diff options
Diffstat (limited to 'llvm/lib/Target/RISCV')
| -rw-r--r-- | llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp | 49 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.h | 3 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/MCTargetDesc/RISCVFixupKinds.h | 21 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp | 13 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVISelLowering.cpp | 170 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td | 15 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td | 2 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td | 2 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td | 22 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td | 12 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp | 4 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp | 24 | ||||
| -rw-r--r-- | llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp | 17 |
13 files changed, 194 insertions, 160 deletions
diff --git a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp index 82e3b5c..eb7460e 100644 --- a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp +++ b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.cpp @@ -327,19 +327,19 @@ bool RISCVAsmBackend::relaxAlign(MCFragment &F, unsigned &Size) { bool RISCVAsmBackend::relaxDwarfLineAddr(MCFragment &F, bool &WasRelaxed) const { - MCContext &C = getContext(); - int64_t LineDelta = F.getDwarfLineDelta(); const MCExpr &AddrDelta = F.getDwarfAddrDelta(); - SmallVector<MCFixup, 1> Fixups; size_t OldSize = F.getVarSize(); int64_t Value; + // If the label difference can be resolved, use the default handling, which + // utilizes a shorter special opcode. + if (AddrDelta.evaluateAsAbsolute(Value, *Asm)) + return false; [[maybe_unused]] bool IsAbsolute = AddrDelta.evaluateKnownAbsolute(Value, *Asm); assert(IsAbsolute && "CFA with invalid expression"); - Fixups.clear(); SmallVector<char> Data; raw_svector_ostream OS(Data); @@ -349,33 +349,21 @@ bool RISCVAsmBackend::relaxDwarfLineAddr(MCFragment &F, encodeSLEB128(LineDelta, OS); } - unsigned Offset; - std::pair<MCFixupKind, MCFixupKind> Fixup; - // According to the DWARF specification, the `DW_LNS_fixed_advance_pc` opcode // takes a single unsigned half (unencoded) operand. The maximum encodable // value is therefore 65535. Set a conservative upper bound for relaxation. + unsigned PCBytes; if (Value > 60000) { - unsigned PtrSize = C.getAsmInfo()->getCodePointerSize(); - - OS << uint8_t(dwarf::DW_LNS_extended_op); - encodeULEB128(PtrSize + 1, OS); - - OS << uint8_t(dwarf::DW_LNE_set_address); - Offset = OS.tell(); - assert((PtrSize == 4 || PtrSize == 8) && "Unexpected pointer size"); - Fixup = RISCV::getRelocPairForSize(PtrSize); - OS.write_zeros(PtrSize); + PCBytes = getContext().getAsmInfo()->getCodePointerSize(); + OS << uint8_t(dwarf::DW_LNS_extended_op) << uint8_t(PCBytes + 1) + << uint8_t(dwarf::DW_LNE_set_address); + OS.write_zeros(PCBytes); } else { + PCBytes = 2; OS << uint8_t(dwarf::DW_LNS_fixed_advance_pc); - Offset = OS.tell(); - Fixup = RISCV::getRelocPairForSize(2); support::endian::write<uint16_t>(OS, 0, llvm::endianness::little); } - - const MCBinaryExpr &MBE = cast<MCBinaryExpr>(AddrDelta); - Fixups.push_back(MCFixup::create(Offset, MBE.getLHS(), std::get<0>(Fixup))); - Fixups.push_back(MCFixup::create(Offset, MBE.getRHS(), std::get<1>(Fixup))); + auto Offset = OS.tell() - PCBytes; if (LineDelta == INT64_MAX) { OS << uint8_t(dwarf::DW_LNS_extended_op); @@ -386,7 +374,8 @@ bool RISCVAsmBackend::relaxDwarfLineAddr(MCFragment &F, } F.setVarContents(Data); - F.setVarFixups(Fixups); + F.setVarFixups({MCFixup::create(Offset, &AddrDelta, + MCFixup::getDataKindForSize(PCBytes))}); WasRelaxed = OldSize != Data.size(); return true; } @@ -881,9 +870,8 @@ bool RISCVAsmBackend::addReloc(const MCFragment &F, const MCFixup &Fixup, } void RISCVAsmBackend::applyFixup(const MCFragment &F, const MCFixup &Fixup, - const MCValue &Target, - MutableArrayRef<char> Data, uint64_t Value, - bool IsResolved) { + const MCValue &Target, uint8_t *Data, + uint64_t Value, bool IsResolved) { IsResolved = addReloc(F, Fixup, Target, Value, IsResolved); MCFixupKind Kind = Fixup.getKind(); if (mc::isRelocation(Kind)) @@ -898,15 +886,14 @@ void RISCVAsmBackend::applyFixup(const MCFragment &F, const MCFixup &Fixup, // Shift the value into position. Value <<= Info.TargetOffset; - unsigned Offset = Fixup.getOffset(); unsigned NumBytes = alignTo(Info.TargetSize + Info.TargetOffset, 8) / 8; - - assert(Offset + NumBytes <= Data.size() && "Invalid fixup offset!"); + assert(Fixup.getOffset() + NumBytes <= F.getSize() && + "Invalid fixup offset!"); // For each byte of the fragment that the fixup touches, mask in the // bits from the fixup value. for (unsigned i = 0; i != NumBytes; ++i) { - Data[Offset + i] |= uint8_t((Value >> (i * 8)) & 0xff); + Data[i] |= uint8_t((Value >> (i * 8)) & 0xff); } } diff --git a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.h b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.h index d97d632..adec1ec 100644 --- a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.h +++ b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVAsmBackend.h @@ -46,8 +46,7 @@ public: void maybeAddVendorReloc(const MCFragment &, const MCFixup &); void applyFixup(const MCFragment &, const MCFixup &, const MCValue &Target, - MutableArrayRef<char> Data, uint64_t Value, - bool IsResolved) override; + uint8_t *Data, uint64_t Value, bool IsResolved) override; std::unique_ptr<MCObjectTargetWriter> createObjectTargetWriter() const override; diff --git a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVFixupKinds.h b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVFixupKinds.h index f816561c..98c8738 100644 --- a/llvm/lib/Target/RISCV/MCTargetDesc/RISCVFixupKinds.h +++ b/llvm/lib/Target/RISCV/MCTargetDesc/RISCVFixupKinds.h @@ -68,27 +68,6 @@ enum Fixups { fixup_riscv_invalid, NumTargetFixupKinds = fixup_riscv_invalid - FirstTargetFixupKind }; - -static inline std::pair<MCFixupKind, MCFixupKind> -getRelocPairForSize(unsigned Size) { - switch (Size) { - default: - llvm_unreachable("unsupported fixup size"); - case 1: - return std::make_pair(FirstLiteralRelocationKind + ELF::R_RISCV_ADD8, - FirstLiteralRelocationKind + ELF::R_RISCV_SUB8); - case 2: - return std::make_pair(FirstLiteralRelocationKind + ELF::R_RISCV_ADD16, - FirstLiteralRelocationKind + ELF::R_RISCV_SUB16); - case 4: - return std::make_pair(FirstLiteralRelocationKind + ELF::R_RISCV_ADD32, - FirstLiteralRelocationKind + ELF::R_RISCV_SUB32); - case 8: - return std::make_pair(FirstLiteralRelocationKind + ELF::R_RISCV_ADD64, - FirstLiteralRelocationKind + ELF::R_RISCV_SUB64); - } -} - } // end namespace llvm::RISCV #endif diff --git a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp index 7b0fe5f..5998653 100644 --- a/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp +++ b/llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp @@ -2827,6 +2827,8 @@ static bool selectConstantAddr(SelectionDAG *CurDAG, const SDLoc &DL, static bool isWorthFoldingAdd(SDValue Add) { for (auto *User : Add->users()) { if (User->getOpcode() != ISD::LOAD && User->getOpcode() != ISD::STORE && + User->getOpcode() != RISCVISD::LD_RV32 && + User->getOpcode() != RISCVISD::SD_RV32 && User->getOpcode() != ISD::ATOMIC_LOAD && User->getOpcode() != ISD::ATOMIC_STORE) return false; @@ -2841,6 +2843,9 @@ static bool isWorthFoldingAdd(SDValue Add) { if (User->getOpcode() == ISD::ATOMIC_STORE && cast<AtomicSDNode>(User)->getVal() == Add) return false; + if (User->getOpcode() == RISCVISD::SD_RV32 && + (User->getOperand(0) == Add || User->getOperand(1) == Add)) + return false; if (isStrongerThanMonotonic(cast<MemSDNode>(User)->getSuccessOrdering())) return false; } @@ -2942,8 +2947,8 @@ bool RISCVDAGToDAGISel::SelectAddrRegImm(SDValue Addr, SDValue &Base, /// Similar to SelectAddrRegImm, except that the offset is restricted to uimm9. bool RISCVDAGToDAGISel::SelectAddrRegImm9(SDValue Addr, SDValue &Base, SDValue &Offset) { - // FIXME: Support FrameIndex. Need to teach eliminateFrameIndex that only - // a 9-bit immediate can be folded. + if (SelectAddrFrameIndex(Addr, Base, Offset)) + return true; SDLoc DL(Addr); MVT VT = Addr.getSimpleValueType(); @@ -2953,8 +2958,8 @@ bool RISCVDAGToDAGISel::SelectAddrRegImm9(SDValue Addr, SDValue &Base, if (isUInt<9>(CVal)) { Base = Addr.getOperand(0); - // FIXME: Support FrameIndex. Need to teach eliminateFrameIndex that only - // a 9-bit immediate can be folded. + if (auto *FIN = dyn_cast<FrameIndexSDNode>(Base)) + Base = CurDAG->getTargetFrameIndex(FIN->getIndex(), VT); Offset = CurDAG->getSignedTargetConstant(CVal, DL, VT); return true; } diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp index b47d89b..adbfbeb 100644 --- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp +++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp @@ -1819,6 +1819,13 @@ bool RISCVTargetLowering::getTgtMemIntrinsic(IntrinsicInfo &Info, case Intrinsic::riscv_seg6_load_mask: case Intrinsic::riscv_seg7_load_mask: case Intrinsic::riscv_seg8_load_mask: + case Intrinsic::riscv_sseg2_load_mask: + case Intrinsic::riscv_sseg3_load_mask: + case Intrinsic::riscv_sseg4_load_mask: + case Intrinsic::riscv_sseg5_load_mask: + case Intrinsic::riscv_sseg6_load_mask: + case Intrinsic::riscv_sseg7_load_mask: + case Intrinsic::riscv_sseg8_load_mask: return SetRVVLoadStoreInfo(/*PtrOp*/ 0, /*IsStore*/ false, /*IsUnitStrided*/ false, /*UsePtrVal*/ true); case Intrinsic::riscv_seg2_store_mask: @@ -10938,6 +10945,97 @@ static inline SDValue getVCIXISDNodeVOID(SDValue &Op, SelectionDAG &DAG, return DAG.getNode(Type, SDLoc(Op), Op.getValueType(), Operands); } +static SDValue +lowerFixedVectorSegLoadIntrinsics(unsigned IntNo, SDValue Op, + const RISCVSubtarget &Subtarget, + SelectionDAG &DAG) { + bool IsStrided; + switch (IntNo) { + case Intrinsic::riscv_seg2_load_mask: + case Intrinsic::riscv_seg3_load_mask: + case Intrinsic::riscv_seg4_load_mask: + case Intrinsic::riscv_seg5_load_mask: + case Intrinsic::riscv_seg6_load_mask: + case Intrinsic::riscv_seg7_load_mask: + case Intrinsic::riscv_seg8_load_mask: + IsStrided = false; + break; + case Intrinsic::riscv_sseg2_load_mask: + case Intrinsic::riscv_sseg3_load_mask: + case Intrinsic::riscv_sseg4_load_mask: + case Intrinsic::riscv_sseg5_load_mask: + case Intrinsic::riscv_sseg6_load_mask: + case Intrinsic::riscv_sseg7_load_mask: + case