1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -slp-vectorizer -slp-threshold=-10000 < %s | FileCheck %s --check-prefixes=CHECK,THRESHOLD 3; RUN: opt -S -slp-vectorizer -slp-threshold=0 < %s | FileCheck %s --check-prefixes=CHECK,NOTHRESHOLD 4; RUN: opt -S -slp-vectorizer -slp-threshold=-10000 -slp-min-tree-size=0 < %s | FileCheck %s --check-prefixes=CHECK,MINTREESIZE 5 6target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-n8:16:32:64-S128" 7target triple = "x86_64-apple-macosx10.8.0" 8 9define <4 x float> @simple_select(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 10; CHECK-LABEL: @simple_select( 11; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer 12; CHECK-NEXT: [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]] 13; CHECK-NEXT: ret <4 x float> [[TMP2]] 14; 15 %c0 = extractelement <4 x i32> %c, i32 0 16 %c1 = extractelement <4 x i32> %c, i32 1 17 %c2 = extractelement <4 x i32> %c, i32 2 18 %c3 = extractelement <4 x i32> %c, i32 3 19 %a0 = extractelement <4 x float> %a, i32 0 20 %a1 = extractelement <4 x float> %a, i32 1 21 %a2 = extractelement <4 x float> %a, i32 2 22 %a3 = extractelement <4 x float> %a, i32 3 23 %b0 = extractelement <4 x float> %b, i32 0 24 %b1 = extractelement <4 x float> %b, i32 1 25 %b2 = extractelement <4 x float> %b, i32 2 26 %b3 = extractelement <4 x float> %b, i32 3 27 %cmp0 = icmp ne i32 %c0, 0 28 %cmp1 = icmp ne i32 %c1, 0 29 %cmp2 = icmp ne i32 %c2, 0 30 %cmp3 = icmp ne i32 %c3, 0 31 %s0 = select i1 %cmp0, float %a0, float %b0 32 %s1 = select i1 %cmp1, float %a1, float %b1 33 %s2 = select i1 %cmp2, float %a2, float %b2 34 %s3 = select i1 %cmp3, float %a3, float %b3 35 %ra = insertelement <4 x float> undef, float %s0, i32 0 36 %rb = insertelement <4 x float> %ra, float %s1, i32 1 37 %rc = insertelement <4 x float> %rb, float %s2, i32 2 38 %rd = insertelement <4 x float> %rc, float %s3, i32 3 39 ret <4 x float> %rd 40} 41 42define <8 x float> @simple_select2(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 43; CHECK-LABEL: @simple_select2( 44; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer 45; CHECK-NEXT: [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]] 46; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <8 x i32> <i32 0, i32 undef, i32 1, i32 undef, i32 2, i32 undef, i32 undef, i32 3> 47; CHECK-NEXT: [[RD1:%.*]] = shufflevector <8 x float> undef, <8 x float> [[TMP3]], <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 6, i32 15> 48; CHECK-NEXT: ret <8 x float> [[RD1]] 49; 50 %c0 = extractelement <4 x i32> %c, i32 0 51 %c1 = extractelement <4 x i32> %c, i32 1 52 %c2 = extractelement <4 x i32> %c, i32 2 53 %c3 = extractelement <4 x i32> %c, i32 3 54 %a0 = extractelement <4 x float> %a, i32 0 55 %a1 = extractelement <4 x float> %a, i32 1 56 %a2 = extractelement <4 x float> %a, i32 2 57 %a3 = extractelement <4 x float> %a, i32 3 58 %b0 = extractelement <4 x float> %b, i32 0 59 %b1 = extractelement <4 x float> %b, i32 1 60 %b2 = extractelement <4 x float> %b, i32 2 61 %b3 = extractelement <4 x float> %b, i32 3 62 %cmp0 = icmp ne i32 %c0, 0 63 %cmp1 = icmp ne i32 %c1, 0 64 %cmp2 = icmp ne i32 %c2, 0 65 %cmp3 = icmp ne i32 %c3, 0 66 %s0 = select i1 %cmp0, float %a0, float %b0 67 %s1 = select i1 %cmp1, float %a1, float %b1 68 %s2 = select i1 %cmp2, float %a2, float %b2 69 %s3 = select i1 %cmp3, float %a3, float %b3 70 %ra = insertelement <8 x float> undef, float %s0, i32 0 71 %rb = insertelement <8 x float> %ra, float %s1, i32 2 72 %rc = insertelement <8 x float> %rb, float %s2, i32 4 73 %rd = insertelement <8 x float> %rc, float %s3, i32 7 74 ret <8 x float> %rd 75} 76 77declare void @llvm.assume(i1) nounwind 78 79; This entire tree is ephemeral, don't vectorize any of it. 80define <4 x float> @simple_select_eph(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 81; THRESHOLD-LABEL: @simple_select_eph( 82; THRESHOLD-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0 83; THRESHOLD-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1 84; THRESHOLD-NEXT: [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2 85; THRESHOLD-NEXT: [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3 86; THRESHOLD-NEXT: [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0 87; THRESHOLD-NEXT: [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1 88; THRESHOLD-NEXT: [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2 89; THRESHOLD-NEXT: [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3 90; THRESHOLD-NEXT: [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0 91; THRESHOLD-NEXT: [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1 92; THRESHOLD-NEXT: [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2 93; THRESHOLD-NEXT: [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3 94; THRESHOLD-NEXT: [[CMP0:%.