[mlir][vector] Add n-d deinterleave lowering #94237
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
mub-at-arm
requested review from
hanhanW,
nicolasvasilache,
banach-space and
dcaballe
as code owners
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-vector Author: Mubashar Ahmad (mub-at-arm) ChangesThis patch implements the lowering for vector From: To: Full diff: https://github.com/llvm/llvm-project/pull/94237.diff 2 Files Affected:
diff --git a/mlir/lib/Dialect/Vector/Transforms/LowerVectorInterleave.cpp b/mlir/lib/Dialect/Vector/Transforms/LowerVectorInterleave.cpp
index 77c97b2f1497c..557837426d855 100644
--- a/mlir/lib/Dialect/Vector/Transforms/LowerVectorInterleave.cpp
+++ b/mlir/lib/Dialect/Vector/Transforms/LowerVectorInterleave.cpp
@@ -79,6 +79,42 @@ class UnrollInterleaveOp final : public OpRewritePattern<vector::InterleaveOp> {
int64_t targetRank = 1;
};
+class UnrollDeinterleaveOp final : public OpRewritePattern<vector::DeinterleaveOp> {
+public:
+ UnrollDeinterleaveOp(int64_t targetRank, MLIRContext *context,
+ PatternBenefit benefit = 1)
+ : OpRewritePattern(context, benefit), targetRank(targetRank) {};
+
+ LogicalResult matchAndRewrite(vector::DeinterleaveOp op,
+ PatternRewriter &rewriter) const override {
+ VectorType resultType = op.getResultVectorType();
+ auto unrollIterator = vector::createUnrollIterator(resultType, targetRank);
+ if (!unrollIterator)
+ return failure();
+
+ auto loc = op.getLoc();
+ Value evenResult = rewriter.create<arith::ConstantOp>(
+ loc, resultType, rewriter.getZeroAttr(resultType));
+ Value oddResult = rewriter.create<arith::ConstantOp>(
+ loc, resultType, rewriter.getZeroAttr(resultType));
+
+ for (auto position : *unrollIterator) {
+ auto extractSrc = rewriter.create<vector::ExtractOp>(
+ loc, op.getSource(), position);
+ auto deinterleave = rewriter.create<vector::DeinterleaveOp>(
+ loc, extractSrc);
+ evenResult = rewriter.create<vector::InsertOp>(
+ loc, deinterleave.getRes1(), evenResult, position);
+ oddResult = rewriter.create<vector::InsertOp>(
+ loc, deinterleave.getRes2(), oddResult, position);
+ }
+ rewriter.replaceOp(op, ValueRange{evenResult, oddResult});
+ return success();
+ }
+
+private:
+ int64_t targetRank = 1;
+};
/// Rewrite vector.interleave op into an equivalent vector.shuffle op, when
/// applicable: `sourceType` must be 1D and non-scalable.
///
@@ -117,6 +153,7 @@ struct InterleaveToShuffle final : OpRewritePattern<vector::InterleaveOp> {
void mlir::vector::populateVectorInterleaveLoweringPatterns(
RewritePatternSet &patterns, int64_t targetRank, PatternBenefit benefit) {
patterns.add<UnrollInterleaveOp>(targetRank, patterns.getContext(), benefit);
+ patterns.add<UnrollDeinterleaveOp>(targetRank, patterns.getContext(), benefit);
}
void mlir::vector::populateVectorInterleaveToShufflePatterns(
diff --git a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
index 245edb6789d30..21f4872bb2cd9 100644
