#include "llvm/Support/CommandLine.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Transforms/Utils/BuildLibCalls.h"
+#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
using namespace PatternMatch;
//===----------------------------------------------------------------------===//
static bool ignoreCallingConv(LibFunc::Func Func) {
- switch (Func) {
- case LibFunc::abs:
- case LibFunc::labs:
- case LibFunc::llabs:
- case LibFunc::strlen:
- return true;
- default:
- return false;
- }
- llvm_unreachable("All cases should be covered in the switch.");
+ return Func == LibFunc::abs || Func == LibFunc::labs ||
+ Func == LibFunc::llabs || Func == LibFunc::strlen;
}
/// isOnlyUsedInZeroEqualityComparison - Return true if it only matters that the
}
/// \brief Check whether the overloaded unary floating point function
-/// corresponing to \a Ty is available.
+/// corresponding to \a Ty is available.
static bool hasUnaryFloatFn(const TargetLibraryInfo *TLI, Type *Ty,
LibFunc::Func DoubleFn, LibFunc::Func FloatFn,
LibFunc::Func LongDoubleFn) {
}
}
+/// \brief Check whether we can use unsafe floating point math for
+/// the function passed as input.
+static bool canUseUnsafeFPMath(Function *F) {
+
+ // FIXME: For finer-grain optimization, we need intrinsics to have the same
+ // fast-math flag decorations that are applied to FP instructions. For now,
+ // we have to rely on the function-level unsafe-fp-math attribute to do this
+ // optimization because there's no other way to express that the sqrt can be
+ // reassociated.
+ if (F->hasFnAttribute("unsafe-fp-math")) {
+ Attribute Attr = F->getFnAttribute("unsafe-fp-math");
+ if (Attr.getValueAsString() == "true")
+ return true;
+ }
+ return false;
+}
+
/// \brief Returns whether \p F matches the signature expected for the
/// string/memory copying library function \p Func.
/// Acceptable functions are st[rp][n]?cpy, memove, memcpy, and memset.
Value *LibCallSimplifier::optimizeStpCpy(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
- // Verify the "stpcpy" function prototype.
- FunctionType *FT = Callee->getFunctionType();
-
if (!checkStringCopyLibFuncSignature(Callee, LibFunc::stpcpy))
return nullptr;
if (Len == 0)
return nullptr;
- Type *PT = FT->getParamType(0);
+ Type *PT = Callee->getFunctionType()->getParamType(0);
Value *LenV = ConstantInt::get(DL.getIntPtrType(PT), Len);
Value *DstEnd =
B.CreateGEP(B.getInt8Ty(), Dst, ConstantInt::get(DL.getIntPtrType(PT), Len - 1));
Value *LibCallSimplifier::optimizeStrNCpy(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
-
if (!checkStringCopyLibFuncSignature(Callee, LibFunc::strncpy))
return nullptr;
if (Len > SrcLen + 1)
return nullptr;
- Type *PT = FT->getParamType(0);
+ Type *PT = Callee->getFunctionType()->getParamType(0);
// strncpy(x, s, c) -> memcpy(x, s, c, 1) [s and c are constant]
B.CreateMemCpy(Dst, Src, ConstantInt::get(DL.getIntPtrType(PT), Len), 1);
return B.CreateSub(LHSV, RHSV, "chardiff");
}
