1 ; RUN: opt < %s -sroa -S | FileCheck %s 2 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-n8:16:32:64" 3 4 %S1 = type { i64, [42 x float] } 5 6 define i32 @test1(<4 x i32> %x, <4 x i32> %y) { 7 ; CHECK-LABEL: @test1( 8 entry: 9 %a = alloca [2 x <4 x i32>] 10 ; CHECK-NOT: alloca 11 12 %a.x = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0 13 store <4 x i32> %x, <4 x i32>* %a.x 14 %a.y = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1 15 store <4 x i32> %y, <4 x i32>* %a.y 16 ; CHECK-NOT: store 17 18 %a.tmp1 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0, i64 2 19 %tmp1 = load i32* %a.tmp1 20 %a.tmp2 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 3 21 %tmp2 = load i32* %a.tmp2 22 %a.tmp3 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 0 23 %tmp3 = load i32* %a.tmp3 24 ; CHECK-NOT: load 25 ; CHECK: extractelement <4 x i32> %x, i32 2 26 ; CHECK-NEXT: extractelement <4 x i32> %y, i32 3 27 ; CHECK-NEXT: extractelement <4 x i32> %y, i32 0 28 29 %tmp4 = add i32 %tmp1, %tmp2 30 %tmp5 = add i32 %tmp3, %tmp4 31 ret i32 %tmp5 32 ; CHECK-NEXT: add 33 ; CHECK-NEXT: add 34 ; CHECK-NEXT: ret 35 } 36 37 define i32 @test2(<4 x i32> %x, <4 x i32> %y) { 38 ; CHECK-LABEL: @test2( 39 entry: 40 %a = alloca [2 x <4 x i32>] 41 ; CHECK-NOT: alloca 42 43 %a.x = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0 44 store <4 x i32> %x, <4 x i32>* %a.x 45 %a.y = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1 46 store <4 x i32> %y, <4 x i32>* %a.y 47 ; CHECK-NOT: store 48 49 %a.tmp1 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0, i64 2 50 %tmp1 = load i32* %a.tmp1 51 %a.tmp2 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 3 52 %tmp2 = load i32* %a.tmp2 53 %a.tmp3 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 0 54 %a.tmp3.cast = bitcast i32* %a.tmp3 to <2 x i32>* 55 %tmp3.vec = load <2 x i32>* %a.tmp3.cast 56 %tmp3 = extractelement <2 x i32> %tmp3.vec, i32 0 57 ; CHECK-NOT: load 58 ; CHECK: %[[extract1:.*]] = extractelement <4 x i32> %x, i32 2 59 ; CHECK-NEXT: %[[extract2:.*]] = extractelement <4 x i32> %y, i32 3 60 ; CHECK-NEXT: %[[extract3:.*]] = shufflevector <4 x i32> %y, <4 x i32> undef, <2 x i32> <i32 0, i32 1> 61 ; CHECK-NEXT: %[[extract4:.*]] = extractelement <2 x i32> %[[extract3]], i32 0 62 63 %tmp4 = add i32 %tmp1, %tmp2 64 %tmp5 = add i32 %tmp3, %tmp4 65 ret i32 %tmp5 66 ; CHECK-NEXT: %[[sum1:.*]] = add i32 %[[extract1]], %[[extract2]] 67 ; CHECK-NEXT: %[[sum2:.