[MLIR][ROCDL] Add math.clampf -> rocdl.fmed3 conversion #163259
Merged
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
Signed-off-by: Keshav Vinayak Jha <[email protected]>
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-gpu Author: Keshav Vinayak Jha (keshavvinayak01) ChangesAdded Pattern for lowering Solves #15072 Patch is 52.62 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/163259.diff 5 Files Affected:
diff --git a/mlir/include/mlir/Conversion/MathToROCDL/MathToROCDL.h b/mlir/include/mlir/Conversion/MathToROCDL/MathToROCDL.h
index 46573e7966ccc..770f257d89bd5 100644
--- a/mlir/include/mlir/Conversion/MathToROCDL/MathToROCDL.h
+++ b/mlir/include/mlir/Conversion/MathToROCDL/MathToROCDL.h
@@ -9,6 +9,7 @@
#define MLIR_CONVERSION_MATHTOROCDL_MATHTOROCDL_H_
#include "mlir/Conversion/LLVMCommon/TypeConverter.h"
+#include "mlir/Dialect/AMDGPU/Utils/Chipset.h"
#include "mlir/IR/PatternMatch.h"
#include <memory>
@@ -20,7 +21,8 @@ class Pass;
/// Populate the given list with patterns that convert from Math to ROCDL calls.
void populateMathToROCDLConversionPatterns(const LLVMTypeConverter &converter,
- RewritePatternSet &patterns);
+ RewritePatternSet &patterns,
+ amdgpu::Chipset chipset);
} // namespace mlir
#endif // MLIR_CONVERSION_MATHTOROCDL_MATHTOROCDL_H_
diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td
index 3c18ecc753d0f..c3fd397e258ae 100644
--- a/mlir/include/mlir/Conversion/Passes.td
+++ b/mlir/include/mlir/Conversion/Passes.td
@@ -755,6 +755,14 @@ def ConvertMathToLibmPass : Pass<"convert-math-to-libm", "ModuleOp"> {
"func::FuncDialect",
"vector::VectorDialect",
];
+ let options = [
+ Option<"chipset", "chipset", "std::string",
+
+
+ /*default=*/"\"gfx000\"",
+ "Chipset that these operations will run on">
+ ];
+
}
//===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
index b215211e131d4..c03f3a5d3889c 100644
--- a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
+++ b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
@@ -484,5 +484,5 @@ void mlir::populateGpuToROCDLConversionPatterns(
GPUSubgroupBroadcastOpToROCDL>(converter);
patterns.add<GPUSubgroupSizeOpToROCDL>(converter, chipset);
- populateMathToROCDLConversionPatterns(converter, patterns);
+ populateMathToROCDLConversionPatterns(converter, patterns, chipset);
}
diff --git a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
index df219f3ff4f6e..ceb3d22c6bd59 100644
--- a/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
+++ b/mlir/lib/Conversion/MathToROCDL/MathToROCDL.cpp
@@ -10,6 +10,7 @@
#include "mlir/Conversion/GPUCommon/GPUCommonPass.h"
#include "mlir/Conversion/LLVMCommon/LoweringOptions.h"
#include "mlir/Conversion/LLVMCommon/TypeConverter.h"
+#include "mlir/Dialect/AMDGPU/Utils/Chipset.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "mlir/Dialect/LLVMIR/ROCDLDialect.h"
@@ -42,8 +43,39 @@ static void populateOpPatterns(const LLVMTypeConverter &converter,
f32ApproxFunc, f16Func);
}
+struct ClampFOpConversion final
+ : public ConvertOpToLLVMPattern<math::ClampFOp> {
+ using ConvertOpToLLVMPattern::ConvertOpToLLVMPattern;
+ ClampFOpConversion(const LLVMTypeConverter &converter,
+ amdgpu::Chipset chipset)
+ : ConvertOpToLLVMPattern<math::ClampFOp>(converter), chipset(chipset) {}
+
+ LogicalResult
+ matchAndRewrite(math::ClampFOp op, OpAdaptor adaptor,
+ ConversionPatternRewriter &rewriter) const override {
+ // V_MED3_F16/F32 only exists in gfx9+ artchitectures
+ if (chipset.majorVersion < 9) {
+ return rewriter.notifyMatchFailure(
+ op, ("pre-gfx9 (gfx" + std::to_string(chipset.majorVersion) +
+ "): V_MED_F16 / V_MED3_F32 not supported."));
+ }
+ rewriter.replaceOpWithNewOp<ROCDL::FMed3Op>(op, op.getType(), op.getValue(),
+ op.getMin(), op.getMax());
+ return success();
+ }
+ amdgpu::Chipset chipset;
+};
+
+static void addChipsetDependentPatterns(const LLVMTypeConverter &converter,
+ RewritePatternSet &patterns,
+ amdgpu::Chipset chipset) {
+
+ patterns.add<ClampFOpConversion>(converter, chipset);
+}
+
