[mlir][math] Add missing trig math-to-llvm conversion patterns

[ashermancinelli]

asin, acos, atan, and atan2 were being lowered to libm calls instead of llvm intrinsics. Add the conversion patterns to handle these intrinsics and update tests to expect this.

NOTE: I don't know what the difference between the fast and relaxed versions of the fir tests. I followed the surrounding patterns, but relaxed and fast look identical. I expected precise to use the libm C call with no FMFs, relaxed to use the math dialect with no FMFs, and the fast to use the math dialect with FMFs.

NOTE: I don't know who to tag for math dialect changes, so feel free to tag someone else if you do!

[llvmbot]

@llvm/pr-subscribers-mlir

@llvm/pr-subscribers-flang-codegen

Author: Asher Mancinelli (ashermancinelli)

Changes

asin, acos, atan, and atan2 were being lowered to libm calls instead of llvm intrinsics. Add the conversion patterns to handle these intrinsics and update tests to expect this.

NOTE: I don't know what the difference between the fast and relaxed versions of the fir tests. I followed the surrounding patterns, but relaxed and fast look identical. I expected precise to use the libm C call with no FMFs, relaxed to use the math dialect with no FMFs, and the fast to use the math dialect with FMFs.

NOTE: I don't know who to tag for math dialect changes, so feel free to tag someone else if you do!

---

Full diff: https://github.com/llvm/llvm-project/pull/141069.diff

3 Files Affected:

• (modified) flang/test/Intrinsics/math-codegen.fir (+162-8)
• (modified) mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp (+10-2)
• (modified) mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir (+78)

