[llvm-branch-commits] [mlir] 75d9a46 - [MLIR] Add atan and atan2 lowerings to CUDA intrinsics

[Frederik Gossen via llvm-branch-commits]

Author: Frederik Gossen
Date: 2020-12-14T10:45:28+01:00
New Revision: 75d9a46090249ed8abfafea090e686b4fecf182f

URL: https://github.com/llvm/llvm-project/commit/75d9a46090249ed8abfafea090e686b4fecf182f
DIFF: https://github.com/llvm/llvm-project/commit/75d9a46090249ed8abfafea090e686b4fecf182f.diff

LOG: [MLIR] Add atan and atan2 lowerings to CUDA intrinsics

Differential Revision: https://reviews.llvm.org/D93124

Added: 
    

Modified: 
    mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
    mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir

Removed: 
    


################################################################################
diff  --git a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
index 3e90894e2fe9..a8b4d074a08e 100644
--- a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
+++ b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
@@ -173,6 +173,10 @@ void mlir::populateGpuToNVVMConversionPatterns(
               GPUFuncOpLowering<0>>(converter);
   patterns.insert<OpToFuncCallLowering<AbsFOp>>(converter, "__nv_fabsf",
                                                 "__nv_fabs");
+  patterns.insert<OpToFuncCallLowering<AtanOp>>(converter, "__nv_atanf",
+                                                "__nv_atan");
+  patterns.insert<OpToFuncCallLowering<Atan2Op>>(converter, "__nv_atan2f",
+                                                 "__nv_atan2");
   patterns.insert<OpToFuncCallLowering<CeilFOp>>(converter, "__nv_ceilf",
                                                  "__nv_ceil");
   patterns.insert<OpToFuncCallLowering<CosOp>>(converter, "__nv_cosf",

diff  --git a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
index 1c0257de0d55..d8c4d7e064e5 100644
--- a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
+++ b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
@@ -335,6 +335,47 @@ gpu.module @test_module {
 
 // -----
 
+gpu.module @test_module {
+  // CHECK: llvm.func @__nv_atanf(!llvm.float) -> !llvm.float
+  // CHECK: llvm.func @__nv_atan(!llvm.double) -> !llvm.double
+  // CHECK-LABEL: func @gpu_atan
+  func @gpu_atan(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64)
+      -> (f16, f32, f64) {
+    %result16 = std.atan %arg_f16 : f16
+    // CHECK: llvm.fpext %{{.*}} : !llvm.half to !llvm.float
+    // CHECK-NEXT: llvm.call @__nv_atanf(%{{.*}}) : (!llvm.float) -> !llvm.float
+    // CHECK-NEXT: llvm.fptrunc %{{.*}} : !llvm.float to !llvm.half
+    %result32 = std.atan %arg_f32 : f32
+    // CHECK: llvm.call @__nv_atanf(%{{.*}}) : (!llvm.float) -> !llvm.float
+    %result64 = std.atan %arg_f64 : f64
+    // CHECK: llvm.call @__nv_atan(%{{.*}}) : (!llvm.double) -> !llvm.double
+    std.return %result16, %result32, %result64 : f16, f32, f64
+  }
+}
+
+// -----
+
+gpu.module @test_module {
+  // CHECK: llvm.func @__nv_atan2f(!llvm.float, !llvm.float) -> !llvm.float
+  // CHECK: llvm.func @__nv_atan2(!llvm.double, !llvm.double) -> !llvm.double
+  // CHECK-LABEL: func @gpu_atan2
+  func @gpu_atan2(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64)
+      -> (f16, f32, f64) {
+    %result16 = std.atan2 %arg_f16, %arg_f16 : f16
+    // CHECK: llvm.fpext %{{.*}} : !llvm.half to !llvm.float
+    // CHECK: llvm.fpext %{{.*}} : !llvm.half to !llvm.float
+    // CHECK-NEXT: llvm.call @__nv_atan2f(%{{.*}}) : (!llvm.float, !llvm.float) -> !llvm.float
+    // CHECK-NEXT: llvm.fptrunc %{{.*}} : !llvm.float to !llvm.half
+    %result32 = std.atan2 %arg_f32, %arg_f32 : f32
+    // CHECK: llvm.call @__nv_atan2f(%{{.*}}) : (!llvm.float, !llvm.float) -> !llvm.float
+    %result64 = std.atan2 %arg_f64, %arg_f64 : f64
+    // CHECK: llvm.call @__nv_atan2(%{{.*}}) : (!llvm.double, !llvm.double) -> !llvm.double
+    std.return %result16, %result32, %result64 : f16, f32, f64
+  }
+}
+
+// -----
+
 // Test that we handled properly operation with SymbolTable other than module op
 gpu.module @test_module {
   "test.symbol_scope"() ({