[Mlir-commits] [mlir] [MLIR][Math][GPU] Add lowering of absi and fpowi to libdevice (PR #123422)

[William Moses]

https://github.com/wsmoses created https://github.com/llvm/llvm-project/pull/123422

None

>From 8461886b0b780d9a7d57ec6f4cc7a625e127eb26 Mon Sep 17 00:00:00 2001
From: "William S. Moses" <gh at wsmoses.com>
Date: Fri, 17 Jan 2025 17:29:34 -0600
Subject: [PATCH] [MLIR][Math][GPU] Add lowering of absi and fpowi to libdevice

---
 .../GPUToNVVM/LowerGpuOpsToNVVMOps.cpp        |  2 ++
 .../Conversion/GPUToNVVM/gpu-to-nvvm.mlir     | 24 +++++++++++++++++++
 2 files changed, 26 insertions(+)

diff --git a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
index 2768929f460e2e..bbffec02009ea6 100644
--- a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
+++ b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
@@ -509,6 +509,7 @@ void mlir::populateGpuToNVVMConversionPatterns(
 
   populateOpPatterns<arith::RemFOp>(converter, patterns, "__nv_fmodf",
                                     "__nv_fmod");
+  populateOpPatterns<math::AbsIOp>(converter, patterns, "__nv_abs");
   populateOpPatterns<math::AbsFOp>(converter, patterns, "__nv_fabsf",
                                    "__nv_fabs");
   populateOpPatterns<math::AcosOp>(converter, patterns, "__nv_acosf",
@@ -555,6 +556,7 @@ void mlir::populateGpuToNVVMConversionPatterns(
                                    "__nv_log2", "__nv_fast_log2f");
   populateOpPatterns<math::PowFOp>(converter, patterns, "__nv_powf", "__nv_pow",
                                    "__nv_fast_powf");
+  populateOpPatterns<math::FPowIOp>(converter, patterns, "__nv_powif", "__nv_powi");
   populateOpPatterns<math::RoundOp>(converter, patterns, "__nv_roundf",
                                     "__nv_round");
   populateOpPatterns<math::RoundEvenOp>(converter, patterns, "__nv_rintf",
diff --git a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
index f52dd6c0d0ce30..b3df7186dab1ef 100644
--- a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
+++ b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
@@ -1033,3 +1033,27 @@ module attributes {transform.with_named_sequence} {
     transform.yield
   }
 }
+
+
+gpu.module @test_module_52 {
+  // CHECK: llvm.func @__nv_abs(i32) -> i32
+  // CHECK-LABEL: func @gpu_abs
+  func.func @gpu_fabs(%arg_i32 : i32) -> (i32) {
+    %result32 = math.absi %arg_i32 : i32
+    // CHECK: llvm.call @__nv_abs(%{{.*}}) : (i32) -> i32
+    func.return %result32 : i32
+  }
+}
+
+gpu.module @test_module_26 {
+  // CHECK: llvm.func @__nv_powif(f32, i32) -> f32
+  // CHECK: llvm.func @__nv_powi(f64, i32) -> f64
+  // CHECK-LABEL: func @gpu_powi
+  func.func @gpu_powi(%arg_f32 : f32, %arg_f64 : f64, %arg_i32 : i32) -> (f32, f64) {
+    %result32 = math.fpowi %arg_f32, %arg_i32 : f32
+    // CHECK: llvm.call @__nv_powf(%{{.*}}, %{{.*}}) : (f32, i32) -> f32
+    %result64 = math.fpowi %arg_f64, %arg_i64 : f64
+    // CHECK: llvm.call @__nv_pow(%{{.*}}, %{{.*}}) : (f64, i32) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}