|
6 | 6 | // |
7 | 7 | //===----------------------------------------------------------------------===// |
8 | 8 |
|
9 | | - |
10 | 9 | #include "./Tiling.hpp" |
11 | 10 | #include "gc/Dialect/Arith/Utils/EasyBuild.h" |
12 | 11 | #include "gc/Dialect/Linalgx/LinalgxOps.h" |
@@ -45,11 +44,44 @@ namespace gc { |
45 | 44 | #include "gc/Transforms/Passes.h.inc" |
46 | 45 |
|
47 | 46 | namespace { |
| 47 | + |
| 48 | +struct MHAToFlashAttention |
| 49 | + : public OpInterfaceRewritePattern<linalg::LinalgOp> { |
| 50 | + using OpInterfaceRewritePattern<linalg::LinalgOp>::OpInterfaceRewritePattern; |
| 51 | + |
| 52 | + LogicalResult matchAndRewrite(linalg::LinalgOp linalgOp, |
| 53 | + PatternRewriter &rewriter) const override { |
| 54 | + if (!llvm::isa<linalgx::ScaledDotProductAttentionOp>(linalgOp)) |
| 55 | + return failure(); |
| 56 | + if (linalgOp.hasPureBufferSemantics()) |
| 57 | + return failure(); |
| 58 | + } |
| 59 | +}; |
| 60 | + |
48 | 61 | struct FlashAttentionConversion |
49 | 62 | : public impl::FlashAttentionConversionBase<FlashAttentionConversion> { |
50 | 63 | public: |
51 | 64 | void runOnOperation() final { |
52 | | - return; |
| 65 | + auto &ctx = getContext(); |
| 66 | + IRRewriter rewriter(&ctx); |
| 67 | + RewritePatternSet patterns(&ctx); |
| 68 | + |
| 69 | + patterns.add<MHAToFlashAttention>(patterns.getContext()); |
| 70 | + // linalg::populateLinalgTilingCanonicalizationPatterns(patterns); |
| 71 | + // linalg::ControlDropUnitDims options; |
| 72 | + // options.rankReductionStrategy = |
| 73 | + // linalg::ControlDropUnitDims::RankReductionStrategy::ExtractInsertSlice; |
| 74 | + // linalg::populateFoldUnitExtentDimsPatterns(patterns, options); |
| 75 | + // tensor::populateMergeConsecutiveInsertExtractSlicePatterns(patterns); |
| 76 | + |
| 77 | + // for (auto *dialect : ctx.getLoadedDialects()) |
| 78 | + // dialect->getCanonicalizationPatterns(patterns); |
| 79 | + // for (RegisteredOperationName op : ctx.getRegisteredOperations()) |
| 80 | + // op.getCanonicalizationPatterns(patterns, &ctx); |
| 81 | + if (failed(applyPatternsAndFoldGreedily(getOperation(), |
| 82 | + std::move(patterns)))) { |
| 83 | + return signalPassFailure(); |
| 84 | + } |
53 | 85 | } |
54 | 86 | }; |
55 | 87 |
|
|
0 commit comments