1 ; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -instcombine -S | FileCheck %s 2 3 target 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 target triple = "x86_64-apple-macosx10.9.0" 5 6 %struct.coordinate = type { i32, i32 } 7 8 ; Make sure that we don't generate a wide load when accessing the struct. 9 ; struct coordinate { 10 ; int x; 11 ; int y; 12 ; }; 13 ; 14 ; 15 ; int foo(struct coordinate *A, int n) { 16 ; 17 ; int sum = 0; 18 ; for (int i = 0; i < n; ++i) 19 ; sum += A[i].x; 20 ; 21 ; return sum; 22 ; } 23 24 ;CHECK-LABEL: @foo( 25 ;CHECK-NOT: load <4 x i32> 26 ;CHECK: ret 27 define i32 @foo(%struct.coordinate* nocapture %A, i32 %n) nounwind uwtable readonly ssp { 28 entry: 29 %cmp4 = icmp sgt i32 %n, 0 30 br i1 %cmp4, label %for.body, label %for.end 31 32 for.body: ; preds = %entry, %for.body 33 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 34 %sum.05 = phi i32 [ %add, %for.body ], [ 0, %entry ] 35 %x = getelementptr inbounds %struct.coordinate, %struct.coordinate* %A, i64 %indvars.iv, i32 0 36 %0 = load i32, i32* %x, align 4 37 %add = add nsw i32 %0, %sum.05 38 %indvars.iv.next = add i64 %indvars.iv, 1 39 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 40 %exitcond = icmp eq i32 %lftr.wideiv, %n 41 br i1 %exitcond, label %for.end, label %for.body 42 43 for.end: ; preds = %for.body, %entry 44 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ] 45 ret i32 %sum.0.lcssa 46 } 47 48 %struct.lit = type { i32 } 49 50 ; Verify that we still vectorize the access if the struct has the same size as 51 ; the loaded element. 52 ; struct lit { 53 ; int x; 54 ; }; 55 ; 56 ; 57 ; int bar(struct lit *A, int n) { 58 ; 59 ; int sum = 0; 60 ; for (int i = 0; i < n; ++i) 61 ; sum += A[i].x; 62 ; 63 ; return sum; 64 ; } 65 66 ;CHECK-LABEL: @bar( 67 ;CHECK: load <4 x i32> 68 ;CHECK: ret 69 define i32 @bar(%struct.lit* nocapture %A, i32 %n) nounwind uwtable readonly ssp { 70 entry: 71 %cmp4 = icmp sgt i32 %n, 0 72 br i1 %cmp4, label %for.body, label %for.end 73 74 for.body: ; preds = %entry, %for.body 75 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ] 76 %sum.05 = phi i32 [ %add, %for.body ], [ 0, %entry ] 77 %x = getelementptr inbounds %struct.lit, %struct.lit* %A, i64 %indvars.iv, i32 0 78 %0 = load i32, i32* %x, align 4 79 %add = add nsw i32 %0, %sum.05 80 %indvars.iv.next = add i64 %indvars.iv, 1 81 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 82 %exitcond = icmp eq i32 %lftr.wideiv, %n 83 br i1 %exitcond, label %for.end, label %for.body 84 85 for.end: ; preds = %for.body, %entry 86 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add, %for.body ] 87 ret i32 %sum.0.lcssa 88 } 89
