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=-popcnt,+sse2 | FileCheck %s -check-prefixes=CHECK,SSE,SSE2,NOPOPCNT 3 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+popcnt,+sse2 | FileCheck %s -check-prefixes=CHECK,SSE,SSE2,POPCNT 4 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+popcnt,+sse4.2 | FileCheck %s -check-prefixes=CHECK,POPCNT,SSE,SSE42 5 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+popcnt,+avx | FileCheck %s -check-prefixes=CHECK,POPCNT,AVX,AVX1 6 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+popcnt,+avx2 | FileCheck %s -check-prefixes=CHECK,POPCNT,AVX,AVX2 7 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+popcnt,+avx512f | FileCheck %s -check-prefixes=CHECK,POPCNT,AVX512,AVX512F 8 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -cost-model -analyze -mattr=+popcnt,+avx512vl,+avx512bw,+avx512dq | FileCheck %s -check-prefixes=CHECK,POPCNT,AVX512,AVX512BW 9 10 ; Verify the cost of scalar population count instructions. 11 12 declare i64 @llvm.ctpop.i64(i64) 13 declare i32 @llvm.ctpop.i32(i32) 14 declare i16 @llvm.ctpop.i16(i16) 15 declare i8 @llvm.ctpop.i8(i8) 16 17 define i64 @var_ctpop_i64(i64 %a) { 18 ; NOPOPCNT-LABEL: 'var_ctpop_i64' 19 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %ctpop = call i64 @llvm.ctpop.i64(i64 %a) 20 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %ctpop 21 ; 22 ; POPCNT-LABEL: 'var_ctpop_i64' 23 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ctpop = call i64 @llvm.ctpop.i64(i64 %a) 24 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %ctpop 25 ; 26 %ctpop = call i64 @llvm.ctpop.i64(i64 %a) 27 ret i64 %ctpop 28 } 29 30 define i32 @var_ctpop_i32(i32 %a) { 31 ; NOPOPCNT-LABEL: 'var_ctpop_i32' 32 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %ctpop = call i32 @llvm.ctpop.i32(i32 %a) 33 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %ctpop 34 ; 35 ; POPCNT-LABEL: 'var_ctpop_i32' 36 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ctpop = call i32 @llvm.ctpop.i32(i32 %a) 37 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %ctpop 38 ; 39 %ctpop = call i32 @llvm.ctpop.i32(i32 %a) 40 ret i32 %ctpop 41 } 42 43 define i16 @var_ctpop_i16(i16 %a) { 44 ; NOPOPCNT-LABEL: 'var_ctpop_i16' 45 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %ctpop = call i16 @llvm.ctpop.i16(i16 %a) 46 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %ctpop 47 ; 48 ; POPCNT-LABEL: 'var_ctpop_i16' 49 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ctpop = call i16 @llvm.ctpop.i16(i16 %a) 50 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %ctpop 51 ; 52 %ctpop = call i16 @llvm.ctpop.i16(i16 %a) 53 ret i16 %ctpop 54 } 55 56 define i8 @var_ctpop_i8(i8 %a) { 57 ; NOPOPCNT-LABEL: 'var_ctpop_i8' 58 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %ctpop = call i8 @llvm.ctpop.i8(i8 %a) 59 ; NOPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i8 %ctpop 60 ; 61 ; POPCNT-LABEL: 'var_ctpop_i8' 62 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %ctpop = call i8 @llvm.ctpop.i8(i8 %a) 63 ; POPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i8 %ctpop 64 ; 65 %ctpop = call i8 @llvm.ctpop.i8(i8 %a) 66 ret i8 %ctpop 67 } 68 69 ; Verify the cost of vector population count instructions. 70 71 declare <2 x i64> @llvm.ctpop.v2i64(<2 x i64>) 72 declare <4 x i32> @llvm.ctpop.v4i32(<4 x i32>) 73 declare <8 x i16> @llvm.ctpop.v8i16(<8 x i16>) 74 declare <16 x i8> @llvm.ctpop.v16i8(<16 x i8>) 75 76 declare <4 x i64> @llvm.ctpop.v4i64(<4 x i64>) 77 declare <8 x i32> @llvm.ctpop.v8i32(<8 x i32>) 78 declare <16 x i16> @llvm.ctpop.v16i16(<16 x i16>) 79 declare <32 x i8> @llvm.ctpop.v32i8(<32 x i8>) 80 81 declare <8 x i64> @llvm.ctpop.v8i64(<8 x i64>) 82 declare <16 x i32> @llvm.ctpop.v16i32(<16 x i32>) 83 declare <32 x i16> @llvm.ctpop.v32i16(<32 x i16>) 84 declare <64 x i8> @llvm.ctpop.v64i8(<64 x i8>) 85 86 define <2 x i64> @var_ctpop_v2i64(<2 x i64> %a) { 87 ; SSE2-LABEL: 'var_ctpop_v2i64' 88 ; SSE2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a) 89 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %ctpop 90 ; 91 ; SSE42-LABEL: 'var_ctpop_v2i64' 92 ; SSE42-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a) 