+ // memcmp(S1,S2,N/8)==0 -> (*(intN_t*)S1 != *(intN_t*)S2)==0
+ if (DL.isLegalInteger(Len * 8) && isOnlyUsedInZeroEqualityComparison(CI)) {
+
+ IntegerType *IntType = IntegerType::get(CI->getContext(), Len * 8);
+ unsigned PrefAlignment = DL.getPrefTypeAlignment(IntType);
+
+ if (getKnownAlignment(LHS, DL, CI) >= PrefAlignment &&
+ getKnownAlignment(RHS, DL, CI) >= PrefAlignment) {
+
+ Type *LHSPtrTy =
+ IntType->getPointerTo(LHS->getType()->getPointerAddressSpace());
+ Type *RHSPtrTy =
+ IntType->getPointerTo(RHS->getType()->getPointerAddressSpace());
+
+ Value *LHSV = B.CreateLoad(B.CreateBitCast(LHS, LHSPtrTy, "lhsc"), "lhsv");
+ Value *RHSV = B.CreateLoad(B.CreateBitCast(RHS, RHSPtrTy, "rhsc"), "rhsv");
+
+ return B.CreateZExt(B.CreateICmpNE(LHSV, RHSV), CI->getType(), "memcmp");
+ }
+ }
+
// Constant folding: memcmp(x, y, l) -> cnst (all arguments are constant)
StringRef LHSStr, RHSStr;
if (getConstantStringInfo(LHS, LHSStr) &&
// floor((double)floatval) -> (double)floorf(floatval)
if (Callee->isIntrinsic()) {
Module *M = CI->getParent()->getParent()->getParent();
- Intrinsic::ID IID = (Intrinsic::ID) Callee->getIntrinsicID();
+ Intrinsic::ID IID = Callee->getIntrinsicID();
Function *F = Intrinsic::getDeclaration(M, IID, B.getFloatTy());
V = B.CreateCall(F, V);
} else {
Value *LibCallSimplifier::optimizeCos(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
Value *Ret = nullptr;
- if (UnsafeFPShrink && Callee->getName() == "cos" && TLI->has(LibFunc::cosf)) {
+ StringRef Name = Callee->getName();
+ if (UnsafeFPShrink && Name == "cos" && hasFloatVersion(Name))
Ret = optimizeUnaryDoubleFP(CI, B, true);
- }
FunctionType *FT = Callee->getFunctionType();
// Just make sure this has 1 argument of FP type, which matches the
Value *LibCallSimplifier::optimizePow(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
-
Value *Ret = nullptr;
- if (UnsafeFPShrink && Callee->getName() == "pow" && TLI->has(LibFunc::powf)) {
+ StringRef Name = Callee->getName();
+ if (UnsafeFPShrink && Name == "pow" && hasFloatVersion(Name))
Ret = optimizeUnaryDoubleFP(CI, B, true);
- }
FunctionType *FT = Callee->getFunctionType();
// Just make sure this has 2 arguments of the same FP type, which match the
if (Op1C->isExactlyValue(2.0) &&
hasUnaryFloatFn(TLI, Op1->getType(), LibFunc::exp2, LibFunc::exp2f,
LibFunc::exp2l))
- return EmitUnaryFloatFnCall(Op2, "exp2", B, Callee->getAttributes());
+ return EmitUnaryFloatFnCall(Op2, TLI->getName(LibFunc::exp2), B,
+ Callee->getAttributes());
// pow(10.0, x) -> exp10(x)
if (Op1C->isExactlyValue(10.0) &&
hasUnaryFloatFn(TLI, Op1->getType(), LibFunc::exp10, LibFunc::exp10f,
Callee->getAttributes());
}
+ // pow(exp(x), y) -> exp(x*y)
+ // pow(exp2(x), y) -> exp2(x * y)
+ // We enable these only under fast-math. Besides rounding
+ // differences the transformation changes overflow and
+ // underflow behavior quite dramatically.
+ // Example: x = 1000, y = 0.001.
+ // pow(exp(x), y) = pow(inf, 0.001) = inf, whereas exp(x*y) = exp(1).
