1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -S -slp-vectorizer -slp-threshold=-18 -dce -instcombine -pass-remarks-output=%t < %s | FileCheck %s 3; RUN: cat %t | FileCheck -check-prefix=YAML %s 4; RUN: opt -S -passes='slp-vectorizer,dce,instcombine' -slp-threshold=-18 -pass-remarks-output=%t < %s | FileCheck %s 5; RUN: cat %t | FileCheck -check-prefix=YAML %s 6 7 8target datalayout = "e-m:e-i32:64-i128:128-n32:64-S128" 9target triple = "aarch64--linux-gnu" 10 11; These tests check that we remove from consideration pairs of seed 12; getelementptrs when they are known to have a constant difference. Such pairs 13; are likely not good candidates for vectorization since one can be computed 14; from the other. We use an unprofitable threshold to force vectorization. 15; 16; int getelementptr(int *g, int n, int w, int x, int y, int z) { 17; int sum = 0; 18; for (int i = 0; i < n ; ++i) { 19; sum += g[2*i + w]; sum += g[2*i + x]; 20; sum += g[2*i + y]; sum += g[2*i + z]; 21; } 22; return sum; 23; } 24; 25 26; YAML-LABEL: Function: getelementptr_4x32 27; YAML-NEXT: Args: 28; YAML-NEXT: - String: 'SLP vectorized with cost ' 29; YAML-NEXT: - Cost: '11' 30; YAML-NEXT: - String: ' and with tree size ' 31; YAML-NEXT: - TreeSize: '5' 32 33; YAML: --- !Passed 34; YAML-NEXT: Pass: slp-vectorizer 35; YAML-NEXT: Name: VectorizedList 36; YAML-NEXT: Function: getelementptr_4x32 37; YAML-NEXT: Args: 38; YAML-NEXT: - String: 'SLP vectorized with cost ' 39; YAML-NEXT: - Cost: '16' 40; YAML-NEXT: - String: ' and with tree size ' 41; YAML-NEXT: - TreeSize: '3' 42 43define i32 @getelementptr_4x32(i32* nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) { 44; CHECK-LABEL: @getelementptr_4x32( 45; CHECK-NEXT: entry: 46; CHECK-NEXT: [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0 47; CHECK-NEXT: br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]] 48; CHECK: for.body.preheader: 49; CHECK-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> <i32 0, i32 undef, i32 undef, i32 undef>, i32 [[X:%.*]], i32 1 50; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> [[TMP0]], i32 [[Y:%.*]], i32 2 51; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[Z:%.*]], i32 3 52; CHECK-NEXT: br label [[FOR_BODY:%.*]] 53; CHECK: for.cond.cleanup.loopexit: 54; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i32> [[TMP21:%.*]], i32 1 55; CHECK-NEXT: br label [[FOR_COND_CLEANUP]] 56; CHECK: for.cond.cleanup: 57; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP3]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ] 58; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]] 59; CHECK: for.body: 60; CHECK-NEXT: [[TMP4:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP21]], [[FOR_BODY]] ] 61; CHECK-NEXT: [[TMP5:%.*]] = extractelement <2 x i32> [[TMP4]], i32 0 62; CHECK-NEXT: [[T4:%.*]] = shl nsw i32 [[TMP5]], 1 63; CHECK-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> undef, i32 [[T4]], i32 0 64; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <4 x i32> [[TMP6]], <4 x i32> undef, <4 x i32> zeroinitializer 65; CHECK-NEXT: [[TMP8:%.*]] = add nsw <4 x i32> [[TMP7]], [[TMP2]] 66; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> [[TMP8]], i32 0 67; CHECK-NEXT: [[TMP10:%.