: +; RV32-NEXT: srli a1, a6, 1 +; RV32-NEXT: or a1, a6, a1 +; RV32-NEXT: srli a3, a1, 2 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: srli a3, a1, 4 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: srli a3, a1, 8 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: srli a3, a1, 16 +; RV32-NEXT: or a1, a1, a3 +; RV32-NEXT: not a1, a1 +; RV32-NEXT: srli a3, a1, 1 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: sub a1, a1, a3 +; RV32-NEXT: and a3, a1, t3 +; RV32-NEXT: srli a1, a1, 2 +; RV32-NEXT: and a1, a1, t3 +; RV32-NEXT: add a1, a3, a1 +; RV32-NEXT: srli a3, a1, 4 +; RV32-NEXT: add a1, a1, a3 +; RV32-NEXT: and a1, a1, t2 +; RV32-NEXT: slli a3, a1, 8 +; RV32-NEXT: add a1, a1, a3 +; RV32-NEXT: slli a3, a1, 16 +; RV32-NEXT: add a1, a1, a3 +; RV32-NEXT: srli a3, a1, 24 +; RV32-NEXT: .LBB2_11: # %_udiv-special-cases +; RV32-NEXT: or a1, s9, ra +; RV32-NEXT: or s0, s8, s11 +; RV32-NEXT: or s1, a6, a7 +; RV32-NEXT: or s2, t1, t0 +; RV32-NEXT: sltu t6, a0, t6 +; RV32-NEXT: addi s3, s3, -1 +; RV32-NEXT: addi a0, a3, 64 +; RV32-NEXT: or s4, t0, a7 +; RV32-NEXT: sltu s5, a0, a3 +; RV32-NEXT: snez s6, s4 +; RV32-NEXT: addi s6, s6, -1 +; RV32-NEXT: bnez a7, .LBB2_13 +; RV32-NEXT: # %bb.12: # %_udiv-special-cases +; RV32-NEXT: srli a3, t0, 1 +; RV32-NEXT: or a3, t0, a3 +; RV32-NEXT: srli a4, a3, 2 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 8 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 16 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: not a3, a3 +; RV32-NEXT: srli a4, a3, 1 +; RV32-NEXT: and a4, a4, t4 +; RV32-NEXT: sub a3, a3, a4 +; RV32-NEXT: and a4, a3, t3 +; RV32-NEXT: srli a3, a3, 2 +; RV32-NEXT: and a3, a3, t3 +; RV32-NEXT: add a3, a4, a3 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: and a3, a3, t2 +; RV32-NEXT: slli a4, a3, 8 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: slli a4, a3, 16 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: srli a3, a3, 24 +; RV32-NEXT: addi a3, a3, 32 +; RV32-NEXT: j .LBB2_14 +; RV32-NEXT: .LBB2_13: +; RV32-NEXT: srli a3, a7, 1 +; RV32-NEXT: or a3, a7, a3 +; RV32-NEXT: srli a4, a3, 2 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 8 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 16 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: not a3, a3 +; RV32-NEXT: srli a4, a3, 1 +; RV32-NEXT: and a4, a4, t4 +; RV32-NEXT: sub a3, a3, a4 +; RV32-NEXT: and a4, a3, t3 +; RV32-NEXT: srli a3, a3, 2 +; RV32-NEXT: and a3, a3, t3 +; RV32-NEXT: add a3, a4, a3 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: and a3, a3, t2 +; RV32-NEXT: slli a4, a3, 8 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: slli a4, a3, 16 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: srli a3, a3, 24 +; RV32-NEXT: .LBB2_14: # %_udiv-special-cases +; RV32-NEXT: or s0, s0, a1 +; RV32-NEXT: or a5, s2, s1 +; RV32-NEXT: and a1, s3, t6 +; RV32-NEXT: and a4, s6, s5 +; RV32-NEXT: bnez s4, .LBB2_16 +; RV32-NEXT: # %bb.15: # %_udiv-special-cases +; RV32-NEXT: mv a3, a0 +; RV32-NEXT: .LBB2_16: # %_udiv-special-cases +; RV32-NEXT: seqz a0, s0 +; RV32-NEXT: seqz a5, a5 +; RV32-NEXT: sltu t2, t5, a3 +; RV32-NEXT: sub t4, a1, a4 +; RV32-NEXT: mv t3, t2 +; RV32-NEXT: beq a1, a4, .LBB2_18 +; RV32-NEXT: # %bb.17: # %_udiv-special-cases +; RV32-NEXT: sltu t3, a1, a4 +; RV32-NEXT: .LBB2_18: # %_udiv-special-cases +; RV32-NEXT: sub t2, t4, t2 +; RV32-NEXT: or a0, a0, a5 +; RV32-NEXT: neg t4, t3 +; RV32-NEXT: seqz t6, t3 +; RV32-NEXT: addi t6, t6, -1 +; RV32-NEXT: or a1, t4, t6 +; RV32-NEXT: sub t3, t5, a3 +; RV32-NEXT: beqz a1, .LBB2_20 +; RV32-NEXT: # %bb.19: # %_udiv-special-cases +; RV32-NEXT: snez a1, a1 +; RV32-NEXT: j .LBB2_21 +; RV32-NEXT: .LBB2_20: +; RV32-NEXT: snez a1, t2 +; RV32-NEXT: sltiu a3, t3, 128 +; RV32-NEXT: xori a3, a3, 1 +; RV32-NEXT: or a1, a3, a1 +; RV32-NEXT: .LBB2_21: # %_udiv-special-cases +; RV32-NEXT: or a5, a0, a1 +; RV32-NEXT: addi a3, a5, -1 +; RV32-NEXT: and a0, a3, a7 +; RV32-NEXT: and a1, a3, t0 +; RV32-NEXT: and a4, a3, a6 +; RV32-NEXT: and a3, a3, t1 +; RV32-NEXT: bnez a5, .LBB2_26 +; RV32-NEXT: # %bb.22: # %_udiv-special-cases +; RV32-NEXT: xori a5, t3, 127 +; RV32-NEXT: or a5, a5, t4 +; RV32-NEXT: or t5, t2, t6 +; RV32-NEXT: or a5, a5, t5 +; RV32-NEXT: beqz a5, .LBB2_26 +; RV32-NEXT: # %bb.23: # %udiv-bb1 +; RV32-NEXT: sw s7, 12(sp) # 4-byte Folded Spill +; RV32-NEXT: addi a1, t3, 1 +; RV32-NEXT: sw zero, 72(sp) +; RV32-NEXT: sw zero, 76(sp) +; RV32-NEXT: sw zero, 80(sp) +; RV32-NEXT: sw zero, 84(sp) +; RV32-NEXT: sw t1, 88(sp) +; RV32-NEXT: sw a6, 92(sp) +; RV32-NEXT: sw t0, 96(sp) +; RV32-NEXT: sw a7, 100(sp) +; RV32-NEXT: li a0, 127 +; RV32-NEXT: addi a2, sp, 88 +; RV32-NEXT: seqz a3, a1 +; RV32-NEXT: sub a0, a0, t3 +; RV32-NEXT: add t2, t2, a3 +; RV32-NEXT: andi a3, a0, 31 +; RV32-NEXT: srli a0, a0, 3 +; RV32-NEXT: or a4, a1, t2 +; RV32-NEXT: xori a3, a3, 31 +; RV32-NEXT: andi a0, a0, 12 +; RV32-NEXT: seqz t5, a4 +; RV32-NEXT: sub a2, a2, a0 +; RV32-NEXT: add t5, t4, t5 +; RV32-NEXT: lw a0, 0(a2) +; RV32-NEXT: lw a4, 4(a2) +; RV32-NEXT: lw a5, 8(a2) +; RV32-NEXT: lw a2, 12(a2) +; RV32-NEXT: sltu t4, t5, t4 +; RV32-NEXT: or s0, a1, t5 +; RV32-NEXT: add t4, t6, t4 +; RV32-NEXT: or t6, t2, t4 +; RV32-NEXT: or s0, s0, t6 +; RV32-NEXT: srli t6, a5, 1 +; RV32-NEXT: srli s1, a4, 1 +; RV32-NEXT: srli s2, a0, 1 +; RV32-NEXT: srl t6, t6, a3 +; RV32-NEXT: srl s1, s1, a3 +; RV32-NEXT: srl a3, s2, a3 +; RV32-NEXT: not t3, t3 +; RV32-NEXT: sll a2, a2, t3 +; RV32-NEXT: or s2, a2, t6 +; RV32-NEXT: sll a2, a5, t3 +; RV32-NEXT: sll a4, a4, t3 +; RV32-NEXT: or s1, a2, s1 +; RV32-NEXT: or t6, a4, a3 +; RV32-NEXT: sll t3, a0, t3 +; RV32-NEXT: bnez s0, .LBB2_27 +; RV32-NEXT: # %bb.24: +; RV32-NEXT: li s6, 0 +; RV32-NEXT: li s7, 0 +; RV32-NEXT: li s8, 0 +; RV32-NEXT: .LBB2_25: # %udiv-loop-exit +; RV32-NEXT: srli a0, s1, 31 +; RV32-NEXT: slli s2, s2, 1 +; RV32-NEXT: or a0, s2, a0 +; RV32-NEXT: srli a1, t6, 31 +; RV32-NEXT: slli s1, s1, 1 +; RV32-NEXT: or a1, s1, a1 +; RV32-NEXT: srli a2, t3, 31 +; RV32-NEXT: slli t6, t6, 1 +; RV32-NEXT: slli a3, t3, 1 +; RV32-NEXT: or a3, s0, a3 +; RV32-NEXT: or a2, s6, a2 +; RV32-NEXT: or a4, a2, t6 +; RV32-NEXT: or a1, s7, a1 +; RV32-NEXT: or a0, s8, a0 +; RV32-NEXT: lw s7, 12(sp) # 4-byte Folded Reload +; RV32-NEXT: .LBB2_26: # %udiv-end +; RV32-NEXT: sw a3, 0(s7) +; RV32-NEXT: sw a4, 4(s7) +; RV32-NEXT: sw a1, 8(s7) +; RV32-NEXT: sw a0, 12(s7) +; RV32-NEXT: lw ra, 156(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s0, 152(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s1, 148(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s2, 144(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s3, 140(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s4, 136(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s5, 132(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s6, 128(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s7, 124(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s8, 120(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s9, 116(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s10, 112(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s11, 108(sp) # 4-byte Folded Reload +; RV32-NEXT: addi sp, sp, 160 +; RV32-NEXT: ret +; RV32-NEXT: .LBB2_27: # %udiv-preheader +; RV32-NEXT: li s0, 0 +; RV32-NEXT: li s5, 0 +; RV32-NEXT: li s3, 0 +; RV32-NEXT: li s4, 0 +; RV32-NEXT: sw zero, 56(sp) +; RV32-NEXT: sw zero, 60(sp) +; RV32-NEXT: sw zero, 64(sp) +; RV32-NEXT: sw zero, 68(sp) +; RV32-NEXT: sw t1, 40(sp) +; RV32-NEXT: sw a6, 44(sp) +; RV32-NEXT: sw t0, 48(sp) +; RV32-NEXT: sw a7, 52(sp) +; RV32-NEXT: srli a0, a1, 3 +; RV32-NEXT: addi a2, sp, 40 +; RV32-NEXT: andi a0, a0, 12 +; RV32-NEXT: add a0, a2, a0 +; RV32-NEXT: lw a2, 4(a0) +; RV32-NEXT: lw a3, 8(a0) +; RV32-NEXT: lw a4, 12(a0) +; RV32-NEXT: lw a0, 0(a0) +; RV32-NEXT: andi a5, a1, 31 +; RV32-NEXT: xori a5, a5, 31 +; RV32-NEXT: slli a6, a4, 1 +; RV32-NEXT: slli a7, a3, 1 +; RV32-NEXT: slli t0, a2, 1 +; RV32-NEXT: sll a6, a6, a5 +; RV32-NEXT: sll a7, a7, a5 +; RV32-NEXT: sll a5, t0, a5 +; RV32-NEXT: seqz t0, s8 +; RV32-NEXT: srl a3, a3, a1 +; RV32-NEXT: or s10, a3, a6 +; RV32-NEXT: or a3, s8, s9 +; RV32-NEXT: sw s9, 32(sp) # 4-byte Folded Spill +; RV32-NEXT: sub a6, s9, t0 +; RV32-NEXT: seqz a3, a3 +; RV32-NEXT: srl a2, a2, a1 +; RV32-NEXT: or s9, a2, a7 +; RV32-NEXT: sub a7, s11, a3 +; RV32-NEXT: sw s11, 28(sp) # 4-byte Folded Spill +; RV32-NEXT: sltu a2, s11, a3 +; RV32-NEXT: sw ra, 24(sp) # 4-byte Folded Spill +; RV32-NEXT: sub a2, ra, a2 +; RV32-NEXT: sw a2, 20(sp) # 4-byte Folded Spill +; RV32-NEXT: srl a0, a0, a1 +; RV32-NEXT: srl ra, a4, a1 +; RV32-NEXT: or t1, a0, a5 +; RV32-NEXT: sw s8, 36(sp) # 4-byte Folded Spill +; RV32-NEXT: addi s8, s8, -1 +; RV32-NEXT: sw s8, 16(sp) # 4-byte Folded Spill +; RV32-NEXT: li s7, 0 +; RV32-NEXT: li s8, 0 +; RV32-NEXT: j .LBB2_29 +; RV32-NEXT: .LBB2_28: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB2_29 Depth=1 +; RV32-NEXT: li s6, 0 +; RV32-NEXT: sub a0, a0, a5 +; RV32-NEXT: srli a5, s1, 31 +; RV32-NEXT: slli s2, s2, 1 +; RV32-NEXT: or a5, s2, a5 +; RV32-NEXT: srli s2, t6, 31 +; RV32-NEXT: slli s1, s1, 1 +; RV32-NEXT: or s1, s1, s2 +; RV32-NEXT: srli s2, t3, 31 +; RV32-NEXT: slli t6, t6, 1 +; RV32-NEXT: slli t3, t3, 1 +; RV32-NEXT: or t6, t6, s2 +; RV32-NEXT: lw a2, 28(sp) # 4-byte Folded Reload +; RV32-NEXT: and s2, s10, a2 +; RV32-NEXT: or t3, s0, t3 +; RV32-NEXT: sub a2, a3, s2 +; RV32-NEXT: sltu a3, a3, s2 +; RV32-NEXT: lw t0, 24(sp) # 4-byte Folded Reload +; RV32-NEXT: and s0, s10, t0 +; RV32-NEXT: sub t0, s9, s0 +; RV32-NEXT: or s2, a1, t2 +; RV32-NEXT: sub s9, a0, a4 +; RV32-NEXT: seqz a0, a1 +; RV32-NEXT: sub t2, t2, a0 +; RV32-NEXT: or t6, s5, t6 +; RV32-NEXT: addi a1, a1, -1 +; RV32-NEXT: andi s0, s10, 1 +; RV32-NEXT: seqz a0, s2 +; RV32-NEXT: or s1, s3, s1 +; RV32-NEXT: or s2, s4, a5 +; RV32-NEXT: sub s10, a2, ra +; RV32-NEXT: sltu a2, a2, ra +; RV32-NEXT: sub a3, t0, a3 +; RV32-NEXT: sltu a4, t5, a0 +; RV32-NEXT: sub t5, t5, a0 +; RV32-NEXT: sub ra, a3, a2 +; RV32-NEXT: sub t4, t4, a4 +; RV32-NEXT: or a0, t2, t4 +; RV32-NEXT: or a2, a1, t5 +; RV32-NEXT: or a0, a2, a0 +; RV32-NEXT: sub t1, s11, t1 +; RV32-NEXT: li s5, 0 +; RV32-NEXT: li s3, 0 +; RV32-NEXT: li s4, 0 +; RV32-NEXT: beqz a0, .LBB2_25 +; RV32-NEXT: .LBB2_29: # %udiv-do-while +; RV32-NEXT: # =>This Inner Loop Header: Depth=1 +; RV32-NEXT: srli a0, t1, 31 +; RV32-NEXT: slli a3, s9, 1 +; RV32-NEXT: slli t1, t1, 1 +; RV32-NEXT: or a0, a3, a0 +; RV32-NEXT: srli a3, s2, 31 +; RV32-NEXT: or s11, t1, a3 +; RV32-NEXT: beq a6, a0, .LBB2_31 +; RV32-NEXT: # %bb.30: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB2_29 Depth=1 +; RV32-NEXT: sltu a4, a6, a0 +; RV32-NEXT: j .LBB2_32 +; RV32-NEXT: .LBB2_31: # in Loop: Header=BB2_29 Depth=1 +; RV32-NEXT: lw a2, 16(sp) # 4-byte Folded Reload +; RV32-NEXT: sltu a4, a2, s11 +; RV32-NEXT: .LBB2_32: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB2_29 Depth=1 +; RV32-NEXT: lw a2, 36(sp) # 4-byte Folded Reload +; RV32-NEXT: srli a3, s10, 31 +; RV32-NEXT: slli ra, ra, 1 +; RV32-NEXT: srli a5, s9, 31 +; RV32-NEXT: slli s10, s10, 1 +; RV32-NEXT: or s9, ra, a3 +; RV32-NEXT: or a3, s10, a5 +; RV32-NEXT: sub a5, a7, a3 +; RV32-NEXT: sltu t1, a7, a3 +; RV32-NEXT: lw t0, 20(sp) # 4-byte Folded Reload +; RV32-NEXT: sub s6, t0, s9 +; RV32-NEXT: sltu a4, a5, a4 +; RV32-NEXT: sub a5, s6, t1 +; RV32-NEXT: sub a5, a5, a4 +; RV32-NEXT: srai s10, a5, 31 +; RV32-NEXT: and t1, s10, a2 +; RV32-NEXT: lw a2, 32(sp) # 4-byte Folded Reload +; RV32-NEXT: and a5, s10, a2 +; RV32-NEXT: sltu a4, s11, t1 +; RV32-NEXT: mv ra, a4 +; RV32-NEXT: beq a0, a5, .LBB2_28 +; RV32-NEXT: # %bb.33: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB2_29 Depth=1 +; RV32-NEXT: sltu ra, a0, a5 +; RV32-NEXT: j .LBB2_28 +; +; RV64-LABEL: udiv_i128: +; RV64: # %bb.0: +; RV64-NEXT: addi sp, sp, -16 +; RV64-NEXT: sd ra, 8(sp) # 8-byte Folded Spill +; RV64-NEXT: call __udivti3 +; RV64-NEXT: ld ra, 8(sp) # 8-byte Folded Reload +; RV64-NEXT: addi sp, sp, 16 +; RV64-NEXT: ret %res = udiv i128 %x, %y ret i128 %res } define i129 @udiv_i129(i129 %x, i129 %y) nounwind { -; CHECK-LABEL: udiv_i129: -; CHECK-NOT: call{{.*}}div +; RV32-LABEL: udiv_i129: +; RV32: # %bb.0: # %_udiv-special-cases +; RV32-NEXT: addi sp, sp, -240 +; RV32-NEXT: sw ra, 236(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s0, 232(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s1, 228(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s2, 224(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s3, 220(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s4, 216(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s5, 212(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s6, 208(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s7, 204(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s8, 200(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s9, 196(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s10, 192(sp) # 4-byte Folded Spill +; RV32-NEXT: sw s11, 188(sp) # 4-byte Folded Spill +; RV32-NEXT: mv ra, a0 +; RV32-NEXT: lw t2, 16(a2) +; RV32-NEXT: lw a4, 0(a2) +; RV32-NEXT: lw a5, 4(a2) +; RV32-NEXT: lw a6, 8(a2) +; RV32-NEXT: lw a0, 12(a2) +; RV32-NEXT: sw a0, 24(sp) # 4-byte Folded Spill +; RV32-NEXT: lui a0, 349525 +; RV32-NEXT: lui a2, 209715 +; RV32-NEXT: lui a3, 61681 +; RV32-NEXT: addi t5, a0, 1365 +; RV32-NEXT: addi t4, a2, 819 +; RV32-NEXT: addi t3, a3, -241 +; RV32-NEXT: sw a6, 28(sp) # 4-byte Folded Spill +; RV32-NEXT: slli a0, a6, 31 +; RV32-NEXT: srli a2, a5, 1 +; RV32-NEXT: sw a5, 20(sp) # 4-byte Folded Spill +; RV32-NEXT: slli a3, a5, 31 +; RV32-NEXT: or a0, a2, a0 +; RV32-NEXT: sw a4, 32(sp) # 4-byte Folded Spill +; RV32-NEXT: srli a2, a4, 1 +; RV32-NEXT: or a2, a2, a3 +; RV32-NEXT: bnez a0, .LBB3_2 +; RV32-NEXT: # %bb.1: # %_udiv-special-cases +; RV32-NEXT: srli a3, a2, 1 +; RV32-NEXT: or a3, a2, a3 +; RV32-NEXT: srli a4, a3, 2 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 8 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 16 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: not a3, a3 +; RV32-NEXT: srli a4, a3, 1 +; RV32-NEXT: and a4, a4, t5 +; RV32-NEXT: sub a3, a3, a4 +; RV32-NEXT: and a4, a3, t4 +; RV32-NEXT: srli a3, a3, 2 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: add a3, a4, a3 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: and a3, a3, t3 +; RV32-NEXT: slli a4, a3, 8 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: slli a4, a3, 16 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: srli a3, a3, 24 +; RV32-NEXT: addi a6, a3, 32 +; RV32-NEXT: j .LBB3_3 +; RV32-NEXT: .LBB3_2: +; RV32-NEXT: srli a3, a0, 1 +; RV32-NEXT: or a3, a0, a3 +; RV32-NEXT: srli a4, a3, 2 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 8 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 16 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: not a3, a3 +; RV32-NEXT: srli a4, a3, 1 +; RV32-NEXT: and a4, a4, t5 +; RV32-NEXT: sub a3, a3, a4 +; RV32-NEXT: and a4, a3, t4 +; RV32-NEXT: srli a3, a3, 2 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: add a3, a4, a3 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: and a3, a3, t3 +; RV32-NEXT: slli a4, a3, 8 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: slli a4, a3, 16 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: srli a6, a3, 24 +; RV32-NEXT: .LBB3_3: # %_udiv-special-cases +; RV32-NEXT: lw a7, 24(sp) # 4-byte Folded Reload +; RV32-NEXT: srli a3, a7, 1 +; RV32-NEXT: slli a5, t2, 31 +; RV32-NEXT: slli a7, a7, 31 +; RV32-NEXT: lw a4, 28(sp) # 4-byte Folded Reload +; RV32-NEXT: srli t0, a4, 1 +; RV32-NEXT: lw a4, 32(sp) # 4-byte Folded Reload +; RV32-NEXT: slli a4, a4, 31 +; RV32-NEXT: li s2, 64 +; RV32-NEXT: bnez a4, .LBB3_5 +; RV32-NEXT: # %bb.4: # %_udiv-special-cases +; RV32-NEXT: li t6, 64 +; RV32-NEXT: j .LBB3_6 +; RV32-NEXT: .LBB3_5: +; RV32-NEXT: srli t1, a4, 1 +; RV32-NEXT: or t1, a4, t1 +; RV32-NEXT: srli t6, t1, 2 +; RV32-NEXT: or t1, t1, t6 +; RV32-NEXT: srli t6, t1, 4 +; RV32-NEXT: or t1, t1, t6 +; RV32-NEXT: srli t6, t1, 8 +; RV32-NEXT: or t1, t1, t6 +; RV32-NEXT: srli t6, t1, 16 +; RV32-NEXT: or t1, t1, t6 +; RV32-NEXT: not t1, t1 +; RV32-NEXT: srli t6, t1, 1 +; RV32-NEXT: and t6, t6, t5 +; RV32-NEXT: sub t1, t1, t6 +; RV32-NEXT: and t6, t1, t4 +; RV32-NEXT: srli t1, t1, 2 +; RV32-NEXT: and t1, t1, t4 +; RV32-NEXT: add t1, t6, t1 +; RV32-NEXT: srli t6, t1, 4 +; RV32-NEXT: add t1, t1, t6 +; RV32-NEXT: and t1, t1, t3 +; RV32-NEXT: slli t6, t1, 8 +; RV32-NEXT: add t1, t1, t6 +; RV32-NEXT: slli t6, t1, 16 +; RV32-NEXT: add t1, t1, t6 +; RV32-NEXT: srli t6, t1, 24 +; RV32-NEXT: .LBB3_6: # %_udiv-special-cases +; RV32-NEXT: or t1, a5, a3 +; RV32-NEXT: or a7, t0, a7 +; RV32-NEXT: bnez a4, .LBB3_8 +; RV32-NEXT: # %bb.7: # %_udiv-special-cases +; RV32-NEXT: li t6, 128 +; RV32-NEXT: .LBB3_8: # %_udiv-special-cases +; RV32-NEXT: or a5, a7, t1 +; RV32-NEXT: addi a4, a6, 64 +; RV32-NEXT: addi a3, t6, 128 +; RV32-NEXT: or a0, a0, t1 +; RV32-NEXT: or a2, a2, a7 +; RV32-NEXT: or s3, a2, a0 +; RV32-NEXT: sltu s0, a3, t6 +; RV32-NEXT: bnez s3, .LBB3_11 +; RV32-NEXT: # %bb.9: # %_udiv-special-cases +; RV32-NEXT: mv t6, s0 +; RV32-NEXT: beqz t1, .LBB3_12 +; RV32-NEXT: .LBB3_10: +; RV32-NEXT: srli a0, t1, 1 +; RV32-NEXT: or a0, t1, a0 +; RV32-NEXT: srli a2, a0, 2 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: srli a2, a0, 4 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: srli a2, a0, 8 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: srli a2, a0, 16 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: not a0, a0 +; RV32-NEXT: srli a2, a0, 1 +; RV32-NEXT: and a2, a2, t5 +; RV32-NEXT: sub a0, a0, a2 +; RV32-NEXT: and a2, a0, t4 +; RV32-NEXT: srli a0, a0, 2 +; RV32-NEXT: and a0, a0, t4 +; RV32-NEXT: add a0, a2, a0 +; RV32-NEXT: srli a2, a0, 4 +; RV32-NEXT: add a0, a0, a2 +; RV32-NEXT: and a0, a0, t3 +; RV32-NEXT: slli a2, a0, 8 +; RV32-NEXT: add a0, a0, a2 +; RV32-NEXT: slli a2, a0, 16 +; RV32-NEXT: add a0, a0, a2 +; RV32-NEXT: srli s1, a0, 24 +; RV32-NEXT: beqz a5, .LBB3_13 +; RV32-NEXT: j .LBB3_14 +; RV32-NEXT: .LBB3_11: +; RV32-NEXT: snez a0, a5 +; RV32-NEXT: sltu a2, a4, a6 +; RV32-NEXT: addi a0, a0, -1 +; RV32-NEXT: and t6, a0, a2 +; RV32-NEXT: bnez t1, .LBB3_10 +; RV32-NEXT: .LBB3_12: # %_udiv-special-cases +; RV32-NEXT: srli a0, a7, 1 +; RV32-NEXT: or a0, a7, a0 +; RV32-NEXT: srli a2, a0, 2 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: srli a2, a0, 4 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: srli a2, a0, 8 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: srli a2, a0, 16 +; RV32-NEXT: or a0, a0, a2 +; RV32-NEXT: not a0, a0 +; RV32-NEXT: srli a2, a0, 1 +; RV32-NEXT: and a2, a2, t5 +; RV32-NEXT: sub a0, a0, a2 +; RV32-NEXT: and a2, a0, t4 +; RV32-NEXT: srli a0, a0, 2 +; RV32-NEXT: and a0, a0, t4 +; RV32-NEXT: add a0, a2, a0 +; RV32-NEXT: srli a2, a0, 4 +; RV32-NEXT: add a0, a0, a2 +; RV32-NEXT: and a0, a0, t3 +; RV32-NEXT: slli a2, a0, 8 +; RV32-NEXT: add a0, a0, a2 +; RV32-NEXT: slli a2, a0, 16 +; RV32-NEXT: add a0, a0, a2 +; RV32-NEXT: srli a0, a0, 24 +; RV32-NEXT: addi s1, a0, 32 +; RV32-NEXT: bnez a5, .LBB3_14 +; RV32-NEXT: .LBB3_13: # %_udiv-special-cases +; RV32-NEXT: mv s1, a4 +; RV32-NEXT: .LBB3_14: # %_udiv-special-cases +; RV32-NEXT: lw a7, 0(a1) +; RV32-NEXT: lw t0, 4(a1) +; RV32-NEXT: lw a6, 8(a1) +; RV32-NEXT: bnez s3, .LBB3_16 +; RV32-NEXT: # %bb.15: # %_udiv-special-cases +; RV32-NEXT: mv s1, a3 +; RV32-NEXT: .LBB3_16: # %_udiv-special-cases +; RV32-NEXT: lw t1, 12(a1) +; RV32-NEXT: lw a1, 16(a1) +; RV32-NEXT: slli a0, a6, 31 +; RV32-NEXT: srli a2, t0, 1 +; RV32-NEXT: or a0, a2, a0 +; RV32-NEXT: slli a2, t0, 31 +; RV32-NEXT: srli a3, a7, 1 +; RV32-NEXT: or a2, a3, a2 +; RV32-NEXT: bnez a0, .LBB3_18 +; RV32-NEXT: # %bb.17: # %_udiv-special-cases +; RV32-NEXT: srli a3, a2, 1 +; RV32-NEXT: or a3, a2, a3 +; RV32-NEXT: srli a4, a3, 2 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 8 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 16 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: not a3, a3 +; RV32-NEXT: srli a4, a3, 1 +; RV32-NEXT: and a4, a4, t5 +; RV32-NEXT: sub a3, a3, a4 +; RV32-NEXT: and a4, a3, t4 +; RV32-NEXT: srli a3, a3, 2 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: add a3, a4, a3 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: and a3, a3, t3 +; RV32-NEXT: slli a4, a3, 8 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: slli a4, a3, 16 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: srli a3, a3, 24 +; RV32-NEXT: addi s5, a3, 32 +; RV32-NEXT: j .LBB3_19 +; RV32-NEXT: .LBB3_18: +; RV32-NEXT: srli a3, a0, 1 +; RV32-NEXT: or a3, a0, a3 +; RV32-NEXT: srli a4, a3, 2 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 8 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: srli a4, a3, 16 +; RV32-NEXT: or a3, a3, a4 +; RV32-NEXT: not a3, a3 +; RV32-NEXT: srli a4, a3, 1 +; RV32-NEXT: and a4, a4, t5 +; RV32-NEXT: sub a3, a3, a4 +; RV32-NEXT: and a4, a3, t4 +; RV32-NEXT: srli a3, a3, 2 +; RV32-NEXT: and a3, a3, t4 +; RV32-NEXT: add a3, a4, a3 +; RV32-NEXT: srli a4, a3, 4 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: and a3, a3, t3 +; RV32-NEXT: slli a4, a3, 8 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: slli a4, a3, 16 +; RV32-NEXT: add a3, a3, a4 +; RV32-NEXT: srli s5, a3, 24 +; RV32-NEXT: .LBB3_19: # %_udiv-special-cases +; RV32-NEXT: srli a3, t1, 1 +; RV32-NEXT: slli a4, a1, 31 +; RV32-NEXT: slli a5, t1, 31 +; RV32-NEXT: slli s4, a7, 31 +; RV32-NEXT: srli s6, a6, 1 +; RV32-NEXT: beqz s4, .LBB3_21 +; RV32-NEXT: # %bb.20: +; RV32-NEXT: srli s2, s4, 1 +; RV32-NEXT: or s2, s4, s2 +; RV32-NEXT: srli s7, s2, 2 +; RV32-NEXT: or s2, s2, s7 +; RV32-NEXT: srli s7, s2, 4 +; RV32-NEXT: or s2, s2, s7 +; RV32-NEXT: srli s7, s2, 8 +; RV32-NEXT: or s2, s2, s7 +; RV32-NEXT: srli s7, s2, 16 +; RV32-NEXT: or s2, s2, s7 +; RV32-NEXT: not s2, s2 +; RV32-NEXT: srli s7, s2, 1 +; RV32-NEXT: and s7, s7, t5 +; RV32-NEXT: sub s2, s2, s7 +; RV32-NEXT: and s7, s2, t4 +; RV32-NEXT: srli s2, s2, 2 +; RV32-NEXT: and s2, s2, t4 +; RV32-NEXT: add s2, s7, s2 +; RV32-NEXT: srli s7, s2, 4 +; RV32-NEXT: add s2, s2, s7 +; RV32-NEXT: and s2, s2, t3 +; RV32-NEXT: slli s7, s2, 8 +; RV32-NEXT: add s2, s2, s7 +; RV32-NEXT: slli s7, s2, 16 +; RV32-NEXT: add s2, s2, s7 +; RV32-NEXT: srli s2, s2, 24 +; RV32-NEXT: .LBB3_21: # %_udiv-special-cases +; RV32-NEXT: or s7, a4, a3 +; RV32-NEXT: or s6, s6, a5 +; RV32-NEXT: bnez s4, .LBB3_23 +; RV32-NEXT: # %bb.22: # %_udiv-special-cases +; RV32-NEXT: li s2, 128 +; RV32-NEXT: .LBB3_23: # %_udiv-special-cases +; RV32-NEXT: or s4, s6, s7 +; RV32-NEXT: addi a5, s5, 64 +; RV32-NEXT: addi a3, s2, 128 +; RV32-NEXT: or a0, a0, s7 +; RV32-NEXT: or a4, a2, s6 +; RV32-NEXT: or a4, a4, a0 +; RV32-NEXT: sltu a0, a3, s2 +; RV32-NEXT: bnez a4, .LBB3_26 +; RV32-NEXT: # %bb.24: # %_udiv-special-cases +; RV32-NEXT: mv a2, a0 +; RV32-NEXT: snez s2, s3 +; RV32-NEXT: beqz s7, .LBB3_27 +; RV32-NEXT: .LBB3_25: +; RV32-NEXT: srli s3, s7, 1 +; RV32-NEXT: or s3, s7, s3 +; RV32-NEXT: srli s5, s3, 2 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli s5, s3, 4 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli s5, s3, 8 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli s5, s3, 16 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: not s3, s3 +; RV32-NEXT: srli s5, s3, 1 +; RV32-NEXT: and t5, s5, t5 +; RV32-NEXT: sub t5, s3, t5 +; RV32-NEXT: and s3, t5, t4 +; RV32-NEXT: srli t5, t5, 2 +; RV32-NEXT: and t4, t5, t4 +; RV32-NEXT: add t4, s3, t4 +; RV32-NEXT: srli t5, t4, 4 +; RV32-NEXT: add t4, t4, t5 +; RV32-NEXT: and t3, t4, t3 +; RV32-NEXT: slli t4, t3, 8 +; RV32-NEXT: add t3, t3, t4 +; RV32-NEXT: slli t4, t3, 16 +; RV32-NEXT: add t3, t3, t4 +; RV32-NEXT: srli t3, t3, 24 +; RV32-NEXT: j .LBB3_28 +; RV32-NEXT: .LBB3_26: +; RV32-NEXT: snez a2, s4 +; RV32-NEXT: sltu s2, a5, s5 +; RV32-NEXT: addi a2, a2, -1 +; RV32-NEXT: and a2, a2, s2 +; RV32-NEXT: snez s2, s3 +; RV32-NEXT: bnez s7, .LBB3_25 +; RV32-NEXT: .LBB3_27: # %_udiv-special-cases +; RV32-NEXT: srli s3, s6, 1 +; RV32-NEXT: or s3, s6, s3 +; RV32-NEXT: srli s5, s3, 2 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli s5, s3, 4 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli s5, s3, 8 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: srli s5, s3, 16 +; RV32-NEXT: or s3, s3, s5 +; RV32-NEXT: not s3, s3 +; RV32-NEXT: srli s5, s3, 1 +; RV32-NEXT: and t5, s5, t5 +; RV32-NEXT: sub t5, s3, t5 +; RV32-NEXT: and s3, t5, t4 +; RV32-NEXT: srli t5, t5, 2 +; RV32-NEXT: and t4, t5, t4 +; RV32-NEXT: add t4, s3, t4 +; RV32-NEXT: srli t5, t4, 4 +; RV32-NEXT: add t4, t4, t5 +; RV32-NEXT: and t3, t4, t3 +; RV32-NEXT: slli t4, t3, 8 +; RV32-NEXT: add t3, t3, t4 +; RV32-NEXT: slli t4, t3, 16 +; RV32-NEXT: add t3, t3, t4 +; RV32-NEXT: srli t3, t3, 24 +; RV32-NEXT: addi t3, t3, 32 +; RV32-NEXT: .LBB3_28: # %_udiv-special-cases +; RV32-NEXT: xori t4, s0, 1 +; RV32-NEXT: addi s2, s2, -1 +; RV32-NEXT: bnez s4, .LBB3_30 +; RV32-NEXT: # %bb.29: # %_udiv-special-cases +; RV32-NEXT: mv t3, a5 +; RV32-NEXT: .LBB3_30: # %_udiv-special-cases +; RV32-NEXT: andi s11, a1, 1 +; RV32-NEXT: andi s8, t2, 1 +; RV32-NEXT: lw a1, 32(sp) # 4-byte Folded Reload +; RV32-NEXT: lw a5, 28(sp) # 4-byte Folded Reload +; RV32-NEXT: or s9, a1, a5 +; RV32-NEXT: or t2, a7, a6 +; RV32-NEXT: neg a1, t4 +; RV32-NEXT: and s0, s2, s0 +; RV32-NEXT: bnez a4, .LBB3_32 +; RV32-NEXT: # %bb.31: # %_udiv-special-cases +; RV32-NEXT: mv t3, a3 +; RV32-NEXT: .LBB3_32: # %_udiv-special-cases +; RV32-NEXT: lw a3, 20(sp) # 4-byte Folded Reload +; RV32-NEXT: lw a5, 24(sp) # 4-byte Folded Reload +; RV32-NEXT: or s10, a3, a5 +; RV32-NEXT: or a5, s9, s8 +; RV32-NEXT: or t4, t0, t1 +; RV32-NEXT: or t5, t2, s11 +; RV32-NEXT: and a1, s0, a1 +; RV32-NEXT: xori a3, a0, 1 +; RV32-NEXT: snez a4, a4 +; RV32-NEXT: neg a3, a3 +; RV32-NEXT: addi a4, a4, -1 +; RV32-NEXT: and a0, a4, a0 +; RV32-NEXT: sltu a4, s1, t3 +; RV32-NEXT: and t2, a0, a3 +; RV32-NEXT: mv a3, a4 +; RV32-NEXT: beq t6, a2, .LBB3_34 +; RV32-NEXT: # %bb.33: # %_udiv-special-cases +; RV32-NEXT: sltu a3, t6, a2 +; RV32-NEXT: .LBB3_34: # %_udiv-special-cases +; RV32-NEXT: or a0, a5, s10 +; RV32-NEXT: or t5, t5, t4 +; RV32-NEXT: sltu t4, a1, t2 +; RV32-NEXT: mv s0, a3 +; RV32-NEXT: beq a1, t2, .LBB3_36 +; RV32-NEXT: # %bb.35: # %_udiv-special-cases +; RV32-NEXT: mv s0, t4 +; RV32-NEXT: .LBB3_36: # %_udiv-special-cases +; RV32-NEXT: seqz a5, a0 +; RV32-NEXT: seqz t5, t5 +; RV32-NEXT: andi a0, s0, 1 +; RV32-NEXT: sub a2, t6, a2 +; RV32-NEXT: sub a1, a1, t2 +; RV32-NEXT: sub t2, a2, a4 +; RV32-NEXT: sltu a2, a1, a3 +; RV32-NEXT: add a2, t4, a2 +; RV32-NEXT: neg t4, a2 +; RV32-NEXT: sub a4, a1, a3 +; RV32-NEXT: or a1, a4, t4 +; RV32-NEXT: sub a3, s1, t3 +; RV32-NEXT: beqz a1, .LBB3_38 +; RV32-NEXT: # %bb.37: # %_udiv-special-cases +; RV32-NEXT: snez a1, a1 +; RV32-NEXT: or a2, a5, t5 +; RV32-NEXT: bnez a0, .LBB3_39 +; RV32-NEXT: j .LBB3_40 +; RV32-NEXT: .LBB3_38: +; RV32-NEXT: snez a1, t2 +; RV32-NEXT: sltiu a2, a3, 129 +; RV32-NEXT: xori a2, a2, 1 +; RV32-NEXT: or a1, a2, a1 +; RV32-NEXT: or a2, a5, t5 +; RV32-NEXT: beqz a0, .LBB3_40 +; RV32-NEXT: .LBB3_39: # %_udiv-special-cases +; RV32-NEXT: mv a1, a0 +; RV32-NEXT: .LBB3_40: # %_udiv-special-cases +; RV32-NEXT: or t6, a2, a1 +; RV32-NEXT: addi a1, t6, -1 +; RV32-NEXT: and a2, s11, a1 +; RV32-NEXT: and a5, a1, t1 +; RV32-NEXT: and t3, a1, a6 +; RV32-NEXT: and t5, a1, t0 +; RV32-NEXT: and a1, a1, a7 +; RV32-NEXT: bnez t6, .LBB3_57 +; RV32-NEXT: # %bb.41: # %_udiv-special-cases +; RV32-NEXT: or t6, t2, t4 +; RV32-NEXT: xori s0, a3, 128 +; RV32-NEXT: or s0, s0, a0 +; RV32-NEXT: or s0, s0, a4 +; RV32-NEXT: or t6, s0, t6 +; RV32-NEXT: beqz t6, .LBB3_57 +; RV32-NEXT: # %bb.42: # %udiv-bb1 +; RV32-NEXT: sw ra, 8(sp) # 4-byte Folded Spill +; RV32-NEXT: addi a1, a3, 1 +; RV32-NEXT: sw zero, 136(sp) +; RV32-NEXT: sw zero, 140(sp) +; RV32-NEXT: sw zero, 144(sp) +; RV32-NEXT: sw zero, 148(sp) +; RV32-NEXT: sw zero, 120(sp) +; RV32-NEXT: sw zero, 124(sp) +; RV32-NEXT: sw zero, 128(sp) +; RV32-NEXT: sw zero, 132(sp) +; RV32-NEXT: sw a7, 152(sp) +; RV32-NEXT: sw t0, 156(sp) +; RV32-NEXT: sw a6, 160(sp) +; RV32-NEXT: sw t1, 164(sp) +; RV32-NEXT: sw s11, 168(sp) +; RV32-NEXT: li a5, 128 +; RV32-NEXT: addi t3, sp, 152 +; RV32-NEXT: neg a2, a3 +; RV32-NEXT: seqz t5, a1 +; RV32-NEXT: sub a5, a5, a3 +; RV32-NEXT: add t2, t2, t5 +; RV32-NEXT: andi a3, a5, 31 +; RV32-NEXT: srli t5, a5, 3 +; RV32-NEXT: or t6, a1, t2 +; RV32-NEXT: xori a5, a3, 31 +; RV32-NEXT: andi a3, t5, 28 +; RV32-NEXT: seqz t6, t6 +; RV32-NEXT: sub ra, t3, a3 +; RV32-NEXT: add t6, a4, t6 +; RV32-NEXT: lw t3, 0(ra) +; RV32-NEXT: lw s0, 4(ra) +; RV32-NEXT: lw s1, 8(ra) +; RV32-NEXT: lw a3, 12(ra) +; RV32-NEXT: sltu a4, t6, a4 +; RV32-NEXT: or t5, a1, t6 +; RV32-NEXT: add t4, t4, a4 +; RV32-NEXT: or a4, t2, t4 +; RV32-NEXT: or a4, t5, a4 +; RV32-NEXT: srli t5, s1, 1 +; RV32-NEXT: seqz s2, a4 +; RV32-NEXT: add a0, a0, s2 +; RV32-NEXT: sll s2, a3, a2 +; RV32-NEXT: srl t5, t5, a5 +; RV32-NEXT: or t5, s2, t5 +; RV32-NEXT: srli s2, s0, 1 +; RV32-NEXT: sll s1, s1, a2 +; RV32-NEXT: srl s2, s2, a5 +; RV32-NEXT: or s2, s1, s2 +; RV32-NEXT: srli s1, t3, 1 +; RV32-NEXT: sll s0, s0, a2 +; RV32-NEXT: srl s1, s1, a5 +; RV32-NEXT: andi s3, a0, 1 +; RV32-NEXT: or s1, s0, s1 +; RV32-NEXT: or a0, a4, s3 +; RV32-NEXT: sll t3, t3, a2 +; RV32-NEXT: beqz a0, .LBB3_55 +; RV32-NEXT: # %bb.43: # %udiv-preheader +; RV32-NEXT: sw zero, 52(sp) # 4-byte Folded Spill +; RV32-NEXT: sw zero, 48(sp) # 4-byte Folded Spill +; RV32-NEXT: sw zero, 44(sp) # 4-byte Folded Spill +; RV32-NEXT: sw zero, 40(sp) # 4-byte Folded Spill +; RV32-NEXT: li s7, 0 +; RV32-NEXT: srli a3, a3, 1 +; RV32-NEXT: lw a0, 16(ra) +; RV32-NEXT: sw zero, 104(sp) +; RV32-NEXT: sw zero, 108(sp) +; RV32-NEXT: sw zero, 112(sp) +; RV32-NEXT: sw zero, 116(sp) +; RV32-NEXT: sw zero, 88(sp) +; RV32-NEXT: sw zero, 92(sp) +; RV32-NEXT: sw zero, 96(sp) +; RV32-NEXT: sw zero, 100(sp) +; RV32-NEXT: sw s11, 72(sp) +; RV32-NEXT: sw zero, 76(sp) +; RV32-NEXT: sw zero, 80(sp) +; RV32-NEXT: sw zero, 84(sp) +; RV32-NEXT: sw a7, 56(sp) +; RV32-NEXT: sw t0, 60(sp) +; RV32-NEXT: sw a6, 64(sp) +; RV32-NEXT: sw t1, 68(sp) +; RV32-NEXT: srli a4, a1, 3 +; RV32-NEXT: addi a6, sp, 56 +; RV32-NEXT: andi a7, a1, 31 +; RV32-NEXT: or t0, s9, s10 +; RV32-NEXT: srl a3, a3, a5 +; RV32-NEXT: andi a4, a4, 28 +; RV32-NEXT: xori a5, a7, 31 +; RV32-NEXT: snez a7, t0 +; RV32-NEXT: add a4, a6, a4 +; RV32-NEXT: add a7, s8, a7 +; RV32-NEXT: lw a6, 16(a4) +; RV32-NEXT: lw t0, 0(a4) +; RV32-NEXT: lw t1, 4(a4) +; RV32-NEXT: lw s0, 8(a4) +; RV32-NEXT: lw a4, 12(a4) +; RV32-NEXT: sll a0, a0, a2 +; RV32-NEXT: or a3, a0, a3 +; RV32-NEXT: slli a6, a6, 1 +; RV32-NEXT: slli a0, a4, 1 +; RV32-NEXT: slli a2, s0, 1 +; RV32-NEXT: slli s4, t1, 1 +; RV32-NEXT: sll a6, a6, a5 +; RV32-NEXT: sll a0, a0, a5 +; RV32-NEXT: sll s8, a2, a5 +; RV32-NEXT: sll s4, s4, a5 +; RV32-NEXT: srl a2, a4, a1 +; RV32-NEXT: or ra, a2, a6 +; RV32-NEXT: lw a6, 32(sp) # 4-byte Folded Reload +; RV32-NEXT: seqz a4, a6 +; RV32-NEXT: srl a2, s0, a1 +; RV32-NEXT: or a2, a2, a0 +; RV32-NEXT: lw a5, 20(sp) # 4-byte Folded Reload +; RV32-NEXT: or a0, a6, a5 +; RV32-NEXT: sub s5, a5, a4 +; RV32-NEXT: seqz a4, a0 +; RV32-NEXT: srl a0, t1, a1 +; RV32-NEXT: or a0, a0, s8 +; RV32-NEXT: lw a5, 28(sp) # 4-byte Folded Reload +; RV32-NEXT: sub t1, a5, a4 +; RV32-NEXT: sw t1, 36(sp) # 4-byte Folded Spill +; RV32-NEXT: sltu a4, a5, a4 +; RV32-NEXT: addi a7, a7, 1 +; RV32-NEXT: lw a5, 24(sp) # 4-byte Folded Reload +; RV32-NEXT: sub s6, a5, a4 +; RV32-NEXT: andi a4, a7, 1 +; RV32-NEXT: sw a4, 16(sp) # 4-byte Folded Spill +; RV32-NEXT: andi a5, a3, 1 +; RV32-NEXT: srl a3, t0, a1 +; RV32-NEXT: or a4, a3, s4 +; RV32-NEXT: addi a6, a6, -1 +; RV32-NEXT: sw a6, 12(sp) # 4-byte Folded Spill +; RV32-NEXT: li s11, 0 +; RV32-NEXT: li s10, 0 +; RV32-NEXT: j .LBB3_45 +; RV32-NEXT: .LBB3_44: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: lw s0, 28(sp) # 4-byte Folded Reload +; RV32-NEXT: and s0, a5, s0 +; RV32-NEXT: xor s8, t1, a7 +; RV32-NEXT: xor s9, a2, s0 +; RV32-NEXT: or s8, s9, s8 +; RV32-NEXT: li s9, 0 +; RV32-NEXT: li s8, 0 +; RV32-NEXT: sltu s4, a2, s0 +; RV32-NEXT: sub s0, a2, s0 +; RV32-NEXT: sub a7, t1, a7 +; RV32-NEXT: srli a2, s2, 31 +; RV32-NEXT: sub a0, a0, t0 +; RV32-NEXT: slli t0, t5, 1 +; RV32-NEXT: or t0, t0, a2 +; RV32-NEXT: srli a2, s1, 31 +; RV32-NEXT: slli s2, s2, 1 +; RV32-NEXT: or t1, s2, a2 +; RV32-NEXT: srli a2, t3, 31 +; RV32-NEXT: slli s1, s1, 1 +; RV32-NEXT: or s1, s1, a2 +; RV32-NEXT: slli t3, t3, 1 +; RV32-NEXT: lw a2, 52(sp) # 4-byte Folded Reload +; RV32-NEXT: or t3, a2, t3 +; RV32-NEXT: srli a2, t5, 31 +; RV32-NEXT: or s7, s7, a2 +; RV32-NEXT: sub a2, s0, ra +; RV32-NEXT: sltu s0, s0, ra +; RV32-NEXT: or t5, a1, t6 +; RV32-NEXT: sub a7, a7, s4 +; RV32-NEXT: or s2, t2, t4 +; RV32-NEXT: sub a0, a0, a6 +; RV32-NEXT: or a6, a1, t2 +; RV32-NEXT: or s4, t5, s2 +; RV32-NEXT: seqz t5, a1 +; RV32-NEXT: addi a1, a1, -1 +; RV32-NEXT: andi a5, a5, 1 +; RV32-NEXT: sw a5, 52(sp) # 4-byte Folded Spill +; RV32-NEXT: seqz a6, a6 +; RV32-NEXT: sub t2, t2, t5 +; RV32-NEXT: lw a5, 48(sp) # 4-byte Folded Reload +; RV32-NEXT: or s1, a5, s1 +; RV32-NEXT: lw a5, 44(sp) # 4-byte Folded Reload +; RV32-NEXT: or s2, a5, t1 +; RV32-NEXT: lw a5, 40(sp) # 4-byte Folded Reload +; RV32-NEXT: or t5, a5, t0 +; RV32-NEXT: andi a5, s7, 1 +; RV32-NEXT: sub ra, a7, s0 +; RV32-NEXT: snez a7, s4 +; RV32-NEXT: sltu t0, t6, a6 +; RV32-NEXT: sub t6, t6, a6 +; RV32-NEXT: add a7, s3, a7 +; RV32-NEXT: sub t4, t4, t0 +; RV32-NEXT: or a6, a1, t6 +; RV32-NEXT: addi a7, a7, 1 +; RV32-NEXT: or t0, t2, t4 +; RV32-NEXT: andi s3, a7, 1 +; RV32-NEXT: or a6, a6, t0 +; RV32-NEXT: or a6, a6, s3 +; RV32-NEXT: sub a4, a4, a3 +; RV32-NEXT: sw zero, 48(sp) # 4-byte Folded Spill +; RV32-NEXT: sw zero, 44(sp) # 4-byte Folded Spill +; RV32-NEXT: sw zero, 40(sp) # 4-byte Folded Spill +; RV32-NEXT: li s7, 0 +; RV32-NEXT: beqz a6, .LBB3_56 +; RV32-NEXT: .LBB3_45: # %udiv-do-while +; RV32-NEXT: # =>This Inner Loop Header: Depth=1 +; RV32-NEXT: srli a3, a2, 31 +; RV32-NEXT: slli a6, ra, 1 +; RV32-NEXT: or t1, a6, a3 +; RV32-NEXT: srli a3, a0, 31 +; RV32-NEXT: slli a2, a2, 1 +; RV32-NEXT: or a2, a2, a3 +; RV32-NEXT: beq s6, t1, .LBB3_47 +; RV32-NEXT: # %bb.46: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: sltu a3, s6, t1 +; RV32-NEXT: j .LBB3_48 +; RV32-NEXT: .LBB3_47: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: lw a3, 36(sp) # 4-byte Folded Reload +; RV32-NEXT: sltu a3, a3, a2 +; RV32-NEXT: .LBB3_48: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: srli a6, a4, 31 +; RV32-NEXT: slli a0, a0, 1 +; RV32-NEXT: slli a4, a4, 1 +; RV32-NEXT: or a0, a0, a6 +; RV32-NEXT: andi a5, a5, 1 +; RV32-NEXT: or a4, a4, a5 +; RV32-NEXT: beq s5, a0, .LBB3_50 +; RV32-NEXT: # %bb.49: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: sltu a5, s5, a0 +; RV32-NEXT: j .LBB3_51 +; RV32-NEXT: .LBB3_50: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: lw a5, 12(sp) # 4-byte Folded Reload +; RV32-NEXT: sltu a5, a5, a4 +; RV32-NEXT: .LBB3_51: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: lw a6, 36(sp) # 4-byte Folded Reload +; RV32-NEXT: xor a6, a6, a2 +; RV32-NEXT: xor a7, s6, t1 +; RV32-NEXT: or a6, a6, a7 +; RV32-NEXT: beqz a6, .LBB3_53 +; RV32-NEXT: # %bb.52: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: mv a5, a3 +; RV32-NEXT: .LBB3_53: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: srli a3, ra, 31 +; RV32-NEXT: lw a6, 16(sp) # 4-byte Folded Reload +; RV32-NEXT: sub a3, a6, a3 +; RV32-NEXT: sub a3, a3, a5 +; RV32-NEXT: slli a3, a3, 31 +; RV32-NEXT: srai a5, a3, 31 +; RV32-NEXT: lw a3, 24(sp) # 4-byte Folded Reload +; RV32-NEXT: and a7, a5, a3 +; RV32-NEXT: lw a3, 32(sp) # 4-byte Folded Reload +; RV32-NEXT: and a3, a5, a3 +; RV32-NEXT: lw a6, 20(sp) # 4-byte Folded Reload +; RV32-NEXT: and t0, a5, a6 +; RV32-NEXT: sltu a6, a4, a3 +; RV32-NEXT: mv ra, a6 +; RV32-NEXT: beq a0, t0, .LBB3_44 +; RV32-NEXT: # %bb.54: # %udiv-do-while +; RV32-NEXT: # in Loop: Header=BB3_45 Depth=1 +; RV32-NEXT: sltu ra, a0, t0 +; RV32-NEXT: j .LBB3_44 +; RV32-NEXT: .LBB3_55: +; RV32-NEXT: sw zero, 52(sp) # 4-byte Folded Spill +; RV32-NEXT: li s11, 0 +; RV32-NEXT: li s9, 0 +; RV32-NEXT: li s10, 0 +; RV32-NEXT: li s8, 0 +; RV32-NEXT: .LBB3_56: # %udiv-loop-exit +; RV32-NEXT: srli a0, s2, 31 +; RV32-NEXT: slli a1, t5, 1 +; RV32-NEXT: or a0, a1, a0 +; RV32-NEXT: srli a1, s1, 31 +; RV32-NEXT: slli s2, s2, 1 +; RV32-NEXT: or a2, s2, a1 +; RV32-NEXT: srli a3, t3, 31 +; RV32-NEXT: slli s1, s1, 1 +; RV32-NEXT: srli a4, t5, 31 +; RV32-NEXT: slli t3, t3, 1 +; RV32-NEXT: lw a1, 52(sp) # 4-byte Folded Reload +; RV32-NEXT: or a1, a1, t3 +; RV32-NEXT: or a3, s11, a3 +; RV32-NEXT: or a4, s8, a4 +; RV32-NEXT: or t5, a3, s1 +; RV32-NEXT: or t3, s9, a2 +; RV32-NEXT: or a5, s10, a0 +; RV32-NEXT: andi a2, a4, 1 +; RV32-NEXT: lw ra, 8(sp) # 4-byte Folded Reload +; RV32-NEXT: .LBB3_57: # %udiv-end +; RV32-NEXT: sw a1, 0(ra) +; RV32-NEXT: sw t5, 4(ra) +; RV32-NEXT: sw t3, 8(ra) +; RV32-NEXT: sw a5, 12(ra) +; RV32-NEXT: andi a2, a2, 1 +; RV32-NEXT: sb a2, 16(ra) +; RV32-NEXT: lw ra, 236(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s0, 232(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s1, 228(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s2, 224(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s3, 220(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s4, 216(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s5, 212(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s6, 208(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s7, 204(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s8, 200(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s9, 196(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s10, 192(sp) # 4-byte Folded Reload +; RV32-NEXT: lw s11, 188(sp) # 4-byte Folded Reload +; RV32-NEXT: addi sp, sp, 240 +; RV32-NEXT: ret +; +; RV64-LABEL: udiv_i129: +; RV64: # %bb.0: # %_udiv-special-cases +; RV64-NEXT: ld a3, 0(a2) +; RV64-NEXT: ld a4, 8(a2) +; RV64-NEXT: ld t1, 16(a2) +; RV64-NEXT: lui a2, 349525 +; RV64-NEXT: lui a5, 209715 +; RV64-NEXT: lui a6, 61681 +; RV64-NEXT: addi t0, a2, 1365 +; RV64-NEXT: addi a7, a5, 819 +; RV64-NEXT: addi a6, a6, -241 +; RV64-NEXT: slli a2, t0, 32 +; RV64-NEXT: slli a5, a7, 32 +; RV64-NEXT: slli t2, a6, 32 +; RV64-NEXT: add t0, t0, a2 +; RV64-NEXT: add a7, a7, a5 +; RV64-NEXT: add a6, a6, t2 +; RV64-NEXT: srli a2, a4, 1 +; RV64-NEXT: slli a5, t1, 63 +; RV64-NEXT: slli t2, a4, 63 +; RV64-NEXT: or t3, a5, a2 +; RV64-NEXT: srli a2, a3, 1 +; RV64-NEXT: or t4, a2, t2 +; RV64-NEXT: bnez t3, .LBB3_2 +; RV64-NEXT: # %bb.1: # %_udiv-special-cases +; RV64-NEXT: srli a2, t4, 1 +; RV64-NEXT: or a2, t4, a2 +; RV64-NEXT: srli a5, a2, 2 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 4 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 8 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 16 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 32 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: not a2, a2 +; RV64-NEXT: srli a5, a2, 1 +; RV64-NEXT: and a5, a5, t0 +; RV64-NEXT: sub a2, a2, a5 +; RV64-NEXT: and a5, a2, a7 +; RV64-NEXT: srli a2, a2, 2 +; RV64-NEXT: and a2, a2, a7 +; RV64-NEXT: add a2, a5, a2 +; RV64-NEXT: srli a5, a2, 4 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: and a2, a2, a6 +; RV64-NEXT: slli a5, a2, 8 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: slli a5, a2, 16 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: slli a5, a2, 32 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: srli a2, a2, 56 +; RV64-NEXT: addi t2, a2, 64 +; RV64-NEXT: j .LBB3_3 +; RV64-NEXT: .LBB3_2: +; RV64-NEXT: srli a2, t3, 1 +; RV64-NEXT: or a2, t3, a2 +; RV64-NEXT: srli a5, a2, 2 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 4 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 8 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 16 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 32 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: not a2, a2 +; RV64-NEXT: srli a5, a2, 1 +; RV64-NEXT: and a5, a5, t0 +; RV64-NEXT: sub a2, a2, a5 +; RV64-NEXT: and a5, a2, a7 +; RV64-NEXT: srli a2, a2, 2 +; RV64-NEXT: and a2, a2, a7 +; RV64-NEXT: add a2, a5, a2 +; RV64-NEXT: srli a5, a2, 4 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: and a2, a2, a6 +; RV64-NEXT: slli a5, a2, 8 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: slli a5, a2, 16 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: slli a5, a2, 32 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: srli t2, a2, 56 +; RV64-NEXT: .LBB3_3: # %_udiv-special-cases +; RV64-NEXT: addi sp, sp, -192 +; RV64-NEXT: sd s0, 184(sp) # 8-byte Folded Spill +; RV64-NEXT: sd s1, 176(sp) # 8-byte Folded Spill +; RV64-NEXT: sd s2, 168(sp) # 8-byte Folded Spill +; RV64-NEXT: sd s3, 160(sp) # 8-byte Folded Spill +; RV64-NEXT: sd s4, 152(sp) # 8-byte Folded Spill +; RV64-NEXT: sd s5, 144(sp) # 8-byte Folded Spill +; RV64-NEXT: sd s6, 136(sp) # 8-byte Folded Spill +; RV64-NEXT: slli a2, a3, 63 +; RV64-NEXT: li t5, 128 +; RV64-NEXT: bnez a2, .LBB3_5 +; RV64-NEXT: # %bb.4: # %_udiv-special-cases +; RV64-NEXT: li s0, 128 +; RV64-NEXT: j .LBB3_6 +; RV64-NEXT: .LBB3_5: +; RV64-NEXT: srli a5, a2, 1 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 2 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 4 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 8 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 16 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: srli a5, a2, 32 +; RV64-NEXT: or a2, a2, a5 +; RV64-NEXT: not a2, a2 +; RV64-NEXT: srli a5, a2, 1 +; RV64-NEXT: and a5, a5, t0 +; RV64-NEXT: sub a2, a2, a5 +; RV64-NEXT: and a5, a2, a7 +; RV64-NEXT: srli a2, a2, 2 +; RV64-NEXT: and a2, a2, a7 +; RV64-NEXT: add a2, a5, a2 +; RV64-NEXT: srli a5, a2, 4 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: and a2, a2, a6 +; RV64-NEXT: slli a5, a2, 8 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: slli a5, a2, 16 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: slli a5, a2, 32 +; RV64-NEXT: add a2, a2, a5 +; RV64-NEXT: srli s0, a2, 56 +; RV64-NEXT: .LBB3_6: # %_udiv-special-cases +; RV64-NEXT: ld a5, 0(a1) +; RV64-NEXT: ld a2, 8(a1) +; RV64-NEXT: ld s2, 16(a1) +; RV64-NEXT: or a1, t4, t3 +; RV64-NEXT: addi s1, s0, 128 +; RV64-NEXT: bnez a1, .LBB3_8 +; RV64-NEXT: # %bb.7: # %_udiv-special-cases +; RV64-NEXT: mv t2, s1 +; RV64-NEXT: .LBB3_8: # %_udiv-special-cases +; RV64-NEXT: snez s3, a1 +; RV64-NEXT: srli a1, a2, 1 +; RV64-NEXT: slli t3, s2, 63 +; RV64-NEXT: slli t4, a2, 63 +; RV64-NEXT: or a1, t3, a1 +; RV64-NEXT: srli t3, a5, 1 +; RV64-NEXT: or t6, t3, t4 +; RV64-NEXT: bnez a1, .LBB3_10 +; RV64-NEXT: # %bb.9: # %_udiv-special-cases +; RV64-NEXT: srli t3, t6, 1 +; RV64-NEXT: or t3, t6, t3 +; RV64-NEXT: srli t4, t3, 2 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 4 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 8 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 16 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 32 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: not t3, t3 +; RV64-NEXT: srli t4, t3, 1 +; RV64-NEXT: and t4, t4, t0 +; RV64-NEXT: sub t3, t3, t4 +; RV64-NEXT: and t4, t3, a7 +; RV64-NEXT: srli t3, t3, 2 +; RV64-NEXT: and t3, t3, a7 +; RV64-NEXT: add t3, t4, t3 +; RV64-NEXT: srli t4, t3, 4 