1; RUN: opt < %s -mattr=avx -force-vector-width=2 -force-vector-interleave=1 -loop-vectorize -simplifycfg -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"
4target triple = "x86_64-apple-macosx10.8.0"
5
6; CHECK-LABEL: predicated_sdiv_masked_load
7;
8; This test ensures that we don't scalarize the predicated load. Since the load
9; can be vectorized with predication, scalarizing it would cause its pointer
10; operand to become non-uniform.
11;
12; CHECK: vector.body:
13; CHECK:   %wide.masked.load = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32
14; CHECK:   br i1 {{.*}}, label %[[IF0:.+]], label %[[CONT0:.+]]
15; CHECK: [[IF0]]:
16; CHECK:   %[[T0:.+]] = extractelement <2 x i32> %wide.masked.load, i32 0
17; CHECK:   %[[T1:.+]] = sdiv i32 %[[T0]], %x
18; CHECK:   %[[T2:.+]] = insertelement <2 x i32> undef, i32 %[[T1]], i32 0
19; CHECK:   br label %[[CONT0]]
20; CHECK: [[CONT0]]:
21; CHECK:   %[[T3:.+]] = phi <2 x i32> [ undef, %vector.body ], [ %[[T2]], %[[IF0]] ]
22; CHECK:   br i1 {{.*}}, label %[[IF1:.+]], label %[[CONT1:.+]]
23; CHECK: [[IF1]]:
24; CHECK:   %[[T4:.+]] = extractelement <2 x i32> %wide.masked.load, i32 1
25; CHECK:   %[[T5:.+]] = sdiv i32 %[[T4]], %x
26; CHECK:   %[[T6:.+]] = insertelement <2 x i32> %[[T3]], i32 %[[T5]], i32 1
27; CHECK:   br label %[[CONT1]]
28; CHECK: [[CONT1]]:
29; CHECK:   phi <2 x i32> [ %[[T3]], %[[CONT0]] ], [ %[[T6]], %[[IF1]] ]
30; CHECK:   br i1 {{.*}}, label %middle.block, label %vector.body
31
32define i32 @predicated_sdiv_masked_load(i32* %a, i32* %b, i32 %x, i1 %c) {
33entry:
34  br label %for.body
35
36for.body:
37  %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
38  %r = phi i32 [ 0, %entry ], [ %tmp7, %for.inc ]
39  %tmp0 = getelementptr inbounds i32, i32* %a, i64 %i
40  %tmp1 = load i32, i32* %tmp0, align 4
41  br i1 %c, label %if.then, label %for.inc
42
43if.then:
44  %tmp2 = getelementptr inbounds i32, i32* %b, i64 %i
45  %tmp3 = load i32, i32* %tmp2, align 4
46  %tmp4 = sdiv i32 %tmp3, %x
47  %tmp5 = add nsw i32 %tmp4, %tmp1
48  br label %for.inc
49
50for.inc:
51  %tmp6 = phi i32 [ %tmp1, %for.body ], [ %tmp5, %if.then]
52  %tmp7 = add i32 %r, %tmp6
53  %i.next = add nuw nsw i64 %i, 1
54  %cond = icmp eq i64 %i.next, 10000
55  br i1 %cond, label %for.end, label %for.body
56
57for.end:
58  %tmp8 = phi i32 [ %tmp7, %for.inc ]
59  ret i32 %tmp8
60}
61