1 // RUN: mlir-opt %s -affine-super-vectorizer-test="vector-shape-ratio=4,8" 2>&1 | FileCheck %s
2 // RUN: mlir-opt %s -affine-super-vectorizer-test="vector-shape-ratio=2,5,2" 2>&1 | FileCheck %s -check-prefix=TEST-3x4x5x8
3 // RUN: mlir-opt %s -affine-super-vectorizer-test=vectorize-affine-loop-nest 2>&1 | FileCheck %s -check-prefix=VECNEST
5 func.func @vector_add_2d(%arg0: index, %arg1: index) -> f32 {
6 // Nothing should be matched in this first block.
7 // CHECK-NOT:matched: {{.*}} = memref.alloc{{.*}}
8 // CHECK-NOT:matched: {{.*}} = arith.constant 0{{.*}}
9 // CHECK-NOT:matched: {{.*}} = arith.constant 1{{.*}}
10 %0 = memref.alloc(%arg0, %arg1) : memref<?x?xf32>
11 %1 = memref.alloc(%arg0, %arg1) : memref<?x?xf32>
12 %2 = memref.alloc(%arg0, %arg1) : memref<?x?xf32>
13 %c0 = arith.constant 0 : index
14 %cst = arith.constant 1.000000e+00 : f32
16 // CHECK:matched: {{.*}} arith.constant dense{{.*}} with shape ratio: 2, 32
17 %cst_1 = arith.constant dense<1.000000e+00> : vector<8x256xf32>
18 // CHECK:matched: {{.*}} arith.constant dense{{.*}} with shape ratio: 1, 3, 7, 2, 1
19 %cst_a = arith.constant dense<1.000000e+00> : vector<1x3x7x8x8xf32>
20 // CHECK-NOT:matched: {{.*}} arith.constant dense{{.*}} with shape ratio: 1, 3, 7, 1{{.*}}
21 %cst_b = arith.constant dense<1.000000e+00> : vector<1x3x7x4x4xf32>
22 // TEST-3x4x5x8:matched: {{.*}} arith.constant dense{{.*}} with shape ratio: 3, 2, 1, 4
23 %cst_c = arith.constant dense<1.000000e+00> : vector<3x4x5x8xf32>
24 // TEST-3x4x5x8-NOT:matched: {{.*}} arith.constant dense{{.*}} with shape ratio{{.*}}
25 %cst_d = arith.constant dense<1.000000e+00> : vector<3x4x4x8xf32>
26 // TEST-3x4x5x8:matched: {{.*}} arith.constant dense{{.*}} with shape ratio: 1, 1, 2, 16
27 %cst_e = arith.constant dense<1.000000e+00> : vector<1x2x10x32xf32>
29 // Nothing should be matched in this last block.
30 // CHECK-NOT:matched: {{.*}} = arith.constant 7{{.*}}
31 // CHECK-NOT:matched: {{.*}} = arith.constant 42{{.*}}
32 // CHECK-NOT:matched: {{.*}} = memref.load{{.*}}
33 // CHECK-NOT:matched: return {{.*}}
34 %c7 = arith.constant 7 : index
35 %c42 = arith.constant 42 : index
36 %9 = memref.load %2[%c7, %c42] : memref<?x?xf32>
40 // VECNEST-LABEL: func @double_loop_nest
41 func.func @double_loop_nest(%a: memref<20x30xf32>, %b: memref<20xf32>) {
43 affine.for %i = 0 to 20 {
44 %b_ld = affine.load %b[%i] : memref<20xf32>
45 affine.for %j = 0 to 30 {
46 %a_ld = affine.load %a[%i, %j] : memref<20x30xf32>
47 affine.store %a_ld, %a[%i, %j] : memref<20x30xf32>
49 affine.store %b_ld, %b[%i] : memref<20xf32>
55 // VECNEST: affine.for %{{.*}} = 0 to 20 step 4 {
56 // VECNEST: vector.transfer_read
57 // VECNEST-NEXT: affine.for %{{.*}} = 0 to 30 {
58 // VECNEST: vector.transfer_read
59 // VECNEST-NEXT: vector.transfer_write %{{.*}}, %{{.*}}[%{{.*}}, %{{.*}}] {permutation_map = #{{.*}}}
61 // VECNEST-NEXT: vector.transfer_write