void mlir::populateMathToROCDLConversionPatterns(
- const LLVMTypeConverter &converter, RewritePatternSet &patterns) {
+ const LLVMTypeConverter &converter, RewritePatternSet &patterns,
+ amdgpu::Chipset chipset) {
// Handled by mathToLLVM: math::AbsIOp
// Handled by mathToLLVM: math::AbsFOp
// Handled by mathToLLVM: math::CopySignOp
@@ -118,27 +150,31 @@ void mlir::populateMathToROCDLConversionPatterns(
// worth creating a separate pass for it.
populateOpPatterns<arith::RemFOp>(converter, patterns, "__ocml_fmod_f32",
"__ocml_fmod_f64", "__ocml_fmod_f16");
+
+ addChipsetDependentPatterns(converter, patterns, chipset);
}
-namespace {
-struct ConvertMathToROCDLPass
- : public impl::ConvertMathToROCDLBase<ConvertMathToROCDLPass> {
- ConvertMathToROCDLPass() = default;
+struct ConvertMathToROCDLPass final
+ : impl::ConvertMathToROCDLBase<ConvertMathToROCDLPass> {
+ using impl::ConvertMathToROCDLBase<
+ ConvertMathToROCDLPass>::ConvertMathToROCDLBase;
+
void runOnOperation() override;
};
-} // namespace
void ConvertMathToROCDLPass::runOnOperation() {
auto m = getOperation();
MLIRContext *ctx = m.getContext();
+ FailureOr<amdgpu::Chipset> maybeChipset = amdgpu::Chipset::parse(chipset);
RewritePatternSet patterns(&getContext());
LowerToLLVMOptions options(ctx, DataLayout(m));
LLVMTypeConverter converter(ctx, options);
- populateMathToROCDLConversionPatterns(converter, patterns);
+ populateMathToROCDLConversionPatterns(converter, patterns, *maybeChipset);
ConversionTarget target(getContext());
- target.addLegalDialect<BuiltinDialect, func::FuncDialect,
- vector::VectorDialect, LLVM::LLVMDialect>();
+ target
+ .addLegalDialect<BuiltinDialect, func::FuncDialect, vector::VectorDialect,
+ LLVM::LLVMDialect, ROCDL::ROCDLDialect>();
target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::Exp2Op, LLVM::FAbsOp,
LLVM::FCeilOp, LLVM::FFloorOp, LLVM::FRemOp, LLVM::LogOp,
LLVM::Log10Op, LLVM::Log2Op, LLVM::PowOp, LLVM::SinOp,
diff --git a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
index dbff23339d8b3..29851e2de5cb2 100644
--- a/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
+++ b/mlir/test/Conversion/MathToROCDL/math-to-rocdl.mlir
@@ -1,18 +1,40 @@
-// RUN: mlir-opt %s -convert-math-to-rocdl -allow-unregistered-dialect -split-input-file | FileCheck %s
+// RUN: mlir-opt %s -allow-unregistered-dialect -split-input-file
+// -pass-pipeline='builtin.module(convert-math-to-rocdl{chipset=gfx803})' |
+// FileCheck %s --check-prefix=PRE9 RUN: mlir-opt %s -allow-unregistered-dialect
+// -split-input-file
+// -pass-pipeline='builtin.module(convert-math-to-rocdl{chipset=gfx942})' |
+// FileCheck %s --check-prefix=POST9
module @test_module {
// CHECK: llvm.func @__ocml_fmod_f16(f16, f16) -> f16
// CHECK: llvm.func @__ocml_fmod_f32(f32, f32) -> f32
// CHECK: llvm.func @__ocml_fmod_f64(f64, f64) -> f64
// CHECK-LABEL: func @arith_remf
- func.func @arith_remf(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = arith.remf %arg_f16, %arg_f16 : f16
- // CHECK: llvm.call @__ocml_fmod_f16(%{{.*}}, %{{.*}}) : (f16, f16) -> f16
- %result32 = arith.remf %arg_f32, %arg_f32 : f32
- // CHECK: llvm.call @__ocml_fmod_f32(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
- %result64 = arith.remf %arg_f64, %arg_f64 : f64
- // CHECK: llvm.call @__ocml_fmod_f64(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @arith_remf(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = arith.remf % arg_f16,
+ %
+ arg_f16 : f16
+ // CHECK: llvm.call @__ocml_fmod_f16(%{{.*}}, %{{.*}}) :
+ // (f16, f16) -> f16
+ %
+ result32 = arith.remf % arg_f32,
+ %
+ arg_f32 : f32
+ // CHECK: llvm.call @__ocml_fmod_f32(%{{.*}}, %{{.*}}) :
+ // (f32, f32) -> f32
+ %
+ result64 = arith.remf % arg_f64,
+ %
+ arg_f64 : f64
+ // CHECK: llvm.call @__ocml_fmod_f64(%{{.*}}, %{{.*}}) :
+ // (f64, f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -23,14 +45,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_acos_f32(f32) -> f32
// CHECK: llvm.func @__ocml_acos_f64(f64) -> f64
