[mlir][Math][LLVMSPV] Add Math to LLVM-SPV conversion pass by akhilgoe · Pull Request #198370 · llvm/llvm-project

akhilgoe · 2026-05-18T18:33:09Z

This PR adds a conversion pass with patterns to convert supported math ops to SPIR-V builtin calls for the chosen extended instruction set. For OpenCL these lowerings correspond to OpExtInst calls into the OpenCL SPIR-V extended instruction set via mangled __spirv_ocl_ entry points for f32/f64 variants.

llvmorg-github-actions · 2026-05-18T18:33:58Z

@llvm/pr-subscribers-mlir

Author: Akhil Goel (akhilgoe)

Changes

This PR adds a conversion pass with patterns to convert supported math ops to SPIR-V builtin calls for the chosen extended instruction set. For OpenCL these lowerings correspond to OpExtInst calls into the OpenCL SPIR-V extended instruction set via mangled __spirv_ocl_ entry points for f32/f64 variants.

Patch is 31.43 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/198370.diff

8 Files Affected:

(added) mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h (+27)
(modified) mlir/include/mlir/Conversion/Passes.h (+1)
(modified) mlir/include/mlir/Conversion/Passes.td (+22)
(modified) mlir/lib/Conversion/CMakeLists.txt (+1)
(modified) mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h (+11-7)
(added) mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt (+24)
(added) mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp (+143)
(added) mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir (+414)

diff --git a/mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h b/mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h
new file mode 100644
index 0000000000000..448b65600f930
--- /dev/null
+++ b/mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h
@@ -0,0 +1,27 @@
+//===- MathToLLVMSPV.h - Utils for converting Math to LLVMSPV -------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#ifndef MLIR_CONVERSION_MATHTOLLVMSPV_MATHTOLLVMSPV_H_
+#define MLIR_CONVERSION_MATHTOLLVMSPV_MATHTOLLVMSPV_H_
+
+#include "mlir/Conversion/LLVMCommon/TypeConverter.h"
+#include "mlir/IR/PatternMatch.h"
+
+namespace mlir {
+class Pass;
+
+#define GEN_PASS_DECL_CONVERTMATHTOLLVMSPV
+#include "mlir/Conversion/Passes.h.inc"
+
+/// Populate the given list with patterns that convert from Math to OCL LLVM-SPV
+/// builtin calls.
+void populateMathToOCLExtSetLLVMSPVConversionPatterns(
+    const LLVMTypeConverter &converter, RewritePatternSet &patterns,
+    PatternBenefit benefit = 1);
+} // namespace mlir
+
+#endif // MLIR_CONVERSION_MATHTOLLVMSPV_MATHTOLLVMSPV_H_
diff --git a/mlir/include/mlir/Conversion/Passes.h b/mlir/include/mlir/Conversion/Passes.h
index a54b98004c3b6..8f6e080ad55c0 100644
--- a/mlir/include/mlir/Conversion/Passes.h
+++ b/mlir/include/mlir/Conversion/Passes.h
@@ -48,6 +48,7 @@
 #include "mlir/Conversion/MathToEmitC/MathToEmitCPass.h"
 #include "mlir/Conversion/MathToFuncs/MathToFuncs.h"
 #include "mlir/Conversion/MathToLLVM/MathToLLVM.h"
+#include "mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h"
 #include "mlir/Conversion/MathToLibm/MathToLibm.h"
 #include "mlir/Conversion/MathToNVVM/MathToNVVM.h"
 #include "mlir/Conversion/MathToROCDL/MathToROCDL.h"
diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td
index d401b56c7602d..548b1351fae02 100644
--- a/mlir/include/mlir/Conversion/Passes.td
+++ b/mlir/include/mlir/Conversion/Passes.td
@@ -838,6 +838,28 @@ def ConvertMathToLLVMPass : Pass<"convert-math-to-llvm"> {
   ];
 }
 
+//===----------------------------------------------------------------------===//
+// MathToLLVMSPV
+//===----------------------------------------------------------------------===//
+
+def ConvertMathToLLVMSPV : Pass<"convert-math-to-llvm-spv", "ModuleOp"> {
+  let summary = "Convert Math dialect to LLVM SPV builtin calls";
+  let description = [{
+    This pass converts supported Math ops to function calls for SPIR-V
+    math intrinsics.
+
+    The extensionSetName option specifies the instruction set chosen for
+    math op lowerings.  
+  }];
+  let dependentDialects = [
+    "func::FuncDialect",
+    "LLVM::LLVMDialect",
+    "vector::VectorDialect"];
+  let options = [Option<"extensionSetName", "extension-set-name", "std::string",
+                        /*default=*/"\"\"",
+                        "SPIR-V Extension set to use for math lowering">];
+}
+
 //===----------------------------------------------------------------------===//
 // MathToLibm
 //===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Conversion/CMakeLists.txt b/mlir/lib/Conversion/CMakeLists.txt
index e17988b12cade..7a2e745a3a64c 100644
--- a/mlir/lib/Conversion/CMakeLists.txt
+++ b/mlir/lib/Conversion/CMakeLists.txt
@@ -39,6 +39,7 @@ add_subdirectory(MathToEmitC)
 add_subdirectory(MathToFuncs)
 add_subdirectory(MathToLibm)
 add_subdirectory(MathToLLVM)
+add_subdirectory(MathToLLVMSPV)
 add_subdirectory(MathToNVVM)
 add_subdirectory(MathToROCDL)
 add_subdirectory(MathToSPIRV)
diff --git a/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h b/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h
index 9f36e5c369d06..cb9b6da071839 100644
--- a/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h
+++ b/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h
@@ -54,14 +54,14 @@ using has_get_fastmath_t = decltype(std::declval<T>().getFastmath());
 template <typename SourceOp>
 struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
 public:
-  explicit OpToFuncCallLowering(const LLVMTypeConverter &lowering,
-                                StringRef f32Func, StringRef f64Func,
-                                StringRef f32ApproxFunc, StringRef f16Func,
-                                StringRef i32Func = "",
-                                PatternBenefit benefit = 1)
+  explicit OpToFuncCallLowering(
+      const LLVMTypeConverter &lowering, StringRef f32Func, StringRef f64Func,
+      StringRef f32ApproxFunc, StringRef f16Func, StringRef i32Func = "",
+      PatternBenefit benefit = 1,
+      LLVM::cconv::CConv cconv = LLVM::cconv::CConv::C)
       : ConvertOpToLLVMPattern<SourceOp>(lowering, benefit), f32Func(f32Func),
         f64Func(f64Func), f32ApproxFunc(f32ApproxFunc), f16Func(f16Func),
-        i32Func(i32Func) {}
+        i32Func(i32Func), cconv(cconv) {}
 