Intrinsic::riscv_sseg8_load_mask: + IsStrided = true; + break; + default: + llvm_unreachable("unexpected intrinsic ID"); + }; + + static const Intrinsic::ID VlsegInts[7] = { + Intrinsic::riscv_vlseg2_mask, Intrinsic::riscv_vlseg3_mask, + Intrinsic::riscv_vlseg4_mask, Intrinsic::riscv_vlseg5_mask, + Intrinsic::riscv_vlseg6_mask, Intrinsic::riscv_vlseg7_mask, + Intrinsic::riscv_vlseg8_mask}; + static const Intrinsic::ID VlssegInts[7] = { + Intrinsic::riscv_vlsseg2_mask, Intrinsic::riscv_vlsseg3_mask, + Intrinsic::riscv_vlsseg4_mask, Intrinsic::riscv_vlsseg5_mask, + Intrinsic::riscv_vlsseg6_mask, Intrinsic::riscv_vlsseg7_mask, + Intrinsic::riscv_vlsseg8_mask}; + + SDLoc DL(Op); + unsigned NF = Op->getNumValues() - 1; + assert(NF >= 2 && NF <= 8 && "Unexpected seg number"); + MVT XLenVT = Subtarget.getXLenVT(); + MVT VT = Op->getSimpleValueType(0); + MVT ContainerVT = ::getContainerForFixedLengthVector(DAG, VT, Subtarget); + unsigned Sz = NF * ContainerVT.getVectorMinNumElements() * + ContainerVT.getScalarSizeInBits(); + EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, NF); + + // Operands: (chain, int_id, pointer, mask, vl) or + // (chain, int_id, pointer, offset, mask, vl) + SDValue VL = Op.getOperand(Op.getNumOperands() - 1); + SDValue Mask = Op.getOperand(Op.getNumOperands() - 2); + MVT MaskVT = Mask.getSimpleValueType(); + MVT MaskContainerVT = + ::getContainerForFixedLengthVector(DAG, MaskVT, Subtarget); + Mask = convertToScalableVector(MaskContainerVT, Mask, DAG, Subtarget); + + SDValue IntID = DAG.getTargetConstant( + IsStrided ? VlssegInts[NF - 2] : VlsegInts[NF - 2], DL, XLenVT); + auto *Load = cast<MemIntrinsicSDNode>(Op); + + SDVTList VTs = DAG.getVTList({VecTupTy, MVT::Other}); + SmallVector<SDValue, 9> Ops = { + Load->getChain(), + IntID, + DAG.getUNDEF(VecTupTy), + Op.getOperand(2), + Mask, + VL, + DAG.getTargetConstant( + RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC, DL, XLenVT), + DAG.getTargetConstant(Log2_64(VT.getScalarSizeInBits()), DL, XLenVT)}; + // Insert the stride operand. + if (IsStrided) + Ops.insert(std::next(Ops.begin(), 4), Op.getOperand(3)); + + SDValue Result = + DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, + Load->getMemoryVT(), Load->getMemOperand()); + SmallVector<SDValue, 9> Results; + for (unsigned int RetIdx = 0; RetIdx < NF; RetIdx++) { + SDValue SubVec = DAG.getNode(RISCVISD::TUPLE_EXTRACT, DL, ContainerVT, + Result.getValue(0), + DAG.getTargetConstant(RetIdx, DL, MVT::i32)); + Results.push_back(convertFromScalableVector(VT, SubVec, DAG, Subtarget)); + } + Results.push_back(Result.getValue(1)); + return DAG.getMergeValues(Results, DL); +} + SDValue RISCVTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG) const { unsigned IntNo = Op.getConstantOperandVal(1); @@ -10950,57 +11048,16 @@ SDValue RISCVTargetLowering::LowerINTRINSIC_W_CHAIN(SDValue Op, case Intrinsic::riscv_seg5_load_mask: case Intrinsic::riscv_seg6_load_mask: case Intrinsic::riscv_seg7_load_mask: - case Intrinsic::riscv_seg8_load_mask: { - SDLoc DL(Op); - static const Intrinsic::ID VlsegInts[7] = { - Intrinsic::riscv_vlseg2_mask, Intrinsic::riscv_vlseg3_mask, - Intrinsic::riscv_vlseg4_mask, Intrinsic::riscv_vlseg5_mask, - Intrinsic::riscv_vlseg6_mask, Intrinsic::riscv_vlseg7_mask, - Intrinsic::riscv_vlseg8_mask}; - unsigned NF = Op->getNumValues() - 1; - assert(NF >= 2 && NF <= 8 && "Unexpected seg number"); - MVT XLenVT = Subtarget.getXLenVT(); - MVT VT = Op->getSimpleValueType(0); - MVT ContainerVT = getContainerForFixedLengthVector(VT); - unsigned Sz = NF * ContainerVT.getVectorMinNumElements() * - ContainerVT.getScalarSizeInBits(); - EVT VecTupTy = MVT::getRISCVVectorTupleVT(Sz, NF); - - // Operands: (chain, int_id, pointer, mask, vl) - SDValue VL = Op.getOperand(Op.getNumOperands() - 1); - SDValue Mask = Op.getOperand(3); - MVT MaskVT = Mask.getSimpleValueType(); - MVT MaskContainerVT = - ::getContainerForFixedLengthVector(DAG, MaskVT, Subtarget); - Mask = convertToScalableVector(MaskContainerVT, Mask, DAG, Subtarget); - - SDValue IntID = DAG.getTargetConstant(VlsegInts[NF - 2], DL, XLenVT); - auto *Load = cast<MemIntrinsicSDNode>(Op); + case Intrinsic::riscv_seg8_load_mask: + case Intrinsic::riscv_sseg2_load_mask: + case Intrinsic::riscv_sseg3_load_mask: + case Intrinsic::riscv_sseg4_load_mask: + case Intrinsic::riscv_sseg5_load_mask: + case Intrinsic::riscv_sseg6_load_mask: + case Intrinsic::riscv_sseg7_load_mask: + case Intrinsic::riscv_sseg8_load_mask: + return lowerFixedVectorSegLoadIntrinsics(IntNo, Op, Subtarget, DAG); - SDVTList VTs = DAG.getVTList({VecTupTy, MVT::Other}); - SDValue Ops[] = { - Load->getChain(), - IntID, - DAG.getUNDEF(VecTupTy), - Op.getOperand(2), - Mask, - VL, - DAG.getTargetConstant( - RISCVVType::TAIL_AGNOSTIC | RISCVVType::MASK_AGNOSTIC, DL, XLenVT), - DAG.getTargetConstant(Log2_64(VT.getScalarSizeInBits()), DL, XLenVT)}; - SDValue Result = - DAG.getMemIntrinsicNode(ISD::INTRINSIC_W_CHAIN, DL, VTs, Ops, - Load->getMemoryVT(), Load->getMemOperand()); - SmallVector<SDValue, 9> Results; - for (unsigned int RetIdx = 0; RetIdx < NF; RetIdx++) { - SDValue SubVec = DAG.getNode(RISCVISD::TUPLE_EXTRACT, DL, ContainerVT, - Result.getValue(0), - DAG.getTargetConstant(RetIdx, DL, MVT::i32)); - Results.push_back(convertFromScalableVector(VT, SubVec, DAG, Subtarget)); - } - Results.push_back(Result.getValue(1)); - return DAG.getMergeValues(Results, DL); - } case Intrinsic::riscv_sf_vc_v_x_se: return getVCIXISDNodeWCHAIN(Op, DAG, RISCVISD::SF_VC_V_X_SE); case Intrinsic::riscv_sf_vc_v_i_se: @@ -22725,8 +22782,14 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI, bool IsVarArg = CLI.IsVarArg; EVT PtrVT = getPointerTy(DAG.getDataLayout()); MVT XLenVT = Subtarget.getXLenVT(); + const CallBase *CB = CLI.CB; MachineFunction &MF = DAG.getMachineFunction(); + MachineFunction::CallSiteInfo CSInfo; + + // Set type id for call site info. + if (MF.getTarget().Options.EmitCallGraphSection && CB && CB->isIndirectCall()) + CSInfo = MachineFunction::CallSiteInfo(*CB); // Analyze the operands of the call, assigning locations to each operand. SmallVector<CCValAssign, 16> ArgLocs; @@ -22984,6 +23047,9 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI, if (CLI.CFIType) Ret.getNode()->setCFIType(CLI.CFIType->getZExtValue()); DAG.addNoMergeSiteInfo(Ret.getNode(), CLI.NoMerge); + if (MF.getTarget().Options.EmitCallGraphSection && CB && + CB->isIndirectCall()) + DAG.addCallSiteInfo(Ret.getNode(), std::move(CSInfo)); return Ret; } @@ -22991,6 +23057,10 @@ SDValue RISCVTargetLowering::LowerCall(CallLoweringInfo &CLI, Chain = DAG.getNode(CallOpc, DL, NodeTys, Ops); if (CLI.CFIType) Chain.getNode()->setCFIType(CLI.CFIType->getZExtValue()); + + if (MF.getTarget().Options.EmitCallGraphSection && CB && CB->isIndirectCall()) + DAG.addCallSiteInfo(Chain.getNode(), std::move(CSInfo)); + DAG.addNoMergeSiteInfo(Chain.getNode(), CLI.NoMerge); Glue = Chain.getValue(1); diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td index 