Lines Matching refs:vector
1 // RUN: mlir-opt %s -test-vector-transpose-lowering=eltwise=1 -split-input-file | FileCheck %s --ch…
2 // RUN: mlir-opt %s -test-vector-transpose-lowering=shuffle=1 -split-input-file | FileCheck %s --ch…
3 // RUN: mlir-opt %s -test-vector-transpose-lowering=flat=1 -split-input-file | FileCheck %s --check…
4 // RUN: mlir-opt %s -test-vector-transpose-lowering=avx2=1 -split-input-file | FileCheck %s --check…
7 // ELTWISE-SAME: %[[A:.*]]: vector<2x3xf32>
8 // ELTWISE: %[[Z:.*]] = arith.constant dense<0.000000e+00> : vector<3x2xf32>
9 // ELTWISE: %[[T0:.*]] = vector.extract %[[A]][0, 0] : vector<2x3xf32>
10 // ELTWISE: %[[T1:.*]] = vector.insert %[[T0]], %[[Z]] [0, 0] : f32 into vector<3x2xf32>
11 // ELTWISE: %[[T2:.*]] = vector.extract %[[A]][0, 1] : vector<2x3xf32>
12 // ELTWISE: %[[T3:.*]] = vector.insert %[[T2]], %[[T1]] [1, 0] : f32 into vector<3x2xf32>
13 // ELTWISE: %[[T4:.*]] = vector.extract %[[A]][0, 2] : vector<2x3xf32>
14 // ELTWISE: %[[T5:.*]] = vector.insert %[[T4]], %[[T3]] [2, 0] : f32 into vector<3x2xf32>
15 // ELTWISE: %[[T6:.*]] = vector.extract %[[A]][1, 0] : vector<2x3xf32>
16 // ELTWISE: %[[T7:.*]] = vector.insert %[[T6]], %[[T5]] [0, 1] : f32 into vector<3x2xf32>
17 // ELTWISE: %[[T8:.*]] = vector.extract %[[A]][1, 1] : vector<2x3xf32>
18 // ELTWISE: %[[T9:.*]] = vector.insert %[[T8]], %[[T7]] [1, 1] : f32 into vector<3x2xf32>
19 // ELTWISE: %[[T10:.*]] = vector.extract %[[A]][1, 2] : vector<2x3xf32>
20 // ELTWISE: %[[T11:.*]] = vector.insert %[[T10]], %[[T9]] [2, 1] : f32 into vector<3x2xf32>
21 // ELTWISE: return %[[T11]] : vector<3x2xf32>
22 func.func @transpose23(%arg0: vector<2x3xf32>) -> vector<3x2xf32> {
23 %0 = vector.transpose %arg0, [1, 0] : vector<2x3xf32> to vector<3x2xf32>
24 return %0 : vector<3x2xf32>
31 func.func @transpose(%arg0: vector<2x4xf32>) -> vector<4x2xf32> {
32 // SHUFFLE: vector.shape_cast %{{.*}} : vector<2x4xf32> to vector<8xf32>
37 // SHUFFLE-NEXT: vector.shuffle %{{.*}} [0, 4, 1, 5, 2, 6, 3, 7] : vector<8xf32>, vector<8xf32>
38 // SHUFFLE-NEXT: vector.shape_cast %{{.*}} : vector<8xf32> to vector<4x2xf32>
40 // FLAT: vector.shape_cast {{.*}} : vector<2x4xf32> to vector<8xf32>
41 …// FLAT: vector.flat_transpose %{{.*}} {columns = 2 : i32, rows = 4 : i32} : vector<8xf32> -…
42 // FLAT: vector.shape_cast {{.*}} : vector<8xf32> to vector<4x2xf32>
43 %0 = vector.transpose %arg0, [1, 0] : vector<2x4xf32> to vector<4x2xf32>
44 return %0 : vector<4x2xf32>
50 func.func @transpose4x8xf32(%arg0: vector<4x8xf32>) -> vector<8x4xf32> {
51 // AVX2: vector.extract {{.*}}[0]
52 // AVX2-NEXT: vector.extract {{.*}}[1]
53 // AVX2-NEXT: vector.extract {{.*}}[2]
54 // AVX2-NEXT: vector.extract {{.*}}[3]
55 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
56 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
57 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
58 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
59 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
60 // AVX2-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
61 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
62 // AVX2-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
63 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
64 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
65 // AVX2-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
66 // AVX2-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
67 // AVX2-NEXT: vector.insert {{.*}}[0]
68 // AVX2-NEXT: vector.insert {{.*}}[1]
69 // AVX2-NEXT: vector.insert {{.*}}[2]
70 // AVX2-NEXT: vector.insert {{.*}}[3]
71 // AVX2-NEXT: vector.shape_cast {{.*}} vector<4x8xf32> to vector<32xf32>
72 // AVX2-NEXT: vector.shape_cast {{.*}} vector<32xf32> to vector<8x4xf32>
73 %0 = vector.transpose %arg0, [1, 0] : vector<4x8xf32> to vector<8x4xf32>
74 return %0 : vector<8x4xf32>
80 func.func @transpose021_1x4x8xf32(%arg0: vector<1x4x8xf32>) -> vector<1x8x4xf32> {