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: and t3, t3, a6 +; RV64-NEXT: slli t4, t3, 8 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: slli t4, t3, 16 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: slli t4, t3, 32 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: srli t3, t3, 56 +; RV64-NEXT: addi s4, t3, 64 +; RV64-NEXT: j .LBB3_11 +; RV64-NEXT: .LBB3_10: +; RV64-NEXT: srli t3, a1, 1 +; RV64-NEXT: or t3, a1, t3 +; RV64-NEXT: srli t4, t3, 2 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 4 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 8 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 16 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: srli t4, t3, 32 +; RV64-NEXT: or t3, t3, t4 +; RV64-NEXT: not t3, t3 +; RV64-NEXT: srli t4, t3, 1 +; RV64-NEXT: and t4, t4, t0 +; RV64-NEXT: sub t3, t3, t4 +; RV64-NEXT: and t4, t3, a7 +; RV64-NEXT: srli t3, t3, 2 +; RV64-NEXT: and t3, t3, a7 +; RV64-NEXT: add t3, t4, t3 +; RV64-NEXT: srli t4, t3, 4 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: and t3, t3, a6 +; RV64-NEXT: slli t4, t3, 8 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: slli t4, t3, 16 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: slli t4, t3, 32 +; RV64-NEXT: add t3, t3, t4 +; RV64-NEXT: srli s4, t3, 56 +; RV64-NEXT: .LBB3_11: # %_udiv-special-cases +; RV64-NEXT: andi t4, s2, 1 +; RV64-NEXT: andi t1, t1, 1 +; RV64-NEXT: or t3, a3, a4 +; RV64-NEXT: or s2, a5, a2 +; RV64-NEXT: sltu s0, s1, s0 +; RV64-NEXT: slli s1, a5, 63 +; RV64-NEXT: addi s3, s3, -1 +; RV64-NEXT: beqz s1, .LBB3_13 +; RV64-NEXT: # %bb.12: +; RV64-NEXT: srli t5, s1, 1 +; RV64-NEXT: or t5, s1, t5 +; RV64-NEXT: srli s1, t5, 2 +; RV64-NEXT: or t5, t5, s1 +; RV64-NEXT: srli s1, t5, 4 +; RV64-NEXT: or t5, t5, s1 +; RV64-NEXT: srli s1, t5, 8 +; RV64-NEXT: or t5, t5, s1 +; RV64-NEXT: srli s1, t5, 16 +; RV64-NEXT: or t5, t5, s1 +; RV64-NEXT: srli s1, t5, 32 +; RV64-NEXT: or t5, t5, s1 +; RV64-NEXT: not t5, t5 +; RV64-NEXT: srli s1, t5, 1 +; RV64-NEXT: and t0, s1, t0 +; RV64-NEXT: sub t0, t5, t0 +; RV64-NEXT: and t5, t0, a7 +; RV64-NEXT: srli t0, t0, 2 +; RV64-NEXT: and a7, t0, a7 +; RV64-NEXT: add a7, t5, a7 +; RV64-NEXT: srli t0, a7, 4 +; RV64-NEXT: add a7, a7, t0 +; RV64-NEXT: and a6, a7, a6 +; RV64-NEXT: slli a7, a6, 8 +; RV64-NEXT: add a6, a6, a7 +; RV64-NEXT: slli a7, a6, 16 +; RV64-NEXT: add a6, a6, a7 +; RV64-NEXT: slli a7, a6, 32 +; RV64-NEXT: add a6, a6, a7 +; RV64-NEXT: srli t5, a6, 56 +; RV64-NEXT: .LBB3_13: # %_udiv-special-cases +; RV64-NEXT: or t0, t3, t1 +; RV64-NEXT: or a6, s2, t4 +; RV64-NEXT: and a7, s3, s0 +; RV64-NEXT: or t6, t6, a1 +; RV64-NEXT: addi s0, t5, 128 +; RV64-NEXT: bnez t6, .LBB3_15 +; RV64-NEXT: # %bb.14: # %_udiv-special-cases +; RV64-NEXT: mv s4, s0 +; RV64-NEXT: .LBB3_15: # %_udiv-special-cases +; RV64-NEXT: seqz a1, t0 +; RV64-NEXT: sltu t0, s0, t5 +; RV64-NEXT: snez t5, t6 +; RV64-NEXT: addi t5, t5, -1 +; RV64-NEXT: and t0, t5, t0 +; RV64-NEXT: sltu t5, t2, s4 +; RV64-NEXT: seqz a6, a6 +; RV64-NEXT: mv t6, t5 +; RV64-NEXT: beq a7, t0, .LBB3_17 +; RV64-NEXT: # %bb.16: # %_udiv-special-cases +; RV64-NEXT: sltu t6, a7, t0 +; RV64-NEXT: .LBB3_17: # %_udiv-special-cases +; RV64-NEXT: or a1, a1, a6 +; RV64-NEXT: andi a6, t6, 1 +; RV64-NEXT: sub a7, a7, t0 +; RV64-NEXT: sub t5, a7, t5 +; RV64-NEXT: sub a7, t2, s4 +; RV64-NEXT: beqz a6, .LBB3_19 +; RV64-NEXT: # %bb.18: # %_udiv-special-cases +; RV64-NEXT: mv t0, a6 +; RV64-NEXT: j .LBB3_20 +; RV64-NEXT: .LBB3_19: +; RV64-NEXT: sltiu t0, a7, 129 +; RV64-NEXT: xori t0, t0, 1 +; RV64-NEXT: snez t2, t5 +; RV64-NEXT: or t0, t0, t2 +; RV64-NEXT: .LBB3_20: # %_udiv-special-cases +; RV64-NEXT: or t6, a1, t0 +; RV64-NEXT: addi a1, t6, -1 +; RV64-NEXT: and t2, t4, a1 +; RV64-NEXT: and t0, a1, a2 +; RV64-NEXT: and a1, a1, a5 +; RV64-NEXT: bnez t6, .LBB3_30 +; RV64-NEXT: # %bb.21: # %_udiv-special-cases +; RV64-NEXT: xori t6, a7, 128 +; RV64-NEXT: or t6, t6, a6 +; RV64-NEXT: or t6, t6, t5 +; RV64-NEXT: beqz t6, .LBB3_30 +; RV64-NEXT: # %bb.22: # %udiv-bb1 +; RV64-NEXT: addi a1, a7, 1 +; RV64-NEXT: sd zero, 64(sp) +; RV64-NEXT: sd zero, 72(sp) +; RV64-NEXT: sd zero, 80(sp) +; RV64-NEXT: sd zero, 88(sp) +; RV64-NEXT: sd a5, 96(sp) +; RV64-NEXT: sd a2, 104(sp) +; RV64-NEXT: sd t4, 112(sp) +; RV64-NEXT: li t0, 128 +; RV64-NEXT: addi t2, sp, 96 +; RV64-NEXT: neg s1, a7 +; RV64-NEXT: seqz t6, a1 +; RV64-NEXT: sub a7, t0, a7 +; RV64-NEXT: add t5, t5, t6 +; RV64-NEXT: andi t0, a7, 63 +; RV64-NEXT: srli a7, a7, 3 +; RV64-NEXT: or t6, a1, t5 +; RV64-NEXT: xori s2, t0, 63 +; RV64-NEXT: andi a7, a7, 24 +; RV64-NEXT: seqz t0, t6 +; RV64-NEXT: sub s3, t2, a7 +; RV64-NEXT: add a6, a6, t0 +; RV64-NEXT: ld t2, 0(s3) +; RV64-NEXT: ld s4, 8(s3) +; RV64-NEXT: andi a7, a6, 1 +; RV64-NEXT: or t6, t6, a7 +; RV64-NEXT: srli a6, t2, 1 +; RV64-NEXT: sll t0, s4, s1 +; RV64-NEXT: srl a6, a6, s2 +; RV64-NEXT: or t0, t0, a6 +; RV64-NEXT: sll a6, t2, s1 +; RV64-NEXT: li t2, 0 +; RV64-NEXT: beqz t6, .LBB3_28 +; RV64-NEXT: # %bb.23: # %udiv-preheader +; RV64-NEXT: li t6, 0 +; RV64-NEXT: li s0, 0 +; RV64-NEXT: srli s4, s4, 1 +; RV64-NEXT: ld s3, 16(s3) +; RV64-NEXT: sd zero, 32(sp) +; RV64-NEXT: sd zero, 40(sp) +; RV64-NEXT: sd zero, 48(sp) +; RV64-NEXT: sd zero, 56(sp) +; RV64-NEXT: sd a5, 0(sp) +; RV64-NEXT: sd a2, 8(sp) +; RV64-NEXT: sd t4, 16(sp) +; RV64-NEXT: sd zero, 24(sp) +; RV64-NEXT: srli a2, a1, 3 +; RV64-NEXT: srl a5, s4, s2 +; RV64-NEXT: mv t4, sp +; RV64-NEXT: snez t3, t3 +; RV64-NEXT: andi a2, a2, 24 +; RV64-NEXT: add t1, t1, t3 +; RV64-NEXT: add a2, t4, a2 +; RV64-NEXT: ld t3, 0(a2) +; RV64-NEXT: ld t4, 8(a2) +; RV64-NEXT: ld a2, 16(a2) +; RV64-NEXT: sll s1, s3, s1 +; RV64-NEXT: andi s2, a1, 63 +; RV64-NEXT: xori s2, s2, 63 +; RV64-NEXT: or s3, s1, a5 +; RV64-NEXT: slli a2, a2, 1 +; RV64-NEXT: slli a5, t4, 1 +; RV64-NEXT: sll a2, a2, s2 +; RV64-NEXT: sll s2, a5, s2 +; RV64-NEXT: srl s1, t4, a1 +; RV64-NEXT: or s1, s1, a2 +; RV64-NEXT: seqz a2, a3 +; RV64-NEXT: sub a2, a4, a2 +; RV64-NEXT: addi a5, t1, 1 +; RV64-NEXT: andi a5, a5, 1 +; RV64-NEXT: andi s3, s3, 1 +; RV64-NEXT: srl t1, t3, a1 +; RV64-NEXT: or s2, t1, s2 +; RV64-NEXT: addi t1, a3, -1 +; RV64-NEXT: j .LBB3_26 +; RV64-NEXT: .LBB3_24: # %udiv-do-while +; RV64-NEXT: # in Loop: Header=BB3_26 Depth=1 +; RV64-NEXT: sltu t3, a2, s4 +; RV64-NEXT: .LBB3_25: # %udiv-do-while +; RV64-NEXT: # in Loop: Header=BB3_26 Depth=1 +; RV64-NEXT: srli s1, s1, 63 +; RV64-NEXT: sub t4, a5, s1 +; RV64-NEXT: sub t3, t4, t3 +; RV64-NEXT: slli t3, t3, 63 +; RV64-NEXT: srai s1, t3, 63 +; RV64-NEXT: and s3, s1, a4 +; RV64-NEXT: li t3, 0 +; RV64-NEXT: li t4, 0 +; RV64-NEXT: srli s5, a6, 63 +; RV64-NEXT: sub s4, s4, s3 +; RV64-NEXT: slli s3, t0, 1 +; RV64-NEXT: or s3, s3, s5 +; RV64-NEXT: srli t0, t0, 63 +; RV64-NEXT: slli a6, a6, 1 +; RV64-NEXT: or a6, t2, a6 +; RV64-NEXT: seqz t2, a1 +; RV64-NEXT: or s0, s0, t0 +; RV64-NEXT: or s5, a1, t5 +; RV64-NEXT: sub t5, t5, t2 +; RV64-NEXT: and s6, s1, a3 +; RV64-NEXT: addi a1, a1, -1 +; RV64-NEXT: andi t2, s1, 1 +; RV64-NEXT: or t0, t6, s3 +; RV64-NEXT: sltu t6, s2, s6 +; RV64-NEXT: snez s5, s5 +; RV64-NEXT: andi s3, s0, 1 +; RV64-NEXT: sub s1, s4, t6 +; RV64-NEXT: add a7, a7, s5 +; RV64-NEXT: addi a7, a7, 1 +; RV64-NEXT: andi a7, a7, 1 +; RV64-NEXT: or t6, a1, t5 +; RV64-NEXT: or s4, t6, a7 +; RV64-NEXT: sub s2, s2, s6 +; RV64-NEXT: li t6, 0 +; RV64-NEXT: li s0, 0 +; RV64-NEXT: beqz s4, .LBB3_29 +; RV64-NEXT: .LBB3_26: # %udiv-do-while +; RV64-NEXT: # =>This Inner Loop Header: Depth=1 +; RV64-NEXT: srli t3, s2, 63 +; RV64-NEXT: slli t4, s1, 1 +; RV64-NEXT: slli s2, s2, 1 +; RV64-NEXT: or s4, t4, t3 +; RV64-NEXT: andi t3, s3, 1 +; RV64-NEXT: or s2, s2, t3 +; RV64-NEXT: bne a2, s4, .LBB3_24 +; RV64-NEXT: # %bb.27: # in Loop: Header=BB3_26 Depth=1 +; RV64-NEXT: sltu t3, t1, s2 +; RV64-NEXT: j .LBB3_25 +; RV64-NEXT: .LBB3_28: +; RV64-NEXT: li t3, 0 +; RV64-NEXT: li t4, 0 +; RV64-NEXT: .LBB3_29: # %udiv-loop-exit +; RV64-NEXT: srli a2, a6, 63 +; RV64-NEXT: slli a3, t0, 1 +; RV64-NEXT: srli a4, t0, 63 +; RV64-NEXT: slli a6, a6, 1 +; RV64-NEXT: or a1, t2, a6 +; RV64-NEXT: or a2, t3, a2 +; RV64-NEXT: or a4, t4, a4 +; RV64-NEXT: or t0, a2, a3 +; RV64-NEXT: andi t2, a4, 1 +; RV64-NEXT: .LBB3_30: # %udiv-end +; RV64-NEXT: andi a2, t2, 1 +; RV64-NEXT: sd a1, 0(a0) +; RV64-NEXT: sd t0, 8(a0) +; RV64-NEXT: sb a2, 16(a0) +; RV64-NEXT: ld s0, 184(sp) # 8-byte Folded Reload +; RV64-NEXT: ld s1, 176(sp) # 8-byte Folded Reload +; RV64-NEXT: ld s2, 168(sp) # 8-byte Folded Reload +; RV64-NEXT: ld s3, 160(sp) # 8-byte Folded Reload +; RV64-NEXT: ld s4, 152(sp) # 8-byte Folded Reload +; RV64-NEXT: ld s5, 144(sp) # 8-byte Folded Reload +; RV64-NEXT: ld s6, 136(sp) # 8-byte Folded Reload +; RV64-NEXT: addi sp, sp, 192 +; RV64-NEXT: ret %res = udiv i129 %x, %y ret i129 %res } diff --git a/llvm/test/CodeGen/WebAssembly/simd-relaxed-dot.ll b/llvm/test/CodeGen/WebAssembly/simd-relaxed-dot.ll new file mode 100644 index 0000000..9716cbe --- /dev/null +++ b/llvm/test/CodeGen/WebAssembly/simd-relaxed-dot.ll @@ -0,0 +1,104 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 +; RUN: llc < %s -disable-wasm-fallthrough-return-opt -wasm-disable-explicit-locals -wasm-keep-registers -mattr=+fp16,+simd128,+relaxed-simd | FileCheck %s + +target triple = "wasm32" +; relaxed_dot stands for relaxed_dot_i8x16_i7x16_s, as in td +; relaxed_dot_add stands for i32x4.relaxed_dot_i8x16_i7x16_add_s, as in td + +define <8 x i16> @relaxed_dot_sext_1(<16 x i8> %a, <16 x i8> %b) { +; CHECK-LABEL: relaxed_dot_sext_1: +; CHECK: .functype relaxed_dot_sext_1 (v128, v128) -> (v128) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: i16x8.relaxed_dot_i8x16_i7x16_s $push0=, $0, $1 +; CHECK-NEXT: return $pop0 + %sext1 = sext <16 x i8> %a to <16 x i16> + %sext2 = sext <16 x i8> %b to <16 x i16> + %mul = mul <16 x i16> %sext1, %sext2 + %shuffle1 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> + %shuffle2 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> + %res = add <8 x i16> %shuffle1, %shuffle2 + ret <8 x i16> %res +} + + +define <8 x i16> @relaxed_dot_sext_2(<16 x i8> %a, <16 x i8> %b) { +; CHECK-LABEL: relaxed_dot_sext_2: +; CHECK: .functype relaxed_dot_sext_2 (v128, v128) -> (v128) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: i16x8.relaxed_dot_i8x16_i7x16_s $push0=, $0, $1 +; CHECK-NEXT: return $pop0 + %sext1 = sext <16 x i8> %a to <16 x i16> + %sext2 = sext <16 x i8> %b to <16 x i16> + %mul = mul <16 x i16> %sext1, %sext2 + %shuffle1 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> + %shuffle2 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> + %res = add <8 x i16> %shuffle2, %shuffle1 + ret <8 x i16> %res +} + +define <8 x i16> @relaxed_dot_sext_self(<16 x i8> %v) { +; CHECK-LABEL: relaxed_dot_sext_self: +; CHECK: .functype relaxed_dot_sext_self (v128) -> (v128) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: i16x8.relaxed_dot_i8x16_i7x16_s $push0=, $0, $0 +; CHECK-NEXT: return $pop0 + %sext = sext <16 x i8> %v to <16 x i16> + %mul = mul <16 x i16> %sext, %sext + %shuffle1 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> + %shuffle2 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> + %res = add <8 x i16> %shuffle1, %shuffle2 + ret <8 x i16> %res +} + +define <4 x i32> @relaxed_dot_add_from_relaxed_dot(<16 x i8> %a, <16 x i8> %b, <4 x i32> %c) { +; CHECK-LABEL: relaxed_dot_add_from_relaxed_dot: +; CHECK: .functype relaxed_dot_add_from_relaxed_dot (v128, v128, v128) -> (v128) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: i32x4.relaxed_dot_i8x16_i7x16_add_s $push0=, $0, $1, $2 +; CHECK-NEXT: return $pop0 + %relaxed_dot_call = call <8 x i16> @llvm.wasm.relaxed.dot.i8x16.i7x16.signed(<16 x i8> %a, <16 x i8> %b) + %sext = call <4 x i32> @llvm.wasm.extadd.pairwise.signed.v4i32(<8 x i16> %relaxed_dot_call) + %res = add <4 x i32> %sext, %c + ret <4 x i32> %res +} + +; INFO: Negative test +define <8 x i16> @relaxed_dot_zext(<16 x i8> %a, <16 x i8> %b) { +; CHECK-LABEL: relaxed_dot_zext: +; CHECK: .functype relaxed_dot_zext (v128, v128) -> (v128) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: i16x8.extmul_low_i8x16_u $push6=, $0, $1 +; CHECK-NEXT: local.tee $push5=, $2=, $pop6 +; CHECK-NEXT: i16x8.extmul_high_i8x16_u $push4=, $0, $1 +; CHECK-NEXT: local.tee $push3=, $1=, $pop4 +; CHECK-NEXT: i8x16.shuffle $push1=, $pop5, $pop3, 0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29 +; CHECK-NEXT: i8x16.shuffle $push0=, $2, $1, 2, 3, 6, 7, 10, 11, 14, 15, 18, 19, 22, 23, 26, 27, 30, 31 +; CHECK-NEXT: i16x8.add $push2=, $pop1, $pop0 +; CHECK-NEXT: return $pop2 + %zext1 = zext <16 x i8> %a to <16 x i16> + %zext2 = zext <16 x i8> %b to <16 x i16> + %mul = mul <16 x i16> %zext1, %zext2 + %shuffle1 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14> + %shuffle2 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15> + %res = add <8 x i16> %shuffle1, %shuffle2 + ret <8 x i16> %res + +} + +; INFO: Negative test +define <8 x i16> @relaxed_dot_wrong_shuffle(<16 x i8> %a, <16 x i8> %b) { +; CHECK-LABEL: relaxed_dot_wrong_shuffle: +; CHECK: .functype relaxed_dot_wrong_shuffle (v128, v128) -> (v128) +; CHECK-NEXT: # %bb.0: +; CHECK-NEXT: i16x8.extmul_low_i8x16_s $push1=, $0, $1 +; CHECK-NEXT: i16x8.extmul_high_i8x16_s $push0=, $0, $1 +; CHECK-NEXT: i16x8.add $push2=, $pop1, $pop0 +; CHECK-NEXT: return $pop2 + %sext1 = sext <16 x i8> %a to <16 x i16> + %sext2 = sext <16 x i8> %b to <16 x i16> + %mul = mul <16 x i16> %sext1, %sext2 + %shuffle1 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> + %shuffle2 = shufflevector <16 x i16> %mul, <16 x i16> poison, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> + %res = add <8 x i16> %shuffle1, %shuffle2 + ret <8 x i16> %res +} diff --git a/llvm/test/CodeGen/X86/cpus-intel.ll b/llvm/test/CodeGen/X86/cpus-intel.ll index 71253c8..646629d 100644 --- a/llvm/test/CodeGen/X86/cpus-intel.ll +++ b/llvm/test/CodeGen/X86/cpus-intel.ll @@ -39,6 +39,7 @@ ; RUN: llc < %s -o /dev/null -mtriple=i686-unknown-unknown -mcpu=gracemont 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=i686-unknown-unknown -mcpu=pantherlake 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=i686-unknown-unknown -mcpu=wildcatlake 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty +; RUN: llc < %s -o /dev/null -mtriple=i686-unknown-unknown -mcpu=novalake 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=i686-unknown-unknown -mcpu=clearwaterforest 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=i686-unknown-unknown -mcpu=diamondrapids 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty @@ -106,6 +107,7 @@ ; RUN: llc < %s -o /dev/null -mtriple=x86_64-unknown-unknown -mcpu=gracemont 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=x86_64-unknown-unknown -mcpu=pantherlake 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=x86_64-unknown-unknown -mcpu=wildcatlake 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty +; RUN: llc < %s -o /dev/null -mtriple=x86_64-unknown-unknown -mcpu=novalake 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=x86_64-unknown-unknown -mcpu=clearwaterforest 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty ; RUN: llc < %s -o /dev/null -mtriple=x86_64-unknown-unknown -mcpu=diamondrapids 2>&1 | FileCheck %s --check-prefix=CHECK-NO-ERROR --allow-empty diff --git a/llvm/test/CodeGen/X86/isel-fpclass.ll b/llvm/test/CodeGen/X86/isel-fpclass.ll index c2b7068..df04b67 100644 --- a/llvm/test/CodeGen/X86/isel-fpclass.ll +++ b/llvm/test/CodeGen/X86/isel-fpclass.ll @@ -1,10 +1,10 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 -; RUN: llc < %s -mtriple=i686-linux | FileCheck %s -check-prefixes=X86,X86-SDAGISEL +; RUN: llc < %s -mtriple=i686-linux | FileCheck %s -check-prefixes=X86 ; RUN: llc < %s -mtriple=x86_64-linux | FileCheck %s -check-prefixes=X64,X64-SDAGISEL ; RUN: llc < %s -mtriple=i686-linux -fast-isel -fast-isel-abort=1 | FileCheck %s -check-prefixes=X86-FASTISEL ; RUN: llc < %s -mtriple=x86_64-linux -fast-isel -fast-isel-abort=1 | FileCheck %s -check-prefixes=X64,X64-FASTISEL -; RUN: llc < %s -mtriple=i686-linux -global-isel -global-isel-abort=1 | FileCheck %s -check-prefixes=X86,X86-GISEL -; RUN: llc < %s -mtriple=x86_64-linux -global-isel -global-isel-abort=1 | FileCheck %s -check-prefixes=X64-GISEL +; RUN: llc < %s -mtriple=i686-linux -global-isel -global-isel-abort=2 | FileCheck %s -check-prefixes=X86 +; RUN: llc < %s -mtriple=x86_64-linux -global-isel -global-isel-abort=2 | FileCheck %s -check-prefixes=X64,X64-GISEL define i1 @isnone_f(float %x) nounwind { ; X86-LABEL: isnone_f: @@ -23,11 +23,6 @@ define i1 @isnone_f(float %x) nounwind { ; X86-FASTISEL-NEXT: fstp %st(0) ; X86-FASTISEL-NEXT: xorl %eax, %eax ; X86-FASTISEL-NEXT: retl -; -; X64-GISEL-LABEL: isnone_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: xorl %eax, %eax -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 0) ret i1 %0 @@ -50,27 +45,22 @@ define i1 @isany_f(float %x) nounwind { ; X86-FASTISEL-NEXT: fstp %st(0) ; X86-FASTISEL-NEXT: movb $1, %al ; X86-FASTISEL-NEXT: retl -; -; X64-GISEL-LABEL: isany_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movb $1, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 1023) ret i1 %0 } define i1 @issignaling_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: issignaling_f: -; X86-SDAGISEL: # %bb.0: -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X86-SDAGISEL-NEXT: setl %cl -; X86-SDAGISEL-NEXT: cmpl $2139095041, %eax # imm = 0x7F800001 -; X86-SDAGISEL-NEXT: setge %al -; X86-SDAGISEL-NEXT: andb %cl, %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: issignaling_f: +; X86: # %bb.0: +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 +; X86-NEXT: setl %cl +; X86-NEXT: cmpl $2139095041, %eax # imm = 0x7F800001 +; X86-NEXT: setge %al +; X86-NEXT: andb %cl, %al +; X86-NEXT: retl ; ; X64-LABEL: issignaling_f: ; X64: # %bb.0: @@ -97,44 +87,18 @@ define i1 @issignaling_f(float %x) nounwind { ; X86-FASTISEL-NEXT: andb %cl, %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: issignaling_f: -; X86-GISEL: # %bb.0: -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: seta %dl -; X86-GISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X86-GISEL-NEXT: setb %al -; X86-GISEL-NEXT: andb %dl, %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: issignaling_f: -; X64-GISEL: # %bb.0: -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: seta %dl -; X64-GISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X64-GISEL-NEXT: setb %al -; X64-GISEL-NEXT: andb %dl, %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq %a0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 1) ; "snan" ret i1 %a0 } define i1 @isquiet_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: isquiet_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X86-SDAGISEL-NEXT: setge %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: isquiet_f: +; X86: # %bb.0: # %entry +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 +; X86-NEXT: setge %al +; X86-NEXT: retl ; ; X64-LABEL: isquiet_f: ; X64: # %bb.0: # %entry @@ -155,39 +119,19 @@ define i1 @issignaling_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setge %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: isquiet_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X86-GISEL-NEXT: setae %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: isquiet_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X64-GISEL-NEXT: setae %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 2) ; "qnan" ret i1 %0 } define i1 @not_isquiet_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: not_isquiet_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X86-SDAGISEL-NEXT: setl %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: not_isquiet_f: +; X86: # %bb.0: # %entry +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 +; X86-NEXT: setl %al +; X86-NEXT: retl ; ; X64-LABEL: not_isquiet_f: ; X64: # %bb.0: # %entry @@ -208,57 +152,19 @@ define i1 @not_isquiet_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setl %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: not_isquiet_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: setb %dl -; X86-GISEL-NEXT: orb %cl, %dl -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: sete %cl -; X86-GISEL-NEXT: orb %dl, %cl -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: seta %dl -; X86-GISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X86-GISEL-NEXT: setb %al -; X86-GISEL-NEXT: andb %dl, %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: not_isquiet_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: setb %dl -; X64-GISEL-NEXT: orb %cl, %dl -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: sete %cl -; X64-GISEL-NEXT: orb %dl, %cl -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: seta %dl -; X64-GISEL-NEXT: cmpl $2143289344, %eax # imm = 0x7FC00000 -; X64-GISEL-NEXT: setb %al -; X64-GISEL-NEXT: andb %dl, %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 1021) ; ~"qnan" ret i1 %0 } define i1 @isinf_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: isinf_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-SDAGISEL-NEXT: sete %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: isinf_f: +; X86: # %bb.0: # %entry +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 +; X86-NEXT: sete %al +; X86-NEXT: retl ; ; X64-LABEL: isinf_f: ; X64: # %bb.0: # %entry @@ -279,39 +185,19 @@ define i1 @isinf_f(float %x) nounwind { ; X86-FASTISEL-NEXT: sete %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: isinf_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: sete %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: isinf_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: sete %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 