93 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %ctpop 94 ; 95 ; AVX-LABEL: 'var_ctpop_v2i64' 96 ; AVX-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a) 97 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %ctpop 98 ; 99 ; AVX512-LABEL: 'var_ctpop_v2i64' 100 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a) 101 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %ctpop 102 ; 103 %ctpop = call <2 x i64> @llvm.ctpop.v2i64(<2 x i64> %a) 104 ret <2 x i64> %ctpop 105 } 106 107 define <4 x i64> @var_ctpop_v4i64(<4 x i64> %a) { 108 ; SSE2-LABEL: 'var_ctpop_v4i64' 109 ; SSE2-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a) 110 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %ctpop 111 ; 112 ; SSE42-LABEL: 'var_ctpop_v4i64' 113 ; SSE42-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a) 114 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %ctpop 115 ; 116 ; AVX1-LABEL: 'var_ctpop_v4i64' 117 ; AVX1-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a) 118 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %ctpop 119 ; 120 ; AVX2-LABEL: 'var_ctpop_v4i64' 121 ; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a) 122 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %ctpop 123 ; 124 ; AVX512-LABEL: 'var_ctpop_v4i64' 125 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a) 126 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %ctpop 127 ; 128 %ctpop = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> %a) 129 ret <4 x i64> %ctpop 130 } 131 132 define <8 x i64> @var_ctpop_v8i64(<8 x i64> %a) { 133 ; SSE2-LABEL: 'var_ctpop_v8i64' 134 ; SSE2-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 135 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %ctpop 136 ; 137 ; SSE42-LABEL: 'var_ctpop_v8i64' 138 ; SSE42-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 139 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %ctpop 140 ; 141 ; AVX1-LABEL: 'var_ctpop_v8i64' 142 ; AVX1-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 143 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %ctpop 144 ; 145 ; AVX2-LABEL: 'var_ctpop_v8i64' 146 ; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 147 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %ctpop 148 ; 149 ; AVX512F-LABEL: 'var_ctpop_v8i64' 150 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 151 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %ctpop 152 ; 153 ; AVX512BW-LABEL: 'var_ctpop_v8i64' 154 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 155 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %ctpop 156 ; 157 %ctpop = call <8 x i64> @llvm.ctpop.v8i64(<8 x i64> %a) 158 ret <8 x i64> %ctpop 159 } 160 161 define <4 x i32> @var_ctpop_v4i32(<4 x i32> %a) { 162 ; SSE2-LABEL: 'var_ctpop_v4i32' 163 ; SSE2-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %ctpop = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a) 164 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %ctpop 165 ; 166 ; SSE42-LABEL: 'var_ctpop_v4i32' 167 ; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %ctpop = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a) 168 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %ctpop 169 ; 170 ; AVX-LABEL: 'var_ctpop_v4i32' 171 ; AVX-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %ctpop = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a) 172 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %ctpop 173 ; 174 ; AVX512-LABEL: 'var_ctpop_v4i32' 175 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %ctpop = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a) 176 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %ctpop 177 ; 178 %ctpop = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> %a) 179 ret <4 x i32> %ctpop 180 } 181 182 define <8 x i32> @var_ctpop_v8i32(<8 x i32> %a) { 183 ; SSE2-LABEL: 'var_ctpop_v8i32' 184 ; SSE2-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a) 185 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %ctpop 186 ; 187 ; SSE42-LABEL: 'var_ctpop_v8i32' 188 ; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a) 189 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %ctpop 190 ; 191 ; AVX1-LABEL: 'var_ctpop_v8i32' 