1// RUN: mlir-opt %s -tensor-copy-insertion -split-input-file | FileCheck %s
2// RUN: mlir-opt %s -tensor-copy-insertion="bufferize-function-boundaries allow-return-allocs" -split-input-file | FileCheck %s --check-prefix=CHECK-FUNC
3// RUN: mlir-opt %s -tensor-copy-insertion="create-deallocs=0" -split-input-file | FileCheck %s --check-prefix=CHECK-NO-DEALLOC
4
5// CHECK-LABEL: func @read_after_write_conflict(
6//  CHECK-SAME:     %[[t:.*]]: tensor<?xf32>
7// CHECK-FUNC-LABEL: func @read_after_write_conflict(
8// CHECK-NO-DEALLOC-LABEL: func @read_after_write_conflict(
9func.func @read_after_write_conflict(%t: tensor<?xf32>, %idx: index, %f: f32)
10  -> (tensor<?xf32>, tensor<?xf32>)
11{
12  // CHECK: %[[copy:.*]] = bufferization.alloc_tensor() copy(%[[t]]) {bufferization.escape = [false]} : tensor<?xf32>
13  // CHECK-FUNC: bufferization.alloc_tensor() copy(%{{.*}}) {bufferization.escape = [true]} : tensor<?xf32>
14  // CHECK-NO-DEALLOC: bufferization.alloc_tensor() copy(%{{.*}}) {bufferization.escape = [true]} : tensor<?xf32>
15  // CHECK: %[[insert:.*]] = tensor.insert %{{.*}} into %[[copy]]
16  %0 = tensor.insert %f into %t[%idx] : tensor<?xf32>
17  // CHECK: return %[[insert]], %[[t]]
18  return %0, %t : tensor<?xf32>, tensor<?xf32>
19}
20
21// -----
22
23// CHECK-LABEL: func @return_alloc_tensor
24// CHECK-FUNC-LABEL: func @return_alloc_tensor
25// CHECK-NO-DEALLOC-LABEL: func @return_alloc_tensor
26func.func @return_alloc_tensor() -> (tensor<5xf32>) {
27  // CHECK: bufferization.alloc_tensor() {bufferization.escape = [false]} : tensor<5xf32>
28  // CHECK-FUNC: bufferization.alloc_tensor() {bufferization.escape = [true]} : tensor<5xf32>
29  // CHECK-NO-DEALLOC: bufferization.alloc_tensor() {bufferization.escape = [true]} : tensor<5xf32>
30  %0 = bufferization.alloc_tensor() : tensor<5xf32>
31  return %0 : tensor<5xf32>
32}
33
34// -----
35
36// CHECK-LABEL: func @do_not_copy_undefined_tensor
37// CHECK-NO-DEALLOC-LABEL: func @do_not_copy_undefined_tensor
38func.func @do_not_copy_undefined_tensor(%f: f32, %idx: index)
39  -> (tensor<5xf32>, tensor<5xf32>)
40{
41  // CHECK: bufferization.alloc_tensor() {bufferization.escape = [false]} : tensor<5xf32>
42  // The second alloc_tensor should not have a copy operand.
43  // CHECK: bufferization.alloc_tensor() {bufferization.escape = [false], memory_space = 0 : ui64} : tensor<5xf32>
44
45  // CHECK-NO-DEALLOC: bufferization.alloc_tensor() {bufferization.escape = [true]} : tensor<5xf32>
46  // CHECK-NO-DEALLOC: bufferization.alloc_tensor() {bufferization.escape = [true], memory_space = 0 : ui64} : tensor<5xf32>
47  %0 = bufferization.alloc_tensor() : tensor<5xf32>
48  %1 = tensor.insert %f into %0[%idx] : tensor<5xf32>
49  return %0, %1 : tensor<5xf32>, tensor<5xf32>
50}
51
52// -----
53
54// CHECK-LABEL: func @do_not_copy_when_overwritten
55func.func @do_not_copy_when_overwritten(%t: tensor<5xf32>, %f: f32)
56  -> (tensor<5xf32>, tensor<5xf32>)
57{
58  // CHECK: %[[alloc:.*]] = bufferization.alloc_tensor() {bufferization.escape = [false], memory_space = 0 : ui64} : tensor<5xf32>
59  // CHECK: linalg.generic {{.*}} outs(%[[alloc]] : tensor<5xf32>)
60  %r = linalg.generic {
61    indexing_maps = [affine_map<(d0) -> (d0)>],
62    iterator_types = ["parallel"]}
63    outs(%t : tensor<5xf32>) {
64      ^bb0(%arg0 : f32) :
65        linalg.yield %f : f32
66    } -> tensor<5xf32>
67  return %t, %r : tensor<5xf32>, tensor<5xf32>
68}
69
70// -----
71
72// CHECK-LABEL: func @do_not_copy_when_result_not_read
73func.func @do_not_copy_when_result_not_read(%t: tensor<5xf32>, %f: f32)
74  -> (tensor<3xf32>)
75{
76  %0 = tensor.extract_slice %t[0][3][1] : tensor<5xf32> to tensor<3xf32>
77  // CHECK: %[[alloc:.*]] = bufferization.alloc_tensor() {bufferization.escape = [false], memory_space = 0 : ui64} : tensor<3xf32>
78  // CHECK: linalg.generic {{.*}} outs(%[[alloc]] : tensor<3xf32>)
79  %r = linalg.generic {
80    indexing_maps = [affine_map<(d0) -> (d0)>],
81    iterator_types = ["parallel"]}
82    outs(%0 : tensor<3xf32>) {
83      ^bb0(%arg0 : f32) :
84        linalg.yield %f : f32
85    } -> tensor<3xf32>
86  return %r : tensor<3xf32>
87}
88