MLIR: add SinOp Lowering to __nv_sinf and __nv_sin

[qliu93]

Summary

Added lowering rule from SinOp to __nv_sinf and __nv_sin

[ftynse]

You likely want to declare LLVM::SinOp as illegal in the conversion setup ~50 lines above.

Also, please make sure to upload the patch with full context or use arcanist.

[qliu93]

addIllegalOp

[qliu93]

Hi, I add addIllegalOp and the corresponding test, please review again, thanks 😃

[ftynse]

mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp

141

Let's not things irrelevant to the current commit (TanhOp)

[qliu93]

Remove irrelevant TanhOp

[qliu93]

This is the first time that I submit codes into llvm project. After that this revision is accepted, what should I do next to merge into master branch? Thanks in advance.

[ftynse]

I can land it for you, but I will need your name and email as they should appear on the commit. (Using arcanist - https://llvm.org/docs/Phabricator.html#requesting-a-review-via-the-command-line - automates this).

[qliu93]

My name is "Qingyi Liu" and my email is [e-mail]. Thanks for adding commit message for me : ) I will learn to use arcanist and do it by myself next time.

mlir/lib/Conversion/GPUToNVVM/LowerGpuOpsToNVVMOps.cpp

void runOnOperation() override {
     applyPatternsAndFoldGreedily(m, std::move(patterns));
 
     populateStdToLLVMConversionPatterns(converter, llvmPatterns);
     populateGpuToNVVMConversionPatterns(converter, llvmPatterns);
     LLVMConversionTarget target(getContext());
     target.addIllegalDialect<gpu::GPUDialect>();
     target.addIllegalOp<LLVM::CosOp, LLVM::ExpOp, LLVM::FAbsOp, LLVM::FCeilOp,
                         LLVM::FFloorOp, LLVM::LogOp, LLVM::Log10Op,
-                        LLVM::Log2Op>();
+                        LLVM::Log2Op, LLVM::SinOp>();
     target.addIllegalOp<FuncOp>();
     target.addLegalDialect<NVVM::NVVMDialect>();
     // TODO: Remove once we support replacing non-root ops.
     target.addLegalOp<gpu::YieldOp, gpu::GPUModuleOp, gpu::ModuleEndOp>();
     if (failed(applyPartialConversion(m, target, std::move(llvmPatterns))))
       signalPassFailure();
   }
 };
void mlir::populateGpuToNVVMConversionPatterns(
   patterns.insert<OpToFuncCallLowering<FloorFOp>>(converter, "__nv_floorf",
                                                   "__nv_floor");
   patterns.insert<OpToFuncCallLowering<LogOp>>(converter, "__nv_logf",
                                                "__nv_log");
   patterns.insert<OpToFuncCallLowering<Log10Op>>(converter, "__nv_log10f",
                                                  "__nv_log10");
   patterns.insert<OpToFuncCallLowering<Log2Op>>(converter, "__nv_log2f",
                                                 "__nv_log2");
+  patterns.insert<OpToFuncCallLowering<SinOp>>(converter, "__nv_sinf",
+                                               "__nv_sin");
   patterns.insert<OpToFuncCallLowering<TanhOp>>(converter, "__nv_tanhf",
                                                 "__nv_tanh");
 }
 
 std::unique_ptr<OperationPass<gpu::GPUModuleOp>>
 mlir::createLowerGpuOpsToNVVMOpsPass(unsigned indexBitwidth) {
   return std::make_unique<LowerGpuOpsToNVVMOpsPass>(indexBitwidth);
 }

mlir/test/Conversion/GPUToNVVM/gpu-to-nvvm.mlir

func @gpu_log2(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
     // CHECK: llvm.call @__nv_log2(%{{.*}}) : (!llvm.double) -> !llvm.double
     std.return %result32, %result64 : f32, f64
   }
 }
 
 // -----
 
 gpu.module @test_module {
+  // CHECK: llvm.func @__nv_sinf(!llvm.float) -> !llvm.float
+  // CHECK: llvm.func @__nv_sin(!llvm.double) -> !llvm.double
+  // CHECK-LABEL: func @gpu_sin
+  func @gpu_sin(%arg_f32 : f32, %arg_f64 : f64) -> (f32, f64) {
+    %result32 = std.sin %arg_f32 : f32
+    // CHECK: llvm.call @__nv_sinf(%{{.*}}) : (!llvm.float) -> !llvm.float
+    %result64 = std.sin %arg_f64 : f64
+    // CHECK: llvm.call @__nv_sin(%{{.*}}) : (!llvm.double) -> !llvm.double
+    std.return %result32, %result64 : f32, f64
+  }
+}
+
+// -----
+
+gpu.module @test_module {
   // CHECK: llvm.func @__nv_tanhf(!llvm.float) -> !llvm.float
   // CHECK: llvm.func @__nv_tanh(!llvm.double) -> !llvm.double
   // CHECK-LABEL: func @gpu_tanh
   func @gpu_tanh(%arg_f16 : f16, %arg_f32 : f32, %arg_f64 : f64) -> (f16, f32, f64) {
     %result16 = std.tanh %arg_f16 : f16
     // CHECK: llvm.fpext %{{.*}} : !llvm.half to !llvm.float
     // CHECK-NEXT: llvm.call @__nv_tanhf(%{{.*}}) : (!llvm.float) -> !llvm.float
     // CHECK-NEXT: llvm.fptrunc %{{.*}} : !llvm.float to !llvm.half