516) ; 0x204 = "inf" ret i1 %0 } define i1 @not_isinf_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: not_isinf_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-SDAGISEL-NEXT: setne %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: not_isinf_f: +; X86: # %bb.0: # %entry +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 +; X86-NEXT: setne %al +; X86-NEXT: retl ; ; X64-LABEL: not_isinf_f: ; X64: # %bb.0: # %entry @@ -332,43 +218,17 @@ define i1 @not_isinf_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setne %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: not_isinf_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: setb %dl -; X86-GISEL-NEXT: orb %cl, %dl -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: seta %al -; X86-GISEL-NEXT: orb %dl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: not_isinf_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: setb %dl -; X64-GISEL-NEXT: orb %cl, %dl -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: seta %al -; X64-GISEL-NEXT: orb %dl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 507) ; ~0x204 = "~inf" ret i1 %0 } define i1 @is_plus_inf_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: is_plus_inf_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: cmpl $2139095040, {{[0-9]+}}(%esp) # imm = 0x7F800000 -; X86-SDAGISEL-NEXT: sete %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: is_plus_inf_f: +; X86: # %bb.0: # %entry +; X86-NEXT: cmpl $2139095040, {{[0-9]+}}(%esp) # imm = 0x7F800000 +; X86-NEXT: sete %al +; X86-NEXT: retl ; ; X64-LABEL: is_plus_inf_f: ; X64: # %bb.0: # %entry @@ -386,34 +246,17 @@ define i1 @is_plus_inf_f(float %x) nounwind { ; X86-FASTISEL-NEXT: sete %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: is_plus_inf_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, {{[0-9]+}}(%esp) # imm = 0x7F800000 -; X86-GISEL-NEXT: sete %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: is_plus_inf_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: sete %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 512) ; 0x200 = "+inf" ret i1 %0 } define i1 @is_minus_inf_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: is_minus_inf_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: cmpl $-8388608, {{[0-9]+}}(%esp) # imm = 0xFF800000 -; X86-SDAGISEL-NEXT: sete %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: is_minus_inf_f: +; X86: # %bb.0: # %entry +; X86-NEXT: cmpl $-8388608, {{[0-9]+}}(%esp) # imm = 0xFF800000 +; X86-NEXT: sete %al +; X86-NEXT: retl ; ; X64-LABEL: is_minus_inf_f: ; X64: # %bb.0: # %entry @@ -431,34 +274,17 @@ define i1 @is_minus_inf_f(float %x) nounwind { ; X86-FASTISEL-NEXT: sete %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: is_minus_inf_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $-8388608, {{[0-9]+}}(%esp) # imm = 0xFF800000 -; X86-GISEL-NEXT: sete %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: is_minus_inf_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: cmpl $-8388608, %eax # imm = 0xFF800000 -; X64-GISEL-NEXT: sete %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 4) ; "-inf" ret i1 %0 } define i1 @not_is_minus_inf_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: not_is_minus_inf_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: cmpl $-8388608, {{[0-9]+}}(%esp) # imm = 0xFF800000 -; X86-SDAGISEL-NEXT: setne %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: not_is_minus_inf_f: +; X86: # %bb.0: # %entry +; X86-NEXT: cmpl $-8388608, {{[0-9]+}}(%esp) # imm = 0xFF800000 +; X86-NEXT: setne %al +; X86-NEXT: retl ; ; X64-LABEL: not_is_minus_inf_f: ; X64: # %bb.0: # %entry @@ -476,55 +302,19 @@ define i1 @not_is_minus_inf_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setne %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: not_is_minus_inf_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: pushl %ebx -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: movl %eax, %ecx -; X86-GISEL-NEXT: andl $2147483647, %ecx # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %edx, %edx -; X86-GISEL-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000 -; X86-GISEL-NEXT: setb %bl -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: sete %ah -; X86-GISEL-NEXT: orb %dl, %ah -; X86-GISEL-NEXT: orb %bl, %ah -; X86-GISEL-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000 -; X86-GISEL-NEXT: seta %al -; X86-GISEL-NEXT: orb %ah, %al -; X86-GISEL-NEXT: popl %ebx -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: not_is_minus_inf_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: movl %eax, %ecx -; X64-GISEL-NEXT: andl $2147483647, %ecx # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %edx, %edx -; X64-GISEL-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000 -; X64-GISEL-NEXT: setb %sil -; X64-GISEL-NEXT: orb %dl, %sil -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: sete %dl -; X64-GISEL-NEXT: cmpl $2139095040, %ecx # imm = 0x7F800000 -; X64-GISEL-NEXT: seta %al -; X64-GISEL-NEXT: orb %dl, %al -; X64-GISEL-NEXT: orb %sil, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 1019) ; ~"-inf" ret i1 %0 } define i1 @isfinite_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: isfinite_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-SDAGISEL-NEXT: setl %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: isfinite_f: +; X86: # %bb.0: # %entry +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 +; X86-NEXT: setl %al +; X86-NEXT: retl ; ; X64-LABEL: isfinite_f: ; X64: # %bb.0: # %entry @@ -545,39 +335,19 @@ define i1 @isfinite_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setl %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: isfinite_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: setb %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: isfinite_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: setb %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 504) ; 0x1f8 = "finite" ret i1 %0 } define i1 @not_isfinite_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: not_isfinite_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-SDAGISEL-NEXT: andl {{[0-9]+}}(%esp), %eax -; X86-SDAGISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-SDAGISEL-NEXT: setge %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: not_isfinite_f: +; X86: # %bb.0: # %entry +; X86-NEXT: movl $2147483647, %eax # imm = 0x7FFFFFFF +; X86-NEXT: andl {{[0-9]+}}(%esp), %eax +; X86-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 +; X86-NEXT: setge %al +; X86-NEXT: retl ; ; X64-LABEL: not_isfinite_f: ; X64: # %bb.0: # %entry @@ -598,43 +368,17 @@ define i1 @not_isfinite_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setge %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: not_isfinite_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: sete %dl -; X86-GISEL-NEXT: orb %cl, %dl -; X86-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X86-GISEL-NEXT: seta %al -; X86-GISEL-NEXT: orb %dl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: not_isfinite_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: andl $2147483647, %eax # imm = 0x7FFFFFFF -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: sete %dl -; X64-GISEL-NEXT: orb %cl, %dl -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: seta %al -; X64-GISEL-NEXT: orb %dl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 519) ; ~0x1f8 = "~finite" ret i1 %0 } define i1 @is_plus_finite_f(float %x) nounwind { -; X86-SDAGISEL-LABEL: is_plus_finite_f: -; X86-SDAGISEL: # %bb.0: # %entry -; X86-SDAGISEL-NEXT: cmpl $2139095040, {{[0-9]+}}(%esp) # imm = 0x7F800000 -; X86-SDAGISEL-NEXT: setb %al -; X86-SDAGISEL-NEXT: retl +; X86-LABEL: is_plus_finite_f: +; X86: # %bb.0: # %entry +; X86-NEXT: cmpl $2139095040, {{[0-9]+}}(%esp) # imm = 0x7F800000 +; X86-NEXT: setb %al +; X86-NEXT: retl ; ; X64-LABEL: is_plus_finite_f: ; X64: # %bb.0: # %entry @@ -652,23 +396,6 @@ define i1 @is_plus_finite_f(float %x) nounwind { ; X86-FASTISEL-NEXT: setb %al ; X86-FASTISEL-NEXT: popl %ecx ; X86-FASTISEL-NEXT: retl -; -; X86-GISEL-LABEL: is_plus_finite_f: -; X86-GISEL: # %bb.0: # %entry -; X86-GISEL-NEXT: xorl %ecx, %ecx -; X86-GISEL-NEXT: cmpl $2139095040, {{[0-9]+}}(%esp) # imm = 0x7F800000 -; X86-GISEL-NEXT: setb %al -; X86-GISEL-NEXT: orb %cl, %al -; X86-GISEL-NEXT: retl -; -; X64-GISEL-LABEL: is_plus_finite_f: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: xorl %ecx, %ecx -; X64-GISEL-NEXT: movd %xmm0, %eax -; X64-GISEL-NEXT: cmpl $2139095040, %eax # imm = 0x7F800000 -; X64-GISEL-NEXT: setb %al -; X64-GISEL-NEXT: orb %cl, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f32(float %x, i32 448) ; 0x1c0 = "+finite" ret i1 %0 @@ -691,11 +418,6 @@ define i1 @isnone_d(double %x) nounwind { ; X86-FASTISEL-NEXT: fstp %st(0) ; X86-FASTISEL-NEXT: xorl %eax, %eax ; X86-FASTISEL-NEXT: retl -; -; X64-GISEL-LABEL: isnone_d: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: xorl %eax, %eax -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 0) ret i1 %0 @@ -718,11 +440,6 @@ define i1 @isany_d(double %x) nounwind { ; X86-FASTISEL-NEXT: fstp %st(0) ; X86-FASTISEL-NEXT: movb $1, %al ; X86-FASTISEL-NEXT: retl -; -; X64-GISEL-LABEL: isany_d: -; X64-GISEL: # %bb.0: # %entry -; X64-GISEL-NEXT: movb $1, %al -; X64-GISEL-NEXT: retq entry: %0 = tail call i1 @llvm.is.fpclass.f64(double %x, i32 1023) ret i1 %0 diff --git a/llvm/test/CodeGen/X86/min-legal-vector-width.ll b/llvm/test/CodeGen/X86/min-legal-vector-width.ll index d752659..04f0a65 100644 --- a/llvm/test/CodeGen/X86/min-legal-vector-width.ll +++ b/llvm/test/CodeGen/X86/min-legal-vector-width.ll @@ -1,5 +1,5 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skylake-avx512 -mattr=prefer-256-bit | FileCheck %s --check-prefixes=CHECK,CHECK-SKX +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skylake-avx512 -mattr=prefer-256-bit | FileCheck %s --check-prefixes=CHECK,CHECK-SKX,CHECK-SKX-NOVBMI ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skylake-avx512 -mattr=prefer-256-bit,avx512vbmi | FileCheck %s --check-prefixes=CHECK,CHECK-SKX,CHECK-SKX-VBMI ; Make sure CPUs default to prefer-256-bit. avx512vnni isn't interesting as it just adds an isel peephole for vpmaddwd+vpaddd ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=skylake-avx512 | FileCheck %s --check-prefixes=CHECK,CHECK-AVX512 @@ -883,6 +883,30 @@ define <16 x i16> @test_16f32tosb_512(ptr %ptr, <16 x i16> %passthru) "min-legal } define dso_local void @mul256(ptr %a, ptr %b, ptr %c) "min-legal-vector-width"="256" { +; CHECK-SKX-NOVBMI-LABEL: mul256: +; CHECK-SKX-NOVBMI: # %bb.0: +; CHECK-SKX-NOVBMI-NEXT: vmovdqa (%rdi), %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vmovdqa 32(%rdi), %ymm1 +; CHECK-SKX-NOVBMI-NEXT: vmovdqa (%rsi), %ymm2 +; CHECK-SKX-NOVBMI-NEXT: vmovdqa 32(%rsi), %ymm3 +; CHECK-SKX-NOVBMI-NEXT: vpbroadcastd {{.*#+}} ymm4 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; CHECK-SKX-NOVBMI-NEXT: vpand %ymm4, %ymm3, %ymm5 +; CHECK-SKX-NOVBMI-NEXT: vpmaddubsw %ymm5, %ymm1, %ymm5 +; CHECK-SKX-NOVBMI-NEXT: vpandn %ymm3, %ymm4, %ymm3 +; CHECK-SKX-NOVBMI-NEXT: vpmaddubsw %ymm3, %ymm1, %ymm1 +; CHECK-SKX-NOVBMI-NEXT: vpsllw $8, %ymm1, %ymm1 +; CHECK-SKX-NOVBMI-NEXT: vpternlogq {{.*#+}} ymm1 = ymm1 | (ymm5 & ymm4) +; CHECK-SKX-NOVBMI-NEXT: vpand %ymm4, %ymm2, %ymm3 +; CHECK-SKX-NOVBMI-NEXT: vpmaddubsw %ymm3, %ymm0, %ymm3 +; CHECK-SKX-NOVBMI-NEXT: vpandn %ymm2, %ymm4, %ymm2 +; CHECK-SKX-NOVBMI-NEXT: vpmaddubsw %ymm2, %ymm0, %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vpsllw $8, %ymm0, %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vpternlogq {{.*#+}} ymm0 = ymm0 | (ymm3 & ymm4) +; CHECK-SKX-NOVBMI-NEXT: vmovdqa %ymm0, (%rdx) +; CHECK-SKX-NOVBMI-NEXT: vmovdqa %ymm1, 32(%rdx) +; CHECK-SKX-NOVBMI-NEXT: vzeroupper +; CHECK-SKX-NOVBMI-NEXT: retq +; ; CHECK-SKX-VBMI-LABEL: mul256: ; CHECK-SKX-VBMI: # %bb.0: ; CHECK-SKX-VBMI-NEXT: vmovdqa (%rdi), %ymm0 @@ -960,6 +984,21 @@ define dso_local void @mul256(ptr %a, ptr %b, ptr %c) "min-legal-vector-width"=" } define dso_local void @mul512(ptr %a, ptr %b, ptr %c) "min-legal-vector-width"="512" { +; CHECK-SKX-NOVBMI-LABEL: mul512: +; CHECK-SKX-NOVBMI: # %bb.0: +; CHECK-SKX-NOVBMI-NEXT: vmovdqa64 (%rdi), %zmm0 +; CHECK-SKX-NOVBMI-NEXT: vmovdqa64 (%rsi), %zmm1 +; CHECK-SKX-NOVBMI-NEXT: vpbroadcastd {{.*#+}} zmm2 = [255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255,255] +; CHECK-SKX-NOVBMI-NEXT: vpandq %zmm2, %zmm1, %zmm3 +; CHECK-SKX-NOVBMI-NEXT: vpmaddubsw %zmm3, %zmm0, %zmm3 +; CHECK-SKX-NOVBMI-NEXT: vpandnq %zmm1, %zmm2, %zmm1 +; CHECK-SKX-NOVBMI-NEXT: vpmaddubsw %zmm1, %zmm0, %zmm0 +; CHECK-SKX-NOVBMI-NEXT: vpsllw $8, %zmm0, %zmm0 +; CHECK-SKX-NOVBMI-NEXT: vpternlogq {{.*#+}} zmm0 = zmm0 | (zmm3 & zmm2) +; CHECK-SKX-NOVBMI-NEXT: vmovdqa64 %zmm0, (%rdx) +; CHECK-SKX-NOVBMI-NEXT: vzeroupper +; CHECK-SKX-NOVBMI-NEXT: retq +; ; CHECK-SKX-VBMI-LABEL: mul512: ; CHECK-SKX-VBMI: # %bb.0: ; CHECK-SKX-VBMI-NEXT: vmovdqa64 (%rdi), %zmm0 @@ -1137,6 +1176,14 @@ define <16 x i16> @trunc_v16i32_v16i16_zeroes(ptr %x) nounwind "min-legal-vector } define <32 x i8> @trunc_v32i16_v32i8_zeroes(ptr %x) nounwind "min-legal-vector-width"="256" { +; CHECK-SKX-NOVBMI-LABEL: trunc_v32i16_v32i8_zeroes: +; CHECK-SKX-NOVBMI: # %bb.0: +; CHECK-SKX-NOVBMI-NEXT: vpsrlw $8, 32(%rdi), %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vpsrlw $8, (%rdi), %ymm1 +; CHECK-SKX-NOVBMI-NEXT: vpackuswb %ymm0, %ymm1, %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3] +; CHECK-SKX-NOVBMI-NEXT: retq +; ; CHECK-SKX-VBMI-LABEL: trunc_v32i16_v32i8_zeroes: ; CHECK-SKX-VBMI: # %bb.0: ; CHECK-SKX-VBMI-NEXT: vmovdqa (%rdi), %ymm1 @@ -1192,6 +1239,14 @@ define <16 x i16> @trunc_v16i32_v16i16_sign(ptr %x) nounwind "min-legal-vector-w } define <32 x i8> @trunc_v32i16_v32i8_sign(ptr %x) nounwind "min-legal-vector-width"="256" { +; CHECK-SKX-NOVBMI-LABEL: trunc_v32i16_v32i8_sign: +; CHECK-SKX-NOVBMI: # %bb.0: +; CHECK-SKX-NOVBMI-NEXT: vpsrlw $8, 32(%rdi), %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vpsrlw $8, (%rdi), %ymm1 +; CHECK-SKX-NOVBMI-NEXT: vpackuswb %ymm0, %ymm1, %ymm0 +; CHECK-SKX-NOVBMI-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3] +; CHECK-SKX-NOVBMI-NEXT: retq +; ; CHECK-SKX-VBMI-LABEL: trunc_v32i16_v32i8_sign: ; CHECK-SKX-VBMI: # %bb.0: ; CHECK-SKX-VBMI-NEXT: vmovdqa (%rdi), %ymm1 diff --git a/llvm/test/DebugInfo/AArch64/callsite.mir b/llvm/test/DebugInfo/AArch64/callsite.mir new file mode 100644 index 0000000..e3bd764 --- /dev/null +++ b/llvm/test/DebugInfo/AArch64/callsite.mir @@ -0,0 +1,68 @@ +# This test should not crash when generating call-site information. +# It was created to make sure that if isCopyLikeInstr in TargetInstrInfo.h +# returns an undef Dest Reg or Src Reg, we don't try to get a SubReg for it. + +# RUN: llc --mtriple=arm64e-apple-ios -start-before=aarch64-asm-printer %s -filetype=obj -o /dev/null --emit-call-site-info +--- | + %struct.rtyuio = type { i8 } + define noundef i32 @aserty(ptr noundef %0, ptr noundef %1) local_unnamed_addr #0 !dbg !23 { + ret i32 0 + } + define void @asdfgh(ptr noundef %0, ptr noundef %1, i8 noundef zeroext %2) local_unnamed_addr #0 !dbg !53 { + %4 = alloca ptr + %5 = call ptr @llvm.stackguard() + %6 = alloca %struct.rtyuio + %7 = icmp eq ptr %1, null + br i1 %7, label %10, label %8 + %9 = tail call i8 @polkiokl(ptr noundef %0) #6 + br label %10 + ret void + } + declare i8 @polkiokl(ptr noundef) local_unnamed_addr #2 + !llvm.module.flags = !{!2, !8} + !llvm.dbg.cu = !{!9} + !2 = !{i32 2, !"Debug Info Version", i32 3} + !8 = !{i32 7, !"frame-pointer", i32 1} + !9 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus_11, file: !10, emissionKind: FullDebug, sysroot: "/") + !10 = !DIFile(filename: "a.cpp", directory: "/") + !23 = distinct !DISubprogram(type: !27, flags: DIFlagPrototyped | DIFlagAllCallsDescribed, unit: !9, retainedNodes: !46) + !24 = distinct !DICompositeType(tag: DW_TAG_class_type, identifier: "yshscbshhdvcm") + !27 = !DISubroutineType(types: !28) + !28 = !{} + !30 = !DIDerivedType(tag: DW_TAG_typedef, baseType: !33) + !33 = distinct !DICompositeType(tag: DW_TAG_structure_type, identifier: "tyruwyeuiwiybabd") + !36 = !DISubroutineType(types: !37) + !37 = !{} + !46 = !{} + !47 = !DILocalVariable(scope: !23, type: !48, flags: DIFlagArtificial | DIFlagObjectPointer) + !48 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !24, size: 64) + !49 = !DILocalVariable(scope: !23, type: !30) + !50 = !DILocation(scope: !23) + !51 = !DILocation(scope: !23) + !53 = distinct !DISubprogram(type: !36, unit: !9, retainedNodes: !54) + !54 = !{} +name: aserty +stack: + - { id: 0, name: '', type: spill-slot, offset: -8, size: 8, alignment: 8, + debug-info-variable: '', debug-info-expression: '', debug-info-location: '' } + - { id: 1, name: '', type: spill-slot, offset: -16, size: 8, alignment: 8, + debug-info-variable: '', debug-info-expression: '', debug-info-location: '' } +callSites: + - { bb: 0, offset: 9, fwdArgRegs: + - { arg: 2, reg: '$w2' } } +body: | + bb.0 (%ir-block.2): + DBG_VALUE $x0, $noreg, !47, !DIExpression(), debug-location !50 + DBG_VALUE $x1, $noreg, !49, !DIExpression(), debug-location !50 + frame-setup PACIBSP implicit-def $lr, implicit killed $lr, implicit $sp + early-clobber $sp = frame-setup STPXpre $fp, killed $lr, $sp, -2 :: (store (s64) into %stack.1), (store (s64) into %stack.0) + $fp = frame-setup ADDXri $sp, 0, 0 + frame-setup CFI_INSTRUCTION def_cfa $w29, 16 + frame-setup CFI_INSTRUCTION offset $w30, -8 + frame-setup CFI_INSTRUCTION offset $w29, -16 + $x2 = ORRXrs $xzr, undef $noreg, 0, implicit $wzr, debug-location !51 + BL @asdfgh, csr_darwin_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit killed $x0, implicit killed $x1, implicit killed $w2, implicit-def $sp, debug-location !51 +... +name: asdfgh +body: | + bb.2 (%ir-block.10): diff --git a/llvm/test/LTO/X86/memprof-supports-hot-cold-new.ll b/llvm/test/LTO/X86/memprof-supports-hot-cold-new.ll index 3ed68e8..c3a75f6 100644 --- a/llvm/test/LTO/X86/memprof-supports-hot-cold-new.ll +++ b/llvm/test/LTO/X86/memprof-supports-hot-cold-new.ll @@ -13,14 +13,14 @@ ; RUN: -r=%t.o,main,plx \ ; RUN: -r=%t.o,_Znam, \ ; RUN: -memprof-dump-ccg \ -; RUN: -save-temps \ -; RUN: -o %t.out 2>&1 | FileCheck %s --check-prefix=DUMP -; DUMP: Callsite Context Graph: +; RUN: -print-before=memprof-context-disambiguation \ +; RUN: -o %t.out 2>&1 | FileCheck %s --check-prefix=DUMP --check-prefix=IR -; RUN: llvm-dis %t.out.0.0.preopt.bc -o - | FileCheck %s --check-prefix=IR ; IR: !memprof {{.*}} !callsite ; IR: "memprof"="cold" +; DUMP: Callsite Context Graph: + ;; Next check without -supports-hot-cold-new, we should not perform ;; context disambiguation, and we should strip memprof metadata and ;; attributes before optimization. @@ -28,13 +28,16 @@ ; RUN: -r=%t.o,main,plx \ ; RUN: -r=%t.o,_Znam, \ ; RUN: -memprof-dump-ccg \ -; RUN: -save-temps \ +; RUN: -print-before=memprof-context-disambiguation \ ; RUN: -o %t.out 2>&1 | FileCheck %s --allow-empty \ -; RUN: --implicit-check-not "Callsite Context Graph:" +; RUN: --implicit-check-not "Callsite Context Graph:" \ +; RUN: --implicit-check-not "!memprof" --implicit-check-not "!callsite" \ +; RUN: --implicit-check-not "memprof"="cold" -; RUN: llvm-dis %t.out.0.0.preopt.bc -o - | FileCheck %s \ -; RUN: --implicit-check-not "!memprof" --implicit-check-not "!callsite" \ -; RUN: --implicit-check-not "memprof"="cold" +;; Ensure the attributes and metadata are stripped when running a non-LTO pipeline. +; RUN: opt -O3 %t.o -S | FileCheck %s \ +; RUN: --implicit-check-not "!memprof" --implicit-check-not "!callsite" \ +; RUN: --implicit-check-not "memprof"="cold" source_filename = "memprof-supports-hot-cold-new.ll" target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" diff --git a/llvm/test/MC/Disassembler/X86/apx/pushp-popp.txt b/llvm/test/MC/Disassembler/X86/apx/pushp-popp.txt index 4ec534f..fa40fe6 100644 --- a/llvm/test/MC/Disassembler/X86/apx/pushp-popp.txt +++ b/llvm/test/MC/Disassembler/X86/apx/pushp-popp.txt @@ -17,6 +17,10 @@ # INTEL: pushp r16 0xd5,0x18,0x50 +# ATT: pushq %r16 +# INTEL: push r16 +0xd5,0x10,0x50 + # ATT: popp %rax # INTEL: popp rax 0xd5,0x08,0x58 @@ -32,3 +36,7 @@ # ATT: popp %r16 # INTEL: popp r16 0xd5,0x18,0x58 + +# ATT: popq %r16 +# INTEL: pop r16 +0xd5,0x10,0x58 diff --git a/llvm/test/MC/X86/apx/pushp-popp-att.s b/llvm/test/MC/X86/apx/pushp-popp-att.s index a810744..d638034 100644 --- a/llvm/test/MC/X86/apx/pushp-popp-att.s +++ b/llvm/test/MC/X86/apx/pushp-popp-att.s @@ -1,7 +1,7 @@ # RUN: llvm-mc -triple x86_64 -show-encoding %s | FileCheck %s # RUN: not llvm-mc -triple i386 -show-encoding %s 2>&1 | FileCheck %s --check-prefix=ERROR -# ERROR-COUNT-8: error: +# ERROR-COUNT-10: error: # ERROR-NOT: error: # CHECK: pushp %rax @@ -16,6 +16,9 @@ # CHECK: pushp %r16 # CHECK: encoding: [0xd5,0x18,0x50] pushp %r16 +# CHECK: pushq %r16 +# CHECK: encoding: [0xd5,0x10,0x50] + pushq %r16 # CHECK: popp %rax # CHECK: encoding: [0xd5,0x08,0x58] @@ -29,3 +32,6 @@ # CHECK: popp %r16 # CHECK: encoding: [0xd5,0x18,0x58] popp %r16 +# CHECK: popq %r16 +# CHECK: encoding: [0xd5,0x10,0x58] + popq %r16 diff --git a/llvm/test/ThinLTO/X86/memprof-supports-hot-cold-new.ll b/llvm/test/ThinLTO/X86/memprof-supports-hot-cold-new.ll index 7a4d860..fe2a002 100644 --- a/llvm/test/ThinLTO/X86/memprof-supports-hot-cold-new.ll +++ b/llvm/test/ThinLTO/X86/memprof-supports-hot-cold-new.ll @@ -17,11 +17,12 @@ ; RUN: -r=%t/foo.o,foo,plx \ ; RUN: -r=%t/foo.o,_Znam, \ ; RUN: -memprof-dump-ccg \ -; RUN: -save-temps \ -; RUN: -o %t.out 2>&1 | FileCheck %s --check-prefix=DUMP +; RUN: -print-before=memprof-context-disambiguation \ +; RUN: -thinlto-threads=1 \ +; RUN: -o %t.out 2>&1 | FileCheck %s --check-prefix=DUMP --check-prefix=IR + ; DUMP: Callsite Context Graph: -; RUN: llvm-dis %t.out.1.3.import.bc -o - | FileCheck %s --check-prefix=IR ; IR: @main() ; IR: !memprof {{.