81 // AVX2: vector.extract {{.*}}[0, 0]
82 // AVX2-NEXT: vector.extract {{.*}}[0, 1]
83 // AVX2-NEXT: vector.extract {{.*}}[0, 2]
84 // AVX2-NEXT: vector.extract {{.*}}[0, 3]
85 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
86 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
87 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
88 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
89 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
90 // AVX2-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
91 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 8, 9, 4, 5, 12, 13] : vector<8xf32>, vector<8xf32>
92 // AVX2-NEXT: vector.shuffle {{.*}} [2, 3, 10, 11, 6, 7, 14, 15] : vector<8xf32>, vector<8xf32>
93 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
94 // AVX2-NEXT: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
95 // AVX2-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
96 // AVX2-NEXT: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
97 // AVX2-NEXT: vector.insert {{.*}}[0]
98 // AVX2-NEXT: vector.insert {{.*}}[1]
99 // AVX2-NEXT: vector.insert {{.*}}[2]
100 // AVX2-NEXT: vector.insert {{.*}}[3]
101 // AVX2-NEXT: vector.shape_cast {{.*}} vector<4x8xf32> to vector<32xf32>
102 // AVX2-NEXT: vector.shape_cast {{.*}} vector<32xf32> to vector<1x8x4xf32>
103 %0 = vector.transpose %arg0, [0, 2, 1] : vector<1x4x8xf32> to vector<1x8x4xf32>
104 return %0 : vector<1x8x4xf32>
110 func.func @transpose8x8xf32(%arg0: vector<8x8xf32>) -> vector<8x8xf32> {
111 // AVX2: vector.extract {{.*}}[0]
112 // AVX2-NEXT: vector.extract {{.*}}[1]
113 // AVX2-NEXT: vector.extract {{.*}}[2]
114 // AVX2-NEXT: vector.extract {{.*}}[3]
115 // AVX2-NEXT: vector.extract {{.*}}[4]
116 // AVX2-NEXT: vector.extract {{.*}}[5]
117 // AVX2-NEXT: vector.extract {{.*}}[6]
118 // AVX2-NEXT: vector.extract {{.*}}[7]
119 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
120 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
121 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
122 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
123 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
124 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
125 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
126 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
127 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
128 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
129 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
130 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
131 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
132 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
133 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
134 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
135 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
136 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
137 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
138 // AVX2-NEXT: vector.insert {{.*}}[0]
139 // AVX2-NEXT: vector.insert {{.*}}[1]
140 // AVX2-NEXT: vector.insert {{.*}}[2]
141 // AVX2-NEXT: vector.insert {{.*}}[3]
142 // AVX2-NEXT: vector.insert {{.*}}[4]
143 // AVX2-NEXT: vector.insert {{.*}}[5]
144 // AVX2-NEXT: vector.insert {{.*}}[6]
145 // AVX2-NEXT: vector.insert {{.*}}[7]
146 %0 = vector.transpose %arg0, [1, 0] : vector<8x8xf32> to vector<8x8xf32>
147 return %0 : vector<8x8xf32>
153 func.func @transpose021_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<1x8x8xf32> {
154 // AVX2: vector.extract {{.*}}[0, 0]
155 // AVX2-NEXT: vector.extract {{.*}}[0, 1]
156 // AVX2-NEXT: vector.extract {{.*}}[0, 2]
157 // AVX2-NEXT: vector.extract {{.*}}[0, 3]
158 // AVX2-NEXT: vector.extract {{.*}}[0, 4]
159 // AVX2-NEXT: vector.extract {{.*}}[0, 5]
160 // AVX2-NEXT: vector.extract {{.*}}[0, 6]
161 // AVX2-NEXT: vector.extract {{.*}}[0, 7]
