xref: /llvm-project/mlir/test/Dialect/Vector/vector-scan-transforms.mlir (revision 9816edc9f3ce198d41e364dd3467caa839a0c220) 
1// RUN: mlir-opt %s --test-vector-scan-lowering | FileCheck %s
2
3// CHECK-LABEL: func @scan1d_inc
4// CHECK-SAME: %[[ARG0:.*]]: vector<2xi32>,
5// CHECK-SAME: %[[ARG1:.*]]: vector<i32>
6// CHECK:      %[[A:.*]] = arith.constant dense<0> : vector<2xi32>
7// CHECK:      %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0], sizes = [1], strides = [1]} : vector<2xi32> to vector<1xi32>
8// CHECK:      %[[C:.*]] = vector.insert_strided_slice %[[B]], %[[A]] {offsets = [0], strides = [1]} : vector<1xi32> into vector<2xi32>
9// CHECK:      %[[D:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [1], sizes = [1], strides = [1]} : vector<2xi32> to vector<1xi32>
10// CHECK:      %[[E:.*]] = arith.addi %[[B]], %[[D]] : vector<1xi32>
11// CHECK:      %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[C]] {offsets = [1], strides = [1]} : vector<1xi32> into vector<2xi32>
12// CHECK:      %[[G:.*]] = vector.extract %[[E]][0] : i32 from vector<1xi32>
13// CHECK:      %[[H:.*]] = vector.broadcast %[[G]] : i32 to vector<i32>
14// CHECK:      return %[[F]], %[[H]] : vector<2xi32>, vector<i32>
15func.func @scan1d_inc(%arg0 : vector<2xi32>, %arg1 : vector<i32>) -> (vector<2xi32>, vector<i32>) {
16  %0:2 = vector.scan <add>, %arg0, %arg1 {inclusive = true, reduction_dim = 0} :
17    vector<2xi32>, vector<i32>
18  return %0#0, %0#1 : vector<2xi32>, vector<i32>
19}
20
21// CHECK-LABEL: func @scan1d_exc
22// CHECK-SAME: %[[ARG0:.*]]: vector<2xi32>,
23// CHECK-SAME: %[[ARG1:.*]]: vector<i32>
24// CHECK:      %[[A:.*]] = arith.constant dense<0> : vector<2xi32>
25// CHECK:      %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0], sizes = [1], strides = [1]} : vector<2xi32> to vector<1xi32>
26// CHECK:      %[[C:.*]] = vector.broadcast %[[ARG1]] : vector<i32> to vector<1xi32>
27// CHECK:      %[[D:.*]] = vector.insert_strided_slice %[[C]], %[[A]] {offsets = [0], strides = [1]} : vector<1xi32> into vector<2xi32>
28// CHECK:      %[[E:.*]] = arith.addi %[[C]], %[[B]] : vector<1xi32>
29// CHECK:      %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[D]] {offsets = [1], strides = [1]} : vector<1xi32> into vector<2xi32>
30// CHECK:      %[[G:.*]] = vector.extract %[[E]][0] : i32 from vector<1xi32>
31// CHECK:      %[[H:.*]] = vector.broadcast %[[G]] : i32 to vector<i32>
32// CHECK:      return %[[F]], %[[H]] : vector<2xi32>, vector<i32>
33func.func @scan1d_exc(%arg0 : vector<2xi32>, %arg1 : vector<i32>) -> (vector<2xi32>, vector<i32>) {
34  %0:2 = vector.scan <add>, %arg0, %arg1 {inclusive = false, reduction_dim = 0} :
35    vector<2xi32>, vector<i32>
36  return %0#0, %0#1 : vector<2xi32>, vector<i32>
37}
38
39// CHECK-LABEL: func @scan2d_mul_dim0
40// CHECK-SAME: %[[ARG0:.*]]: vector<2x3xi32>,
41// CHECK-SAME: %[[ARG1:.*]]: vector<3xi32>
42// CHECK:      %[[A:.*]] = arith.constant dense<0> : vector<2x3xi32>
43// CHECK:      %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 0], sizes = [1, 3], strides = [1, 1]} : vector<2x3xi32> to vector<1x3xi32>
44// CHECK:      %[[C:.*]] = vector.insert_strided_slice %[[B]], %[[A]] {offsets = [0, 0], strides = [1, 1]} : vector<1x3xi32> into vector<2x3xi32>
45// CHECK:      %[[D:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [1, 0], sizes = [1, 3], strides = [1, 1]} : vector<2x3xi32> to vector<1x3xi32>
46// CHECK:      %[[E:.*]] = arith.muli %[[B]], %[[D]] : vector<1x3xi32>
47// CHECK:      %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[C]] {offsets = [1, 0], strides = [1, 1]} : vector<1x3xi32> into vector<2x3xi32>