*}} !callsite ; IR: @_Znam(i64 0) #[[ATTR:[0-9]+]] @@ -41,13 +42,12 @@ ; RUN: -r=%t/foo.o,foo,plx \ ; RUN: -r=%t/foo.o,_Znam, \ ; RUN: -memprof-dump-ccg \ -; RUN: -save-temps \ +; RUN: -print-before=memprof-context-disambiguation \ +; RUN: -thinlto-threads=1 \ ; RUN: -o %t.out 2>&1 | FileCheck %s --allow-empty \ -; RUN: --implicit-check-not "Callsite Context Graph:" - -; RUN: llvm-dis %t.out.1.3.import.bc -o - | FileCheck %s \ -; RUN: --implicit-check-not "!memprof" --implicit-check-not "!callsite" \ -; RUN: --implicit-check-not "memprof"="cold" +; RUN: --implicit-check-not "Callsite Context Graph:" \ +; RUN: --implicit-check-not "!memprof" --implicit-check-not "!callsite" \ +; RUN: --implicit-check-not "memprof"="cold" ;--- main.ll target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" diff --git a/llvm/test/Transforms/GVN/PRE/pre-load.ll b/llvm/test/Transforms/GVN/PRE/pre-load.ll index 5a07f9f..afa1354 100644 --- a/llvm/test/Transforms/GVN/PRE/pre-load.ll +++ b/llvm/test/Transforms/GVN/PRE/pre-load.ll @@ -1503,3 +1503,51 @@ wrong: exit: ret void } + +; Allow the load to be made available on the edge (%entry, %if.end) as part of PRE, +; but ensure `%identical.l` is not hoisted to its predecessor due to the local +; dependency with the call. + +define i32 @test24(ptr noalias %p, ptr noalias %q, i1 %c) { +; MDEP-LABEL: @test24( +; MDEP-NEXT: entry: +; MDEP-NEXT: br i1 [[C:%.*]], label [[ENTRY_IF_END_CRIT_EDGE:%.*]], label [[IF_THEN:%.*]] +; MDEP: entry.if.end_crit_edge: +; MDEP-NEXT: [[VV_PRE:%.*]] = load i32, ptr [[X:%.*]], align 4 +; MDEP-NEXT: br label [[IF_END:%.*]] +; MDEP: if.then: +; MDEP-NEXT: call void @opaque(ptr [[X]]) +; MDEP-NEXT: [[UU:%.*]] = load i32, ptr [[X]], align 4 +; MDEP-NEXT: store i32 [[UU]], ptr [[R:%.*]], align 4 +; MDEP-NEXT: br label [[IF_END]] +; MDEP: if.end: +; MDEP-NEXT: [[VV:%.*]] = phi i32 [ [[VV_PRE]], [[ENTRY_IF_END_CRIT_EDGE]] ], [ [[UU]], [[IF_THEN]] ] +; MDEP-NEXT: ret i32 [[VV]] +; +; MSSA-LABEL: @test24( +; MSSA-NEXT: entry: +; MSSA-NEXT: br i1 [[C:%.*]], label [[IF_END:%.*]], label [[IF_THEN:%.*]] +; MSSA: if.then: +; MSSA-NEXT: call void @opaque(ptr [[X:%.*]]) +; MSSA-NEXT: [[UU:%.*]] = load i32, ptr [[X]], align 4 +; MSSA-NEXT: store i32 [[UU]], ptr [[R:%.*]], align 4 +; MSSA-NEXT: br label [[IF_END]] +; MSSA: if.end: +; MSSA-NEXT: [[VV:%.*]] = load i32, ptr [[X]], align 4 +; MSSA-NEXT: ret i32 [[VV]] +; +entry: + br i1 %c, label %if.end, label %if.then + +if.then: + call void @opaque(ptr %p) + %identical.l = load i32, ptr %p, align 4 + store i32 %identical.l, ptr %q, align 4 + br label %if.end + +if.end: + %l = load i32, ptr %p, align 4 + ret i32 %l +} + +declare void @opaque(ptr) nounwind willreturn diff --git a/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll b/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll index cb4e07e..9b9bc68 100644 --- a/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll +++ b/llvm/test/Transforms/IndVarSimplify/X86/overflow-intrinsics.ll @@ -60,8 +60,7 @@ define void @f_sadd_overflow(ptr %a) { ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[CONT:.*]] ], [ 2147483645, %[[ENTRY]] ] ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]] ; CHECK-NEXT: store i8 0, ptr [[ARRAYIDX]], align 1 -; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV]], 2147483647 -; CHECK-NEXT: br i1 [[EXITCOND]], label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] +; CHECK-NEXT: br i1 true, label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] ; CHECK: [[TRAP]]: ; CHECK-NEXT: tail call void @llvm.trap(), !nosanitize [[META0]] ; CHECK-NEXT: unreachable, !nosanitize [[META0]] @@ -150,8 +149,7 @@ define void @f_uadd_overflow(ptr %a) { ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[CONT:.*]] ], [ -6, %[[ENTRY]] ] ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]] ; CHECK-NEXT: store i8 0, ptr [[ARRAYIDX]], align 1 -; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV]], -1 -; CHECK-NEXT: br i1 [[EXITCOND]], label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] +; CHECK-NEXT: br i1 true, label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] ; CHECK: [[TRAP]]: ; CHECK-NEXT: tail call void @llvm.trap(), !nosanitize [[META0]] ; CHECK-NEXT: unreachable, !nosanitize [[META0]] @@ -243,10 +241,7 @@ define void @f_ssub_overflow(ptr nocapture %a) { ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[CONT:.*]] ], [ -2147483642, %[[ENTRY]] ] ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]] ; CHECK-NEXT: store i8 0, ptr [[ARRAYIDX]], align 1 -; CHECK-NEXT: [[TMP0:%.*]] = trunc nsw i64 [[INDVARS_IV]] to i32 -; CHECK-NEXT: [[TMP1:%.*]] = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 [[TMP0]], i32 1) -; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { i32, i1 } [[TMP1]], 1 -; CHECK-NEXT: br i1 [[TMP2]], label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] +; CHECK-NEXT: br i1 true, label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] ; CHECK: [[TRAP]]: ; CHECK-NEXT: tail call void @llvm.trap(), !nosanitize [[META0]] ; CHECK-NEXT: unreachable, !nosanitize [[META0]] @@ -339,10 +334,7 @@ define void @f_usub_overflow(ptr nocapture %a) { ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], %[[CONT:.*]] ], [ 15, %[[ENTRY]] ] ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]] ; CHECK-NEXT: store i8 0, ptr [[ARRAYIDX]], align 1 -; CHECK-NEXT: [[TMP0:%.*]] = trunc nuw nsw i64 [[INDVARS_IV]] to i32 -; CHECK-NEXT: [[TMP1:%.*]] = tail call { i32, i1 } @llvm.usub.with.overflow.i32(i32 [[TMP0]], i32 1) -; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { i32, i1 } [[TMP1]], 1 -; CHECK-NEXT: br i1 [[TMP2]], label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] +; CHECK-NEXT: br i1 true, label %[[TRAP:.*]], label %[[CONT]], !nosanitize [[META0]] ; CHECK: [[TRAP]]: ; CHECK-NEXT: tail call void @llvm.trap(), !nosanitize [[META0]] ; CHECK-NEXT: unreachable, !nosanitize [[META0]] diff --git a/llvm/test/Transforms/IndVarSimplify/unreachable-exit.ll b/llvm/test/Transforms/IndVarSimplify/unreachable-exit.ll new file mode 100644 index 0000000..b9c9228 --- /dev/null +++ b/llvm/test/Transforms/IndVarSimplify/unreachable-exit.ll @@ -0,0 +1,738 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 +; RUN: opt -S -passes=indvars < %s | FileCheck %s + +define void @optimize_trap(i32 %block_size) { +; CHECK-LABEL: define void @optimize_trap( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[BLOCK_SIZE]], -1 +; CHECK-NEXT: [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP1]], i32 3) +; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 3, [[UMIN]] +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: br i1 [[TMP2]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @llvm.trap() +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP4:%.*]] = xor i8 [[TMP3]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @llvm.trap() + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_atomic(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_atomic( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @llvm.trap() +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP4:%.*]] = xor i8 [[TMP3]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store atomic i8 [[TMP4]], ptr [[ARRAYIDX7]] unordered, align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @llvm.trap() + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store atomic i8 %1, ptr %arrayidx7 unordered, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_volatile(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_volatile( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @llvm.trap() +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP4:%.*]] = xor i8 [[TMP3]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store volatile i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @llvm.trap() + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store volatile i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_call(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_call( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @llvm.trap() +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP4:%.*]] = xor i8 [[TMP3]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: call void @x(ptr null) +; CHECK-NEXT: store volatile i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @llvm.trap() + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + call void @x(ptr null) + store volatile i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @optimize_ubsan_trap(i32 %block_size) { +; CHECK-LABEL: define void @optimize_ubsan_trap( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[BLOCK_SIZE]], -1 +; CHECK-NEXT: [[UMIN:%.*]] = call i32 @llvm.umin.i32(i32 [[TMP1]], i32 3) +; CHECK-NEXT: [[TMP2:%.*]] = icmp eq i32 3, [[UMIN]] +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: br i1 [[TMP2]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @llvm.ubsantrap(i8 1) +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP4:%.*]] = xor i8 [[TMP3]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @llvm.ubsantrap(i8 1) + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_arbitrary_call(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_arbitrary_call( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @noreturn_with_argmem(ptr [[FOO_ARR]]) +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP3:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP4:%.*]] = xor i8 [[TMP3]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP4]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @noreturn_with_argmem(ptr %foo_arr) + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_two_exits(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_two_exits( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[P:%.*]] = call i1 @pred() +; CHECK-NEXT: br i1 [[P]], label %[[FOR_BODY_CONT:.*]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; CHECK: [[FOR_BODY_CONT]]: +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @noreturn(ptr [[FOO_ARR]]) +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[TMP0]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP1]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %p = call i1 @pred() + br i1 %p, label %for.body.cont, label %for.cond.cleanup.loopexit + +for.body.cont: ; preds = %for.body.preheader, %if.end4 + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @noreturn(ptr %foo_arr) + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_two_exits2(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_two_exits2( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[FOR_BODY_CONT:.*]] +; CHECK: [[FOR_BODY_CONT]]: +; CHECK-NEXT: [[P:%.*]] = call i1 @pred() +; CHECK-NEXT: br i1 [[P]], label %[[IF_END4]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: call void @noreturn(ptr [[FOO_ARR]]) +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[TMP0]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP1]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %for.body.cont + +for.body.cont: ; preds = %for.body.preheader, %if.end4 + %p = call i1 @pred() + br i1 %p, label %if.end4, label %for.cond.cleanup.loopexit + +if.then: ; preds = %for.body + call void @noreturn(ptr %foo_arr) + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_depdendent_ubsan_trap(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_depdendent_ubsan_trap( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: [[I_015_LCSSA:%.*]] = phi i32 [ [[I_015]], %[[FOR_BODY]] ] +; CHECK-NEXT: call void @noreturn_with_i32(i32 [[I_015_LCSSA]]) +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[TMP0]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP1]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + call void @noreturn_with_i32(i32 %i.015) + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + +define void @no_optimize_depdendent_load_trap(i32 %block_size) { +; CHECK-LABEL: define void @no_optimize_depdendent_load_trap( +; CHECK-SAME: i32 [[BLOCK_SIZE:%.*]]) { +; CHECK-NEXT: [[ENTRY:.*:]] +; CHECK-NEXT: [[FOO_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: [[BAR_ARR:%.*]] = alloca [2 x i8], align 16 +; CHECK-NEXT: call void @x(ptr nonnull [[FOO_ARR]]) +; CHECK-NEXT: [[CMP14_NOT:%.*]] = icmp eq i32 [[BLOCK_SIZE]], 0 +; CHECK-NEXT: br i1 [[CMP14_NOT]], label %[[FOR_COND_CLEANUP:.*]], label %[[FOR_BODY_PREHEADER:.*]] +; CHECK: [[FOR_BODY_PREHEADER]]: +; CHECK-NEXT: br label %[[FOR_BODY:.*]] +; CHECK: [[FOR_COND_CLEANUP_LOOPEXIT:.*]]: +; CHECK-NEXT: br label %[[FOR_COND_CLEANUP]] +; CHECK: [[FOR_COND_CLEANUP]]: +; CHECK-NEXT: call void @x(ptr nonnull [[BAR_ARR]]) +; CHECK-NEXT: ret void +; CHECK: [[FOR_BODY]]: +; CHECK-NEXT: [[I_015:%.*]] = phi i32 [ [[INC:%.*]], %[[IF_END4:.*]] ], [ 0, %[[FOR_BODY_PREHEADER]] ] +; CHECK-NEXT: [[CMP1:%.*]] = icmp samesign ugt i32 [[I_015]], 2 +; CHECK-NEXT: br i1 [[CMP1]], label %[[IF_THEN:.*]], label %[[IF_END4]] +; CHECK: [[IF_THEN]]: +; CHECK-NEXT: [[I_015_LCSSA:%.*]] = load i8, ptr [[FOO_ARR]], align 1 +; CHECK-NEXT: call void @noreturn_with_i8(i8 [[I_015_LCSSA]]) +; CHECK-NEXT: unreachable +; CHECK: [[IF_END4]]: +; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds nuw [1024 x i8], ptr [[FOO_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: [[TMP0:%.*]] = load i8, ptr [[ARRAYIDX]], align 1 +; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[TMP0]], 54 +; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds nuw [1025 x i8], ptr [[BAR_ARR]], i64 0, i32 [[I_015]] +; CHECK-NEXT: store i8 [[TMP1]], ptr [[ARRAYIDX7]], align 1 +; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_015]], 1 +; CHECK-NEXT: [[CMP:%.*]] = icmp ult i32 [[INC]], [[BLOCK_SIZE]] +; CHECK-NEXT: br i1 [[CMP]], label %[[FOR_BODY]], label %[[FOR_COND_CLEANUP_LOOPEXIT]] +; +entry: + %foo_arr = alloca [2 x i8], align 16 + %bar_arr = alloca [2 x i8], align 16 + call void @x(ptr nonnull %foo_arr) + %cmp14.not = icmp eq i32 %block_size, 0 + br i1 %cmp14.not, label %for.cond.cleanup, label %for.body.preheader + +for.body.preheader: ; preds = %entry + br label %for.body + +for.cond.cleanup.loopexit: ; preds = %if.end4 + br label %for.cond.cleanup + +for.cond.cleanup: ; preds = %for.cond.cleanup.loopexit, %entry + call void @x(ptr nonnull %bar_arr) + ret void + +for.body: ; preds = %for.body.preheader, %if.end4 + %i.015 = phi i32 [ %inc, %if.end4 ], [ 0, %for.body.preheader ] + %cmp1 = icmp samesign ugt i32 %i.015, 2 + br i1 %cmp1, label %if.then, label %if.end4 + +if.then: ; preds = %for.body + %r = load i8, ptr %foo_arr, align 1 + call void @noreturn_with_i8(i8 %r) + unreachable + +if.end4: ; preds = %for.body + %arrayidx = getelementptr inbounds nuw [1024 x i8], ptr %foo_arr, i64 0, i32 %i.015 + %0 = load i8, ptr %arrayidx, align 1 + %1 = xor i8 %0, 54 + %arrayidx7 = getelementptr inbounds nuw [1025 x i8], ptr %bar_arr, i64 0, i32 %i.015 + store i8 %1, ptr %arrayidx7, align 1 + %inc = add nuw nsw i32 %i.015, 1 + %cmp = icmp ult i32 %inc, %block_size + br i1 %cmp, label %for.body, label %for.cond.cleanup.loopexit +} + + +declare void @x(ptr noundef) local_unnamed_addr +declare i1 @pred() local_unnamed_addr + +declare void @llvm.trap() #0 +declare void @noreturn(ptr) #0 +declare void @noreturn_with_i32(i32) #0 +declare void @noreturn_with_i8(i8) #0 +declare void @noreturn_with_argmem(ptr) #1 + +attributes #0 = { cold noreturn nounwind memory(inaccessiblemem: write) } +attributes #1 = { cold noreturn nounwind memory(argmem: read) } diff --git a/llvm/test/Transforms/InterleavedAccess/AArch64/scalable-deinterleave-intrinsics.ll b/llvm/test/Transforms/InterleavedAccess/AArch64/scalable-deinterleave-intrinsics.ll index ed9fba3..22ab79d 100644 --- a/llvm/test/Transforms/InterleavedAccess/AArch64/scalable-deinterleave-intrinsics.ll +++ b/llvm/test/Transforms/InterleavedAccess/AArch64/scalable-deinterleave-intrinsics.ll @@ -289,6 +289,225 @@ define void @interleave_wide_nxdouble_factor2(ptr %ptr, <vscale x 4 x double> %l ret void } +define void @deinterleave1_nxi64_factor3(ptr %ptr, <vscale x 4 x i64>* %s1, <vscale x 4 x i64>* %s2, <vscale x 4 x i64>* %s3) #0 { +; CHECK-LABEL: define void @deinterleave1_nxi64_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0 +; CHECK-NEXT: [[LDN1:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP1]]) +; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 0 +; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0) +; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 1 +; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0) +; CHECK-NEXT: [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 2 +; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP6]], i64 0) +; CHECK-NEXT: [[TMP8:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3 +; CHECK-NEXT: [[LDN2:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP8]]) +; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 0 +; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP9]], i64 2) +; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 1 +; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP11]], i64 2) +; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 2 +; CHECK-NEXT: [[TMP14:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP13]], i64 2) +; CHECK-NEXT: [[TMP15:%.*]] = insertvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } poison, <vscale x 4 x i64> [[TMP10]], 0 +; CHECK-NEXT: [[TMP16:%.*]] = insertvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP15]], <vscale x 4 x i64> [[TMP12]], 1 +; CHECK-NEXT: [[TMP17:%.*]] = insertvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP16]], <vscale x 4 x i64> [[TMP14]], 2 +; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP17]], 0 +; CHECK-NEXT: [[TMP19:%.*]] = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP17]], 1 +; CHECK-NEXT: [[TMP20:%.*]] = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP17]], 2 +; CHECK-NEXT: store <vscale x 4 x i64> [[TMP18]], ptr [[S1]], align 32 +; CHECK-NEXT: store <vscale x 4 x i64> [[TMP19]], ptr [[S2]], align 32 +; CHECK-NEXT: store <vscale x 4 x i64> [[TMP20]], ptr [[S3]], align 32 +; CHECK-NEXT: ret void +; + %wide.vec = load <vscale x 12 x i64>, ptr %ptr, align 8 + %ldN = tail call { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave3.nxv12i64(<vscale x 12 x i64> %wide.vec) + + %3 = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } %ldN, 0 + %4 = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } %ldN, 1 + %5 = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } %ldN, 2 + + store <vscale x 4 x i64> %3, <vscale x 4 x i64>* %s1 + store <vscale x 4 x i64> %4, <vscale x 4 x i64>* %s2 + store <vscale x 4 x i64> %5, <vscale x 4 x i64>* %s3 + ret void +} + +define void @deinterleave2_nxi64_factor3(ptr %ptr, <vscale x 8 x i64>* %s1, <vscale x 8 x i64>* %s2, <vscale x 8 x i64>* %s3) #0 { +; CHECK-LABEL: define void @deinterleave2_nxi64_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0 +; CHECK-NEXT: [[LDN1:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP1]]) +; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 0 +; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0) +; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 1 +; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0) +; CHECK-NEXT: [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 2 +; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP6]], i64 0) +; CHECK-NEXT: [[TMP8:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3 +; CHECK-NEXT: [[LDN2:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP8]]) +; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 0 +; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP9]], i64 2) +; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 1 +; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP11]], i64 2) +; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 2 +; CHECK-NEXT: [[TMP14:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP13]], i64 2) +; CHECK-NEXT: [[TMP15:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 6 +; CHECK-NEXT: [[LDN3:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP15]]) +; CHECK-NEXT: [[TMP16:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN3]], 0 +; CHECK-NEXT: [[TMP17:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP16]], i64 4) +; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN3]], 1 +; CHECK-NEXT: [[TMP19:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP18]], i64 4) +; CHECK-NEXT: [[TMP20:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN3]], 2 +; CHECK-NEXT: [[TMP21:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP20]], i64 4) +; CHECK-NEXT: [[TMP22:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 9 +; CHECK-NEXT: [[LDN4:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP22]]) +; CHECK-NEXT: [[TMP23:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN4]], 0 +; CHECK-NEXT: [[TMP24:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP17]], <vscale x 2 x i64> [[TMP23]], i64 6) +; CHECK-NEXT: [[TMP25:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN4]], 1 +; CHECK-NEXT: [[TMP26:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP19]], <vscale x 2 x i64> [[TMP25]], i64 6) +; CHECK-NEXT: [[TMP27:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN4]], 2 +; CHECK-NEXT: [[TMP28:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP21]], <vscale x 2 x i64> [[TMP27]], i64 6) +; CHECK-NEXT: [[TMP29:%.*]] = insertvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } poison, <vscale x 8 x i64> [[TMP24]], 0 +; CHECK-NEXT: [[TMP30:%.