   LogicalResult
   matchAndRewrite(SourceOp op, typename SourceOp::Adaptor adaptor,
@@ -104,6 +104,7 @@ struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
     LLVMFuncOp funcOp = appendOrGetFuncOp(funcName, funcType, op);
     auto callOp =
         LLVM::CallOp::create(rewriter, op->getLoc(), funcOp, castedOperands);
+    callOp.setCConv(cconv);
 
     if (resultType == adaptor.getOperands().front().getType()) {
       rewriter.replaceOp(op, {callOp.getResult()});
@@ -171,7 +172,9 @@ struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
     // location as debug info metadata inside of a function cannot be used
     // outside of that function.
     auto globalloc = op->getLoc()->findInstanceOfOrUnknown<FileLineColLoc>();
-    return LLVMFuncOp::create(b, globalloc, funcName, funcType);
+    auto newFuncOp = LLVMFuncOp::create(b, globalloc, funcName, funcType);
+    newFuncOp.setCConv(cconv);
+    return newFuncOp;
   }
 
   StringRef getFunctionName(Type type, SourceOp op) const {
@@ -202,6 +205,7 @@ struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
   const std::string f32ApproxFunc;
   const std::string f16Func;
   const std::string i32Func;
+  const LLVM::cconv::CConv cconv;
 };
 