*]] = add i32 %[[extract4]], %[[sum1]] 68 ; CHECK-NEXT: ret i32 %[[sum2]] 69 } 70 71 define i32 @test3(<4 x i32> %x, <4 x i32> %y) { 72 ; CHECK-LABEL: @test3( 73 entry: 74 %a = alloca [2 x <4 x i32>] 75 ; CHECK-NOT: alloca 76 77 %a.x = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0 78 store <4 x i32> %x, <4 x i32>* %a.x 79 %a.y = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1 80 store <4 x i32> %y, <4 x i32>* %a.y 81 ; CHECK-NOT: store 82 83 %a.y.cast = bitcast <4 x i32>* %a.y to i8* 84 call void @llvm.memset.p0i8.i32(i8* %a.y.cast, i8 0, i32 16, i32 1, i1 false) 85 ; CHECK-NOT: memset 86 87 %a.tmp1 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0, i64 2 88 %a.tmp1.cast = bitcast i32* %a.tmp1 to i8* 89 call void @llvm.memset.p0i8.i32(i8* %a.tmp1.cast, i8 -1, i32 4, i32 1, i1 false) 90 %tmp1 = load i32* %a.tmp1 91 %a.tmp2 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 3 92 %tmp2 = load i32* %a.tmp2 93 %a.tmp3 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 0 94 %tmp3 = load i32* %a.tmp3 95 ; CHECK-NOT: load 96 ; CHECK: %[[insert:.*]] = insertelement <4 x i32> %x, i32 -1, i32 2 97 ; CHECK-NEXT: extractelement <4 x i32> %[[insert]], i32 2 98 ; CHECK-NEXT: extractelement <4 x i32> zeroinitializer, i32 3 99 ; CHECK-NEXT: extractelement <4 x i32> zeroinitializer, i32 0 100 101 %tmp4 = add i32 %tmp1, %tmp2 102 %tmp5 = add i32 %tmp3, %tmp4 103 ret i32 %tmp5 104 ; CHECK-NEXT: add 105 ; CHECK-NEXT: add 106 ; CHECK-NEXT: ret 107 } 108 109 define i32 @test4(<4 x i32> %x, <4 x i32> %y, <4 x i32>* %z) { 110 ; CHECK-LABEL: @test4( 111 entry: 112 %a = alloca [2 x <4 x i32>] 113 ; CHECK-NOT: alloca 114 115 %a.x = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0 116 store <4 x i32> %x, <4 x i32>* %a.x 117 %a.y = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1 118 store <4 x i32> %y, <4 x i32>* %a.y 119 ; CHECK-NOT: store 120 121 %a.y.cast = bitcast <4 x i32>* %a.y to i8* 122 %z.cast = bitcast <4 x i32>* %z to i8* 123 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.y.cast, i8* %z.cast, i32 16, i32 1, i1 false) 124 ; CHECK-NOT: memcpy 125 126 %a.tmp1 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0, i64 2 127 %a.tmp1.cast = bitcast i32* %a.tmp1 to i8* 128 %z.tmp1 = getelementptr inbounds <4 x i32>* %z, i64 0, i64 2 129 %z.tmp1.cast = bitcast i32* %z.tmp1 to i8* 130 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.tmp1.cast, i8* %z.tmp1.cast, i32 4, i32 1, i1 false) 131 %tmp1 = load i32* %a.tmp1 132 %a.tmp2 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 3 133 %tmp2 = load i32* %a.tmp2 134 %a.tmp3 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 0 135 %tmp3 = load i32* %a.tmp3 136 ; CHECK-NOT: memcpy 137 ; CHECK: %[[load:.*]] = load <4 x i32>* %z 138 ; CHECK-NEXT: %[[gep:.*]] = getelementptr inbounds <4 x i32>* %z, i64 0, i64 2 139 ; CHECK-NEXT: %[[element_load:.*]] = load i32* %[[gep]] 140 ; CHECK-NEXT: %[[insert:.