-//===-- PPCTargetTransformInfo.cpp - PPC specific TTI pass ----------------===//
+//===-- PPCTargetTransformInfo.cpp - PPC specific TTI ---------------------===//
//
// The LLVM Compiler Infrastructure
//
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
-/// \file
-/// This file implements a TargetTransformInfo analysis pass specific to the
-/// PPC target machine. It uses the target's detailed information to provide
-/// more precise answers to certain TTI queries, while letting the target
-/// independent and default TTI implementations handle the rest.
-///
-//===----------------------------------------------------------------------===//
-#include "PPC.h"
-#include "PPCTargetMachine.h"
+#include "PPCTargetTransformInfo.h"
#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/CodeGen/BasicTTIImpl.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Target/CostTable.h"
static cl::opt<bool> DisablePPCConstHoist("disable-ppc-constant-hoisting",
cl::desc("disable constant hoisting on PPC"), cl::init(false), cl::Hidden);
-// Declare the pass initialization routine locally as target-specific passes
-// don't have a target-wide initialization entry point, and so we rely on the
-// pass constructor initialization.
-namespace llvm {
-void initializePPCTTIPass(PassRegistry &);
-}
-
-namespace {
-
-class PPCTTI final : public ImmutablePass, public TargetTransformInfo {
- const PPCSubtarget *ST;
- const PPCTargetLowering *TLI;
-
-public:
- PPCTTI() : ImmutablePass(ID), ST(nullptr), TLI(nullptr) {
- llvm_unreachable("This pass cannot be directly constructed");
- }
-
- PPCTTI(const PPCTargetMachine *TM)
- : ImmutablePass(ID), ST(TM->getSubtargetImpl()),
- TLI(TM->getSubtargetImpl()->getTargetLowering()) {
- initializePPCTTIPass(*PassRegistry::getPassRegistry());
- }
-
- virtual void initializePass() override {
- pushTTIStack(this);
- }
-
- virtual void getAnalysisUsage(AnalysisUsage &AU) const override {
- TargetTransformInfo::getAnalysisUsage(AU);
- }
-
- /// Pass identification.
- static char ID;
-
- /// Provide necessary pointer adjustments for the two base classes.
- virtual void *getAdjustedAnalysisPointer(const void *ID) override {
- if (ID == &TargetTransformInfo::ID)
- return (TargetTransformInfo*)this;
- return this;
- }
-
- /// \name Scalar TTI Implementations
- /// @{
- unsigned getIntImmCost(const APInt &Imm, Type *Ty) const override;
-
- unsigned getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
- Type *Ty) const override;
- unsigned getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
- Type *Ty) const override;
-
- virtual PopcntSupportKind
- getPopcntSupport(unsigned TyWidth) const override;
- virtual void getUnrollingPreferences(
- Loop *L, UnrollingPreferences &UP) const override;
-
- /// @}
-
- /// \name Vector TTI Implementations
- /// @{
-
- virtual unsigned getNumberOfRegisters(bool Vector) const override;
- virtual unsigned getRegisterBitWidth(bool Vector) const override;
- virtual unsigned getMaximumUnrollFactor() const override;
- virtual unsigned
- getArithmeticInstrCost(unsigned Opcode, Type *Ty, OperandValueKind,
- OperandValueKind, OperandValueProperties,
- OperandValueProperties) const override;
- virtual unsigned getShuffleCost(ShuffleKind Kind, Type *Tp,
- int Index, Type *SubTp) const override;
- virtual unsigned getCastInstrCost(unsigned Opcode, Type *Dst,
- Type *Src) const override;
- virtual unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
- Type *CondTy) const override;
- virtual unsigned getVectorInstrCost(unsigned Opcode, Type *Val,
- unsigned Index) const override;
- virtual unsigned getMemoryOpCost(unsigned Opcode, Type *Src,
- unsigned Alignment,
- unsigned AddressSpace) const override;
-
- /// @}
-};
-
-} // end anonymous namespace
-
-INITIALIZE_AG_PASS(PPCTTI, TargetTransformInfo, "ppctti",
- "PPC Target Transform Info", true, true, false)
-char PPCTTI::ID = 0;
-
-ImmutablePass *
-llvm::createPPCTargetTransformInfoPass(const PPCTargetMachine *TM) {
- return new PPCTTI(TM);
-}
-
-
//===----------------------------------------------------------------------===//
//
// PPC cost model.
