@@ -234,19 +234,25 @@ builtin.func @fuse_indexed(%arg0: tensor<24x12xi32>,
234
234
235
235
// CHECK-DAG: #[[MAP0:.*]] = affine_map<(d0, d1) -> (d0 + d1)>
236
236
// CHECK-DAG: #[[MAP1:.*]] = affine_map<(d0, d1) -> (8, -d0 - d1 + 17)>
237
- // CHECK-DAG: #[[MAP2:.*]] = affine_map<(d0, d1, d2) -> (d0, -d1 - d2 + 18 )>
237
+ // CHECK-DAG: #[[MAP2:.*]] = affine_map<(d0, d1, d2) -> (d0, -d1 - d2 + 17 )>
238
238
#map0 = affine_map <(d0 , d1 ) -> (d0 , d0 + d1 )>
239
239
#map1 = affine_map <(d0 , d1 ) -> (d0 , d1 )>
240
240
241
241
// CHECK: fuse_non_rectangular
242
- // CHECK-SAME: %[[ARG0:[0-9a-zA-Z]*]]: tensor<10x18xf32 >
243
- func @fuse_non_rectangular (%arg0: tensor <10 x 18 x f32 >,
242
+ // CHECK-SAME: %[[ARG0:[0-9a-zA-Z]*]]: tensor<10x17xf32 >
243
+ func @fuse_non_rectangular (%arg0: tensor <10 x 17 x f32 >,
244
244
%arg1: tensor <10 x8 xf32 >) -> tensor <10 x8 xf32 > {
245
+
246
+ // CHECK-DAG: %[[C0:.*]] = constant 0 : index
247
+ // CHECK-DAG: %[[C4:.*]] = constant 4 : index
248
+ // CHECK-DAG: %[[C5:.*]] = constant 5 : index
249
+ // CHECK-DAG: %[[C8:.*]] = constant 8 : index
250
+ // CHECK-DAG: %[[C10:.*]] = constant 10 : index
245
251
%cst = constant 0.000000e+00 : f32
246
- %0 = linalg.fill (%cst , %arg0 ) : f32 , tensor <10 x 18 x f32 > -> tensor <10 x 18 x f32 >
252
+ %0 = linalg.fill (%cst , %arg0 ) : f32 , tensor <10 x 17 x f32 > -> tensor <10 x 17 x f32 >
247
253
248
- // CHECK: scf.for %[[IV0:[0-9a-zA-Z]*]] = %c0 to %c8 step %c4
249
- // CHECK: scf.for %[[IV1:[0-9a-zA-Z]*]] = %c0 to %c10 step %c5
254
+ // CHECK: scf.for %[[IV0:[0-9a-zA-Z]*]] = %[[C0]] to %[[C8]] step %[[C4]]
255
+ // CHECK: scf.for %[[IV1:[0-9a-zA-Z]*]] = %[[C0]] to %[[C10]] step %[[C5]]
250
256
251
257
// Compute producer on a hyper rectangular bounding box. Along the second dimenson,
252
258
// the offset is set to the sum of the induction variables, and the upper bound
@@ -259,7 +265,7 @@ func @fuse_non_rectangular(%arg0: tensor<10x18xf32>,
259
265
// CHECK-SAME: %[[IV1]], %[[SUM]]
260
266
// CHECK-SAME: , %[[UB1]]
261
267
// CHECK: %[[T1:.*]] = linalg.fill(%{{.*}}, %[[T0]])
262
- %1 = linalg.generic {index ing_maps = [#map0 , #map1 ], iterator_types = [" parallel" , " parallel" ]} ins (%0 : tensor <10 x 18 x f32 >) outs (%arg1 : tensor <10 x8 xf32 >) {
268
+ %1 = linalg.generic {index ing_maps = [#map0 , #map1 ], iterator_types = [" parallel" , " parallel" ]} ins (%0 : tensor <10 x 17 x f32 >) outs (%arg1 : tensor <10 x8 xf32 >) {
263
269
^bb0 (%arg2: f32 , %arg3: f32 ): // no predecessors
264
270
%2 = addf %arg2 , %arg3 : f32
265
271
linalg.yield %2 : f32
0 commit comments