xref: /llvm-project/mlir/test/Dialect/Vector/CPU/X86/vector-transpose-lowering.mlir (revision 0bacffbbfc081b4147ac935512e2c5da9e3c06f0) 
1// RUN: mlir-opt %s --transform-interpreter --split-input-file | FileCheck %s
2
3// NOTE: This file tests lowerings that are implemented in the X86Vector
4// dialect. Since X86 does not support scalable vectors, all examples in this
5// file use fixed-width vectors.
6
7// CHECK-LABEL: func @transpose4x8
8func.func @transpose4x8xf32(%arg0: vector<4x8xf32>) -> vector<8x4xf32> {
9  //      CHECK: vector.extract {{.*}}[0]
10  // CHECK-NEXT: vector.extract {{.*}}[1]
11  // CHECK-NEXT: vector.extract {{.*}}[2]
12  // CHECK-NEXT: vector.extract {{.*}}[3]
13  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
14  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
15  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
16  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
17  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
18  // CHECK-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
19  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
20  // CHECK-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
21  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
22  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
23  // CHECK-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
24  // CHECK-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
25  // CHECK-NEXT: vector.insert {{.*}}[0]
26  // CHECK-NEXT: vector.insert {{.*}}[1]
27  // CHECK-NEXT: vector.insert {{.*}}[2]
28  // CHECK-NEXT: vector.insert {{.*}}[3]
29  // CHECK-NEXT: vector.shape_cast {{.*}} vector<4x8xf32> to vector<32xf32>
30  // CHECK-NEXT: vector.shape_cast {{.*}} vector<32xf32> to vector<8x4xf32>
31  %0 = vector.transpose %arg0, [1, 0] : vector<4x8xf32> to vector<8x4xf32>
32  return %0 : vector<8x4xf32>
33}
34
35// CHECK-LABEL: func @transpose021_1x4x8
36func.func @transpose021_1x4x8xf32(%arg0: vector<1x4x8xf32>) -> vector<1x8x4xf32> {
37  //      CHECK: vector.extract {{.*}}[0, 0]
38  // CHECK-NEXT: vector.extract {{.*}}[0, 1]
39  // CHECK-NEXT: vector.extract {{.*}}[0, 2]
40  // CHECK-NEXT: vector.extract {{.*}}[0, 3]
41  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
42  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
43  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
44  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
45  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
46  // CHECK-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
47  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
48  // CHECK-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
49  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
50  // CHECK-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
51  // CHECK-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
52  // CHECK-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
53  // CHECK-NEXT: vector.insert {{.*}}[0]
54  // CHECK-NEXT: vector.insert {{.*}}[1]
55  // CHECK-NEXT: vector.insert {{.*}}[2]
56  // CHECK-NEXT: vector.insert {{.*}}[3]
57  // CHECK-NEXT: vector.shape_cast {{.*}} vector<4x8xf32> to vector<32xf32>
58  // CHECK-NEXT: vector.shape_cast {{.*}} vector<32xf32> to vector<1x8x4xf32>
59  %0 = vector.transpose %arg0, [0, 2, 1] : vector<1x4x8xf32> to vector<1x8x4xf32>
60  return %0 : vector<1x8x4xf32>
61}
62
63// CHECK-LABEL: func @transpose8x8
64func.func @transpose8x8xf32(%arg0: vector<8x8xf32>) -> vector<8x8xf32> {
65  //      CHECK: vector.extract {{.*}}[0]
66  // CHECK-NEXT: vector.extract {{.*}}[1]
67  // CHECK-NEXT: vector.extract {{.*}}[2]
68  // CHECK-NEXT: vector.extract {{.*}}[3]
69  // CHECK-NEXT: vector.extract {{.*}}[4]
70  // CHECK-NEXT: vector.extract {{.*}}[5]
71  // CHECK-NEXT: vector.extract {{.*}}[6]
72  // CHECK-NEXT: vector.extract {{.*}}[7]
73  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
74  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
75  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
76  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
77  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
78  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
79  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
80  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
81  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
82  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
83  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
84  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
85  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
86  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
87  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
88  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
89  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
90  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
91  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
92  // CHECK-NEXT: vector.insert {{.*}}[0]
93  // CHECK-NEXT: vector.insert {{.*}}[1]
94  // CHECK-NEXT: vector.insert {{.*}}[2]
95  // CHECK-NEXT: vector.insert {{.*}}[3]
96  // CHECK-NEXT: vector.insert {{.*}}[4]
97  // CHECK-NEXT: vector.insert {{.*}}[5]