 } // namespace mlir
diff --git a/mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt b/mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt
new file mode 100644
index 0000000000000..34279187b1c21
--- /dev/null
+++ b/mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt
@@ -0,0 +1,24 @@
+add_mlir_conversion_library(MLIRMathToLLVMSPV
+  MathToLLVMSPV.cpp
+
+  ADDITIONAL_HEADER_DIRS
+  ${MLIR_MAIN_INCLUDE_DIR}/mlir/Conversion/MathToLLVMSPV
+
+  DEPENDS
+  MLIRConversionPassIncGen
+
+  LINK_COMPONENTS
+  Core
+
+  LINK_LIBS PUBLIC
+  MLIRDialectUtils
+  MLIRFuncDialect
+  MLIRGPUToGPURuntimeTransforms
+  MLIRMathDialect
+  MLIRLLVMCommonConversion
+  MLIRLLVMDialect
+  MLIRPass
+  MLIRTransformUtils
+  MLIRVectorDialect
+  MLIRVectorUtils
+  )
diff --git a/mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp b/mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp
new file mode 100644
index 0000000000000..82a32b3966981
--- /dev/null
+++ b/mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp
@@ -0,0 +1,143 @@
+//===-- MathToLLVMSPV.cpp - conversion from Math to SPIR-V builtin calls --===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h"
+#include "mlir/Dialect/Func/IR/FuncOps.h"
+#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/Dialect/Math/IR/Math.h"
+#include "mlir/Dialect/Vector/IR/VectorOps.h"
+#include "mlir/IR/BuiltinDialect.h"
+#include "mlir/Pass/Pass.h"
+
+#include "../GPUCommon/GPUOpsLowering.h"
+#include "../GPUCommon/OpToFuncCallLowering.h"
+
+namespace mlir {
+#define GEN_PASS_DEF_CONVERTMATHTOLLVMSPV
+#include "mlir/Conversion/Passes.h.inc"
+} // namespace mlir
+
+using namespace mlir;
+
+#define DEBUG_TYPE "math-to-llvm-spv"
+
+static bool isExtensionSetSupported(StringRef name) {
+  return name == "OpenCL.std";
+}
+
+template <typename OpTy>
+static void populateOpPatterns(const LLVMTypeConverter &converter,
+                               RewritePatternSet &patterns,
+                               PatternBenefit benefit, StringRef f32Func,
+                               StringRef f64Func) {
+  patterns.add<ScalarizeVectorOpLowering<OpTy>>(converter, benefit);
+  patterns.add<OpToFuncCallLowering<OpTy>>(converter, f32Func, f64Func,
+                                           /*f32ApproxFunc=*/"", /*f16Func=*/"",
+                                           /*i32Func=*/"", benefit,
+                                           LLVM::cconv::CConv::SPIR_FUNC);
+}
+
+template <typename OpTy>
+static void populateOCLExtSetOpPatterns(const LLVMTypeConverter &converter,
+                                        RewritePatternSet &patterns,
+                                        PatternBenefit benefit,
+                                        StringRef opName) {
+  std::string mangledName =
+      "_Z" + std::to_string(12 + opName.size()) + "__spirv_ocl_" + opName.str();
+  populateOpPatterns<OpTy>(converter, patterns, benefit, mangledName + "f",
+                           mangledName + "d");
+}
+
+void mlir::populateMathToOCLExtSetLLVMSPVConversionPatterns(
+    const LLVMTypeConverter &converter, RewritePatternSet &patterns,
+    PatternBenefit benefit) {
+  populateOCLExtSetOpPatterns<math::AcosOp>(converter, patterns, benefit,
+                                            "acos");
+  populateOCLExtSetOpPatterns<math::AcoshOp>(converter, patterns, benefit,
+                                             "acosh");
+  populateOCLExtSetOpPatterns<math::AsinOp>(converter, patterns, benefit,
+                                            "asin");
+  populateOCLExtSetOpPatterns<math::AsinhOp>(converter, patterns, benefit,
+                                             "asinh");
+  populateOCLExtSetOpPatterns<math::AtanOp>(converter, patterns, benefit,
+                                            "atan");
+  populateOCLExtSetOpPatterns<math::Atan2Op>(converter, patterns, benefit,
+                                             "atan2");
+  populateOCLExtSetOpPatterns<math::AtanhOp>(converter, patterns, benefit,
+                                             "atanh");
+  populateOCLExtSetOpPatterns<math::CbrtOp>(converter, patterns, benefit,
+                                            "cbrt");
+  populateOCLExtSetOpPatterns<math::CopySignOp>(converter, patterns, benefit,
+                                                "copysign");
+  populateOCLExtSetOpPatterns<math::CosOp>(converter, patterns, benefit, "cos");
+  populateOCLExtSetOpPatterns<math::CoshOp>(converter, patterns, benefit,
+                                            "cosh");
+  populateOCLExtSetOpPatterns<math::ErfOp>(converter, patterns, benefit, "erf");
+  populateOCLExtSetOpPatterns<math::ErfcOp>(converter, patterns, benefit,
+                                            "erfc");
+  populateOCLExtSetOpPatterns<math::ExpOp>(converter, patterns, benefit, "exp");
+  populateOCLExtSetOpPatterns<math::Exp2Op>(converter, patterns, benefit,
+                                            "exp2");
+  populateOCLExtSetOpPatterns<math::ExpM1Op>(converter, patterns, benefit,
+                                             "expm1");
+  populateOCLExtSetOpPatterns<math::LogOp>(converter, patterns, benefit, "log");
+  populateOCLExtSetOpPatterns<math::Log10Op>(converter, patterns, benefit,
+                                             "log10");
+  populateOCLExtSetOpPatterns<math::Log1pOp>(converter, patterns, benefit,
+                                             "log1p");
+  populateOCLExtSetOpPatterns<math::Log2Op>(converter, patterns, benefit,
+                                            "log2");
+  populateOCLExtSetOpPatterns<math::PowFOp>(converter, patterns, benefit,
+                                            "pow");
+  populateOCLExtSetOpPatterns<math::RsqrtOp>(converter, patterns, benefit,
+                                             "rsqrt");
+  populateOCLExtSetOpPatterns<math::SinOp>(converter, patterns, benefit, "sin");
+  populateOCLExtSetOpPatterns<math::SinhOp>(converter, patterns, benefit,
+                                            "sinh");
+  populateOCLExtSetOpPatterns<math::SqrtOp>(converter, patterns, benefit,
+                                            "sqrt");
+  populateOCLExtSetOpPatterns<math::TanOp>(converter, patterns, benefit, "tan");
+  populateOCLExtSetOpPatterns<math::TanhOp>(converter, patterns, benefit,
+                                            "tanh");
+}
+
+namespace {
+struct ConvertMathToLLVMSPVPass final
+    : impl::ConvertMathToLLVMSPVBase<ConvertMathToLLVMSPVPass> {
+  using impl::ConvertMathToLLVMSPVBase<
+      ConvertMathToLLVMSPVPass>::ConvertMathToLLVMSPVBase;
+
+  void runOnOperation() override;
+};
+} // namespace
+
+void ConvertMathToLLVMSPVPass::runOnOperation() {
+  auto m = getOperation();
+  MLIRContext *ctx = m.getContext();
+
+  if (!isExtensionSetSupported(extensionSetName)) {
+    m.emitError() << "Unsupported extension set '" << extensionSetName << "'!";
+    return signalPassFailure();
+  }
+
+  RewritePatternSet patterns(&getContext());
+  LowerToLLVMOptions options(ctx, DataLayout(m));
+  LLVMTypeConverter converter(ctx, options);
+  ConversionTarget target(getContext());
+  target.addLegalDialect<BuiltinDialect, func::FuncDialect,
+                         vector::VectorDialect, LLVM::LLVMDialect>();
+  if (extensionSetName == "OpenCL.std") {
+    populateMathToOCLExtSetLLVMSPVConversionPatterns(converter, patterns,
+                                                     /*benefit=*/1);
+    target
+        .addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::Exp2Op, LLVM::LogOp,
+                      LLVM::Log10Op, LLVM::Log2Op, LLVM::SinOp, LLVM::SqrtOp>();
+  }
+  if (failed(applyPartialConversion(m, target, std::move(patterns))))
+    signalPassFailure();
+}
diff --git a/mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir b/mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir
new file mode 100644
index 0000000000000..d47927aa2b5ad
--- /dev/null
+++ b/mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir
@@ -0,0 +1,414 @@
+// RUN: mlir-opt %s -split-input-file -convert-math-to-llvm-spv='extension-set-name=OpenCL.std' -gpu-module-to-binary | FileCheck %s 
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_copysignf(f32, f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atan2f(f32, f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_powf(f32, f32) -> f32
+  // CHECK-LABEL: func @math_bin_f32
+  func.func @math_bin_f32(%arg_f32_1 : f32, %arg_f32_2 : f32) -> (f32, f32, f32) {
+    %result1 = math.copysign %arg_f32_1, %arg_f32_2 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_copysignf(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
+    %result2 = math.atan2 %arg_f32_1, %arg_f32_2 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atan2f(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
+    %result3 = math.powf %arg_f32_1, %arg_f32_2 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_powf(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
+    func.return %result1, %result2, %result3 : f32, f32, f32
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_copysignd(f64, f64) -> f64
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atan2d(f64, f64) -> f64
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_powd(f64, f64) -> f64
+  // CHECK-LABEL: func @math_bin_f64
+  func.func @math_bin_f64(%arg_f64_1 : f64, %arg_f64_2 : f64) -> (f64, f64, f64) {
+    %result1 = math.copysign %arg_f64_1, %arg_f64_2 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_copysignd(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
+    %result2 = math.atan2 %arg_f64_1, %arg_f64_2 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atan2d(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
+    %result3 = math.powf %arg_f64_1, %arg_f64_2 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_powd(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
+    func.return %result1, %result2, %result3 : f64, f64, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acosf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acosd(f64) -> f64
+  // CHECK-LABEL: func @math_acos
+  func.func @math_acos(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.acos %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acosf(%{{.*}}) : (f32) -> f32
+    %result64 = math.acos %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acosd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acoshf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acoshd(f64) -> f64
+  // CHECK-LABEL: func @math_acosh
+  func.func @math_acosh(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.acosh %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acoshf(%{{.*}}) : (f32) -> f32
+    %result64 = math.acosh %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acoshd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asinf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asind(f64) -> f64
+  // CHECK-LABEL: func @math_asin
+  func.func @math_asin(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.asin %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asinf(%{{.*}}) : (f32) -> f32
+    %result64 = math.asin %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asind(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asinhf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asinhd(f64) -> f64
+  // CHECK-LABEL: func @math_asinh
+  func.func @math_asinh(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.asinh %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asinhf(%{{.*}}) : (f32) -> f32
+    %result64 = math.asinh %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asinhd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atanf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atand(f64) -> f64
+  // CHECK-LABEL: func @math_atan
+  func.func @math_atan(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.atan %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atanf(%{{.*}}) : (f32) -> f32
+    %result64 = math.atan %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atand(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atanhf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atanhd(f64) -> f64
+  // CHECK-LABEL: func @math_atanh
+  func.func @math_atanh(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.atanh %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atanhf(%{{.*}}) : (f32) -> f32
+    %result64 = math.atanh %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atanhd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_cbrtf(f32...
[truncated]

