Skip to content

Commit 548b2fe

Browse files
authored
Use Shuffle16x16 lowering from upstream (#745)
1 parent 8835e8d commit 548b2fe

File tree

2 files changed

+73
-3
lines changed

2 files changed

+73
-3
lines changed

lib/Transforms/VectorLinearize.cpp

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -235,7 +235,6 @@ struct VectorInsertOpConversion final
235235
return rewriter.notifyMatchFailure(insertOp,
236236
"dynamic position is not supported.");
237237
auto srcTy = insertOp.getSourceType();
238-
auto src = insertOp.getSource();
239238
auto srcAsVec = mlir::dyn_cast<mlir::VectorType>(srcTy);
240239
uint64_t srcSize = 0;
241240
if (srcAsVec) {
@@ -245,7 +244,6 @@ struct VectorInsertOpConversion final
245244
"scalars are not supported.");
246245
}
247246

248-
auto dst = insertOp.getDest();
249247
auto dstShape = insertOp.getDestVectorType().getShape();
250248
const auto dstSize = insertOp.getDestVectorType().getNumElements();
251249
auto dstSizeForOffsets = dstSize;
@@ -304,10 +302,11 @@ struct VectorLinearizePass final
304302
VectorExtractOpConversion, VectorInsertOpConversion>(
305303
typeConverter, context);
306304

305+
// Shuffle16x16 will fallback to Shuffle1D for non 16x16 sizes.
307306
mlir::vector::populateVectorTransposeLoweringPatterns(
308307
patterns,
309308
mlir::vector::VectorTransformsOptions().setVectorTransposeLowering(
310-
mlir::vector::VectorTransposeLowering::Shuffle1D));
309+
mlir::vector::VectorTransposeLowering::Shuffle16x16));
311310
unsigned targetVectBitWidth = std::numeric_limits<unsigned>::max();
312311
mlir::vector::populateVectorLinearizeTypeConversionsAndLegality(
313312
typeConverter, patterns, target, targetVectBitWidth);

test/Transforms/vector-linearize.mlir

Lines changed: 71 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -142,3 +142,74 @@ func.func @test_vector_transpose(%arg: vector<2x8xf32>) -> vector<8x2xf32> {
142142
%0 = vector.transpose %arg, [1, 0] : vector<2x8xf32> to vector<8x2xf32>
143143
return %0 : vector<8x2xf32>
144144
}
145+
146+
// -----
147+
// CHECK-LABEL: test_vector_transpose_16x16
148+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
149+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
150+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
151+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
152+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
153+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
154+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
155+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
156+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
157+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
158+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
159+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
160+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
161+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
162+
// CHECK: vector.shuffle {{.*}} [0, 16, 1, 17, 4, 20, 5, 21, 8, 24, 9, 25, 12, 28, 13, 29] : vector<16xf32>, vector<16xf32>
163+
// CHECK: vector.shuffle {{.*}} [2, 18, 3, 19, 6, 22, 7, 23, 10, 26, 11, 27, 14, 30, 15, 31] : vector<16xf32>, vector<16xf32>
164+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
165+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
166+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
167+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
168+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
169+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
170+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
171+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
172+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
173+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
174+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
175+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
176+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
177+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
178+
// CHECK: vector.shuffle {{.*}} [0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29] : vector<16xf32>, vector<16xf32>
179+
// CHECK: vector.shuffle {{.*}} [2, 3, 18, 19, 6, 7, 22, 23, 10, 11, 26, 27, 14, 15, 30, 31] : vector<16xf32>, vector<16xf32>
180+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
181+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
182+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
183+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
184+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
185+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
186+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
187+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
188+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
189+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
190+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
191+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
192+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
193+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
194+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
195+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
196+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
197+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
198+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
199+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
200+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
201+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
202+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
203+
// CHECK: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11, 16, 17, 18, 19, 24, 25, 26, 27] : vector<16xf32>, vector<16xf32>
204+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
205+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
206+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
207+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
208+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
209+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
210+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
211+
// CHECK: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31] : vector<16xf32>, vector<16xf32>
212+
func.func @test_vector_transpose_16x16(%arg: vector<16x16xf32>) -> vector<16x16xf32> {
213+
%0 = vector.transpose %arg, [1, 0] : vector<16x16xf32> to vector<16x16xf32>
214+
return %0 : vector<16x16xf32>
215+
}

0 commit comments

Comments
 (0)