AMDGPU: Reduce cost of f64 copysign #141944
Merged
+23
−21
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This was referenced May 29, 2025
This was referenced May 29, 2025
|
@llvm/pr-subscribers-llvm-analysis @llvm/pr-subscribers-backend-amdgpu Author: Matt Arsenault (arsenm) ChangesThe real implementation is 1 real instruction plus a constant Full diff: https://github.com/llvm/llvm-project/pull/141944.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
index 0dbaf7c548f89..c1ccc8f6798a6 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp
@@ -718,9 +718,6 @@ GCNTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
MVT::SimpleValueType SLT = LT.second.getScalarType().SimpleTy;
- if (SLT == MVT::f64)
- return LT.first * NElts * get64BitInstrCost(CostKind);
-
if ((ST->hasVOP3PInsts() && (SLT == MVT::f16 || SLT == MVT::i16)) ||
(ST->hasPackedFP32Ops() && SLT == MVT::f32))
NElts = (NElts + 1) / 2;
@@ -731,6 +728,11 @@ GCNTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
switch (ICA.getID()) {
case Intrinsic::fma:
case Intrinsic::fmuladd:
+ if (SLT == MVT::f64) {
+ InstRate = get64BitInstrCost(CostKind);
+ break;
+ }
+
if ((SLT == MVT::f32 && ST->hasFastFMAF32()) || SLT == MVT::f16)
InstRate = getFullRateInstrCost();
else {
@@ -741,8 +743,8 @@ GCNTTIImpl::getIntrinsicInstrCost(const IntrinsicCostAttributes &ICA,
case Intrinsic::copysign:
return NElts * getFullRateInstrCost();
case Intrinsic::canonicalize: {
- assert(SLT != MVT::f64);
- InstRate = getFullRateInstrCost();
+ InstRate =
+ SLT == MVT::f64 ? get64BitInstrCost(CostKind) : getFullRateInstrCost();
break;
}
case Intrinsic::uadd_sat:
diff --git a/llvm/test/Analysis/CostModel/AMDGPU/copysign.ll b/llvm/test/Analysis/CostModel/AMDGPU/copysign.ll
index 334bb341a3c3e..5b042a8a04603 100644
--- a/llvm/test/Analysis/CostModel/AMDGPU/copysign.ll
+++ b/llvm/test/Analysis/CostModel/AMDGPU/copysign.ll
@@ -245,25 +245,25 @@ define void @copysign_bf16() {
define void @copysign_f64() {
; ALL-LABEL: 'copysign_f64'
-; ALL-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %f64 = call double @llvm.copysign.f64(double undef, double undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v2f64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %v3f64 = call <3 x double> @llvm.copysign.v3f64(<3 x double> undef, <3 x double> undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v4f64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %v5f64 = call <5 x double> @llvm.copysign.v5f64(<5 x double> undef, <5 x double> undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 96 for instruction: %v8f64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 256 for instruction: %v9f64 = call <9 x double> @llvm.copysign.v9f64(<9 x double> undef, <9 x double> undef)
-; ALL-NEXT: Cost Model: Found an estimated cost of 320 for instruction: %v16f64 = call <16 x double> @llvm.copysign.v16f64(<16 x double> undef, <16 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call double @llvm.copysign.f64(double undef, double undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v3f64 = call <3 x double> @llvm.copysign.v3f64(<3 x double> undef, <3 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v4f64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v5f64 = call <5 x double> @llvm.copysign.v5f64(<5 x double> undef, <5 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v8f64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v9f64 = call <9 x double> @llvm.copysign.v9f64(<9 x double> undef, <9 x double> undef)
+; ALL-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v16f64 = call <16 x double> @llvm.copysign.v16f64(<16 x double> undef, <16 x double> undef)
; ALL-NEXT: Cost Model: Found an estimated cost of 10 for instruction: ret void
;
; ALL-SIZE-LABEL: 'copysign_f64'
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %f64 = call double @llvm.copysign.f64(double undef, double undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v2f64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %v3f64 = call <3 x double> @llvm.copysign.v3f64(<3 x double> undef, <3 x double> undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v4f64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %v5f64 = call <5 x double> @llvm.copysign.v5f64(<5 x double> undef, <5 x double> undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %v8f64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %v9f64 = call <9 x double> @llvm.copysign.v9f64(<9 x double> undef, <9 x double> undef)
-; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 160 for instruction: %v16f64 = call <16 x double> @llvm.copysign.v16f64(<16 x double> undef, <16 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %f64 = call double @llvm.copysign.f64(double undef, double undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %v2f64 = call <2 x double> @llvm.copysign.v2f64(<2 x double> undef, <2 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %v3f64 = call <3 x double> @llvm.copysign.v3f64(<3 x double> undef, <3 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %v4f64 = call <4 x double> @llvm.copysign.v4f64(<4 x double> undef, <4 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v5f64 = call <5 x double> @llvm.copysign.v5f64(<5 x double> undef, <5 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %v8f64 = call <8 x double> @llvm.copysign.v8f64(<8 x double> undef, <8 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v9f64 = call <9 x double> @llvm.copysign.v9f64(<9 x double> undef, <9 x double> undef)
+; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %v16f64 = call <16 x double> @llvm.copysign.v16f64(<16 x double> undef, <16 x double> undef)
; ALL-SIZE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: ret void
;
%f64 = call double @llvm.copysign.f64(double undef, double undef)
|
Pierre-vh
reviewed
Jun 2, 2025
| assert(SLT != MVT::f64); | ||
| InstRate = getFullRateInstrCost(); | ||
| InstRate = | ||
| SLT == MVT::f64 ? get64BitInstrCost(CostKind) : getFullRateInstrCost(); | ||
| break; | ||
| } | ||
| case Intrinsic::uadd_sat: |
There was a problem hiding this comment.
are those cases below fine with handling f64 now?
There was a problem hiding this comment.
They are only integer intrinsics
Pierre-vh
approved these changes
Jun 4, 2025
arsenm
force-pushed
the
users/arsenm/amdgpu/fix-cost-of-16-bit-ops
branch
from
017304b to
d990f79
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/reduce-cost-f64-copysign
branch
from
19ab42a to
598db89
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/fix-cost-of-16-bit-ops
branch
from
d990f79 to
e4fcaab
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/reduce-cost-f64-copysign
branch
from
598db89 to
0ddc81d
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/fix-cost-of-16-bit-ops
branch
from
e4fcaab to
7fbe4e2
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/reduce-cost-f64-copysign
branch
from
0ddc81d to
641ab37
Compare
arsenm
force-pushed
the
users/arsenm/amdgpu/fix-cost-of-16-bit-ops
branch
3 times, most recently
from
5fdd877 to
90d3969
Compare
Base automatically changed from
users/arsenm/amdgpu/fix-cost-of-16-bit-ops
to
main
June 17, 2025 23:07
The real implementation is 1 real instruction plus a constant materialize. Call that a 1, it's not a real f64 operation.
arsenm
force-pushed
the
users/arsenm/amdgpu/reduce-cost-f64-copysign
branch
from
641ab37 to
9f4bc8d
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The real implementation is 1 real instruction plus a constant
materialize. Call that a 1, it's not a real f64 operation.