llvmorg-github-actions · 2026-05-18T18:33:59Z

@llvm/pr-subscribers-mlir-gpu

Author: Akhil Goel (akhilgoe)

Changes

This PR adds a conversion pass with patterns to convert supported math ops to SPIR-V builtin calls for the chosen extended instruction set. For OpenCL these lowerings correspond to OpExtInst calls into the OpenCL SPIR-V extended instruction set via mangled __spirv_ocl_ entry points for f32/f64 variants.

Patch is 31.43 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/198370.diff

8 Files Affected:

(added) mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h (+27)
(modified) mlir/include/mlir/Conversion/Passes.h (+1)
(modified) mlir/include/mlir/Conversion/Passes.td (+22)
(modified) mlir/lib/Conversion/CMakeLists.txt (+1)
(modified) mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h (+11-7)
(added) mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt (+24)
(added) mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp (+143)
(added) mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir (+414)

diff --git a/mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h b/mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h
new file mode 100644
index 0000000000000..448b65600f930
--- /dev/null
+++ b/mlir/include/mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h
@@ -0,0 +1,27 @@
+//===- MathToLLVMSPV.h - Utils for converting Math to LLVMSPV -------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+#ifndef MLIR_CONVERSION_MATHTOLLVMSPV_MATHTOLLVMSPV_H_
+#define MLIR_CONVERSION_MATHTOLLVMSPV_MATHTOLLVMSPV_H_
+
+#include "mlir/Conversion/LLVMCommon/TypeConverter.h"
+#include "mlir/IR/PatternMatch.h"
+
+namespace mlir {
+class Pass;
+
+#define GEN_PASS_DECL_CONVERTMATHTOLLVMSPV
+#include "mlir/Conversion/Passes.h.inc"
+
+/// Populate the given list with patterns that convert from Math to OCL LLVM-SPV
+/// builtin calls.
+void populateMathToOCLExtSetLLVMSPVConversionPatterns(
+    const LLVMTypeConverter &converter, RewritePatternSet &patterns,
+    PatternBenefit benefit = 1);
+} // namespace mlir
+
+#endif // MLIR_CONVERSION_MATHTOLLVMSPV_MATHTOLLVMSPV_H_
diff --git a/mlir/include/mlir/Conversion/Passes.h b/mlir/include/mlir/Conversion/Passes.h
index a54b98004c3b6..8f6e080ad55c0 100644
--- a/mlir/include/mlir/Conversion/Passes.h
+++ b/mlir/include/mlir/Conversion/Passes.h
@@ -48,6 +48,7 @@
 #include "mlir/Conversion/MathToEmitC/MathToEmitCPass.h"
 #include "mlir/Conversion/MathToFuncs/MathToFuncs.h"
 #include "mlir/Conversion/MathToLLVM/MathToLLVM.h"
+#include "mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h"
 #include "mlir/Conversion/MathToLibm/MathToLibm.h"
 #include "mlir/Conversion/MathToNVVM/MathToNVVM.h"
 #include "mlir/Conversion/MathToROCDL/MathToROCDL.h"
diff --git a/mlir/include/mlir/Conversion/Passes.td b/mlir/include/mlir/Conversion/Passes.td
index d401b56c7602d..548b1351fae02 100644
--- a/mlir/include/mlir/Conversion/Passes.td
+++ b/mlir/include/mlir/Conversion/Passes.td
@@ -838,6 +838,28 @@ def ConvertMathToLLVMPass : Pass<"convert-math-to-llvm"> {
   ];
 }
 
