1 ; RUN: opt < %s -mattr=avx -force-vector-width=2 -force-vector-interleave=1 -loop-vectorize -simplifycfg -S | FileCheck %s
2 ; RUN: opt -mcpu=skylake-avx512 -S -force-vector-width=8 -force-vector-interleave=1 -loop-vectorize < %s | FileCheck %s --check-prefix=SINK-GATHER
4 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"
5 target triple = "x86_64-apple-macosx10.8.0"
7 ; CHECK-LABEL: predicated_sdiv_masked_load
9 ; This test ensures that we don't scalarize the predicated load. Since the load
10 ; can be vectorized with predication, scalarizing it would cause its pointer
11 ; operand to become non-uniform.
14 ; CHECK: %wide.masked.load = call <2 x i32> @llvm.masked.load.v2i32.p0v2i32
15 ; CHECK: br i1 {{.*}}, label %[[IF0:.+]], label %[[CONT0:.+]]
17 ; CHECK: %[[T0:.+]] = extractelement <2 x i32> %wide.masked.load, i32 0
18 ; CHECK: %[[T1:.+]] = sdiv i32 %[[T0]], %x
19 ; CHECK: %[[T2:.+]] = insertelement <2 x i32> undef, i32 %[[T1]], i32 0
20 ; CHECK: br label %[[CONT0]]
22 ; CHECK: %[[T3:.+]] = phi <2 x i32> [ undef, %vector.body ], [ %[[T2]], %[[IF0]] ]
23 ; CHECK: br i1 {{.*}}, label %[[IF1:.+]], label %[[CONT1:.+]]
25 ; CHECK: %[[T4:.+]] = extractelement <2 x i32> %wide.masked.load, i32 1
26 ; CHECK: %[[T5:.+]] = sdiv i32 %[[T4]], %x
27 ; CHECK: %[[T6:.+]] = insertelement <2 x i32> %[[T3]], i32 %[[T5]], i32 1
28 ; CHECK: br label %[[CONT1]]
30 ; CHECK: phi <2 x i32> [ %[[T3]], %[[CONT0]] ], [ %[[T6]], %[[IF1]] ]
31 ; CHECK: br i1 {{.*}}, label %middle.block, label %vector.body
33 define i32 @predicated_sdiv_masked_load(i32* %a, i32* %b, i32 %x, i1 %c) {
38 %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
39 %r = phi i32 [ 0, %entry ], [ %tmp7, %for.inc ]
40 %tmp0 = getelementptr inbounds i32, i32* %a, i64 %i
41 %tmp1 = load i32, i32* %tmp0, align 4
42 br i1 %c, label %if.then, label %for.inc
45 %tmp2 = getelementptr inbounds i32, i32* %b, i64 %i
46 %tmp3 = load i32, i32* %tmp2, align 4
47 %tmp4 = sdiv i32 %tmp3, %x
48 %tmp5 = add nsw i32 %tmp4, %tmp1
52 %tmp6 = phi i32 [ %tmp1, %for.body ], [ %tmp5, %if.then]
53 %tmp7 = add i32 %r, %tmp6
54 %i.next = add nuw nsw i64 %i, 1
55 %cond = icmp eq i64 %i.next, 10000
56 br i1 %cond, label %for.end, label %for.body
59 %tmp8 = phi i32 [ %tmp7, %for.inc ]
63 ; This test ensures that a load, which would have been widened otherwise is
64 ; instead scalarized if Cost-Model so decided as part of its
65 ; sink-scalar-operands optimization for predicated instructions.
67 ; SINK-GATHER: vector.body:
68 ; SINK-GATHER: pred.udiv.if:
69 ; SINK-GATHER: %[[T0:.+]] = load i32, i32* %{{.*}}, align 4
70 ; SINK-GATHER: %{{.*}} = udiv i32 %[[T0]], %{{.*}}
71 ; SINK-GATHER: pred.udiv.continue:
72 define i32 @scalarize_and_sink_gather(i32* %a, i1 %c, i32 %x, i64 %n) {
77 %i = phi i64 [ 0, %entry ], [ %i.next, %for.inc ]
78 %r = phi i32 [ 0, %entry ], [ %tmp6, %for.inc ]
80 br i1 %c, label %if.then, label %for.inc
83 %tmp0 = getelementptr inbounds i32, i32* %a, i64 %i7
84 %tmp2 = load i32, i32* %tmp0, align 4
85 %tmp4 = udiv i32 %tmp2, %x
89 %tmp5 = phi i32 [ %x, %for.body ], [ %tmp4, %if.then]
90 %tmp6 = add i32 %r, %tmp5
91 %i.next = add nuw nsw i64 %i, 1
92 %cond = icmp slt i64 %i.next, %n
93 br i1 %cond, label %for.body, label %for.end
96 %tmp7 = phi i32 [ %tmp6, %for.inc ]