Skip to content

[SYCL] Enable FTZ operations for CUDA/PTX backend via -fcuda-flush-denormals-to-zero #6411

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
steffenlarsen merged 6 commits into from
Aug 23, 2022
Merged

[SYCL] Enable FTZ operations for CUDA/PTX backend via -fcuda-flush-denormals-to-zero #6411

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
f7f2faa
Use -fcuda-flush-denormals-to-zero for SYCL programs
P-Jensen Feb 15, 2022
d78d213
Fixed syntax error
P-Jensen Jun 29, 2022
2b0c3dc
Merge branch 'ftz' into cuda-ftz
P-Jensen Jul 6, 2022
01185af
Added test: clang/test/Driver/ftz-cuda.c
P-Jensen Jul 7, 2022
77c2fd9
Added CodeGenSYCL test
P-Jensen Aug 22, 2022
e28b133
Fixed a couple of test cases that were failing
P-Jensen Aug 23, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 5 additions & 3 deletions clang/lib/CodeGen/CodeGenModule.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -815,13 +815,15 @@ void CodeGenModule::Release() {
llvm::MDString::get(Ctx, CodeGenOpts.MemoryProfileOutput));
}

if ((LangOpts.CUDAIsDevice || LangOpts.isSYCL()) && getTriple().isNVPTX()) {
if ((LangOpts.CUDAIsDevice || LangOpts.SYCLIsDevice) && getTriple().isNVPTX()) {
// Indicate whether __nvvm_reflect should be configured to flush denormal
// floating point values to 0. (This corresponds to its "__CUDA_FTZ"
// property.)
getModule().addModuleFlag(llvm::Module::Override, "nvvm-reflect-ftz",
CodeGenOpts.FP32DenormalMode.Output !=
llvm::DenormalMode::IEEE);
(CodeGenOpts.FP32DenormalMode.Output !=
llvm::DenormalMode::IEEE) ||
(CodeGenOpts.FPDenormalMode.Output !=
llvm::DenormalMode::IEEE));
getModule().addModuleFlag(llvm::Module::Override, "nvvm-reflect-prec-sqrt",
getTarget().getTargetOpts().NVVMCudaPrecSqrt);
}
Expand Down
5 changes: 5 additions & 0 deletions clang/lib/Driver/ToolChains/Clang.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3003,6 +3003,11 @@ static void RenderFloatingPointOptions(const ToolChain &TC, const Driver &D,
RoundingMathPresent = false;
break;

case options::OPT_fcuda_flush_denormals_to_zero:
case options::OPT_fgpu_flush_denormals_to_zero:
DenormalFP32Math = llvm::DenormalMode::getPreserveSign();
break;
Copy link
Contributor

@mdtoguchi mdtoguchi Jul 6, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could you add a driver test verifying the expect values when the options are set?

Copy link
Contributor Author

@P-Jensen P-Jensen Jul 7, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@mdtoguchi Do you prefer a separate test for this, or should I add it to the clang/test/Driver/fp-model.c?


case options::OPT_fdenormal_fp_math_EQ:
DenormalFPMath = llvm::parseDenormalFPAttribute(A->getValue());
DenormalFP32Math = DenormalFPMath;
Expand Down
18 changes: 18 additions & 0 deletions clang/test/CodeGenSYCL/flush-denormals.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
// RUN: %clang_cc1 -fcuda-is-device -fdenormal-fp-math-f32=preserve-sign \
// RUN: -triple nvptx-nvidia-cuda -emit-llvm -o - %s | \
// RUN: FileCheck -check-prefixes=FTZ32,PTXFTZ32 %s

// RUN: %clang_cc1 -fcuda-is-device -fdenormal-fp-math=preserve-sign \
// RUN: -triple nvptx-nvidia-cuda -emit-llvm -o - %s | \
// RUN: FileCheck -check-prefixes=FTZ,PTXFTZ %s

// CHECK-LABEL: define void @_Z3foov() #0
void foo() {}

// FTZ32: attributes #0 = {{.*}} "denormal-fp-math-f32"="preserve-sign,preserve-sign"
// PTXFTZ32:!llvm.module.flags = !{{{.*}}, [[MODFLAG:![0-9]+]], {{.*}}}
// PTXFTZ32:[[MODFLAG]] = !{i32 4, !"nvvm-reflect-ftz", i32 1}

// FTZ: attributes #0 = {{.*}} "denormal-fp-math"="preserve-sign,preserve-sign"
// PTXFTZ:!llvm.module.flags = !{{{.*}}, [[MODFLAG:![0-9]+]], {{.*}}}
// PTXFTZ:[[MODFLAG]] = !{i32 4, !"nvvm-reflect-ftz", i32 1}
4 changes: 4 additions & 0 deletions clang/test/Driver/ftz-cuda.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,4 @@
// RUN: %clang -### -fcuda-flush-denormals-to-zero -c %s 2>&1 \
// RUN: | FileCheck --check-prefix=CHECK-FTZ %s
// CHECK-FTZ: "-cc1"
// CHECK-FTZ: "-fdenormal-fp-math-f32=preserve-sign,preserve-sign"