+//===----------------------------------------------------------------------===//
+// MathToLLVMSPV
+//===----------------------------------------------------------------------===//
+
+def ConvertMathToLLVMSPV : Pass<"convert-math-to-llvm-spv", "ModuleOp"> {
+  let summary = "Convert Math dialect to LLVM SPV builtin calls";
+  let description = [{
+    This pass converts supported Math ops to function calls for SPIR-V
+    math intrinsics.
+
+    The extensionSetName option specifies the instruction set chosen for
+    math op lowerings.  
+  }];
+  let dependentDialects = [
+    "func::FuncDialect",
+    "LLVM::LLVMDialect",
+    "vector::VectorDialect"];
+  let options = [Option<"extensionSetName", "extension-set-name", "std::string",
+                        /*default=*/"\"\"",
+                        "SPIR-V Extension set to use for math lowering">];
+}
+
 //===----------------------------------------------------------------------===//
 // MathToLibm
 //===----------------------------------------------------------------------===//
diff --git a/mlir/lib/Conversion/CMakeLists.txt b/mlir/lib/Conversion/CMakeLists.txt
index e17988b12cade..7a2e745a3a64c 100644
--- a/mlir/lib/Conversion/CMakeLists.txt
+++ b/mlir/lib/Conversion/CMakeLists.txt
@@ -39,6 +39,7 @@ add_subdirectory(MathToEmitC)
 add_subdirectory(MathToFuncs)
 add_subdirectory(MathToLibm)
 add_subdirectory(MathToLLVM)
+add_subdirectory(MathToLLVMSPV)
 add_subdirectory(MathToNVVM)
 add_subdirectory(MathToROCDL)
 add_subdirectory(MathToSPIRV)
diff --git a/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h b/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h
index 9f36e5c369d06..cb9b6da071839 100644
--- a/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h
+++ b/mlir/lib/Conversion/GPUCommon/OpToFuncCallLowering.h
@@ -54,14 +54,14 @@ using has_get_fastmath_t = decltype(std::declval<T>().getFastmath());
 template <typename SourceOp>
 struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
 public:
-  explicit OpToFuncCallLowering(const LLVMTypeConverter &lowering,
-                                StringRef f32Func, StringRef f64Func,
-                                StringRef f32ApproxFunc, StringRef f16Func,
-                                StringRef i32Func = "",
-                                PatternBenefit benefit = 1)
+  explicit OpToFuncCallLowering(
+      const LLVMTypeConverter &lowering, StringRef f32Func, StringRef f64Func,
+      StringRef f32ApproxFunc, StringRef f16Func, StringRef i32Func = "",
+      PatternBenefit benefit = 1,
+      LLVM::cconv::CConv cconv = LLVM::cconv::CConv::C)
       : ConvertOpToLLVMPattern<SourceOp>(lowering, benefit), f32Func(f32Func),
         f64Func(f64Func), f32ApproxFunc(f32ApproxFunc), f16Func(f16Func),
-        i32Func(i32Func) {}
+        i32Func(i32Func), cconv(cconv) {}
 
   LogicalResult
   matchAndRewrite(SourceOp op, typename SourceOp::Adaptor adaptor,
@@ -104,6 +104,7 @@ struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
     LLVMFuncOp funcOp = appendOrGetFuncOp(funcName, funcType, op);
     auto callOp =
         LLVM::CallOp::create(rewriter, op->getLoc(), funcOp, castedOperands);
+    callOp.setCConv(cconv);
 
     if (resultType == adaptor.getOperands().front().getType()) {
       rewriter.replaceOp(op, {callOp.getResult()});
@@ -171,7 +172,9 @@ struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
     // location as debug info metadata inside of a function cannot be used
     // outside of that function.
     auto globalloc = op->getLoc()->findInstanceOfOrUnknown<FileLineColLoc>();
-    return LLVMFuncOp::create(b, globalloc, funcName, funcType);
+    auto newFuncOp = LLVMFuncOp::create(b, globalloc, funcName, funcType);
+    newFuncOp.setCConv(cconv);
+    return newFuncOp;
   }
 
   StringRef getFunctionName(Type type, SourceOp op) const {
@@ -202,6 +205,7 @@ struct OpToFuncCallLowering : public ConvertOpToLLVMPattern<SourceOp> {
   const std::string f32ApproxFunc;
   const std::string f16Func;
   const std::string i32Func;
+  const LLVM::cconv::CConv cconv;
 };
 