diff --git a/flang/test/Intrinsics/math-codegen.fir b/flang/test/Intrinsics/math-codegen.fir index c45c6b23e897e..b7c4e07130662 100644 --- a/flang/test/Intrinsics/math-codegen.fir +++ b/flang/test/Intrinsics/math-codegen.fir @@ -378,13 +378,167 @@ func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { func.func private @llvm.round.f32(f32) -> f32 func.func private @llvm.round.f64(f64) -> f64 +//--- asin_fast.fir +// RUN: fir-opt %t/asin_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/asin_fast.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.asin %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.asin %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- asin_relaxed.fir +// RUN: fir-opt %t/asin_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/asin_relaxed.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.asin %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.asin %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- asin_precise.fir +// RUN: fir-opt %t/asin_precise.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/asin_precise.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @asinf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @asin({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = fir.call @asinf(%1) : (f32) -> f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = fir.call @asin(%1) : (f64) -> f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} +func.func private @asinf(f32) -> f32 +func.func private @asin(f64) -> f64 + +//--- acos_fast.fir +// RUN: fir-opt %t/acos_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/acos_fast.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.acos %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.acos %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- acos_relaxed.fir +// RUN: fir-opt %t/acos_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/acos_relaxed.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.acos %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.acos %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- acos_precise.fir +// RUN: fir-opt %t/acos_precise.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/acos_precise.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @acosf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @acos({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = fir.call @acosf(%1) : (f32) -> f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = fir.call @acos(%1) : (f64) -> f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} +func.func private @acosf(f32) -> f32 +func.func private @acos(f64) -> f64 + //--- atan_fast.fir // RUN: fir-opt %t/atan_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan_fast.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atanf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} @@ -406,10 +560,10 @@ func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { //--- atan_relaxed.fir // RUN: fir-opt %t/atan_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan_relaxed.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atanf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} @@ -458,10 +612,10 @@ func.func private @atan(f64) -> f64 //--- atan2_fast.fir // RUN: fir-opt %t/atan2_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan2_fast.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2f({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}, %arg1: !fir.ref<f32> {fir.bindc_name = "y"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} @@ -485,10 +639,10 @@ func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}, %arg1: !fi //--- atan2_relaxed.fir // RUN: fir-opt %t/atan2_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan2_relaxed.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2f({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}, %arg1: !fir.ref<f32> {fir.bindc_name = "y"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} diff --git a/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp b/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp index 97da96afac4cd..b42bb773f53ee 100644 --- a/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp +++ b/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp @@ -42,6 +42,7 @@ using CopySignOpLowering = ConvertFMFMathToLLVMPattern<math::CopySignOp, LLVM::CopySignOp>; using CosOpLowering = ConvertFMFMathToLLVMPattern<math::CosOp, LLVM::CosOp>; using CoshOpLowering = ConvertFMFMathToLLVMPattern<math::CoshOp, LLVM::CoshOp>; +using AcosOpLowering = ConvertFMFMathToLLVMPattern<math::AcosOp, LLVM::ACosOp>; using CtPopFOpLowering = VectorConvertToLLVMPattern<math::CtPopOp, LLVM::CtPopOp>; using Exp2OpLowering = ConvertFMFMathToLLVMPattern<math::Exp2Op, LLVM::Exp2Op>; @@ -62,12 +63,15 @@ using RoundOpLowering = ConvertFMFMathToLLVMPattern<math::RoundOp, LLVM::RoundOp>; using SinOpLowering = ConvertFMFMathToLLVMPattern<math::SinOp, LLVM::SinOp>; using SinhOpLowering = ConvertFMFMathToLLVMPattern<math::SinhOp, LLVM::SinhOp>; +using ASinOpLowering = ConvertFMFMathToLLVMPattern<math::AsinOp, LLVM::ASinOp>; using SqrtOpLowering = ConvertFMFMathToLLVMPattern<math::SqrtOp, LLVM::SqrtOp>; using FTruncOpLowering = ConvertFMFMathToLLVMPattern<math::TruncOp, LLVM::FTruncOp>; using TanOpLowering = ConvertFMFMathToLLVMPattern<math::TanOp, LLVM::TanOp>; using TanhOpLowering = ConvertFMFMathToLLVMPattern<math::TanhOp, LLVM::TanhOp>; - +using ATanOpLowering = ConvertFMFMathToLLVMPattern<math::AtanOp, LLVM::ATanOp>; +using ATan2OpLowering = + ConvertFMFMathToLLVMPattern<math::Atan2Op, LLVM::ATan2Op>; // A `CtLz/CtTz/absi(a)` is converted into `CtLz/CtTz/absi(a, false)`. template <typename MathOp, typename LLVMOp> struct IntOpWithFlagLowering : public ConvertOpToLLVMPattern<MathOp> { @@ -353,6 +357,7 @@ void mlir::populateMathToLLVMConversionPatterns( CopySignOpLowering, CosOpLowering, CoshOpLowering, + AcosOpLowering, CountLeadingZerosOpLowering, CountTrailingZerosOpLowering, CtPopFOpLowering, @@ -371,10 +376,13 @@ void mlir::populateMathToLLVMConversionPatterns( RsqrtOpLowering, SinOpLowering, SinhOpLowering, + ASinOpLowering, SqrtOpLowering, FTruncOpLowering, TanOpLowering, - TanhOpLowering + TanhOpLowering, + ATanOpLowering, + ATan2OpLowering >(converter, benefit); // clang-format on } diff --git a/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir b/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir index 974743a55932b..537fb967ef0e1 100644 --- a/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir +++ b/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir @@ -177,6 +177,84 @@ func.func @trigonometrics(%arg0: f32) { // ----- +// CHECK-LABEL: func @inverse_trigonometrics +// CHECK-SAME: [[ARG0:%.+]]: f32 +func.func @inverse_trigonometrics(%arg0: f32) { + // CHECK: llvm.intr.asin([[ARG0]]) : (f32) -> f32 + %0 = math.asin %arg0 : f32 + + // CHECK: llvm.intr.acos([[ARG0]]) : (f32) -> f32 + %1 = math.acos %arg0 : f32 + + // CHECK: llvm.intr.atan([[ARG0]]) : (f32) -> f32 + %2 = math.atan %arg0 : f32 + func.return +} + +// ----- + +// CHECK-LABEL: func @atan2 +// CHECK-SAME: [[ARG0:%.+]]: f32, [[ARG1:%.+]]: f32 +func.func @atan2(%arg0: f32, %arg1: f32) { + // CHECK: llvm.intr.atan2([[ARG0]], [[ARG1]]) : (f32, f32) -> f32 + %0 = math.atan2 %arg0, %arg1 : f32 + func.return +} + +// ----- + +// CHECK-LABEL: func @inverse_trigonometrics_vector +// CHECK-SAME: [[ARG0:%.+]]: vector<4xf32> +func.func @inverse_trigonometrics_vector(%arg0: vector<4xf32>) { + // CHECK: llvm.intr.asin([[ARG0]]) : (vector<4xf32>) -> vector<4xf32> + %0 = math.asin %arg0 : vector<4xf32> + + // CHECK: llvm.intr.acos([[ARG0]]) : (vector<4xf32>) -> vector<4xf32> + %1 = math.acos %arg0 : vector<4xf32> + + // CHECK: llvm.intr.atan([[ARG0]]) : (vector<4xf32>) -> vector<4xf32> + %2 = math.atan %arg0 : vector<4xf32> + func.return +} + +// ----- + +// CHECK-LABEL: func @atan2_vector +// CHECK-SAME: [[ARG0:%.+]]: vector<4xf32>, [[ARG1:%.+]]: vector<4xf32> +func.func @atan2_vector(%arg0: vector<4xf32>, %arg1: vector<4xf32>) { + // CHECK: llvm.intr.atan2([[ARG0]], [[ARG1]]) : (vector<4xf32>, vector<4xf32>) -> vector<4xf32> + %0 = math.atan2 %arg0, %arg1 : vector<4xf32> + func.return +} + +// ----- + +// CHECK-LABEL: func @inverse_trigonometrics_fmf +// CHECK-SAME: [[ARG0:%.+]]: f32 +func.func @inverse_trigonometrics_fmf(%arg0: f32) { + // CHECK: llvm.intr.asin([[ARG0]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32) -> f32 + %0 = math.asin %arg0 fastmath<fast> : f32 + + // CHECK: llvm.intr.acos([[ARG0]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32) -> f32 + %1 = math.acos %arg0 fastmath<fast> : f32 + + // CHECK: llvm.intr.atan([[ARG0]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32) -> f32 + %2 = math.atan %arg0 fastmath<fast> : f32 + func.return +} + +// ----- + +// CHECK-LABEL: func @atan2_fmf +// CHECK-SAME: [[ARG0:%.+]]: f32, [[ARG1:%.+]]: f32 +func.func @atan2_fmf(%arg0: f32, %arg1: f32) { + // CHECK: llvm.intr.atan2([[ARG0]], [[ARG1]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32, f32) -> f32 + %0 = math.atan2 %arg0, %arg1 fastmath<fast> : f32 + func.return +} + +// ----- + // CHECK-LABEL: func @hyperbolics // CHECK-SAME: [[ARG0:%.+]]: f32 func.func @hyperbolics(%arg0: f32) { 