// CHECK-LABEL: func @math_acos
- func.func @math_acos(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.acos %arg_f16 : f16
- // CHECK: llvm.call @__ocml_acos_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.acos %arg_f32 : f32
- // CHECK: llvm.call @__ocml_acos_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.acos %arg_f64 : f64
- // CHECK: llvm.call @__ocml_acos_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_acos(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.acos %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_acos_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.acos %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_acos_f32(%{{.*}}) : (f32) -> f32
+ %
+ result64 = math.acos %
+ arg_f64
+ : f64
+ // CHECK: llvm.call @__ocml_acos_f64(%{{.*}}) : (f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -41,14 +77,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_acosh_f32(f32) -> f32
// CHECK: llvm.func @__ocml_acosh_f64(f64) -> f64
// CHECK-LABEL: func @math_acosh
- func.func @math_acosh(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.acosh %arg_f16 : f16
- // CHECK: llvm.call @__ocml_acosh_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.acosh %arg_f32 : f32
- // CHECK: llvm.call @__ocml_acosh_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.acosh %arg_f64 : f64
- // CHECK: llvm.call @__ocml_acosh_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_acosh(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.acosh %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_acosh_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.acosh %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_acosh_f32(%{{.*}}) : (f32) -> f32
+ %
+ result64 = math.acosh %
+ arg_f64
+ : f64
+ // CHECK: llvm.call @__ocml_acosh_f64(%{{.*}}) : (f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -59,14 +109,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_asin_f32(f32) -> f32
// CHECK: llvm.func @__ocml_asin_f64(f64) -> f64
// CHECK-LABEL: func @math_asin
- func.func @math_asin(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.asin %arg_f16 : f16
- // CHECK: llvm.call @__ocml_asin_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.asin %arg_f32 : f32
- // CHECK: llvm.call @__ocml_asin_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.asin %arg_f64 : f64
- // CHECK: llvm.call @__ocml_asin_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_asin(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.asin %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_asin_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.asin %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_asin_f32(%{{.*}}) : (f32) -> f32
+ %
+ result64 = math.asin %
+ arg_f64
+ : f64
+ // CHECK: llvm.call @__ocml_asin_f64(%{{.*}}) : (f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -77,14 +141,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_asinh_f32(f32) -> f32
// CHECK: llvm.func @__ocml_asinh_f64(f64) -> f64
// CHECK-LABEL: func @math_asinh
- func.func @math_asinh(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.asinh %arg_f16 : f16
- // CHECK: llvm.call @__ocml_asinh_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.asinh %arg_f32 : f32
- // CHECK: llvm.call @__ocml_asinh_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.asinh %arg_f64 : f64
- // CHECK: llvm.call @__ocml_asinh_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_asinh(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.asinh %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_asinh_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.asinh %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_asinh_f32(%{{.*}}) : (f32) -> f32
+ %
+ result64 = math.asinh %
+ arg_f64
+ : f64
+ // CHECK: llvm.call @__ocml_asinh_f64(%{{.*}}) : (f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -95,14 +173,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_atan_f32(f32) -> f32
// CHECK: llvm.func @__ocml_atan_f64(f64) -> f64