--- a/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
+++ b/mlir/test/Conversion/VectorToLLVM/vector-to-llvm.mlir
@@ -2564,3 +2564,39 @@ func.func @vector_deinterleave_1d_scalable(%a: vector<[4]xi32>) -> (vector<[2]xi
%0, %1 = vector.deinterleave %a : vector<[4]xi32> -> vector<[2]xi32>
return %0, %1 : vector<[2]xi32>, vector<[2]xi32>
}
+
+// CHECK-LABEL: @vector_deinterleave_2d
+// CHECK-SAME: %[[SRC:.*]]: vector<2x8xf32>) -> (vector<2x4xf32>, vector<2x4xf32>)
+func.func @vector_deinterleave_2d(%a: vector<2x8xf32>) -> (vector<2x4xf32>, vector<2x4xf32>) {
+ // CHECK: %[[EXTRACT_ONE:.*]] = llvm.extractvalue %{{.*}}[0] : !llvm.array<2 x vector<8xf32>>
+ // CHECK: %[[POISON_ONE:.*]] = llvm.mlir.poison : vector<8xf32>
+ // CHECK: %[[SHUFFLE_A:.*]] = llvm.shufflevector %[[EXTRACT_ONE]], %[[POISON_ONE]] [0, 2, 4, 6] : vector<8xf32>
+ // CHECK: %[[SHUFFLE_B:.*]] = llvm.shufflevector %[[EXTRACT_ONE]], %[[POISON_ONE]] [1, 3, 5, 7] : vector<8xf32>
+ // CHECK: %[[INSERT_A:.*]] = llvm.insertvalue %[[SHUFFLE_A]], %{{.*}}[0] : !llvm.array<2 x vector<4xf32>>
+ // CHECK: %[[INSERT_B:.*]] = llvm.insertvalue %[[SHUFFLE_B]], %{{.*}}[0] : !llvm.array<2 x vector<4xf32>>
+ // CHECK: %[[EXTRACT_TWO:.*]] = llvm.extractvalue %{{.*}}[1] : !llvm.array<2 x vector<8xf32>>
+ // CHECK: %[[POISON_TWO:.*]] = llvm.mlir.poison : vector<8xf32>
+ // CHECK: %[[SHUFFLE_C:.*]] = llvm.shufflevector %[[EXTRACT_TWO]], %[[POISON_TWO]] [0, 2, 4, 6] : vector<8xf32>
+ // CHECK: %[[SHUFFLE_D:.*]] = llvm.shufflevector %[[EXTRACT_TWO]], %[[POISON_TWO]] [1, 3, 5, 7] : vector<8xf32>
+ // CHECK: %[[INSERT_C:.*]] = llvm.insertvalue %[[SHUFFLE_C]], %[[INSERT_A]][1] : !llvm.array<2 x vector<4xf32>>
+ // CHECK: %[[INSERT_D:.*]] = llvm.insertvalue %[[SHUFFLE_D]], %[[INSERT_B]][1] : !llvm.array<2 x vector<4xf32>>
+ %0, %1 = vector.deinterleave %a : vector<2x8xf32> -> vector<2x4xf32>
+ return %0, %1 : vector<2x4xf32>, vector<2x4xf32>
+}
+
+func.func @vector_deinterleave_2d_scalable(%a: vector<2x[8]xf32>) -> (vector<2x[4]xf32>, vector<2x[4]xf32>) {
+ // CHECK: %[[EXTRACT_A:.*]] = llvm.extractvalue %{{.*}}[0] : !llvm.array<2 x vector<[8]xf32>>
+ // CHECK: %[[VECTOR_ONE:.*]] = "llvm.intr.vector.deinterleave2"(%[[EXTRACT_ONE]]) : (vector<[8]xf32>) -> !llvm.struct<(vector<[4]xf32>, vector<[4]xf32>)>
+ // CHECK: %[[EXTRACT_B:.*]] = llvm.extractvalue %[[VECTOR_ONE]][0] : !llvm.struct<(vector<[4]xf32>, vector<[4]xf32>)>
+ // CHECK: %[[EXTRACT_C:.*]] = llvm.extractvalue %[[VECTOR_ONE]][1] : !llvm.struct<(vector<[4]xf32>, vector<[4]xf32>)>
+ // CHECK: %[[INSERT_A:.*]] = llvm.insertvalue %[[EXTRACT_B]], %{{.*}}[0] : !llvm.array<2 x vector<[4]xf32>>
+ // CHECK: %[[INSERT_B:.*]] = llvm.insertvalue %[[EXTRACT_C]], %{{.*}}[0] : !llvm.array<2 x vector<[4]xf32>>
+ // CHECK: %[[EXTRACT_D:.*]] = llvm.extractvalue %{{.*}}[1] : !llvm.array<2 x vector<[8]xf32>>
+ // CHECK: %[[VECTOR_TWO:.*]] = "llvm.intr.vector.deinterleave2"(%[[EXTRACT_D]]) : (vector<[8]xf32>) -> !llvm.struct<(vector<[4]xf32>, vector<[4]xf32>)>
