1; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -instcombine -S | FileCheck %s 2 3target 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-f80:128:128-n8:16:32:64-S128" 4 5%struct.coordinate = type { i32, i32 } 6 7; Make sure that we don't generate a wide load when accessing the struct. 8; struct coordinate { 9; int x; 10; int y; 11; }; 12; 13; 14; int foo(struct coordinate *A, int n) { 15; 16; int sum = 0; 17; for (int i = 0; i < n; ++i) 18; sum += A[i].x; 19; 20; return sum; 21; } 22 23;CHECK-LABEL: @foo( 24;CHECK-NOT: load <4 x i32> 25;CHECK: ret 26define i32 @foo(%struct.coordinate* nocapture %A, i32 %n) nounwind uwtable readonly ssp { 27entry: 28 %cmp4 = icmp sgt i32 %n, 0 29 br i1 %cmp4, label %for.body, label %for.end 30 31for.body: ; preds = %entry, %for.body 32 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 33 %sum.05 = phi i32 [ %add, %for.body ], [ 0, %entry ] 34 %x = getelementptr inbounds %struct.coordinate, %struct.coordinate* %A, i64 %indvars.iv, i32 0 35 %0 = load i32, i32* %x, align 4 36 %add = add nsw i32 %0, %sum.05 37 %indvars.iv.next = add i64 %indvars.iv, 1 38 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 39 %exitcond = icmp eq i32 %lftr.wideiv, %n 40 br i1 %exitcond, label %for.end, label %for.body 41 42for.end: ; preds = %for.body, %entry 43 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ] 44 ret i32 %sum.0.lcssa 45} 46 47%struct.lit = type { i32 } 48 49; Verify that we still vectorize the access if the struct has the same size as 50; the loaded element. 51; struct lit { 52; int x; 53; }; 54; 55; 56; int bar(struct lit *A, int n) { 57; 58; int sum = 0; 59; for (int i = 0; i < n; ++i) 60; sum += A[i].x; 61; 62; return sum; 63; } 64 65;CHECK-LABEL: @bar( 66;CHECK: load <4 x i32> 67;CHECK: ret 68define i32 @bar(%struct.lit* nocapture %A, i32 %n) nounwind uwtable readonly ssp { 69entry: 70 %cmp4 = icmp sgt i32 %n, 0 71 br i1 %cmp4, label %for.body, label %for.end 72 73for.body: ; preds = %entry, %for.body 74 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 75 %sum.05 = phi i32 [ %add, %for.body ], [ 0, %entry ] 76 %x = getelementptr inbounds %struct.lit, %struct.lit* %A, i64 %indvars.iv, i32 0 77 %0 = load i32, i32* %x, align 4 78 %add = add nsw i32 %0, %sum.05 79 %indvars.iv.next = add i64 %indvars.iv, 1 80 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 81 %exitcond = icmp eq i32 %lftr.wideiv, %n 82 br i1 %exitcond, label %for.end, label %for.body 83 84for.end: ; preds = %for.body, %entry 85 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ] 86 ret i32 %sum.0.lcssa 87} 88