+ if (canUseUnsafeFPMath(CI->getParent()->getParent())) {
+ if (auto *OpC = dyn_cast<CallInst>(Op1)) {
+ IRBuilder<>::FastMathFlagGuard Guard(B);
+ FastMathFlags FMF;
+ FMF.setUnsafeAlgebra();
+ B.SetFastMathFlags(FMF);
+
+ LibFunc::Func Func;
+ Function *Callee = OpC->getCalledFunction();
+ StringRef FuncName = Callee->getName();
+
+ if (TLI->getLibFunc(FuncName, Func) && TLI->has(Func) &&
+ (Func == LibFunc::exp || Func == LibFunc::exp2))
+ return EmitUnaryFloatFnCall(
+ B.CreateFMul(OpC->getArgOperand(0), Op2, "mul"), FuncName, B,
+ Callee->getAttributes());
+ }
+ }
+
ConstantFP *Op2C = dyn_cast<ConstantFP>(Op2);
if (!Op2C)
return Ret;
Value *LibCallSimplifier::optimizeExp2(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
Function *Caller = CI->getParent()->getParent();
-
Value *Ret = nullptr;
- if (UnsafeFPShrink && Callee->getName() == "exp2" &&
- TLI->has(LibFunc::exp2f)) {
+ StringRef Name = Callee->getName();
+ if (UnsafeFPShrink && Name == "exp2" && hasFloatVersion(Name))
Ret = optimizeUnaryDoubleFP(CI, B, true);
- }
FunctionType *FT = Callee->getFunctionType();
// Just make sure this has 1 argument of FP type, which matches the
Value *LibCallSimplifier::optimizeFabs(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
-
Value *Ret = nullptr;
- if (Callee->getName() == "fabs" && TLI->has(LibFunc::fabsf)) {
+ StringRef Name = Callee->getName();
+ if (Name == "fabs" && hasFloatVersion(Name))
Ret = optimizeUnaryDoubleFP(CI, B, false);
- }
FunctionType *FT = Callee->getFunctionType();
// Make sure this has 1 argument of FP type which matches the result type.
return Ret;
}
+Value *LibCallSimplifier::optimizeFMinFMax(CallInst *CI, IRBuilder<> &B) {
+ // If we can shrink the call to a float function rather than a double
+ // function, do that first.
+ Function *Callee = CI->getCalledFunction();
+ StringRef Name = Callee->getName();
+ if ((Name == "fmin" && hasFloatVersion(Name)) ||
+ (Name == "fmax" && hasFloatVersion(Name))) {
+ Value *Ret = optimizeBinaryDoubleFP(CI, B);
+ if (Ret)
+ return Ret;
+ }
+
+ // Make sure this has 2 arguments of FP type which match the result type.
+ FunctionType *FT = Callee->getFunctionType();
+ if (FT->getNumParams() != 2 || FT->getReturnType() != FT->getParamType(0) ||
+ FT->getParamType(0) != FT->getParamType(1) ||
+ !FT->getParamType(0)->isFloatingPointTy())
+ return nullptr;
+
+ IRBuilder<>::FastMathFlagGuard Guard(B);
+ FastMathFlags FMF;
+ Function *F = CI->getParent()->getParent();
+ if (canUseUnsafeFPMath(F)) {
+ // Unsafe algebra sets all fast-math-flags to true.
+ FMF.setUnsafeAlgebra();
+ } else {
+ // At a minimum, no-nans-fp-math must be true.
+ Attribute Attr = F->getFnAttribute("no-nans-fp-math");
+ if (Attr.getValueAsString() != "true")
+ return nullptr;
+ // No-signed-zeros is implied by the definitions of fmax/fmin themselves:
+ // "Ideally, fmax would be sensitive to the sign of zero, for example
+ // fmax(-0. 0, +0. 0) would return +0; however, implementation in software
+ // might be impractical."
+ FMF.setNoSignedZeros();
+ FMF.setNoNaNs();
+ }
+ B.SetFastMathFlags(FMF);
+
+ // We have a relaxed floating-point environment. We can ignore NaN-handling
+ // and transform to a compare and select. We do not have to consider errno or
+ // exceptions, because fmin/fmax do not have those.