162 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
163 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
164 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
165 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
166 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
167 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
168 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
169 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
170 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
171 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
172 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
173 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
174 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
175 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
176 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
177 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
178 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
179 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
180 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
181 // AVX2-NEXT: vector.insert {{.*}}[0]
182 // AVX2-NEXT: vector.insert {{.*}}[1]
183 // AVX2-NEXT: vector.insert {{.*}}[2]
184 // AVX2-NEXT: vector.insert {{.*}}[3]
185 // AVX2-NEXT: vector.insert {{.*}}[4]
186 // AVX2-NEXT: vector.insert {{.*}}[5]
187 // AVX2-NEXT: vector.insert {{.*}}[6]
188 // AVX2-NEXT: vector.insert {{.*}}[7]
189 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<1x8x8xf32>
190 %0 = vector.transpose %arg0, [0, 2, 1] : vector<1x8x8xf32> to vector<1x8x8xf32>
191 return %0 : vector<1x8x8xf32>
197 func.func @transpose120_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<1x8x8xf32> {
198 // AVX2: vector.extract {{.*}}[0, 0]
199 // AVX2-NEXT: vector.extract {{.*}}[1, 0]
200 // AVX2-NEXT: vector.extract {{.*}}[2, 0]
201 // AVX2-NEXT: vector.extract {{.*}}[3, 0]
202 // AVX2-NEXT: vector.extract {{.*}}[4, 0]
203 // AVX2-NEXT: vector.extract {{.*}}[5, 0]
204 // AVX2-NEXT: vector.extract {{.*}}[6, 0]
205 // AVX2-NEXT: vector.extract {{.*}}[7, 0]
206 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
207 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
208 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
209 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
210 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
211 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
212 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
213 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
214 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
215 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
216 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
217 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
218 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
219 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
220 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
221 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
222 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
223 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
224 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
225 // AVX2-NEXT: vector.insert {{.*}}[0]
226 // AVX2-NEXT: vector.insert {{.*}}[1]
227 // AVX2-NEXT: vector.insert {{.*}}[2]
228 // AVX2-NEXT: vector.insert {{.*}}[3]
229 // AVX2-NEXT: vector.insert {{.*}}[4]
230 // AVX2-NEXT: vector.insert {{.*}}[5]
231 // AVX2-NEXT: vector.insert {{.*}}[6]
232 // AVX2-NEXT: vector.insert {{.*}}[7]
233 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<1x8x8xf32>
234 %0 = vector.transpose %arg0, [1, 2, 0] : vector<8x1x8xf32> to vector<1x8x8xf32>
235 return %0 : vector<1x8x8xf32>
241 func.func @transpose120_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<8x1x8xf32> {
242 // AVX2: vector.shape_cast %{{.*}} : vector<8x8x1xf32> to vector<8x8xf32>
