422 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: %if x86-registered-target %{ opt < %s -passes=slp-vectorizer -S -mtriple=x86_64-unknown-linux-gnu | FileCheck %s %}3; RUN: %if aarch64-registered-target %{ opt < %s -passes=slp-vectorizer -S -mtriple=aarch64-unknown-linux-gnu | FileCheck %s %}4 5@b = common global [4 x i32] zeroinitializer, align 166@c = common global [4 x i32] zeroinitializer, align 167@d = common global [4 x i32] zeroinitializer, align 168@e = common global [4 x i32] zeroinitializer, align 169@a = common global [4 x i32] zeroinitializer, align 1610@fb = common global [4 x float] zeroinitializer, align 1611@fc = common global [4 x float] zeroinitializer, align 1612@fa = common global [4 x float] zeroinitializer, align 1613@fd = common global [4 x float] zeroinitializer, align 1614 15; Function Attrs: nounwind uwtable16define void @addsub() #0 {17; CHECK-LABEL: @addsub(18; CHECK-NEXT: entry:19; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr @b, align 420; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @c, align 421; CHECK-NEXT: [[TMP2:%.*]] = add nsw <4 x i32> [[TMP0]], [[TMP1]]22; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr @d, align 423; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr @e, align 424; CHECK-NEXT: [[TMP5:%.*]] = add nsw <4 x i32> [[TMP3]], [[TMP4]]25; CHECK-NEXT: [[TMP6:%.*]] = add nsw <4 x i32> [[TMP2]], [[TMP5]]26; CHECK-NEXT: [[TMP7:%.*]] = sub nsw <4 x i32> [[TMP2]], [[TMP5]]27; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> [[TMP7]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>28; CHECK-NEXT: store <4 x i32> [[TMP8]], ptr @a, align 429; CHECK-NEXT: ret void30;31entry:32 %0 = load i32, ptr @b, align 433 %1 = load i32, ptr @c, align 434 %add = add nsw i32 %0, %135 %2 = load i32, ptr @d, align 436 %3 = load i32, ptr @e, align 437 %add1 = add nsw i32 %2, %338 %add2 = add nsw i32 %add, %add139 store i32 %add2, ptr @a, align 440 %4 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 1), align 441 %5 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 1), align 442 %add3 = add nsw i32 %4, %543 %6 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 1), align 444 %7 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 1), align 445 %add4 = add nsw i32 %6, %746 %sub = sub nsw i32 %add3, %add447 store i32 %sub, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 1), align 448 %8 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 2), align 449 %9 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 2), align 450 %add5 = add nsw i32 %8, %951 %10 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 2), align 452 %11 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 2), align 453 %add6 = add nsw i32 %10, %1154 %add7 = add nsw i32 %add5, %add655 store i32 %add7, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 2), align 456 %12 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 3), align 457 %13 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 3), align 458 %add8 = add nsw i32 %12, %1359 %14 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 3), align 460 %15 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 3), align 461 %add9 = add nsw i32 %14, %1562 %sub10 = sub nsw i32 %add8, %add963 store i32 %sub10, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 3), align 464 ret void65}66 67define void @addsub_freeze() #0 {68; CHECK-LABEL: @addsub_freeze(69; CHECK-NEXT: entry:70; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr @b, align 471; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @c, align 472; CHECK-NEXT: [[TMP2:%.*]] = add nsw <4 x i32> [[TMP0]], [[TMP1]]73; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr @d, align 474; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr @e, align 475; CHECK-NEXT: [[TMP5:%.*]] = add nsw <4 x i32> [[TMP3]], [[TMP4]]76; CHECK-NEXT: [[TMP6:%.*]] = add nsw <4 x i32> [[TMP2]], [[TMP5]]77; CHECK-NEXT: [[TMP7:%.*]] = sub nsw <4 x i32> [[TMP2]], [[TMP5]]78; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> [[TMP7]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>79; CHECK-NEXT: [[TMP9:%.