[MLIR][ROCDL] Add conversion of math.erfc to AMD GPU library calls #128899
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
llvmbot
added
backend:AMDGPU
mlir
flang
|
@llvm/pr-subscribers-backend-amdgpu @llvm/pr-subscribers-flang-fir-hlfir Author: Jan Leyonberg (jsjodin) ChangesThis patch adds a pattern to convert the math.erfc operation to AMD GPU library calls. Depends on: #128897 for the flang test Full diff: https://github.com/llvm/llvm-project/pull/128899.diff 3 Files Affected:
diff --git a/flang/test/Lower/OpenMP/math-amdgpu.f90 b/flang/test/Lower/OpenMP/math-amdgpu.f90
index 116768ba9412a..55a14f7d31378 100644
--- a/flang/test/Lower/OpenMP/math-amdgpu.f90
+++ b/flang/test/Lower/OpenMP/math-amdgpu.f90
@@ -99,6 +99,20 @@ subroutine omp_erf_f64(x, y)
y = erf(x)
end subroutine omp_erf_f64
+subroutine omp_erfc_f32(x, y)
+!$omp declare target
+ real :: x, y
+!CHECK: call float @__ocml_erfc_f32(float {{.*}})
+ y = erfc(x)
+end subroutine omp_erfc_f32
+
+subroutine omp_erfc_f64(x, y)
+!$omp declare target
+ real(8) :: x, y
+!CHECK: call double @__ocml_erfc_f64(double {{.*}})
+ y = erfc(x)
+end subroutine omp_erfc_f64
+
subroutine omp_exp_f32(x, y)
!$omp declare target
real :: x, y
diff --git a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
index 838eef30a938f..e065c804507ac 100644
--- a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
+++ b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
@@ -113,6 +113,8 @@ void mlir::populateMathToROCDLConversionPatterns(
"__ocml_tan_f64", "__ocml_tan_f16");
populateOpPatterns<math::ErfOp>(converter, patterns, "__ocml_erf_f32",
"__ocml_erf_f64", "__ocml_erf_f16");
+ populateOpPatterns<math::ErfcOp>(converter, patterns, "__ocml_erfc_f32",
+ "__ocml_erfc_f64", "__ocml_erfc_f16");
populateOpPatterns<math::FPowIOp>(converter, patterns, "__ocml_pown_f32",
"__ocml_pown_f64", "__ocml_pown_f16");
// Single arith pattern that needs a ROCDL call, probably not
diff --git a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
index 313d7b086731e..dbff23339d8b3 100644
--- a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
+++ b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
@@ -462,6 +462,24 @@ module @test_module {
// -----
+module @test_module {
+ // CHECK: llvm.func @__ocml_erfc_f16(f16) -> f16
+ // CHECK: llvm.func @__ocml_erfc_f32(f32) -> f32
+ // CHECK: llvm.func @__ocml_erfc_f64(f64) -> f64
+ // CHECK-LABEL: func @math_erfc
+ func.func @math_erfc(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
+ %result16 = math.erfc %arg_f16 : f16
+ // CHECK: llvm.call @__ocml_erfc_f16(%{{.*}}) : (f16) -> f16
+ %result32 = math.erfc %arg_f32 : f32
+ // CHECK: llvm.call @__ocml_erfc_f32(%{{.*}}) : (f32) -> f32
+ %result64 = math.erfc %arg_f64 : f64
+ // CHECK: llvm.call @__ocml_erfc_f64(%{{.*}}) : (f64) -> f64
+ func.return %result16, %result32, %result64 : f16, f32, f64
+ }
+}
+
+// -----
+
module @test_module {
// CHECK: llvm.func @__ocml_sin_f16(f16) -> f16
// CHECK: llvm.func @__ocml_sin_f32(f32) -> f32
|
|
@llvm/pr-subscribers-mlir Author: Jan Leyonberg (jsjodin) ChangesThis patch adds a pattern to convert the math.erfc operation to AMD GPU library calls. Depends on: #128897 for the flang test Full diff: https://github.com/llvm/llvm-project/pull/128899.diff 3 Files Affected:
diff --git a/flang/test/Lower/OpenMP/math-amdgpu.f90 b/flang/test/Lower/OpenMP/math-amdgpu.f90
index 116768ba9412a..55a14f7d31378 100644
--- a/flang/test/Lower/OpenMP/math-amdgpu.f90
+++ b/flang/test/Lower/OpenMP/math-amdgpu.f90
@@ -99,6 +99,20 @@ subroutine omp_erf_f64(x, y)
y = erf(x)
end subroutine omp_erf_f64
+subroutine omp_erfc_f32(x, y)
+!$omp declare target
+ real :: x, y
+!CHECK: call float @__ocml_erfc_f32(float {{.*}})
+ y = erfc(x)
+end subroutine omp_erfc_f32
+
+subroutine omp_erfc_f64(x, y)
+!$omp declare target
+ real(8) :: x, y
+!CHECK: call double @__ocml_erfc_f64(double {{.*}})
+ y = erfc(x)
+end subroutine omp_erfc_f64
+
subroutine omp_exp_f32(x, y)
!$omp declare target
real :: x, y
diff --git a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
index 838eef30a938f..e065c804507ac 100644
--- a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
+++ b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
@@ -113,6 +113,8 @@ void mlir::populateMathToROCDLConversionPatterns(
"__ocml_tan_f64", "__ocml_tan_f16");
populateOpPatterns<math::ErfOp>(converter, patterns, "__ocml_erf_f32",
"__ocml_erf_f64", "__ocml_erf_f16");
+ populateOpPatterns<math::ErfcOp>(converter, patterns, "__ocml_erfc_f32",
+ "__ocml_erfc_f64", "__ocml_erfc_f16");
