Debian Codegraph @

mlir/test/Dialect/Linalg/elementwise/roundtrip.mlir

raw
1 // RUN: mlir-opt %s -split-input-file | FileCheck %s
2 //
3 // Note - the functions are named @{unary|binary}_{identity|transpose|broadcast|transpose_a|...}_{exp|mul|div|..}
4
5 // CHECK: @unary_identity_exp(%[[A:.+]]: tensor<8x16x32xf32>, %[[B:.+]]: tensor<8x16x32xf32>)
6 // CHECK: %{{.*}} = linalg.elementwise kind=#linalg.elementwise_kind<exp>
7 // CHECK-SAME ins(%[[A:.+]] : tensor<8x16x32xf32>) outs(%[[B:.+]] : tensor<8x16x32xf32>)
8 //
9 func.func @unary_identity_exp(%A : tensor<8x16x32xf32>, %B: tensor<8x16x32xf32>) -> tensor<8x16x32xf32> {
10 %r = linalg.elementwise
11 kind=#linalg.elementwise_kind<exp>
12 ins(%A : tensor<8x16x32xf32>)
13 outs(%B: tensor<8x16x32xf32>) -> tensor<8x16x32xf32>
14 return %r : tensor<8x16x32xf32>
15 }
16
17 // -----
18
19 // CHECK-DAG: #[[IDENTITY:.+]] = affine_map<(d0, d1, d2) -> (d0, d1, d2)>
20 // CHECK-DAG: #[[PROJECTION:.+]] = affine_map<(d0, d1, d2) -> (d2, d1)>
21 //
22 // CHECK: @unary_projection_tanh(%[[A:.+]]: tensor<?x16xf32>,
23 // CHECK-SAME: %[[B:.+]]: tensor<8x16x?xf32>) -> tensor<8x16x?xf32> {
24 // CHECK: {{.*}} = linalg.elementwise kind=#linalg.elementwise_kind<tanh>
25 // CHECK-SAME: indexing_maps = [#[[PROJECTION]], #[[IDENTITY]]]
26 // CHECK-SAME: ins(%[[A]] : tensor<?x16xf32>) outs(%[[B]] : tensor<8x16x?xf32>) -> tensor<8x16x?xf32>
27 //
28 func.func @unary_projection_tanh(%A: tensor<?x16xf32>,
29 %B: tensor<8x16x?xf32>) -> tensor<8x16x?xf32> {
30 %r = linalg.elementwise
31 kind=#linalg.elementwise_kind<tanh>
32 indexing_maps = [affine_map<(d0, d1, d2) -> (d2, d1)>,
33 affine_map<(d0, d1, d2) -> (d0, d1, d2)>]
34 ins(%A : tensor<?x16xf32>)
35 outs(%B: tensor<8x16x?xf32>) -> tensor<8x16x?xf32>
36 return %r : tensor<8x16x?xf32>
37 }
38
39 // -----
40
41 // CHECK: @binary_identity_div(%[[A:.+]]: tensor<16x8xf32>, %[[B:.+]]: tensor<16x8xf32>,
42 // CHECK-SAME: %[[C:.+]]: tensor<16x8xf32>) -> tensor<16x8xf32> {
43 // CHECK: {{.*}} = linalg.elementwise
44 // CHECK-SAME: kind=#linalg.elementwise_kind<div>
45 // CHECK-SAME: ins(%[[A]], %[[B]] : tensor<16x8xf32>, tensor<16x8xf32>)
46 // CHECK-SAME: outs(%[[C]] : tensor<16x8xf32>) -> tensor<16x8xf32>
47 //
48 func.func @binary_identity_div(%A: tensor<16x8xf32>, %B: tensor<16x8xf32>,
49 %C: tensor<16x8xf32>) -> tensor<16x8xf32> {
50 %r = linalg.elementwise
51 kind=#linalg.elementwise_kind<div>
52 ins(%A, %B: tensor<16x8xf32>, tensor<16x8xf32>)
53 outs(%C: tensor<16x8xf32>) -> tensor<16x8xf32>
54 return %r : tensor<16x8xf32>
55 }
56
57 // -----
58
59 // CHECK: @binary_identity_mul_5Di(%[[A]]: tensor<1x2x3x4x5xi32>,
60 // CHECK-SAME: %[[B:.+]]: tensor<1x2x3x4x5xi32>,
61 // CHECK-SAME: %[[C:.+]]: tensor<1x2x3x4x5xi32>) -> tensor<1x2x3x4x5xi32> {
62 // CHECK: {{.*}} = linalg.elementwise
63 // CHECK-SAME: kind=#linalg.elementwise_kind<mul>
64 // CHECK-SAME: ins(%[[A]], %[[B]] : tensor<1x2x3x4x5xi32>, tensor<1x2x3x4x5xi32>)
65 // CHECK-SAME: outs(%[[C]] : tensor<1x2x3x4x5xi32>) -> tensor<1x2x3x4x5xi32>
66 //
67 func.func @binary_identity_mul_5Di(%A: tensor<1x2x3x4x5xi32>, %B: tensor<1x2x3x4x5xi32>,
68 %C: tensor<1x2x3x4x5xi32>) -> tensor<1x2x3x4x5xi32> {
69 %r = linalg.elementwise
70 kind=#linalg.elementwise_kind<mul>
71 ins(%A, %B: tensor<1x2x3x4x5xi32>, tensor<1x2x3x4x5xi32>)
72 outs(%C: tensor<1x2x3x4x5xi32>) -> tensor<1x2x3x4x5xi32>
73 return %r : tensor<1x2x3x4x5xi32>
74 }
75
76 // -----
77
78 // CHECK: @redundant_maps
79 // CHECK-NOT: indexing_maps
80 //
81 #map = affine_map<(d0, d1, d2, d3, d4) -> (d0, d1, d2, d3, d4)>
82 func.func @redundant_maps(%A: tensor<1x2x3x4x5xi32>, %B: tensor<1x2x3x4x5xi32>,
83 %C: tensor<1x2x3x4x5xi32>) -> tensor<1x2x3x4x5xi32> {
84 %r = linalg.elementwise
85 kind=#linalg.elementwise_kind<mul>
86 indexing_maps = [#map, #map, #map]
87 ins(%A, %B: tensor<1x2x3x4x5xi32>, tensor<1x2x3x4x5xi32>)
88 outs(%C: tensor<1x2x3x4x5xi32>) -> tensor<1x2x3x4x5xi32>
89 return %r : tensor<1x2x3x4x5xi32>
90 }