RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-11-27 13:40:30 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[CodeGen] Teach LLVM how to lower @llvm.{min,max}num to {MIN,MAX}NAN

Browse Source 
The behavior of {MIN,MAX}NAN differs from that of {MIN,MAX}NUM when only
one of the inputs is NaN: -NUM will return the non-NaN argument while
-NAN would return NaN.

It is desirable to lower to @llvm.{min,max}num to -NAN if they don't
have a native instruction for -NUM.  Notably, ARMv7 NEON's vmin has the
-NAN semantics.

N.B.  Of course, it is only safe to do this if the intrinsic call is
marked nnan.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@266279 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
David Majnemer 2016-04-14 07:13:24 +00:00
parent c1f6e319fe
commit 1f0cfcc0d1
10 changed files with 152 additions and 75 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
include/llvm/Analysis/TargetTransformInfo.h
View File

@ -25,6 +25,7 @@
#include "llvm/ADT/Optional.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/PassManager.h"
#include "llvm/Pass.h"
#include "llvm/Support/DataTypes.h"
@ -518,11 +519,11 @@ public:
/// \returns The cost of Intrinsic instructions. Types analysis only.
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys) const;
ArrayRef<Type *> Tys, FastMathFlags FMF) const;
/// \returns The cost of Intrinsic instructions. Analyses the real arguments.
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args) const;
ArrayRef<Value *> Args, FastMathFlags FMF) const;
/// \returns The cost of Call instructions.
int getCallInstrCost(Function *F, Type *RetTy, ArrayRef<Type *> Tys) const;
@ -664,9 +665,11 @@ public:
virtual int getReductionCost(unsigned Opcode, Type *Ty,
bool IsPairwiseForm) = 0;
virtual int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys) = 0;
ArrayRef<Type *> Tys,
FastMathFlags FMF) = 0;
virtual int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args) = 0;
ArrayRef<Value *> Args,
FastMathFlags FMF) = 0;
virtual int getCallInstrCost(Function *F, Type *RetTy,
ArrayRef<Type *> Tys) = 0;
virtual unsigned getNumberOfParts(Type *Tp) = 0;
@ -861,13 +864,14 @@ public:
bool IsPairwiseForm) override {
return Impl.getReductionCost(Opcode, Ty, IsPairwiseForm);
}
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys) override {
return Impl.getIntrinsicInstrCost(ID, RetTy, Tys);
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy, ArrayRef<Type *> Tys,
FastMathFlags FMF) override {
return Impl.getIntrinsicInstrCost(ID, RetTy, Tys, FMF);
}
int getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args) override {
return Impl.getIntrinsicInstrCost(ID, RetTy, Args);
ArrayRef<Value *> Args,
FastMathFlags FMF) override {
return Impl.getIntrinsicInstrCost(ID, RetTy, Args, FMF);
}
int getCallInstrCost(Function *F, Type *RetTy,
ArrayRef<Type *> Tys) override {

4
include/llvm/Analysis/TargetTransformInfoImpl.h
View File

@ -324,11 +324,11 @@ public:
}
unsigned getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys) {
ArrayRef<Type *> Tys, FastMathFlags FMF) {
return 1;
}
unsigned getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args) {
ArrayRef<Value *> Args, FastMathFlags FMF) {
return 1;
}