// CHECK-LABEL: func @math_atan
- func.func @math_atan(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.atan %arg_f16 : f16
- // CHECK: llvm.call @__ocml_atan_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.atan %arg_f32 : f32
- // CHECK: llvm.call @__ocml_atan_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.atan %arg_f64 : f64
- // CHECK: llvm.call @__ocml_atan_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_atan(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.atan %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_atan_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.atan %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_atan_f32(%{{.*}}) : (f32) -> f32
+ %
+ result64 = math.atan %
+ arg_f64
+ : f64
+ // CHECK: llvm.call @__ocml_atan_f64(%{{.*}}) : (f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -113,14 +205,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_atanh_f32(f32) -> f32
// CHECK: llvm.func @__ocml_atanh_f64(f64) -> f64
// CHECK-LABEL: func @math_atanh
- func.func @math_atanh(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.atanh %arg_f16 : f16
- // CHECK: llvm.call @__ocml_atanh_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.atanh %arg_f32 : f32
- // CHECK: llvm.call @__ocml_atanh_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.atanh %arg_f64 : f64
- // CHECK: llvm.call @__ocml_atanh_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_atanh(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.atanh %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_atanh_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.atanh %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_atanh_f32(%{{.*}}) : (f32) -> f32
+ %
+ result64 = math.atanh %
+ arg_f64
+ : f64
+ // CHECK: llvm.call @__ocml_atanh_f64(%{{.*}}) : (f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -131,14 +237,31 @@ module @test_module {
// CHECK: llvm.func @__ocml_atan2_f32(f32, f32) -> f32
// CHECK: llvm.func @__ocml_atan2_f64(f64, f64) -> f64
// CHECK-LABEL: func @math_atan2
- func.func @math_atan2(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.atan2 %arg_f16, %arg_f16 : f16
- // CHECK: llvm.call @__ocml_atan2_f16(%{{.*}}, %{{.*}}) : (f16, f16) -> f16
- %result32 = math.atan2 %arg_f32, %arg_f32 : f32
- // CHECK: llvm.call @__ocml_atan2_f32(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
- %result64 = math.atan2 %arg_f64, %arg_f64 : f64
- // CHECK: llvm.call @__ocml_atan2_f64(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_atan2(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.atan2 % arg_f16,
+ %
+ arg_f16 : f16
+ // CHECK: llvm.call @__ocml_atan2_f16(%{{.*}}, %{{.*}}) :
+ // (f16, f16) -> f16
+ %
+ result32 = math.atan2 % arg_f32,
+ %
+ arg_f32 : f32
+ // CHECK: llvm.call @__ocml_atan2_f32(%{{.*}}, %{{.*}}) :
+ // (f32, f32) -> f32
+ %
+ result64 = math.atan2 % arg_f64,
+ %
+ arg_f64 : f64
+ // CHECK: llvm.call @__ocml_atan2_f64(%{{.*}}, %{{.*}})
+ // : (f64, f64) -> f64
+ func.return %
+ result16,
+ % result32, % result64 : f16, f32, f64
}
}
@@ -149,14 +272,28 @@ module @test_module {
// CHECK: llvm.func @__ocml_cbrt_f32(f32) -> f32
// CHECK: llvm.func @__ocml_cbrt_f64(f64) -> f64
// CHECK-LABEL: func @math_cbrt
- func.func @math_cbrt(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
- %result16 = math.cbrt %arg_f16 : f16
- // CHECK: llvm.call @__ocml_cbrt_f16(%{{.*}}) : (f16) -> f16
- %result32 = math.cbrt %arg_f32 : f32
- // CHECK: llvm.call @__ocml_cbrt_f32(%{{.*}}) : (f32) -> f32
- %result64 = math.cbrt %arg_f64 : f64
- // CHECK: llvm.call @__ocml_cbrt_f64(%{{.*}}) : (f64) -> f64
- func.return %result16, %result32, %result64 : f16, f32, f64
+ func.func @math_cbrt(% arg_f16
+ : f16, % arg_f32
+ : f32, % arg_f64
+ : f64)
+ ->(f16, f32, f64) {
+ % result16 = math.cbrt %
+ arg_f16
+ : f16
+ // CHECK: llvm.call @__ocml_cbrt_f16(%{{.*}}) : (f16) -> f16
+ %
+ result32 = math.cbrt %
+ arg_f32
+ : f32
+ // CHECK: llvm.call @__ocml_cbrt_f32(%{{.*}}) : (f...