*]] = icmp ne i32 [[C0]], 0 95; THRESHOLD-NEXT: [[CMP1:%.*]] = icmp ne i32 [[C1]], 0 96; THRESHOLD-NEXT: [[CMP2:%.*]] = icmp ne i32 [[C2]], 0 97; THRESHOLD-NEXT: [[CMP3:%.*]] = icmp ne i32 [[C3]], 0 98; THRESHOLD-NEXT: [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]] 99; THRESHOLD-NEXT: [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]] 100; THRESHOLD-NEXT: [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]] 101; THRESHOLD-NEXT: [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]] 102; THRESHOLD-NEXT: [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0 103; THRESHOLD-NEXT: [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1 104; THRESHOLD-NEXT: [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2 105; THRESHOLD-NEXT: [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3 106; THRESHOLD-NEXT: [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0 107; THRESHOLD-NEXT: [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1 108; THRESHOLD-NEXT: [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2 109; THRESHOLD-NEXT: [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3 110; THRESHOLD-NEXT: [[Q4:%.*]] = fadd float [[Q0]], [[Q1]] 111; THRESHOLD-NEXT: [[Q5:%.*]] = fadd float [[Q2]], [[Q3]] 112; THRESHOLD-NEXT: [[Q6:%.*]] = fadd float [[Q4]], [[Q5]] 113; THRESHOLD-NEXT: [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]] 114; THRESHOLD-NEXT: call void @llvm.assume(i1 [[QI]]) 115; THRESHOLD-NEXT: ret <4 x float> undef 116; 117; NOTHRESHOLD-LABEL: @simple_select_eph( 118; NOTHRESHOLD-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0 119; NOTHRESHOLD-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1 120; NOTHRESHOLD-NEXT: [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2 121; NOTHRESHOLD-NEXT: [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3 122; NOTHRESHOLD-NEXT: [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0 123; NOTHRESHOLD-NEXT: [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1 124; NOTHRESHOLD-NEXT: [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2 125; NOTHRESHOLD-NEXT: [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3 126; NOTHRESHOLD-NEXT: [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0 127; NOTHRESHOLD-NEXT: [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1 128; NOTHRESHOLD-NEXT: [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2 129; NOTHRESHOLD-NEXT: [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3 130; NOTHRESHOLD-NEXT: [[CMP0:%.*]] = icmp ne i32 [[C0]], 0 131; NOTHRESHOLD-NEXT: [[CMP1:%.*]] = icmp ne i32 [[C1]], 0 132; NOTHRESHOLD-NEXT: [[CMP2:%.*]] = icmp ne i32 [[C2]], 0 133; NOTHRESHOLD-NEXT: [[CMP3:%.*]] = icmp ne i32 [[C3]], 0 134; NOTHRESHOLD-NEXT: [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]] 135; NOTHRESHOLD-NEXT: [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]] 136; NOTHRESHOLD-NEXT: [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]] 137; NOTHRESHOLD-NEXT: [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]] 138; NOTHRESHOLD-NEXT: [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0 139; NOTHRESHOLD-NEXT: [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1 140; NOTHRESHOLD-NEXT: [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2 141; NOTHRESHOLD-NEXT: [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3 142; NOTHRESHOLD-NEXT: [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0 143; NOTHRESHOLD-NEXT: [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1 144; NOTHRESHOLD-NEXT: [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2 145; NOTHRESHOLD-NEXT: [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3 146; NOTHRESHOLD-NEXT: [[Q4:%.*]] = fadd float [[Q0]], [[Q1]] 147; NOTHRESHOLD-NEXT: [[Q5:%.*]] = fadd float [[Q2]], [[Q3]] 148; NOTHRESHOLD-NEXT: [[Q6:%.*]] = fadd float [[Q4]], [[Q5]] 149; NOTHRESHOLD-NEXT: [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]] 150; NOTHRESHOLD-NEXT: call void @llvm.assume(i1 [[QI]]) 151; NOTHRESHOLD-NEXT: ret <4 x float> undef 152; 153; MINTREESIZE-LABEL: @simple_select_eph( 154; MINTREESIZE-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0 155; MINTREESIZE-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1 156; MINTREESIZE-NEXT: [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2 157; MINTREESIZE-NEXT: [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3 158; MINTREESIZE-NEXT: [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0 159; MINTREESIZE-NEXT: [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1 160; MINTREESIZE-NEXT: [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2 161; MINTREESIZE-NEXT: [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3 162; MINTREESIZE-NEXT: [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0 163; MINTREESIZE-NEXT: [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1 164; MINTREESIZE-NEXT: [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2 165; MINTREESIZE-NEXT: [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3 166; MINTREESIZE-NEXT: [[CMP0:%.