 } // namespace mlir
diff --git a/mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt b/mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt
new file mode 100644
index 0000000000000..34279187b1c21
--- /dev/null
+++ b/mlir/lib/Conversion/MathToLLVMSPV/CMakeLists.txt
@@ -0,0 +1,24 @@
+add_mlir_conversion_library(MLIRMathToLLVMSPV
+  MathToLLVMSPV.cpp
+
+  ADDITIONAL_HEADER_DIRS
+  ${MLIR_MAIN_INCLUDE_DIR}/mlir/Conversion/MathToLLVMSPV
+
+  DEPENDS
+  MLIRConversionPassIncGen
+
+  LINK_COMPONENTS
+  Core
+
+  LINK_LIBS PUBLIC
+  MLIRDialectUtils
+  MLIRFuncDialect
+  MLIRGPUToGPURuntimeTransforms
+  MLIRMathDialect
+  MLIRLLVMCommonConversion
+  MLIRLLVMDialect
+  MLIRPass
+  MLIRTransformUtils
+  MLIRVectorDialect
+  MLIRVectorUtils
+  )
diff --git a/mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp b/mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp
new file mode 100644
index 0000000000000..82a32b3966981
--- /dev/null
+++ b/mlir/lib/Conversion/MathToLLVMSPV/MathToLLVMSPV.cpp
@@ -0,0 +1,143 @@
+//===-- MathToLLVMSPV.cpp - conversion from Math to SPIR-V builtin calls --===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "mlir/Conversion/MathToLLVMSPV/MathToLLVMSPV.h"
+#include "mlir/Dialect/Func/IR/FuncOps.h"
+#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
+#include "mlir/Dialect/Math/IR/Math.h"
+#include "mlir/Dialect/Vector/IR/VectorOps.h"
+#include "mlir/IR/BuiltinDialect.h"
+#include "mlir/Pass/Pass.h"
+
+#include "../GPUCommon/GPUOpsLowering.h"
+#include "../GPUCommon/OpToFuncCallLowering.h"
+
+namespace mlir {
+#define GEN_PASS_DEF_CONVERTMATHTOLLVMSPV
+#include "mlir/Conversion/Passes.h.inc"
+} // namespace mlir
+
+using namespace mlir;
+
+#define DEBUG_TYPE "math-to-llvm-spv"
+
+static bool isExtensionSetSupported(StringRef name) {
+  return name == "OpenCL.std";
+}
+
+template <typename OpTy>
+static void populateOpPatterns(const LLVMTypeConverter &converter,
+                               RewritePatternSet &patterns,
+                               PatternBenefit benefit, StringRef f32Func,
+                               StringRef f64Func) {
+  patterns.add<ScalarizeVectorOpLowering<OpTy>>(converter, benefit);
+  patterns.add<OpToFuncCallLowering<OpTy>>(converter, f32Func, f64Func,
+                                           /*f32ApproxFunc=*/"", /*f16Func=*/"",
+                                           /*i32Func=*/"", benefit,
+                                           LLVM::cconv::CConv::SPIR_FUNC);
+}
+
+template <typename OpTy>
+static void populateOCLExtSetOpPatterns(const LLVMTypeConverter &converter,
+                                        RewritePatternSet &patterns,
+                                        PatternBenefit benefit,
+                                        StringRef opName) {
+  std::string mangledName =
+      "_Z" + std::to_string(12 + opName.size()) + "__spirv_ocl_" + opName.str();
+  populateOpPatterns<OpTy>(converter, patterns, benefit, mangledName + "f",
+                           mangledName + "d");
+}
+
+void mlir::populateMathToOCLExtSetLLVMSPVConversionPatterns(
+    const LLVMTypeConverter &converter, RewritePatternSet &patterns,
+    PatternBenefit benefit) {
+  populateOCLExtSetOpPatterns<math::AcosOp>(converter, patterns, benefit,
+                                            "acos");
+  populateOCLExtSetOpPatterns<math::AcoshOp>(converter, patterns, benefit,
+                                             "acosh");
+  populateOCLExtSetOpPatterns<math::AsinOp>(converter, patterns, benefit,
+                                            "asin");
+  populateOCLExtSetOpPatterns<math::AsinhOp>(converter, patterns, benefit,
+                                             "asinh");
+  populateOCLExtSetOpPatterns<math::AtanOp>(converter, patterns, benefit,
+                                            "atan");
+  populateOCLExtSetOpPatterns<math::Atan2Op>(converter, patterns, benefit,
+                                             "atan2");
+  populateOCLExtSetOpPatterns<math::AtanhOp>(converter, patterns, benefit,
+                                             "atanh");
+  populateOCLExtSetOpPatterns<math::CbrtOp>(converter, patterns, benefit,
+                                            "cbrt");
+  populateOCLExtSetOpPatterns<math::CopySignOp>(converter, patterns, benefit,
+                                                "copysign");
+  populateOCLExtSetOpPatterns<math::CosOp>(converter, patterns, benefit, "cos");
+  populateOCLExtSetOpPatterns<math::CoshOp>(converter, patterns, benefit,
+                                            "cosh");
+  populateOCLExtSetOpPatterns<math::ErfOp>(converter, patterns, benefit, "erf");
+  populateOCLExtSetOpPatterns<math::ErfcOp>(converter, patterns, benefit,
+                                            "erfc");
+  populateOCLExtSetOpPatterns<math::ExpOp>(converter, patterns, benefit, "exp");
+  populateOCLExtSetOpPatterns<math::Exp2Op>(converter, patterns, benefit,
+                                            "exp2");
+  populateOCLExtSetOpPatterns<math::ExpM1Op>(converter, patterns, benefit,
+                                             "expm1");
+  populateOCLExtSetOpPatterns<math::LogOp>(converter, patterns, benefit, "log");
+  populateOCLExtSetOpPatterns<math::Log10Op>(converter, patterns, benefit,
+                                             "log10");
+  populateOCLExtSetOpPatterns<math::Log1pOp>(converter, patterns, benefit,
+                                             "log1p");
+  populateOCLExtSetOpPatterns<math::Log2Op>(converter, patterns, benefit,
+                                            "log2");
+  populateOCLExtSetOpPatterns<math::PowFOp>(converter, patterns, benefit,
+                                            "pow");
+  populateOCLExtSetOpPatterns<math::RsqrtOp>(converter, patterns, benefit,
+                                             "rsqrt");
+  populateOCLExtSetOpPatterns<math::SinOp>(converter, patterns, benefit, "sin");
+  populateOCLExtSetOpPatterns<math::SinhOp>(converter, patterns, benefit,
+                                            "sinh");
+  populateOCLExtSetOpPatterns<math::SqrtOp>(converter, patterns, benefit,
+                                            "sqrt");
+  populateOCLExtSetOpPatterns<math::TanOp>(converter, patterns, benefit, "tan");
+  populateOCLExtSetOpPatterns<math::TanhOp>(converter, patterns, benefit,