105
include/llvm/CodeGen/BasicTTIImpl.h
View File

@ -587,13 +587,14 @@ public:
/// Get intrinsic cost based on arguments
unsigned getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
ArrayRef<Value *> Args) {
ArrayRef<Value *> Args, FastMathFlags FMF) {
switch (IID) {
default: {
SmallVector<Type *, 4> Types;
for (Value *Op : Args)
Types.push_back(Op->getType());
return static_cast<T *>(this)->getIntrinsicInstrCost(IID, RetTy, Types);
return static_cast<T *>(this)->getIntrinsicInstrCost(IID, RetTy, Types,
FMF);
}
case Intrinsic::masked_scatter: {
Value *Mask = Args[3];
@ -619,8 +620,8 @@ public:
/// Get intrinsic cost based on argument types
unsigned getIntrinsicInstrCost(Intrinsic::ID IID, Type *RetTy,
ArrayRef<Type *> Tys) {
unsigned ISD = 0;
ArrayRef<Type *> Tys, FastMathFlags FMF) {
SmallVector<unsigned, 2> ISDs;
unsigned SingleCallCost = 10; // Library call cost. Make it expensive.
switch (IID) {
default: {
@ -647,74 +648,78 @@ public:
return 1; // Return cost of a scalar intrinsic. Assume it to be cheap.
unsigned ScalarCost = static_cast<T *>(this)->getIntrinsicInstrCost(
IID, ScalarRetTy, ScalarTys);
IID, ScalarRetTy, ScalarTys, FMF);
return ScalarCalls * ScalarCost + ScalarizationCost;
}
// Look for intrinsics that can be lowered directly or turned into a scalar
// intrinsic call.
case Intrinsic::sqrt:
ISD = ISD::FSQRT;
ISDs.push_back(ISD::FSQRT);
break;
case Intrinsic::sin:
ISD = ISD::FSIN;
ISDs.push_back(ISD::FSIN);
break;
case Intrinsic::cos:
ISD = ISD::FCOS;
ISDs.push_back(ISD::FCOS);
break;
case Intrinsic::exp:
ISD = ISD::FEXP;
ISDs.push_back(ISD::FEXP);
break;
case Intrinsic::exp2:
ISD = ISD::FEXP2;
ISDs.push_back(ISD::FEXP2);
break;
case Intrinsic::log:
ISD = ISD::FLOG;
ISDs.push_back(ISD::FLOG);
break;
case Intrinsic::log10:
ISD = ISD::FLOG10;
ISDs.push_back(ISD::FLOG10);
break;
case Intrinsic::log2:
ISD = ISD::FLOG2;
ISDs.push_back(ISD::FLOG2);
break;
case Intrinsic::fabs:
ISD = ISD::FABS;
ISDs.push_back(ISD::FABS);
break;
case Intrinsic::minnum:
ISD = ISD::FMINNUM;
ISDs.push_back(ISD::FMINNUM);
if (FMF.noNaNs())
ISDs.push_back(ISD::FMINNAN);
break;
case Intrinsic::maxnum:
ISD = ISD::FMAXNUM;
ISDs.push_back(ISD::FMAXNUM);
if (FMF.noNaNs())
ISDs.push_back(ISD::FMAXNAN);
break;
case Intrinsic::copysign:
ISD = ISD::FCOPYSIGN;
ISDs.push_back(ISD::FCOPYSIGN);
break;
case Intrinsic::floor:
ISD = ISD::FFLOOR;
ISDs.push_back(ISD::FFLOOR);
break;
case Intrinsic::ceil:
ISD = ISD::FCEIL;
ISDs.push_back(ISD::FCEIL);
break;
case Intrinsic::trunc:
ISD = ISD::FTRUNC;
ISDs.push_back(ISD::FTRUNC);
break;
case Intrinsic::nearbyint:
ISD = ISD::FNEARBYINT;
ISDs.push_back(ISD::FNEARBYINT);
break;
case Intrinsic::rint:
ISD = ISD::FRINT;
ISDs.push_back(ISD::FRINT);
break;
case Intrinsic::round:
ISD = ISD::FROUND;
ISDs.push_back(ISD::FROUND);
break;
case Intrinsic::pow:
ISD = ISD::FPOW;
ISDs.push_back(ISD::FPOW);
break;
case Intrinsic::fma:
ISD = ISD::FMA;
ISDs.push_back(ISD::FMA);
break;
case Intrinsic::fmuladd:
ISD = ISD::FMA;
ISDs.push_back(ISD::FMA);
break;
// FIXME: We should return 0 whenever getIntrinsicCost == TCC_Free.
case Intrinsic::lifetime_start:
@ -727,7 +732,7 @@ public:
return static_cast<T *>(this)
->getMaskedMemoryOpCost(Instruction::Load, RetTy, 0, 0);
case Intrinsic::ctpop:
ISD = ISD::CTPOP;
ISDs.push_back(ISD::CTPOP);
// In case of legalization use TCC_Expensive. This is cheaper than a
// library call but still not a cheap instruction.
SingleCallCost = TargetTransformInfo::TCC_Expensive;
@ -738,26 +743,36 @@ public:
const TargetLoweringBase *TLI = getTLI();
std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(DL, RetTy);
if (TLI->isOperationLegalOrPromote(ISD, LT.second)) {
if (IID == Intrinsic::fabs &&
TLI->isFAbsFree(LT.second)) {
return 0;
SmallVector<unsigned, 2> LegalCost;
SmallVector<unsigned, 2> CustomCost;
for (unsigned ISD : ISDs) {
if (TLI->isOperationLegalOrPromote(ISD, LT.second)) {
if (IID == Intrinsic::fabs && TLI->isFAbsFree(LT.second)) {
return 0;
}
// The operation is legal. Assume it costs 1.
// If the type is split to multiple registers, assume that there is some
// overhead to this.
// TODO: Once we have extract/insert subvector cost we need to use them.
if (LT.first > 1)
LegalCost.push_back(LT.first * 2);
else
LegalCost.push_back(LT.first * 1);
} else if (!TLI->isOperationExpand(ISD, LT.second)) {
// If the operation is custom lowered then assume
// that the code is twice as expensive.
CustomCost.push_back(LT.first * 2);
}
// The operation is legal. Assume it costs 1.
// If the type is split to multiple registers, assume that there is some
// overhead to this.
// TODO: Once we have extract/insert subvector cost we need to use them.
if (LT.first > 1)
return LT.first * 2;
return LT.first * 1;
}
if (!TLI->isOperationExpand(ISD, LT.second)) {
// If the operation is custom lowered then assume
// thare the code is twice as expensive.
return LT.first * 2;
}
auto MinLegalCostI = std::min_element(LegalCost.begin(), LegalCost.end());
if (MinLegalCostI != LegalCost.end())
return *MinLegalCostI;
auto MinCustomCostI = std::min_element(CustomCost.begin(), CustomCost.end());
if (MinCustomCostI != CustomCost.end())
return *MinCustomCostI;
// If we can't lower fmuladd into an FMA estimate the cost as a floating
// point mul followed by an add.
@ -781,7 +796,7 @@ public:
ScalarTys.push_back(Ty);
}
unsigned ScalarCost = static_cast<T *>(this)->getIntrinsicInstrCost(
IID, RetTy->getScalarType(), ScalarTys);
IID, RetTy->getScalarType(), ScalarTys, FMF);
for (unsigned i = 0, ie = Tys.size(); i != ie; ++i) {
if (Tys[i]->isVectorTy()) {
ScalarizationCost += getScalarizationOverhead(Tys[i], false, true);