*]] = icmp ne i32 [[C0]], 0 167; MINTREESIZE-NEXT: [[CMP1:%.*]] = icmp ne i32 [[C1]], 0 168; MINTREESIZE-NEXT: [[CMP2:%.*]] = icmp ne i32 [[C2]], 0 169; MINTREESIZE-NEXT: [[CMP3:%.*]] = icmp ne i32 [[C3]], 0 170; MINTREESIZE-NEXT: [[S0:%.*]] = select i1 [[CMP0]], float [[A0]], float [[B0]] 171; MINTREESIZE-NEXT: [[S1:%.*]] = select i1 [[CMP1]], float [[A1]], float [[B1]] 172; MINTREESIZE-NEXT: [[S2:%.*]] = select i1 [[CMP2]], float [[A2]], float [[B2]] 173; MINTREESIZE-NEXT: [[S3:%.*]] = select i1 [[CMP3]], float [[A3]], float [[B3]] 174; MINTREESIZE-NEXT: [[RA:%.*]] = insertelement <4 x float> undef, float [[S0]], i32 0 175; MINTREESIZE-NEXT: [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[S1]], i32 1 176; MINTREESIZE-NEXT: [[RC:%.*]] = insertelement <4 x float> [[RB]], float [[S2]], i32 2 177; MINTREESIZE-NEXT: [[RD:%.*]] = insertelement <4 x float> [[RC]], float [[S3]], i32 3 178; MINTREESIZE-NEXT: [[Q0:%.*]] = extractelement <4 x float> [[RD]], i32 0 179; MINTREESIZE-NEXT: [[Q1:%.*]] = extractelement <4 x float> [[RD]], i32 1 180; MINTREESIZE-NEXT: [[Q2:%.*]] = extractelement <4 x float> [[RD]], i32 2 181; MINTREESIZE-NEXT: [[Q3:%.*]] = extractelement <4 x float> [[RD]], i32 3 182; MINTREESIZE-NEXT: [[TMP1:%.*]] = insertelement <2 x float> poison, float [[Q2]], i32 0 183; MINTREESIZE-NEXT: [[TMP2:%.*]] = insertelement <2 x float> [[TMP1]], float [[Q3]], i32 1 184; MINTREESIZE-NEXT: [[Q4:%.*]] = fadd float [[Q0]], [[Q1]] 185; MINTREESIZE-NEXT: [[Q5:%.*]] = fadd float [[Q2]], [[Q3]] 186; MINTREESIZE-NEXT: [[TMP3:%.*]] = insertelement <2 x float> poison, float [[Q4]], i32 0 187; MINTREESIZE-NEXT: [[TMP4:%.*]] = insertelement <2 x float> [[TMP3]], float [[Q5]], i32 1 188; MINTREESIZE-NEXT: [[Q6:%.*]] = fadd float [[Q4]], [[Q5]] 189; MINTREESIZE-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[Q6]], i32 0 190; MINTREESIZE-NEXT: [[TMP6:%.*]] = insertelement <2 x float> [[TMP5]], float [[Q5]], i32 1 191; MINTREESIZE-NEXT: [[QI:%.*]] = fcmp olt float [[Q6]], [[Q5]] 192; MINTREESIZE-NEXT: call void @llvm.assume(i1 [[QI]]) 193; MINTREESIZE-NEXT: ret <4 x float> undef 194; 195 %c0 = extractelement <4 x i32> %c, i32 0 196 %c1 = extractelement <4 x i32> %c, i32 1 197 %c2 = extractelement <4 x i32> %c, i32 2 198 %c3 = extractelement <4 x i32> %c, i32 3 199 %a0 = extractelement <4 x float> %a, i32 0 200 %a1 = extractelement <4 x float> %a, i32 1 201 %a2 = extractelement <4 x float> %a, i32 2 202 %a3 = extractelement <4 x float> %a, i32 3 203 %b0 = extractelement <4 x float> %b, i32 0 204 %b1 = extractelement <4 x float> %b, i32 1 205 %b2 = extractelement <4 x float> %b, i32 2 206 %b3 = extractelement <4 x float> %b, i32 3 207 %cmp0 = icmp ne i32 %c0, 0 208 %cmp1 = icmp ne i32 %c1, 0 209 %cmp2 = icmp ne i32 %c2, 0 210 %cmp3 = icmp ne i32 %c3, 0 211 %s0 = select i1 %cmp0, float %a0, float %b0 212 %s1 = select i1 %cmp1, float %a1, float %b1 213 %s2 = select i1 %cmp2, float %a2, float %b2 214 %s3 = select i1 %cmp3, float %a3, float %b3 215 %ra = insertelement <4 x float> undef, float %s0, i32 0 216 %rb = insertelement <4 x float> %ra, float %s1, i32 1 217 %rc = insertelement <4 x float> %rb, float %s2, i32 2 218 %rd = insertelement <4 x float> %rc, float %s3, i32 3 219 %q0 = extractelement <4 x float> %rd, i32 0 220 %q1 = extractelement <4 x float> %rd, i32 1 221 %q2 = extractelement <4 x float> %rd, i32 2 222 %q3 = extractelement <4 x float> %rd, i32 3 223 %q4 = fadd float %q0, %q1 224 %q5 = fadd float %q2, %q3 225 %q6 = fadd float %q4, %q5 226 %qi = fcmp olt float %q6, %q5 227 call void @llvm.assume(i1 %qi) 228 ret <4 x float> undef 229} 230 231; Insert in an order different from the vector indices to make sure it 232; doesn't matter 233define <4 x float> @simple_select_insert_out_of_order(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 234; CHECK-LABEL: @simple_select_insert_out_of_order( 235; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer 236; CHECK-NEXT: [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]] 237; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> poison, <4 x i32> <i32 2, i32 1, i32 0, i32 3> 238; CHECK-NEXT: ret <4 x float> [[TMP3]] 239; 240 %c0 = extractelement <4 x i32> %c, i32 0 241 %c1 = extractelement <4 x i32> %c, i32 1 242 %c2 = extractelement <4 x i32> %c, i32 2 243 %c3 = extractelement <4 x i32> %c, i32 3 244 %a0 = extractelement <4 x float> %a, i32 0 245 %a1 = extractelement <4 x float> %a, i32 1 246 %a2 = extractelement <4 x float> %a, i32 2 247 %a3 = extractelement <4 x float> %a, i32 3 248 %b0 = extractelement <4 x float> %b, i32 0 249 %b1 = extractelement <4 x float> %b, i32 1 250 %b2 = extractelement <4 x float> %b, i32 2 251 %b3 = extractelement <4 x float> %b, i32 3 252 %cmp0 = icmp ne i32 %c0, 0 253 %cmp1 = icmp ne i32 %c1, 0 254 %cmp2 = icmp ne i32 %c2, 0 255 %cmp3 = icmp ne i32 %c3, 0 256 %s0 = select i1 %cmp0, float %a0, float %b0 257 %s1 = select i1 %cmp1, float %a1, float %b1 258 %s2 = select i1 %cmp2, float %a2, float %b2 259 %s3 = select i1 %cmp3, float %a3, float %b3 260 %ra = insertelement <4 x float> undef, float %s0, i32 2 261 %rb = insertelement <4 x float> %ra, float %s1, i32 1 262 %rc = insertelement <4 x float> %rb, float %s2, i32 0 263 %rd = insertelement <4 x float> %rc, float %s3, i32 3 264 ret <4 x float> %rd 265} 266 267declare void @v4f32_user(<4 x float>) #0 268declare void @f32_user(float) #0 269 270; Multiple users of the final constructed vector 271define <4 x float> @simple_select_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 272; CHECK-LABEL: @simple_select_users( 273; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <4 x i32> [[C:%.*]], zeroinitializer 274; CHECK-NEXT: [[TMP2:%.*]] = select <4 x i1> [[TMP1]], <4 x float> [[A:%.*]], <4 x float> [[B:%.*]] 275; CHECK-NEXT: call void @v4f32_user(<4 x float> [[TMP2]]) #[[ATTR0:[0-9]+]] 276; CHECK-NEXT: ret <4 x float> [[TMP2]] 277; 278 %c0 = extractelement <4 x i32> %c, i32 0 279 %c1 = extractelement <4 x i32> %c, i32 1 280 %c2 = extractelement <4 x i32> %c, i32 2 281 %c3 = extractelement <4 x i32> %c, i32 3 282 %a0 = extractelement <4 x float> %a, i32 0 283 %a1 = extractelement <4 x float> %a, i32 1 284 %a2 = extractelement <4 x float> %a, i32 2 285 %a3 = extractelement <4 x float> %a, i32 3 286 %b0 = extractelement <4 x float> %b, i32 0 287 %b1 = extractelement <4 x float> %b, i32 1 288 %b2 = extractelement <4 x float> %b, i32 2 289 %b3 = extractelement <4 x float> %b, i32 3 290 %cmp0 = icmp ne i32 %c0, 0 291 %cmp1 = icmp ne i32 %c1, 0 292 %cmp2 = icmp ne i32 %c2, 0 293 %cmp3 = icmp ne i32 %c3, 0 294 %s0 = select i1 %cmp0, float %a0, float %b0 295 %s1 = select i1 %cmp1, float %a1, float %b1 296 %s2 = select i1 %cmp2, float %a2, float %b2 297 %s3 = select i1 %cmp3, float %a3, float %b3 298 %ra = insertelement <4 x float> undef, float %s0, i32 0 299 %rb = insertelement <4 x float> %ra, float %s1, i32 1 300 %rc = insertelement <4 x float> %rb, float %s2, i32 2 301 %rd = insertelement <4 x float> %rc, float %s3, i32 3 302 call void @v4f32_user(<4 x float> %rd) #0 303 ret <4 x float> %rd 304} 305 306; Unused insertelement 307define <4 x float> @simple_select_no_users(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 308; CHECK-LABEL: @simple_select_no_users( 309; CHECK-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0 310; CHECK-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1 311; CHECK-NEXT: [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2 312; CHECK-NEXT: [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3 313; CHECK-NEXT: [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0 314; CHECK-NEXT: [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1 315; CHECK-NEXT: [[A2:%.*]] = extractelement <4 x float> [[A]], i32 2 316; CHECK-NEXT: [[A3:%.*]] = extractelement <4 x float> [[A]], i32 3 317; CHECK-NEXT: [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0 318; CHECK-NEXT: [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1 319; CHECK-NEXT: [[B2:%.*]] = extractelement <4 x float> [[B]], i32 2 320; CHECK-NEXT: [[B3:%.*]] = extractelement <4 x float> [[B]], i32 3 321; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> poison, i32 [[C0]], i32 0 322; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1 323; CHECK-NEXT: [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer 324; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x i32> poison, i32 [[C2]], i32 0 325; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x i32> [[TMP4]], i32 [[C3]], i32 1 326; CHECK-NEXT: [[TMP6:%.*]] = icmp ne <2 x i32> [[TMP5]], zeroinitializer 327; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x float> poison, float [[A0]], i32 0 328; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x float> [[TMP7]], float [[A1]], i32 1 329; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x float> poison, float [[B0]], i32 0 330; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[B1]], i32 1 331; CHECK-NEXT: [[TMP11:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP8]], <2 x float> [[TMP10]] 332; CHECK-NEXT: [[TMP12:%.*]] = insertelement <2 x float> poison, float [[A2]], i32 0 333; CHECK-NEXT: [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[A3]], i32 1 334; CHECK-NEXT: [[TMP14:%.*]] = insertelement <2 x float> poison, float [[B2]], i32 0 335; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[B3]], i32 1 336; CHECK-NEXT: [[TMP16:%.