192 ; AVX1-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a) 193 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %ctpop 194 ; 195 ; AVX2-LABEL: 'var_ctpop_v8i32' 196 ; AVX2-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a) 197 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %ctpop 198 ; 199 ; AVX512-LABEL: 'var_ctpop_v8i32' 200 ; AVX512-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a) 201 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %ctpop 202 ; 203 %ctpop = call <8 x i32> @llvm.ctpop.v8i32(<8 x i32> %a) 204 ret <8 x i32> %ctpop 205 } 206 207 define <16 x i32> @var_ctpop_v16i32(<16 x i32> %a) { 208 ; SSE2-LABEL: 'var_ctpop_v16i32' 209 ; SSE2-NEXT: Cost Model: Found an estimated cost of 60 for instruction: %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 210 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %ctpop 211 ; 212 ; SSE42-LABEL: 'var_ctpop_v16i32' 213 ; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 214 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %ctpop 215 ; 216 ; AVX1-LABEL: 'var_ctpop_v16i32' 217 ; AVX1-NEXT: Cost Model: Found an estimated cost of 48 for instruction: %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 218 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %ctpop 219 ; 220 ; AVX2-LABEL: 'var_ctpop_v16i32' 221 ; AVX2-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 222 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %ctpop 223 ; 224 ; AVX512F-LABEL: 'var_ctpop_v16i32' 225 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 226 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %ctpop 227 ; 228 ; AVX512BW-LABEL: 'var_ctpop_v16i32' 229 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 230 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %ctpop 231 ; 232 %ctpop = call <16 x i32> @llvm.ctpop.v16i32(<16 x i32> %a) 233 ret <16 x i32> %ctpop 234 } 235 236 define <8 x i16> @var_ctpop_v8i16(<8 x i16> %a) { 237 ; SSE2-LABEL: 'var_ctpop_v8i16' 238 ; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a) 239 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %ctpop 240 ; 241 ; SSE42-LABEL: 'var_ctpop_v8i16' 242 ; SSE42-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a) 243 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %ctpop 244 ; 245 ; AVX-LABEL: 'var_ctpop_v8i16' 246 ; AVX-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a) 247 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %ctpop 248 ; 249 ; AVX512-LABEL: 'var_ctpop_v8i16' 250 ; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a) 251 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %ctpop 252 ; 253 %ctpop = call <8 x i16> @llvm.ctpop.v8i16(<8 x i16> %a) 254 ret <8 x i16> %ctpop 255 } 256 257 define <16 x i16> @var_ctpop_v16i16(<16 x i16> %a) { 258 ; SSE2-LABEL: 'var_ctpop_v16i16' 259 ; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a) 260 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %ctpop 261 ; 262 ; SSE42-LABEL: 'var_ctpop_v16i16' 263 ; SSE42-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a) 264 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %ctpop 265 ; 266 ; AVX1-LABEL: 'var_ctpop_v16i16' 267 ; AVX1-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a) 268 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %ctpop 269 ; 270 ; AVX2-LABEL: 'var_ctpop_v16i16' 271 ; AVX2-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a) 272 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %ctpop 273 ; 274 ; AVX512-LABEL: 'var_ctpop_v16i16' 275 ; AVX512-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a) 276 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %ctpop 277 ; 278 %ctpop = call <16 x i16> @llvm.ctpop.v16i16(<16 x i16> %a) 279 ret <16 x i16> %ctpop 280 } 281 282 define <32 x i16> @var_ctpop_v32i16(<32 x i16> %a) { 283 ; SSE2-LABEL: 'var_ctpop_v32i16' 284 ; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 285 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %ctpop 286 ; 287 ; SSE42-LABEL: 'var_ctpop_v32i16' 288 ; SSE42-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 289 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %ctpop 290 ; 291 ; AVX1-LABEL: 'var_ctpop_v32i16' 292 ; AVX1-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 293 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %ctpop 294 ; 295 ; AVX2-LABEL: 'var_ctpop_v32i16' 296 ; AVX2-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 297 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %ctpop 298 ; 299 ; AVX512F-LABEL: 'var_ctpop_v32i16' 300 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 301 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %ctpop 302 ; 303 ; AVX512BW-LABEL: 'var_ctpop_v32i16' 304 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 305 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %ctpop 306 ; 307 %ctpop = call <32 x i16> @llvm.ctpop.v32i16(<32 x i16> %a) 308 ret <32 x i16> %ctpop 309 } 310 311 define <16 x i8> @var_ctpop_v16i8(<16 x i8> %a) { 312 ; SSE2-LABEL: 'var_ctpop_v16i8' 313 ; SSE2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %ctpop = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a) 314 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %ctpop 315 ; 316 ; SSE42-LABEL: 'var_ctpop_v16i8' 317 ; SSE42-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %ctpop = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a) 318 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %ctpop 319 ; 320 ; AVX-LABEL: 'var_ctpop_v16i8' 321 ; AVX-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %ctpop = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a) 322 ; AVX-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %ctpop 323 ; 324 ; AVX512-LABEL: 'var_ctpop_v16i8' 325 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %ctpop = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a) 326 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %ctpop 327 ; 328 %ctpop = call <16 x i8> @llvm.ctpop.v16i8(<16 x i8> %a) 329 ret <16 x i8> %ctpop 330 } 331 332 define <32 x i8> @var_ctpop_v32i8(<32 x i8> %a) { 333 ; SSE2-LABEL: 'var_ctpop_v32i8' 334 ; SSE2-NEXT: Cost Model: Found an estimated cost of 20 for instruction: %ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a) 335 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %ctpop 336 ; 337 ; SSE42-LABEL: 'var_ctpop_v32i8' 338 ; SSE42-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a) 339 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %ctpop 340 ; 341 ; AVX1-LABEL: 'var_ctpop_v32i8' 342 ; AVX1-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a) 343 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %ctpop 344 ; 345 ; AVX2-LABEL: 'var_ctpop_v32i8' 346 ; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a) 347 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %ctpop 348 ; 349 ; AVX512-LABEL: 'var_ctpop_v32i8' 350 ; AVX512-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a) 351 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %ctpop 352 ; 353 %ctpop = call <32 x i8> @llvm.ctpop.v32i8(<32 x i8> %a) 354 ret <32 x i8> %ctpop 355 } 356 357 define <64 x i8> @var_ctpop_v64i8(<64 x i8> %a) { 358 ; SSE2-LABEL: 'var_ctpop_v64i8' 359 ; SSE2-NEXT: Cost Model: Found an estimated cost of 40 for instruction: %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 360 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %ctpop 361 ; 362 ; SSE42-LABEL: 'var_ctpop_v64i8' 363 ; SSE42-NEXT: Cost Model: Found an estimated cost of 24 for instruction: %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 364 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %ctpop 365 ; 366 ; AVX1-LABEL: 'var_ctpop_v64i8' 367 ; AVX1-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 368 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %ctpop 369 ; 370 ; AVX2-LABEL: 'var_ctpop_v64i8' 371 ; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 372 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %ctpop 373 ; 374 ; AVX512F-LABEL: 'var_ctpop_v64i8' 375 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 376 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %ctpop 377 ; 378 ; AVX512BW-LABEL: 'var_ctpop_v64i8' 379 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 380 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %ctpop 381 ; 382 %ctpop = call <64 x i8> @llvm.ctpop.v64i8(<64 x i8> %a) 383 ret <64 x i8> %ctpop 384 } 385