[llvmbot]

@llvm/pr-subscribers-mlir-math

Author: Asher Mancinelli (ashermancinelli)

Changes

asin, acos, atan, and atan2 were being lowered to libm calls instead of llvm intrinsics. Add the conversion patterns to handle these intrinsics and update tests to expect this.

NOTE: I don't know what the difference between the fast and relaxed versions of the fir tests. I followed the surrounding patterns, but relaxed and fast look identical. I expected precise to use the libm C call with no FMFs, relaxed to use the math dialect with no FMFs, and the fast to use the math dialect with FMFs.

NOTE: I don't know who to tag for math dialect changes, so feel free to tag someone else if you do!

---

Full diff: https://github.com/llvm/llvm-project/pull/141069.diff

3 Files Affected:

• (modified) flang/test/Intrinsics/math-codegen.fir (+162-8)
• (modified) mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp (+10-2)
• (modified) mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir (+78)

diff --git a/flang/test/Intrinsics/math-codegen.fir b/flang/test/Intrinsics/math-codegen.fir index c45c6b23e897e..b7c4e07130662 100644 --- a/flang/test/Intrinsics/math-codegen.fir +++ b/flang/test/Intrinsics/math-codegen.fir @@ -378,13 +378,167 @@ func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { func.func private @llvm.round.f32(f32) -> f32 func.func private @llvm.round.f64(f64) -> f64 +//--- asin_fast.fir +// RUN: fir-opt %t/asin_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/asin_fast.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.asin %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.asin %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- asin_relaxed.fir +// RUN: fir-opt %t/asin_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/asin_relaxed.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.asin({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.asin %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.asin %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- asin_precise.fir +// RUN: fir-opt %t/asin_precise.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/asin_precise.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @asinf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @asin({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = fir.call @asinf(%1) : (f32) -> f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = fir.call @asin(%1) : (f64) -> f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} +func.func private @asinf(f32) -> f32 +func.func private @asin(f64) -> f64 + +//--- acos_fast.fir +// RUN: fir-opt %t/acos_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/acos_fast.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.acos %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.acos %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- acos_relaxed.fir +// RUN: fir-opt %t/acos_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/acos_relaxed.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.acos({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = math.acos %1 : f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = math.acos %1 : f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} + +//--- acos_precise.fir +// RUN: fir-opt %t/acos_precise.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/acos_precise.fir +// CHECK: @_QPtest_real4 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @acosf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 + +// CHECK: @_QPtest_real8 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @acos({{%[A-Za-z0-9._]+}}) : (f64) -> f64 + +func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { + %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} + %1 = fir.load %arg0 : !fir.ref<f32> + %2 = fir.call @acosf(%1) : (f32) -> f32 + fir.store %2 to %0 : !fir.ref<f32> + %3 = fir.load %0 : !fir.ref<f32> + return %3 : f32 +} +func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { + %0 = fir.alloca f64 {bindc_name = "test_real8", uniq_name = "_QFtest_real8Etest_real8"} + %1 = fir.load %arg0 : !fir.ref<f64> + %2 = fir.call @acos(%1) : (f64) -> f64 + fir.store %2 to %0 : !fir.ref<f64> + %3 = fir.load %0 : !fir.ref<f64> + return %3 : f64 +} +func.func private @acosf(f32) -> f32 +func.func private @acos(f64) -> f64 + //--- atan_fast.fir // RUN: fir-opt %t/atan_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan_fast.