Lower math.cbrt to NVVM/ROCDL.

Reviewed By: pifon2a

Differential Revision: https://reviews.llvm.org/D141270

diff --git a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
index a822cce..5050f6b 100644 (file)
--- a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
+++ b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
@@ -266,6 +266,8 @@ void mlir::populateGpuToNVVMConversionPatterns(LLVMTypeConverter &converter,
                                    "__nv_atan");
   populateOpPatterns<math::Atan2Op>(converter, patterns, "__nv_atan2f",
                                     "__nv_atan2");
+  populateOpPatterns<math::CbrtOp>(converter, patterns, "__nv_cbrtf",
+                                   "__nv_cbrt");
   populateOpPatterns<math::CeilOp>(converter, patterns, "__nv_ceilf",
                                    "__nv_ceil");
   populateOpPatterns<math::CosOp>(converter, patterns, "__nv_cosf", "__nv_cos");

diff --git a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
index 1f34fe6..fead289 100644 (file)
--- a/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
+++ b/mlir/lib/Conversion/GPUToROCDL/LowerGpuOpsToROCDLOps.cpp
@@ -224,6 +224,8 @@ void mlir::populateGpuToROCDLConversionPatterns(
                                    "__ocml_atan_f64");
   populateOpPatterns<math::Atan2Op>(converter, patterns, "__ocml_atan2_f32",
                                     "__ocml_atan2_f64");
+  populateOpPatterns<math::CbrtOp>(converter, patterns, "__ocml_cbrt_f32",
+                                   "__ocml_cbrt_f64");
   populateOpPatterns<math::CeilOp>(converter, patterns, "__ocml_ceil_f32",
                                    "__ocml_ceil_f64");
   populateOpPatterns<math::CosOp>(converter, patterns, "__ocml_cos_f32",

diff --git a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
index 230cd4a..40035fa 100644 (file)
--- a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
+++ b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
@@ -181,6 +181,21 @@ gpu.module @test_module {
 // -----
 
 gpu.module @test_module {
+  // CHECK: llvm.func @__nv_cbrtf(f32) -> f32
+  // CHECK: llvm.func @__nv_cbrt(f64) -> f64
+  // CHECK-LABEL: func @gpu_cbrt
+  func.func @gpu_cbrt(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.cbrt %arg_f32 : f32
+    // CHECK: llvm.call @__nv_cbrtf(%{{.*}}) : (f32) -> f32
+    %result64 = math.cbrt %arg_f64 : f64
+    // CHECK: llvm.call @__nv_cbrt(%{{.*}}) : (f64) -> f64
+    func.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+gpu.module @test_module {
   // CHECK: llvm.func @__nv_ceilf(f32) -> f32
   // CHECK: llvm.func @__nv_ceil(f64) -> f64
   // CHECK-LABEL: func @gpu_ceil
@@ -526,7 +541,7 @@ gpu.module @test_module {
   gpu.func @test_printf(%arg0: i32, %arg1: f32) {
     // CHECK-NEXT: %[[FORMATSTR:.*]] = llvm.mlir.addressof @[[$PRINT_GLOBAL1]] : !llvm.ptr<array<11 x i8>>
     // CHECK-NEXT: %[[FORMATSTART:.*]] = llvm.getelementptr %[[FORMATSTR]][0, 0] : (!llvm.ptr<array<11 x i8>>) -> !llvm.ptr<i8>
-    // CHECK-NEXT: %[[EXT:.+]] = llvm.fpext %[[ARG1]] : f32 to f64 
+    // CHECK-NEXT: %[[EXT:.+]] = llvm.fpext %[[ARG1]] : f32 to f64
     // CHECK-NEXT: %[[O:.*]] = llvm.mlir.constant(1 : index) : i64
     // CHECK-NEXT: %[[ALLOC:.*]] = llvm.alloca %[[O]] x !llvm.struct<(i32, f64)> : (i64) -> !llvm.ptr<struct<(i32, f64)>>
     // CHECK-NEXT: %[[EL0:.*]] = llvm.getelementptr %[[ALLOC]][0, 0] : (!llvm.ptr<struct<(i32, f64)>>) -> !llvm.ptr<i32>

diff --git a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
index 5751d76..0403cdb 100644 (file)
--- a/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
+++ b/mlir/test/Conversion/GPUToROCDL/gpu-to-rocdl.mlir
@@ -130,6 +130,21 @@ gpu.module @test_module {
 // -----
 
 gpu.module @test_module {
+  // CHECK: llvm.func @__ocml_cbrt_f32(f32) -> f32
+  // CHECK: llvm.func @__ocml_cbrt_f64(f64) -> f64
+  // CHECK-LABEL: func @gpu_cbrt
+  func.func @gpu_cbrt(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %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 %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+gpu.module @test_module {
   // CHECK: llvm.func @__ocml_ceil_f32(f32) -> f32
   // CHECK: llvm.func @__ocml_ceil_f64(f64) -> f64
   // CHECK-LABEL: func @gpu_ceil