98  // CHECK-NEXT: vector.insert {{.*}}[6]
99  // CHECK-NEXT: vector.insert {{.*}}[7]
100  %0 = vector.transpose %arg0, [1, 0] : vector<8x8xf32> to vector<8x8xf32>
101  return %0 : vector<8x8xf32>
102}
103
104// CHECK-LABEL: func @transpose021_1x8x8
105func.func @transpose021_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<1x8x8xf32> {
106  //      CHECK: vector.extract {{.*}}[0, 0]
107  // CHECK-NEXT: vector.extract {{.*}}[0, 1]
108  // CHECK-NEXT: vector.extract {{.*}}[0, 2]
109  // CHECK-NEXT: vector.extract {{.*}}[0, 3]
110  // CHECK-NEXT: vector.extract {{.*}}[0, 4]
111  // CHECK-NEXT: vector.extract {{.*}}[0, 5]
112  // CHECK-NEXT: vector.extract {{.*}}[0, 6]
113  // CHECK-NEXT: vector.extract {{.*}}[0, 7]
114  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
115  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
116  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
117  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
118  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
119  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
120  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
121  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
122  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
123  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
124  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
125  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
126  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
127  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
128  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
129  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
130  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
131  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
132  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
133  // CHECK-NEXT: vector.insert {{.*}}[0]
134  // CHECK-NEXT: vector.insert {{.*}}[1]
135  // CHECK-NEXT: vector.insert {{.*}}[2]
136  // CHECK-NEXT: vector.insert {{.*}}[3]
137  // CHECK-NEXT: vector.insert {{.*}}[4]
138  // CHECK-NEXT: vector.insert {{.*}}[5]
139  // CHECK-NEXT: vector.insert {{.*}}[6]
140  // CHECK-NEXT: vector.insert {{.*}}[7]
141  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<1x8x8xf32>
142  %0 = vector.transpose %arg0, [0, 2, 1] : vector<1x8x8xf32> to vector<1x8x8xf32>
143  return %0 : vector<1x8x8xf32>
144}
145
146// CHECK-LABEL: func @transpose120_8x1x8
147func.func @transpose120_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<1x8x8xf32> {
148  //      CHECK: vector.extract {{.*}}[0, 0]
149  // CHECK-NEXT: vector.extract {{.*}}[1, 0]
150  // CHECK-NEXT: vector.extract {{.*}}[2, 0]
151  // CHECK-NEXT: vector.extract {{.*}}[3, 0]
152  // CHECK-NEXT: vector.extract {{.*}}[4, 0]
153  // CHECK-NEXT: vector.extract {{.*}}[5, 0]
154  // CHECK-NEXT: vector.extract {{.*}}[6, 0]
155  // CHECK-NEXT: vector.extract {{.*}}[7, 0]
156  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
157  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
158  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
159  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
160  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
161  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
162  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
163  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
164  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
165  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
166  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
167  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
168  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
169  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
170  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
171  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
172  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
173  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
174  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
175  // CHECK-NEXT: vector.insert {{.*}}[0]
176  // CHECK-NEXT: vector.insert {{.*}}[1]
177  // CHECK-NEXT: vector.insert {{.*}}[2]
178  // CHECK-NEXT: vector.insert {{.*}}[3]
179  // CHECK-NEXT: vector.insert {{.*}}[4]
180  // CHECK-NEXT: vector.insert {{.*}}[5]
181  // CHECK-NEXT: vector.insert {{.*}}[6]
182  // CHECK-NEXT: vector.insert {{.*}}[7]
183  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<1x8x8xf32>
184  %0 = vector.transpose %arg0, [1, 2, 0] : vector<8x1x8xf32> to vector<1x8x8xf32>
185  return %0 : vector<1x8x8xf32>
186}
187
188// CHECK-LABEL: func @transpose120_8x8x1
189func.func @transpose120_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<8x1x8xf32> {
190  //      CHECK: vector.shape_cast %{{.*}} : vector<8x8x1xf32> to vector<8x8xf32>
191  // CHECK-NEXT: vector.extract {{.*}}[0]
192  // CHECK-NEXT: vector.extract {{.*}}[1]
193  // CHECK-NEXT: vector.extract {{.*}}[2]
194  // CHECK-NEXT: vector.extract {{.*}}[3]
195  // CHECK-NEXT: vector.extract {{.*}}[4]
196  // CHECK-NEXT: vector.extract {{.*}}[5]
197  // CHECK-NEXT: vector.extract {{.*}}[6]
198  // CHECK-NEXT: vector.extract {{.*}}[7]