243 // AVX2-NEXT: vector.extract {{.*}}[0]
244 // AVX2-NEXT: vector.extract {{.*}}[1]
245 // AVX2-NEXT: vector.extract {{.*}}[2]
246 // AVX2-NEXT: vector.extract {{.*}}[3]
247 // AVX2-NEXT: vector.extract {{.*}}[4]
248 // AVX2-NEXT: vector.extract {{.*}}[5]
249 // AVX2-NEXT: vector.extract {{.*}}[6]
250 // AVX2-NEXT: vector.extract {{.*}}[7]
251 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
252 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
253 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
254 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
255 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
256 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
257 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
258 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
259 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
260 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
261 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
262 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
263 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
264 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
265 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
266 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
267 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
268 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
269 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
270 // AVX2-NEXT: vector.insert {{.*}}[0]
271 // AVX2-NEXT: vector.insert {{.*}}[1]
272 // AVX2-NEXT: vector.insert {{.*}}[2]
273 // AVX2-NEXT: vector.insert {{.*}}[3]
274 // AVX2-NEXT: vector.insert {{.*}}[4]
275 // AVX2-NEXT: vector.insert {{.*}}[5]
276 // AVX2-NEXT: vector.insert {{.*}}[6]
277 // AVX2-NEXT: vector.insert {{.*}}[7]
278 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x1x8xf32>
279 %0 = vector.transpose %arg0, [1, 2, 0] : vector<8x8x1xf32> to vector<8x1x8xf32>
280 return %0 : vector<8x1x8xf32>
286 func.func @transpose102_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<8x8x1xf32> {
287 // AVX2: vector.shape_cast %{{.*}} : vector<8x8x1xf32> to vector<8x8xf32>
288 // AVX2-NEXT: vector.extract {{.*}}[0]
289 // AVX2-NEXT: vector.extract {{.*}}[1]
290 // AVX2-NEXT: vector.extract {{.*}}[2]
291 // AVX2-NEXT: vector.extract {{.*}}[3]
292 // AVX2-NEXT: vector.extract {{.*}}[4]
293 // AVX2-NEXT: vector.extract {{.*}}[5]
294 // AVX2-NEXT: vector.extract {{.*}}[6]
295 // AVX2-NEXT: vector.extract {{.*}}[7]
296 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
297 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
298 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
299 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
300 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
301 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
302 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
303 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
304 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
305 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
306 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
307 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
308 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
309 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
310 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
311 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
312 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
313 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
314 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
315 // AVX2-NEXT: vector.insert {{.*}}[0]
316 // AVX2-NEXT: vector.insert {{.*}}[1]
317 // AVX2-NEXT: vector.insert {{.*}}[2]
318 // AVX2-NEXT: vector.insert {{.*}}[3]
319 // AVX2-NEXT: vector.insert {{.*}}[4]
320 // AVX2-NEXT: vector.insert {{.*}}[5]
321 // AVX2-NEXT: vector.insert {{.*}}[6]
322 // AVX2-NEXT: vector.insert {{.*}}[7]
