1 ; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py 2 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE2 3 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+sse4.2 | FileCheck %s --check-prefixes=CHECK,SSE,SSE42 4 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1 5 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2 6 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+xop,+avx | FileCheck %s --check-prefixes=CHECK,XOP,XOPAVX1 7 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+xop,+avx2 | FileCheck %s --check-prefixes=CHECK,XOP,XOPAVX2 8 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F 9 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f,+avx512dq | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F 10 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BW 11 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512F 12 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f,+avx512dq,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512VL 13 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+avx512f,+avx512bw,+avx512vl | FileCheck %s --check-prefixes=CHECK,AVX512,AVX512BWVL 14 ; 15 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mcpu=slm | FileCheck %s --check-prefixes=CHECK,SSE,SSE42 16 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mcpu=goldmont | FileCheck %s --check-prefixes=CHECK,SSE,SSE42 17 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mcpu=btver2 | FileCheck %s --check-prefixes=BTVER2 18 19 ; Verify the cost of vector shift left instructions. 20 21 ; 22 ; Variable Shifts 23 ; 24 25 define <2 x i64> @var_shift_v2i64(<2 x i64> %a, <2 x i64> %b) { 26 ; SSE-LABEL: 'var_shift_v2i64' 27 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b 28 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 29 ; 30 ; AVX1-LABEL: 'var_shift_v2i64' 31 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b 32 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 33 ; 34 ; AVX2-LABEL: 'var_shift_v2i64' 35 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b 36 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 37 ; 38 ; XOP-LABEL: 'var_shift_v2i64' 39 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b 40 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 41 ; 42 ; AVX512-LABEL: 'var_shift_v2i64' 43 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %b 44 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 45 ; 46 ; BTVER2-LABEL: 'var_shift_v2i64' 47 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, %b 48 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 49 ; 50 %shift = shl <2 x i64> %a, %b 51 ret <2 x i64> %shift 52 } 53 54 define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) { 55 ; SSE-LABEL: 'var_shift_v4i64' 56 ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <4 x i64> %a, %b 57 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 58 ; 59 ; AVX1-LABEL: 'var_shift_v4i64' 60 ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, %b 61 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 62 ; 63 ; AVX2-LABEL: 'var_shift_v4i64' 64 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b 65 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 66 ; 67 ; XOPAVX1-LABEL: 'var_shift_v4i64' 68 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %b 69 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 70 ; 71 ; XOPAVX2-LABEL: 'var_shift_v4i64' 72 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b 73 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 74 ; 75 ; AVX512-LABEL: 'var_shift_v4i64' 76 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %b 77 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 78 ; 79 ; BTVER2-LABEL: 'var_shift_v4i64' 80 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, %b 81 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 82 ; 83 %shift = shl <4 x i64> %a, %b 84 ret <4 x i64> %shift 85 } 86 87 define <8 x i64> @var_shift_v8i64(<8 x i64> %a, <8 x i64> %b) { 88 ; SSE-LABEL: 'var_shift_v8i64' 89 ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <8 x i64> %a, %b 90 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 91 ; 92 ; AVX1-LABEL: 'var_shift_v8i64' 93 ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, %b 94 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 95 ; 96 ; AVX2-LABEL: 'var_shift_v8i64' 97 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b 98 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 99 ; 100 ; XOPAVX1-LABEL: 'var_shift_v8i64' 101 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %b 102 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 103 ; 104 ; XOPAVX2-LABEL: 'var_shift_v8i64' 105 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %b 106 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 107 ; 108 ; AVX512-LABEL: 'var_shift_v8i64' 109 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %b 110 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 111 ; 112 ; BTVER2-LABEL: 'var_shift_v8i64' 113 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, %b 114 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 115 ; 116 %shift = shl <8 x i64> %a, %b 117 ret <8 x i64> %shift 118 } 119 120 define <4 x i32> @var_shift_v4i32(<4 x i32> %a, <4 x i32> %b) { 121 ; SSE2-LABEL: 'var_shift_v4i32' 122 ; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i32> %a, %b 123 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 124 ; 125 ; SSE42-LABEL: 'var_shift_v4i32' 126 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b 127 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 128 ; 129 ; AVX1-LABEL: 'var_shift_v4i32' 130 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b 131 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 132 ; 133 ; AVX2-LABEL: 'var_shift_v4i32' 134 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b 135 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 136 ; 137 ; XOP-LABEL: 'var_shift_v4i32' 138 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b 139 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 140 ; 141 ; AVX512-LABEL: 'var_shift_v4i32' 142 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %b 143 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 144 ; 145 ; BTVER2-LABEL: 'var_shift_v4i32' 146 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i32> %a, %b 147 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 148 ; 149 %shift = shl <4 x i32> %a, %b 150 ret <4 x i32> %shift 151 } 152 153 define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) { 154 ; SSE2-LABEL: 'var_shift_v8i32' 155 ; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i32> %a, %b 156 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 157 ; 158 ; SSE42-LABEL: 'var_shift_v8i32' 159 ; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i32> %a, %b 160 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 161 ; 162 ; AVX1-LABEL: 'var_shift_v8i32' 163 ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i32> %a, %b 164 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 165 ; 166 ; AVX2-LABEL: 'var_shift_v8i32' 167 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b 168 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 169 ; 170 ; XOPAVX1-LABEL: 'var_shift_v8i32' 171 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %b 172 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 173 ; 174 ; XOPAVX2-LABEL: 'var_shift_v8i32' 175 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b 176 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 177 ; 178 ; AVX512-LABEL: 'var_shift_v8i32' 179 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %b 180 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 181 ; 182 ; BTVER2-LABEL: 'var_shift_v8i32' 183 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <8 x i32> %a, %b 184 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 185 ; 186 %shift = shl <8 x i32> %a, %b 187 ret <8 x i32> %shift 188 } 189 190 define <16 x i32> @var_shift_v16i32(<16 x i32> %a, <16 x i32> %b) { 191 ; SSE2-LABEL: 'var_shift_v16i32' 192 ; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %shift = shl <16 x i32> %a, %b 193 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 194 ; 195 ; SSE42-LABEL: 'var_shift_v16i32' 196 ; SSE42-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <16 x i32> %a, %b 197 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 198 ; 199 ; AVX1-LABEL: 'var_shift_v16i32' 200 ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <16 x i32> %a, %b 201 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 202 ; 203 ; AVX2-LABEL: 'var_shift_v16i32' 204 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, %b 205 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 206 ; 207 ; XOPAVX1-LABEL: 'var_shift_v16i32' 208 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %b 209 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 210 ; 211 ; XOPAVX2-LABEL: 'var_shift_v16i32' 212 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, %b 213 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 214 ; 215 ; AVX512-LABEL: 'var_shift_v16i32' 216 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %b 217 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 218 ; 219 ; BTVER2-LABEL: 'var_shift_v16i32' 220 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <16 x i32> %a, %b 221 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 222 ; 223 %shift = shl <16 x i32> %a, %b 224 ret <16 x i32> %shift 225 } 226 227 define <8 x i16> @var_shift_v8i16(<8 x i16> %a, <8 x i16> %b) { 228 ; SSE2-LABEL: 'var_shift_v8i16' 229 ; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %shift = shl <8 x i16> %a, %b 230 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 231 ; 232 ; SSE42-LABEL: 'var_shift_v8i16' 233 ; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i16> %a, %b 234 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 235 ; 236 ; AVX-LABEL: 'var_shift_v8i16' 237 ; AVX-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i16> %a, %b 238 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 239 ; 240 ; XOP-LABEL: 'var_shift_v8i16' 241 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b 242 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 243 ; 244 ; AVX512F-LABEL: 'var_shift_v8i16' 245 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i16> %a, %b 246 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 247 ; 248 ; AVX512BW-LABEL: 'var_shift_v8i16' 249 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b 250 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 251 ; 252 ; AVX512VL-LABEL: 'var_shift_v8i16' 253 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i16> %a, %b 254 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 255 ; 256 ; AVX512BWVL-LABEL: 'var_shift_v8i16' 257 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %b 258 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 259 ; 260 ; BTVER2-LABEL: 'var_shift_v8i16' 261 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %shift = shl <8 x i16> %a, %b 262 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 263 ; 264 %shift = shl <8 x i16> %a, %b 265 ret <8 x i16> %shift 266 } 267 268 define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) { 269 ; SSE2-LABEL: 'var_shift_v16i16' 270 ; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %shift = shl <16 x i16> %a, %b 271 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 272 ; 273 ; SSE42-LABEL: 'var_shift_v16i16' 274 ; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %shift = shl <16 x i16> %a, %b 275 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 276 ; 277 ; AVX1-LABEL: 'var_shift_v16i16' 278 ; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <16 x i16> %a, %b 279 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 