*]] = select <2 x i1> [[TMP6]], <2 x float> [[TMP13]], <2 x float> [[TMP15]] 337; CHECK-NEXT: [[TMP17:%.*]] = shufflevector <2 x float> [[TMP11]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 338; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <2 x float> [[TMP16]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef> 339; CHECK-NEXT: [[RD1:%.*]] = shufflevector <4 x float> undef, <4 x float> [[TMP18]], <4 x i32> <i32 0, i32 1, i32 4, i32 5> 340; CHECK-NEXT: ret <4 x float> [[RD1]] 341; 342 %c0 = extractelement <4 x i32> %c, i32 0 343 %c1 = extractelement <4 x i32> %c, i32 1 344 %c2 = extractelement <4 x i32> %c, i32 2 345 %c3 = extractelement <4 x i32> %c, i32 3 346 %a0 = extractelement <4 x float> %a, i32 0 347 %a1 = extractelement <4 x float> %a, i32 1 348 %a2 = extractelement <4 x float> %a, i32 2 349 %a3 = extractelement <4 x float> %a, i32 3 350 %b0 = extractelement <4 x float> %b, i32 0 351 %b1 = extractelement <4 x float> %b, i32 1 352 %b2 = extractelement <4 x float> %b, i32 2 353 %b3 = extractelement <4 x float> %b, i32 3 354 %cmp0 = icmp ne i32 %c0, 0 355 %cmp1 = icmp ne i32 %c1, 0 356 %cmp2 = icmp ne i32 %c2, 0 357 %cmp3 = icmp ne i32 %c3, 0 358 %s0 = select i1 %cmp0, float %a0, float %b0 359 %s1 = select i1 %cmp1, float %a1, float %b1 360 %s2 = select i1 %cmp2, float %a2, float %b2 361 %s3 = select i1 %cmp3, float %a3, float %b3 362 %ra = insertelement <4 x float> undef, float %s0, i32 0 363 %rb = insertelement <4 x float> %ra, float %s1, i32 1 364 %rc = insertelement <4 x float> undef, float %s2, i32 2 365 %rd = insertelement <4 x float> %rc, float %s3, i32 3 366 ret <4 x float> %rd 367} 368 369; Make sure infinite loop doesn't happen which I ran into when trying 370; to do this backwards this backwards 371define <4 x i32> @reconstruct(<4 x i32> %c) #0 { 372; CHECK-LABEL: @reconstruct( 373; CHECK-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0 374; CHECK-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1 375; CHECK-NEXT: [[C2:%.*]] = extractelement <4 x i32> [[C]], i32 2 376; CHECK-NEXT: [[C3:%.*]] = extractelement <4 x i32> [[C]], i32 3 377; CHECK-NEXT: [[RA:%.*]] = insertelement <4 x i32> undef, i32 [[C0]], i32 0 378; CHECK-NEXT: [[RB:%.*]] = insertelement <4 x i32> [[RA]], i32 [[C1]], i32 1 379; CHECK-NEXT: [[RC:%.*]] = insertelement <4 x i32> [[RB]], i32 [[C2]], i32 2 380; CHECK-NEXT: [[RD:%.*]] = insertelement <4 x i32> [[RC]], i32 [[C3]], i32 3 381; CHECK-NEXT: ret <4 x i32> [[RD]] 382; 383 %c0 = extractelement <4 x i32> %c, i32 0 384 %c1 = extractelement <4 x i32> %c, i32 1 385 %c2 = extractelement <4 x i32> %c, i32 2 386 %c3 = extractelement <4 x i32> %c, i32 3 387 %ra = insertelement <4 x i32> undef, i32 %c0, i32 0 388 %rb = insertelement <4 x i32> %ra, i32 %c1, i32 1 389 %rc = insertelement <4 x i32> %rb, i32 %c2, i32 2 390 %rd = insertelement <4 x i32> %rc, i32 %c3, i32 3 391 ret <4 x i32> %rd 392} 393 394define <2 x float> @simple_select_v2(<2 x float> %a, <2 x float> %b, <2 x i32> %c) #0 { 395; CHECK-LABEL: @simple_select_v2( 396; CHECK-NEXT: [[TMP1:%.*]] = icmp ne <2 x i32> [[C:%.*]], zeroinitializer 397; CHECK-NEXT: [[TMP2:%.*]] = select <2 x i1> [[TMP1]], <2 x float> [[A:%.*]], <2 x float> [[B:%.*]] 398; CHECK-NEXT: ret <2 x float> [[TMP2]] 399; 400 %c0 = extractelement <2 x i32> %c, i32 0 401 %c1 = extractelement <2 x i32> %c, i32 1 402 %a0 = extractelement <2 x float> %a, i32 0 403 %a1 = extractelement <2 x float> %a, i32 1 404 %b0 = extractelement <2 x float> %b, i32 0 405 %b1 = extractelement <2 x float> %b, i32 1 406 %cmp0 = icmp ne i32 %c0, 0 407 %cmp1 = icmp ne i32 %c1, 0 408 %s0 = select i1 %cmp0, float %a0, float %b0 409 %s1 = select i1 %cmp1, float %a1, float %b1 410 %ra = insertelement <2 x float> undef, float %s0, i32 0 411 %rb = insertelement <2 x float> %ra, float %s1, i32 1 412 ret <2 x float> %rb 413} 414 415; Make sure when we construct partial vectors, we don't keep 416; re-visiting the insertelement chains starting with undef 417; (low cost threshold needed to force this to happen) 418define <4 x float> @simple_select_partial_vector(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { 419; CHECK-LABEL: @simple_select_partial_vector( 420; CHECK-NEXT: [[C0:%.*]] = extractelement <4 x i32> [[C:%.*]], i32 0 421; CHECK-NEXT: [[C1:%.*]] = extractelement <4 x i32> [[C]], i32 1 422; CHECK-NEXT: [[A0:%.*]] = extractelement <4 x float> [[A:%.*]], i32 0 423; CHECK-NEXT: [[A1:%.*]] = extractelement <4 x float> [[A]], i32 1 424; CHECK-NEXT: [[B0:%.*]] = extractelement <4 x float> [[B:%.*]], i32 0 425; CHECK-NEXT: [[B1:%.*]] = extractelement <4 x float> [[B]], i32 1 426; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> undef, i32 [[C0]], i32 0 427; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i32> [[TMP1]], i32 [[C1]], i32 1 428; CHECK-NEXT: [[TMP3:%.