*]] = insertelement <4 x i32> %x, i32 %[[element_load]], i32 2 141 ; CHECK-NEXT: extractelement <4 x i32> %[[insert]], i32 2 142 ; CHECK-NEXT: extractelement <4 x i32> %[[load]], i32 3 143 ; CHECK-NEXT: extractelement <4 x i32> %[[load]], i32 0 144 145 %tmp4 = add i32 %tmp1, %tmp2 146 %tmp5 = add i32 %tmp3, %tmp4 147 ret i32 %tmp5 148 ; CHECK-NEXT: add 149 ; CHECK-NEXT: add 150 ; CHECK-NEXT: ret 151 } 152 153 define i32 @test5(<4 x i32> %x, <4 x i32> %y, <4 x i32>* %z) { 154 ; CHECK-LABEL: @test5( 155 ; The same as the above, but with reversed source and destination for the 156 ; element memcpy, and a self copy. 157 entry: 158 %a = alloca [2 x <4 x i32>] 159 ; CHECK-NOT: alloca 160 161 %a.x = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0 162 store <4 x i32> %x, <4 x i32>* %a.x 163 %a.y = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1 164 store <4 x i32> %y, <4 x i32>* %a.y 165 ; CHECK-NOT: store 166 167 %a.y.cast = bitcast <4 x i32>* %a.y to i8* 168 %a.x.cast = bitcast <4 x i32>* %a.x to i8* 169 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.x.cast, i8* %a.y.cast, i32 16, i32 1, i1 false) 170 ; CHECK-NOT: memcpy 171 172 %a.tmp1 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 0, i64 2 173 %a.tmp1.cast = bitcast i32* %a.tmp1 to i8* 174 %z.tmp1 = getelementptr inbounds <4 x i32>* %z, i64 0, i64 2 175 %z.tmp1.cast = bitcast i32* %z.tmp1 to i8* 176 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %z.tmp1.cast, i8* %a.tmp1.cast, i32 4, i32 1, i1 false) 177 %tmp1 = load i32* %a.tmp1 178 %a.tmp2 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 3 179 %tmp2 = load i32* %a.tmp2 180 %a.tmp3 = getelementptr inbounds [2 x <4 x i32>]* %a, i64 0, i64 1, i64 0 181 %tmp3 = load i32* %a.tmp3 182 ; CHECK-NOT: memcpy 183 ; CHECK: %[[gep:.*]] = getelementptr inbounds <4 x i32>* %z, i64 0, i64 2 184 ; CHECK-NEXT: %[[extract:.*]] = extractelement <4 x i32> %y, i32 2 185 ; CHECK-NEXT: store i32 %[[extract]], i32* %[[gep]] 186 ; CHECK-NEXT: extractelement <4 x i32> %y, i32 2 187 ; CHECK-NEXT: extractelement <4 x i32> %y, i32 3 188 ; CHECK-NEXT: extractelement <4 x i32> %y, i32 0 189 190 %tmp4 = add i32 %tmp1, %tmp2 191 %tmp5 = add i32 %tmp3, %tmp4 192 ret i32 %tmp5 193 ; CHECK-NEXT: add 194 ; CHECK-NEXT: add 195 ; CHECK-NEXT: ret 196 } 197 198 declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind 199 declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i32, i1) nounwind 200 201 define i64 @test6(<4 x i64> %x, <4 x i64> %y, i64 %n) { 202 ; CHECK-LABEL: @test6( 203 ; The old scalarrepl pass would wrongly drop the store to the second alloca. 204 ; PR13254 205 %tmp = alloca { <4 x i64>, <4 x i64> } 206 %p0 = getelementptr inbounds { <4 x i64>, <4 x i64> }* %tmp, i32 0, i32 0 207 store <4 x i64> %x, <4 x i64>* %p0 208 ; CHECK: store <4 x i64> %x, 209 %p1 = getelementptr inbounds { <4 x i64>, <4 x i64> }* %tmp, i32 0, i32 1 210 store <4 x i64> %y, <4 x i64>* %p1 211 ; CHECK: store <4 x i64> %y, 212 %addr = getelementptr inbounds { <4 x i64>, <4 x i64> }* %tmp, i32 0, i32 0, i64 %n 213 %res = load i64* %addr, align 4 214 ret i64 %res 215 } 216 217 define <4 x i32> @test_subvec_store() { 218 ; CHECK-LABEL: @test_subvec_store( 219 entry: 220 %a = alloca <4 x i32> 221 ; CHECK-NOT: alloca 222 223 %a.gep0 = getelementptr <4 x i32>* %a, i32 0, i32 0 224 %a.cast0 = bitcast i32* %a.gep0 to <2 x i32>* 225 store <2 x i32> <i32 0, i32 0>, <2 x i32>* %a.cast0 226 ; CHECK-NOT: store 227 ; CHECK: select <4 x i1> <i1 true, i1 true, i1 false, i1 false> 228 229 %a.gep1 = getelementptr <4 x i32>* %a, i32 0, i32 1 230 %a.cast1 = bitcast i32* %a.gep1 to <2 x i32>* 231 store <2 x i32> <i32 1, i32 1>, <2 x i32>* %a.cast1 232 ; CHECK-NEXT: select <4 x i1> <i1 false, i1 true, i1 true, i1 false> 233 234 %a.gep2 = getelementptr <4 x i32>* %a, i32 0, i32 2 235 %a.cast2 = bitcast i32* %a.gep2 to <2 x i32>* 236 store <2 x i32> <i32 2, i32 2>, <2 x i32>* %a.cast2 237 ; CHECK-NEXT: select <4 x i1> <i1 false, i1 false, i1 true, i1 true> 238 239 %a.gep3 = getelementptr <4 x i32>* %a, i32 0, i32 3 240 store i32 3, i32* %a.gep3 241 ; CHECK-NEXT: insertelement <4 x i32> 242 243 %ret = load <4 x i32>* %a 244 245 ret <4 x i32> %ret 246 ; CHECK-NEXT: ret <4 x i32> 247 } 248 249 define <4 x i32> @test_subvec_load() { 250 ; CHECK-LABEL: @test_subvec_load( 251 entry: 252 %a = alloca <4 x i32> 253 ; CHECK-NOT: alloca 254 store <4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32>* %a 255 ; CHECK-NOT: store 256 257 %a.gep0 = getelementptr <4 x i32>* %a, i32 0, i32 0 258 %a.cast0 = bitcast i32* %a.gep0 to <2 x i32>* 259 %first = load <2 x i32>* %a.cast0 260 ; CHECK-NOT: load 261 ; CHECK: %[[extract1:.*]] = shufflevector <4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> undef, <2 x i32> <i32 0, i32 1> 262 263 %a.gep1 = getelementptr <4 x i32>* %a, i32 0, i32 1 264 %a.cast1 = bitcast i32* %a.gep1 to <2 x i32>* 265 %second = load <2 x i32>* %a.cast1 266 ; CHECK-NEXT: %[[extract2:.*]] = shufflevector <4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> undef, <2 x i32> <i32 1, i32 2> 267 268 %a.gep2 = getelementptr <4 x i32>* %a, i32 0, i32 2 269 %a.cast2 = bitcast i32* %a.gep2 to <2 x i32>* 270 %third = load <2 x i32>* %a.cast2 271 ; CHECK-NEXT: %[[extract3:.*]] = shufflevector <4 x i32> <i32 0, i32 1, i32 2, i32 3>, <4 x i32> undef, <2 x i32> <i32 2, i32 3> 272 273 %tmp = shufflevector <2 x i32> %first, <2 x i32> %second, <2 x i32> <i32 0, i32 2> 274 %ret = shufflevector <2 x i32> %tmp, <2 x i32> %third, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 275 ; CHECK-NEXT: %[[tmp:.*]] = shufflevector <2 x i32> %[[extract1]], <2 x i32> %[[extract2]], <2 x i32> <i32 0, i32 2> 276 ; CHECK-NEXT: %[[ret:.