[truncated]
|
Signed-off-by: Keshav Vinayak Jha <[email protected]>
fabianmcg
reviewed
Oct 13, 2025
|
To reference @krzysz00 comments:
Applied changes as per other suggestions. |
fabianmcg
changed the title
Signed-off-by: Keshav Vinayak Jha <[email protected]>
krzysz00
reviewed
Oct 13, 2025
fabianmcg
reviewed
Oct 14, 2025
1. Added lit test for 1D and 2D vectors 2. Added unrolling support for ND inputs Signed-off-by: Keshav Vinayak Jha <[email protected]>
Signed-off-by: Keshav Vinayak Jha <[email protected]>
Signed-off-by: Keshav Vinayak Jha <[email protected]>
Signed-off-by: Keshav Vinayak Jha <[email protected]>
Signed-off-by: Keshav Vinayak Jha <[email protected]>
krzysz00
reviewed
Oct 14, 2025
Signed-off-by: Keshav Vinayak Jha <[email protected]>
krzysz00
approved these changes
Oct 14, 2025
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/205/builds/24997 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/204/builds/25020 Here is the relevant piece of the build log for the reference |
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/203/builds/26208 Here is the relevant piece of the build log for the reference |
fabianmcg
added a commit
that referenced
this pull request
Oct 14, 2025
)" This reverts commit 1e6df64.
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/138/builds/20429 Here is the relevant piece of the build log for the reference |
fabianmcg
reviewed
Oct 14, 2025
Comment on lines
+201
to
+207
| if (!chipset.empty()) { | ||
| FailureOr<amdgpu::Chipset> maybeChipset = amdgpu::Chipset::parse(chipset); | ||
| if (failed(maybeChipset)) { | ||
| return signalPassFailure(); | ||
| } | ||
| populateMathToROCDLConversionPatterns(converter, patterns, *maybeChipset); | ||
| } |
There was a problem hiding this comment.
This is wrong, the call to populateMathToROCDLConversionPatterns shouldn't be guarded by an if.
llvm-sync bot
pushed a commit
to arm/arm-toolchain
that referenced
this pull request
Oct 14, 2025
…rsion" (#163447) Reverts llvm/llvm-project#163259. Reverting due to missing link libraries causing failures in shared build bots.
akadutta
pushed a commit
to akadutta/llvm-project
that referenced
this pull request
Oct 14, 2025
Added Pattern for lowering `Math::ClampFOp` to `ROCDL::FMED3`. Also added `chipset` option to `MathToRocdl` pass to check for arch support ISA instructions Solves [llvm#15072](llvm#157052) Reapplies llvm#160100 --------- Signed-off-by: Keshav Vinayak Jha <[email protected]>
akadutta
pushed a commit
to akadutta/llvm-project
that referenced
this pull request
Oct 14, 2025
…m#163447) Reverts llvm#163259. Reverting due to missing link libraries causing failures in shared build bots.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
Added Pattern for lowering
Math::ClampFOptoROCDL::FMED3.Also added
chipetoption toMathToRocdlpass to check for arch support ISA instructionsSolves #15072
Reapplies #160100