*]] = insertvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP29]], <vscale x 8 x i64> [[TMP26]], 1 +; CHECK-NEXT: [[TMP31:%.*]] = insertvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP30]], <vscale x 8 x i64> [[TMP28]], 2 +; CHECK-NEXT: [[TMP32:%.*]] = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP31]], 0 +; CHECK-NEXT: [[TMP33:%.*]] = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP31]], 1 +; CHECK-NEXT: [[TMP34:%.*]] = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP31]], 2 +; CHECK-NEXT: store <vscale x 8 x i64> [[TMP32]], ptr [[S1]], align 64 +; CHECK-NEXT: store <vscale x 8 x i64> [[TMP33]], ptr [[S2]], align 64 +; CHECK-NEXT: store <vscale x 8 x i64> [[TMP34]], ptr [[S3]], align 64 +; CHECK-NEXT: ret void +; + %wide.vec = load <vscale x 24 x i64>, ptr %ptr, align 8 + %ldN = tail call { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } @llvm.vector.deinterleave3.nxv24i64(<vscale x 24 x i64> %wide.vec) + + %3 = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } %ldN, 0 + %4 = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } %ldN, 1 + %5 = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } %ldN, 2 + + store <vscale x 8 x i64> %3, <vscale x 8 x i64>* %s1 + store <vscale x 8 x i64> %4, <vscale x 8 x i64>* %s2 + store <vscale x 8 x i64> %5, <vscale x 8 x i64>* %s3 + ret void +} + +define void @deinterleave_neg1_nxi64_factor3(ptr %ptr, <vscale x 1 x i64>* %s1, <vscale x 1 x i64>* %s2, <vscale x 1 x i64>* %s3) #0 { +; CHECK-LABEL: define void @deinterleave_neg1_nxi64_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 3 x i64>, ptr [[PTR]], align 8 +; CHECK-NEXT: [[LDN:%.*]] = tail call { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave3.nxv3i64(<vscale x 3 x i64> [[WIDE_VEC]]) +; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } [[LDN]], 0 +; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } [[LDN]], 1 +; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } [[LDN]], 2 +; CHECK-NEXT: store <vscale x 1 x i64> [[TMP1]], ptr [[S1]], align 8 +; CHECK-NEXT: store <vscale x 1 x i64> [[TMP2]], ptr [[S2]], align 8 +; CHECK-NEXT: store <vscale x 1 x i64> [[TMP3]], ptr [[S3]], align 8 +; CHECK-NEXT: ret void +; + %wide.vec = load <vscale x 3 x i64>, ptr %ptr, align 8 + %ldN = tail call { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave3.nxv3i64(<vscale x 3 x i64> %wide.vec) + + %3 = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } %ldN, 0 + %4 = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } %ldN, 1 + %5 = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } %ldN, 2 + + store <vscale x 1 x i64> %3, <vscale x 1 x i64>* %s1 + store <vscale x 1 x i64> %4, <vscale x 1 x i64>* %s2 + store <vscale x 1 x i64> %5, <vscale x 1 x i64>* %s3 + ret void +} + +define void @deinterleave_neg2_nxi8_factor3(ptr %ptr, <vscale x 8 x i8>* %s1, <vscale x 8 x i8>* %s2, <vscale x 8 x i8>* %s3) #0 { +; CHECK-LABEL: define void @deinterleave_neg2_nxi8_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 24 x i8>, ptr [[PTR]], align 8 +; CHECK-NEXT: [[LDN:%.*]] = tail call { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } @llvm.vector.deinterleave3.nxv24i8(<vscale x 24 x i8> [[WIDE_VEC]]) +; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } [[LDN]], 0 +; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } [[LDN]], 1 +; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } [[LDN]], 2 +; CHECK-NEXT: store <vscale x 8 x i8> [[TMP1]], ptr [[S1]], align 8 +; CHECK-NEXT: store <vscale x 8 x i8> [[TMP2]], ptr [[S2]], align 8 +; CHECK-NEXT: store <vscale x 8 x i8> [[TMP3]], ptr [[S3]], align 8 +; CHECK-NEXT: ret void +; + %wide.vec = load <vscale x 24 x i8>, ptr %ptr, align 8 + %ldN = tail call { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } @llvm.vector.deinterleave3.nxv12i8(<vscale x 24 x i8> %wide.vec) + + %3 = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } %ldN, 0 + %4 = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } %ldN, 1 + %5 = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } %ldN, 2 + + store <vscale x 8 x i8> %3, <vscale x 8 x i8>* %s1 + store <vscale x 8 x i8> %4, <vscale x 8 x i8>* %s2 + store <vscale x 8 x i8> %5, <vscale x 8 x i8>* %s3 + ret void +} + +define void @interleave1_nxi64_factor3(ptr %ptr, <vscale x 8 x i64> %s1, <vscale x 8 x i64> %s2, <vscale x 8 x i64> %s3) #0 { +; CHECK-LABEL: define void @interleave1_nxi64_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 8 x i64> [[S1:%.*]], <vscale x 8 x i64> [[S2:%.*]], <vscale x 8 x i64> [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 0) +; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 0) +; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 0) +; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP1]]) +; CHECK-NEXT: [[TMP5:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3 +; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 2) +; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 2) +; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 2) +; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP5]]) +; CHECK-NEXT: [[TMP9:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 6 +; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 4) +; CHECK-NEXT: [[TMP11:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 4) +; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 4) +; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], <vscale x 2 x i64> [[TMP12]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP9]]) +; CHECK-NEXT: [[TMP13:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 9 +; CHECK-NEXT: [[TMP14:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 6) +; CHECK-NEXT: [[TMP15:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 6) +; CHECK-NEXT: [[TMP16:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 6) +; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], <vscale x 2 x i64> [[TMP16]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP13]]) +; CHECK-NEXT: ret void +; + %interleave = tail call <vscale x 24 x i64> @llvm.vector.interleave3.nxv24i64(<vscale x 8 x i64> %s1, <vscale x 8 x i64> %s2, <vscale x 8 x i64> %s3) + + store <vscale x 24 x i64> %interleave, ptr %ptr, align 4 + ret void +} + +define void @interleave2_nxi64_factor3(ptr %ptr, <vscale x 4 x i64> %s1, <vscale x 4 x i64> %s2, <vscale x 4 x i64> %s3) #0 { +; CHECK-LABEL: define void @interleave2_nxi64_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 4 x i64> [[S1:%.*]], <vscale x 4 x i64> [[S2:%.*]], <vscale x 4 x i64> [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0 +; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S1]], i64 0) +; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S2]], i64 0) +; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S3]], i64 0) +; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP1]]) +; CHECK-NEXT: [[TMP5:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3 +; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S1]], i64 2) +; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S2]], i64 2) +; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S3]], i64 2) +; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP5]]) +; CHECK-NEXT: ret void +; + %interleave = tail call <vscale x 12 x i64> @llvm.vector.interleave3.nxv12i64(<vscale x 4 x i64> %s1, <vscale x 4 x i64> %s2, <vscale x 4 x i64> %s3) + + store <vscale x 12 x i64> %interleave, ptr %ptr, align 4 + ret void +} + +define void @interleave_neg_nxi8_factor3(ptr %ptr, <vscale x 8 x i8> %s1, <vscale x 8 x i8> %s2, <vscale x 8 x i8> %s3) #0 { +; CHECK-LABEL: define void @interleave_neg_nxi8_factor3 +; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 8 x i8> [[S1:%.*]], <vscale x 8 x i8> [[S2:%.*]], <vscale x 8 x i8> [[S3:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[INTERLEAVE:%.*]] = tail call <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8> [[S1]], <vscale x 8 x i8> [[S2]], <vscale x 8 x i8> [[S3]]) +; CHECK-NEXT: store <vscale x 24 x i8> [[INTERLEAVE]], ptr [[PTR]], align 4 +; CHECK-NEXT: ret void +; + %interleave = tail call <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8> %s1, <vscale x 8 x i8> %s2, <vscale x 8 x i8> %s3) + + store <vscale x 24 x i8> %interleave, ptr %ptr, align 4 + ret void +} + 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>) @@ -312,4 +531,15 @@ declare <vscale x 4 x ptr> @llvm.vector.interleave2.nxv4p0(<vscale x 2 x ptr>, < ; Larger interleaves to test 'legalization' declare <vscale x 8 x double> @llvm.vector.interleave2.nxv8f64(<vscale x 4 x double>, <vscale x 4 x double>) +; De-Interleaves with Factor=3 +declare { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave3.nxv3i64(<vscale x 3 x i64>) +declare { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave3.nxv12i64(<vscale x 12 x i64>) +declare { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } @llvm.vector.deinterleave3.nxv24i64(<vscale x 24 x i64>) +declare { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } @llvm.vector.deinterleave3.nxv24i8(<vscale x 24 x i8>) + +; Interleaves with Factor=3 +declare <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8>) +declare <vscale x 24 x i64> @llvm.vector.interleave3.nxv24i64(<vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64>) +declare <vscale x 12 x i64> @llvm.vector.interleave3.nxv12i64(<vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64>) + attributes #0 = { vscale_range(1,16) "target-features"="+sve" } diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll index 6cf11be..6fe6883 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/conditional-branches-cost.ll @@ -660,16 +660,17 @@ define void @low_trip_count_fold_tail_scalarized_store(ptr %dst) { ; COMMON-NEXT: store i8 6, ptr [[TMP6]], align 1 ; COMMON-NEXT: br label %[[PRED_STORE_CONTINUE12]] ; COMMON: [[PRED_STORE_CONTINUE12]]: -; COMMON-NEXT: br i1 false, label %[[PRED_STORE_IF13:.*]], label %[[EXIT1:.*]] +; COMMON-NEXT: br i1 false, label %[[PRED_STORE_IF13:.*]], label %[[EXIT:.*]] ; COMMON: [[PRED_STORE_IF13]]: ; COMMON-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[DST]], i64 7 ; COMMON-NEXT: store i8 7, ptr [[TMP7]], align 1 -; COMMON-NEXT: br label %[[EXIT1]] +; COMMON-NEXT: br label %[[EXIT]] +; COMMON: [[EXIT]]: +; COMMON-NEXT: br label %[[SCALAR_PH:.*]] +; COMMON: [[SCALAR_PH]]: +; COMMON-NEXT: br label %[[EXIT1:.*]] ; COMMON: [[EXIT1]]: -; COMMON-NEXT: br label %[[SCALAR_PH1:.*]] -; COMMON: [[SCALAR_PH1]]: -; COMMON-NEXT: br [[EXIT:label %.*]] -; COMMON: [[SCALAR_PH:.*:]] +; COMMON-NEXT: ret void ; entry: br label %loop diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/interleave-with-gaps.ll b/llvm/test/Transforms/LoopVectorize/AArch64/interleave-with-gaps.ll index 93e71af..e3e4833 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/interleave-with-gaps.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/interleave-with-gaps.ll @@ -42,7 +42,7 @@ define i64 @vector_loop_with_remaining_iterations(ptr %src, ptr noalias %dst, i3 ; CHECK-NEXT: br label %[[VEC_EPILOG_ITER_CHECK:.*]] ; CHECK: [[VEC_EPILOG_ITER_CHECK]]: ; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ule i64 1, [[TMP1]] -; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]] ; CHECK: [[VEC_EPILOG_PH]]: ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 16, %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP13]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ] @@ -80,7 +80,7 @@ define i64 @vector_loop_with_remaining_iterations(ptr %src, ptr noalias %dst, i3 ; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX7]], [[TMP17]] ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT6]] ; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC]] -; CHECK-NEXT: br i1 [[TMP36]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP36]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]: ; CHECK-NEXT: [[TMP37:%.*]] = call i64 @llvm.vector.reduce.or.nxv2i64(<vscale x 2 x i64> [[TMP35]]) ; CHECK-NEXT: br label %[[VEC_EPILOG_SCALAR_PH]] @@ -104,7 +104,7 @@ define i64 @vector_loop_with_remaining_iterations(ptr %src, ptr noalias %dst, i3 ; CHECK-NEXT: [[RED_NEXT]] = or i64 [[RED]], [[MIN_EXT]] ; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1 ; CHECK-NEXT: [[EXITCOND_NOT_I_I:%.*]] = icmp eq i64 [[IV_NEXT]], 17 -; CHECK-NEXT: br i1 [[EXITCOND_NOT_I_I]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP4:![0-9]+]] +; CHECK-NEXT: br i1 [[EXITCOND_NOT_I_I]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP5:![0-9]+]] ; CHECK: [[EXIT]]: ; CHECK-NEXT: [[RED_NEXT_LCSSA:%.*]] = phi i64 [ [[RED_NEXT]], %[[LOOP]] ] ; CHECK-NEXT: ret i64 [[RED_NEXT_LCSSA]] @@ -167,13 +167,13 @@ define i64 @main_vector_loop_fixed_with_no_remaining_iterations(ptr %src, ptr no ; CHECK-NEXT: [[TMP11]] = or <16 x i64> [[VEC_PHI]], [[TMP10]] ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16 -; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: ; CHECK-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.or.v16i64(<16 x i64> [[TMP11]]) ; CHECK-NEXT: br label %[[VEC_EPILOG_ITER_CHECK:.*]] ; CHECK: [[VEC_EPILOG_ITER_CHECK]]: ; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ule i64 1, [[TMP1]] -; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF3]] ; CHECK: [[VEC_EPILOG_PH]]: ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 16, %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP13]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ] @@ -211,7 +211,7 @@ define i64 @main_vector_loop_fixed_with_no_remaining_iterations(ptr %src, ptr no ; CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX7]], [[TMP17]] ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT6]] ; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC]] -; CHECK-NEXT: br i1 [[TMP36]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP36]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]: ; CHECK-NEXT: [[TMP37:%.*]] = call i64 @llvm.vector.reduce.or.nxv2i64(<vscale x 2 x i64> [[TMP35]]) ; CHECK-NEXT: br label %[[VEC_EPILOG_SCALAR_PH]] @@ -235,7 +235,7 @@ define i64 @main_vector_loop_fixed_with_no_remaining_iterations(ptr %src, ptr no ; CHECK-NEXT: [[RED_NEXT]] = or i64 [[RED]], [[MIN_EXT]] ; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1 ; CHECK-NEXT: [[EXITCOND_NOT_I_I:%.*]] = icmp eq i64 [[IV_NEXT]], 17 -; CHECK-NEXT: br i1 [[EXITCOND_NOT_I_I]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP7:![0-9]+]] +; CHECK-NEXT: br i1 [[EXITCOND_NOT_I_I]], label %[[EXIT:.*]], label %[[LOOP]], !llvm.loop [[LOOP8:![0-9]+]] ; CHECK: [[EXIT]]: ; CHECK-NEXT: [[RED_NEXT_LCSSA:%.*]] = phi i64 [ [[RED_NEXT]], %[[LOOP]] ] ; CHECK-NEXT: ret i64 [[RED_NEXT_LCSSA]] @@ -308,7 +308,7 @@ define void @main_vector_loop_fixed_single_vector_iteration_with_runtime_checks( ; CHECK-NEXT: store i64 0, ptr [[L]], align 8 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 ; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4 -; CHECK-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: ; CHECK-NEXT: br label %[[SCALAR_PH:.*]] ; CHECK: [[SCALAR_PH]]: @@ -332,7 +332,7 @@ define void @main_vector_loop_fixed_single_vector_iteration_with_runtime_checks( ; CHECK-NEXT: store i64 0, ptr [[L]], align 8 ; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV1]], 2 ; CHECK-NEXT: [[EC:%.*]] = icmp ult i64 [[IV1]], 14 -; CHECK-NEXT: br i1 [[EC]], label %[[LOOP]], label %[[EXIT:.*]], !llvm.loop [[LOOP10:![0-9]+]] +; CHECK-NEXT: br i1 [[EC]], label %[[LOOP]], label %[[EXIT:.*]], !llvm.loop [[LOOP11:![0-9]+]] ; CHECK: [[EXIT]]: ; CHECK-NEXT: ret void ; diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll b/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll index e424649..75b18ff 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/low_trip_count_predicates.ll @@ -541,3 +541,22 @@ exit: ; preds = %for.body ; CHECK-VS1: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} ; CHECK-VS1: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} ; CHECK-VS1: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} +; CHECK-VS1: [[PROF3]] = !{!"branch_weights", i32 8, i32 8} +; CHECK-VS1: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]} +; CHECK-VS1: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]]} +; CHECK-VS1: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]} +; CHECK-VS1: [[LOOP7]] = distinct !{[[LOOP7]], [[META1]]} +; CHECK-VS1: [[LOOP8]] = distinct !{[[LOOP8]], [[META1]], [[META2]]} +; CHECK-VS1: [[PROF9]] = !{!"branch_weights", i32 10, i32 30} +;. +; CHECK-VS2: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]} +; CHECK-VS2: [[META1]] = !{!"llvm.loop.isvectorized", i32 1} +; CHECK-VS2: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"} +; CHECK-VS2: [[PROF3]] = !{!"branch_weights", i32 8, i32 8} +; CHECK-VS2: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]} +; CHECK-VS2: [[LOOP5]] = distinct !{[[LOOP5]], [[META1]]} +; CHECK-VS2: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]} +; CHECK-VS2: [[LOOP7]] = distinct !{[[LOOP7]], [[META1]]} +; CHECK-VS2: [[LOOP8]] = distinct !{[[LOOP8]], [[META1]], [[META2]]} +; CHECK-VS2: [[PROF9]] = !{!"branch_weights", i32 10, i32 30} +;. diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/multiple-result-intrinsics.ll b/llvm/test/Transforms/LoopVectorize/AArch64/multiple-result-intrinsics.ll index a6e0f8a..300f5d9 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/multiple-result-intrinsics.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/multiple-result-intrinsics.ll @@ -40,6 +40,7 @@ define void @sincos_f32(ptr noalias %in, ptr noalias writeonly %out_a, ptr noali ; CHECK-ARMPL: [[ENTRY:.*:]] ; CHECK-ARMPL: [[VECTOR_PH:.*:]] ; CHECK-ARMPL: [[VECTOR_BODY:.*:]] +; CHECK-ARMPL: [[VECTOR_BODY1:.*:]] ; CHECK-ARMPL: [[TMP12:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.sincos.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD:%.*]]) ; CHECK-ARMPL: [[TMP13:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.sincos.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD1:%.*]]) ; CHECK-ARMPL: [[TMP14:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP12]], 0 @@ -53,6 +54,15 @@ define void @sincos_f32(ptr noalias %in, ptr noalias writeonly %out_a, ptr noali ; CHECK-ARMPL: [[MIDDLE_BLOCK:.*:]] ; CHECK-ARMPL: [[SCALAR_PH:.*:]] ; CHECK-ARMPL: [[FOR_BODY:.*:]] +; CHECK-ARMPL: [[VEC_EPILOG_VECTOR_BODY:.*:]] +; CHECK-ARMPL: [[TMP29:%.*]] = call { <4 x float>, <4 x float> } @llvm.sincos.v4f32(<4 x float> [[WIDE_LOAD3:%.*]]) +; CHECK-ARMPL: [[TMP25:%.*]] = extractvalue { <4 x float>, <4 x float> } [[TMP29]], 0 +; CHECK-ARMPL: [[TMP26:%.*]] = extractvalue { <4 x float>, <4 x float> } [[TMP29]], 1 +; CHECK-ARMPL: store <4 x float> [[TMP25]], ptr [[TMP30:%.*]], align 4 +; CHECK-ARMPL: store <4 x float> [[TMP26]], ptr [[TMP28:%.*]], align 4 +; CHECK-ARMPL: [[VEC_EPILOG_MIDDLE_BLOCK:.*:]] +; CHECK-ARMPL: [[VEC_EPILOG_SCALAR_PH:.*:]] +; CHECK-ARMPL: [[FOR_BODY1:.*:]] ; CHECK-ARMPL: [[CALL:%.*]] = tail call { float, float } @llvm.sincos.f32(float [[IN_VAL:%.*]]) ; CHECK-ARMPL: [[EXTRACT_A:%.*]] = extractvalue { float, float } [[CALL]], 0 ; CHECK-ARMPL: [[EXTRACT_B:%.*]] = extractvalue { float, float } [[CALL]], 1 @@ -262,6 +272,7 @@ define void @modf_f32(ptr noalias %in, ptr noalias writeonly %out_a, ptr noalias ; CHECK-ARMPL: [[ENTRY:.*:]] ; CHECK-ARMPL: [[VECTOR_PH:.*:]] ; CHECK-ARMPL: [[VECTOR_BODY:.*:]] +; CHECK-ARMPL: [[VECTOR_BODY1:.*:]] ; CHECK-ARMPL: [[TMP12:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.modf.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD:%.*]]) ; CHECK-ARMPL: [[TMP13:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.modf.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD1:%.*]]) ; CHECK-ARMPL: [[TMP14:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP12]], 0 @@ -275,6 +286,15 @@ define void @modf_f32(ptr noalias %in, ptr noalias writeonly %out_a, ptr noalias ; CHECK-ARMPL: [[MIDDLE_BLOCK:.*:]] ; CHECK-ARMPL: [[SCALAR_PH:.*:]] ; CHECK-ARMPL: [[FOR_BODY:.*:]] +; CHECK-ARMPL: [[VEC_EPILOG_VECTOR_BODY:.*:]] +; CHECK-ARMPL: [[TMP29:%.*]] = call { <4 x float>, <4 x float> } @llvm.modf.v4f32(<4 x float> [[WIDE_LOAD3:%.*]]) +; CHECK-ARMPL: [[TMP25:%.*]] = extractvalue { <4 x float>, <4 x float> } [[TMP29]], 0 +; CHECK-ARMPL: [[TMP26:%.*]] = extractvalue { <4 x float>, <4 x float> } [[TMP29]], 1 +; CHECK-ARMPL: store <4 x float> [[TMP25]], ptr [[TMP30:%.*]], align 4 +; CHECK-ARMPL: store <4 x float> [[TMP26]], ptr [[TMP28:%.*]], align 4 +; CHECK-ARMPL: [[VEC_EPILOG_MIDDLE_BLOCK:.*:]] +; CHECK-ARMPL: [[VEC_EPILOG_SCALAR_PH:.*:]] +; CHECK-ARMPL: [[FOR_BODY1:.*:]] ; CHECK-ARMPL: [[CALL:%.*]] = tail call { float, float } @llvm.modf.f32(float [[IN_VAL:%.*]]) ; CHECK-ARMPL: [[EXTRACT_A:%.*]] = extractvalue { float, float } [[CALL]], 0 ; CHECK-ARMPL: [[EXTRACT_B:%.*]] = extractvalue { float, float } [[CALL]], 1 @@ -412,6 +432,7 @@ define void @sincospi_f32(ptr noalias %in, ptr noalias writeonly %out_a, ptr noa ; CHECK-ARMPL: [[ENTRY:.*:]] ; CHECK-ARMPL: [[VECTOR_PH:.*:]] ; CHECK-ARMPL: [[VECTOR_BODY:.*:]] +; CHECK-ARMPL: [[VECTOR_BODY1:.*:]] ; CHECK-ARMPL: [[TMP12:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.sincospi.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD:%.*]]) ; CHECK-ARMPL: [[TMP13:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.sincospi.nxv4f32(<vscale x 4 x float> [[WIDE_LOAD1:%.*]]) ; CHECK-ARMPL: [[TMP14:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[TMP12]], 0 @@ -425,6 +446,15 @@ define void @sincospi_f32(ptr noalias %in, ptr noalias writeonly %out_a, ptr noa ; CHECK-ARMPL: [[MIDDLE_BLOCK:.*:]] ; CHECK-ARMPL: [[SCALAR_PH:.*:]] ; CHECK-ARMPL: [[FOR_BODY:.*:]] +; CHECK-ARMPL: [[VEC_EPILOG_VECTOR_BODY:.*:]] +; CHECK-ARMPL: [[TMP29:%.*]] = call { <4 x float>, <4 x float> } @llvm.sincospi.v4f32(<4 x float> [[WIDE_LOAD3:%.