populateOpPatterns<math::FPowIOp>(converter, patterns, "__ocml_pown_f32",
"__ocml_pown_f64", "__ocml_pown_f16");
// Single arith pattern that needs a ROCDL call, probably not
diff --git a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
index 313d7b086731e..dbff23339d8b3 100644
--- a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
+++ b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
@@ -462,6 +462,24 @@ module @test_module {
// -----
+module @test_module {
+ // CHECK: llvm.func @__ocml_erfc_f16(f16) -> f16
+ // CHECK: llvm.func @__ocml_erfc_f32(f32) -> f32
+ // CHECK: llvm.func @__ocml_erfc_f64(f64) -> f64
+ // CHECK-LABEL: func @math_erfc
+ func.func @math_erfc(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
+ %result16 = math.erfc %arg_f16 : f16
+ // CHECK: llvm.call @__ocml_erfc_f16(%{{.*}}) : (f16) -> f16
+ %result32 = math.erfc %arg_f32 : f32
+ // CHECK: llvm.call @__ocml_erfc_f32(%{{.*}}) : (f32) -> f32
+ %result64 = math.erfc %arg_f64 : f64
+ // CHECK: llvm.call @__ocml_erfc_f64(%{{.*}}) : (f64) -> f64
+ func.return %result16, %result32, %result64 : f16, f32, f64
+ }
+}
+
+// -----
+
module @test_module {
// CHECK: llvm.func @__ocml_sin_f16(f16) -> f16
// CHECK: llvm.func @__ocml_sin_f32(f32) -> f32
|
|
@llvm/pr-subscribers-flang-openmp Author: Jan Leyonberg (jsjodin) ChangesThis patch adds a pattern to convert the math.erfc operation to AMD GPU library calls. Depends on: #128897 for the flang test Full diff: https://github.com/llvm/llvm-project/pull/128899.diff 3 Files Affected:
diff --git a/flang/test/Lower/OpenMP/math-amdgpu.f90 b/flang/test/Lower/OpenMP/math-amdgpu.f90
index 116768ba9412a..55a14f7d31378 100644
--- a/flang/test/Lower/OpenMP/math-amdgpu.f90
+++ b/flang/test/Lower/OpenMP/math-amdgpu.f90
@@ -99,6 +99,20 @@ subroutine omp_erf_f64(x, y)
y = erf(x)
end subroutine omp_erf_f64
+subroutine omp_erfc_f32(x, y)
+!$omp declare target
+ real :: x, y
+!CHECK: call float @__ocml_erfc_f32(float {{.*}})
+ y = erfc(x)
+end subroutine omp_erfc_f32
+
+subroutine omp_erfc_f64(x, y)
+!$omp declare target
+ real(8) :: x, y
+!CHECK: call double @__ocml_erfc_f64(double {{.*}})
+ y = erfc(x)
+end subroutine omp_erfc_f64
+
subroutine omp_exp_f32(x, y)
!$omp declare target
real :: x, y
diff --git a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
index 838eef30a938f..e065c804507ac 100644
--- a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
+++ b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
@@ -113,6 +113,8 @@ void mlir::populateMathToROCDLConversionPatterns(
"__ocml_tan_f64", "__ocml_tan_f16");
populateOpPatterns<math::ErfOp>(converter, patterns, "__ocml_erf_f32",
"__ocml_erf_f64", "__ocml_erf_f16");
+ populateOpPatterns<math::ErfcOp>(converter, patterns, "__ocml_erfc_f32",
+ "__ocml_erfc_f64", "__ocml_erfc_f16");
populateOpPatterns<math::FPowIOp>(converter, patterns, "__ocml_pown_f32",
"__ocml_pown_f64", "__ocml_pown_f16");
// Single arith pattern that needs a ROCDL call, probably not
diff --git a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
index 313d7b086731e..dbff23339d8b3 100644
--- a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
+++ b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
@@ -462,6 +462,24 @@ module @test_module {
// -----
+module @test_module {
+ // CHECK: llvm.func @__ocml_erfc_f16(f16) -> f16
+ // CHECK: llvm.func @__ocml_erfc_f32(f32) -> f32
+ // CHECK: llvm.func @__ocml_erfc_f64(f64) -> f64
+ // CHECK-LABEL: func @math_erfc
+ func.func @math_erfc(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
+ %result16 = math.erfc %arg_f16 : f16
+ // CHECK: llvm.call @__ocml_erfc_f16(%{{.*}}) : (f16) -> f16
+ %result32 = math.erfc %arg_f32 : f32
+ // CHECK: llvm.call @__ocml_erfc_f32(%{{.*}}) : (f32) -> f32
+ %result64 = math.erfc %arg_f64 : f64
+ // CHECK: llvm.call @__ocml_erfc_f64(%{{.*}}) : (f64) -> f64
+ func.return %result16, %result32, %result64 : f16, f32, f64
+ }
+}
+
+// -----
+
module @test_module {
// CHECK: llvm.func @__ocml_sin_f16(f16) -> f16
// CHECK: llvm.func @__ocml_sin_f32(f32) -> f32
|
krzysz00
approved these changes
Feb 26, 2025
jsjodin
force-pushed
the
jleyonberg/rocdlerfc
branch
from
f8d8b4d to
5ee67e6
Compare
This patch adds a pattern to convert the math.erfc operation to AMD GPU library calls.
jsjodin
force-pushed
the
jleyonberg/rocdlerfc
branch
from
5ee67e6 to
d6d4851
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.
This patch adds a pattern to convert the math.erfc operation to AMD GPU library calls.
Depends on: #128897 for the flang test