+ Value *Op0 = CI->getArgOperand(0);
+ Value *Op1 = CI->getArgOperand(1);
+ Value *Cmp = Callee->getName().startswith("fmin") ?
+ B.CreateFCmpOLT(Op0, Op1) : B.CreateFCmpOGT(Op0, Op1);
+ return B.CreateSelect(Cmp, Op0, Op1);
+}
+
Value *LibCallSimplifier::optimizeSqrt(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
if (TLI->has(LibFunc::sqrtf) && (Callee->getName() == "sqrt" ||
Callee->getIntrinsicID() == Intrinsic::sqrt))
Ret = optimizeUnaryDoubleFP(CI, B, true);
+ if (!canUseUnsafeFPMath(CI->getParent()->getParent()))
+ return Ret;
- // FIXME: For finer-grain optimization, we need intrinsics to have the same
- // fast-math flag decorations that are applied to FP instructions. For now,
- // we have to rely on the function-level unsafe-fp-math attribute to do this
- // optimization because there's no other way to express that the sqrt can be
- // reassociated.
- Function *F = CI->getParent()->getParent();
- if (F->hasFnAttribute("unsafe-fp-math")) {
- // Check for unsafe-fp-math = true.
- Attribute Attr = F->getFnAttribute("unsafe-fp-math");
- if (Attr.getValueAsString() != "true")
- return Ret;
- }
Value *Op = CI->getArgOperand(0);
if (Instruction *I = dyn_cast<Instruction>(Op)) {
if (I->getOpcode() == Instruction::FMul && I->hasUnsafeAlgebra()) {
// and multiply.
// FIXME: We're not checking the sqrt because it doesn't have
// fast-math-flags (see earlier comment).
- IRBuilder<true, ConstantFolder,
- IRBuilderDefaultInserter<true> >::FastMathFlagGuard Guard(B);
+ IRBuilder<>::FastMathFlagGuard Guard(B);
B.SetFastMathFlags(I->getFastMathFlags());
// If we found a repeated factor, hoist it out of the square root and
// replace it with the fabs of that factor.
return Ret;
}
+Value *LibCallSimplifier::optimizeTan(CallInst *CI, IRBuilder<> &B) {
+ Function *Callee = CI->getCalledFunction();
+ Value *Ret = nullptr;
+ StringRef Name = Callee->getName();
+ if (UnsafeFPShrink && Name == "tan" && hasFloatVersion(Name))
+ Ret = optimizeUnaryDoubleFP(CI, B, true);
+ FunctionType *FT = Callee->getFunctionType();
+
+ // Just make sure this has 1 argument of FP type, which matches the
+ // result type.
+ if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
+ !FT->getParamType(0)->isFloatingPointTy())
+ return Ret;
+
+ if (!canUseUnsafeFPMath(CI->getParent()->getParent()))
+ return Ret;
+ Value *Op1 = CI->getArgOperand(0);
+ auto *OpC = dyn_cast<CallInst>(Op1);
+ if (!OpC)
+ return Ret;
+
+ // tan(atan(x)) -> x
+ // tanf(atanf(x)) -> x
+ // tanl(atanl(x)) -> x
+ LibFunc::Func Func;
+ Function *F = OpC->getCalledFunction();
+ StringRef FuncName = F->getName();
+ if (TLI->getLibFunc(FuncName, Func) && TLI->has(Func) &&
+ ((Func == LibFunc::atan && Callee->getName() == "tan") ||
+ (Func == LibFunc::atanf && Callee->getName() == "tanf") ||
+ (Func == LibFunc::atanl && Callee->getName() == "tanl")))
+ Ret = OpC->getArgOperand(0);
+ return Ret;
+}
+
static bool isTrigLibCall(CallInst *CI);
static void insertSinCosCall(IRBuilder<> &B, Function *OrigCallee, Value *Arg,
bool UseFloat, Value *&Sin, Value *&Cos,
void LibCallSimplifier::replaceTrigInsts(SmallVectorImpl<CallInst *> &Calls,
Value *Res) {
- for (SmallVectorImpl<CallInst *>::iterator I = Calls.begin(), E = Calls.end();
- I != E; ++I) {
- replaceAllUsesWith(*I, Res);
- }
+ for (CallInst *C : Calls)
+ replaceAllUsesWith(C, Res);
}
void insertSinCosCall(IRBuilder<> &B, Function *OrigCallee, Value *Arg,
if (Instruction *ArgInst = dyn_cast<Instruction>(Arg)) {
// If the argument is an instruction, it must dominate all uses so put our
// sincos call there.