199  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
200  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
201  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
202  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
203  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
204  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
205  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
206  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
207  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
208  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
209  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
210  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
211  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
212  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
213  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
214  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
215  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
216  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
217  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
218  // CHECK-NEXT: vector.insert {{.*}}[0]
219  // CHECK-NEXT: vector.insert {{.*}}[1]
220  // CHECK-NEXT: vector.insert {{.*}}[2]
221  // CHECK-NEXT: vector.insert {{.*}}[3]
222  // CHECK-NEXT: vector.insert {{.*}}[4]
223  // CHECK-NEXT: vector.insert {{.*}}[5]
224  // CHECK-NEXT: vector.insert {{.*}}[6]
225  // CHECK-NEXT: vector.insert {{.*}}[7]
226  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x1x8xf32>
227  %0 = vector.transpose %arg0, [1, 2, 0] : vector<8x8x1xf32> to vector<8x1x8xf32>
228  return %0 : vector<8x1x8xf32>
229}
230
231// CHECK-LABEL: func @transpose102_8x8x1
232func.func @transpose102_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<8x8x1xf32> {
233  //      CHECK: vector.shape_cast %{{.*}} : vector<8x8x1xf32> to vector<8x8xf32>
234  // CHECK-NEXT: vector.extract {{.*}}[0]
235  // CHECK-NEXT: vector.extract {{.*}}[1]
236  // CHECK-NEXT: vector.extract {{.*}}[2]
237  // CHECK-NEXT: vector.extract {{.*}}[3]
238  // CHECK-NEXT: vector.extract {{.*}}[4]
239  // CHECK-NEXT: vector.extract {{.*}}[5]
240  // CHECK-NEXT: vector.extract {{.*}}[6]
241  // CHECK-NEXT: vector.extract {{.*}}[7]
242  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
243  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
244  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
245  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
246  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
247  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
248  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
249  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
250  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
251  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
252  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
253  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
254  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
255  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
256  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
257  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
258  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
259  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
260  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
261  // CHECK-NEXT: vector.insert {{.*}}[0]
262  // CHECK-NEXT: vector.insert {{.*}}[1]
263  // CHECK-NEXT: vector.insert {{.*}}[2]
264  // CHECK-NEXT: vector.insert {{.*}}[3]
265  // CHECK-NEXT: vector.insert {{.*}}[4]
266  // CHECK-NEXT: vector.insert {{.*}}[5]
267  // CHECK-NEXT: vector.insert {{.*}}[6]
268  // CHECK-NEXT: vector.insert {{.*}}[7]
269  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x8x1xf32>
270  %0 = vector.transpose %arg0, [1, 0, 2] : vector<8x8x1xf32> to vector<8x8x1xf32>
271  return %0 : vector<8x8x1xf32>
272}
273
274// CHECK-LABEL: func @transpose201_8x1x8
275func.func @transpose201_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<8x8x1xf32> {
276  //      CHECK: vector.extract {{.*}}[0, 0]
277  // CHECK-NEXT: vector.extract {{.*}}[1, 0]
278  // CHECK-NEXT: vector.extract {{.*}}[2, 0]
279  // CHECK-NEXT: vector.extract {{.*}}[3, 0]
280  // CHECK-NEXT: vector.extract {{.*}}[4, 0]
281  // CHECK-NEXT: vector.extract {{.*}}[5, 0]
282  // CHECK-NEXT: vector.extract {{.*}}[6, 0]
283  // CHECK-NEXT: vector.extract {{.*}}[7, 0]
284  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
285  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
286  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
287  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
288  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
289  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
290  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
291  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
292  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
293  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
294  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
295  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