//
//===----------------------------------------------------------------------===//
-PPCTTI::PopcntSupportKind PPCTTI::getPopcntSupport(unsigned TyWidth) const {
+TargetTransformInfo::PopcntSupportKind
+PPCTTIImpl::getPopcntSupport(unsigned TyWidth) {
assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
if (ST->hasPOPCNTD() && TyWidth <= 64)
- return PSK_FastHardware;
- return PSK_Software;
+ return TTI::PSK_FastHardware;
+ return TTI::PSK_Software;
}
-unsigned PPCTTI::getIntImmCost(const APInt &Imm, Type *Ty) const {
+int PPCTTIImpl::getIntImmCost(const APInt &Imm, Type *Ty) {
if (DisablePPCConstHoist)
- return TargetTransformInfo::getIntImmCost(Imm, Ty);
+ return BaseT::getIntImmCost(Imm, Ty);
assert(Ty->isIntegerTy());
return ~0U;
if (Imm == 0)
- return TCC_Free;
+ return TTI::TCC_Free;
if (Imm.getBitWidth() <= 64) {
if (isInt<16>(Imm.getSExtValue()))
- return TCC_Basic;
+ return TTI::TCC_Basic;
if (isInt<32>(Imm.getSExtValue())) {
// A constant that can be materialized using lis.
if ((Imm.getZExtValue() & 0xFFFF) == 0)
- return TCC_Basic;
+ return TTI::TCC_Basic;
- return 2 * TCC_Basic;
+ return 2 * TTI::TCC_Basic;
}
}
- return 4 * TCC_Basic;
+ return 4 * TTI::TCC_Basic;
}
-unsigned PPCTTI::getIntImmCost(Intrinsic::ID IID, unsigned Idx,
- const APInt &Imm, Type *Ty) const {
+int PPCTTIImpl::getIntImmCost(Intrinsic::ID IID, unsigned Idx, const APInt &Imm,
+ Type *Ty) {
if (DisablePPCConstHoist)
- return TargetTransformInfo::getIntImmCost(IID, Idx, Imm, Ty);
+ return BaseT::getIntImmCost(IID, Idx, Imm, Ty);
assert(Ty->isIntegerTy());
return ~0U;
switch (IID) {
- default: return TCC_Free;
+ default:
+ return TTI::TCC_Free;
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow:
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
if ((Idx == 1) && Imm.getBitWidth() <= 64 && isInt<16>(Imm.getSExtValue()))
- return TCC_Free;
+ return TTI::TCC_Free;
+ break;
+ case Intrinsic::experimental_stackmap:
+ if ((Idx < 2) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
+ return TTI::TCC_Free;
+ break;
+ case Intrinsic::experimental_patchpoint_void:
+ case Intrinsic::experimental_patchpoint_i64:
+ if ((Idx < 4) || (Imm.getBitWidth() <= 64 && isInt<64>(Imm.getSExtValue())))
+ return TTI::TCC_Free;
break;
}
- return PPCTTI::getIntImmCost(Imm, Ty);
+ return PPCTTIImpl::getIntImmCost(Imm, Ty);
}
-unsigned PPCTTI::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
- Type *Ty) const {
+int PPCTTIImpl::getIntImmCost(unsigned Opcode, unsigned Idx, const APInt &Imm,
+ Type *Ty) {
if (DisablePPCConstHoist)
- return TargetTransformInfo::getIntImmCost(Opcode, Idx, Imm, Ty);
+ return BaseT::getIntImmCost(Opcode, Idx, Imm, Ty);
assert(Ty->isIntegerTy());
bool ShiftedFree = false, RunFree = false, UnsignedFree = false,
ZeroFree = false;
switch (Opcode) {
- default: return TCC_Free;
+ default:
+ return TTI::TCC_Free;
case Instruction::GetElementPtr:
// Always hoist the base address of a GetElementPtr. This prevents the
// creation of new constants for every base constant that gets constant
// folded with the offset.
if (Idx == 0)
- return 2 * TCC_Basic;
- return TCC_Free;
+ return 2 * TTI::TCC_Basic;
+ return TTI::TCC_Free;
case Instruction::And:
RunFree = true; // (for the rotate-and-mask instructions)
// Fallthrough...