6afc942d..03e6f43 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVPseudos.td @@ -1510,21 +1510,6 @@ class VPseudoTiedBinaryCarryIn<VReg RetClass, let VLMul = MInfo.value; } -class VPseudoTernaryNoMask<VReg RetClass, - RegisterClass Op1Class, - DAGOperand Op2Class, - string Constraint> : - RISCVVPseudo<(outs RetClass:$rd), - (ins RetClass:$rs3, Op1Class:$rs1, Op2Class:$rs2, - AVL:$vl, sew:$sew)> { - let mayLoad = 0; - let mayStore = 0; - let hasSideEffects = 0; - let Constraints = !interleave([Constraint, "$rd = $rs3"], ","); - let HasVLOp = 1; - let HasSEWOp = 1; -} - class VPseudoTernaryNoMaskWithPolicy<VReg RetClass, RegisterClass Op1Class, DAGOperand Op2Class, diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td index 31ea2de..cc2977c 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVSDPatterns.td @@ -910,7 +910,7 @@ foreach vti = AllIntegerVectors in { foreach vti = I64IntegerVectors in { let Predicates = [HasVInstructionsI64] in { def : Pat<(add (vti.Vector vti.RegClass:$rs1), - (vti.Vector (SplatPat_imm64_neg i64:$rs2))), + (vti.Vector (SplatPat_imm64_neg (i64 GPR:$rs2)))), (!cast<Instruction>("PseudoVSUB_VX_"#vti.LMul.MX) (vti.Vector (IMPLICIT_DEF)), vti.RegClass:$rs1, diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td index 695223b..acbccdd 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoVVLPatterns.td @@ -2123,7 +2123,7 @@ foreach vti = AllIntegerVectors in { foreach vti = I64IntegerVectors in { let Predicates = [HasVInstructionsI64] in { def : Pat<(riscv_add_vl (vti.Vector vti.RegClass:$rs1), - (vti.Vector (SplatPat_imm64_neg i64:$rs2)), + (vti.Vector (SplatPat_imm64_neg (i64 GPR:$rs2))), vti.RegClass:$passthru, (vti.Mask VMV0:$vm), VLOpFrag), (!cast<Instruction>("PseudoVSUB_VX_"#vti.LMul.MX#"_MASK") vti.RegClass:$passthru, vti.RegClass:$rs1, diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td b/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td index c0f7ab1..4c31ce4 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoXTHead.td @@ -590,12 +590,12 @@ let Predicates = [HasVendorXTHeadBb, IsRV64] in { def : PatGprImm<riscv_rorw, TH_SRRIW, uimm5>; def : Pat<(riscv_rolw GPR:$rs1, uimm5:$rs2), (TH_SRRIW GPR:$rs1, (ImmSubFrom32 uimm5:$rs2))>; -def : Pat<(sra (bswap i64:$rs1), (i64 32)), - (TH_REVW i64:$rs1)>; -def : Pat<(binop_allwusers<srl> (bswap i64:$rs1), (i64 32)), - (TH_REVW i64:$rs1)>; -def : Pat<(riscv_clzw i64:$rs1), - (TH_FF0 (i64 (SLLI (i64 (XORI i64:$rs1, -1)), 32)))>; +def : Pat<(i64 (sra (bswap GPR:$rs1), (i64 32))), + (TH_REVW GPR:$rs1)>; +def : Pat<(binop_allwusers<srl> (bswap GPR:$rs1), (i64 32)), + (TH_REVW GPR:$rs1)>; +def : Pat<(riscv_clzw GPR:$rs1), + (TH_FF0 (i64 (SLLI (i64 (XORI GPR:$rs1, -1)), 32)))>; } // Predicates = [HasVendorXTHeadBb, IsRV64] let Predicates = [HasVendorXTHeadBs] in { @@ -697,11 +697,13 @@ def uimm2_4 : Operand<XLenVT>, ImmLeaf<XLenVT, [{ }], uimm2_4_XFORM>; let Predicates = [HasVendorXTHeadMemPair, IsRV64] in { -def : Pat<(th_lwud i64:$rs1, uimm2_3:$uimm2_3), (TH_LWUD i64:$rs1, uimm2_3:$uimm2_3, 3)>; -def : Pat<(th_ldd i64:$rs1, uimm2_4:$uimm2_4), (TH_LDD i64:$rs1, uimm2_4:$uimm2_4, 4)>; +def : Pat<(th_lwud GPR:$rs1, (i64 uimm2_3:$uimm2_3)), + (TH_LWUD GPR:$rs1, uimm2_3:$uimm2_3, 3)>; +def : Pat<(th_ldd GPR:$rs1, (i64 uimm2_4:$uimm2_4)), + (TH_LDD GPR:$rs1, uimm2_4:$uimm2_4, 4)>; -def : Pat<(th_sdd i64:$rd1, i64:$rd2, i64:$rs1, uimm2_4:$uimm2_4), - (TH_SDD i64:$rd1, i64:$rd2, i64:$rs1, uimm2_4:$uimm2_4, 4)>; +def : Pat<(th_sdd (i64 