280 ; 281 ; AVX2-LABEL: 'var_shift_v16i16' 282 ; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b 283 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 284 ; 285 ; XOP-LABEL: 'var_shift_v16i16' 286 ; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, %b 287 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 288 ; 289 ; AVX512F-LABEL: 'var_shift_v16i16' 290 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b 291 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 292 ; 293 ; AVX512BW-LABEL: 'var_shift_v16i16' 294 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b 295 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 296 ; 297 ; AVX512VL-LABEL: 'var_shift_v16i16' 298 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <16 x i16> %a, %b 299 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 300 ; 301 ; AVX512BWVL-LABEL: 'var_shift_v16i16' 302 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %b 303 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 304 ; 305 ; BTVER2-LABEL: 'var_shift_v16i16' 306 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %shift = shl <16 x i16> %a, %b 307 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 308 ; 309 %shift = shl <16 x i16> %a, %b 310 ret <16 x i16> %shift 311 } 312 313 define <32 x i16> @var_shift_v32i16(<32 x i16> %a, <32 x i16> %b) { 314 ; SSE2-LABEL: 'var_shift_v32i16' 315 ; SSE2-NEXT: Cost Model: Found an estimated cost of 128 for instruction: %shift = shl <32 x i16> %a, %b 316 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 317 ; 318 ; SSE42-LABEL: 'var_shift_v32i16' 319 ; SSE42-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %shift = shl <32 x i16> %a, %b 320 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 321 ; 322 ; AVX1-LABEL: 'var_shift_v32i16' 323 ; AVX1-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <32 x i16> %a, %b 324 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 325 ; 326 ; AVX2-LABEL: 'var_shift_v32i16' 327 ; AVX2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, %b 328 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 329 ; 330 ; XOP-LABEL: 'var_shift_v32i16' 331 ; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, %b 332 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 333 ; 334 ; AVX512F-LABEL: 'var_shift_v32i16' 335 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, %b 336 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 337 ; 338 ; AVX512BW-LABEL: 'var_shift_v32i16' 339 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b 340 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 341 ; 342 ; AVX512VL-LABEL: 'var_shift_v32i16' 343 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <32 x i16> %a, %b 344 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 345 ; 346 ; AVX512BWVL-LABEL: 'var_shift_v32i16' 347 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %b 348 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 349 ; 350 ; BTVER2-LABEL: 'var_shift_v32i16' 351 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %shift = shl <32 x i16> %a, %b 352 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 353 ; 354 %shift = shl <32 x i16> %a, %b 355 ret <32 x i16> %shift 356 } 357 358 define <16 x i8> @var_shift_v16i8(<16 x i8> %a, <16 x i8> %b) { 359 ; SSE2-LABEL: 'var_shift_v16i8' 360 ; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = shl <16 x i8> %a, %b 361 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 362 ; 363 ; SSE42-LABEL: 'var_shift_v16i8' 364 ; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %b 365 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 366 ; 367 ; AVX-LABEL: 'var_shift_v16i8' 368 ; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %b 369 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 370 ; 371 ; XOP-LABEL: 'var_shift_v16i8' 372 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %b 373 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 374 ; 375 ; AVX512-LABEL: 'var_shift_v16i8' 376 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %b 377 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 378 ; 379 ; BTVER2-LABEL: 'var_shift_v16i8' 380 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %b 381 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 382 ; 383 %shift = shl <16 x i8> %a, %b 384 ret <16 x i8> %shift 385 } 386 387 define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) { 388 ; SSE2-LABEL: 'var_shift_v32i8' 389 ; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = shl <32 x i8> %a, %b 390 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 391 ; 392 ; SSE42-LABEL: 'var_shift_v32i8' 393 ; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, %b 394 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 395 ; 396 ; AVX1-LABEL: 'var_shift_v32i8' 397 ; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <32 x i8> %a, %b 398 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 399 ; 400 ; AVX2-LABEL: 'var_shift_v32i8' 401 ; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, %b 402 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 403 ; 404 ; XOP-LABEL: 'var_shift_v32i8' 405 ; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i8> %a, %b 406 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 407 ; 408 ; AVX512-LABEL: 'var_shift_v32i8' 409 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, %b 410 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 411 ; 412 ; BTVER2-LABEL: 'var_shift_v32i8' 413 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <32 x i8> %a, %b 414 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 415 ; 416 %shift = shl <32 x i8> %a, %b 417 ret <32 x i8> %shift 418 } 419 420 define <64 x i8> @var_shift_v64i8(<64 x i8> %a, <64 x i8> %b) { 421 ; SSE2-LABEL: 'var_shift_v64i8' 422 ; SSE2-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = shl <64 x i8> %a, %b 423 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 424 ; 425 ; SSE42-LABEL: 'var_shift_v64i8' 426 ; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, %b 427 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 428 ; 429 ; AVX1-LABEL: 'var_shift_v64i8' 430 ; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <64 x i8> %a, %b 431 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 432 ; 433 ; AVX2-LABEL: 'var_shift_v64i8' 434 ; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %b 435 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 436 ; 437 ; XOP-LABEL: 'var_shift_v64i8' 438 ; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, %b 439 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 440 ; 441 ; AVX512F-LABEL: 'var_shift_v64i8' 442 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %b 443 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 444 ; 445 ; AVX512BW-LABEL: 'var_shift_v64i8' 446 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <64 x i8> %a, %b 447 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 448 ; 449 ; AVX512VL-LABEL: 'var_shift_v64i8' 450 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %b 451 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 452 ; 453 ; AVX512BWVL-LABEL: 'var_shift_v64i8' 454 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <64 x i8> %a, %b 455 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 456 ; 457 ; BTVER2-LABEL: 'var_shift_v64i8' 458 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <64 x i8> %a, %b 459 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 460 ; 461 %shift = shl <64 x i8> %a, %b 462 ret <64 x i8> %shift 463 } 464 465 ; 466 ; Uniform Variable Shifts 467 ; 468 469 define <2 x i64> @splatvar_shift_v2i64(<2 x i64> %a, i64 %b) { 470 ; CHECK-LABEL: 'splatvar_shift_v2i64' 471 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <2 x i64> undef, i64 %b, i32 0 472 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer 473 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %splat 474 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 475 ; 476 ; BTVER2-LABEL: 'splatvar_shift_v2i64' 477 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <2 x i64> undef, i64 %b, i32 0 478 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer 479 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, %splat 480 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 481 ; 482 %insert = insertelement <2 x i64> undef, i64 %b, i32 0 483 %splat = shufflevector <2 x i64> %insert, <2 x i64> undef, <2 x i32> zeroinitializer 484 %shift = shl <2 x i64> %a, %splat 485 ret <2 x i64> %shift 486 } 487 488 define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, i64 %b) { 489 ; SSE-LABEL: 'splatvar_shift_v4i64' 490 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 491 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 492 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, %splat 493 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 494 ; 495 ; AVX1-LABEL: 'splatvar_shift_v4i64' 496 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 497 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 498 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %splat 499 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 500 ; 501 ; AVX2-LABEL: 'splatvar_shift_v4i64' 502 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 503 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 504 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %splat 505 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 506 ; 507 ; XOPAVX1-LABEL: 'splatvar_shift_v4i64' 508 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 509 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 510 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %splat 511 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 512 ; 513 ; XOPAVX2-LABEL: 'splatvar_shift_v4i64' 514 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 515 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 516 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %splat 517 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 518 ; 519 ; AVX512-LABEL: 'splatvar_shift_v4i64' 520 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 521 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 522 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, %splat 523 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 524 ; 525 ; BTVER2-LABEL: 'splatvar_shift_v4i64' 526 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i64> undef, i64 %b, i32 0 527 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 528 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, %splat 529 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 530 ; 531 %insert = insertelement <4 x i64> undef, i64 %b, i32 0 532 %splat = shufflevector <4 x i64> %insert, <4 x i64> undef, <4 x i32> zeroinitializer 533 %shift = shl <4 x i64> %a, %splat 534 ret <4 x i64> %shift 535 } 536 537 define <8 x i64> @splatvar_shift_v8i64(<8 x i64> %a, i64 %b) { 538 ; SSE-LABEL: 'splatvar_shift_v8i64' 539 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 540 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 541 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, %splat 542 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 543 ; 544 ; AVX1-LABEL: 'splatvar_shift_v8i64' 545 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 546 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 547 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %splat 548 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 549 ; 550 ; AVX2-LABEL: 'splatvar_shift_v8i64' 551 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 552 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 553 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %splat 554 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 