323 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x8x1xf32>
324 %0 = vector.transpose %arg0, [1, 0, 2] : vector<8x8x1xf32> to vector<8x8x1xf32>
325 return %0 : vector<8x8x1xf32>
331 func.func @transpose201_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<8x8x1xf32> {
332 // AVX2: vector.extract {{.*}}[0, 0]
333 // AVX2-NEXT: vector.extract {{.*}}[1, 0]
334 // AVX2-NEXT: vector.extract {{.*}}[2, 0]
335 // AVX2-NEXT: vector.extract {{.*}}[3, 0]
336 // AVX2-NEXT: vector.extract {{.*}}[4, 0]
337 // AVX2-NEXT: vector.extract {{.*}}[5, 0]
338 // AVX2-NEXT: vector.extract {{.*}}[6, 0]
339 // AVX2-NEXT: vector.extract {{.*}}[7, 0]
340 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
341 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
342 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
343 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
344 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
345 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
346 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
347 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
348 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
349 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
350 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
351 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
352 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
353 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
354 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
355 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
356 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
357 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
358 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
359 // AVX2-NEXT: vector.insert {{.*}}[0]
360 // AVX2-NEXT: vector.insert {{.*}}[1]
361 // AVX2-NEXT: vector.insert {{.*}}[2]
362 // AVX2-NEXT: vector.insert {{.*}}[3]
363 // AVX2-NEXT: vector.insert {{.*}}[4]
364 // AVX2-NEXT: vector.insert {{.*}}[5]
365 // AVX2-NEXT: vector.insert {{.*}}[6]
366 // AVX2-NEXT: vector.insert {{.*}}[7]
367 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x8x1xf32>
368 %0 = vector.transpose %arg0, [2, 0, 1] : vector<8x1x8xf32> to vector<8x8x1xf32>
369 return %0 : vector<8x8x1xf32>
375 func.func @transpose201_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<8x1x8xf32> {
376 // AVX2: vector.extract {{.*}}[0, 0]
377 // AVX2-NEXT: vector.extract {{.*}}[0, 1]
378 // AVX2-NEXT: vector.extract {{.*}}[0, 2]
379 // AVX2-NEXT: vector.extract {{.*}}[0, 3]
380 // AVX2-NEXT: vector.extract {{.*}}[0, 4]
381 // AVX2-NEXT: vector.extract {{.*}}[0, 5]
382 // AVX2-NEXT: vector.extract {{.*}}[0, 6]
383 // AVX2-NEXT: vector.extract {{.*}}[0, 7]
384 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
385 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
386 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
387 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
388 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
389 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
390 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
391 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
392 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
393 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
394 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
395 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
396 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
397 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
398 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
399 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
400 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
401 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
402 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
403 // AVX2-NEXT: vector.insert {{.*}}[0]