*]] = sext i32 [[TMP9]] to i64 68; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[G:%.*]], i64 [[TMP10]] 69; CHECK-NEXT: [[T6:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 70; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i32> [[TMP4]], i32 1 71; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP11]] 72; CHECK-NEXT: [[TMP12:%.*]] = extractelement <4 x i32> [[TMP8]], i32 1 73; CHECK-NEXT: [[TMP13:%.*]] = sext i32 [[TMP12]] to i64 74; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP13]] 75; CHECK-NEXT: [[T8:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4 76; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]] 77; CHECK-NEXT: [[TMP14:%.*]] = extractelement <4 x i32> [[TMP8]], i32 2 78; CHECK-NEXT: [[TMP15:%.*]] = sext i32 [[TMP14]] to i64 79; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP15]] 80; CHECK-NEXT: [[T10:%.*]] = load i32, i32* [[ARRAYIDX10]], align 4 81; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]] 82; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x i32> [[TMP8]], i32 3 83; CHECK-NEXT: [[TMP17:%.*]] = sext i32 [[TMP16]] to i64 84; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP17]] 85; CHECK-NEXT: [[T12:%.*]] = load i32, i32* [[ARRAYIDX15]], align 4 86; CHECK-NEXT: [[TMP18:%.*]] = insertelement <2 x i32> undef, i32 [[TMP5]], i32 0 87; CHECK-NEXT: [[TMP19:%.*]] = insertelement <2 x i32> [[TMP18]], i32 [[ADD11]], i32 1 88; CHECK-NEXT: [[TMP20:%.*]] = insertelement <2 x i32> <i32 1, i32 undef>, i32 [[T12]], i32 1 89; CHECK-NEXT: [[TMP21]] = add nsw <2 x i32> [[TMP19]], [[TMP20]] 90; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x i32> [[TMP21]], i32 0 91; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP22]], [[N]] 92; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]] 93; 94entry: 95 %cmp31 = icmp sgt i32 %n, 0 96 br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup 97 98for.body.preheader: 99 br label %for.body 100 101for.cond.cleanup.loopexit: 102 br label %for.cond.cleanup 103 104for.cond.cleanup: 105 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ] 106 ret i32 %sum.0.lcssa 107 108for.body: 109 %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ] 110 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ] 111 %t4 = shl nsw i32 %indvars.iv, 1 112 %t5 = add nsw i32 %t4, 0 113 %arrayidx = getelementptr inbounds i32, i32* %g, i32 %t5 114 %t6 = load i32, i32* %arrayidx, align 4 115 %add1 = add nsw i32 %t6, %sum.032 116 %t7 = add nsw i32 %t4, %x 117 %arrayidx5 = getelementptr inbounds i32, i32* %g, i32 %t7 118 %t8 = load i32, i32* %arrayidx5, align 4 119 %add6 = add nsw i32 %add1, %t8 120 %t9 = add nsw i32 %t4, %y 121 %arrayidx10 = getelementptr inbounds i32, i32* %g, i32 %t9 122 %t10 = load i32, i32* %arrayidx10, align 4 123 %add11 = add nsw i32 %add6, %t10 124 %t11 = add nsw i32 %t4, %z 125 %arrayidx15 = getelementptr inbounds i32, i32* %g, i32 %t11 126 %t12 = load i32, i32* %arrayidx15, align 4 127 %add16 = add nsw i32 %add11, %t12 128 %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1 129 %exitcond = icmp eq i32 %indvars.iv.next , %n 130 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body 131} 132 133; YAML-LABEL: Function: getelementptr_2x32 134; YAML-NEXT: Args: 135; YAML-NEXT: - String: 'SLP vectorized with