1 ; RUN: opt < %s -force-vector-width=4 -force-vector-interleave=2 -passes=loop-vectorize,instcombine -S | FileCheck %s
2 ; RUN: opt < %s -force-vector-width=4 -force-vector-interleave=2 -passes=loop-vectorize -S | FileCheck %s --check-prefix=NO-IC
4 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
6 ; CHECK-LABEL: @scalar_after_vectorization_0
9 ; CHECK: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
10 ; CHECK: [[OFFSET_IDX:%.+]] = or disjoint i64 %index, 1
11 ; CHECK: %[[T2:.+]] = add nuw nsw i64 [[OFFSET_IDX]], %tmp0
12 ; CHECK: %[[T3:.+]] = sub nsw i64 %[[T2]], %x
13 ; CHECK: %[[T4:.+]] = getelementptr inbounds i32, ptr %a, i64 %[[T3]]
14 ; CHECK: %[[T6:.+]] = getelementptr inbounds nuw i8, ptr %[[T4]], i64 16
15 ; CHECK: load <4 x i32>, ptr %[[T4]], align 4
16 ; CHECK: load <4 x i32>, ptr %[[T6]], align 4
17 ; CHECK: br {{.*}}, label %middle.block, label %vector.body
19 ; NO-IC-LABEL: @scalar_after_vectorization_0
22 ; NO-IC: %index = phi i64 [ 0, %vector.ph ], [ %index.next, %vector.body ]
23 ; NO-IC: [[OFFSET_IDX:%.+]] = add i64 1, %index
24 ; NO-IC: %[[T2:.+]] = add i64 [[OFFSET_IDX]], 0
25 ; NO-IC: %[[T4:.+]] = add nuw nsw i64 %[[T2]], %tmp0
26 ; NO-IC: %[[T6:.+]] = sub nsw i64 %[[T4]], %x
27 ; NO-IC: %[[T8:.+]] = getelementptr inbounds i32, ptr %a, i64 %[[T6]]
28 ; NO-IC: %[[T10:.+]] = getelementptr inbounds i32, ptr %[[T8]], i32 0
29 ; NO-IC: %[[T12:.+]] = getelementptr inbounds i32, ptr %[[T8]], i32 4
30 ; NO-IC: load <4 x i32>, ptr %[[T10]], align 4
31 ; NO-IC: load <4 x i32>, ptr %[[T12]], align 4
32 ; NO-IC: br {{.*}}, label %middle.block, label %vector.body
34 define void @scalar_after_vectorization_0(ptr noalias %a, ptr noalias %b, i64 %x, i64 %y) {
40 %i = phi i64 [ 1, %outer.ph ], [ %i.next, %inner.end ]
41 %tmp0 = mul nuw nsw i64 %i, %x
48 %j = phi i64 [ 1, %inner.ph ], [ %j.next, %inner.body ]
49 %tmp1 = add nuw nsw i64 %j, %tmp0
50 %tmp2 = sub nsw i64 %tmp1, %x
51 %tmp3 = getelementptr inbounds i32, ptr %a, i64 %tmp2
52 %tmp4 = load i32, ptr %tmp3, align 4
53 %tmp5 = getelementptr inbounds i32, ptr %b, i64 %tmp1
54 store i32 %tmp4, ptr %tmp5, align 4
55 %j.next = add i64 %j, 1
56 %cond.j = icmp slt i64 %j.next, %y
57 br i1 %cond.j, label %inner.body, label %inner.end
60 %i.next = add i64 %i, 1
61 %cond.i = icmp slt i64 %i.next, %y
62 br i1 %cond.i, label %outer.body, label %outer.end