*]] = icmp ne <2 x i32> [[TMP2]], zeroinitializer 429; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x float> undef, float [[A0]], i32 0 430; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x float> [[TMP4]], float [[A1]], i32 1 431; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x float> undef, float [[B0]], i32 0 432; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x float> [[TMP6]], float [[B1]], i32 1 433; CHECK-NEXT: [[TMP8:%.*]] = select <2 x i1> [[TMP3]], <2 x float> [[TMP5]], <2 x float> [[TMP7]] 434; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x float> [[TMP8]], i32 0 435; CHECK-NEXT: [[RA:%.*]] = insertelement <4 x float> undef, float [[TMP9]], i32 0 436; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x float> [[TMP8]], i32 1 437; CHECK-NEXT: [[RB:%.*]] = insertelement <4 x float> [[RA]], float [[TMP10]], i32 1 438; CHECK-NEXT: ret <4 x float> [[RB]] 439; 440 %c0 = extractelement <4 x i32> %c, i32 0 441 %c1 = extractelement <4 x i32> %c, i32 1 442 %a0 = extractelement <4 x float> %a, i32 0 443 %a1 = extractelement <4 x float> %a, i32 1 444 %b0 = extractelement <4 x float> %b, i32 0 445 %b1 = extractelement <4 x float> %b, i32 1 446 %1 = insertelement <2 x i32> undef, i32 %c0, i32 0 447 %2 = insertelement <2 x i32> %1, i32 %c1, i32 1 448 %3 = icmp ne <2 x i32> %2, zeroinitializer 449 %4 = insertelement <2 x float> undef, float %a0, i32 0 450 %5 = insertelement <2 x float> %4, float %a1, i32 1 451 %6 = insertelement <2 x float> undef, float %b0, i32 0 452 %7 = insertelement <2 x float> %6, float %b1, i32 1 453 %8 = select <2 x i1> %3, <2 x float> %5, <2 x float> %7 454 %9 = extractelement <2 x float> %8, i32 0 455 %ra = insertelement <4 x float> undef, float %9, i32 0 456 %10 = extractelement <2 x float> %8, i32 1 457 %rb = insertelement <4 x float> %ra, float %10, i32 1 458 ret <4 x float> %rb 459} 460 461; Make sure that vectorization happens even if insertelements operations 462; must be rescheduled. The case here is from compiling Julia. 463define <4 x float> @reschedule_extract(<4 x float> %a, <4 x float> %b) { 464; THRESHOLD-LABEL: @reschedule_extract( 465; THRESHOLD-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]] 466; THRESHOLD-NEXT: ret <4 x float> [[TMP1]] 467; 468; NOTHRESHOLD-LABEL: @reschedule_extract( 469; NOTHRESHOLD-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]] 470; NOTHRESHOLD-NEXT: ret <4 x float> [[TMP1]] 471; 472; MINTREESIZE-LABEL: @reschedule_extract( 473; MINTREESIZE-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[B:%.*]], i32 3 474; MINTREESIZE-NEXT: [[TMP2:%.*]] = extractelement <4 x float> [[B]], i32 2 475; MINTREESIZE-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[B]], i32 1 476; MINTREESIZE-NEXT: [[TMP4:%.*]] = extractelement <4 x float> [[B]], i32 0 477; MINTREESIZE-NEXT: [[TMP5:%.*]] = extractelement <4 x float> [[A:%.*]], i32 3 478; MINTREESIZE-NEXT: [[TMP6:%.*]] = extractelement <4 x float> [[A]], i32 2 479; MINTREESIZE-NEXT: [[TMP7:%.*]] = extractelement <4 x float> [[A]], i32 1 480; MINTREESIZE-NEXT: [[TMP8:%.*]] = extractelement <4 x float> [[A]], i32 0 481; MINTREESIZE-NEXT: [[TMP9:%.*]] = insertelement <2 x float> poison, float [[TMP8]], i32 0 482; MINTREESIZE-NEXT: [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[TMP4]], i32 1 483; MINTREESIZE-NEXT: [[TMP11:%.*]] = fadd <4 x float> [[A]], [[B]] 484; MINTREESIZE-NEXT: [[TMP12:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i32 0 485; MINTREESIZE-NEXT: [[TMP13:%.*]] = insertelement <2 x float> [[TMP12]], float [[TMP3]], i32 1 486; MINTREESIZE-NEXT: [[TMP14:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i32 0 487; MINTREESIZE-NEXT: [[TMP15:%.*]] = insertelement <2 x float> [[TMP14]], float [[TMP2]], i32 1 488; MINTREESIZE-NEXT: [[TMP16:%.*]] = insertelement <2 x float> poison, float [[TMP5]], i32 0 489; MINTREESIZE-NEXT: [[TMP17:%.*]] = insertelement <2 x float> [[TMP16]], float [[TMP1]], i32 1 490; MINTREESIZE-NEXT: ret <4 x float> [[TMP11]] 491; 492 %a0 = extractelement <4 x float> %a, i32 0 493 %b0 = extractelement <4 x float> %b, i32 0 494 %c0 = fadd float %a0, %b0 495 %v0 = insertelement <4 x float> undef, float %c0, i32 0 496 %a1 = extractelement <4 x float> %a, i32 1 497 %b1 = extractelement <4 x float> %b, i32 1 498 %c1 = fadd float %a1, %b1 499 %v1 = insertelement <4 x float> %v0, float %c1, i32 1 500 %a2 = extractelement <4 x float> %a, i32 2 501 %b2 = extractelement <4 x float> %b, i32 2 502 %c2 = fadd float %a2, %b2 503 %v2 = insertelement <4 x float> %v1, float %c2, i32 2 504 %a3 = extractelement <4 x float> %a, i32 3 505 %b3 = extractelement <4 x float> %b, i32 3 506 %c3 = fadd float %a3, %b3 507 %v3 = insertelement <4 x float> %v2, float %c3, i32 3 508 ret <4 x float> %v3 509} 510 511; Check that cost model for vectorization takes credit for 512; instructions that are erased. 