555 ; 556 ; XOPAVX1-LABEL: 'splatvar_shift_v8i64' 557 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 558 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 559 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %splat 560 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 561 ; 562 ; XOPAVX2-LABEL: 'splatvar_shift_v8i64' 563 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 564 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 565 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, %splat 566 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 567 ; 568 ; AVX512-LABEL: 'splatvar_shift_v8i64' 569 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 570 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 571 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, %splat 572 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 573 ; 574 ; BTVER2-LABEL: 'splatvar_shift_v8i64' 575 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i64> undef, i64 %b, i32 0 576 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 577 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, %splat 578 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 579 ; 580 %insert = insertelement <8 x i64> undef, i64 %b, i32 0 581 %splat = shufflevector <8 x i64> %insert, <8 x i64> undef, <8 x i32> zeroinitializer 582 %shift = shl <8 x i64> %a, %splat 583 ret <8 x i64> %shift 584 } 585 586 define <4 x i32> @splatvar_shift_v4i32(<4 x i32> %a, i32 %b) { 587 ; CHECK-LABEL: 'splatvar_shift_v4i32' 588 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i32> undef, i32 %b, i32 0 589 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer 590 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %splat 591 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 592 ; 593 ; BTVER2-LABEL: 'splatvar_shift_v4i32' 594 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <4 x i32> undef, i32 %b, i32 0 595 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer 596 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, %splat 597 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 598 ; 599 %insert = insertelement <4 x i32> undef, i32 %b, i32 0 600 %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer 601 %shift = shl <4 x i32> %a, %splat 602 ret <4 x i32> %shift 603 } 604 605 define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, i32 %b) { 606 ; SSE-LABEL: 'splatvar_shift_v8i32' 607 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 608 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 609 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, %splat 610 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 611 ; 612 ; AVX1-LABEL: 'splatvar_shift_v8i32' 613 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 614 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 615 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %splat 616 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 617 ; 618 ; AVX2-LABEL: 'splatvar_shift_v8i32' 619 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 620 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 621 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %splat 622 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 623 ; 624 ; XOPAVX1-LABEL: 'splatvar_shift_v8i32' 625 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 626 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 627 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %splat 628 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 629 ; 630 ; XOPAVX2-LABEL: 'splatvar_shift_v8i32' 631 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 632 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 633 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %splat 634 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 635 ; 636 ; AVX512-LABEL: 'splatvar_shift_v8i32' 637 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 638 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 639 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, %splat 640 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 641 ; 642 ; BTVER2-LABEL: 'splatvar_shift_v8i32' 643 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i32> undef, i32 %b, i32 0 644 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 645 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, %splat 646 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 647 ; 648 %insert = insertelement <8 x i32> undef, i32 %b, i32 0 649 %splat = shufflevector <8 x i32> %insert, <8 x i32> undef, <8 x i32> zeroinitializer 650 %shift = shl <8 x i32> %a, %splat 651 ret <8 x i32> %shift 652 } 653 654 define <16 x i32> @splatvar_shift_v16i32(<16 x i32> %a, i32 %b) { 655 ; SSE-LABEL: 'splatvar_shift_v16i32' 656 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 657 ; SSE-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 658 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, %splat 659 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 660 ; 661 ; AVX1-LABEL: 'splatvar_shift_v16i32' 662 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 663 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 664 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %splat 665 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 666 ; 667 ; AVX2-LABEL: 'splatvar_shift_v16i32' 668 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 669 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 670 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, %splat 671 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 672 ; 673 ; XOPAVX1-LABEL: 'splatvar_shift_v16i32' 674 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 675 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 676 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %splat 677 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 678 ; 679 ; XOPAVX2-LABEL: 'splatvar_shift_v16i32' 680 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 681 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 682 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, %splat 683 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 684 ; 685 ; AVX512-LABEL: 'splatvar_shift_v16i32' 686 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 687 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 688 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, %splat 689 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 690 ; 691 ; BTVER2-LABEL: 'splatvar_shift_v16i32' 692 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i32> undef, i32 %b, i32 0 693 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 694 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, %splat 695 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 696 ; 697 %insert = insertelement <16 x i32> undef, i32 %b, i32 0 698 %splat = shufflevector <16 x i32> %insert, <16 x i32> undef, <16 x i32> zeroinitializer 699 %shift = shl <16 x i32> %a, %splat 700 ret <16 x i32> %shift 701 } 702 703 define <8 x i16> @splatvar_shift_v8i16(<8 x i16> %a, i16 %b) { 704 ; SSE2-LABEL: 'splatvar_shift_v8i16' 705 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i16> undef, i16 %b, i32 0 706 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 707 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %splat 708 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 709 ; 710 ; SSE42-LABEL: 'splatvar_shift_v8i16' 711 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i16> undef, i16 %b, i32 0 712 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 713 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %splat 714 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 715 ; 716 ; AVX-LABEL: 'splatvar_shift_v8i16' 717 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i16> undef, i16 %b, i32 0 718 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 719 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %splat 720 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 721 ; 722 ; XOP-LABEL: 'splatvar_shift_v8i16' 723 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i16> undef, i16 %b, i32 0 724 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 725 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %splat 726 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 727 ; 728 ; AVX512-LABEL: 'splatvar_shift_v8i16' 729 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i16> undef, i16 %b, i32 0 730 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 731 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %splat 732 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 733 ; 734 ; BTVER2-LABEL: 'splatvar_shift_v8i16' 735 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <8 x i16> undef, i16 %b, i32 0 736 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 737 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, %splat 738 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 739 ; 740 %insert = insertelement <8 x i16> undef, i16 %b, i32 0 741 %splat = shufflevector <8 x i16> %insert, <8 x i16> undef, <8 x i32> zeroinitializer 742 %shift = shl <8 x i16> %a, %splat 743 ret <8 x i16> %shift 744 } 745 746 define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, i16 %b) { 747 ; SSE2-LABEL: 'splatvar_shift_v16i16' 748 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 749 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 750 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, %splat 751 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 752 ; 753 ; SSE42-LABEL: 'splatvar_shift_v16i16' 754 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 755 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 756 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, %splat 757 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 758 ; 759 ; AVX1-LABEL: 'splatvar_shift_v16i16' 760 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 761 ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 762 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, %splat 763 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 764 ; 765 ; AVX2-LABEL: 'splatvar_shift_v16i16' 766 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 767 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 768 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %splat 769 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 770 ; 771 ; XOPAVX1-LABEL: 'splatvar_shift_v16i16' 772 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 773 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 774 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, %splat 775 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 776 ; 777 ; XOPAVX2-LABEL: 'splatvar_shift_v16i16' 778 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 779 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 780 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %splat 781 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 782 ; 783 ; AVX512-LABEL: 'splatvar_shift_v16i16' 784 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 785 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 786 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, %splat 787 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 788 ; 789 ; BTVER2-LABEL: 'splatvar_shift_v16i16' 790 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i16> undef, i16 %b, i32 0 791 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 792 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, %splat 793 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 794 ; 795 %insert = insertelement <16 x i16> undef, i16 %b, i32 0 796 %splat = shufflevector <16 x i16> %insert, <16 x i16> undef, <16 x i32> zeroinitializer 797 %shift = shl <16 x i16> %a, %splat 798 ret <16 x i16> %shift 799 } 800 801 define <32 x i16> @splatvar_shift_v32i16(<32 x i16> %a, i16 %b) { 802 ; SSE2-LABEL: 'splatvar_shift_v32i16' 803 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 804 ; SSE2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 805 ; SSE2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, %splat 806 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 807 ; 808 ; SSE42-LABEL: 'splatvar_shift_v32i16' 809 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 810 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 811 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, %splat 812 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 813 ; 814 ; AVX1-LABEL: 'splatvar_shift_v32i16' 815 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 816 ; AVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 817 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, %splat 818 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 819 ; 820 ; AVX2-LABEL: 'splatvar_shift_v32i16' 821 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 822 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 823 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, %splat 824 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 825 ; 826 ; XOPAVX1-LABEL: 'splatvar_shift_v32i16' 827 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 828 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 829 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, %splat 830 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 831 ; 832 ; XOPAVX2-LABEL: 'splatvar_shift_v32i16' 833 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 834 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 835 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, %splat 836 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 837 ; 838 ; AVX512F-LABEL: 'splatvar_shift_v32i16' 839 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 840 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 841 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, %splat 842 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 843 ; 844 ; AVX512BW-LABEL: 'splatvar_shift_v32i16' 845 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 846 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 847 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %splat 848 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 849 ; 850 ; AVX512VL-LABEL: 'splatvar_shift_v32i16' 851 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 852 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 853 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, %splat 854 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 855 ; 856 ; AVX512BWVL-LABEL: 'splatvar_shift_v32i16' 857 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 858 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 859 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, %splat 860 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 861 ; 862 ; BTVER2-LABEL: 'splatvar_shift_v32i16' 863 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i16> undef, i16 %b, i32 0 864 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 865 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, %splat 866 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 867 ; 868 %insert = insertelement <32 x i16> undef, i16 %b, i32 0 869 %splat = shufflevector <32 x i16> %insert, <32 x i16> undef, <32 x i32> zeroinitializer 870 %shift = shl <32 x i16> %a, %splat 871 ret <32 x i16> %shift 872 } 873 874 define <16 x i8> @splatvar_shift_v16i8(<16 x i8> %a, i8 %b) { 875 ; SSE2-LABEL: 'splatvar_shift_v16i8' 876 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0 877 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 878 ; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = shl <16 x i8> %a, %splat 879 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 880 ; 881 ; SSE42-LABEL: 'splatvar_shift_v16i8' 882 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0 883 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 884 ; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %splat 885 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 886 ; 887 ; AVX-LABEL: 'splatvar_shift_v16i8' 888 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0 889 ; AVX-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 890 ; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %splat 891 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 892 ; 893 ; XOP-LABEL: 'splatvar_shift_v16i8' 894 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0 895 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 896 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, %splat 897 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 898 ; 899 ; AVX512-LABEL: 'splatvar_shift_v16i8' 900 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0 901 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 902 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %splat 903 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 904 ; 905 ; BTVER2-LABEL: 'splatvar_shift_v16i8' 906 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <16 x i8> undef, i8 %b, i32 0 907 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 908 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, %splat 909 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 910 ; 911 %insert = insertelement <16 x i8> undef, i8 %b, i32 0 912 %splat = shufflevector <16 x i8> %insert, <16 x i8> undef, <16 x i32> zeroinitializer 913 %shift = shl <16 x i8> %a, %splat 914 ret <16 x i8> %shift 915 } 916 917 define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, i8 %b) { 918 ; SSE2-LABEL: 'splatvar_shift_v32i8' 919 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 920 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 921 ; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = shl <32 x i8> %a, %splat 922 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 923 ; 924 ; SSE42-LABEL: 'splatvar_shift_v32i8' 925 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 926 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 927 ; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, %splat 928 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 929 ; 930 ; AVX1-LABEL: 'splatvar_shift_v32i8' 931 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 932 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 933 ; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <32 x i8> %a, %splat 934 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 935 ; 936 ; AVX2-LABEL: 'splatvar_shift_v32i8' 937 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 938 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 939 ; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, %splat 940 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 941 ; 942 ; XOPAVX1-LABEL: 'splatvar_shift_v32i8' 943 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 944 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 945 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i8> %a, %splat 946 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 947 ; 948 ; XOPAVX2-LABEL: 'splatvar_shift_v32i8' 949 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 950 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 951 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i8> %a, %splat 952 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 953 ; 954 ; AVX512-LABEL: 'splatvar_shift_v32i8' 955 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 956 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 957 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, %splat 958 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 959 ; 960 ; BTVER2-LABEL: 'splatvar_shift_v32i8' 961 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <32 x i8> undef, i8 %b, i32 0 962 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 963 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <32 x i8> %a, %splat 964 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 965 ; 966 %insert = insertelement <32 x i8> undef, i8 %b, i32 0 967 %splat = shufflevector <32 x i8> %insert, <32 x i8> undef, <32 x i32> zeroinitializer 968 %shift = shl <32 x i8> %a, %splat 969 ret <32 x i8> %shift 970 } 971 972 define <64 x i8> @splatvar_shift_v64i8(<64 x i8> %a, i8 %b) { 973 ; SSE2-LABEL: 'splatvar_shift_v64i8' 974 ; SSE2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 975 ; SSE2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 976 ; SSE2-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = shl <64 x i8> %a, %splat 977 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 978 ; 979 ; SSE42-LABEL: 'splatvar_shift_v64i8' 980 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 981 ; SSE42-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 982 ; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, %splat 983 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 984 ; 985 ; AVX1-LABEL: 'splatvar_shift_v64i8' 986 ; AVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 987 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 988 ; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <64 x i8> %a, %splat 989 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 990 ; 991 ; AVX2-LABEL: 'splatvar_shift_v64i8' 992 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 993 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 994 ; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %splat 995 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 996 ; 997 ; XOPAVX1-LABEL: 'splatvar_shift_v64i8' 998 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 999 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1000 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, %splat 1001 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1002 ; 1003 ; XOPAVX2-LABEL: 'splatvar_shift_v64i8' 1004 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1005 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1006 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, %splat 1007 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1008 ; 1009 ; AVX512F-LABEL: 'splatvar_shift_v64i8' 1010 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1011 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1012 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %splat 1013 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1014 ; 1015 ; AVX512BW-LABEL: 'splatvar_shift_v64i8' 1016 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1017 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1018 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <64 x i8> %a, %splat 1019 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1020 ; 1021 ; AVX512VL-LABEL: 'splatvar_shift_v64i8' 1022 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1023 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1024 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, %splat 1025 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1026 ; 1027 ; AVX512BWVL-LABEL: 'splatvar_shift_v64i8' 1028 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1029 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1030 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <64 x i8> %a, %splat 1031 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1032 ; 1033 ; BTVER2-LABEL: 'splatvar_shift_v64i8' 1034 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1035 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1036 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <64 x i8> %a, %splat 1037 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1038 ; 1039 %insert = insertelement <64 x i8> undef, i8 %b, i32 0 1040 %splat = shufflevector <64 x i8> %insert, <64 x i8> undef, <64 x i32> zeroinitializer 1041 %shift = shl <64 x i8> %a, %splat 1042 ret <64 x i8> %shift 1043 } 1044 1045 ; 1046 ; Constant Shifts 1047 ; 1048 1049 define <2 x i64> @constant_shift_v2i64(<2 x i64> %a) { 1050 ; SSE-LABEL: 'constant_shift_v2i64' 1051 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7> 1052 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1053 ; 1054 ; AVX1-LABEL: 'constant_shift_v2i64' 1055 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7> 1056 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1057 ; 1058 ; AVX2-LABEL: 'constant_shift_v2i64' 1059 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7> 1060 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1061 ; 1062 ; XOP-LABEL: 'constant_shift_v2i64' 1063 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7> 1064 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1065 ; 1066 ; AVX512-LABEL: 'constant_shift_v2i64' 1067 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7> 1068 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1069 ; 1070 ; BTVER2-LABEL: 'constant_shift_v2i64' 1071 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <2 x i64> %a, <i64 1, i64 7> 1072 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1073 ; 1074 %shift = shl <2 x i64> %a, <i64 1, i64 7> 1075 ret <2 x i64> %shift 1076 } 1077 1078 define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) { 1079 ; SSE-LABEL: 'constant_shift_v4i64' 1080 ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1081 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1082 ; 1083 ; AVX1-LABEL: 'constant_shift_v4i64' 1084 ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1085 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1086 ; 1087 ; AVX2-LABEL: 'constant_shift_v4i64' 1088 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1089 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1090 ; 1091 ; XOPAVX1-LABEL: 'constant_shift_v4i64' 1092 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1093 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1094 ; 1095 ; XOPAVX2-LABEL: 'constant_shift_v4i64' 1096 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1097 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1098 ; 1099 ; AVX512-LABEL: 'constant_shift_v4i64' 1100 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1101 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1102 ; 1103 ; BTVER2-LABEL: 'constant_shift_v4i64' 1104 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1105 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1106 ; 1107 %shift = shl <4 x i64> %a, <i64 1, i64 7, i64 15, i64 31> 1108 ret <4 x i64> %shift 1109 } 1110 1111 define <8 x i64> @constant_shift_v8i64(<8 x i64> %a) { 1112 ; SSE-LABEL: 'constant_shift_v8i64' 1113 ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1114 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1115 ; 1116 ; AVX1-LABEL: 'constant_shift_v8i64' 1117 ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1118 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1119 ; 1120 ; AVX2-LABEL: 'constant_shift_v8i64' 1121 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1122 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1123 ; 1124 ; XOPAVX1-LABEL: 'constant_shift_v8i64' 1125 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1126 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1127 ; 1128 ; XOPAVX2-LABEL: 'constant_shift_v8i64' 1129 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1130 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1131 ; 1132 ; AVX512-LABEL: 'constant_shift_v8i64' 1133 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1134 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1135 ; 1136 ; BTVER2-LABEL: 'constant_shift_v8i64' 1137 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1138 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1139 ; 1140 %shift = shl <8 x i64> %a, <i64 1, i64 7, i64 15, i64 31, i64 1, i64 7, i64 15, i64 31> 1141 ret <8 x i64> %shift 1142 } 1143 1144 define <4 x i32> @constant_shift_v4i32(<4 x i32> %a) { 1145 ; SSE2-LABEL: 'constant_shift_v4i32' 1146 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1147 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1148 ; 1149 ; SSE42-LABEL: 'constant_shift_v4i32' 1150 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1151 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1152 ; 1153 ; AVX1-LABEL: 'constant_shift_v4i32' 1154 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1155 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1156 ; 1157 ; AVX2-LABEL: 'constant_shift_v4i32' 1158 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1159 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1160 ; 1161 ; XOP-LABEL: 'constant_shift_v4i32' 1162 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1163 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1164 ; 1165 ; AVX512-LABEL: 'constant_shift_v4i32' 1166 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1167 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1168 ; 1169 ; BTVER2-LABEL: 'constant_shift_v4i32' 1170 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1171 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1172 ; 1173 %shift = shl <4 x i32> %a, <i32 4, i32 5, i32 6, i32 7> 1174 ret <4 x i32> %shift 1175 } 1176 1177 define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) { 1178 ; SSE2-LABEL: 'constant_shift_v8i32' 1179 ; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1180 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1181 ; 1182 ; SSE42-LABEL: 'constant_shift_v8i32' 1183 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1184 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1185 ; 1186 ; AVX1-LABEL: 'constant_shift_v8i32' 1187 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1188 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1189 ; 1190 ; AVX2-LABEL: 'constant_shift_v8i32' 1191 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1192 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1193 ; 1194 ; XOPAVX1-LABEL: 'constant_shift_v8i32' 1195 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1196 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1197 ; 1198 ; XOPAVX2-LABEL: 'constant_shift_v8i32' 1199 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1200 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1201 ; 1202 ; AVX512-LABEL: 'constant_shift_v8i32' 1203 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1204 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1205 ; 1206 ; BTVER2-LABEL: 'constant_shift_v8i32' 1207 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1208 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1209 ; 1210 %shift = shl <8 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1211 ret <8 x i32> %shift 1212 } 1213 1214 define <16 x i32> @constant_shift_v16i32(<16 x i32> %a) { 1215 ; SSE2-LABEL: 'constant_shift_v16i32' 1216 ; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1217 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1218 ; 1219 ; SSE42-LABEL: 'constant_shift_v16i32' 1220 ; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1221 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1222 ; 1223 ; AVX1-LABEL: 'constant_shift_v16i32' 1224 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1225 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1226 ; 1227 ; AVX2-LABEL: 'constant_shift_v16i32' 1228 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1229 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1230 ; 1231 ; XOPAVX1-LABEL: 'constant_shift_v16i32' 1232 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1233 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1234 ; 1235 ; XOPAVX2-LABEL: 'constant_shift_v16i32' 1236 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1237 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1238 ; 1239 ; AVX512-LABEL: 'constant_shift_v16i32' 1240 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1241 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1242 ; 1243 ; BTVER2-LABEL: 'constant_shift_v16i32' 1244 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1245 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1246 ; 1247 %shift = shl <16 x i32> %a, <i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 0, i32 1, i32 2, i32 3> 1248 ret <16 x i32> %shift 1249 } 1250 1251 define <8 x i16> @constant_shift_v8i16(<8 x i16> %a) { 1252 ; CHECK-LABEL: 'constant_shift_v8i16' 1253 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1254 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 1255 ; 1256 ; BTVER2-LABEL: 'constant_shift_v8i16' 1257 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1258 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 1259 ; 1260 %shift = shl <8 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1261 ret <8 x i16> %shift 1262 } 1263 1264 define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) { 1265 ; SSE-LABEL: 'constant_shift_v16i16' 1266 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1267 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1268 ; 1269 ; AVX1-LABEL: 'constant_shift_v16i16' 1270 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1271 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1272 ; 1273 ; AVX2-LABEL: 'constant_shift_v16i16' 1274 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1275 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1276 ; 1277 ; XOPAVX1-LABEL: 'constant_shift_v16i16' 1278 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1279 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1280 ; 1281 ; XOPAVX2-LABEL: 'constant_shift_v16i16' 1282 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1283 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1284 ; 1285 ; AVX512-LABEL: 'constant_shift_v16i16' 1286 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1287 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1288 ; 1289 ; BTVER2-LABEL: 'constant_shift_v16i16' 1290 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1291 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1292 ; 1293 %shift = shl <16 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1294 ret <16 x i16> %shift 1295 } 1296 1297 define <32 x i16> @constant_shift_v32i16(<32 x i16> %a) { 1298 ; SSE-LABEL: 'constant_shift_v32i16' 1299 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1300 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1301 ; 1302 ; AVX1-LABEL: 'constant_shift_v32i16' 1303 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1304 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1305 ; 1306 ; AVX2-LABEL: 'constant_shift_v32i16' 1307 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1308 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1309 ; 1310 ; XOPAVX1-LABEL: 'constant_shift_v32i16' 1311 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1312 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1313 ; 1314 ; XOPAVX2-LABEL: 'constant_shift_v32i16' 1315 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1316 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1317 ; 1318 ; AVX512F-LABEL: 'constant_shift_v32i16' 1319 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1320 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1321 ; 1322 ; AVX512BW-LABEL: 'constant_shift_v32i16' 1323 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1324 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1325 ; 1326 ; AVX512VL-LABEL: 'constant_shift_v32i16' 1327 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1328 