cost ' 136; YAML-NEXT: - Cost: '11' 137; YAML-NEXT: - String: ' and with tree size ' 138; YAML-NEXT: - TreeSize: '5' 139 140; YAML: --- !Passed 141; YAML-NEXT: Pass: slp-vectorizer 142; YAML-NEXT: Name: VectorizedList 143; YAML-NEXT: Function: getelementptr_2x32 144; YAML-NEXT: Args: 145; YAML-NEXT: - String: 'SLP vectorized with cost ' 146; YAML-NEXT: - Cost: '6' 147; YAML-NEXT: - String: ' and with tree size ' 148; YAML-NEXT: - TreeSize: '3' 149 150define i32 @getelementptr_2x32(i32* nocapture readonly %g, i32 %n, i32 %x, i32 %y, i32 %z) { 151; CHECK-LABEL: @getelementptr_2x32( 152; CHECK-NEXT: entry: 153; CHECK-NEXT: [[CMP31:%.*]] = icmp sgt i32 [[N:%.*]], 0 154; CHECK-NEXT: br i1 [[CMP31]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]] 155; CHECK: for.body.preheader: 156; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i32> undef, i32 [[Y:%.*]], i32 0 157; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i32> [[TMP0]], i32 [[Z:%.*]], i32 1 158; CHECK-NEXT: br label [[FOR_BODY:%.*]] 159; CHECK: for.cond.cleanup.loopexit: 160; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x i32> [[TMP18:%.*]], i32 1 161; CHECK-NEXT: br label [[FOR_COND_CLEANUP]] 162; CHECK: for.cond.cleanup: 163; CHECK-NEXT: [[SUM_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[TMP2]], [[FOR_COND_CLEANUP_LOOPEXIT:%.*]] ] 164; CHECK-NEXT: ret i32 [[SUM_0_LCSSA]] 165; CHECK: for.body: 166; CHECK-NEXT: [[TMP3:%.*]] = phi <2 x i32> [ zeroinitializer, [[FOR_BODY_PREHEADER]] ], [ [[TMP18]], [[FOR_BODY]] ] 167; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i32> [[TMP3]], i32 0 168; CHECK-NEXT: [[T4:%.*]] = shl nsw i32 [[TMP4]], 1 169; CHECK-NEXT: [[TMP5:%.*]] = sext i32 [[T4]] to i64 170; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[G:%.*]], i64 [[TMP5]] 171; CHECK-NEXT: [[T6:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 172; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1 173; CHECK-NEXT: [[ADD1:%.*]] = add nsw i32 [[T6]], [[TMP6]] 174; CHECK-NEXT: [[T7:%.*]] = or i32 [[T4]], 1 175; CHECK-NEXT: [[TMP7:%.*]] = sext i32 [[T7]] to i64 176; CHECK-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP7]] 177; CHECK-NEXT: [[T8:%.*]] = load i32, i32* [[ARRAYIDX5]], align 4 178; CHECK-NEXT: [[ADD6:%.*]] = add nsw i32 [[ADD1]], [[T8]] 179; CHECK-NEXT: [[TMP8:%.*]] = insertelement <2 x i32> undef, i32 [[T4]], i32 0 180; CHECK-NEXT: [[TMP9:%.*]] = shufflevector <2 x i32> [[TMP8]], <2 x i32> undef, <2 x i32> zeroinitializer 181; CHECK-NEXT: [[TMP10:%.*]] = add nsw <2 x i32> [[TMP9]], [[TMP1]] 182; CHECK-NEXT: [[TMP11:%.*]] = extractelement <2 x i32> [[TMP10]], i32 0 183; CHECK-NEXT: [[TMP12:%.*]] = sext i32 [[TMP11]] to i64 184; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP12]] 185; CHECK-NEXT: [[T10:%.*]] = load i32, i32* [[ARRAYIDX10]], align 4 186; CHECK-NEXT: [[ADD11:%.*]] = add nsw i32 [[ADD6]], [[T10]] 187; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i32> [[TMP10]], i32 1 188; CHECK-NEXT: [[TMP14:%.*]] = sext i32 [[TMP13]] to i64 189; CHECK-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds i32, i32* [[G]], i64 [[TMP14]] 190; CHECK-NEXT: [[T12:%.*]] = load i32, i32* [[ARRAYIDX15]], align 4 191; CHECK-NEXT: [[TMP15:%.*]] = insertelement <2 x i32> undef, i32 [[TMP4]], i32 0 192; CHECK-NEXT: [[TMP16:%.