404 // AVX2-NEXT: vector.insert {{.*}}[1]
405 // AVX2-NEXT: vector.insert {{.*}}[2]
406 // AVX2-NEXT: vector.insert {{.*}}[3]
407 // AVX2-NEXT: vector.insert {{.*}}[4]
408 // AVX2-NEXT: vector.insert {{.*}}[5]
409 // AVX2-NEXT: vector.insert {{.*}}[6]
410 // AVX2-NEXT: vector.insert {{.*}}[7]
411 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x1x8xf32>
412 %0 = vector.transpose %arg0, [2, 0, 1] : vector<1x8x8xf32> to vector<8x1x8xf32>
413 return %0 : vector<8x1x8xf32>
419 func.func @transpose210_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<8x1x8xf32> {
420 // AVX2: vector.extract {{.*}}[0, 0]
421 // AVX2-NEXT: vector.extract {{.*}}[1, 0]
422 // AVX2-NEXT: vector.extract {{.*}}[2, 0]
423 // AVX2-NEXT: vector.extract {{.*}}[3, 0]
424 // AVX2-NEXT: vector.extract {{.*}}[4, 0]
425 // AVX2-NEXT: vector.extract {{.*}}[5, 0]
426 // AVX2-NEXT: vector.extract {{.*}}[6, 0]
427 // AVX2-NEXT: vector.extract {{.*}}[7, 0]
428 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
429 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
430 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
431 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
432 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
433 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
434 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
435 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
436 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
437 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
438 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
439 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
440 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
441 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
442 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
443 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
444 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
445 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
446 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
447 // AVX2-NEXT: vector.insert {{.*}}[0]
448 // AVX2-NEXT: vector.insert {{.*}}[1]
449 // AVX2-NEXT: vector.insert {{.*}}[2]
450 // AVX2-NEXT: vector.insert {{.*}}[3]
451 // AVX2-NEXT: vector.insert {{.*}}[4]
452 // AVX2-NEXT: vector.insert {{.*}}[5]
453 // AVX2-NEXT: vector.insert {{.*}}[6]
454 // AVX2-NEXT: vector.insert {{.*}}[7]
455 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x1x8xf32>
456 %0 = vector.transpose %arg0, [2, 1, 0] : vector<8x1x8xf32> to vector<8x1x8xf32>
457 return %0 : vector<8x1x8xf32>
463 func.func @transpose210_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<1x8x8xf32> {
464 // AVX2: vector.shape_cast %{{.*}} : vector<8x8x1xf32> to vector<8x8xf32>
465 // AVX2-NEXT: vector.extract {{.*}}[0]
466 // AVX2-NEXT: vector.extract {{.*}}[1]
467 // AVX2-NEXT: vector.extract {{.*}}[2]
468 // AVX2-NEXT: vector.extract {{.*}}[3]
469 // AVX2-NEXT: vector.extract {{.*}}[4]
470 // AVX2-NEXT: vector.extract {{.*}}[5]
471 // AVX2-NEXT: vector.extract {{.*}}[6]
472 // AVX2-NEXT: vector.extract {{.*}}[7]
473 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
474 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
475 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
476 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
477 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
478 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
479 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
480 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
481 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
482 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
483 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
484 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
485 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
486 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