296  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
297  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
298  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
299  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
300  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
301  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
302  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
303  // CHECK-NEXT: vector.insert {{.*}}[0]
304  // CHECK-NEXT: vector.insert {{.*}}[1]
305  // CHECK-NEXT: vector.insert {{.*}}[2]
306  // CHECK-NEXT: vector.insert {{.*}}[3]
307  // CHECK-NEXT: vector.insert {{.*}}[4]
308  // CHECK-NEXT: vector.insert {{.*}}[5]
309  // CHECK-NEXT: vector.insert {{.*}}[6]
310  // CHECK-NEXT: vector.insert {{.*}}[7]
311  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x8x1xf32>
312  %0 = vector.transpose %arg0, [2, 0, 1] : vector<8x1x8xf32> to vector<8x8x1xf32>
313  return %0 : vector<8x8x1xf32>
314}
315
316// CHECK-LABEL: func @transpose201_1x8x8
317func.func @transpose201_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<8x1x8xf32> {
318  //      CHECK: vector.extract {{.*}}[0, 0]
319  // CHECK-NEXT: vector.extract {{.*}}[0, 1]
320  // CHECK-NEXT: vector.extract {{.*}}[0, 2]
321  // CHECK-NEXT: vector.extract {{.*}}[0, 3]
322  // CHECK-NEXT: vector.extract {{.*}}[0, 4]
323  // CHECK-NEXT: vector.extract {{.*}}[0, 5]
324  // CHECK-NEXT: vector.extract {{.*}}[0, 6]
325  // CHECK-NEXT: vector.extract {{.*}}[0, 7]
326  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
327  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
328  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
329  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
330  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
331  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
332  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
333  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
334  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
335  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
336  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
337  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
338  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
339  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
340  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
341  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
342  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
343  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
344  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
345  // CHECK-NEXT: vector.insert {{.*}}[0]
346  // CHECK-NEXT: vector.insert {{.*}}[1]
347  // CHECK-NEXT: vector.insert {{.*}}[2]
348  // CHECK-NEXT: vector.insert {{.*}}[3]
349  // CHECK-NEXT: vector.insert {{.*}}[4]
350  // CHECK-NEXT: vector.insert {{.*}}[5]
351  // CHECK-NEXT: vector.insert {{.*}}[6]
352  // CHECK-NEXT: vector.insert {{.*}}[7]
353  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x1x8xf32>
354  %0 = vector.transpose %arg0, [2, 0, 1] : vector<1x8x8xf32> to vector<8x1x8xf32>
355  return %0 : vector<8x1x8xf32>
356}
357
358// CHECK-LABEL: func @transpose210_8x1x8
359func.func @transpose210_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<8x1x8xf32> {
360  //      CHECK: vector.extract {{.*}}[0, 0]
361  // CHECK-NEXT: vector.extract {{.*}}[1, 0]
362  // CHECK-NEXT: vector.extract {{.*}}[2, 0]
363  // CHECK-NEXT: vector.extract {{.*}}[3, 0]
364  // CHECK-NEXT: vector.extract {{.*}}[4, 0]
365  // CHECK-NEXT: vector.extract {{.*}}[5, 0]
366  // CHECK-NEXT: vector.extract {{.*}}[6, 0]
367  // CHECK-NEXT: vector.extract {{.*}}[7, 0]
368  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
369  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
370  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
371  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
372  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
373  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
374  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
375  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
376  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
377  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
378  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
379  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
380  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
381  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
382  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
383  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
384  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
385  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
386  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
387  // CHECK-NEXT: vector.insert {{.*}}[0]
388  // CHECK-NEXT: vector.insert {{.*}}[1]
389  // CHECK-NEXT: vector.insert {{.*}}[2]
390  // CHECK-NEXT: vector.insert {{.*}}[3]
391  // CHECK-NEXT: vector.insert {{.*}}[4]
392  // CHECK-NEXT: vector.insert {{.*}}[5]
393  // CHECK-NEXT: vector.insert {{.*}}[6]
394  // CHECK-NEXT: vector.insert {{.*}}[7]
395  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x1x8xf32>