*]] = freeze <4 x i32> [[TMP8]]80; CHECK-NEXT: store <4 x i32> [[TMP9]], ptr @a, align 481; CHECK-NEXT: ret void82;83entry:84 %0 = load i32, ptr @b, align 485 %1 = load i32, ptr @c, align 486 %add = add nsw i32 %0, %187 %2 = load i32, ptr @d, align 488 %3 = load i32, ptr @e, align 489 %add1 = add nsw i32 %2, %390 %add2 = add nsw i32 %add, %add191 %freeze.add2 = freeze i32 %add292 store i32 %freeze.add2, ptr @a, align 493 %4 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 1), align 494 %5 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 1), align 495 %add3 = add nsw i32 %4, %596 %6 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 1), align 497 %7 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 1), align 498 %add4 = add nsw i32 %6, %799 %sub = sub nsw i32 %add3, %add4100 %freeze.sub = freeze i32 %sub101 store i32 %freeze.sub, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 1), align 4102 %8 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 2), align 4103 %9 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 2), align 4104 %add5 = add nsw i32 %8, %9105 %10 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 2), align 4106 %11 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 2), align 4107 %add6 = add nsw i32 %10, %11108 %add7 = add nsw i32 %add5, %add6109 %freeze.add7 = freeze i32 %add7110 store i32 %freeze.add7, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 2), align 4111 %12 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 3), align 4112 %13 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 3), align 4113 %add8 = add nsw i32 %12, %13114 %14 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 3), align 4115 %15 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 3), align 4116 %add9 = add nsw i32 %14, %15117 %sub10 = sub nsw i32 %add8, %add9118 %freeze.sub10 = freeze i32 %sub10119 store i32 %freeze.sub10, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 3), align 4120 ret void121}122 123; Function Attrs: nounwind uwtable124define void @subadd() #0 {125; CHECK-LABEL: @subadd(126; CHECK-NEXT: entry:127; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr @b, align 4128; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @c, align 4129; CHECK-NEXT: [[TMP2:%.*]] = add nsw <4 x i32> [[TMP0]], [[TMP1]]130; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr @d, align 4131; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr @e, align 4132; CHECK-NEXT: [[TMP5:%.*]] = add nsw <4 x i32> [[TMP3]], [[TMP4]]133; CHECK-NEXT: [[TMP6:%.*]] = sub nsw <4 x i32> [[TMP2]], [[TMP5]]134; CHECK-NEXT: [[TMP7:%.*]] = add nsw <4 x i32> [[TMP2]], [[TMP5]]135; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> [[TMP7]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>136; CHECK-NEXT: store <4 x i32> [[TMP8]], ptr @a, align 4137; CHECK-NEXT: ret void138;139entry:140 %0 = load i32, ptr @b, align 4141 %1 = load i32, ptr @c, align 4142 %add = add nsw i32 %0, %1143 %2 = load i32, ptr @d, align 4144 %3 = load i32, ptr @e, align 4145 %add1 = add nsw i32 %2, %3146 %sub = sub nsw i32 %add, %add1147 store i32 %sub, ptr @a, align 4148 %4 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 1), align 4149 %5 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 1), align 4150 %add2 = add nsw i32 %4, %5151 %6 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 1), align 4152 %7 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 1), align 4153 %add3 = add nsw i32 %6, %7154 %add4 = add nsw i32 %add2, %add3155 store i32 %add4, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 1), align 4156 %8 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 2), align 4157 %9 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 2), align 4158 %add5 = add nsw i32 %8, %9159 %10 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 2), align 4160 %11 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 2), align 4161 %add6 = add nsw i32 %10, %11162 %sub7 = sub nsw i32 %add5, %add6163 store i32 %sub7, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 2), align 4164 %12 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @b, i32 0, i64 3), align 4165 %13 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @c, i32 0, i64 3), align 4166 %add8 = add nsw i32 %12, %13167 %14 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @d, i32 0, i64 3), align 4168 %15 = load i32, ptr getelementptr inbounds ([4 x i32], ptr @e, i32 0, i64 3), align 4169 %add9 = add nsw i32 %14, %15170 %add10 = add nsw i32 %add8, %add9171 store i32 %add10, ptr getelementptr inbounds ([4 x i32], ptr @a, i32 0, i64 3), align 4172 ret void173}174 175; Function Attrs: nounwind uwtable176define void @faddfsub() #0 {177; CHECK-LABEL: @faddfsub(178; CHECK-NEXT: entry:179; CHECK-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr @fb, align 4180; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @fc, align 4181; CHECK-NEXT: [[TMP2:%.