xref: /llvm-project/mlir/lib/Conversion/VectorToLLVM/ConvertVectorToLLVMPass.cpp (revision 039b969b32b64b64123dce30dd28ec4e343d893f)
1 //===- VectorToLLVM.cpp - Conversion from Vector to the LLVM dialect ------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "mlir/Conversion/VectorToLLVM/ConvertVectorToLLVM.h"
10 
11 #include "../PassDetail.h"
12 
13 #include "mlir/Conversion/LLVMCommon/ConversionTarget.h"
14 #include "mlir/Conversion/LLVMCommon/TypeConverter.h"
15 #include "mlir/Dialect/AMX/AMXDialect.h"
16 #include "mlir/Dialect/AMX/Transforms.h"
17 #include "mlir/Dialect/Arithmetic/IR/Arithmetic.h"
18 #include "mlir/Dialect/ArmNeon/ArmNeonDialect.h"
19 #include "mlir/Dialect/ArmSVE/ArmSVEDialect.h"
20 #include "mlir/Dialect/ArmSVE/Transforms.h"
21 #include "mlir/Dialect/Func/IR/FuncOps.h"
22 #include "mlir/Dialect/LLVMIR/LLVMDialect.h"
23 #include "mlir/Dialect/MemRef/IR/MemRef.h"
24 #include "mlir/Dialect/Vector/Transforms/VectorRewritePatterns.h"
25 #include "mlir/Dialect/X86Vector/Transforms.h"
26 #include "mlir/Dialect/X86Vector/X86VectorDialect.h"
27 #include "mlir/Transforms/GreedyPatternRewriteDriver.h"
28 
29 using namespace mlir;
30 using namespace mlir::vector;
31 
32 namespace {
33 struct LowerVectorToLLVMPass
34     : public ConvertVectorToLLVMBase<LowerVectorToLLVMPass> {
35   LowerVectorToLLVMPass(const LowerVectorToLLVMOptions &options) {
36     this->reassociateFPReductions = options.reassociateFPReductions;
37     this->force32BitVectorIndices = options.force32BitVectorIndices;
38     this->armNeon = options.armNeon;
39     this->armSVE = options.armSVE;
40     this->amx = options.amx;
41     this->x86Vector = options.x86Vector;
42   }
43   // Override explicitly to allow conditional dialect dependence.
44   void getDependentDialects(DialectRegistry &registry) const override {
45     registry.insert<LLVM::LLVMDialect>();
46     registry.insert<arith::ArithmeticDialect>();
47     registry.insert<memref::MemRefDialect>();
48     if (armNeon)
49       registry.insert<arm_neon::ArmNeonDialect>();
50     if (armSVE)
51       registry.insert<arm_sve::ArmSVEDialect>();
52     if (amx)
53       registry.insert<amx::AMXDialect>();
54     if (x86Vector)
55       registry.insert<x86vector::X86VectorDialect>();
56   }
57   void runOnOperation() override;
58 };
59 } // namespace
60 
61 void LowerVectorToLLVMPass::runOnOperation() {
62   // Perform progressive lowering of operations on slices and
63   // all contraction operations. Also applies folding and DCE.
64   {
65     RewritePatternSet patterns(&getContext());
66     populateVectorToVectorCanonicalizationPatterns(patterns);
67     populateVectorBroadcastLoweringPatterns(patterns);
68     populateVectorContractLoweringPatterns(patterns);
69     populateVectorMaskOpLoweringPatterns(patterns);
70     populateVectorShapeCastLoweringPatterns(patterns);
71     populateVectorTransposeLoweringPatterns(patterns);
72     // Vector transfer ops with rank > 1 should be lowered with VectorToSCF.
73     populateVectorTransferLoweringPatterns(patterns, /*maxTransferRank=*/1);
74     (void)applyPatternsAndFoldGreedily(getOperation(), std::move(patterns));
75   }
76 
77   // Convert to the LLVM IR dialect.
78   LLVMTypeConverter converter(&getContext());
79   RewritePatternSet patterns(&getContext());
80   populateVectorMaskMaterializationPatterns(patterns, force32BitVectorIndices);
81   populateVectorTransferLoweringPatterns(patterns);
82   populateVectorToLLVMMatrixConversionPatterns(converter, patterns);
83   populateVectorToLLVMConversionPatterns(
84       converter, patterns, reassociateFPReductions, force32BitVectorIndices);
85   populateVectorToLLVMMatrixConversionPatterns(converter, patterns);
86 
87   // Architecture specific augmentations.
88   LLVMConversionTarget target(getContext());
89   target.addLegalDialect<arith::ArithmeticDialect>();
90   target.addLegalDialect<memref::MemRefDialect>();
91   target.addLegalOp<UnrealizedConversionCastOp>();
92   if (armNeon) {
93     // TODO: we may or may not want to include in-dialect lowering to
94     // LLVM-compatible operations here. So far, all operations in the dialect
95     // can be translated to LLVM IR so there is no conversion necessary.
96     target.addLegalDialect<arm_neon::ArmNeonDialect>();
97   }
98   if (armSVE) {
99     configureArmSVELegalizeForExportTarget(target);
100     populateArmSVELegalizeForLLVMExportPatterns(converter, patterns);
101   }
102   if (amx) {
103     configureAMXLegalizeForExportTarget(target);
104     populateAMXLegalizeForLLVMExportPatterns(converter, patterns);
105   }
106   if (x86Vector) {
107     configureX86VectorLegalizeForExportTarget(target);
108     populateX86VectorLegalizeForLLVMExportPatterns(converter, patterns);
109   }
110 
111   if (failed(
112           applyPartialConversion(getOperation(), target, std::move(patterns))))
113     signalPassFailure();
114 }
115 
116 std::unique_ptr<OperationPass<ModuleOp>>
117 mlir::createConvertVectorToLLVMPass(const LowerVectorToLLVMOptions &options) {
118   return std::make_unique<LowerVectorToLLVMPass>(options);
119 }
120