513define <4 x float> @take_credit(<4 x float> %a, <4 x float> %b) { 514; THRESHOLD-LABEL: @take_credit( 515; THRESHOLD-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]] 516; THRESHOLD-NEXT: ret <4 x float> [[TMP1]] 517; 518; NOTHRESHOLD-LABEL: @take_credit( 519; NOTHRESHOLD-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[A:%.*]], [[B:%.*]] 520; NOTHRESHOLD-NEXT: ret <4 x float> [[TMP1]] 521; 522; MINTREESIZE-LABEL: @take_credit( 523; MINTREESIZE-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[B:%.*]], i32 3 524; MINTREESIZE-NEXT: [[TMP2:%.*]] = extractelement <4 x float> [[B]], i32 2 525; MINTREESIZE-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[B]], i32 1 526; MINTREESIZE-NEXT: [[TMP4:%.*]] = extractelement <4 x float> [[B]], i32 0 527; MINTREESIZE-NEXT: [[TMP5:%.*]] = extractelement <4 x float> [[A:%.*]], i32 3 528; MINTREESIZE-NEXT: [[TMP6:%.*]] = extractelement <4 x float> [[A]], i32 2 529; MINTREESIZE-NEXT: [[TMP7:%.*]] = extractelement <4 x float> [[A]], i32 1 530; MINTREESIZE-NEXT: [[TMP8:%.*]] = extractelement <4 x float> [[A]], i32 0 531; MINTREESIZE-NEXT: [[TMP9:%.*]] = insertelement <2 x float> poison, float [[TMP8]], i32 0 532; MINTREESIZE-NEXT: [[TMP10:%.*]] = insertelement <2 x float> [[TMP9]], float [[TMP4]], i32 1 533; MINTREESIZE-NEXT: [[TMP11:%.*]] = insertelement <2 x float> poison, float [[TMP7]], i32 0 534; MINTREESIZE-NEXT: [[TMP12:%.*]] = insertelement <2 x float> [[TMP11]], float [[TMP3]], i32 1 535; MINTREESIZE-NEXT: [[TMP13:%.*]] = insertelement <2 x float> poison, float [[TMP6]], i32 0 536; MINTREESIZE-NEXT: [[TMP14:%.*]] = insertelement <2 x float> [[TMP13]], float [[TMP2]], i32 1 537; MINTREESIZE-NEXT: [[TMP15:%.*]] = insertelement <2 x float> poison, float [[TMP5]], i32 0 538; MINTREESIZE-NEXT: [[TMP16:%.*]] = insertelement <2 x float> [[TMP15]], float [[TMP1]], i32 1 539; MINTREESIZE-NEXT: [[TMP17:%.*]] = fadd <4 x float> [[A]], [[B]] 540; MINTREESIZE-NEXT: ret <4 x float> [[TMP17]] 541; 542 %a0 = extractelement <4 x float> %a, i32 0 543 %b0 = extractelement <4 x float> %b, i32 0 544 %c0 = fadd float %a0, %b0 545 %a1 = extractelement <4 x float> %a, i32 1 546 %b1 = extractelement <4 x float> %b, i32 1 547 %c1 = fadd float %a1, %b1 548 %a2 = extractelement <4 x float> %a, i32 2 549 %b2 = extractelement <4 x float> %b, i32 2 550 %c2 = fadd float %a2, %b2 551 %a3 = extractelement <4 x float> %a, i32 3 552 %b3 = extractelement <4 x float> %b, i32 3 553 %c3 = fadd float %a3, %b3 554 %v0 = insertelement <4 x float> undef, float %c0, i32 0 555 %v1 = insertelement <4 x float> %v0, float %c1, i32 1 556 %v2 = insertelement <4 x float> %v1, float %c2, i32 2 557 %v3 = insertelement <4 x float> %v2, float %c3, i32 3 558 ret <4 x float> %v3 559} 560 561; Make sure we handle multiple trees that feed one build vector correctly. 562define <4 x double> @multi_tree(double %w, double %x, double %y, double %z) { 563; CHECK-LABEL: @multi_tree( 564; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> poison, double [[Z:%.*]], i32 0 565; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x double> [[TMP1]], double [[Y:%.*]], i32 1 566; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP2]], double [[X:%.*]], i32 2 567; CHECK-NEXT: [[TMP4:%.*]] = insertelement <4 x double> [[TMP3]], double [[W:%.*]], i32 3 568; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 3.000000e+00, double 2.000000e+00, double 1.000000e+00, double 0.000000e+00> 569; CHECK-NEXT: [[TMP6:%.*]] = fmul <4 x double> [[TMP5]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00> 570; CHECK-NEXT: ret <4 x double> [[TMP6]] 571; 572 %t0 = fadd double %w , 0.000000e+00 573 %t1 = fadd double %x , 1.000000e+00 574 %t2 = fadd double %y , 2.000000e+00 575 %t3 = fadd double %z , 3.000000e+00 576 %t4 = fmul double %t0, 1.000000e+00 577 %i1 = insertelement <4 x double> undef, double %t4, i32 3 578 %t5 = fmul double %t1, 1.000000e+00 579 %i2 = insertelement <4 x double> %i1, double %t5, i32 2 580 %t6 = fmul double %t2, 1.000000e+00 581 %i3 = insertelement <4 x double> %i2, double %t6, i32 1 582 %t7 = fmul double %t3, 1.000000e+00 583 %i4 = insertelement <4 x double> %i3, double %t7, i32 0 584 ret <4 x double> %i4 585} 586 587define <8 x float> @_vadd256(<8 x float> %a, <8 x float> %b) local_unnamed_addr #0 { 588; THRESHOLD-LABEL: @_vadd256( 589; THRESHOLD-NEXT: [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]] 590; THRESHOLD-NEXT: ret <8 x float> [[TMP1]] 591; 592; NOTHRESHOLD-LABEL: @_vadd256( 593; NOTHRESHOLD-NEXT: [[TMP1:%.*]] = fadd <8 x float> [[A:%.*]], [[B:%.*]] 594; NOTHRESHOLD-NEXT: ret <8 x float> [[TMP1]] 595; 596; MINTREESIZE-LABEL: @_vadd256( 597; MINTREESIZE-NEXT: [[TMP1:%.*]] = extractelement <8 x float> [[B:%.*]], i32 7 598; MINTREESIZE-NEXT: [[TMP2:%.