*]] = shufflevector <2 x i32> %[[tmp]], <2 x i32> %[[extract3]], <4 x i32> <i32 0, i32 1, i32 2, i32 3> 277 278 ret <4 x i32> %ret 279 ; CHECK-NEXT: ret <4 x i32> %[[ret]] 280 } 281 282 declare void @llvm.memset.p0i32.i32(i32* nocapture, i32, i32, i32, i1) nounwind 283 284 define <4 x float> @test_subvec_memset() { 285 ; CHECK-LABEL: @test_subvec_memset( 286 entry: 287 %a = alloca <4 x float> 288 ; CHECK-NOT: alloca 289 290 %a.gep0 = getelementptr <4 x float>* %a, i32 0, i32 0 291 %a.cast0 = bitcast float* %a.gep0 to i8* 292 call void @llvm.memset.p0i8.i32(i8* %a.cast0, i8 0, i32 8, i32 0, i1 false) 293 ; CHECK-NOT: store 294 ; CHECK: select <4 x i1> <i1 true, i1 true, i1 false, i1 false> 295 296 %a.gep1 = getelementptr <4 x float>* %a, i32 0, i32 1 297 %a.cast1 = bitcast float* %a.gep1 to i8* 298 call void @llvm.memset.p0i8.i32(i8* %a.cast1, i8 1, i32 8, i32 0, i1 false) 299 ; CHECK-NEXT: select <4 x i1> <i1 false, i1 true, i1 true, i1 false> 300 301 %a.gep2 = getelementptr <4 x float>* %a, i32 0, i32 2 302 %a.cast2 = bitcast float* %a.gep2 to i8* 303 call void @llvm.memset.p0i8.i32(i8* %a.cast2, i8 3, i32 8, i32 0, i1 false) 304 ; CHECK-NEXT: select <4 x i1> <i1 false, i1 false, i1 true, i1 true> 305 306 %a.gep3 = getelementptr <4 x float>* %a, i32 0, i32 3 307 %a.cast3 = bitcast float* %a.gep3 to i8* 308 call void @llvm.memset.p0i8.i32(i8* %a.cast3, i8 7, i32 4, i32 0, i1 false) 309 ; CHECK-NEXT: insertelement <4 x float> 310 311 %ret = load <4 x float>* %a 312 313 ret <4 x float> %ret 314 ; CHECK-NEXT: ret <4 x float> 315 } 316 317 define <4 x float> @test_subvec_memcpy(i8* %x, i8* %y, i8* %z, i8* %f, i8* %out) { 318 ; CHECK-LABEL: @test_subvec_memcpy( 319 entry: 320 %a = alloca <4 x float> 321 ; CHECK-NOT: alloca 322 323 %a.gep0 = getelementptr <4 x float>* %a, i32 0, i32 0 324 %a.cast0 = bitcast float* %a.gep0 to i8* 325 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.cast0, i8* %x, i32 8, i32 0, i1 false) 326 ; CHECK: %[[xptr:.*]] = bitcast i8* %x to <2 x float>* 327 ; CHECK-NEXT: %[[x:.*]] = load <2 x float>* %[[xptr]] 328 ; CHECK-NEXT: %[[expand_x:.*]] = shufflevector <2 x float> %[[x]], <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 329 ; CHECK-NEXT: select <4 x i1> <i1 true, i1 true, i1 false, i1 false> 330 331 %a.gep1 = getelementptr <4 x float>* %a, i32 0, i32 1 332 %a.cast1 = bitcast float* %a.gep1 to i8* 333 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.cast1, i8* %y, i32 8, i32 0, i1 false) 334 ; CHECK-NEXT: %[[yptr:.*]] = bitcast i8* %y to <2 x float>* 335 ; CHECK-NEXT: %[[y:.*]] = load <2 x float>* %[[yptr]] 336 ; CHECK-NEXT: %[[expand_y:.*]] = shufflevector <2 x float> %[[y]], <2 x float> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 undef> 337 ; CHECK-NEXT: select <4 x i1> <i1 false, i1 true, i1 true, i1 false> 338 339 %a.gep2 = getelementptr <4 x float>* %a, i32 0, i32 2 340 %a.cast2 = bitcast float* %a.gep2 to i8* 341 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.cast2, i8* %z, i32 8, i32 0, i1 false) 342 ; CHECK-NEXT: %[[zptr:.*]] = bitcast i8* %z to <2 x float>* 343 ; CHECK-NEXT: %[[z:.*]] = load <2 x float>* %[[zptr]] 344 ; CHECK-NEXT: %[[expand_z:.