*]] = fadd <4 x float> [[TMP0]], [[TMP1]]182; CHECK-NEXT: [[TMP3:%.*]] = fsub <4 x float> [[TMP0]], [[TMP1]]183; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> [[TMP3]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>184; CHECK-NEXT: store <4 x float> [[TMP4]], ptr @fa, align 4185; CHECK-NEXT: ret void186;187entry:188 %0 = load float, ptr @fb, align 4189 %1 = load float, ptr @fc, align 4190 %add = fadd float %0, %1191 store float %add, ptr @fa, align 4192 %2 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 1), align 4193 %3 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 1), align 4194 %sub = fsub float %2, %3195 store float %sub, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 1), align 4196 %4 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 2), align 4197 %5 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 2), align 4198 %add1 = fadd float %4, %5199 store float %add1, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 2), align 4200 %6 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 3), align 4201 %7 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 3), align 4202 %sub2 = fsub float %6, %7203 store float %sub2, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 3), align 4204 ret void205}206 207; Function Attrs: nounwind uwtable208define void @fsubfadd() #0 {209; CHECK-LABEL: @fsubfadd(210; CHECK-NEXT: entry:211; CHECK-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr @fb, align 4212; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @fc, align 4213; CHECK-NEXT: [[TMP2:%.*]] = fsub <4 x float> [[TMP0]], [[TMP1]]214; CHECK-NEXT: [[TMP3:%.*]] = fadd <4 x float> [[TMP0]], [[TMP1]]215; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> [[TMP3]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>216; CHECK-NEXT: store <4 x float> [[TMP4]], ptr @fa, align 4217; CHECK-NEXT: ret void218;219entry:220 %0 = load float, ptr @fb, align 4221 %1 = load float, ptr @fc, align 4222 %sub = fsub float %0, %1223 store float %sub, ptr @fa, align 4224 %2 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 1), align 4225 %3 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 1), align 4226 %add = fadd float %2, %3227 store float %add, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 1), align 4228 %4 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 2), align 4229 %5 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 2), align 4230 %sub1 = fsub float %4, %5231 store float %sub1, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 2), align 4232 %6 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 3), align 4233 %7 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 3), align 4234 %add2 = fadd float %6, %7235 store float %add2, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 3), align 4236 ret void237}238 239; Function Attrs: nounwind uwtable240define void @faddfsub_select() #0 {241; CHECK-LABEL: @faddfsub_select(242; CHECK-NEXT: entry:243; CHECK-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr @fb, align 4244; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @fc, align 4245; CHECK-NEXT: [[TMP2:%.*]] = fadd <4 x float> [[TMP0]], [[TMP1]]246; CHECK-NEXT: [[TMP3:%.