487 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
488 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
489 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
490 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
491 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
492 // AVX2-NEXT: vector.insert {{.*}}[0]
493 // AVX2-NEXT: vector.insert {{.*}}[1]
494 // AVX2-NEXT: vector.insert {{.*}}[2]
495 // AVX2-NEXT: vector.insert {{.*}}[3]
496 // AVX2-NEXT: vector.insert {{.*}}[4]
497 // AVX2-NEXT: vector.insert {{.*}}[5]
498 // AVX2-NEXT: vector.insert {{.*}}[6]
499 // AVX2-NEXT: vector.insert {{.*}}[7]
500 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<1x8x8xf32>
501 %0 = vector.transpose %arg0, [2, 1, 0] : vector<8x8x1xf32> to vector<1x8x8xf32>
502 return %0 : vector<1x8x8xf32>
508 func.func @transpose210_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<8x8x1xf32> {
509 // AVX2: vector.extract {{.*}}[0, 0]
510 // AVX2-NEXT: vector.extract {{.*}}[0, 1]
511 // AVX2-NEXT: vector.extract {{.*}}[0, 2]
512 // AVX2-NEXT: vector.extract {{.*}}[0, 3]
513 // AVX2-NEXT: vector.extract {{.*}}[0, 4]
514 // AVX2-NEXT: vector.extract {{.*}}[0, 5]
515 // AVX2-NEXT: vector.extract {{.*}}[0, 6]
516 // AVX2-NEXT: vector.extract {{.*}}[0, 7]
517 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
518 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
519 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
520 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
521 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
522 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
523 // AVX2-NEXT: vector.shuffle {{.*}} [0, 8, 1, 9, 4, 12, 5, 13] : vector<8xf32>, vector<8xf32>
524 // AVX2-NEXT: vector.shuffle {{.*}} [2, 10, 3, 11, 6, 14, 7, 15] : vector<8xf32>, vector<8xf32>
525 // AVX2-COUNT-4: vector.shuffle {{.*}} [2, 3, 8, 9, 6, 7, 12, 13] : vector<8xf32>, vector<8xf32>
526 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
527 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
528 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
529 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
530 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
531 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
532 … = intel "vblendps $0, $1, $2, 0xcc", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
533 … = intel "vblendps $0, $1, $2, 0x33", "=x,x,x" {{.*}} : (vector<8xf32>, vector<8xf32>) -> vector<8…
534 // AVX2-COUNT-4: vector.shuffle {{.*}} [0, 1, 2, 3, 8, 9, 10, 11] : vector<8xf32>, vector<8xf32>
535 // AVX2-COUNT-4: vector.shuffle {{.*}} [4, 5, 6, 7, 12, 13, 14, 15] : vector<8xf32>, vector<8xf32>
536 // AVX2-NEXT: vector.insert {{.*}}[0]
537 // AVX2-NEXT: vector.insert {{.*}}[1]
538 // AVX2-NEXT: vector.insert {{.*}}[2]
539 // AVX2-NEXT: vector.insert {{.*}}[3]
540 // AVX2-NEXT: vector.insert {{.*}}[4]
541 // AVX2-NEXT: vector.insert {{.*}}[5]
542 // AVX2-NEXT: vector.insert {{.*}}[6]
543 // AVX2-NEXT: vector.insert {{.*}}[7]
544 // AVX2-NEXT: vector.shape_cast %{{.*}} : vector<8x8xf32> to vector<8x8x1xf32>
545 %0 = vector.transpose %arg0, [2, 1, 0] : vector<1x8x8xf32> to vector<8x8x1xf32>
546 return %0 : vector<8x8x1xf32>
551 func.func @do_not_lower_nonf32_to_avx2(%arg0: vector<4x8xi32>) -> vector<8x4xi32> {
552 %0 = vector.transpose %arg0, [1, 0] : vector<4x8xi32> to vector<8x4xi32>
553 return %0 : vector<8x4xi32>
556 // AVX2-NOT: vector.shuffle
561 func.func @transpose021_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<8x8x1xf32> {
562 %0 = vector.transpose %arg0, [0, 2, 1] : vector<8x1x8xf32> to vector<8x8x1xf32>
563 return %0 : vector<8x8x1xf32>
566 // AVX2-NOT: vector.shuffle
571 func.func @transpose021_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<8x1x8xf32> {
572 %0 = vector.transpose %arg0, [0, 2, 1] : vector<8x8x1xf32> to vector<8x1x8xf32>
573 return %0 : vector<8x1x8xf32>
