1 ; RUN: opt -loop-vectorize -force-vector-width=4 -S < %s | FileCheck %s
3 ; This is the test case from PR26314.
4 ; When we were retrying dependence checking with memchecks only,
5 ; the loop-invariant access in the inner loop was incorrectly determined to be wrapping
6 ; because it was not strided in the inner loop.
7 ; Improved wrapping detection allows vectorization in the following case.
16 ; void slow_function (s* const obj, int z) {
17 ; for (int j=0; j<Z; j++) {
18 ; for (int k=0; k<z; k++) {
19 ; int x = obj->v1[k] + obj->v2[j];
27 ; CHECK: <4 x i32>, <4 x i32>
28 ; CHECK: !{!"llvm.loop.isvectorized", i32 1}
30 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
32 %struct.s = type { [32 x i32], [32 x i32], [32 x [32 x i32]] }
34 define void @Test(%struct.s* nocapture %obj, i64 %z) #0 {
35 br label %.outer.preheader
39 %i = phi i64 [ 0, %0 ], [ %i.next, %.outer ]
40 %1 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 1, i64 %i
47 %i.next = add nuw nsw i64 %i, 1
48 %exitcond.outer = icmp eq i64 %i.next, 32
49 br i1 %exitcond.outer, label %.exit, label %.outer.preheader
52 %j = phi i64 [ 0, %.outer.preheader ], [ %j.next, %.inner ]
53 %2 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 0, i64 %j
54 %3 = load i32, i32* %2
55 %4 = load i32, i32* %1
56 %5 = add nsw i32 %4, %3
57 %6 = getelementptr inbounds %struct.s, %struct.s* %obj, i64 0, i32 2, i64 %i, i64 %j
58 %7 = load i32, i32* %6
59 %8 = add nsw i32 %5, %7
61 %j.next = add nuw nsw i64 %j, 1
62 %exitcond.inner = icmp eq i64 %j.next, %z
63 br i1 %exitcond.inner, label %.outer, label %.inner