+                                            "tanh");
+}
+
+namespace {
+struct ConvertMathToLLVMSPVPass final
+    : impl::ConvertMathToLLVMSPVBase<ConvertMathToLLVMSPVPass> {
+  using impl::ConvertMathToLLVMSPVBase<
+      ConvertMathToLLVMSPVPass>::ConvertMathToLLVMSPVBase;
+
+  void runOnOperation() override;
+};
+} // namespace
+
+void ConvertMathToLLVMSPVPass::runOnOperation() {
+  auto m = getOperation();
+  MLIRContext *ctx = m.getContext();
+
+  if (!isExtensionSetSupported(extensionSetName)) {
+    m.emitError() << "Unsupported extension set '" << extensionSetName << "'!";
+    return signalPassFailure();
+  }
+
+  RewritePatternSet patterns(&getContext());
+  LowerToLLVMOptions options(ctx, DataLayout(m));
+  LLVMTypeConverter converter(ctx, options);
+  ConversionTarget target(getContext());
+  target.addLegalDialect<BuiltinDialect, func::FuncDialect,
+                         vector::VectorDialect, LLVM::LLVMDialect>();
+  if (extensionSetName == "OpenCL.std") {
+    populateMathToOCLExtSetLLVMSPVConversionPatterns(converter, patterns,
+                                                     /*benefit=*/1);
+    target
+        .addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::Exp2Op, LLVM::LogOp,
+                      LLVM::Log10Op, LLVM::Log2Op, LLVM::SinOp, LLVM::SqrtOp>();
+  }
+  if (failed(applyPartialConversion(m, target, std::move(patterns))))
+    signalPassFailure();
+}
diff --git a/mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir b/mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir
new file mode 100644
index 0000000000000..d47927aa2b5ad
--- /dev/null
+++ b/mlir/test/Conversion/MathToLLVMSPV/math-to-llvm-spv.mlir
@@ -0,0 +1,414 @@
+// RUN: mlir-opt %s -split-input-file -convert-math-to-llvm-spv='extension-set-name=OpenCL.std' -gpu-module-to-binary | FileCheck %s 
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_copysignf(f32, f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atan2f(f32, f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_powf(f32, f32) -> f32
+  // CHECK-LABEL: func @math_bin_f32
+  func.func @math_bin_f32(%arg_f32_1 : f32, %arg_f32_2 : f32) -> (f32, f32, f32) {
+    %result1 = math.copysign %arg_f32_1, %arg_f32_2 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_copysignf(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
+    %result2 = math.atan2 %arg_f32_1, %arg_f32_2 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atan2f(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
+    %result3 = math.powf %arg_f32_1, %arg_f32_2 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_powf(%{{.*}}, %{{.*}}) : (f32, f32) -> f32
+    func.return %result1, %result2, %result3 : f32, f32, f32
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_copysignd(f64, f64) -> f64
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atan2d(f64, f64) -> f64
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_powd(f64, f64) -> f64
+  // CHECK-LABEL: func @math_bin_f64
+  func.func @math_bin_f64(%arg_f64_1 : f64, %arg_f64_2 : f64) -> (f64, f64, f64) {
+    %result1 = math.copysign %arg_f64_1, %arg_f64_2 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_copysignd(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
+    %result2 = math.atan2 %arg_f64_1, %arg_f64_2 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atan2d(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
+    %result3 = math.powf %arg_f64_1, %arg_f64_2 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_powd(%{{.*}}, %{{.*}}) : (f64, f64) -> f64
+    func.return %result1, %result2, %result3 : f64, f64, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acosf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acosd(f64) -> f64
+  // CHECK-LABEL: func @math_acos
+  func.func @math_acos(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.acos %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acosf(%{{.*}}) : (f32) -> f32
+    %result64 = math.acos %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acosd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acoshf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_acoshd(f64) -> f64
+  // CHECK-LABEL: func @math_acosh
+  func.func @math_acosh(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.acosh %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acoshf(%{{.*}}) : (f32) -> f32
+    %result64 = math.acosh %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_acoshd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asinf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asind(f64) -> f64
+  // CHECK-LABEL: func @math_asin
+  func.func @math_asin(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.asin %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asinf(%{{.*}}) : (f32) -> f32
+    %result64 = math.asin %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asind(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asinhf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_asinhd(f64) -> f64
+  // CHECK-LABEL: func @math_asinh
+  func.func @math_asinh(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.asinh %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asinhf(%{{.*}}) : (f32) -> f32
+    %result64 = math.asinh %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_asinhd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atanf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atand(f64) -> f64
+  // CHECK-LABEL: func @math_atan
+  func.func @math_atan(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.atan %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atanf(%{{.*}}) : (f32) -> f32
+    %result64 = math.atan %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atand(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atanhf(f32) -> f32
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_atanhd(f64) -> f64
+  // CHECK-LABEL: func @math_atanh
+  func.func @math_atanh(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.atanh %arg_f32 : f32
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atanhf(%{{.*}}) : (f32) -> f32
+    %result64 = math.atanh %arg_f64 : f64
+    // CHECK: llvm.call spir_funccc @_Z{{.*}}__spirv_ocl_atanhd(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+module @test_module {
+  // CHECK: llvm.func spir_funccc @_Z{{.*}}__spirv_ocl_cbrtf(f32...
[truncated]