396  %0 = vector.transpose %arg0, [2, 1, 0] : vector<8x1x8xf32> to vector<8x1x8xf32>
397  return %0 : vector<8x1x8xf32>
398}
399
400// CHECK-LABEL: func @transpose210_8x8x1
401func.func @transpose210_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<1x8x8xf32> {
402  //      CHECK: vector.shape_cast %{{.*}} : vector<8x8x1xf32> to vector<8x8xf32>
403  // CHECK-NEXT: vector.extract {{.*}}[0]
404  // CHECK-NEXT: vector.extract {{.*}}[1]
405  // CHECK-NEXT: vector.extract {{.*}}[2]
406  // CHECK-NEXT: vector.extract {{.*}}[3]
407  // CHECK-NEXT: vector.extract {{.*}}[4]
408  // CHECK-NEXT: vector.extract {{.*}}[5]
409  // CHECK-NEXT: vector.extract {{.*}}[6]
410  // CHECK-NEXT: vector.extract {{.*}}[7]
411  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
412  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
413  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
414  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
415  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
416  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
417  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
418  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
419  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
420  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
421  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
422  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
423  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
424  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
425  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
426  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
427  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
428  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
429  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
430  // CHECK-NEXT: vector.insert {{.*}}[0]
431  // CHECK-NEXT: vector.insert {{.*}}[1]
432  // CHECK-NEXT: vector.insert {{.*}}[2]
433  // CHECK-NEXT: vector.insert {{.*}}[3]
434  // CHECK-NEXT: vector.insert {{.*}}[4]
435  // CHECK-NEXT: vector.insert {{.*}}[5]
436  // CHECK-NEXT: vector.insert {{.*}}[6]
437  // CHECK-NEXT: vector.insert {{.*}}[7]
438  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<1x8x8xf32>
439  %0 = vector.transpose %arg0, [2, 1, 0] : vector<8x8x1xf32> to vector<1x8x8xf32>
440  return %0 : vector<1x8x8xf32>
441}
442
443// CHECK-LABEL: func @transpose210_1x8x8
444func.func @transpose210_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<8x8x1xf32> {
445  //      CHECK: vector.extract {{.*}}[0, 0]
446  // CHECK-NEXT: vector.extract {{.*}}[0, 1]
447  // CHECK-NEXT: vector.extract {{.*}}[0, 2]
448  // CHECK-NEXT: vector.extract {{.*}}[0, 3]
449  // CHECK-NEXT: vector.extract {{.*}}[0, 4]
450  // CHECK-NEXT: vector.extract {{.*}}[0, 5]
451  // CHECK-NEXT: vector.extract {{.*}}[0, 6]
452  // CHECK-NEXT: vector.extract {{.*}}[0, 7]
453  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
454  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
455  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
456  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
457  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
458  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
459  // CHECK-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
460  // CHECK-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
461  // CHECK-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
462  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
463  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
464  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
465  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
466  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
467  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
468  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
469  // CHECK-NEXT: llvm.inline_asm asm_dialect = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8xf32>
470  // CHECK-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
471  // CHECK-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
472  // CHECK-NEXT: vector.insert {{.*}}[0]
473  // CHECK-NEXT: vector.insert {{.*}}[1]
474  // CHECK-NEXT: vector.insert {{.*}}[2]
475  // CHECK-NEXT: vector.insert {{.*}}[3]
476  // CHECK-NEXT: vector.insert {{.*}}[4]
477  // CHECK-NEXT: vector.insert {{.*}}[5]
478  // CHECK-NEXT: vector.insert {{.*}}[6]
479  // CHECK-NEXT: vector.insert {{.*}}[7]
480  // CHECK-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x8x1xf32>
481  %0 = vector.transpose %arg0, [2, 1, 0] : vector<1x8x8xf32> to vector<8x8x1xf32>
482  return %0 : vector<8x8x1xf32>
483}
484
485module attributes {transform.with_named_sequence} {
486  transform.named_sequence @__transform_main(%root : !transform.any_op {transform.readonly}) {
487    %func_op = transform.structured.match ops{["func.func"]} in %root : (!transform.any_op) -> !transform.op<"func.func">
488    transform.apply_patterns to %func_op {
489      transform.apply_patterns.vector.lower_transpose avx2_lowering_strategy = true
490    } : !transform.op<"func.func">
491    transform.yield
492  }
493}
494