summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
diff options
context:
space:
mode:
authorMatt Arsenault <Matthew.Arsenault@amd.com>2019-10-31 18:50:30 -0700
committerMatt Arsenault <arsenm2@gmail.com>2019-11-19 19:55:43 +0530
commitdb0ed3e429b55d1730d1ecc253b0643de7fca099 (patch)
tree33f725b9778863f7c737075b4ca9d9e6c43ceb08 /llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
parentea23b6428b88ed50a2cfc91b783f627fa139bb36 (diff)
downloadbcm5719-llvm-db0ed3e429b55d1730d1ecc253b0643de7fca099.tar.gz
bcm5719-llvm-db0ed3e429b55d1730d1ecc253b0643de7fca099.zip
AMDGPU: Refactor treatment of denormal mode
Start moving towards treating this as a property of the calling convention, and not the subtarget. The default denormal mode should not be part of the subtarget, and be moved into a separate function attribute. This patch is still NFC. The denormal mode remains as a subtarget feature for now, but make the necessary changes to switch to using an attribute.
Diffstat (limited to 'llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp')
-rw-r--r--llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp7
1 files changed, 4 insertions, 3 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
index ed495bd40b8..cf908766caa 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUCodeGenPrepare.cpp
@@ -70,6 +70,7 @@ class AMDGPUCodeGenPrepare : public FunctionPass,
Module *Mod = nullptr;
const DataLayout *DL = nullptr;
bool HasUnsafeFPMath = false;
+ bool HasFP32Denormals = false;
/// Copies exact/nsw/nuw flags (if any) from binary operation \p I to
/// binary operation \p V.
@@ -575,7 +576,6 @@ bool AMDGPUCodeGenPrepare::visitFDiv(BinaryOperator &FDiv) {
Value *NewFDiv = nullptr;
- bool HasDenormals = ST->hasFP32Denormals();
if (VectorType *VT = dyn_cast<VectorType>(Ty)) {
NewFDiv = UndefValue::get(VT);
@@ -586,7 +586,7 @@ bool AMDGPUCodeGenPrepare::visitFDiv(BinaryOperator &FDiv) {
Value *DenEltI = Builder.CreateExtractElement(Den, I);
Value *NewElt;
- if (shouldKeepFDivF32(NumEltI, UnsafeDiv, HasDenormals)) {
+ if (shouldKeepFDivF32(NumEltI, UnsafeDiv, HasFP32Denormals)) {
NewElt = Builder.CreateFDiv(NumEltI, DenEltI);
} else {
NewElt = Builder.CreateCall(Decl, { NumEltI, DenEltI });
@@ -595,7 +595,7 @@ bool AMDGPUCodeGenPrepare::visitFDiv(BinaryOperator &FDiv) {
NewFDiv = Builder.CreateInsertElement(NewFDiv, NewElt, I);
}
} else {
- if (!shouldKeepFDivF32(Num, UnsafeDiv, HasDenormals))
+ if (!shouldKeepFDivF32(Num, UnsafeDiv, HasFP32Denormals))
NewFDiv = Builder.CreateCall(Decl, { Num, Den });
}
@@ -1034,6 +1034,7 @@ bool AMDGPUCodeGenPrepare::runOnFunction(Function &F) {
AC = &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
DA = &getAnalysis<LegacyDivergenceAnalysis>();
HasUnsafeFPMath = hasUnsafeFPMath(F);
+ HasFP32Denormals = ST->hasFP32Denormals(F);
bool MadeChange = false;
OpenPOWER on IntegriCloud