*]]) +; CHECK-ARMPL: [[TMP25:%.*]] = extractvalue { <4 x float>, <4 x float> } [[TMP29]], 0 +; CHECK-ARMPL: [[TMP26:%.*]] = extractvalue { <4 x float>, <4 x float> } [[TMP29]], 1 +; CHECK-ARMPL: store <4 x float> [[TMP25]], ptr [[TMP30:%.*]], align 4 +; CHECK-ARMPL: store <4 x float> [[TMP26]], ptr [[TMP28:%.*]], align 4 +; CHECK-ARMPL: [[VEC_EPILOG_MIDDLE_BLOCK:.*:]] +; CHECK-ARMPL: [[VEC_EPILOG_SCALAR_PH:.*:]] +; CHECK-ARMPL: [[FOR_BODY1:.*:]] ; CHECK-ARMPL: [[CALL:%.*]] = tail call { float, float } @llvm.sincospi.f32(float [[IN_VAL:%.*]]) ; CHECK-ARMPL: [[EXTRACT_A:%.*]] = extractvalue { float, float } [[CALL]], 0 ; CHECK-ARMPL: [[EXTRACT_B:%.*]] = extractvalue { float, float } [[CALL]], 1 diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/scalable-struct-return.ll b/llvm/test/Transforms/LoopVectorize/AArch64/scalable-struct-return.ll index 8830ce3..5f79d02 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/scalable-struct-return.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/scalable-struct-return.ll @@ -38,8 +38,9 @@ define void @struct_return_f32_widen(ptr noalias %in, ptr noalias writeonly %out ; CHECK-NEXT: [[TMP14:%.*]] = xor i1 [[TMP13]], true ; CHECK-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[EXIT:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[EXIT:.*]] +; CHECK: [[EXIT]]: +; CHECK-NEXT: ret void ; entry: br label %for.body @@ -96,8 +97,9 @@ define void @struct_return_f64_widen(ptr noalias %in, ptr noalias writeonly %out ; CHECK-NEXT: [[TMP14:%.*]] = xor i1 [[TMP13]], true ; CHECK-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[EXIT:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[EXIT:.*]] +; CHECK: [[EXIT]]: +; CHECK-NEXT: ret void ; entry: br label %for.body diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/sve-low-trip-count.ll b/llvm/test/Transforms/LoopVectorize/AArch64/sve-low-trip-count.ll index d447517..f03f743 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/sve-low-trip-count.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/sve-low-trip-count.ll @@ -29,8 +29,9 @@ define void @trip7_i64(ptr noalias nocapture noundef %dst, ptr noalias nocapture ; CHECK-NEXT: [[COND:%.*]] = xor i1 [[EXTRACT_FIRST_LANE_MASK]], true ; CHECK-NEXT: br i1 [[COND]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[EXIT:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[EXIT:.*]] +; CHECK: [[EXIT]]: +; CHECK-NEXT: ret void ; entry: br label %loop diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/tail-fold-uniform-memops.ll b/llvm/test/Transforms/LoopVectorize/AArch64/tail-fold-uniform-memops.ll index b8f4e84..753847f 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/tail-fold-uniform-memops.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/tail-fold-uniform-memops.ll @@ -33,8 +33,9 @@ define void @uniform_load(ptr noalias %dst, ptr noalias readonly %src, i64 %n) # ; CHECK-NEXT: [[FIRST_LANE_SET:%.*]] = xor i1 [[EXTRACT_FIRST_LANE_MASK]], true ; CHECK-NEXT: br i1 [[FIRST_LANE_SET]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[FOR_END:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[FOR_END:.*]] +; CHECK: [[FOR_END]]: +; CHECK-NEXT: ret void ; entry: @@ -87,8 +88,9 @@ define void @cond_uniform_load(ptr noalias nocapture %dst, ptr nocapture readonl ; CHECK-NEXT: [[TMP9:%.*]] = xor i1 [[TMP8]], true ; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[FOR_END:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[FOR_END:.*]] +; CHECK: [[FOR_END]]: +; CHECK-NEXT: ret void ; entry: br label %for.body diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/safe-dep-distance.ll b/llvm/test/Transforms/LoopVectorize/RISCV/safe-dep-distance.ll index e046816..e84c0d6 100644 --- a/llvm/test/Transforms/LoopVectorize/RISCV/safe-dep-distance.ll +++ b/llvm/test/Transforms/LoopVectorize/RISCV/safe-dep-distance.ll @@ -67,7 +67,7 @@ define void @test_may_clobber(ptr %p) { ; CHECK-NEXT: store <4 x i64> [[WIDE_LOAD]], ptr [[TMP4]], align 32 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 200 -; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: exit: @@ -111,7 +111,7 @@ define void @trivial_due_max_vscale(ptr %p) { ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP11]], [[INDEX]] ; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: exit: @@ -155,7 +155,7 @@ define void @no_high_lmul_or_interleave(ptr %p) { ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP11]], [[INDEX]] ; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: exit: @@ -207,7 +207,7 @@ define void @safe_load_store_distance_not_pow_of_2(i64 %N) { ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <8 x i64> [[VEC_IND]], splat (i64 24) ; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] -; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[SCALAR_PH]] ; CHECK: scalar.ph: @@ -221,7 +221,7 @@ define void @safe_load_store_distance_not_pow_of_2(i64 %N) { ; CHECK-NEXT: store i16 0, ptr [[GEP_OFF]], align 2 ; CHECK-NEXT: [[IV_NEXT]] = add nsw i64 [[IV]], 3 ; CHECK-NEXT: [[CMP:%.*]] = icmp ult i64 [[IV]], [[N]] -; CHECK-NEXT: br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]], !llvm.loop [[LOOP8:![0-9]+]] +; CHECK-NEXT: br i1 [[CMP]], label [[LOOP]], label [[EXIT:%.*]], !llvm.loop [[LOOP7:![0-9]+]] ; CHECK: exit: ; CHECK-NEXT: ret void ; diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll index 2fbc73e..c66d8d6 100644 --- a/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll +++ b/llvm/test/Transforms/LoopVectorize/RISCV/strided-accesses.ll @@ -133,7 +133,7 @@ define void @single_constant_stride_int_iv(ptr %p) { ; CHECK-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[DOTSPLAT]] ; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: exit: @@ -237,7 +237,7 @@ define void @single_constant_stride_ptr_iv(ptr %p) { ; CHECK-NEXT: [[TMP12:%.*]] = mul i64 8, [[TMP9]] ; CHECK-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP12]] ; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: exit: @@ -346,7 +346,7 @@ define void @single_stride_int_scaled(ptr %p, i64 %stride) { ; NOSTRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP11]], [[INDEX]] ; NOSTRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; NOSTRIDED-NEXT: [[TMP9:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; NOSTRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; NOSTRIDED: middle.block: ; NOSTRIDED-NEXT: br label [[EXIT:%.*]] ; NOSTRIDED: scalar.ph: @@ -360,7 +360,7 @@ define void @single_stride_int_scaled(ptr %p, i64 %stride) { ; NOSTRIDED-NEXT: store i32 [[Y0]], ptr [[Q0]], align 4 ; NOSTRIDED-NEXT: [[NEXTI]] = add i64 [[I]], 1 ; NOSTRIDED-NEXT: [[DONE:%.*]] = icmp eq i64 [[NEXTI]], 1024 -; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP7:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP6:![0-9]+]] ; NOSTRIDED: exit: ; NOSTRIDED-NEXT: ret void ; @@ -468,7 +468,7 @@ define void @single_stride_int_iv(ptr %p, i64 %stride) { ; NOSTRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP11]], [[INDEX]] ; NOSTRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; NOSTRIDED-NEXT: [[TMP9:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; NOSTRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; NOSTRIDED: middle.block: ; NOSTRIDED-NEXT: br label [[EXIT:%.*]] ; NOSTRIDED: scalar.ph: @@ -483,7 +483,7 @@ define void @single_stride_int_iv(ptr %p, i64 %stride) { ; NOSTRIDED-NEXT: [[OFFSET_NEXT]] = add nuw nsw i64 [[OFFSET]], [[STRIDE]] ; NOSTRIDED-NEXT: [[NEXTI]] = add i64 [[I]], 1 ; NOSTRIDED-NEXT: [[DONE:%.*]] = icmp eq i64 [[NEXTI]], 1024 -; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP9:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP8:![0-9]+]] ; NOSTRIDED: exit: ; NOSTRIDED-NEXT: ret void ; @@ -640,7 +640,7 @@ define void @double_stride_int_scaled(ptr %p, ptr %p2, i64 %stride) { ; NOSTRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP13]], [[INDEX]] ; NOSTRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP13]] ; NOSTRIDED-NEXT: [[TMP11:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; NOSTRIDED-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] ; NOSTRIDED: middle.block: ; NOSTRIDED-NEXT: br label [[EXIT:%.*]] ; NOSTRIDED: scalar.ph: @@ -656,7 +656,7 @@ define void @double_stride_int_scaled(ptr %p, ptr %p2, i64 %stride) { ; NOSTRIDED-NEXT: store i32 [[Y0]], ptr [[Q1]], align 4 ; NOSTRIDED-NEXT: [[NEXTI]] = add i64 [[I]], 1 ; NOSTRIDED-NEXT: [[DONE:%.*]] = icmp eq i64 [[NEXTI]], 1024 -; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP11:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP10:![0-9]+]] ; NOSTRIDED: exit: ; NOSTRIDED-NEXT: ret void ; @@ -790,14 +790,14 @@ define void @double_stride_int_scaled(ptr %p, ptr %p2, i64 %stride) { ; STRIDED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <vscale x 4 x i64> [[BROADCAST_SPLATINSERT9]], <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer ; STRIDED-NEXT: [[TMP18:%.*]] = mul nuw nsw <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT1]] ; STRIDED-NEXT: [[TMP19:%.*]] = getelementptr i32, ptr [[P]], <vscale x 4 x i64> [[TMP18]] -; STRIDED-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.vp.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP19]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP43]]), !alias.scope [[META6:![0-9]+]] +; STRIDED-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.vp.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[TMP19]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP43]]), !alias.scope [[META5:![0-9]+]] ; STRIDED-NEXT: [[TMP20:%.*]] = add <vscale x 4 x i32> [[WIDE_MASKED_GATHER]], splat (i32 1) ; STRIDED-NEXT: [[TMP21:%.*]] = getelementptr i32, ptr [[P2]], <vscale x 4 x i64> [[TMP18]] -; STRIDED-NEXT: call void @llvm.vp.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[TMP20]], <vscale x 4 x ptr> align 4 [[TMP21]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP43]]), !alias.scope [[META9:![0-9]+]], !noalias [[META6]] +; STRIDED-NEXT: call void @llvm.vp.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[TMP20]], <vscale x 4 x ptr> align 4 [[TMP21]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP43]]), !alias.scope [[META8:![0-9]+]], !noalias [[META5]] ; STRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP44]] ; STRIDED-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; STRIDED-NEXT: [[TMP41:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; STRIDED-NEXT: br i1 [[TMP41]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] +; STRIDED-NEXT: br i1 [[TMP41]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] ; STRIDED: middle.block: ; STRIDED-NEXT: br label [[EXIT:%.*]] ; STRIDED: scalar.ph: @@ -813,7 +813,7 @@ define void @double_stride_int_scaled(ptr %p, ptr %p2, i64 %stride) { ; STRIDED-NEXT: store i32 [[Y0]], ptr [[Q1]], align 4 ; STRIDED-NEXT: [[NEXTI]] = add i64 [[I]], 1 ; STRIDED-NEXT: [[DONE:%.*]] = icmp eq i64 [[NEXTI]], 1024 -; STRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP12:![0-9]+]] +; STRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP11:![0-9]+]] ; STRIDED: exit: ; STRIDED-NEXT: ret void ; @@ -965,7 +965,7 @@ define void @double_stride_int_iv(ptr %p, ptr %p2, i64 %stride) { ; NOSTRIDED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[TMP11]], [[INDEX]] ; NOSTRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; NOSTRIDED-NEXT: [[TMP9:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; NOSTRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] ; NOSTRIDED: middle.block: ; NOSTRIDED-NEXT: br label [[EXIT:%.*]] ; NOSTRIDED: scalar.ph: @@ -981,7 +981,7 @@ define void @double_stride_int_iv(ptr %p, ptr %p2, i64 %stride) { ; NOSTRIDED-NEXT: [[OFFSET_NEXT]] = add nuw nsw i64 [[OFFSET]], [[STRIDE]] ; NOSTRIDED-NEXT: [[NEXTI]] = add i64 [[I]], 1 ; NOSTRIDED-NEXT: [[DONE:%.*]] = icmp eq i64 [[NEXTI]], 1024 -; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP13:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP12:![0-9]+]] ; NOSTRIDED: exit: ; NOSTRIDED-NEXT: ret void ; @@ -1145,16 +1145,16 @@ define void @double_stride_ptr_iv(ptr %p, ptr %p2, i64 %stride) { ; STRIDED-NEXT: [[VECTOR_GEP:%.*]] = getelementptr i8, ptr [[POINTER_PHI11]], <vscale x 4 x i64> [[TMP18]] ; STRIDED-NEXT: [[VECTOR_GEP7:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <vscale x 4 x i64> [[TMP18]] ; STRIDED-NEXT: [[TMP14:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 4, i1 true) -; STRIDED-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.vp.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[VECTOR_GEP7]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP14]]), !alias.scope [[META13:![0-9]+]] +; STRIDED-NEXT: [[WIDE_MASKED_GATHER:%.*]] = call <vscale x 4 x i32> @llvm.vp.gather.nxv4i32.nxv4p0(<vscale x 4 x ptr> align 4 [[VECTOR_GEP7]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP14]]), !alias.scope [[META12:![0-9]+]] ; STRIDED-NEXT: [[TMP30:%.*]] = add <vscale x 4 x i32> [[WIDE_MASKED_GATHER]], splat (i32 1) -; STRIDED-NEXT: call void @llvm.vp.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[TMP30]], <vscale x 4 x ptr> align 4 [[VECTOR_GEP]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP14]]), !alias.scope [[META16:![0-9]+]], !noalias [[META13]] +; STRIDED-NEXT: call void @llvm.vp.scatter.nxv4i32.nxv4p0(<vscale x 4 x i32> [[TMP30]], <vscale x 4 x ptr> align 4 [[VECTOR_GEP]], <vscale x 4 x i1> splat (i1 true), i32 [[TMP14]]), !alias.scope [[META15:![0-9]+]], !noalias [[META12]] ; STRIDED-NEXT: [[TMP16:%.*]] = zext i32 [[TMP14]] to i64 ; STRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP16]] ; STRIDED-NEXT: [[TMP25:%.*]] = mul i64 [[STRIDE]], [[TMP16]] ; STRIDED-NEXT: [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 [[TMP25]] ; STRIDED-NEXT: [[PTR_IND12]] = getelementptr i8, ptr [[POINTER_PHI11]], i64 [[TMP25]] ; STRIDED-NEXT: [[TMP23:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; STRIDED-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] +; STRIDED-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP17:![0-9]+]] ; STRIDED: middle.block: ; STRIDED-NEXT: br label [[EXIT:%.*]] ; STRIDED: scalar.ph: @@ -1170,7 +1170,7 @@ define void @double_stride_ptr_iv(ptr %p, ptr %p2, i64 %stride) { ; STRIDED-NEXT: [[PTR2_NEXT]] = getelementptr inbounds i8, ptr [[PTR2]], i64 [[STRIDE]] ; STRIDED-NEXT: [[NEXTI]] = add i64 [[I]], 1 ; STRIDED-NEXT: [[DONE:%.*]] = icmp eq i64 [[NEXTI]], 1024 -; STRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP19:![0-9]+]] +; STRIDED-NEXT: br i1 [[DONE]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP18:![0-9]+]] ; STRIDED: exit: ; STRIDED-NEXT: ret void ; @@ -1318,7 +1318,7 @@ define void @constant_stride_reinterpret(ptr noalias %in, ptr noalias %out) { ; NOSTRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP3]] ; NOSTRIDED-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; NOSTRIDED-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; NOSTRIDED-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]] +; NOSTRIDED-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP13:![0-9]+]] ; NOSTRIDED: middle.block: ; NOSTRIDED-NEXT: br label [[LOOP:%.*]] ; NOSTRIDED: exit: @@ -1402,7 +1402,7 @@ define void @constant_stride_reinterpret(ptr noalias %in, ptr noalias %out) { ; STRIDED-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP3]] ; STRIDED-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; STRIDED-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; STRIDED-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]] +; STRIDED-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP19:![0-9]+]] ; STRIDED: middle.block: ; STRIDED-NEXT: br label [[LOOP:%.*]] ; STRIDED: exit: diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-interleave.ll b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-interleave.ll index 0c22a9e..46daee4 100644 --- a/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-interleave.ll +++ b/llvm/test/Transforms/LoopVectorize/RISCV/tail-folding-interleave.ll @@ -142,7 +142,7 @@ define i32 @load_factor_4_with_gap(i64 %n, ptr noalias %a) { ; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP5]] ; IF-EVL-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; IF-EVL-NEXT: [[TMP14:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; IF-EVL-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; IF-EVL-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; IF-EVL: middle.block: ; IF-EVL-NEXT: [[TMP15:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP12]]) ; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] @@ -267,7 +267,7 @@ define void @store_factor_4_with_gap(i32 %n, ptr noalias %a) { ; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i32 [[AVL]], [[TMP6]] ; IF-EVL-NEXT: [[VEC_IND_NEXT5]] = add <vscale x 4 x i32> [[VEC_IND2]], [[BROADCAST_SPLAT]] ; IF-EVL-NEXT: [[TMP8:%.*]] = icmp eq i32 [[AVL_NEXT]], 0 -; IF-EVL-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; IF-EVL-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; IF-EVL: middle.block: ; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] ; IF-EVL: exit: @@ -382,7 +382,7 @@ define i32 @load_factor_4_with_tail_gap(i64 %n, ptr noalias %a) { ; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP5]] ; IF-EVL-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; IF-EVL-NEXT: [[TMP14:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; IF-EVL-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; IF-EVL-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; IF-EVL: middle.block: ; IF-EVL-NEXT: [[TMP15:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP12]]) ; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] @@ -508,7 +508,7 @@ define void @store_factor_4_with_tail_gap(i32 %n, ptr noalias %a) { ; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i32 [[AVL]], [[TMP6]] ; IF-EVL-NEXT: [[VEC_IND_NEXT5]] = add <vscale x 4 x i32> [[VEC_IND2]], [[BROADCAST_SPLAT]] ; IF-EVL-NEXT: [[TMP8:%.*]] = icmp eq i32 [[AVL_NEXT]], 0 -; IF-EVL-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] +; IF-EVL-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; IF-EVL: middle.block: ; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] ; IF-EVL: exit: @@ -621,7 +621,7 @@ define i32 @load_factor_4_reverse(i64 %n, ptr noalias %a) { ; IF-EVL-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP7]] ; IF-EVL-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]] ; IF-EVL-NEXT: [[TMP18:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; IF-EVL-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] +; IF-EVL-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; IF-EVL: middle.block: ; IF-EVL-NEXT: [[TMP19:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32(<vscale x 4 x i32> [[TMP16]]) ; IF-EVL-NEXT: br label [[FOR_BODY:%.*]] diff --git a/llvm/test/Transforms/LoopVectorize/RISCV/uniform-load-store.ll b/llvm/test/Transforms/LoopVectorize/RISCV/uniform-load-store.ll index bae97e5..c34417b 100644 --- a/llvm/test/Transforms/LoopVectorize/RISCV/uniform-load-store.ll +++ b/llvm/test/Transforms/LoopVectorize/RISCV/uniform-load-store.ll @@ -129,7 +129,7 @@ define i64 @uniform_load_outside_use(ptr noalias nocapture %a, ptr noalias nocap ; SCALABLE-NEXT: store <vscale x 2 x i64> [[BROADCAST_SPLAT]], ptr [[TMP8]], align 8 ; SCALABLE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; SCALABLE-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] -; SCALABLE-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP10]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: [[CMP_N:%.*]] = icmp eq i64 1025, [[N_VEC]] ; SCALABLE-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] @@ -143,7 +143,7 @@ define i64 @uniform_load_outside_use(ptr noalias nocapture %a, ptr noalias nocap ; SCALABLE-NEXT: store i64 [[V]], ptr [[ARRAYIDX]], align 8 ; SCALABLE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 ; SCALABLE-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1025 -; SCALABLE-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; SCALABLE-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; SCALABLE: [[FOR_END]]: ; SCALABLE-NEXT: [[V_LCSSA:%.*]] = phi i64 [ [[V]], %[[FOR_BODY]] ], [ [[TMP6]], %[[MIDDLE_BLOCK]] ] ; SCALABLE-NEXT: ret i64 [[V_LCSSA]] @@ -204,7 +204,7 @@ define i64 @uniform_load_outside_use(ptr noalias nocapture %a, ptr noalias nocap ; TF-SCALABLE-NEXT: store <vscale x 2 x i64> [[BROADCAST_SPLAT]], ptr [[ARRAYIDX]], align 8 ; TF-SCALABLE-NEXT: [[INDEX_NEXT]] = add nuw i64 [[IV]], [[TMP3]] ; TF-SCALABLE-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] -; TF-SCALABLE-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: [[CMP_N:%.*]] = icmp eq i64 1025, [[N_VEC]] ; TF-SCALABLE-NEXT: br i1 [[CMP_N]], label %[[FOR_END:.*]], label %[[SCALAR_PH]] @@ -218,7 +218,7 @@ define i64 @uniform_load_outside_use(ptr noalias nocapture %a, ptr noalias nocap ; TF-SCALABLE-NEXT: store i64 [[V1]], ptr [[ARRAYIDX1]], align 8 ; TF-SCALABLE-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV1]], 1 ; TF-SCALABLE-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], 1025 -; TF-SCALABLE-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_END]], label %[[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; TF-SCALABLE: [[FOR_END]]: ; TF-SCALABLE-NEXT: [[V_LCSSA:%.*]] = phi i64 [ [[V1]], %[[FOR_BODY]] ], [ [[V]], %[[MIDDLE_BLOCK]] ] ; TF-SCALABLE-NEXT: ret i64 [[V_LCSSA]] @@ -269,7 +269,7 @@ define void @conditional_uniform_load(ptr noalias nocapture %a, ptr noalias noca ; SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP8]] ; SCALABLE-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[DOTSPLAT]] ; SCALABLE-NEXT: [[TMP14:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; SCALABLE-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP14]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: br label %[[FOR_END:.