1 /* Copyright 2020 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 #include <utility>
17
18 #include "mlir-hlo/Transforms/GPUPassDetail.h"
19 #include "mlir-hlo/Transforms/gpu_passes.h"
20 #include "mlir/Conversion/ArithmeticToLLVM/ArithmeticToLLVM.h"
21 #include "mlir/Conversion/ComplexToLLVM/ComplexToLLVM.h"
22 #include "mlir/Conversion/ControlFlowToLLVM/ControlFlowToLLVM.h"
23 #include "mlir/Conversion/FuncToLLVM/ConvertFuncToLLVM.h"
24 #include "mlir/Conversion/GPUToNVVM/GPUToNVVMPass.h"
25 #include "mlir/Conversion/GPUToROCDL/GPUToROCDLPass.h"
26 #include "mlir/Conversion/LLVMCommon/LoweringOptions.h"
27 #include "mlir/Conversion/LLVMCommon/TypeConverter.h"
28 #include "mlir/Conversion/MathToLLVM/MathToLLVM.h"
29 #include "mlir/Conversion/MemRefToLLVM/MemRefToLLVM.h"
30 #include "mlir/Dialect/GPU/IR/GPUDialect.h"
31 #include "mlir/Transforms/DialectConversion.h"
32
33 using namespace mlir;
34
35 namespace {
36
37 /// A pass that does the final lowering to NVVM. It collects all the patterns
38 /// that are currently required, currently mixing std, linalg and gpu.
39 class GpuKernelToNVVMPass
40 : public GpuKernelToNVVMPassBase<GpuKernelToNVVMPass> {
41 void runOnOperation() override;
42 };
43
44 /// A pass that does the final lowering to ROCDL. It collects all the patterns
45 /// that are currently required, currently mixing std, linalg and gpu.
46 class GpuKernelToROCDLPass
47 : public GpuKernelToROCDLPassBase<GpuKernelToROCDLPass> {
48 void runOnOperation() override;
49 };
50
51 } // namespace
52
populateCommonPatterns(LLVMTypeConverter & converter,RewritePatternSet & patterns)53 static void populateCommonPatterns(LLVMTypeConverter& converter,
54 RewritePatternSet& patterns) {
55 arith::populateArithmeticToLLVMConversionPatterns(converter, patterns);
56 populateMathToLLVMConversionPatterns(converter, patterns);
57 populateMemRefToLLVMConversionPatterns(converter, patterns);
58 populateFuncToLLVMConversionPatterns(converter, patterns);
59 cf::populateControlFlowToLLVMConversionPatterns(converter, patterns);
60 populateComplexToLLVMConversionPatterns(converter, patterns);
61 }
62
runOnOperation()63 void GpuKernelToNVVMPass::runOnOperation() {
64 RewritePatternSet patterns(&getContext());
65 LowerToLLVMOptions llvmOpts(&getContext(), DataLayout(getOperation()));
66 LLVMTypeConverter converter(&getContext(), llvmOpts);
67 populateCommonPatterns(converter, patterns);
68 populateGpuToNVVMConversionPatterns(converter, patterns);
69 ConversionTarget target(getContext());
70 configureGpuToNVVMConversionLegality(target);
71 if (failed(
72 applyFullConversion(getOperation(), target, std::move(patterns)))) {
73 signalPassFailure();
74 }
75 }
76
runOnOperation()77 void GpuKernelToROCDLPass::runOnOperation() {
78 RewritePatternSet patterns(&getContext());
79 LLVMTypeConverter converter(&getContext());
80 populateCommonPatterns(converter, patterns);
81 populateGpuToROCDLConversionPatterns(converter, patterns,
82 gpu::amd::Runtime::Unknown);
83 ConversionTarget target(getContext());
84 configureGpuToROCDLConversionLegality(target);
85 if (failed(
86 applyFullConversion(getOperation(), target, std::move(patterns)))) {
87 signalPassFailure();
88 }
89 }
90
91 std::unique_ptr<OperationPass<gpu::GPUModuleOp> >
createGpuKernelToNvvmPass()92 mlir::createGpuKernelToNvvmPass() {
93 return std::make_unique<GpuKernelToNVVMPass>();
94 }
95
96 std::unique_ptr<OperationPass<gpu::GPUModuleOp> >
createGpuKernelToRocdlPass()97 mlir::createGpuKernelToRocdlPass() {
98 return std::make_unique<GpuKernelToROCDLPass>();
99 }
100