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1329 ; 1330 ; AVX512BWVL-LABEL: 'constant_shift_v32i16' 1331 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1332 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1333 ; 1334 ; BTVER2-LABEL: 'constant_shift_v32i16' 1335 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1336 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1337 ; 1338 %shift = shl <32 x i16> %a, <i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7, i16 0, i16 1, i16 2, i16 3, i16 4, i16 5, i16 6, i16 7> 1339 ret <32 x i16> %shift 1340 } 1341 1342 define <16 x i8> @constant_shift_v16i8(<16 x i8> %a) { 1343 ; SSE2-LABEL: 'constant_shift_v16i8' 1344 ; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1345 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1346 ; 1347 ; SSE42-LABEL: 'constant_shift_v16i8' 1348 ; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1349 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1350 ; 1351 ; AVX-LABEL: 'constant_shift_v16i8' 1352 ; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1353 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1354 ; 1355 ; XOP-LABEL: 'constant_shift_v16i8' 1356 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1357 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1358 ; 1359 ; AVX512-LABEL: 'constant_shift_v16i8' 1360 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1361 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1362 ; 1363 ; BTVER2-LABEL: 'constant_shift_v16i8' 1364 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1365 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1366 ; 1367 %shift = shl <16 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1368 ret <16 x i8> %shift 1369 } 1370 1371 define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) { 1372 ; SSE2-LABEL: 'constant_shift_v32i8' 1373 ; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1374 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1375 ; 1376 ; SSE42-LABEL: 'constant_shift_v32i8' 1377 ; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1378 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1379 ; 1380 ; AVX1-LABEL: 'constant_shift_v32i8' 1381 ; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1382 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1383 ; 1384 ; AVX2-LABEL: 'constant_shift_v32i8' 1385 ; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1386 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1387 ; 1388 ; XOP-LABEL: 'constant_shift_v32i8' 1389 ; XOP-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1390 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1391 ; 1392 ; AVX512-LABEL: 'constant_shift_v32i8' 1393 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1394 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1395 ; 1396 ; BTVER2-LABEL: 'constant_shift_v32i8' 1397 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1398 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1399 ; 1400 %shift = shl <32 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1401 ret <32 x i8> %shift 1402 } 1403 1404 define <64 x i8> @constant_shift_v64i8(<64 x i8> %a) { 1405 ; SSE2-LABEL: 'constant_shift_v64i8' 1406 ; SSE2-NEXT: Cost Model: Found an estimated cost of 104 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1407 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1408 ; 1409 ; SSE42-LABEL: 'constant_shift_v64i8' 1410 ; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1411 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1412 ; 1413 ; AVX1-LABEL: 'constant_shift_v64i8' 1414 ; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1415 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1416 ; 1417 ; AVX2-LABEL: 'constant_shift_v64i8' 1418 ; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1419 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1420 ; 1421 ; XOP-LABEL: 'constant_shift_v64i8' 1422 ; XOP-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1423 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1424 ; 1425 ; AVX512F-LABEL: 'constant_shift_v64i8' 1426 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1427 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1428 ; 1429 ; AVX512BW-LABEL: 'constant_shift_v64i8' 1430 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1431 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1432 ; 1433 ; AVX512VL-LABEL: 'constant_shift_v64i8' 1434 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1435 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1436 ; 1437 ; AVX512BWVL-LABEL: 'constant_shift_v64i8' 1438 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1439 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1440 ; 1441 ; BTVER2-LABEL: 'constant_shift_v64i8' 1442 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1443 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1444 ; 1445 %shift = shl <64 x i8> %a, <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0, i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 7, i8 6, i8 5, i8 4, i8 3, i8 2, i8 1, i8 0> 1446 ret <64 x i8> %shift 1447 } 1448 1449 ; 1450 ; Uniform Constant Shifts 1451 ; 1452 1453 define <2 x i64> @splatconstant_shift_v2i64(<2 x i64> %a) { 1454 ; CHECK-LABEL: 'splatconstant_shift_v2i64' 1455 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 7, i64 7> 1456 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1457 ; 1458 ; BTVER2-LABEL: 'splatconstant_shift_v2i64' 1459 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <2 x i64> %a, <i64 7, i64 7> 1460 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shift 1461 ; 1462 %shift = shl <2 x i64> %a, <i64 7, i64 7> 1463 ret <2 x i64> %shift 1464 } 1465 1466 define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) { 1467 ; SSE-LABEL: 'splatconstant_shift_v4i64' 1468 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1469 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1470 ; 1471 ; AVX1-LABEL: 'splatconstant_shift_v4i64' 1472 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1473 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1474 ; 1475 ; AVX2-LABEL: 'splatconstant_shift_v4i64' 1476 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1477 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1478 ; 1479 ; XOPAVX1-LABEL: 'splatconstant_shift_v4i64' 1480 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1481 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1482 ; 1483 ; XOPAVX2-LABEL: 'splatconstant_shift_v4i64' 1484 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1485 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1486 ; 1487 ; AVX512-LABEL: 'splatconstant_shift_v4i64' 1488 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1489 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1490 ; 1491 ; BTVER2-LABEL: 'splatconstant_shift_v4i64' 1492 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1493 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shift 1494 ; 1495 %shift = shl <4 x i64> %a, <i64 7, i64 7, i64 7, i64 7> 1496 ret <4 x i64> %shift 1497 } 1498 1499 define <8 x i64> @splatconstant_shift_v8i64(<8 x i64> %a) { 1500 ; SSE-LABEL: 'splatconstant_shift_v8i64' 1501 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1502 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1503 ; 1504 ; AVX1-LABEL: 'splatconstant_shift_v8i64' 1505 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1506 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1507 ; 1508 ; AVX2-LABEL: 'splatconstant_shift_v8i64' 1509 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1510 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1511 ; 1512 ; XOPAVX1-LABEL: 'splatconstant_shift_v8i64' 1513 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1514 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1515 ; 1516 ; XOPAVX2-LABEL: 'splatconstant_shift_v8i64' 1517 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1518 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1519 ; 1520 ; AVX512-LABEL: 'splatconstant_shift_v8i64' 1521 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1522 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1523 ; 1524 ; BTVER2-LABEL: 'splatconstant_shift_v8i64' 1525 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1526 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shift 1527 ; 1528 %shift = shl <8 x i64> %a, <i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7, i64 7> 1529 ret <8 x i64> %shift 1530 } 1531 1532 define <4 x i32> @splatconstant_shift_v4i32(<4 x i32> %a) { 1533 ; CHECK-LABEL: 'splatconstant_shift_v4i32' 1534 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5> 1535 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1536 ; 1537 ; BTVER2-LABEL: 'splatconstant_shift_v4i32' 1538 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5> 1539 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shift 1540 ; 1541 %shift = shl <4 x i32> %a, <i32 5, i32 5, i32 5, i32 5> 1542 ret <4 x i32> %shift 1543 } 1544 1545 define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) { 1546 ; SSE-LABEL: 'splatconstant_shift_v8i32' 1547 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1548 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1549 ; 1550 ; AVX1-LABEL: 'splatconstant_shift_v8i32' 1551 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1552 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1553 ; 1554 ; AVX2-LABEL: 'splatconstant_shift_v8i32' 1555 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1556 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1557 ; 1558 ; XOPAVX1-LABEL: 'splatconstant_shift_v8i32' 1559 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1560 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1561 ; 1562 ; XOPAVX2-LABEL: 'splatconstant_shift_v8i32' 1563 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1564 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1565 ; 1566 ; AVX512-LABEL: 'splatconstant_shift_v8i32' 1567 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1568 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1569 ; 1570 ; BTVER2-LABEL: 'splatconstant_shift_v8i32' 1571 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1572 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shift 1573 ; 1574 %shift = shl <8 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1575 ret <8 x i32> %shift 1576 } 1577 1578 define <16 x i32> @splatconstant_shift_v16i32(<16 x i32> %a) { 1579 ; SSE-LABEL: 'splatconstant_shift_v16i32' 1580 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1581 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1582 ; 1583 ; AVX1-LABEL: 'splatconstant_shift_v16i32' 1584 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1585 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1586 ; 1587 ; AVX2-LABEL: 'splatconstant_shift_v16i32' 1588 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1589 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1590 ; 1591 ; XOPAVX1-LABEL: 'splatconstant_shift_v16i32' 1592 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1593 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1594 ; 1595 ; XOPAVX2-LABEL: 'splatconstant_shift_v16i32' 1596 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1597 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1598 ; 1599 ; AVX512-LABEL: 'splatconstant_shift_v16i32' 1600 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1601 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1602 ; 1603 ; BTVER2-LABEL: 'splatconstant_shift_v16i32' 1604 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1605 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shift 1606 ; 1607 %shift = shl <16 x i32> %a, <i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5, i32 