*]] = shufflevector <2 x float> %[[z]], <2 x float> undef, <4 x i32> <i32 undef, i32 undef, i32 0, i32 1> 345 ; CHECK-NEXT: select <4 x i1> <i1 false, i1 false, i1 true, i1 true> 346 347 %a.gep3 = getelementptr <4 x float>* %a, i32 0, i32 3 348 %a.cast3 = bitcast float* %a.gep3 to i8* 349 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %a.cast3, i8* %f, i32 4, i32 0, i1 false) 350 ; CHECK-NEXT: %[[fptr:.*]] = bitcast i8* %f to float* 351 ; CHECK-NEXT: %[[f:.*]] = load float* %[[fptr]] 352 ; CHECK-NEXT: %[[insert_f:.*]] = insertelement <4 x float> 353 354 call void @llvm.memcpy.p0i8.p0i8.i32(i8* %out, i8* %a.cast2, i32 8, i32 0, i1 false) 355 ; CHECK-NEXT: %[[outptr:.*]] = bitcast i8* %out to <2 x float>* 356 ; CHECK-NEXT: %[[extract_out:.*]] = shufflevector <4 x float> %[[insert_f]], <4 x float> undef, <2 x i32> <i32 2, i32 3> 357 ; CHECK-NEXT: store <2 x float> %[[extract_out]], <2 x float>* %[[outptr]] 358 359 %ret = load <4 x float>* %a 360 361 ret <4 x float> %ret 362 ; CHECK-NEXT: ret <4 x float> %[[insert_f]] 363 } 364 365 define i32 @PR14212() { 366 ; CHECK-LABEL: @PR14212( 367 ; This caused a crash when "splitting" the load of the i32 in order to promote 368 ; the store of <3 x i8> properly. Heavily reduced from an OpenCL test case. 369 entry: 370 %retval = alloca <3 x i8>, align 4 371 ; CHECK-NOT: alloca 372 373 store <3 x i8> undef, <3 x i8>* %retval, align 4 374 %cast = bitcast <3 x i8>* %retval to i32* 375 %load = load i32* %cast, align 4 376 ret i32 %load 377 ; CHECK: ret i32 378 } 379 380 define <2 x i8> @PR14349.1(i32 %x) { 381 ; CHECK: @PR14349.1 382 ; The first testcase for broken SROA rewriting of split integer loads and 383 ; stores due to smaller vector loads and stores. This particular test ensures 384 ; that we can rewrite a split store of an integer to a store of a vector. 385 entry: 386 %a = alloca i32 387 ; CHECK-NOT: alloca 388 389 store i32 %x, i32* %a 390 ; CHECK-NOT: store 391 392 %cast = bitcast i32* %a to <2 x i8>* 393 %vec = load <2 x i8>* %cast 394 ; CHECK-NOT: load 395 396 ret <2 x i8> %vec 397 ; CHECK: %[[trunc:.*]] = trunc i32 %x to i16 398 ; CHECK: %[[cast:.*]] = bitcast i16 %[[trunc]] to <2 x i8> 399 ; CHECK: ret <2 x i8> %[[cast]] 400 } 401 402 define i32 @PR14349.2(<2 x i8> %x) { 403 ; CHECK: @PR14349.2 404 ; The first testcase for broken SROA rewriting of split integer loads and 405 ; stores due to smaller vector loads and stores. This particular test ensures 406 ; that we can rewrite a split load of an integer to a load of a vector. 407 entry: 408 %a = alloca i32 409 ; CHECK-NOT: alloca 410 411 %cast = bitcast i32* %a to <2 x i8>* 412 store <2 x i8> %x, <2 x i8>* %cast 413 ; CHECK-NOT: store 414 415 %int = load i32* %a 416 ; CHECK-NOT: load 417 418 ret i32 %int 419 ; CHECK: %[[cast:.*]] = bitcast <2 x i8> %x to i16 420 ; CHECK: %[[trunc:.*]] = zext i16 %[[cast]] to i32 421 ; CHECK: %[[insert:.*]] = or i32 %{{.*}}, %[[trunc]] 422 ; CHECK: ret i32 %[[insert]] 423 } 424