*]] = fsub <4 x float> [[TMP0]], [[TMP1]]247; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> [[TMP3]], <4 x i32> <i32 0, i32 1, i32 2, i32 7>248; CHECK-NEXT: store <4 x float> [[TMP4]], ptr @fa, align 4249; CHECK-NEXT: ret void250;251entry:252 %0 = load float, ptr @fb, align 4253 %1 = load float, ptr @fc, align 4254 %add = fadd float %0, %1255 store float %add, ptr @fa, align 4256 %2 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 1), align 4257 %3 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 1), align 4258 %add1 = fadd float %2, %3259 store float %add1, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 1), align 4260 %4 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 2), align 4261 %5 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 2), align 4262 %add2 = fadd float %4, %5263 store float %add2, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 2), align 4264 %6 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 3), align 4265 %7 = load float, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 3), align 4266 %sub = fsub float %6, %7267 store float %sub, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 3), align 4268 ret void269}270 271; Check vectorization of following code for float data type-272; fc[0] = fb[0]+fa[0]; //swapped fb and fa273; fc[1] = fa[1]-fb[1];274; fc[2] = fa[2]+fb[2];275; fc[3] = fa[3]-fb[3];276 277define void @reorder_alt() #0 {278; CHECK-LABEL: @reorder_alt(279; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @fa, align 4280; CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr @fb, align 4281; CHECK-NEXT: [[TMP3:%.*]] = fadd <4 x float> [[TMP1]], [[TMP2]]282; CHECK-NEXT: [[TMP4:%.*]] = fsub <4 x float> [[TMP1]], [[TMP2]]283; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <4 x float> [[TMP3]], <4 x float> [[TMP4]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>284; CHECK-NEXT: store <4 x float> [[TMP5]], ptr @fc, align 4285; CHECK-NEXT: ret void286;287 %1 = load float, ptr @fb, align 4288 %2 = load float, ptr @fa, align 4289 %3 = fadd float %1, %2290 store float %3, ptr @fc, align 4291 %4 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 1), align 4292 %5 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 1), align 4293 %6 = fsub float %4, %5294 store float %6, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 1), align 4295 %7 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 2), align 4296 %8 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 2), align 4297 %9 = fadd float %7, %8298 store float %9, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 2), align 4299 %10 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 3), align 4300 %11 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 3), align 4301 %12 = fsub float %10, %11302 store float %12, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 3), align 4303 ret void304}305 306; Check vectorization of following code for float data type-307; fc[0] = fa[0]+(fb[0]-fd[0]);308; fc[1] = fa[1]-(fb[1]+fd[1]);309; fc[2] = fa[2]+(fb[2]-fd[2]);310; fc[3] = fa[3]-(fd[3]+fb[3]); //swapped fd and fb311 312define void @reorder_alt_subTree() #0 {313; CHECK-LABEL: @reorder_alt_subTree(314; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @fa, align 4315; CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr @fd, align 4316; CHECK-NEXT: [[TMP3:%.*]] = load <4 x float>, ptr @fb, align 4317; CHECK-NEXT: [[TMP4:%.*]] = fsub <4 x float> [[TMP3]], [[TMP2]]318; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x float> [[TMP3]], [[TMP2]]319; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[TMP5]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>320; CHECK-NEXT: [[TMP7:%.*]] = fadd <4 x float> [[TMP1]], [[TMP6]]321; CHECK-NEXT: [[TMP8:%.*]] = fsub <4 x float> [[TMP1]], [[TMP6]]322; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <4 x float> [[TMP7]], <4 x float> [[TMP8]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>323; CHECK-NEXT: store <4 x float> [[TMP9]], ptr @fc, align 4324; CHECK-NEXT: ret void325;326 %1 = load float, ptr @fa, align 4327 %2 = load float, ptr @fb, align 4328 %3 = load float, ptr @fd, align 4329 %4 = fsub float %2, %3330 %5 = fadd float %1, %4331 store float %5, ptr @fc, align 4332 %6 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 1), align 4333 %7 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 1), align 4334 %8 = load float, ptr getelementptr inbounds ([4 x float], ptr @fd, i32 0, i64 1), align 4335 %9 = fadd float %7, %8336 %10 = fsub float %6, %9337 store float %10, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 1), align 4338 %11 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 2), align 4339 %12 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 2), align 4340 %13 = load float, ptr getelementptr inbounds ([4 x float], ptr @fd, i32 0, i64 2), align 4341 %14 = fsub float %12, %13342 %15 = fadd float %11, %14343 store float %15, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 2), align 4344 %16 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 3), align 4345 %17 = load float, ptr getelementptr inbounds ([4 x float], ptr @fd, i32 0, i64 3), align 4346 %18 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 3), align 4347 %19 = fadd float %17, %18348 %20 = fsub float %16, %19349 store float %20, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 3), align 4350 ret void351}352 353; Check vectorization of following code for double data type-354; c[0] = (a[0]+b[0])-d[0];355; c[1] = d[1]+(a[1]+b[1]); //swapped d[1] and (a[1]+b[1])356 357define void @reorder_alt_rightsubTree(ptr nocapture %c, ptr noalias nocapture readonly %a, ptr noalias nocapture readonly %b, ptr noalias nocapture readonly %d) {358; CHECK-LABEL: @reorder_alt_rightsubTree(359; CHECK-NEXT: [[TMP1:%.*]] = load <2 x double>, ptr [[D:%.*]], align 8360; CHECK-NEXT: [[TMP2:%.*]] = load <2 x double>, ptr [[A:%.*]], align 8361; CHECK-NEXT: [[TMP3:%.*]] = load <2 x double>, ptr [[B:%.*]], align 8362; CHECK-NEXT: [[TMP4:%.*]] = fadd <2 x double> [[TMP2]], [[TMP3]]363; CHECK-NEXT: [[TMP5:%.*]] = fsub <2 x double> [[TMP4]], [[TMP1]]364; CHECK-NEXT: [[TMP6:%.*]] = fadd <2 x double> [[TMP4]], [[TMP1]]365; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x double> [[TMP5]], <2 x double> [[TMP6]], <2 x i32> <i32 0, i32 3>366; CHECK-NEXT: store <2 x double> [[TMP7]], ptr [[C:%.*]], align 8367; CHECK-NEXT: ret void368;369 %1 = load double, ptr %a370 %2 = load double, ptr %b371 %3 = fadd double %1, %2372 %4 = load double, ptr %d373 %5 = fsub double %3, %4374 store double %5, ptr %c375 %6 = getelementptr inbounds double, ptr %d, i64 1376 %7 = load double, ptr %6377 %8 = getelementptr inbounds double, ptr %a, i64 1378 %9 = load double, ptr %8379 %10 = getelementptr inbounds double, ptr %b, i64 1380 %11 = load double, ptr %10381 %12 = fadd double %9, %11382 %13 = fadd double %7, %12383 %14 = getelementptr inbounds double, ptr %c, i64 1384 store double %13, ptr %14385 ret void386}387 388define void @vec_shuff_reorder() #0 {389; CHECK-LABEL: @vec_shuff_reorder(390; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr @fb, align 4391; CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr @fa, align 4392; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> [[TMP1]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>393; CHECK-NEXT: [[TMP8:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> [[TMP2]], <4 x i32> <i32 0, i32 1, i32 6, i32 7>394; CHECK-NEXT: [[TMP9:%.*]] = fadd <4 x float> [[TMP6]], [[TMP8]]395; CHECK-NEXT: [[TMP10:%.*]] = fsub <4 x float> [[TMP6]], [[TMP8]]396; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <4 x float> [[TMP9]], <4 x float> [[TMP10]], <4 x i32> <i32 0, i32 5, i32 2, i32 7>397; CHECK-NEXT: store <4 x float> [[TMP11]], ptr @fc, align 4398; CHECK-NEXT: ret void399;400 %1 = load float, ptr @fb, align 4401 %2 = load float, ptr @fa, align 4402 %3 = fadd float %1, %2403 store float %3, ptr @fc, align 4404 %4 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 1), align 4405 %5 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 1), align 4406 %6 = fsub float %4, %5407 store float %6, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 1), align 4408 %7 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 2), align 4409 %8 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 2), align 4410 %9 = fadd float %7, %8411 store float %9, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 2), align 4412 %10 = load float, ptr getelementptr inbounds ([4 x float], ptr @fb, i32 0, i64 3), align 4413 %11 = load float, ptr getelementptr inbounds ([4 x float], ptr @fa, i32 0, i64 3), align 4414 %12 = fsub float %10, %11415 store float %12, ptr getelementptr inbounds ([4 x float], ptr @fc, i32 0, i64 3), align 4416 ret void417}418 419 420attributes #0 = { nounwind }421 422