*]] = extractelement <8 x float> [[B]], i32 6 599; MINTREESIZE-NEXT: [[TMP3:%.*]] = extractelement <8 x float> [[B]], i32 5 600; MINTREESIZE-NEXT: [[TMP4:%.*]] = extractelement <8 x float> [[B]], i32 4 601; MINTREESIZE-NEXT: [[TMP5:%.*]] = extractelement <8 x float> [[B]], i32 3 602; MINTREESIZE-NEXT: [[TMP6:%.*]] = extractelement <8 x float> [[B]], i32 2 603; MINTREESIZE-NEXT: [[TMP7:%.*]] = extractelement <8 x float> [[B]], i32 1 604; MINTREESIZE-NEXT: [[TMP8:%.*]] = extractelement <8 x float> [[B]], i32 0 605; MINTREESIZE-NEXT: [[TMP9:%.*]] = extractelement <8 x float> [[A:%.*]], i32 7 606; MINTREESIZE-NEXT: [[TMP10:%.*]] = extractelement <8 x float> [[A]], i32 6 607; MINTREESIZE-NEXT: [[TMP11:%.*]] = extractelement <8 x float> [[A]], i32 5 608; MINTREESIZE-NEXT: [[TMP12:%.*]] = extractelement <8 x float> [[A]], i32 4 609; MINTREESIZE-NEXT: [[TMP13:%.*]] = extractelement <8 x float> [[A]], i32 3 610; MINTREESIZE-NEXT: [[TMP14:%.*]] = extractelement <8 x float> [[A]], i32 2 611; MINTREESIZE-NEXT: [[TMP15:%.*]] = extractelement <8 x float> [[A]], i32 1 612; MINTREESIZE-NEXT: [[TMP16:%.*]] = extractelement <8 x float> [[A]], i32 0 613; MINTREESIZE-NEXT: [[TMP17:%.*]] = insertelement <2 x float> poison, float [[TMP16]], i32 0 614; MINTREESIZE-NEXT: [[TMP18:%.*]] = insertelement <2 x float> [[TMP17]], float [[TMP8]], i32 1 615; MINTREESIZE-NEXT: [[TMP19:%.*]] = insertelement <2 x float> poison, float [[TMP15]], i32 0 616; MINTREESIZE-NEXT: [[TMP20:%.*]] = insertelement <2 x float> [[TMP19]], float [[TMP7]], i32 1 617; MINTREESIZE-NEXT: [[TMP21:%.*]] = insertelement <2 x float> poison, float [[TMP14]], i32 0 618; MINTREESIZE-NEXT: [[TMP22:%.*]] = insertelement <2 x float> [[TMP21]], float [[TMP6]], i32 1 619; MINTREESIZE-NEXT: [[TMP23:%.*]] = insertelement <2 x float> poison, float [[TMP13]], i32 0 620; MINTREESIZE-NEXT: [[TMP24:%.*]] = insertelement <2 x float> [[TMP23]], float [[TMP5]], i32 1 621; MINTREESIZE-NEXT: [[TMP25:%.*]] = insertelement <2 x float> poison, float [[TMP12]], i32 0 622; MINTREESIZE-NEXT: [[TMP26:%.*]] = insertelement <2 x float> [[TMP25]], float [[TMP4]], i32 1 623; MINTREESIZE-NEXT: [[TMP27:%.*]] = insertelement <2 x float> poison, float [[TMP11]], i32 0 624; MINTREESIZE-NEXT: [[TMP28:%.*]] = insertelement <2 x float> [[TMP27]], float [[TMP3]], i32 1 625; MINTREESIZE-NEXT: [[TMP29:%.*]] = insertelement <2 x float> poison, float [[TMP10]], i32 0 626; MINTREESIZE-NEXT: [[TMP30:%.*]] = insertelement <2 x float> [[TMP29]], float [[TMP2]], i32 1 627; MINTREESIZE-NEXT: [[TMP31:%.*]] = insertelement <2 x float> poison, float [[TMP9]], i32 0 628; MINTREESIZE-NEXT: [[TMP32:%.*]] = insertelement <2 x float> [[TMP31]], float [[TMP1]], i32 1 629; MINTREESIZE-NEXT: [[TMP33:%.*]] = fadd <8 x float> [[A]], [[B]] 630; MINTREESIZE-NEXT: ret <8 x float> [[TMP33]] 631; 632 %vecext = extractelement <8 x float> %a, i32 0 633 %vecext1 = extractelement <8 x float> %b, i32 0 634 %add = fadd float %vecext, %vecext1 635 %vecext2 = extractelement <8 x float> %a, i32 1 636 %vecext3 = extractelement <8 x float> %b, i32 1 637 %add4 = fadd float %vecext2, %vecext3 638 %vecext5 = extractelement <8 x float> %a, i32 2 639 %vecext6 = extractelement <8 x float> %b, i32 2 640 %add7 = fadd float %vecext5, %vecext6 641 %vecext8 = extractelement <8 x float> %a, i32 3 642 %vecext9 = extractelement <8 x float> %b, i32 3 643 %add10 = fadd float %vecext8, %vecext9 644 %vecext11 = extractelement <8 x float> %a, i32 4 645 %vecext12 = extractelement <8 x float> %b, i32 4 646 %add13 = fadd float %vecext11, %vecext12 647 %vecext14 = extractelement <8 x float> %a, i32 5 648 %vecext15 = extractelement <8 x float> %b, i32 5 649 %add16 = fadd float %vecext14, %vecext15 650 %vecext17 = extractelement <8 x float> %a, i32 6 651 %vecext18 = extractelement <8 x float> %b, i32 6 652 %add19 = fadd float %vecext17, %vecext18 653 %vecext20 = extractelement <8 x float> %a, i32 7 654 %vecext21 = extractelement <8 x float> %b, i32 7 655 %add22 = fadd float %vecext20, %vecext21 656 %vecinit.i = insertelement <8 x float> undef, float %add, i32 0 657 %vecinit1.i = insertelement <8 x float> %vecinit.i, float %add4, i32 1 658 %vecinit2.i = insertelement <8 x float> %vecinit1.i, float %add7, i32 2 659 %vecinit3.i = insertelement <8 x float> %vecinit2.i, float %add10, i32 3 660 %vecinit4.i = insertelement <8 x float> %vecinit3.i, float %add13, i32 4 661 %vecinit5.i = insertelement <8 x float> %vecinit4.i, float %add16, i32 5 662 %vecinit6.i = insertelement <8 x float> %vecinit5.i, float %add19, i32 6 663 %vecinit7.i = insertelement <8 x float> %vecinit6.i, float %add22, i32 7 664 ret <8 x float> %vecinit7.i 665} 666 667attributes #0 = { nounwind ssp uwtable "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "unsafe-fp-math"="false" "use-soft-float"="false" } 668