diff options
author | Sanjay Patel <spatel@rotateright.com> | 2014-10-23 21:52:45 +0000 |
---|---|---|
committer | Sanjay Patel <spatel@rotateright.com> | 2014-10-23 21:52:45 +0000 |
commit | 848309da7c1861b328d775b28b03d8f687ebeb5c (patch) | |
tree | 8706329fab0f2e6c19524bb82d49a520ac635af6 /llvm/lib/Transforms | |
parent | ecbe7c03a03bcb04b0a2fb185d331b6b6f404322 (diff) | |
download | bcm5719-llvm-848309da7c1861b328d775b28b03d8f687ebeb5c.tar.gz bcm5719-llvm-848309da7c1861b328d775b28b03d8f687ebeb5c.zip |
Handle sqrt() shrinking in SimplifyLibCalls like any other call
This patch removes a chunk of special case logic for folding
(float)sqrt((double)x) -> sqrtf(x)
in InstCombineCasts and handles it in the mainstream path of SimplifyLibCalls.
No functional change intended, but I loosened the restriction on the existing
sqrt testcases to allow for this optimization even without unsafe-fp-math because
that's the existing behavior.
I also added a missing test case for not shrinking the llvm.sqrt.f64 intrinsic
in case the result is used as a double.
Differential Revision: http://reviews.llvm.org/D5919
llvm-svn: 220514
Diffstat (limited to 'llvm/lib/Transforms')
-rw-r--r-- | llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp | 36 | ||||
-rw-r--r-- | llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp | 19 |
2 files changed, 14 insertions, 41 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp index c16992ff626..cbcc85944a7 100644 --- a/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp +++ b/llvm/lib/Transforms/InstCombine/InstCombineCasts.cpp @@ -1317,42 +1317,6 @@ Instruction *InstCombiner::visitFPTrunc(FPTruncInst &CI) { } } - // Fold (fptrunc (sqrt (fpext x))) -> (sqrtf x) - // Note that we restrict this transformation based on - // TLI->has(LibFunc::sqrtf), even for the sqrt intrinsic, because - // TLI->has(LibFunc::sqrtf) is sufficient to guarantee that the - // single-precision intrinsic can be expanded in the backend. - CallInst *Call = dyn_cast<CallInst>(CI.getOperand(0)); - if (Call && Call->getCalledFunction() && TLI->has(LibFunc::sqrtf) && - (Call->getCalledFunction()->getName() == TLI->getName(LibFunc::sqrt) || - Call->getCalledFunction()->getIntrinsicID() == Intrinsic::sqrt) && - Call->getNumArgOperands() == 1 && - Call->hasOneUse()) { - CastInst *Arg = dyn_cast<CastInst>(Call->getArgOperand(0)); - if (Arg && Arg->getOpcode() == Instruction::FPExt && - CI.getType()->isFloatTy() && - Call->getType()->isDoubleTy() && - Arg->getType()->isDoubleTy() && - Arg->getOperand(0)->getType()->isFloatTy()) { - Function *Callee = Call->getCalledFunction(); - Module *M = CI.getParent()->getParent()->getParent(); - Constant *SqrtfFunc = (Callee->getIntrinsicID() == Intrinsic::sqrt) ? - Intrinsic::getDeclaration(M, Intrinsic::sqrt, Builder->getFloatTy()) : - M->getOrInsertFunction("sqrtf", Callee->getAttributes(), - Builder->getFloatTy(), Builder->getFloatTy(), - NULL); - CallInst *ret = CallInst::Create(SqrtfFunc, Arg->getOperand(0), - "sqrtfcall"); - ret->setAttributes(Callee->getAttributes()); - - - // Remove the old Call. With -fmath-errno, it won't get marked readnone. - ReplaceInstUsesWith(*Call, UndefValue::get(Call->getType())); - EraseInstFromFunction(*Call); - return ret; - } - } - return nullptr; } diff --git a/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp b/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp index caae06c4211..207d3664789 100644 --- a/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp +++ b/llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp @@ -1058,7 +1058,16 @@ Value *LibCallSimplifier::optimizeUnaryDoubleFP(CallInst *CI, IRBuilder<> &B, // floor((double)floatval) -> (double)floorf(floatval) Value *V = Cast->getOperand(0); - V = EmitUnaryFloatFnCall(V, Callee->getName(), B, Callee->getAttributes()); + if (Callee->isIntrinsic()) { + Module *M = CI->getParent()->getParent()->getParent(); + Intrinsic::ID IID = (Intrinsic::ID) Callee->getIntrinsicID(); + Function *F = Intrinsic::getDeclaration(M, IID, B.getFloatTy()); + V = B.CreateCall(F, V); + } else { + // The call is a library call rather than an intrinsic. + V = EmitUnaryFloatFnCall(V, Callee->getName(), B, Callee->getAttributes()); + } + return B.CreateFPExt(V, B.getDoubleTy()); } @@ -1086,6 +1095,7 @@ Value *LibCallSimplifier::optimizeBinaryDoubleFP(CallInst *CI, IRBuilder<> &B) { Value *V = nullptr; Value *V1 = Cast1->getOperand(0); Value *V2 = Cast2->getOperand(0); + // TODO: Handle intrinsics in the same way as in optimizeUnaryDoubleFP(). V = EmitBinaryFloatFnCall(V1, V2, Callee->getName(), B, Callee->getAttributes()); return B.CreateFPExt(V, B.getDoubleTy()); @@ -1267,10 +1277,9 @@ Value *LibCallSimplifier::optimizeSqrt(CallInst *CI, IRBuilder<> &B) { Function *Callee = CI->getCalledFunction(); Value *Ret = nullptr; - if (UnsafeFPShrink && Callee->getName() == "sqrt" && - TLI->has(LibFunc::sqrtf)) { + if (TLI->has(LibFunc::sqrtf) && (Callee->getName() == "sqrt" || + Callee->getIntrinsicID() == Intrinsic::sqrt)) Ret = optimizeUnaryDoubleFP(CI, B, true); - } // FIXME: For finer-grain optimization, we need intrinsics to have the same // fast-math flag decorations that are applied to FP instructions. For now, @@ -2010,7 +2019,7 @@ Value *LibCallSimplifier::optimizeCall(CallInst *CI) { UnsafeFPShrink = true; } - // Next check for intrinsics. + // First, check for intrinsics. if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) { if (!isCallingConvC) return nullptr; |