6
lib/Analysis/CostModel.cpp
View File

@ -504,8 +504,12 @@ unsigned CostModelAnalysis::getInstructionCost(const Instruction *I) const {
for (unsigned J = 0, JE = II->getNumArgOperands(); J != JE; ++J)
Args.push_back(II->getArgOperand(J));
FastMathFlags FMF;
if (auto *FPMO = dyn_cast<FPMathOperator>(II))
FMF = FPMO->getFastMathFlags();
return TTI->getIntrinsicInstrCost(II->getIntrinsicID(), II->getType(),
Args);
Args, FMF);
}
return -1;
default:

10
lib/Analysis/TargetTransformInfo.cpp
View File

@ -315,15 +315,17 @@ int TargetTransformInfo::getInterleavedMemoryOpCost(
}
int TargetTransformInfo::getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Type *> Tys) const {
int Cost = TTIImpl->getIntrinsicInstrCost(ID, RetTy, Tys);
ArrayRef<Type *> Tys,
FastMathFlags FMF) const {
int Cost = TTIImpl->getIntrinsicInstrCost(ID, RetTy, Tys, FMF);
assert(Cost >= 0 && "TTI should not produce negative costs!");
return Cost;
}
int TargetTransformInfo::getIntrinsicInstrCost(Intrinsic::ID ID, Type *RetTy,
ArrayRef<Value *> Args) const {
int Cost = TTIImpl->getIntrinsicInstrCost(ID, RetTy, Args);
ArrayRef<Value *> Args,
FastMathFlags FMF) const {
int Cost = TTIImpl->getIntrinsicInstrCost(ID, RetTy, Args, FMF);
assert(Cost >= 0 && "TTI should not produce negative costs!");
return Cost;
}

22
lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
View File

@ -5143,18 +5143,28 @@ SelectionDAGBuilder::visitIntrinsicCall(const CallInst &I, unsigned Intrinsic) {
getValue(I.getArgOperand(0))));
return nullptr;
}
case Intrinsic::minnum:
setValue(&I, DAG.getNode(ISD::FMINNUM, sdl,
getValue(I.getArgOperand(0)).getValueType(),
case Intrinsic::minnum: {
auto VT = getValue(I.getArgOperand(0)).getValueType();
unsigned Opc =
I.hasNoNaNs() && TLI.isOperationLegalOrCustom(ISD::FMINNAN, VT)
? ISD::FMINNAN
: ISD::FMINNUM;
setValue(&I, DAG.getNode(Opc, sdl, VT,
getValue(I.getArgOperand(0)),
getValue(I.getArgOperand(1))));
return nullptr;
case Intrinsic::maxnum:
setValue(&I, DAG.getNode(ISD::FMAXNUM, sdl,
getValue(I.getArgOperand(0)).getValueType(),
}
case Intrinsic::maxnum: {
auto VT = getValue(I.getArgOperand(0)).getValueType();
unsigned Opc =
I.hasNoNaNs() && TLI.isOperationLegalOrCustom(ISD::FMAXNAN, VT)
? ISD::FMAXNAN
: ISD::FMAXNUM;
setValue(&I, DAG.getNode(Opc, sdl, VT,
getValue(I.getArgOperand(0)),
getValue(I.getArgOperand(1))));
return nullptr;
}
case Intrinsic::copysign:
setValue(&I, DAG.getNode(ISD::FCOPYSIGN, sdl,
getValue(I.getArgOperand(0)).getValueType(),

17
lib/Transforms/Vectorize/BBVectorize.cpp
View File

@ -1117,16 +1117,25 @@ namespace {
}
if (IID && TTI) {
FastMathFlags FMFCI;
if (auto *FPMOCI = dyn_cast<FPMathOperator>(CI))
FMFCI = FPMOCI->getFastMathFlags();
SmallVector<Type*, 4> Tys;
for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i)
Tys.push_back(CI->getArgOperand(i)->getType());
unsigned ICost = TTI->getIntrinsicInstrCost(IID, IT1, Tys);
unsigned ICost = TTI->getIntrinsicInstrCost(IID, IT1, Tys, FMFCI);
Tys.clear();
CallInst *CJ = cast<CallInst>(J);
FastMathFlags FMFCJ;
if (auto *FPMOCJ = dyn_cast<FPMathOperator>(CJ))
FMFCJ = FPMOCJ->getFastMathFlags();
for (unsigned i = 0, ie = CJ->getNumArgOperands(); i != ie; ++i)
Tys.push_back(CJ->getArgOperand(i)->getType());
unsigned JCost = TTI->getIntrinsicInstrCost(IID, JT1, Tys);
unsigned JCost = TTI->getIntrinsicInstrCost(IID, JT1, Tys, FMFCJ);
Tys.clear();
assert(CI->getNumArgOperands() == CJ->getNumArgOperands() &&
@ -1140,8 +1149,10 @@ namespace {
CJ->getArgOperand(i)->getType()));
}
FastMathFlags FMFV = FMFCI;
FMFV &= FMFCJ;
Type *RetTy = getVecTypeForPair(IT1, JT1);
unsigned VCost = TTI->getIntrinsicInstrCost(IID, RetTy, Tys);
unsigned VCost = TTI->getIntrinsicInstrCost(IID, RetTy, Tys, FMFV);
if (VCost > ICost + JCost)
return false;