}
if (ZeroFree && Imm == 0)
- return TCC_Free;
+ return TTI::TCC_Free;
if (Idx == ImmIdx && Imm.getBitWidth() <= 64) {
if (isInt<16>(Imm.getSExtValue()))
- return TCC_Free;
+ return TTI::TCC_Free;
if (RunFree) {
if (Imm.getBitWidth() <= 32 &&
(isShiftedMask_32(Imm.getZExtValue()) ||
isShiftedMask_32(~Imm.getZExtValue())))
- return TCC_Free;
-
+ return TTI::TCC_Free;
if (ST->isPPC64() &&
(isShiftedMask_64(Imm.getZExtValue()) ||
isShiftedMask_64(~Imm.getZExtValue())))
- return TCC_Free;
+ return TTI::TCC_Free;
}
if (UnsignedFree && isUInt<16>(Imm.getZExtValue()))
- return TCC_Free;
+ return TTI::TCC_Free;
if (ShiftedFree && (Imm.getZExtValue() & 0xFFFF) == 0)
- return TCC_Free;
+ return TTI::TCC_Free;
}
- return PPCTTI::getIntImmCost(Imm, Ty);
+ return PPCTTIImpl::getIntImmCost(Imm, Ty);
}
-void PPCTTI::getUnrollingPreferences(Loop *L, UnrollingPreferences &UP) const {
+void PPCTTIImpl::getUnrollingPreferences(Loop *L,
+ TTI::UnrollingPreferences &UP) {
if (ST->getDarwinDirective() == PPC::DIR_A2) {
// The A2 is in-order with a deep pipeline, and concatenation unrolling
// helps expose latency-hiding opportunities to the instruction scheduler.
UP.Partial = UP.Runtime = true;
+
+ // We unroll a lot on the A2 (hundreds of instructions), and the benefits
+ // often outweigh the cost of a division to compute the trip count.
+ UP.AllowExpensiveTripCount = true;
}
+
+ BaseT::getUnrollingPreferences(L, UP);
}
-unsigned PPCTTI::getNumberOfRegisters(bool Vector) const {
- if (Vector && !ST->hasAltivec())
+bool PPCTTIImpl::enableAggressiveInterleaving(bool LoopHasReductions) {
+ // On the A2, always unroll aggressively. For QPX unaligned loads, we depend
+ // on combining the loads generated for consecutive accesses, and failure to
+ // do so is particularly expensive. This makes it much more likely (compared
+ // to only using concatenation unrolling).
+ if (ST->getDarwinDirective() == PPC::DIR_A2)
+ return true;
+
+ return LoopHasReductions;
+}
+
+bool PPCTTIImpl::enableInterleavedAccessVectorization() {
+ return true;
+}
+
+unsigned PPCTTIImpl::getNumberOfRegisters(bool Vector) {
+ if (Vector && !ST->hasAltivec() && !ST->hasQPX())
return 0;
return ST->hasVSX() ? 64 : 32;
}
-unsigned PPCTTI::getRegisterBitWidth(bool Vector) const {
+unsigned PPCTTIImpl::getRegisterBitWidth(bool Vector) {
if (Vector) {
+ if (ST->hasQPX()) return 256;
if (ST->hasAltivec()) return 128;
return 0;
}
}
-unsigned PPCTTI::getMaximumUnrollFactor() const {
+unsigned PPCTTIImpl::getMaxInterleaveFactor(unsigned VF) {
unsigned Directive = ST->getDarwinDirective();
// The 440 has no SIMD support, but floating-point instructions
// have a 5-cycle latency, so unroll by 5x for latency hiding.
if (Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500)
return 1;
+ // For P7 and P8, floating-point instructions have a 6-cycle latency and
+ // there are two execution units, so unroll by 12x for latency hiding.