akhilgoe · 2026-05-18T19:50:42Z

Moving to draft for now.

github-actions · 2026-05-28T16:17:36Z

✅ With the latest revision this PR passed the C/C++ code formatter.

joker-eph · 2026-05-28T22:03:11Z

This PR adds a conversion pass

Does it? I can't spot it...

akhilgoe · 2026-05-28T22:41:29Z

@joker-eph The initial commit added a conversion pass. This morning I moved the patterns to MathToXeVM pass pipeline and restricted to OpenCL transforms.

joker-eph · 2026-05-29T09:25:02Z

+  std::string mangledName =
+      "_Z" + std::to_string(12 + opName.size()) + "__spirv_ocl_" + opName.str();


Suggested change

std::string mangledName =

"_Z" + std::to_string(12 + opName.size()) + "__spirv_ocl_" + opName.str();

std::string prefix = "__spirv_ocl_";

std::string mangledName =

"_Z" + std::to_string(prefix.size() + opName.size()) + prefix + opName.str();

Just a nit to avoid "magic numbers"

That's a great suggestion, thanks!

joker-eph · 2026-05-29T09:25:49Z

Thanks for clarifying: looks like these lowering are specific to XeVM runtime right? Seems right then.

akhilgoe · 2026-05-29T21:51:25Z

Yes, this is relevant for XeVM/SPIR-V targets. XeVM already lowers fastmath ops to native OpenCL intrinsics, so feels natural to move this to the existing pipeline.

Math to LLVM-SPV conversion

df434cb

akhilgoe requested a review from fabianmcg as a code owner May 18, 2026 18:33

llvmorg-github-actions Bot added mlir:gpu mlir labels May 18, 2026

Remove unused flag from test

64a52f2

akhilgoe marked this pull request as draft May 18, 2026 19:51

akhilgoe added 3 commits May 28, 2026 09:19

Move patterns and add more tests

7d3734e

Fix formatting

d4db895

Fix formatting

53fa259

joker-eph reviewed May 29, 2026

View reviewed changes

Add explicit prefix for mangled name

c237467

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[mlir][Math][LLVMSPV] Add Math to LLVM-SPV conversion pass#198370

[mlir][Math][LLVMSPV] Add Math to LLVM-SPV conversion pass#198370
akhilgoe wants to merge 6 commits into
llvm:mainfrom
akhilgoe:akhil/llvmspv_math

akhilgoe commented May 18, 2026

Uh oh!

llvmorg-github-actions Bot commented May 18, 2026

Uh oh!

llvmorg-github-actions Bot commented May 18, 2026

Uh oh!

akhilgoe commented May 18, 2026

Uh oh!

github-actions Bot commented May 28, 2026 •

edited

Loading

Uh oh!

joker-eph commented May 28, 2026

Uh oh!

akhilgoe commented May 28, 2026 •

edited

Loading

Uh oh!

joker-eph May 29, 2026

Uh oh!

akhilgoe May 29, 2026

Uh oh!

akhilgoe May 29, 2026

Uh oh!

joker-eph commented May 29, 2026

Uh oh!

akhilgoe commented May 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		std::string mangledName =
		"_Z" + std::to_string(12 + opName.size()) + "__spirv_ocl_" + opName.str();

Conversation

akhilgoe commented May 18, 2026

Uh oh!

llvmorg-github-actions Bot commented May 18, 2026

Uh oh!

llvmorg-github-actions Bot commented May 18, 2026

Uh oh!

akhilgoe commented May 18, 2026

Uh oh!

github-actions Bot commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joker-eph commented May 28, 2026

Uh oh!

akhilgoe commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

joker-eph May 29, 2026

Choose a reason for hiding this comment

Uh oh!

akhilgoe May 29, 2026

Choose a reason for hiding this comment

Uh oh!

akhilgoe May 29, 2026

Choose a reason for hiding this comment

Uh oh!

joker-eph commented May 29, 2026

Uh oh!

akhilgoe commented May 29, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

github-actions Bot commented May 28, 2026 •

edited

Loading

akhilgoe commented May 28, 2026 •

edited

Loading