*]] = insertelement <2 x i32> [[TMP15]], i32 [[ADD11]], i32 1 193; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x i32> <i32 1, i32 undef>, i32 [[T12]], i32 1 194; CHECK-NEXT: [[TMP18]] = add nsw <2 x i32> [[TMP16]], [[TMP17]] 195; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x i32> [[TMP18]], i32 0 196; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[TMP19]], [[N]] 197; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]] 198; 199entry: 200 %cmp31 = icmp sgt i32 %n, 0 201 br i1 %cmp31, label %for.body.preheader, label %for.cond.cleanup 202 203for.body.preheader: 204 br label %for.body 205 206for.cond.cleanup.loopexit: 207 br label %for.cond.cleanup 208 209for.cond.cleanup: 210 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add16, %for.cond.cleanup.loopexit ] 211 ret i32 %sum.0.lcssa 212 213for.body: 214 %indvars.iv = phi i32 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ] 215 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add16, %for.body ] 216 %t4 = shl nsw i32 %indvars.iv, 1 217 %t5 = add nsw i32 %t4, 0 218 %arrayidx = getelementptr inbounds i32, i32* %g, i32 %t5 219 %t6 = load i32, i32* %arrayidx, align 4 220 %add1 = add nsw i32 %t6, %sum.032 221 %t7 = add nsw i32 %t4, 1 222 %arrayidx5 = getelementptr inbounds i32, i32* %g, i32 %t7 223 %t8 = load i32, i32* %arrayidx5, align 4 224 %add6 = add nsw i32 %add1, %t8 225 %t9 = add nsw i32 %t4, %y 226 %arrayidx10 = getelementptr inbounds i32, i32* %g, i32 %t9 227 %t10 = load i32, i32* %arrayidx10, align 4 228 %add11 = add nsw i32 %add6, %t10 229 %t11 = add nsw i32 %t4, %z 230 %arrayidx15 = getelementptr inbounds i32, i32* %g, i32 %t11 231 %t12 = load i32, i32* %arrayidx15, align 4 232 %add16 = add nsw i32 %add11, %t12 233 %indvars.iv.next = add nuw nsw i32 %indvars.iv, 1 234 %exitcond = icmp eq i32 %indvars.iv.next , %n 235 br i1 %exitcond, label %for.cond.cleanup.loopexit, label %for.body 236} 237 238@global = internal global { i32* } zeroinitializer, align 8 239 240; Make sure we vectorize to maximize the load with when loading i16 and 241; extending it for compute operations. 242define void @test_i16_extend(i16* %p.1, i16* %p.2, i32 %idx.i32) { 243; CHECK-LABEL: @test_i16_extend( 244; CHECK-NEXT: [[P_0:%.*]] = load i32*, i32** getelementptr inbounds ({ i32* }, { i32* }* @global, i64 0, i32 0), align 8 245; CHECK-NEXT: [[IDX_0:%.*]] = zext i32 [[IDX_I32:%.*]] to i64 246; CHECK-NEXT: [[TMP53:%.*]] = getelementptr inbounds i16, i16* [[P_1:%.*]], i64 [[IDX_0]] 247; CHECK-NEXT: [[TMP56:%.*]] = getelementptr inbounds i16, i16* [[P_2:%.*]], i64 [[IDX_0]] 248; CHECK-NEXT: [[TMP1:%.*]] = bitcast i16* [[TMP53]] to <8 x i16>* 249; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i16>, <8 x i16>* [[TMP1]], align 2 250; CHECK-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[TMP2]] to <8 x i32> 251; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[TMP56]] to <8 x i16>* 252; CHECK-NEXT: [[TMP5:%.*]] = load <8 x i16>, <8 x i16>* [[TMP4]], align 2 253; CHECK-NEXT: [[TMP6:%.*]] = zext <8 x i16> [[TMP5]] to <8 x i32> 254; CHECK-NEXT: [[TMP7:%.*]] = sub nsw <8 x i32> [[TMP3]], [[TMP6]] 255; CHECK-NEXT: [[TMP8:%.