+ if (Directive == PPC::DIR_PWR7 ||
+ Directive == PPC::DIR_PWR8)
+ return 12;
+
// For most things, modern systems have two execution units (and
// out-of-order execution).
return 2;
}
-unsigned PPCTTI::getArithmeticInstrCost(
- unsigned Opcode, Type *Ty, OperandValueKind Op1Info,
- OperandValueKind Op2Info, OperandValueProperties Opd1PropInfo,
- OperandValueProperties Opd2PropInfo) const {
+int PPCTTIImpl::getArithmeticInstrCost(
+ unsigned Opcode, Type *Ty, TTI::OperandValueKind Op1Info,
+ TTI::OperandValueKind Op2Info, TTI::OperandValueProperties Opd1PropInfo,
+ TTI::OperandValueProperties Opd2PropInfo) {
assert(TLI->InstructionOpcodeToISD(Opcode) && "Invalid opcode");
// Fallback to the default implementation.
- return TargetTransformInfo::getArithmeticInstrCost(
- Opcode, Ty, Op1Info, Op2Info, Opd1PropInfo, Opd2PropInfo);
+ return BaseT::getArithmeticInstrCost(Opcode, Ty, Op1Info, Op2Info,
+ Opd1PropInfo, Opd2PropInfo);
}
-unsigned PPCTTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,
- Type *SubTp) const {
- return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
+int PPCTTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
+ Type *SubTp) {
+ // Legalize the type.
+ std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);
+
+ // PPC, for both Altivec/VSX and QPX, support cheap arbitrary permutations
+ // (at least in the sense that there need only be one non-loop-invariant
+ // instruction). We need one such shuffle instruction for each actual
+ // register (this is not true for arbitrary shuffles, but is true for the
+ // structured types of shuffles covered by TTI::ShuffleKind).
+ return LT.first;
}
-unsigned PPCTTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
+int PPCTTIImpl::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) {
assert(TLI->InstructionOpcodeToISD(Opcode) && "Invalid opcode");
- return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+ return BaseT::getCastInstrCost(Opcode, Dst, Src);
}
-unsigned PPCTTI::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
- Type *CondTy) const {
- return TargetTransformInfo::getCmpSelInstrCost(Opcode, ValTy, CondTy);
+int PPCTTIImpl::getCmpSelInstrCost(unsigned Opcode, Type *ValTy, Type *CondTy) {
+ return BaseT::getCmpSelInstrCost(Opcode, ValTy, CondTy);
}
-unsigned PPCTTI::getVectorInstrCost(unsigned Opcode, Type *Val,
- unsigned Index) const {
+int PPCTTIImpl::getVectorInstrCost(unsigned Opcode, Type *Val, unsigned Index) {
assert(Val->isVectorTy() && "This must be a vector type");
int ISD = TLI->InstructionOpcodeToISD(Opcode);
if (Index == 0)
return 0;
- return TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+ return BaseT::getVectorInstrCost(Opcode, Val, Index);
+ } else if (ST->hasQPX() && Val->getScalarType()->isFloatingPointTy()) {
+ // Floating point scalars are already located in index #0.
+ if (Index == 0)
+ return 0;
+
+ return BaseT::getVectorInstrCost(Opcode, Val, Index);
}
// Estimated cost of a load-hit-store delay. This was obtained
// these need to be estimated as very costly.
if (ISD == ISD::EXTRACT_VECTOR_ELT ||
ISD == ISD::INSERT_VECTOR_ELT)
- return LHSPenalty +
- TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+ return LHSPenalty + BaseT::getVectorInstrCost(Opcode, Val, Index);
- return TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+ return BaseT::getVectorInstrCost(Opcode, Val, Index);
}
-unsigned PPCTTI::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
- unsigned AddressSpace) const {
+int PPCTTIImpl::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
+ unsigned AddressSpace) {
// Legalize the type.
- std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+ std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, Src);
assert((Opcode == Instruction::Load || Opcode == Instruction::Store) &&
"Invalid Opcode");
- unsigned Cost =
- TargetTransformInfo::getMemoryOpCost(Opcode, Src, Alignment, AddressSpace);
+ int Cost = BaseT::getMemoryOpCost(Opcode, Src, Alignment, AddressSpace);
- // VSX loads/stores support unaligned access.
- if (ST->hasVSX()) {
- if (LT.second == MVT::v2f64 || LT.second == MVT::v2i64)
- return Cost;
- }
-
- bool UnalignedAltivec =
- Src->isVectorTy() &&
- Src->getPrimitiveSizeInBits() >= LT.second.getSizeInBits() &&
- LT.second.getSizeInBits() == 128 &&
- Opcode == Instruction::Load;
+ // Aligned loads and stores are easy.