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atanf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} @@ -406,10 +560,10 @@ func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}) -> f64 { //--- atan_relaxed.fir // RUN: fir-opt %t/atan_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan_relaxed.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atanf({{%[A-Za-z0-9._]+}}) : (f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan({{%[A-Za-z0-9._]+}}) : (f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} @@ -458,10 +612,10 @@ func.func private @atan(f64) -> f64 //--- atan2_fast.fir // RUN: fir-opt %t/atan2_fast.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan2_fast.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2f({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}, %arg1: !fir.ref<f32> {fir.bindc_name = "y"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} @@ -485,10 +639,10 @@ func.func @_QPtest_real8(%arg0: !fir.ref<f64> {fir.bindc_name = "x"}, %arg1: !fi //--- atan2_relaxed.fir // RUN: fir-opt %t/atan2_relaxed.fir --fir-to-llvm-ir="target=x86_64-unknown-linux-gnu" | FileCheck %t/atan2_relaxed.fir // CHECK: @_QPtest_real4 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2f({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f32, f32) -> f32 // CHECK: @_QPtest_real8 -// CHECK: {{%[A-Za-z0-9._]+}} = llvm.call @atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 +// CHECK: {{%[A-Za-z0-9._]+}} = llvm.intr.atan2({{%[A-Za-z0-9._]+}}, {{%[A-Za-z0-9._]+}}) : (f64, f64) -> f64 func.func @_QPtest_real4(%arg0: !fir.ref<f32> {fir.bindc_name = "x"}, %arg1: !fir.ref<f32> {fir.bindc_name = "y"}) -> f32 { %0 = fir.alloca f32 {bindc_name = "test_real4", uniq_name = "_QFtest_real4Etest_real4"} diff --git a/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp b/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp index 97da96afac4cd..b42bb773f53ee 100644 --- a/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp +++ b/mlir/lib/Conversion/MathToLLVM/MathToLLVM.cpp @@ -42,6 +42,7 @@ using CopySignOpLowering = ConvertFMFMathToLLVMPattern<math::CopySignOp, LLVM::CopySignOp>; using CosOpLowering = ConvertFMFMathToLLVMPattern<math::CosOp, LLVM::CosOp>; using CoshOpLowering = ConvertFMFMathToLLVMPattern<math::CoshOp, LLVM::CoshOp>; +using AcosOpLowering = ConvertFMFMathToLLVMPattern<math::AcosOp, LLVM::ACosOp>; using CtPopFOpLowering = VectorConvertToLLVMPattern<math::CtPopOp, LLVM::CtPopOp>; using Exp2OpLowering = ConvertFMFMathToLLVMPattern<math::Exp2Op, LLVM::Exp2Op>; @@ -62,12 +63,15 @@ using RoundOpLowering = ConvertFMFMathToLLVMPattern<math::RoundOp, LLVM::RoundOp>; using SinOpLowering = ConvertFMFMathToLLVMPattern<math::SinOp, LLVM::SinOp>; using SinhOpLowering = ConvertFMFMathToLLVMPattern<math::SinhOp, LLVM::SinhOp>; +using ASinOpLowering = ConvertFMFMathToLLVMPattern<math::AsinOp, LLVM::ASinOp>; using SqrtOpLowering = ConvertFMFMathToLLVMPattern<math::SqrtOp, LLVM::SqrtOp>; using FTruncOpLowering = ConvertFMFMathToLLVMPattern<math::TruncOp, LLVM::FTruncOp>; using TanOpLowering = ConvertFMFMathToLLVMPattern<math::TanOp, LLVM::TanOp>; using TanhOpLowering = ConvertFMFMathToLLVMPattern<math::TanhOp, LLVM::TanhOp>; - +using ATanOpLowering = ConvertFMFMathToLLVMPattern<math::AtanOp, LLVM::ATanOp>; +using ATan2OpLowering = + ConvertFMFMathToLLVMPattern<math::Atan2Op, LLVM::ATan2Op>; // A `CtLz/CtTz/absi(a)` is converted into `CtLz/CtTz/absi(a, false)`. template <typename MathOp, typename LLVMOp> struct IntOpWithFlagLowering : public ConvertOpToLLVMPattern<MathOp> { @@ -353,6 +357,7 @@ void mlir::populateMathToLLVMConversionPatterns( CopySignOpLowering, CosOpLowering, CoshOpLowering, + AcosOpLowering, CountLeadingZerosOpLowering, CountTrailingZerosOpLowering, CtPopFOpLowering, @@ -371,10 +376,13 @@ void mlir::populateMathToLLVMConversionPatterns( RsqrtOpLowering, SinOpLowering, SinhOpLowering, + ASinOpLowering, SqrtOpLowering, FTruncOpLowering, TanOpLowering, - TanhOpLowering + TanhOpLowering, + ATanOpLowering, + ATan2OpLowering >(converter, benefit); // clang-format on } diff --git a/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir b/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir index 974743a55932b..537fb967ef0e1 100644 --- a/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir +++ b/mlir/test/Conversion/MathToLLVM/math-to-llvm.mlir @@ -177,6 +177,84 @@ func.func @trigonometrics(%arg0: f32) { // ----- +// CHECK-LABEL: func @inverse_trigonometrics +// CHECK-SAME: [[ARG0:%.+]]: f32 +func.func @inverse_trigonometrics(%arg0: f32) { + // CHECK: llvm.intr.asin([[ARG0]]) : (f32) -> f32 + %0 = math.asin %arg0 : f32 + + // CHECK: llvm.intr.acos([[ARG0]]) : (f32) -> f32 + %1 = math.acos %arg0 : f32 + + // CHECK: llvm.intr.atan([[ARG0]]) : (f32) -> f32 + %2 = math.atan %arg0 : f32 + func.return +} + +// ----- + +// CHECK-LABEL: func @atan2 +// CHECK-SAME: [[ARG0:%.+]]: f32, [[ARG1:%.+]]: f32 +func.func @atan2(%arg0: f32, %arg1: f32) { + // CHECK: llvm.intr.atan2([[ARG0]], [[ARG1]]) : (f32, f32) -> f32 + %0 = math.atan2 %arg0, %arg1 : f32 + func.return +} + +// ----- + +// CHECK-LABEL: func @inverse_trigonometrics_vector +// CHECK-SAME: [[ARG0:%.+]]: vector<4xf32> +func.func @inverse_trigonometrics_vector(%arg0: vector<4xf32>) { + // CHECK: llvm.intr.asin([[ARG0]]) : (vector<4xf32>) -> vector<4xf32> + %0 = math.asin %arg0 : vector<4xf32> + + // CHECK: llvm.intr.acos([[ARG0]]) : (vector<4xf32>) -> vector<4xf32> + %1 = math.acos %arg0 : vector<4xf32> + + // CHECK: llvm.intr.atan([[ARG0]]) : (vector<4xf32>) -> vector<4xf32> + %2 = math.atan %arg0 : vector<4xf32> + func.return +} + +// ----- + +// CHECK-LABEL: func @atan2_vector +// CHECK-SAME: [[ARG0:%.+]]: vector<4xf32>, [[ARG1:%.+]]: vector<4xf32> +func.func @atan2_vector(%arg0: vector<4xf32>, %arg1: vector<4xf32>) { + // CHECK: llvm.intr.atan2([[ARG0]], [[ARG1]]) : (vector<4xf32>, vector<4xf32>) -> vector<4xf32> + %0 = math.atan2 %arg0, %arg1 : vector<4xf32> + func.return +} + +// ----- + +// CHECK-LABEL: func @inverse_trigonometrics_fmf +// CHECK-SAME: [[ARG0:%.+]]: f32 +func.func @inverse_trigonometrics_fmf(%arg0: f32) { + // CHECK: llvm.intr.asin([[ARG0]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32) -> f32 + %0 = math.asin %arg0 fastmath<fast> : f32 + + // CHECK: llvm.intr.acos([[ARG0]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32) -> f32 + %1 = math.acos %arg0 fastmath<fast> : f32 + + // CHECK: llvm.intr.atan([[ARG0]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32) -> f32 + %2 = math.atan %arg0 fastmath<fast> : f32 + func.return +} + +// ----- + +// CHECK-LABEL: func @atan2_fmf +// CHECK-SAME: [[ARG0:%.+]]: f32, [[ARG1:%.+]]: f32 +func.func @atan2_fmf(%arg0: f32, %arg1: f32) { + // CHECK: llvm.intr.atan2([[ARG0]], [[ARG1]]) {fastmathFlags = #llvm.fastmath<fast>} : (f32, f32) -> f32 + %0 = math.atan2 %arg0, %arg1 fastmath<fast> : f32 + func.return +} + +// ----- + // CHECK-LABEL: func @hyperbolics // CHECK-SAME: [[ARG0:%.+]]: f32 func.func @hyperbolics(%arg0: f32) { 

[jeanPerier]

Thanks, looks good to me.

Nit: should add [mlir][math] to your commit/PR title.

I agree it is a bit odd for the relaxed/fast FIR tests to be exactly the same. Makes sense to me to add some FMF to the fast case and see check how that propagates in a separate patch.

@zero9178 or @matthias-springer for more visibility in MLIR reviewers.