GPR:$rd1), GPR:$rd2, GPR:$rs1, uimm2_4:$uimm2_4), + (TH_SDD GPR:$rd1, GPR:$rd2, GPR:$rs1, uimm2_4:$uimm2_4, 4)>; } let Predicates = [HasVendorXTHeadMemPair] in { diff --git a/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td b/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td index a250ac8..5a5a9ed 100644 --- a/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td +++ b/llvm/lib/Target/RISCV/RISCVInstrInfoZvk.td @@ -206,8 +206,6 @@ let Predicates = [HasStdExtZvksh], RVVConstraint = VS2Constraint in { //===----------------------------------------------------------------------===// defvar I32IntegerVectors = !filter(vti, AllIntegerVectors, !eq(vti.SEW, 32)); -defvar I32I64IntegerVectors = !filter(vti, AllIntegerVectors, - !or(!eq(vti.SEW, 32), !eq(vti.SEW, 64))); class ZvkI32IntegerVectors<string vd_lmul> { list<VTypeInfo> vs2_types = !cond(!eq(vd_lmul, "M8") : !filter(vti, I32IntegerVectors, !le(vti.LMul.octuple, 32)), @@ -1126,16 +1124,16 @@ let Predicates = [HasStdExtZvkned] in { defm : VPatUnaryV_S_NoMaskVectorCrypto<"int_riscv_vaesz", "PseudoVAESZ", I32IntegerVectors>; } // Predicates = [HasStdExtZvkned] -let Predicates = [HasStdExtZvknha] in { +let Predicates = [HasStdExtZvknhaOrZvknhb] in { defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ch", "PseudoVSHA2CH", I32IntegerVectors>; - defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2cl", "PseudoVSHA2CH", I32IntegerVectors>; + defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2cl", "PseudoVSHA2CL", I32IntegerVectors>; defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I32IntegerVectors, isSEWAware=true>; } // Predicates = [HasStdExtZvknha] let Predicates = [HasStdExtZvknhb] in { - defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ch", "PseudoVSHA2CH", I32I64IntegerVectors>; - defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2cl", "PseudoVSHA2CH", I32I64IntegerVectors>; - defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I32I64IntegerVectors, isSEWAware=true>; + defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ch", "PseudoVSHA2CH", I64IntegerVectors>; + defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2cl", "PseudoVSHA2CL", I64IntegerVectors>; + defm : VPatBinaryV_VV_NoMask<"int_riscv_vsha2ms", "PseudoVSHA2MS", I64IntegerVectors, isSEWAware=true>; } // Predicates = [HasStdExtZvknhb] let Predicates = [HasStdExtZvksed] in { diff --git a/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp b/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp index 214536d..7e58b6f 100644 --- a/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp +++ b/llvm/lib/Target/RISCV/RISCVRegisterInfo.cpp @@ -576,6 +576,7 @@ bool RISCVRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, int64_t Val = Offset.getFixed(); int64_t Lo12 = SignExtend64<12>(Val); unsigned Opc = MI.getOpcode(); + if (Opc == RISCV::ADDI && !isInt<12>(Val)) { // We chose to emit the canonical immediate sequence rather than folding // the offset into the using add under the theory that doing so doesn't @@ -588,6 +589,9 @@ bool RISCVRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II, (Lo12 & 0b11111) != 0) { // Prefetch instructions require the offset to be 32 byte aligned. MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0); + } else if (Opc == RISCV::MIPS_PREFETCH && !isUInt<9>(Val)) { + // MIPS Prefetch instructions require the offset to be 9 bits encoded. + MI.getOperand(FIOperandNum + 1).ChangeToImmediate(0); } else if ((Opc == RISCV::PseudoRV32ZdinxLD || Opc == RISCV::PseudoRV32ZdinxSD) && Lo12 >= 2044) { diff --git a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp index 4eef8de..0d5eb86 100644 --- a/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp +++ b/llvm/lib/Target/RISCV/RISCVTargetTransformInfo.cpp @@ -1280,8 +1280,13 @@ RISCVTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA, } else { Ops = {RISCV::VFCVT_X_F_V}; } - return std::max(SrcLT.first, LT.first) * - getRISCVInstructionCost(Ops, LT.second, CostKind); + + // We need to use the source LMUL in the case of a narrowing op, and the + // destination LMUL otherwise. + if (SrcEltSz > DstEltSz) + return SrcLT.first * + getRISCVInstructionCost(Ops, SrcLT.second, CostKind); + return LT.first * getRISCVInstructionCost(Ops, LT.second, CostKind); } break; } @@ -2622,18 +2627,17 @@ void RISCVTTIImpl::getUnrollingPreferences( if (L->getNumBlocks() > 4) return; - // Don't unroll vectorized loops, including the remainder loop - if (getBooleanLoopAttribute(L, "llvm.loop.isvectorized")) - return; - // Scan the loop: don't unroll loops with calls as this could prevent - // inlining. + // inlining. Don't unroll auto-vectorized loops either, though do allow + // unrolling of the scalar remainder. + bool IsVectorized = getBooleanLoopAttribute(L, "llvm.loop.isvectorized"); InstructionCost Cost = 0; for (auto *BB : L->getBlocks()) { for (auto &I : *BB) { - // Initial setting - Don't unroll loops containing vectorized - // instructions. - if (I.getType()->isVectorTy()) + // Both auto-vectorized loops and the scalar remainder have the + // isvectorized attribute, so differentiate between them by the presence + // of vector instructions. + if (IsVectorized && I.getType()->isVectorTy()) return; if (isa<CallInst>(I) || isa<InvokeInst>(I)) { diff --git a/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp b/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp index c1cc19b..050de3d 100644 --- a/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp +++ b/llvm/lib/Target/RISCV/RISCVVectorPeephole.cpp @@ -646,8 +646,7 @@ bool RISCVVectorPeephole::foldVMV_V_V(MachineInstr &MI) { if (!Src || Src->hasUnmodeledSideEffects() || Src->getParent() != MI.getParent() || !RISCVII::isFirstDefTiedToFirstUse(Src->getDesc()) || - !RISCVII::hasVLOp(Src->getDesc().TSFlags) || - !RISCVII::hasVecPolicyOp(Src->getDesc().TSFlags)) + !RISCVII::hasVLOp(Src->getDesc().TSFlags)) return false; // Src's dest needs to have the same EEW as MI's input. @@ -681,12 +680,14 @@ bool RISCVVectorPeephole::foldVMV_V_V(MachineInstr &MI) { *Src->getParent()->getParent())); } - // If MI was tail agnostic and the VL didn't increase, preserve it. - int64_t Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED; - if ((MI.getOperand(5).getImm() & RISCVVType::TAIL_AGNOSTIC) && - RISCV::isVLKnownLE(MI.getOperand(3), SrcVL)) - Policy |= RISCVVType::TAIL_AGNOSTIC; - Src->getOperand(RISCVII::getVecPolicyOpNum(Src->getDesc())).setImm(Policy); + if (RISCVII::hasVecPolicyOp(Src->getDesc().TSFlags)) { + // If MI was tail agnostic and the VL didn't increase, preserve it. + int64_t Policy = RISCVVType::TAIL_UNDISTURBED_MASK_UNDISTURBED; + if ((MI.getOperand(5).getImm() & RISCVVType::TAIL_AGNOSTIC) && + RISCV::isVLKnownLE(MI.getOperand(3), SrcVL)) + Policy |= RISCVVType::TAIL_AGNOSTIC; + Src->getOperand(RISCVII::getVecPolicyOpNum(Src->getDesc())).setImm(Policy); + } MRI->constrainRegClass(Src->getOperand(0).getReg(), MRI->getRegClass(MI.getOperand(0).getReg())); |