+ // CHECK: %[[EXTRACT_E:.*]] = llvm.extractvalue %[[VECTOR_TWO]][0] : !llvm.struct<(vector<[4]xf32>, vector<[4]xf32>)>
+ // CHECK: %[[EXTRACT_F:.*]] = llvm.extractvalue %[[VECTOR_TWO]][1] : !llvm.struct<(vector<[4]xf32>, vector<[4]xf32>)>
+ // CHECK: %[[INSERT_C:.*]] = llvm.insertvalue %[[EXTRACT_E]], %[[INSERT_A]][1] : !llvm.array<2 x vector<[4]xf32>>
+ // CHECK: %[[INSERT_D:.*]] = llvm.insertvalue %[[EXTRACT_F]], %[[INSERT_B]][1] : !llvm.array<2 x vector<[4]xf32>>
+ %0, %1 = vector.deinterleave %a : vector<2x[8]xf32> -> vector<2x[4]xf32>
+ return %0, %1 : vector<2x[4]xf32>, vector<2x[4]xf32>
+}
|
|
✅ With the latest revision this PR passed the C/C++ code formatter. |
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
2 times, most recently
from
c78166b to
8471eb6
Compare
MacDue
reviewed
Jun 4, 2024
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
from
8471eb6 to
b95894d
Compare
MacDue
reviewed
Jun 4, 2024
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
from
b95894d to
c602bf3
Compare
MacDue
reviewed
Jun 5, 2024
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
3 times, most recently
from
471d9a8 to
b0649b5
Compare
MacDue
reviewed
Jun 6, 2024
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
2 times, most recently
from
64bfb63 to
815a1d7
Compare
MacDue
reviewed
Jun 6, 2024
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
2 times, most recently
from
a422db0 to
166b6c9
Compare
MacDue
reviewed
Jun 6, 2024
mlir/test/Dialect/Vector/vector-deinterleave-lowering-transforms.mlir
Outdated
Show resolved
Hide resolved
mlir/test/Dialect/Vector/vector-deinterleave-lowering-transforms.mlir
Outdated
Show resolved
Hide resolved
mlir/test/Dialect/Vector/vector-deinterleave-lowering-transforms.mlir
Outdated
Show resolved
Hide resolved
This patch implements the lowering for vector deinterleave for vector of n-dimensions. Process involves unrolling the n-d vector to a series of one-dimensional vectors. The deinterleave operation is then used on these vectors. From: ``` %0, %1 = vector.deinterleave %a : vector<2x[4]xi8> -> vector<2x[2]xi8> ``` To: ``` %2 = llvm.extractvalue %0[0] : !llvm.array<2 x vector<8xf32>> %3 = llvm.mlir.poison : vector<8xf32> %4 = llvm.shufflevector %2, %3 [0, 2, 4, 6] : vector<8xf32> %5 = llvm.shufflevector %2, %3 [1, 3, 5, 7] : vector<8xf32> %6 = llvm.insertvalue %4, %1[0] : !llvm.array<2 x vector<4xf32>> %7 = llvm.insertvalue %5, %1[0] : !llvm.array<2 x vector<4xf32>> %8 = llvm.extractvalue %0[1] : !llvm.array<2 x vector<8xf32>> ...etc. ```
mub-at-arm
force-pushed
the
nd_deinterleave_lowering
branch
from
166b6c9 to
8da2efc
Compare
Closed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This patch implements the lowering for vector
deinterleave for vector of n-dimensions. Process
involves unrolling the n-d vector to a series
of one-dimensional vectors. The deinterleave
operation is then used on these vectors.
From:
To: