1 ; RUN: opt -mcpu=skx -S -loop-vectorize -instcombine -force-vector-width=8 -force-vector-interleave=1 < %s | FileCheck %s
3 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
4 target triple = "x86_64-pc-linux-gnu"
6 ; Case1: With pragma predicate to force tail-folding.
7 ; All memory opertions are masked.
8 ;void fold_tail(int * restrict p, int * restrict q1, int * restrict q2, int guard) {
9 ; #pragma clang loop vectorize_predicate(enable)
10 ; for(int ix=0; ix < 1021; ++ix) {
12 ; p[ix] = q1[ix] + q2[ix];
17 ;CHECK-LABEL: @fold_tail
19 ;CHECK: call <8 x i32> @llvm.masked.load
20 ;CHECK: call <8 x i32> @llvm.masked.load
21 ;CHECK: call void @llvm.masked.store
23 ; Function Attrs: nofree norecurse nounwind uwtable
24 define dso_local void @fold_tail(i32* noalias nocapture %p, i32* noalias nocapture readonly %q1, i32* noalias nocapture readonly %q2,
25 i32 %guard) local_unnamed_addr #0 {
27 %0 = sext i32 %guard to i64
34 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
35 %cmp1 = icmp sgt i64 %indvars.iv, %0
36 br i1 %cmp1, label %if.then, label %for.inc
39 %arrayidx = getelementptr inbounds i32, i32* %q1, i64 %indvars.iv
40 %1 = load i32, i32* %arrayidx, align 4, !tbaa !2
41 %arrayidx3 = getelementptr inbounds i32, i32* %q2, i64 %indvars.iv
42 %2 = load i32, i32* %arrayidx3, align 4, !tbaa !2
43 %add = add nsw i32 %2, %1
44 %arrayidx5 = getelementptr inbounds i32, i32* %p, i64 %indvars.iv
45 store i32 %add, i32* %arrayidx5, align 4, !tbaa !2
49 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
50 %exitcond = icmp eq i64 %indvars.iv.next, 1021
51 br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !8
54 ; Case2: With pragma assume_safety only the store is masked.
55 ; void assume_safety(int * p, int * q1, int * q2, int guard) {
56 ; #pragma clang loop vectorize(assume_safety)
57 ; for(int ix=0; ix < 1021; ++ix) {
59 ; p[ix] = q1[ix] + q2[ix];
64 ;CHECK-LABEL: @assume_safety
66 ;CHECK-NOT: @llvm.masked.load
67 ;CHECK: call void @llvm.masked.store
69 ; Function Attrs: norecurse nounwind uwtable
70 define void @assume_safety(i32* nocapture, i32* nocapture readonly, i32* nocapture readonly, i32) local_unnamed_addr #0 {
71 %5 = sext i32 %3 to i64
78 %8 = phi i64 [ 0, %4 ], [ %18, %17 ]
79 %9 = icmp sgt i64 %8, %5
80 br i1 %9, label %10, label %17
83 %11 = getelementptr inbounds i32, i32* %1, i64 %8
84 %12 = load i32, i32* %11, align 4, !tbaa !2, !llvm.mem.parallel_loop_access !6
85 %13 = getelementptr inbounds i32, i32* %2, i64 %8
86 %14 = load i32, i32* %13, align 4, !tbaa !2, !llvm.mem.parallel_loop_access !6
87 %15 = add nsw i32 %14, %12
88 %16 = getelementptr inbounds i32, i32* %0, i64 %8
89 store i32 %15, i32* %16, align 4, !tbaa !2, !llvm.mem.parallel_loop_access !6
93 %18 = add nuw nsw i64 %8, 1
94 %19 = icmp eq i64 %18, 1021
95 br i1 %19, label %6, label %7, !llvm.loop !6
98 ; Case3: With pragma assume_safety and pragma predicate both the store and the
100 ; void fold_tail_and_assume_safety(int * p, int * q1, int * q2, int guard) {
101 ; #pragma clang loop vectorize(assume_safety) vectorize_predicate(enable)
102 ; for(int ix=0; ix < 1021; ++ix) {
104 ; p[ix] = q1[ix] + q2[ix];
109 ;CHECK-LABEL: @fold_tail_and_assume_safety
111 ;CHECK: call <8 x i32> @llvm.masked.load
112 ;CHECK: call <8 x i32> @llvm.masked.load
113 ;CHECK: call void @llvm.masked.store
115 ; Function Attrs: nofree norecurse nounwind uwtable
116 define dso_local void @fold_tail_and_assume_safety(i32* noalias nocapture %p, i32* noalias nocapture readonly %q1, i32* noalias nocapture readonly %q2,
117 i32 %guard) local_unnamed_addr #0 {
119 %0 = sext i32 %guard to i64
126 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.inc ]
127 %cmp1 = icmp sgt i64 %indvars.iv, %0
128 br i1 %cmp1, label %if.then, label %for.inc
131 %arrayidx = getelementptr inbounds i32, i32* %q1, i64 %indvars.iv
132 %1 = load i32, i32* %arrayidx, align 4, !tbaa !2, !llvm.access.group !10
133 %arrayidx3 = getelementptr inbounds i32, i32* %q2, i64 %indvars.iv
134 %2 = load i32, i32* %arrayidx3, align 4, !tbaa !2, !llvm.access.group !10
135 %add = add nsw i32 %2, %1
136 %arrayidx5 = getelementptr inbounds i32, i32* %p, i64 %indvars.iv
137 store i32 %add, i32* %arrayidx5, align 4, !tbaa !2, !llvm.access.group !10
141 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
142 %exitcond = icmp eq i64 %indvars.iv.next, 1021
143 br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !11
146 attributes #0 = { norecurse nounwind uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" "unsafe-fp-math"="false" "use-soft-float"="false" }
148 !llvm.module.flags = !{!0}
151 !0 = !{i32 1, !"wchar_size", i32 4}
152 !1 = !{!"clang version 6.0.0-1ubuntu2 (tags/RELEASE_600/final)"}
153 !2 = !{!3, !3, i64 0}
154 !3 = !{!"int", !4, i64 0}
155 !4 = !{!"omnipotent char", !5, i64 0}
156 !5 = !{!"Simple C/C++ TBAA"}
157 !6 = distinct !{!6, !7}
158 !7 = !{!"llvm.loop.vectorize.enable", i1 true}
160 !8 = distinct !{!8, !9}
161 !9 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}
164 !11 = distinct !{!11, !12, !13}
165 !12 = !{!"llvm.loop.parallel_accesses", !10}
166 !13 = !{!"llvm.loop.vectorize.predicate.enable", i1 true}