From ffe6a5832533e71e1edc4fe0eebda421f24fb61a Mon Sep 17 00:00:00 2001
From: Adrian Kuegel <akuegel@google.com>
Date: Thu, 15 Jul 2021 12:04:25 +0200
Subject: [PATCH] [mlir][nvvm]: Add math::Exp2Op lowering to NVVM.

Differential Revision: https://reviews.llvm.org/D106050
---
 mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp |  8 +++++---
 mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir        | 14 ++++++++++++++
 2 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
index 243ed6c..4c8ad15 100644
--- a/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
+++ b/mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp
@@ -188,9 +188,9 @@ void mlir::configureGpuToNVVMConversionLegality(ConversionTarget &target) {
   target.addLegalDialect<::mlir::LLVM::LLVMDialect>();
   target.addLegalDialect<::mlir::NVVM::NVVMDialect>();
   target.addIllegalDialect<gpu::GPUDialect>();
-  target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp,
-                      LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op, LLVM::Log2Op,
-                      LLVM::PowOp, LLVM::SinOp, LLVM::SqrtOp>();
+  target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::Exp2Op, LLVM::FAbsOp,
+                      LLVM::FCeilOp, LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op,
+                      LLVM::Log2Op, LLVM::PowOp, LLVM::SinOp, LLVM::SqrtOp>();
 
   // TODO: Remove once we support replacing non-root ops.
   target.addLegalOp<gpu::YieldOp, gpu::GPUModuleOp, gpu::ModuleEndOp>();
@@ -230,6 +230,8 @@ void mlir::populateGpuToNVVMConversionPatterns(LLVMTypeConverter &converter,
                                                   "__nv_cos");
   patterns.add<OpToFuncCallLowering<math::ExpOp>>(converter, "__nv_expf",
                                                   "__nv_exp");
+  patterns.add<OpToFuncCallLowering<math::Exp2Op>>(converter, "__nv_exp2f",
+                                                   "__nv_exp2");
   patterns.add<OpToFuncCallLowering<math::ExpM1Op>>(converter, "__nv_expm1f",
                                                     "__nv_expm1");
   patterns.add<OpToFuncCallLowering<FloorFOp>>(converter, "__nv_floorf",
diff --git a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
index 1aaa4cd..415090e 100644
--- a/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
+++ b/mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir
@@ -215,6 +215,20 @@ gpu.module @test_module {
 }
 
 // -----
+gpu.module @test_module {
+  // CHECK: llvm.func @__nv_exp2f(f32) -> f32
+  // CHECK: llvm.func @__nv_exp2(f64) -> f64
+  // CHECK-LABEL: func @gpu_exp2
+  func @gpu_exp2(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = math.exp2 %arg_f32 : f32
+    // CHECK: llvm.call @__nv_exp2f(%{{.*}}) : (f32) -> f32
+    %result64 = math.exp2 %arg_f64 : f64
+    // CHECK: llvm.call @__nv_exp2(%{{.*}}) : (f64) -> f64
+    std.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
 
 gpu.module @test_module {
   // CHECK: llvm.func @__nv_logf(f32) -> f32
-- 
2.7.4