48// CHECK:      %[[G:.*]] = vector.shape_cast %[[E]] : vector<1x3xi32> to vector<3xi32>
49// CHECK:      return %[[F]], %[[G]] : vector<2x3xi32>, vector<3xi32>
50func.func @scan2d_mul_dim0(%arg0 : vector<2x3xi32>, %arg1 : vector<3xi32>) -> (vector<2x3xi32>, vector<3xi32>) {
51  %0:2 = vector.scan <mul>, %arg0, %arg1 {inclusive = true, reduction_dim = 0} :
52    vector<2x3xi32>, vector<3xi32>
53  return %0#0, %0#1 : vector<2x3xi32>, vector<3xi32>
54}
55
56// CHECK-LABEL: func @scan2d_mul_dim1
57// CHECK-SAME: %[[ARG0:.*]]: vector<2x3xi32>,
58// CHECK-SAME: %[[ARG1:.*]]: vector<2xi32>
59// CHECK:      %[[A:.*]] = arith.constant dense<0> : vector<2x3xi32>
60// CHECK:      %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 0], sizes = [2, 1], strides = [1, 1]} : vector<2x3xi32> to vector<2x1xi32>
61// CHECK:      %[[C:.*]] = vector.insert_strided_slice %[[B]], %[[A]] {offsets = [0, 0], strides = [1, 1]} : vector<2x1xi32> into vector<2x3xi32>
62// CHECK:      %[[D:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 1], sizes = [2, 1], strides = [1, 1]} : vector<2x3xi32> to vector<2x1xi32>
63// CHECK:      %[[E:.*]] = arith.muli %[[B]], %[[D]] : vector<2x1xi32>
64// CHECK:      %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[C]] {offsets = [0, 1], strides = [1, 1]} : vector<2x1xi32> into vector<2x3xi32>
65// CHECK:      %[[G:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 2], sizes = [2, 1], strides = [1, 1]} : vector<2x3xi32> to vector<2x1xi32>
66// CHECK:      %[[H:.*]] = arith.muli %[[E]], %[[G]] : vector<2x1xi32>
67// CHECK:      %[[I:.*]] = vector.insert_strided_slice %[[H]], %[[F]] {offsets = [0, 2], strides = [1, 1]} : vector<2x1xi32> into vector<2x3xi32>
68// CHECK:      %[[J:.*]] = vector.shape_cast %[[H]] : vector<2x1xi32> to vector<2xi32>
69// CHECK:      return %[[I]], %[[J]] : vector<2x3xi32>, vector<2xi32>
70func.func @scan2d_mul_dim1(%arg0 : vector<2x3xi32>, %arg1 : vector<2xi32>) -> (vector<2x3xi32>, vector<2xi32>) {
71  %0:2 = vector.scan <mul>, %arg0, %arg1 {inclusive = true, reduction_dim = 1} :
72    vector<2x3xi32>, vector<2xi32>
73  return %0#0, %0#1 : vector<2x3xi32>, vector<2xi32>
74}
75
76// CHECK-LABEL: func @scan3d_mul_dim1
77// CHECK-SAME: %[[ARG0:.*]]: vector<4x2x3xf32>,
78// CHECK-SAME: %[[ARG1:.*]]: vector<4x3xf32>
79// CHECK:      %[[A:.*]] = arith.constant dense<0.000000e+00> : vector<4x2x3xf32>
80// CHECK:      %[[B:.*]] = vector.extract_strided_slice %[[ARG0]] {offsets = [0, 0, 0], sizes = [4, 1, 3], strides = [1, 1, 1]} : vector<4x2x3xf32> to vector<4x1x3xf32>
81// CHECK:      %[[C:.*]] = vector.shape_cast %[[ARG1]] : vector<4x3xf32> to vector<4x1x3xf32>
82// CHECK:      %[[D:.*]] = vector.insert_strided_slice %[[C]], %[[A]] {offsets = [0, 0, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
83// CHECK:      %[[E:.*]] = arith.mulf %[[C]], %[[B]] : vector<4x1x3xf32>
84// CHECK:      %[[F:.*]] = vector.insert_strided_slice %[[E]], %[[D]] {offsets = [0, 1, 0], strides = [1, 1, 1]} : vector<4x1x3xf32> into vector<4x2x3xf32>
85// CHECK:      %[[G:.*]] = vector.shape_cast %[[E]] : vector<4x1x3xf32> to vector<4x3xf32>
86// CHECK:      return %[[F]], %[[G]] : vector<4x2x3xf32>, vector<4x3xf32>
87func.func @scan3d_mul_dim1(%arg0 : vector<4x2x3xf32>, %arg1 : vector<4x3xf32>) -> (vector<4x2x3xf32>, vector<4x3xf32>) {
88  %0:2 = vector.scan <mul>, %arg0, %arg1 {inclusive = false, reduction_dim = 1} :
89    vector<4x2x3xf32>, vector<4x3xf32>
90  return %0#0, %0#1 : vector<4x2x3xf32>, vector<4x3xf32>
91}
92