+ unsigned SrcBytes = LT.second.getStoreSize();
+ if (!SrcBytes || !Alignment || Alignment >= SrcBytes)
+ return Cost;
+
+ bool IsAltivecType = ST->hasAltivec() &&
+ (LT.second == MVT::v16i8 || LT.second == MVT::v8i16 ||
+ LT.second == MVT::v4i32 || LT.second == MVT::v4f32);
+ bool IsVSXType = ST->hasVSX() &&
+ (LT.second == MVT::v2f64 || LT.second == MVT::v2i64);
+ bool IsQPXType = ST->hasQPX() &&
+ (LT.second == MVT::v4f64 || LT.second == MVT::v4f32);
+
+ // If we can use the permutation-based load sequence, then this is also
+ // relatively cheap (not counting loop-invariant instructions): one load plus
+ // one permute (the last load in a series has extra cost, but we're
+ // neglecting that here). Note that on the P7, we should do unaligned loads
+ // for Altivec types using the VSX instructions, but that's more expensive
+ // than using the permutation-based load sequence. On the P8, that's no
+ // longer true.
+ if (Opcode == Instruction::Load &&
+ ((!ST->hasP8Vector() && IsAltivecType) || IsQPXType) &&
+ Alignment >= LT.second.getScalarType().getStoreSize())
+ return Cost + LT.first; // Add the cost of the permutations.
+
+ // For VSX, we can do unaligned loads and stores on Altivec/VSX types. On the
+ // P7, unaligned vector loads are more expensive than the permutation-based
+ // load sequence, so that might be used instead, but regardless, the net cost
+ // is about the same (not counting loop-invariant instructions).
+ if (IsVSXType || (ST->hasVSX() && IsAltivecType))
+ return Cost;
// PPC in general does not support unaligned loads and stores. They'll need
// to be decomposed based on the alignment factor.
- unsigned SrcBytes = LT.second.getStoreSize();
- if (SrcBytes && Alignment && Alignment < SrcBytes && !UnalignedAltivec) {
- Cost += LT.first*(SrcBytes/Alignment-1);
-
- // For a vector type, there is also scalarization overhead (only for
- // stores, loads are expanded using the vector-load + permutation sequence,
- // which is much less expensive).
- if (Src->isVectorTy() && Opcode == Instruction::Store)
- for (int i = 0, e = Src->getVectorNumElements(); i < e; ++i)
- Cost += getVectorInstrCost(Instruction::ExtractElement, Src, i);
- }
+
+ // Add the cost of each scalar load or store.
+ Cost += LT.first*(SrcBytes/Alignment-1);
+
+ // For a vector type, there is also scalarization overhead (only for
+ // stores, loads are expanded using the vector-load + permutation sequence,
+ // which is much less expensive).
+ if (Src->isVectorTy() && Opcode == Instruction::Store)
+ for (int i = 0, e = Src->getVectorNumElements(); i < e; ++i)
+ Cost += getVectorInstrCost(Instruction::ExtractElement, Src, i);
+
+ return Cost;
+}
+
+int PPCTTIImpl::getInterleavedMemoryOpCost(unsigned Opcode, Type *VecTy,
+ unsigned Factor,
+ ArrayRef<unsigned> Indices,
+ unsigned Alignment,
+ unsigned AddressSpace) {
+ assert(isa<VectorType>(VecTy) &&
+ "Expect a vector type for interleaved memory op");
+
+ // Legalize the type.
+ std::pair<int, MVT> LT = TLI->getTypeLegalizationCost(DL, VecTy);
+
+ // Firstly, the cost of load/store operation.
+ int Cost = getMemoryOpCost(Opcode, VecTy, Alignment, AddressSpace);
+
+ // PPC, for both Altivec/VSX and QPX, support cheap arbitrary permutations
+ // (at least in the sense that there need only be one non-loop-invariant
+ // instruction). For each result vector, we need one shuffle per incoming
+ // vector (except that the first shuffle can take two incoming vectors
+ // because it does not need to take itself).
+ Cost += Factor*(LT.first-1);
return Cost;
}