IndexIntrinsicsOpLowering.h
2.9 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
//===- IndexIntrinsicsOpLowering.h - GPU IndexOps Lowering class *- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef MLIR_CONVERSION_GPUCOMMON_INDEXINTRINSICSOPLOWERING_H_
#define MLIR_CONVERSION_GPUCOMMON_INDEXINTRINSICSOPLOWERING_H_
#include "mlir/Conversion/StandardToLLVM/ConvertStandardToLLVM.h"
#include "mlir/Dialect/GPU/GPUDialect.h"
#include "mlir/Dialect/LLVMIR/LLVMDialect.h"
#include "llvm/ADT/StringSwitch.h"
namespace mlir {
// Rewriting that replaces Op with XOp, YOp, or ZOp depending on the dimension
// that Op operates on. Op is assumed to return an `std.index` value and
// XOp, YOp and ZOp are assumed to return an `llvm.i32` value. Depending on
// `indexBitwidth`, sign-extend or truncate the resulting value to match the
// bitwidth expected by the consumers of the value.
template <typename Op, typename XOp, typename YOp, typename ZOp>
struct GPUIndexIntrinsicOpLowering : public ConvertToLLVMPattern {
private:
enum dimension { X = 0, Y = 1, Z = 2, invalid };
unsigned indexBitwidth;
static dimension dimensionToIndex(Op op) {
return llvm::StringSwitch<dimension>(op.dimension())
.Case("x", X)
.Case("y", Y)
.Case("z", Z)
.Default(invalid);
}
public:
explicit GPUIndexIntrinsicOpLowering(LLVMTypeConverter &typeConverter)
: ConvertToLLVMPattern(Op::getOperationName(),
typeConverter.getDialect()->getContext(),
typeConverter),
indexBitwidth(typeConverter.getIndexTypeBitwidth()) {}
// Convert the kernel arguments to an LLVM type, preserve the rest.
LogicalResult
matchAndRewrite(Operation *op, ArrayRef<Value> operands,
ConversionPatternRewriter &rewriter) const override {
auto loc = op->getLoc();
MLIRContext *context = rewriter.getContext();
Value newOp;
switch (dimensionToIndex(cast<Op>(op))) {
case X:
newOp = rewriter.create<XOp>(loc, LLVM::LLVMType::getInt32Ty(context));
break;
case Y:
newOp = rewriter.create<YOp>(loc, LLVM::LLVMType::getInt32Ty(context));
break;
case Z:
newOp = rewriter.create<ZOp>(loc, LLVM::LLVMType::getInt32Ty(context));
break;
default:
return failure();
}
if (indexBitwidth > 32) {
newOp = rewriter.create<LLVM::SExtOp>(
loc, LLVM::LLVMType::getIntNTy(context, indexBitwidth), newOp);
} else if (indexBitwidth < 32) {
newOp = rewriter.create<LLVM::TruncOp>(
loc, LLVM::LLVMType::getIntNTy(context, indexBitwidth), newOp);
}
rewriter.replaceOp(op, {newOp});
return success();
}
};
} // namespace mlir
#endif // MLIR_CONVERSION_GPUCOMMON_INDEXINTRINSICSOPLOWERING_H_