14
lib/Transforms/Vectorize/LoopVectorize.cpp
View File

@ -3302,7 +3302,11 @@ static unsigned getVectorIntrinsicCost(CallInst *CI, unsigned VF,
for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i)
Tys.push_back(ToVectorTy(CI->getArgOperand(i)->getType(), VF));
return TTI.getIntrinsicInstrCost(ID, RetTy, Tys);
FastMathFlags FMF;
if (auto *FPMO = dyn_cast<FPMathOperator>(CI))
FMF = FPMO->getFastMathFlags();
return TTI.getIntrinsicInstrCost(ID, RetTy, Tys, FMF);
}
static Type *smallestIntegerVectorType(Type *T1, Type *T2) {
@ -4269,7 +4273,13 @@ void InnerLoopVectorizer::vectorizeBlockInLoop(BasicBlock *BB, PhiVector *PV) {
}
}
assert(VectorF && "Can't create vector function.");
Entry[Part] = Builder.CreateCall(VectorF, Args);
CallInst *V = Builder.CreateCall(VectorF, Args);
if (isa<FPMathOperator>(V))
V->copyFastMathFlags(CI);
Entry[Part] = V;
}
addMetadata(Entry, &*it);

10
lib/Transforms/Vectorize/SLPVectorizer.cpp
View File

@ -1659,10 +1659,14 @@ int BoUpSLP::getEntryCost(TreeEntry *E) {
VecTy->getNumElements()));
}
int ScalarCallCost = VecTy->getNumElements() *
TTI->getIntrinsicInstrCost(ID, ScalarTy, ScalarTys);
FastMathFlags FMF;
if (auto *FPMO = dyn_cast<FPMathOperator>(CI))
FMF = FPMO->getFastMathFlags();
int VecCallCost = TTI->getIntrinsicInstrCost(ID, VecTy, VecTys);
int ScalarCallCost = VecTy->getNumElements() *
TTI->getIntrinsicInstrCost(ID, ScalarTy, ScalarTys, FMF);
int VecCallCost = TTI->getIntrinsicInstrCost(ID, VecTy, VecTys, FMF);
DEBUG(dbgs() << "SLP: Call cost "<< VecCallCost - ScalarCallCost
<< " (" << VecCallCost << "-" << ScalarCallCost << ")"

17
test/CodeGen/ARM/vminmax.ll
View File

@ -291,3 +291,20 @@ declare <8 x i16> @llvm.arm.neon.vmaxu.v8i16(<8 x i16>, <8 x i16>) nounwind read
declare <4 x i32> @llvm.arm.neon.vmaxu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone
declare <4 x float> @llvm.arm.neon.vmaxs.v4f32(<4 x float>, <4 x float>) nounwind readnone
declare float @llvm.maxnum.f32(float %a, float %b)
declare float @llvm.minnum.f32(float %a, float %b)
define float @maxnum(float %a, float %b) {
;CHECK-LABEL: maxnum:
;CHECK: vmax.f32
%r = call nnan float @llvm.maxnum.f32(float %a, float %b)
ret float %r
}
define float @minnum(float %a, float %b) {
;CHECK-LABEL: minnum:
;CHECK: vmin.f32
%r = call nnan float @llvm.minnum.f32(float %a, float %b)
ret float %r
}