*]] ; SCALABLE: [[FOR_END]]: @@ -350,7 +350,7 @@ define void @conditional_uniform_load(ptr noalias nocapture %a, ptr noalias noca ; TF-SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; TF-SCALABLE-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[DOTSPLAT]] ; TF-SCALABLE-NEXT: [[TMP17:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; TF-SCALABLE-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP17]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: br label %[[FOR_END:.*]] ; TF-SCALABLE: [[FOR_END]]: @@ -399,7 +399,7 @@ define void @uniform_load_unaligned(ptr noalias nocapture %a, ptr noalias nocapt ; SCALABLE-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP11]], [[INDEX]] ; SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; SCALABLE-NEXT: [[TMP7:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; SCALABLE-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP7]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: br label %[[FOR_END:.*]] ; SCALABLE: [[FOR_END]]: @@ -457,7 +457,7 @@ define void @uniform_load_unaligned(ptr noalias nocapture %a, ptr noalias nocapt ; TF-SCALABLE-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP8]], [[INDEX]] ; TF-SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP8]] ; TF-SCALABLE-NEXT: [[TMP9:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; TF-SCALABLE-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: br label %[[FOR_END:.*]] ; TF-SCALABLE: [[FOR_END]]: @@ -499,7 +499,7 @@ define void @uniform_store(ptr noalias nocapture %a, ptr noalias nocapture %b, i ; SCALABLE-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP10]], [[INDEX]] ; SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP10]] ; SCALABLE-NEXT: [[TMP6:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; SCALABLE-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: br label %[[FOR_END:.*]] ; SCALABLE: [[FOR_END]]: @@ -557,7 +557,7 @@ define void @uniform_store(ptr noalias nocapture %a, ptr noalias nocapture %b, i ; TF-SCALABLE-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP7]], [[INDEX]] ; TF-SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP7]] ; TF-SCALABLE-NEXT: [[TMP8:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; TF-SCALABLE-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: br label %[[FOR_END:.*]] ; TF-SCALABLE: [[FOR_END]]: @@ -608,7 +608,7 @@ define void @uniform_store_of_loop_varying(ptr noalias nocapture %a, ptr noalias ; SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP8]] ; SCALABLE-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[DOTSPLAT]] ; SCALABLE-NEXT: [[TMP9:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; SCALABLE-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: br label %[[FOR_END:.*]] ; SCALABLE: [[FOR_END]]: @@ -679,7 +679,7 @@ define void @uniform_store_of_loop_varying(ptr noalias nocapture %a, ptr noalias ; TF-SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP13]] ; TF-SCALABLE-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT2]] ; TF-SCALABLE-NEXT: [[TMP12:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; TF-SCALABLE-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: br label %[[FOR_END:.*]] ; TF-SCALABLE: [[FOR_END]]: @@ -731,7 +731,7 @@ define void @conditional_uniform_store(ptr noalias nocapture %a, ptr noalias noc ; SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP14]] ; SCALABLE-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; SCALABLE-NEXT: [[TMP11:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; SCALABLE-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP11]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: br label %[[FOR_END:.*]] ; SCALABLE: [[FOR_END]]: @@ -812,7 +812,7 @@ define void @conditional_uniform_store(ptr noalias nocapture %a, ptr noalias noc ; TF-SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP11]] ; TF-SCALABLE-NEXT: [[VEC_IND_NEXT]] = add <vscale x 2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]] ; TF-SCALABLE-NEXT: [[TMP13:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; TF-SCALABLE-NEXT: br i1 [[TMP13]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP13]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: br label %[[FOR_END:.*]] ; TF-SCALABLE: [[FOR_END]]: @@ -860,7 +860,7 @@ define void @uniform_store_unaligned(ptr noalias nocapture %a, ptr noalias nocap ; SCALABLE-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP10]], [[INDEX]] ; SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP10]] ; SCALABLE-NEXT: [[TMP6:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; SCALABLE-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] +; SCALABLE-NEXT: br i1 [[TMP6]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] ; SCALABLE: [[MIDDLE_BLOCK]]: ; SCALABLE-NEXT: br label %[[FOR_END:.*]] ; SCALABLE: [[FOR_END]]: @@ -918,7 +918,7 @@ define void @uniform_store_unaligned(ptr noalias nocapture %a, ptr noalias nocap ; TF-SCALABLE-NEXT: [[INDEX_EVL_NEXT]] = add nuw i64 [[TMP7]], [[INDEX]] ; TF-SCALABLE-NEXT: [[AVL_NEXT]] = sub nuw i64 [[AVL]], [[TMP7]] ; TF-SCALABLE-NEXT: [[TMP8:%.*]] = icmp eq i64 [[AVL_NEXT]], 0 -; TF-SCALABLE-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] +; TF-SCALABLE-NEXT: br i1 [[TMP8]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] ; TF-SCALABLE: [[MIDDLE_BLOCK]]: ; TF-SCALABLE-NEXT: br label %[[FOR_END:.*]] ; TF-SCALABLE: [[FOR_END]]: diff --git a/llvm/test/Transforms/LoopVectorize/X86/consecutive-ptr-uniforms.ll b/llvm/test/Transforms/LoopVectorize/X86/consecutive-ptr-uniforms.ll index e11b1ad..27d5e64 100644 --- a/llvm/test/Transforms/LoopVectorize/X86/consecutive-ptr-uniforms.ll +++ b/llvm/test/Transforms/LoopVectorize/X86/consecutive-ptr-uniforms.ll @@ -166,7 +166,6 @@ attributes #0 = { "target-cpu"="knl" } ; CHECK: LV: Found uniform instruction: {{%.*}} = icmp eq i32 {{%.*}}, 0 ; CHECK-NOT: LV: Found uniform instruction: {{%.*}} = load i32, ptr {{%.*}}, align 1 ; CHECK: LV: Found not uniform due to requiring predication: {{%.*}} = load i32, ptr {{%.*}}, align 1 -; CHECK: LV: Found scalar instruction: {{%.*}} = getelementptr inbounds [3 x i32], ptr @a, i32 0, i32 {{%.*}} ; ; @a = internal constant [3 x i32] [i32 7, i32 7, i32 0], align 1 @@ -215,8 +214,9 @@ define void @PR40816() #1 { ; FORCE-NEXT: [[TMP15:%.*]] = icmp eq i32 [[INDEX_NEXT]], 4 ; FORCE-NEXT: br i1 [[TMP15]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; FORCE: [[MIDDLE_BLOCK]]: -; FORCE-NEXT: br [[RETURN:label %.*]] -; FORCE: [[SCALAR_PH:.*:]] +; FORCE-NEXT: br label %[[RETURN:.*]] +; FORCE: [[RETURN]]: +; FORCE-NEXT: ret void ; entry: br label %for.body diff --git a/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll b/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll index 9453ad7..725fa49 100644 --- a/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll +++ b/llvm/test/Transforms/LoopVectorize/X86/cost-model.ll @@ -540,6 +540,8 @@ define i64 @cost_assume(ptr %end, i64 %N) { ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 8 ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]] ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne i64 [[N:%.*]], 0 +; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] @@ -551,14 +553,6 @@ define i64 @cost_assume(ptr %end, i64 %N) { ; CHECK-NEXT: [[TMP8]] = add <2 x i64> [[VEC_PHI2]], splat (i64 1) ; CHECK-NEXT: [[TMP9]] = add <2 x i64> [[VEC_PHI3]], splat (i64 1) ; CHECK-NEXT: [[TMP10]] = add <2 x i64> [[VEC_PHI4]], splat (i64 1) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[TMP11]]) ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] diff --git a/llvm/test/Transforms/LoopVectorize/X86/vect.omp.force.small-tc.ll b/llvm/test/Transforms/LoopVectorize/X86/vect.omp.force.small-tc.ll index 28de5c7..56f0b85 100644 --- a/llvm/test/Transforms/LoopVectorize/X86/vect.omp.force.small-tc.ll +++ b/llvm/test/Transforms/LoopVectorize/X86/vect.omp.force.small-tc.ll @@ -58,7 +58,7 @@ define void @vectorized(ptr noalias nocapture %A, ptr noalias nocapture readonly ; CHECK: middle.block: ; CHECK-NEXT: br i1 false, label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] ; CHECK: vec.epilog.iter.check: -; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] +; CHECK-NEXT: br i1 false, label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF5:![0-9]+]] ; CHECK: vec.epilog.ph: ; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 16, [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] ; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] @@ -72,7 +72,7 @@ define void @vectorized(ptr noalias nocapture %A, ptr noalias nocapture readonly ; CHECK-NEXT: store <4 x float> [[TMP21]], ptr [[TMP19]], align 4, !llvm.access.group [[ACC_GRP0]] ; CHECK-NEXT: [[INDEX_NEXT11]] = add nuw i64 [[INDEX8]], 4 ; CHECK-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT11]], 20 -; CHECK-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP22]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK: vec.epilog.middle.block: ; CHECK-NEXT: br i1 true, label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]] ; CHECK: vec.epilog.scalar.ph: @@ -88,7 +88,7 @@ define void @vectorized(ptr noalias nocapture %A, ptr noalias nocapture readonly ; CHECK-NEXT: store float [[ADD]], ptr [[ARRAYIDX2]], align 4, !llvm.access.group [[ACC_GRP0]] ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 20 -; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; CHECK: for.end: ; CHECK-NEXT: ret void ; @@ -132,14 +132,14 @@ define void @vectorized1(ptr noalias nocapture %A, ptr noalias nocapture readonl ; CHECK-NEXT: [[VEC_IV:%.*]] = add <8 x i64> [[BROADCAST_SPLAT]], <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7> ; CHECK-NEXT: [[TMP1:%.*]] = icmp ule <8 x i64> [[VEC_IV]], splat (i64 19) ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[INDEX]] -; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <8 x float> @llvm.masked.load.v8f32.p0(ptr [[TMP2]], i32 4, <8 x i1> [[TMP1]], <8 x float> poison), !llvm.access.group [[ACC_GRP7:![0-9]+]] +; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <8 x float> @llvm.masked.load.v8f32.p0(ptr [[TMP2]], i32 4, <8 x i1> [[TMP1]], <8 x float> poison), !llvm.access.group [[ACC_GRP8:![0-9]+]] ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[INDEX]] -; CHECK-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call <8 x float> @llvm.masked.load.v8f32.p0(ptr [[TMP4]], i32 4, <8 x i1> [[TMP1]], <8 x float> poison), !llvm.access.group [[ACC_GRP7]] +; CHECK-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call <8 x float> @llvm.masked.load.v8f32.p0(ptr [[TMP4]], i32 4, <8 x i1> [[TMP1]], <8 x float> poison), !llvm.access.group [[ACC_GRP8]] ; CHECK-NEXT: [[TMP6:%.*]] = fadd fast <8 x float> [[WIDE_MASKED_LOAD]], [[WIDE_MASKED_LOAD1]] -; CHECK-NEXT: call void @llvm.masked.store.v8f32.p0(<8 x float> [[TMP6]], ptr [[TMP4]], i32 4, <8 x i1> [[TMP1]]), !llvm.access.group [[ACC_GRP7]] +; CHECK-NEXT: call void @llvm.masked.store.v8f32.p0(<8 x float> [[TMP6]], ptr [[TMP4]], i32 4, <8 x i1> [[TMP1]]), !llvm.access.group [[ACC_GRP8]] ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 24 -; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[FOR_BODY:%.*]] ; CHECK: for.end: @@ -180,14 +180,14 @@ define void @vectorized2(ptr noalias nocapture %A, ptr noalias nocapture readonl ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, ptr [[B:%.*]], i64 [[INDEX]] -; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x float>, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP7]] +; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <8 x float>, ptr [[TMP1]], align 4, !llvm.access.group [[ACC_GRP8]] ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[INDEX]] -; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <8 x float>, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP7]] +; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <8 x float>, ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP8]] ; CHECK-NEXT: [[TMP5:%.*]] = fadd fast <8 x float> [[WIDE_LOAD]], [[WIDE_LOAD1]] -; CHECK-NEXT: store <8 x float> [[TMP5]], ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP7]] +; CHECK-NEXT: store <8 x float> [[TMP5]], ptr [[TMP3]], align 4, !llvm.access.group [[ACC_GRP8]] ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 16 -; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP9:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP11:![0-9]+]] ; CHECK: middle.block: ; CHECK-NEXT: br label [[FOR_BODY:%.*]] ; CHECK: for.end: diff --git a/llvm/test/Transforms/LoopVectorize/assume.ll b/llvm/test/Transforms/LoopVectorize/assume.ll index 65c12a1..224ec4a6 100644 --- a/llvm/test/Transforms/LoopVectorize/assume.ll +++ b/llvm/test/Transforms/LoopVectorize/assume.ll @@ -34,8 +34,9 @@ define void @test1(ptr noalias nocapture %a, ptr noalias nocapture readonly %b) ; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1600 ; CHECK-NEXT: br i1 [[TMP12]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[FOR_END:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[FOR_END:.*]] +; CHECK: [[FOR_END]]: +; CHECK-NEXT: ret void ; entry: br label %for.body @@ -73,29 +74,28 @@ define void @test2(ptr noalias %a, ptr noalias %b) { ; CHECK-NEXT: [[MASKCOND4:%.*]] = icmp eq i64 [[MASKEDPTR3]], 0 ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] ; CHECK: [[VECTOR_PH]]: +; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] ; CHECK: [[VECTOR_BODY]]: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) ; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]] ; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds float, ptr [[TMP3]], i32 2 ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, ptr [[TMP3]], align 4 ; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x float>, ptr [[TMP4]], align 4 ; CHECK-NEXT: [[TMP5:%.*]] = fadd <2 x float> [[WIDE_LOAD]], splat (float 1.000000e+00) ; CHECK-NEXT: [[TMP6:%.*]] = fadd <2 x float> [[WIDE_LOAD1]], splat (float 1.000000e+00) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) ; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]] ; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds float, ptr [[TMP7]], i32 2 ; CHECK-NEXT: store <2 x float> [[TMP5]], ptr [[TMP7]], align 4 ; CHECK-NEXT: store <2 x float> [[TMP6]], ptr [[TMP8]], align 4 ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 ; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1600 -; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: -; CHECK-NEXT: br [[FOR_END:label %.*]] -; CHECK: [[SCALAR_PH:.*:]] +; CHECK-NEXT: br label %[[FOR_END:.*]] +; CHECK: [[FOR_END]]: +; CHECK-NEXT: ret void ; entry: %ptrint = ptrtoint ptr %a to i64 @@ -163,7 +163,7 @@ define void @predicated_assume(ptr noalias nocapture readonly %a, ptr noalias no ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 ; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[STEP_ADD]], splat (i64 2) ; CHECK-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] -; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; CHECK-NEXT: br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK: [[MIDDLE_BLOCK]]: ; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] ; CHECK-NEXT: br i1 [[CMP_N]], [[FOR_COND_CLEANUP_LOOPEXIT:label %.*]], label %[[SCALAR_PH]] diff --git a/llvm/test/Transforms/LoopVectorize/operand-bundles.ll b/llvm/test/Transforms/LoopVectorize/operand-bundles.ll index 131e41a..ce07364 100644 --- a/llvm/test/Transforms/LoopVectorize/operand-bundles.ll +++ b/llvm/test/Transforms/LoopVectorize/operand-bundles.ll @@ -189,12 +189,12 @@ define void @assume_cold_operand_bundle(ptr noalias %a, ptr noalias %b) { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: br label %[[VECTOR_PH:.*]] ; CHECK: [[VECTOR_PH]]: +; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "cold"() ] ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] ; CHECK: [[VECTOR_BODY]]: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] ; CHECK-NEXT: [[TMP0:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]] ; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP0]], align 4 -; CHECK-NEXT: tail call void @llvm.assume(i1 true) [ "cold"() ] ; CHECK-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[WIDE_LOAD]], splat (float 1.000000e+00) ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]] ; CHECK-NEXT: store <4 x float> [[TMP1]], ptr [[TMP2]], align 4 diff --git a/llvm/test/Transforms/LoopVectorize/scalable-assume.ll b/llvm/test/Transforms/LoopVectorize/scalable-assume.ll index ad8cd42..667df3a 100644 --- a/llvm/test/Transforms/LoopVectorize/scalable-assume.ll +++ b/llvm/test/Transforms/LoopVectorize/scalable-assume.ll @@ -88,11 +88,11 @@ define void @test2(ptr %a, ptr noalias %b) { ; CHECK-NEXT: [[TMP7:%.*]] = mul nuw i64 [[TMP6]], 4 ; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 1600, [[TMP7]] ; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 1600, [[N_MOD_VF]] +; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) +; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) ; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] ; CHECK: [[VECTOR_BODY]]: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND]]) ; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDEX]] ; CHECK-NEXT: [[TMP11:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-NEXT: [[TMP12:%.*]] = shl nuw i64 [[TMP11]], 1 @@ -101,8 +101,6 @@ define void @test2(ptr %a, ptr noalias %b) { ; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 2 x float>, ptr [[TMP13]], align 4 ; CHECK-NEXT: [[TMP14:%.*]] = fadd <vscale x 2 x float> [[WIDE_LOAD]], splat (float 1.000000e+00) ; CHECK-NEXT: [[TMP15:%.*]] = fadd <vscale x 2 x float> [[WIDE_LOAD3]], splat (float 1.000000e+00) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) -; CHECK-NEXT: tail call void @llvm.assume(i1 [[MASKCOND4]]) ; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds float, ptr [[B]], i64 [[INDEX]] ; CHECK-NEXT: [[TMP17:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-NEXT: [[TMP18:%.*]] = shl nuw i64 [[TMP17]], 1 diff --git a/llvm/test/tools/llvm-mca/RISCV/SiFive7/vgather-vcompress.s b/llvm/test/tools/llvm-mca/RISCV/SiFive7/vrgather-vcompress.s index 4ec1683..4ec1683 100644 --- a/llvm/test/tools/llvm-mca/RISCV/SiFive7/vgather-vcompress.s +++ b/llvm/test/tools/llvm-mca/RISCV/SiFive7/vrgather-vcompress.s diff --git a/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vgather-vcompress.s b/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vrgather-vcompress.s index 5ebed10..5ebed10 100644 --- a/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vgather-vcompress.s +++ b/llvm/test/tools/llvm-mca/RISCV/SiFiveX390/vrgather-vcompress.s diff --git a/llvm/unittests/Analysis/DXILResourceTest.cpp b/llvm/unittests/Analysis/DXILResourceTest.cpp index ee37fad..8c3a213 100644 --- a/llvm/unittests/Analysis/DXILResourceTest.cpp +++ b/llvm/unittests/Analysis/DXILResourceTest.cpp @@ -369,10 +369,8 @@ TEST(DXILResource, AnnotationsAndMetadata) { { StructType *CBufStruct = StructType::create(Context, {Floatx4Ty, Floatx4Ty}, "cb0"); - TargetExtType *CBufLayoutType = - llvm::TargetExtType::get(Context, "dx.Layout", CBufStruct, {32, 0, 16}); ResourceTypeInfo RTI( - llvm::TargetExtType::get(Context, "dx.CBuffer", CBufLayoutType)); + llvm::TargetExtType::get(Context, "dx.CBuffer", CBufStruct)); EXPECT_EQ(RTI.getResourceClass(), ResourceClass::CBuffer); EXPECT_EQ(RTI.getCBufferSize(DL), 32u); EXPECT_EQ(RTI.getResourceKind(), ResourceKind::CBuffer); diff --git a/llvm/utils/TableGen/X86DisassemblerTables.cpp b/llvm/utils/TableGen/X86DisassemblerTables.cpp index ed7a4fe..3414190 100644 --- a/llvm/utils/TableGen/X86DisassemblerTables.cpp +++ b/llvm/utils/TableGen/X86DisassemblerTables.cpp @@ -99,6 +99,7 @@ static inline bool inheritsFrom(InstructionContext child, (noPrefix && inheritsFrom(child, IC_XS, noPrefix))); case IC_64BIT: return (inheritsFrom(child, IC_64BIT_REXW) || + inheritsFrom(child, IC_64BIT_REX2) || (noPrefix && inheritsFrom(child, IC_64BIT_OPSIZE, noPrefix)) || (!AdSize64 && inheritsFrom(child, IC_64BIT_ADSIZE)) || (noPrefix && inheritsFrom(child, IC_64BIT_XD, noPrefix)) || @@ -151,8 +152,10 @@ static inline bool inheritsFrom(InstructionContext child, case IC_64BIT_REXW_XS: case IC_64BIT_REXW_OPSIZE: case IC_64BIT_REXW_ADSIZE: - case IC_64BIT_REX2: + case IC_64BIT_REX2_REXW: return false; + case IC_64BIT_REX2: + return inheritsFrom(child, IC_64BIT_REX2_REXW); case IC_VEX: return (VEX_LIG && WIG && inheritsFrom(child, IC_VEX_L_W)) || (WIG && inheritsFrom(child, IC_VEX_W)) || @@ -980,9 +983,11 @@ void DisassemblerTables::emitContextTable(raw_ostream &o, unsigned &i) const { if ((index & ATTR_EVEXB) && (index & ATTR_EVEXU)) o << "_U"; } - } else if ((index & ATTR_64BIT) && (index & ATTR_REX2)) + } else if ((index & ATTR_64BIT) && (index & ATTR_REX2)) { o << "IC_64BIT_REX2"; - else if ((index & ATTR_64BIT) && (index & ATTR_REXW) && (index & ATTR_XS)) + if (index & ATTR_REXW) + o << "_REXW"; + } else if ((index & ATTR_64BIT) && (index & ATTR_REXW) && (index & ATTR_XS)) o << "IC_64BIT_REXW_XS"; else if ((index & ATTR_64BIT) && (index & ATTR_REXW) && (index & ATTR_XD)) o << "IC_64BIT_REXW_XD"; diff --git a/llvm/utils/TableGen/X86RecognizableInstr.cpp b/llvm/utils/TableGen/X86RecognizableInstr.cpp index e87a1c9..a006888 100644 --- a/llvm/utils/TableGen/X86RecognizableInstr.cpp +++ b/llvm/utils/TableGen/X86RecognizableInstr.cpp @@ -365,6 +365,8 @@ InstructionContext RecognizableInstr::insnContext() const { insnContext = IC_64BIT_XD; else if (OpPrefix == X86Local::XS) insnContext = IC_64BIT_XS; + else if (HasREX_W && ExplicitREX2Prefix) + insnContext = IC_64BIT_REX2_REXW; else if (ExplicitREX2Prefix) insnContext = IC_64BIT_REX2; else if (HasREX_W) |