576 // AVX2-NOT: vector.shuffle
582 func.func @transpose102_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<8x1x8xf32> {
583 // ELTWISE: vector.extract {{.*}}[0, 0] : vector<1x8x8xf32>
584 // ELTWISE-NEXT: vector.insert {{.*}} [0, 0] : vector<8xf32> into vector<8x1x8xf32>
585 // ELTWISE-NEXT: vector.extract {{.*}}[0, 1] : vector<1x8x8xf32>
586 // ELTWISE-NEXT: vector.insert {{.*}} [1, 0] : vector<8xf32> into vector<8x1x8xf32>
587 // ELTWISE-NEXT: vector.extract {{.*}}[0, 2] : vector<1x8x8xf32>
588 // ELTWISE-NEXT: vector.insert {{.*}} [2, 0] : vector<8xf32> into vector<8x1x8xf32>
589 // ELTWISE-NEXT: vector.extract {{.*}}[0, 3] : vector<1x8x8xf32>
590 // ELTWISE-NEXT: vector.insert {{.*}} [3, 0] : vector<8xf32> into vector<8x1x8xf32>
591 // ELTWISE-NEXT: vector.extract {{.*}}[0, 4] : vector<1x8x8xf32>
592 // ELTWISE-NEXT: vector.insert {{.*}} [4, 0] : vector<8xf32> into vector<8x1x8xf32>
593 // ELTWISE-NEXT: vector.extract {{.*}}[0, 5] : vector<1x8x8xf32>
594 // ELTWISE-NEXT: vector.insert {{.*}} [5, 0] : vector<8xf32> into vector<8x1x8xf32>
595 // ELTWISE-NEXT: vector.extract {{.*}}[0, 6] : vector<1x8x8xf32>
596 // ELTWISE-NEXT: vector.insert {{.*}} [6, 0] : vector<8xf32> into vector<8x1x8xf32>
597 // ELTWISE-NEXT: vector.extract {{.*}}[0, 7] : vector<1x8x8xf32>
598 // ELTWISE-NEXT: vector.insert {{.*}} [7, 0] : vector<8xf32> into vector<8x1x8xf32>
599 %0 = vector.transpose %arg0, [1, 0, 2] : vector<1x8x8xf32> to vector<8x1x8xf32>
600 return %0 : vector<8x1x8xf32>
603 // AVX2-NOT: vector.shuffle
609 func.func @transpose102_8x1x8xf32(%arg0: vector<8x1x8xf32>) -> vector<1x8x8xf32> {
610 // ELTWISE: vector.extract {{.*}}[0, 0] : vector<8x1x8xf32>
611 // ELTWISE-NEXT: vector.insert {{.*}} [0, 0] : vector<8xf32> into vector<1x8x8xf32>
612 // ELTWISE-NEXT: vector.extract {{.*}}[1, 0] : vector<8x1x8xf32>
613 // ELTWISE-NEXT: vector.insert {{.*}} [0, 1] : vector<8xf32> into vector<1x8x8xf32>
614 // ELTWISE-NEXT: vector.extract {{.*}}[2, 0] : vector<8x1x8xf32>
615 // ELTWISE-NEXT: vector.insert {{.*}} [0, 2] : vector<8xf32> into vector<1x8x8xf32>
616 // ELTWISE-NEXT: vector.extract {{.*}}[3, 0] : vector<8x1x8xf32>
617 // ELTWISE-NEXT: vector.insert {{.*}} [0, 3] : vector<8xf32> into vector<1x8x8xf32>
618 // ELTWISE-NEXT: vector.extract {{.*}}[4, 0] : vector<8x1x8xf32>
619 // ELTWISE-NEXT: vector.insert {{.*}} [0, 4] : vector<8xf32> into vector<1x8x8xf32>
620 // ELTWISE-NEXT: vector.extract {{.*}}[5, 0] : vector<8x1x8xf32>
621 // ELTWISE-NEXT: vector.insert {{.*}} [0, 5] : vector<8xf32> into vector<1x8x8xf32>
622 // ELTWISE-NEXT: vector.extract {{.*}}[6, 0] : vector<8x1x8xf32>
623 // ELTWISE-NEXT: vector.insert {{.*}} [0, 6] : vector<8xf32> into vector<1x8x8xf32>
624 // ELTWISE-NEXT: vector.extract {{.*}}[7, 0] : vector<8x1x8xf32>
625 // ELTWISE-NEXT: vector.insert {{.*}} [0, 7] : vector<8xf32> into vector<1x8x8xf32>
626 %0 = vector.transpose %arg0, [1, 0, 2] : vector<8x1x8xf32> to vector<1x8x8xf32>
627 return %0 : vector<1x8x8xf32>
630 // AVX2-NOT: vector.shuffle
636 func.func @transpose1023_1x1x8x8xf32(%arg0: vector<1x1x8x8xf32>) -> vector<1x1x8x8xf32> {
638 // ELTWISE: vector.extract {{.*}}[0, 0] : vector<1x1x8x8xf32>
639 // ELTWISE-NEXT: vector.insert {{.*}} [0, 0] : vector<8x8xf32> into vector<1x1x8x8xf32>
640 %0 = vector.transpose %arg0, [1, 0, 2, 3] : vector<1x1x8x8xf32> to vector<1x1x8x8xf32>
641 return %0 : vector<1x1x8x8xf32>
644 // AVX2-NOT: vector.shuffle
649 func.func @transpose120_1x8x8xf32(%arg0: vector<1x8x8xf32>) -> vector<8x8x1xf32> {
651 %0 = vector.transpose %arg0, [1, 2, 0] : vector<1x8x8xf32> to vector<8x8x1xf32>
652 return %0 : vector<8x8x1xf32>
655 // AVX2-NOT: vector.shuffle
660 func.func @transpose201_8x8x1xf32(%arg0: vector<8x8x1xf32>) -> vector<1x8x8xf32> {
661 %0 = vector.transpose %arg0, [2, 0, 1] : vector<8x8x1xf32> to vector<1x8x8xf32>
662 return %0 : vector<1x8x8xf32>
665 // AVX2-NOT: vector.shuffle