- BasicBlock::iterator Loc = ArgInst;
- B.SetInsertPoint(ArgInst->getParent(), ++Loc);
+ B.SetInsertPoint(ArgInst->getParent(), ++ArgInst->getIterator());
} else {
// Otherwise (e.g. for a constant) the beginning of the function is as
// good a place as any.
// Integer Library Call Optimizations
//===----------------------------------------------------------------------===//
+static bool checkIntUnaryReturnAndParam(Function *Callee) {
+ FunctionType *FT = Callee->getFunctionType();
+ return FT->getNumParams() == 1 && FT->getReturnType()->isIntegerTy(32) &&
+ FT->getParamType(0)->isIntegerTy();
+}
+
Value *LibCallSimplifier::optimizeFFS(CallInst *CI, IRBuilder<> &B) {
Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
- // Just make sure this has 2 arguments of the same FP type, which match the
- // result type.
- if (FT->getNumParams() != 1 || !FT->getReturnType()->isIntegerTy(32) ||
- !FT->getParamType(0)->isIntegerTy())
+ if (!checkIntUnaryReturnAndParam(Callee))
return nullptr;
-
Value *Op = CI->getArgOperand(0);
// Constant fold.
Type *ArgType = Op->getType();
Value *F =
Intrinsic::getDeclaration(Callee->getParent(), Intrinsic::cttz, ArgType);
- Value *V = B.CreateCall(F, {Op, B.getFalse()}, "cttz");
+ Value *V = B.CreateCall(F, {Op, B.getTrue()}, "cttz");
V = B.CreateAdd(V, ConstantInt::get(V->getType(), 1));
V = B.CreateIntCast(V, B.getInt32Ty(), false);
}
Value *LibCallSimplifier::optimizeIsDigit(CallInst *CI, IRBuilder<> &B) {
- Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
- // We require integer(i32)
- if (FT->getNumParams() != 1 || !FT->getReturnType()->isIntegerTy() ||
- !FT->getParamType(0)->isIntegerTy(32))
+ if (!checkIntUnaryReturnAndParam(CI->getCalledFunction()))
return nullptr;
// isdigit(c) -> (c-'0') <u 10
}
Value *LibCallSimplifier::optimizeIsAscii(CallInst *CI, IRBuilder<> &B) {
- Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
- // We require integer(i32)
- if (FT->getNumParams() != 1 || !FT->getReturnType()->isIntegerTy() ||
- !FT->getParamType(0)->isIntegerTy(32))
+ if (!checkIntUnaryReturnAndParam(CI->getCalledFunction()))
return nullptr;
// isascii(c) -> c <u 128
}
Value *LibCallSimplifier::optimizeToAscii(CallInst *CI, IRBuilder<> &B) {
- Function *Callee = CI->getCalledFunction();
- FunctionType *FT = Callee->getFunctionType();
- // We require i32(i32)
- if (FT->getNumParams() != 1 || FT->getReturnType() != FT->getParamType(0) ||
- !FT->getParamType(0)->isIntegerTy(32))
+ if (!checkIntUnaryReturnAndParam(CI->getCalledFunction()))
return nullptr;
// toascii(c) -> c & 0x7f
}
static bool isReportingError(Function *Callee, CallInst *CI, int StreamArg) {
- if (!ColdErrorCalls)
- return false;
-
- if (!Callee || !Callee->isDeclaration())
+ if (!ColdErrorCalls || !Callee || !Callee->isDeclaration())
return false;
if (StreamArg < 0)
// Command-line parameter overrides function attribute.
if (EnableUnsafeFPShrink.getNumOccurrences() > 0)
UnsafeFPShrink = EnableUnsafeFPShrink;
- else if (Callee->hasFnAttribute("unsafe-fp-math")) {
- // FIXME: This is the same problem as described in optimizeSqrt().