5> 1608 ret <16 x i32> %shift 1609 } 1610 1611 define <8 x i16> @splatconstant_shift_v8i16(<8 x i16> %a) { 1612 ; CHECK-LABEL: 'splatconstant_shift_v8i16' 1613 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1614 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 1615 ; 1616 ; BTVER2-LABEL: 'splatconstant_shift_v8i16' 1617 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1618 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shift 1619 ; 1620 %shift = shl <8 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1621 ret <8 x i16> %shift 1622 } 1623 1624 define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) { 1625 ; SSE-LABEL: 'splatconstant_shift_v16i16' 1626 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1627 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1628 ; 1629 ; AVX1-LABEL: 'splatconstant_shift_v16i16' 1630 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1631 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1632 ; 1633 ; AVX2-LABEL: 'splatconstant_shift_v16i16' 1634 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1635 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1636 ; 1637 ; XOPAVX1-LABEL: 'splatconstant_shift_v16i16' 1638 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1639 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1640 ; 1641 ; XOPAVX2-LABEL: 'splatconstant_shift_v16i16' 1642 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1643 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1644 ; 1645 ; AVX512-LABEL: 'splatconstant_shift_v16i16' 1646 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1647 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1648 ; 1649 ; BTVER2-LABEL: 'splatconstant_shift_v16i16' 1650 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1651 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shift 1652 ; 1653 %shift = shl <16 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1654 ret <16 x i16> %shift 1655 } 1656 1657 define <32 x i16> @splatconstant_shift_v32i16(<32 x i16> %a) { 1658 ; SSE-LABEL: 'splatconstant_shift_v32i16' 1659 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1660 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1661 ; 1662 ; AVX1-LABEL: 'splatconstant_shift_v32i16' 1663 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1664 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1665 ; 1666 ; AVX2-LABEL: 'splatconstant_shift_v32i16' 1667 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1668 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1669 ; 1670 ; XOPAVX1-LABEL: 'splatconstant_shift_v32i16' 1671 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1672 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1673 ; 1674 ; XOPAVX2-LABEL: 'splatconstant_shift_v32i16' 1675 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1676 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1677 ; 1678 ; AVX512F-LABEL: 'splatconstant_shift_v32i16' 1679 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1680 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1681 ; 1682 ; AVX512BW-LABEL: 'splatconstant_shift_v32i16' 1683 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1684 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1685 ; 1686 ; AVX512VL-LABEL: 'splatconstant_shift_v32i16' 1687 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1688 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1689 ; 1690 ; AVX512BWVL-LABEL: 'splatconstant_shift_v32i16' 1691 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1692 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1693 ; 1694 ; BTVER2-LABEL: 'splatconstant_shift_v32i16' 1695 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1696 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shift 1697 ; 1698 %shift = shl <32 x i16> %a, <i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3, i16 3> 1699 ret <32 x i16> %shift 1700 } 1701 1702 define <16 x i8> @splatconstant_shift_v16i8(<16 x i8> %a) { 1703 ; SSE-LABEL: 'splatconstant_shift_v16i8' 1704 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1705 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1706 ; 1707 ; AVX-LABEL: 'splatconstant_shift_v16i8' 1708 ; AVX-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1709 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1710 ; 1711 ; XOP-LABEL: 'splatconstant_shift_v16i8' 1712 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1713 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1714 ; 1715 ; AVX512-LABEL: 'splatconstant_shift_v16i8' 1716 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1717 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1718 ; 1719 ; BTVER2-LABEL: 'splatconstant_shift_v16i8' 1720 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1721 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %shift 1722 ; 1723 %shift = shl <16 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1724 ret <16 x i8> %shift 1725 } 1726 1727 define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) { 1728 ; SSE-LABEL: 'splatconstant_shift_v32i8' 1729 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1730 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1731 ; 1732 ; AVX1-LABEL: 'splatconstant_shift_v32i8' 1733 ; AVX1-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1734 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1735 ; 1736 ; AVX2-LABEL: 'splatconstant_shift_v32i8' 1737 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1738 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1739 ; 1740 ; XOPAVX1-LABEL: 'splatconstant_shift_v32i8' 1741 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1742 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1743 ; 1744 ; XOPAVX2-LABEL: 'splatconstant_shift_v32i8' 1745 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1746 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1747 ; 1748 ; AVX512-LABEL: 'splatconstant_shift_v32i8' 1749 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1750 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1751 ; 1752 ; BTVER2-LABEL: 'splatconstant_shift_v32i8' 1753 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1754 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %shift 1755 ; 1756 %shift = shl <32 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1757 ret <32 x i8> %shift 1758 } 1759 1760 define <64 x i8> @splatconstant_shift_v64i8(<64 x i8> %a) { 1761 ; SSE-LABEL: 'splatconstant_shift_v64i8' 1762 ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1763 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1764 ; 1765 ; AVX1-LABEL: 'splatconstant_shift_v64i8' 1766 ; AVX1-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1767 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1768 ; 1769 ; AVX2-LABEL: 'splatconstant_shift_v64i8' 1770 ; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1771 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1772 ; 1773 ; XOPAVX1-LABEL: 'splatconstant_shift_v64i8' 1774 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1775 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1776 ; 1777 ; XOPAVX2-LABEL: 'splatconstant_shift_v64i8' 1778 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1779 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1780 ; 1781 ; AVX512F-LABEL: 'splatconstant_shift_v64i8' 1782 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1783 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1784 ; 1785 ; AVX512BW-LABEL: 'splatconstant_shift_v64i8' 1786 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1787 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1788 ; 1789 ; AVX512VL-LABEL: 'splatconstant_shift_v64i8' 1790 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1791 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1792 ; 1793 ; AVX512BWVL-LABEL: 'splatconstant_shift_v64i8' 1794 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1795 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1796 ; 1797 ; BTVER2-LABEL: 'splatconstant_shift_v64i8' 1798 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1799 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %shift 1800 ; 1801 %shift = shl <64 x i8> %a, <i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3, i8 3> 1802 ret <64 x i8> %shift 1803 } 1804 1805 ; 1806 ; Special Cases 1807 ; 1808 1809 ; We always emit a single pmullw in the case of v8i16 vector shifts by 1810 ; non-uniform constant. 1811 1812 define <8 x i16> @test1(<8 x i16> %a) { 1813 ; CHECK-LABEL: 'test1' 1814 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1815 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shl 1816 ; 1817 ; BTVER2-LABEL: 'test1' 1818 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1819 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shl 1820 ; 1821 %shl = shl <8 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1822 ret <8 x i16> %shl 1823 } 1824 1825 define <8 x i16> @test2(<8 x i16> %a) { 1826 ; CHECK-LABEL: 'test2' 1827 ; CHECK-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i16> %a, <i16 0, i16 undef, i16 0, i16 0, i16 1, i16 undef, i16 -1, i16 1> 1828 ; CHECK-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shl 1829 ; 1830 ; BTVER2-LABEL: 'test2' 1831 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i16> %a, <i16 0, i16 undef, i16 0, i16 0, i16 1, i16 undef, i16 -1, i16 1> 1832 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %shl 1833 ; 1834 %shl = shl <8 x i16> %a, <i16 0, i16 undef, i16 0, i16 0, i16 1, i16 undef, i16 -1, i16 1> 1835 ret <8 x i16> %shl 1836 } 1837 1838 ; With SSE4.1, v4i32 shifts can be lowered into a single pmulld instruction. 1839 ; Make sure that the estimated cost is always 2 except for the case where 1840 ; we only have SSE2 support. With SSE2, we are forced to special lower the 1841 ; v4i32 mul as a 2x shuffle, 2x pmuludq, 2x shuffle. 1842 1843 define <4 x i32> @test3(<4 x i32> %a) { 1844 ; SSE2-LABEL: 'test3' 1845 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1846 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1847 ; 1848 ; SSE42-LABEL: 'test3' 1849 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1850 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1851 ; 1852 ; AVX1-LABEL: 'test3' 1853 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1854 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1855 ; 1856 ; AVX2-LABEL: 'test3' 1857 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1858 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1859 ; 1860 ; XOP-LABEL: 'test3' 1861 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1862 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1863 ; 1864 ; AVX512-LABEL: 'test3' 1865 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1866 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1867 ; 1868 ; BTVER2-LABEL: 'test3' 1869 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1870 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1871 ; 1872 %shl = shl <4 x i32> %a, <i32 1, i32 -1, i32 2, i32 -3> 1873 ret <4 x i32> %shl 1874 } 1875 1876 define <4 x i32> @test4(<4 x i32> %a) { 1877 ; SSE2-LABEL: 'test4' 1878 ; SSE2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1879 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1880 ; 1881 ; SSE42-LABEL: 'test4' 1882 ; SSE42-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1883 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1884 ; 1885 ; AVX1-LABEL: 'test4' 1886 ; AVX1-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1887 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1888 ; 1889 ; AVX2-LABEL: 'test4' 1890 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1891 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1892 ; 1893 ; XOP-LABEL: 'test4' 1894 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1895 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1896 ; 1897 ; AVX512-LABEL: 'test4' 1898 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1899 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1900 ; 1901 ; BTVER2-LABEL: 'test4' 1902 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1903 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %shl 1904 ; 1905 %shl = shl <4 x i32> %a, <i32 0, i32 0, i32 1, i32 1> 1906 ret <4 x i32> %shl 1907 } 1908 1909 ; On AVX2 we are able to lower the following shift into a single 1910 ; vpsllvq. Therefore, the expected cost is only 1. 