*]] = extractelement <8 x i32> [[TMP7]], i32 0 256; CHECK-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64 257; CHECK-NEXT: [[TMP60:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP9]] 258; CHECK-NEXT: [[L_1:%.*]] = load i32, i32* [[TMP60]], align 4 259; CHECK-NEXT: [[TMP10:%.*]] = extractelement <8 x i32> [[TMP7]], i32 1 260; CHECK-NEXT: [[TMP11:%.*]] = sext i32 [[TMP10]] to i64 261; CHECK-NEXT: [[TMP71:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP11]] 262; CHECK-NEXT: [[L_2:%.*]] = load i32, i32* [[TMP71]], align 4 263; CHECK-NEXT: [[TMP12:%.*]] = extractelement <8 x i32> [[TMP7]], i32 2 264; CHECK-NEXT: [[TMP13:%.*]] = sext i32 [[TMP12]] to i64 265; CHECK-NEXT: [[TMP82:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP13]] 266; CHECK-NEXT: [[L_3:%.*]] = load i32, i32* [[TMP82]], align 4 267; CHECK-NEXT: [[TMP14:%.*]] = extractelement <8 x i32> [[TMP7]], i32 3 268; CHECK-NEXT: [[TMP15:%.*]] = sext i32 [[TMP14]] to i64 269; CHECK-NEXT: [[TMP93:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP15]] 270; CHECK-NEXT: [[L_4:%.*]] = load i32, i32* [[TMP93]], align 4 271; CHECK-NEXT: [[TMP16:%.*]] = extractelement <8 x i32> [[TMP7]], i32 4 272; CHECK-NEXT: [[TMP17:%.*]] = sext i32 [[TMP16]] to i64 273; CHECK-NEXT: [[TMP104:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP17]] 274; CHECK-NEXT: [[L_5:%.*]] = load i32, i32* [[TMP104]], align 4 275; CHECK-NEXT: [[TMP18:%.*]] = extractelement <8 x i32> [[TMP7]], i32 5 276; CHECK-NEXT: [[TMP19:%.*]] = sext i32 [[TMP18]] to i64 277; CHECK-NEXT: [[TMP115:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP19]] 278; CHECK-NEXT: [[L_6:%.*]] = load i32, i32* [[TMP115]], align 4 279; CHECK-NEXT: [[TMP20:%.*]] = extractelement <8 x i32> [[TMP7]], i32 6 280; CHECK-NEXT: [[TMP21:%.*]] = sext i32 [[TMP20]] to i64 281; CHECK-NEXT: [[TMP126:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP21]] 282; CHECK-NEXT: [[L_7:%.*]] = load i32, i32* [[TMP126]], align 4 283; CHECK-NEXT: [[TMP22:%.*]] = extractelement <8 x i32> [[TMP7]], i32 7 284; CHECK-NEXT: [[TMP23:%.*]] = sext i32 [[TMP22]] to i64 285; CHECK-NEXT: [[TMP137:%.*]] = getelementptr inbounds i32, i32* [[P_0]], i64 [[TMP23]] 286; CHECK-NEXT: [[L_8:%.*]] = load i32, i32* [[TMP137]], align 4 287; CHECK-NEXT: call void @use(i32 [[L_1]], i32 [[L_2]], i32 [[L_3]], i32 [[L_4]], i32 [[L_5]], i32 [[L_6]], i32 [[L_7]], i32 [[L_8]]) 288; CHECK-NEXT: ret void 289; 290 %g = getelementptr inbounds { i32*}, { i32 *}* @global, i64 0, i32 0 291 %p.0 = load i32*, i32** %g, align 8 292 293 %idx.0 = zext i32 %idx.i32 to i64 294 %idx.1 = add nsw i64 %idx.0, 1 295 %idx.2 = add nsw i64 %idx.0, 2 296 %idx.3 = add nsw i64 %idx.0, 3 297 %idx.4 = add nsw i64 %idx.0, 4 298 %idx.5 = add nsw i64 %idx.0, 5 299 %idx.6 = add nsw i64 %idx.0, 6 300 %idx.7 = add nsw i64 %idx.0, 7 301 302 %tmp53 = getelementptr inbounds i16, i16* %p.1, i64 %idx.0 303 %op1.l = load i16, i16* %tmp53, align 2 304 %op1.ext = zext i16 %op1.l to i64 305 %tmp56 = getelementptr inbounds i16, i16* %p.2, i64 %idx.0 306 %op2.l = load i16, i16* %tmp56, align 2 307 %op2.ext = zext i16 %op2.l to i64 308 %sub.1 = sub nsw i64 %op1.ext, %op2.ext 309 310 %tmp60 = getelementptr inbounds i32, i32* %p.0, i64 %sub.1 311 %l.1 = load i32, i32* %tmp60, align 4 312 313 %tmp64 = getelementptr inbounds i16, i16* %p.1, i64 %idx.1 314 %tmp65 = load i16, i16* %tmp64, align 2 315 %tmp66 = zext i16 %tmp65 to i64 316 %tmp67 = getelementptr inbounds i16, i16* %p.2, i64 %idx.1 317 %tmp68 = load i16, i16* %tmp67, align 2 318 %tmp69 = zext i16 %tmp68 to i64 319 %sub.2 = sub nsw i64 %tmp66, %tmp69 320 321 %tmp71 = getelementptr inbounds i32, i32* %p.0, i64 %sub.2 322 %l.2 = load i32, i32* %tmp71, align 4 323 324 %tmp75 = getelementptr inbounds i16, i16* %p.1, i64 %idx.2 325 %tmp76 = load i16, i16* %tmp75, align 2 326 %tmp77 = zext i16 %tmp76 to i64 327 %tmp78 = getelementptr inbounds i16, i16* %p.2, i64 %idx.2 328 %tmp79 = load i16, i16* %tmp78, align 2 329 %tmp80 = zext i16 %tmp79 to i64 330 %sub.3 = sub nsw i64 %tmp77, %tmp80 331 332 %tmp82 = getelementptr inbounds i32, i32* %p.0, i64 %sub.3 333 %l.3 = load i32, i32* %tmp82, align 4 334 335 %tmp86 = getelementptr inbounds i16, i16* %p.1, i64 %idx.3 336 %tmp87 = load i16, i16* %tmp86, align 2 337 %tmp88 = zext i16 %tmp87 to i64 338 %tmp89 = getelementptr inbounds i16, i16* %p.2, i64 %idx.3 339 %tmp90 = load i16, i16* %tmp89, align 2 340 %tmp91 = zext i16 %tmp90 to i64 341 %sub.4 = sub nsw i64 %tmp88, %tmp91 342 343 %tmp93 = getelementptr inbounds i32, i32* %p.0, i64 %sub.4 344 %l.4 = load i32, i32* %tmp93, align 4 345 346 %tmp97 = getelementptr inbounds i16, i16* %p.1, i64 %idx.4 347 %tmp98 = load i16, i16* %tmp97, align 2 348 %tmp99 = zext i16 %tmp98 to i64 349 %tmp100 = getelementptr inbounds i16, i16* %p.2, i64 %idx.4 350 %tmp101 = load i16, i16* %tmp100, align 2 351 %tmp102 = zext i16 %tmp101 to i64 352 %sub.5 = sub nsw i64 %tmp99, %tmp102 353 354 %tmp104 = getelementptr inbounds i32, i32* %p.0, i64 %sub.5 355 %l.5 = load i32, i32* %tmp104, align 4 356 357 %tmp108 = getelementptr inbounds i16, i16* %p.1, i64 %idx.5 358 %tmp109 = load i16, i16* %tmp108, align 2 359 %tmp110 = zext i16 %tmp109 to i64 360 %tmp111 = getelementptr inbounds i16, i16* %p.2, i64 %idx.5 361 %tmp112 = load i16, i16* %tmp111, align 2 362 %tmp113 = zext i16 %tmp112 to i64 363 %sub.6 = sub nsw i64 %tmp110, %tmp113 364 365 %tmp115 = getelementptr inbounds i32, i32* %p.0, i64 %sub.6 366 %l.6 = load i32, i32* %tmp115, align 4 367 368 %tmp119 = getelementptr inbounds i16, i16* %p.1, i64 %idx.6 369 %tmp120 = load i16, i16* %tmp119, align 2 370 %tmp121 = zext i16 %tmp120 to i64 371 %tmp122 = getelementptr inbounds i16, i16* %p.2, i64 %idx.6 372 %tmp123 = load i16, i16* %tmp122, align 2 373 %tmp124 = zext i16 %tmp123 to i64 374 %sub.7 = sub nsw i64 %tmp121, %tmp124 375 376 %tmp126 = getelementptr inbounds i32, i32* %p.0, i64 %sub.7 377 %l.7 = load i32, i32* %tmp126, align 4 378 379 %tmp130 = getelementptr inbounds i16, i16* %p.1, i64 %idx.7 380 %tmp131 = load i16, i16* %tmp130, align 2 381 %tmp132 = zext i16 %tmp131 to i64 382 %tmp133 = getelementptr inbounds i16, i16* %p.2, i64 %idx.7 383 %tmp134 = load i16, i16* %tmp133, align 2 384 %tmp135 = zext i16 %tmp134 to i64 385 %sub.8 = sub nsw i64 %tmp132, %tmp135 386 387 %tmp137 = getelementptr inbounds i32, i32* %p.0, i64 %sub.8 388 %l.8 = load i32, i32* %tmp137, align 4 389 390 call void @use(i32 %l.1, i32 %l.2, i32 %l.3, i32 %l.4, i32 %l.5, i32 %l.6, i32 %l.7, i32 %l.8) 391 ret void 392} 393 394declare void @use(i32, i32, i32, i32, i32, i32, i32, i32) 395