- // If calls gain access to IR-level FMF, then use that instead of a
- // function attribute.
-
- // Check for unsafe-fp-math = true.
- Attribute Attr = Callee->getFnAttribute("unsafe-fp-math");
- if (Attr.getValueAsString() == "true")
- UnsafeFPShrink = true;
- }
+ else if (canUseUnsafeFPMath(Callee))
+ UnsafeFPShrink = true;
// First, check for intrinsics.
if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) {
if (Value *SimplifiedFortifiedCI = FortifiedSimplifier.optimizeCall(CI)) {
// Try to further simplify the result.
CallInst *SimplifiedCI = dyn_cast<CallInst>(SimplifiedFortifiedCI);
- if (SimplifiedCI && SimplifiedCI->getCalledFunction())
- if (Value *V = optimizeStringMemoryLibCall(SimplifiedCI, Builder)) {
+ if (SimplifiedCI && SimplifiedCI->getCalledFunction()) {
+ // Use an IR Builder from SimplifiedCI if available instead of CI
+ // to guarantee we reach all uses we might replace later on.
+ IRBuilder<> TmpBuilder(SimplifiedCI);
+ if (Value *V = optimizeStringMemoryLibCall(SimplifiedCI, TmpBuilder)) {
// If we were able to further simplify, remove the now redundant call.
SimplifiedCI->replaceAllUsesWith(V);
SimplifiedCI->eraseFromParent();
return V;
}
+ }
return SimplifiedFortifiedCI;
}
return optimizeFPuts(CI, Builder);
case LibFunc::puts:
return optimizePuts(CI, Builder);
+ case LibFunc::tan:
+ case LibFunc::tanf:
+ case LibFunc::tanl:
+ return optimizeTan(CI, Builder);
case LibFunc::perror:
return optimizeErrorReporting(CI, Builder);
case LibFunc::vfprintf:
case LibFunc::logb:
case LibFunc::sin:
case LibFunc::sinh:
- case LibFunc::tan:
case LibFunc::tanh:
if (UnsafeFPShrink && hasFloatVersion(FuncName))
return optimizeUnaryDoubleFP(CI, Builder, true);
return nullptr;
case LibFunc::copysign:
- case LibFunc::fmin:
- case LibFunc::fmax:
if (hasFloatVersion(FuncName))
return optimizeBinaryDoubleFP(CI, Builder);
return nullptr;
+ case LibFunc::fminf:
+ case LibFunc::fmin:
+ case LibFunc::fminl:
+ case LibFunc::fmaxf:
+ case LibFunc::fmax:
+ case LibFunc::fmaxl:
+ return optimizeFMinFMax(CI, Builder);
default:
return nullptr;
}
Replacer(I, With);
}
-/*static*/ void LibCallSimplifier::replaceAllUsesWithDefault(Instruction *I,
- Value *With) {
- I->replaceAllUsesWith(With);
- I->eraseFromParent();
-}
-
// TODO:
// Additional cases that we need to add to this file:
//
// cbrt:
// * cbrt(expN(X)) -> expN(x/3)
// * cbrt(sqrt(x)) -> pow(x,1/6)
-// * cbrt(sqrt(x)) -> pow(x,1/9)
+// * cbrt(cbrt(x)) -> pow(x,1/9)
//
// exp, expf, expl:
// * exp(log(x)) -> x
projects / oota-llvm.git / blobdiff