1911 ; In all other cases, this shift is scalarized as the target does not support 1912 ; vpsllv instructions. 1913 1914 define <2 x i64> @test5(<2 x i64> %a) { 1915 ; SSE-LABEL: 'test5' 1916 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3> 1917 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl 1918 ; 1919 ; AVX1-LABEL: 'test5' 1920 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3> 1921 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl 1922 ; 1923 ; AVX2-LABEL: 'test5' 1924 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3> 1925 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl 1926 ; 1927 ; XOP-LABEL: 'test5' 1928 ; XOP-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3> 1929 ; XOP-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl 1930 ; 1931 ; AVX512-LABEL: 'test5' 1932 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3> 1933 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl 1934 ; 1935 ; BTVER2-LABEL: 'test5' 1936 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <2 x i64> %a, <i64 2, i64 3> 1937 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %shl 1938 ; 1939 %shl = shl <2 x i64> %a, <i64 2, i64 3> 1940 ret <2 x i64> %shl 1941 } 1942 1943 ; v16i16 and v8i32 shift left by non-uniform constant are lowered into 1944 ; vector multiply instructions. With AVX (but not AVX2), the vector multiply 1945 ; is lowered into a sequence of: 1 extract + 2 vpmullw + 1 insert. 1946 ; 1947 ; With AVX2, instruction vpmullw works with 256bit quantities and 1948 ; therefore there is no need to split the resulting vector multiply into 1949 ; a sequence of two multiply. 1950 ; 1951 ; With SSE2 and SSE4.1, the vector shift cost for 'test6' is twice 1952 ; the cost computed in the case of 'test1'. That is because the backend 1953 ; simply emits 2 pmullw with no extract/insert. 1954 1955 define <16 x i16> @test6(<16 x i16> %a) { 1956 ; SSE-LABEL: 'test6' 1957 ; SSE-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1958 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1959 ; 1960 ; AVX1-LABEL: 'test6' 1961 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1962 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1963 ; 1964 ; AVX2-LABEL: 'test6' 1965 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1966 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1967 ; 1968 ; XOPAVX1-LABEL: 'test6' 1969 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1970 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1971 ; 1972 ; XOPAVX2-LABEL: 'test6' 1973 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1974 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1975 ; 1976 ; AVX512-LABEL: 'test6' 1977 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1978 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1979 ; 1980 ; BTVER2-LABEL: 'test6' 1981 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1982 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %shl 1983 ; 1984 %shl = shl <16 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 1985 ret <16 x i16> %shl 1986 } 1987 1988 ; With SSE2 and SSE4.1, the vector shift cost for 'test7' is twice 1989 ; the cost computed in the case of 'test3'. That is because the multiply 1990 ; is type-legalized into two 4i32 vector multiply. 1991 1992 define <8 x i32> @test7(<8 x i32> %a) { 1993 ; SSE2-LABEL: 'test7' 1994 ; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 1995 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 1996 ; 1997 ; SSE42-LABEL: 'test7' 1998 ; SSE42-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 1999 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2000 ; 2001 ; AVX1-LABEL: 'test7' 2002 ; AVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2003 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2004 ; 2005 ; AVX2-LABEL: 'test7' 2006 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2007 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2008 ; 2009 ; XOPAVX1-LABEL: 'test7' 2010 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2011 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2012 ; 2013 ; XOPAVX2-LABEL: 'test7' 2014 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2015 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2016 ; 2017 ; AVX512-LABEL: 'test7' 2018 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2019 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2020 ; 2021 ; BTVER2-LABEL: 'test7' 2022 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2023 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %shl 2024 ; 2025 %shl = shl <8 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2026 ret <8 x i32> %shl 2027 } 2028 2029 ; On AVX2 we are able to lower the following shift into a single 2030 ; vpsllvq. Therefore, the expected cost is only 1. 2031 ; In all other cases, this shift is scalarized as the target does not support 2032 ; vpsllv instructions. 2033 2034 define <4 x i64> @test8(<4 x i64> %a) { 2035 ; SSE-LABEL: 'test8' 2036 ; SSE-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2037 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2038 ; 2039 ; AVX1-LABEL: 'test8' 2040 ; AVX1-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2041 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2042 ; 2043 ; AVX2-LABEL: 'test8' 2044 ; AVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2045 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2046 ; 2047 ; XOPAVX1-LABEL: 'test8' 2048 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2049 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2050 ; 2051 ; XOPAVX2-LABEL: 'test8' 2052 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2053 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2054 ; 2055 ; AVX512-LABEL: 'test8' 2056 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2057 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2058 ; 2059 ; BTVER2-LABEL: 'test8' 2060 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2061 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %shl 2062 ; 2063 %shl = shl <4 x i64> %a, <i64 1, i64 2, i64 3, i64 4> 2064 ret <4 x i64> %shl 2065 } 2066 2067 ; Same as 'test6', with the difference that the cost is double. 2068 2069 define <32 x i16> @test9(<32 x i16> %a) { 2070 ; SSE-LABEL: 'test9' 2071 ; SSE-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2072 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2073 ; 2074 ; AVX1-LABEL: 'test9' 2075 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2076 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2077 ; 2078 ; AVX2-LABEL: 'test9' 2079 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2080 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2081 ; 2082 ; XOPAVX1-LABEL: 'test9' 2083 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2084 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2085 ; 2086 ; XOPAVX2-LABEL: 'test9' 2087 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2088 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2089 ; 2090 ; AVX512F-LABEL: 'test9' 2091 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2092 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2093 ; 2094 ; AVX512BW-LABEL: 'test9' 2095 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2096 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2097 ; 2098 ; AVX512VL-LABEL: 'test9' 2099 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2100 ; AVX512VL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2101 ; 2102 ; AVX512BWVL-LABEL: 'test9' 2103 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2104 ; AVX512BWVL-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2105 ; 2106 ; BTVER2-LABEL: 'test9' 2107 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2108 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %shl 2109 ; 2110 %shl = shl <32 x i16> %a, <i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11, i16 1, i16 1, i16 2, i16 3, i16 7, i16 0, i16 9, i16 11> 2111 ret <32 x i16> %shl 2112 } 2113 2114 ; Same as 'test7', except that now the cost is double. 2115 2116 define <16 x i32> @test10(<16 x i32> %a) { 2117 ; SSE2-LABEL: 'test10' 2118 ; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2119 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2120 ; 2121 ; SSE42-LABEL: 'test10' 2122 ; SSE42-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2123 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2124 ; 2125 ; AVX1-LABEL: 'test10' 2126 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2127 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2128 ; 2129 ; AVX2-LABEL: 'test10' 2130 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2131 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2132 ; 2133 ; XOPAVX1-LABEL: 'test10' 2134 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2135 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2136 ; 2137 ; XOPAVX2-LABEL: 'test10' 2138 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2139 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2140 ; 2141 ; AVX512-LABEL: 'test10' 2142 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2143 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2144 ; 2145 ; BTVER2-LABEL: 'test10' 2146 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2147 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %shl 2148 ; 2149 %shl = shl <16 x i32> %a, <i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3, i32 1, i32 1, i32 2, i32 3> 2150 ret <16 x i32> %shl 2151 } 2152 2153 ; On AVX2 we are able to lower the following shift into a sequence of 2154 ; two vpsllvq instructions. Therefore, the expected cost is only 2. 2155 ; In all other cases, this shift is scalarized as we don't have vpsllv 2156 ; instructions. 2157 2158 define <8 x i64> @test11(<8 x i64> %a) { 2159 ; SSE-LABEL: 'test11' 2160 ; SSE-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2161 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2162 ; 2163 ; AVX1-LABEL: 'test11' 2164 ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2165 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2166 ; 2167 ; AVX2-LABEL: 'test11' 2168 ; AVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2169 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2170 ; 2171 ; XOPAVX1-LABEL: 'test11' 2172 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2173 ; XOPAVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2174 ; 2175 ; XOPAVX2-LABEL: 'test11' 2176 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2177 ; XOPAVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2178 ; 2179 ; AVX512-LABEL: 'test11' 2180 ; AVX512-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2181 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2182 ; 2183 ; BTVER2-LABEL: 'test11' 2184 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2185 ; BTVER2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %shl 2186 ; 2187 %shl = shl <8 x i64> %a, <i64 1, i64 1, i64 2, i64 3, i64 1, i64 1, i64 2, i64 3> 2188 ret <8 x i64> %shl 2189 } 2190
