Skip to content

[AMDGPU] Add a target option to disable aggressive FMA fusion #161450

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

Draft
wants to merge 3 commits into
base: main
Choose a base branch
from
Draft

[AMDGPU] Add a target option to disable aggressive FMA fusion #161450

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
aab5e41
[AMDGPU] Add a target option to disable aggressive FMA fusion
yiqian1 Sep 30, 2025
9a47421
Remove an old change
yiqian1 Oct 2, 2025
ada0bb9
Remove an unneeded check
yiqian1 Oct 2, 2025
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
7 changes: 7 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPU.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1238,6 +1238,13 @@ def FeatureSetPrioIncWgInst : SubtargetFeature<"setprio-inc-wg-inst",
// Subtarget Features (options and debugging)
//===------------------------------------------------------------===//

def FeatureDisableAggressiveFMAFusion : SubtargetFeature<
"disable-aggressive-fma-fusion",
"DisableAggressiveFMAFusion",
"true",
"Do not fold fmul and fadd/fsub into fma."
>;

// Ugly hack to accomodate assembling modules with mixed
// wavesizes. Ideally we would have a mapping symbol in assembly which
// would keep track of which sections of code should be treated as
Expand Down
5 changes: 5 additions & 0 deletions llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,6 +76,7 @@ class AMDGPUSubtarget {
bool EnablePromoteAlloca = false;
bool HasTrigReducedRange = false;
bool FastFMAF32 = false;
bool DisableAggressiveFMAFusion = false;
unsigned EUsPerCU = 4;
unsigned MaxWavesPerEU = 10;
unsigned LocalMemorySize = 0;
Expand Down Expand Up @@ -303,6 +304,10 @@ class AMDGPUSubtarget {
return FastFMAF32;
}

bool hasDisableAggressiveFMAFusion() const {
return DisableAggressiveFMAFusion;
}

bool isPromoteAllocaEnabled() const {
return EnablePromoteAlloca;
}
Expand Down
1 change: 1 addition & 0 deletions llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -288,6 +288,7 @@ const FeatureBitset GCNTTIImpl::InlineFeatureIgnoreList = {
AMDGPU::FeatureEnableUnsafeDSOffsetFolding, AMDGPU::FeatureFlatForGlobal,
AMDGPU::FeaturePromoteAlloca, AMDGPU::FeatureUnalignedScratchAccess,
AMDGPU::FeatureUnalignedAccessMode,
AMDGPU::FeatureDisableAggressiveFMAFusion,

AMDGPU::FeatureAutoWaitcntBeforeBarrier,

Expand Down
6 changes: 4 additions & 2 deletions llvm/lib/Target/AMDGPU/SIISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6502,10 +6502,12 @@ bool SITargetLowering::enableAggressiveFMAFusion(EVT VT) const {
// When fma is quarter rate, for f64 where add / sub are at best half rate,
// most of these combines appear to be cycle neutral but save on instruction
// count / code size.
return true;
return !Subtarget->hasDisableAggressiveFMAFusion();
}

bool SITargetLowering::enableAggressiveFMAFusion(LLT Ty) const { return true; }
bool SITargetLowering::enableAggressiveFMAFusion(LLT Ty) const {
return !Subtarget->hasDisableAggressiveFMAFusion();
}

EVT SITargetLowering::getSetCCResultType(const DataLayout &DL, LLVMContext &Ctx,
EVT VT) const {
Expand Down