| 1 | //===- TestXeGPUTransforms.cpp -- Test Vector transforms and lowerings ----===// |
| 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/Dialect/GPU/IR/GPUDialect.h" |
| 10 | #include "mlir/Dialect/Vector/Transforms/VectorTransforms.h" |
| 11 | #include "mlir/Dialect/XeGPU/IR/XeGPU.h" |
| 12 | #include "mlir/Dialect/XeGPU/Transforms/Transforms.h" |
| 13 | #include "mlir/Pass/Pass.h" |
| 14 | #include "mlir/Pass/PassManager.h" |
| 15 | #include "mlir/Transforms/GreedyPatternRewriteDriver.h" |
| 16 | |
| 17 | using namespace mlir; |
| 18 | using namespace mlir::xegpu; |
| 19 | |
| 20 | namespace { |
| 21 | |
| 22 | struct TestXeGPUUnrollingPatterns |
| 23 | : public PassWrapper<TestXeGPUUnrollingPatterns, |
| 24 | OperationPass<gpu::GPUModuleOp>> { |
| 25 | MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(TestXeGPUUnrollingPatterns) |
| 26 | |
| 27 | StringRef getArgument() const final { |
| 28 | return "test-xegpu-unrolling-patterns" ; |
| 29 | } |
| 30 | |
| 31 | StringRef getDescription() const final { |
| 32 | return "Test lowering patterns to unroll ops in the xegpu dialect" ; |
| 33 | } |
| 34 | |
| 35 | void getDependentDialects(::mlir::DialectRegistry ®istry) const override { |
| 36 | registry.insert<memref::MemRefDialect>(); |
| 37 | registry.insert<xegpu::XeGPUDialect>(); |
| 38 | registry.insert<vector::VectorDialect>(); |
| 39 | } |
| 40 | |
| 41 | TestXeGPUUnrollingPatterns() = default; |
| 42 | TestXeGPUUnrollingPatterns(const TestXeGPUUnrollingPatterns &pass) |
| 43 | : PassWrapper(pass) {} |
| 44 | |
| 45 | void runOnOperation() override { |
| 46 | MLIRContext *ctx = &getContext(); |
| 47 | xegpu::UnrollOptions options; |
| 48 | options.setNativeShapeFn( |
| 49 | [&](Operation *op) -> std::optional<SmallVector<int64_t>> { |
| 50 | if (isa<xegpu::CreateNdDescOp, xegpu::UpdateNdOffsetOp, |
| 51 | xegpu::PrefetchNdOp, xegpu::LoadNdOp, xegpu::StoreNdOp>(op)) { |
| 52 | xegpu::TensorDescType tdescTy; |
| 53 | if (auto createNdOp = dyn_cast<xegpu::CreateNdDescOp>(op)) { |
| 54 | tdescTy = createNdOp.getType(); |
| 55 | } else if (auto updateNdOp = |
| 56 | dyn_cast<xegpu::UpdateNdOffsetOp>(op)) { |
| 57 | tdescTy = updateNdOp.getTensorDescType(); |
| 58 | } else if (auto prefetchNdOp = dyn_cast<xegpu::PrefetchNdOp>(op)) { |
| 59 | tdescTy = prefetchNdOp.getTensorDescType(); |
| 60 | } else if (auto loadNdOp = dyn_cast<xegpu::LoadNdOp>(op)) { |
| 61 | tdescTy = loadNdOp.getTensorDescType(); |
| 62 | } else if (auto storeNdOp = dyn_cast<xegpu::StoreNdOp>(op)) { |
| 63 | tdescTy = storeNdOp.getTensorDescType(); |
| 64 | } |
| 65 | |
| 66 | if (auto layout = tdescTy.getLayoutAttr()) { |
| 67 | auto inst_data = layout.getInstData(); |
| 68 | if (inst_data && layout.isSgLayout()) |
| 69 | return SmallVector<int64_t>(inst_data.asArrayRef().begin(), |
| 70 | inst_data.asArrayRef().end()); |
| 71 | } |
| 72 | } |
| 73 | |
| 74 | if (isa<xegpu::DpasOp>(op)) |
| 75 | return SmallVector<int64_t>{8, 16, 16}; |
| 76 | |
| 77 | return std::nullopt; |
| 78 | }); |
| 79 | |
| 80 | options.setUnrolledTypesFn( |
| 81 | [&](ShapedType type, ArrayRef<int64_t> tileShape) -> SmallVector<Type> { |
| 82 | Type elemTy = type.getElementType(); |
| 83 | Type newTy; |
| 84 | |
| 85 | // TensorDescType needs to drop the inst_data field in the layout |
| 86 | // attribute |
| 87 | if (auto tdescTy = dyn_cast<xegpu::TensorDescType>(type)) { |
| 88 | Attribute encoding = tdescTy.getEncoding(); |
| 89 | auto layout = llvm::dyn_cast_if_present<xegpu::LayoutAttr>( |
| 90 | tdescTy.getLayout()); |
| 91 | if (layout) { |
| 92 | if (layout.getLaneLayout() == nullptr) |
| 93 | layout = xegpu::LayoutAttr(); |
| 94 | else |
| 95 | layout = layout.dropInstData(); |
| 96 | } |
| 97 | newTy = xegpu::TensorDescType::get(ctx, tileShape, elemTy, encoding, |
| 98 | layout); |
| 99 | } else { |
| 100 | newTy = type.clone(tileShape, elemTy); |
| 101 | } |
| 102 | |
| 103 | std::optional<SmallVector<int64_t>> ratio = |
| 104 | computeShapeRatio(type.getShape(), tileShape); |
| 105 | assert(ratio && "Expecting the ratio to be valid." ); |
| 106 | return SmallVector<Type>(computeProduct(basis: *ratio), newTy); |
| 107 | }); |
| 108 | |
| 109 | RewritePatternSet patterns(ctx); |
| 110 | |
| 111 | populateXeGPUUnrollPatterns(patterns, options); |
| 112 | (void)applyPatternsGreedily(getOperation(), std::move(patterns)); |
| 113 | } |
| 114 | }; |
| 115 | |
| 116 | } // namespace |
| 117 | |
| 118 | namespace mlir { |
| 119 | namespace test { |
| 120 | void registerTestXeGPULowerings() { |
| 121 | PassRegistration<TestXeGPUUnrollingPatterns>(); |
| 122 | } |
| 123 | } // namespace test |
| 124 | } // namespace mlir |
| 125 | |