1 ; int A[1024], B[1024];
3 ; void foo(int iCount, int c, int jCount)
8 ; #pragma clang loop vectorize(enable) vectorize_width(4)
9 ; for (i = 0; i < iCount; i++) {
11 ; for (j = 0; j < jCount; j++) {
16 ; RUN: opt -S -passes=loop-vectorize -enable-vplan-native-path < %s | FileCheck %s
17 ; CHECK: %[[ZeroTripChk:.*]] = icmp sgt i32 %jCount, 0
18 ; CHECK-LABEL: vector.ph:
19 ; CHECK: %[[CVal0:.*]] = insertelement <4 x i32> poison, i32 %c, i64 0
20 ; CHECK-NEXT: %[[CSplat:.*]] = shufflevector <4 x i32> %[[CVal0]], <4 x i32> poison, <4 x i32> zeroinitializer
22 ; CHECK-LABEL: vector.body:
23 ; CHECK: %[[Ind:.*]] = phi i64 [ 0, %vector.ph ], [ %[[IndNext:.*]], %[[ForInc:.*]] ]
24 ; CHECK: %[[VecInd:.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %vector.ph ], [ %[[VecIndNext:.*]], %[[ForInc]] ]
25 ; CHECK: %[[AAddr:.*]] = getelementptr inbounds [1024 x i32], ptr @A, i64 0, <4 x i64> %[[VecInd]]
26 ; CHECK: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> %[[CSplat]], <4 x ptr> %[[AAddr]], i32 4, <4 x i1> splat (i1 true))
27 ; CHECK: br i1 %[[ZeroTripChk]], label %[[InnerForPh:.*]], label %[[OuterInc:.*]]
29 ; CHECK: [[InnerForPh]]:
30 ; CHECK: %[[WideAVal:.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %[[AAddr]], i32 4, <4 x i1> splat (i1 true), <4 x i32> poison)
31 ; CHECK: %[[VecIndTr:.*]] = trunc <4 x i64> %[[VecInd]] to <4 x i32>
32 ; CHECK: br label %[[InnerForBody:.*]]
34 ; CHECK: [[InnerForBody]]:
35 ; CHECK: %[[InnerInd:.*]] = phi <4 x i64> [ zeroinitializer, %[[InnerForPh]] ], [ %[[InnerIndNext:.*]], %[[InnerForBody]] ]
36 ; CHECK: %[[AccumPhi:.*]] = phi <4 x i32> [ %[[WideAVal]], %[[InnerForPh]] ], [ %[[AccumPhiNext:.*]], %[[InnerForBody]] ]
37 ; CHECK: %[[BAddr:.*]] = getelementptr inbounds [1024 x i32], ptr @B, i64 0, <4 x i64> %[[InnerInd]]
38 ; CHECK: %[[WideBVal:.*]] = call <4 x i32> @llvm.masked.gather.v4i32.v4p0(<4 x ptr> %[[BAddr]], i32 4, <4 x i1> splat (i1 true), <4 x i32> poison)
39 ; CHECK: %[[Add1:.*]] = add nsw <4 x i32> %[[WideBVal]], %[[VecIndTr]]
40 ; CHECK: %[[AccumPhiNext]] = add nsw <4 x i32> %[[Add1]], %[[AccumPhi]]
41 ; CHECK: %[[InnerIndNext]] = add nuw nsw <4 x i64> %[[InnerInd]], splat (i64 1)
42 ; CHECK: %[[InnerVecCond:.*]] = icmp eq <4 x i64> %[[InnerIndNext]], {{.*}}
43 ; CHECK: %[[InnerCond:.+]] = extractelement <4 x i1> %[[InnerVecCond]], i32 0
44 ; CHECK: br i1 %[[InnerCond]], label %[[InnerCrit:.*]], label %[[InnerForBody]]
46 ; CHECK: [[InnerCrit]]:
47 ; CHECK: %[[StorePhi:.*]] = phi <4 x i32> [ %[[AccumPhiNext]], %[[InnerForBody]] ]
48 ; CHECK: call void @llvm.masked.scatter.v4i32.v4p0(<4 x i32> %[[StorePhi]], <4 x ptr> %[[AAddr]], i32 4, <4 x i1> splat (i1 true))
49 ; CHECK: br label %[[ForInc]]
52 ; CHECK: %[[IndNext]] = add nuw i64 %[[Ind]], 4
53 ; CHECK: %[[VecIndNext]] = add <4 x i64> %[[VecInd]], splat (i64 4)
54 ; CHECK: %[[Cmp:.*]] = icmp eq i64 %[[IndNext]], {{.*}}
55 ; CHECK: br i1 %[[Cmp]], label %middle.block, label %vector.body
57 @A = common global [1024 x i32] zeroinitializer, align 16
58 @B = common global [1024 x i32] zeroinitializer, align 16
60 ; Function Attrs: norecurse nounwind uwtable
61 define void @foo(i32 %iCount, i32 %c, i32 %jCount) {
63 %cmp22 = icmp sgt i32 %iCount, 0
64 br i1 %cmp22, label %for.body.lr.ph, label %for.end11
66 for.body.lr.ph: ; preds = %entry
67 %cmp220 = icmp sgt i32 %jCount, 0
68 %wide.trip.count = zext i32 %jCount to i64
69 %wide.trip.count27 = zext i32 %iCount to i64
72 for.body: ; preds = %for.inc9, %for.body.lr.ph
73 %indvars.iv25 = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next26, %for.inc9 ]
74 %arrayidx = getelementptr inbounds [1024 x i32], ptr @A, i64 0, i64 %indvars.iv25
75 store i32 %c, ptr %arrayidx, align 4
76 br i1 %cmp220, label %for.body3.lr.ph, label %for.inc9
78 for.body3.lr.ph: ; preds = %for.body
79 %arrayidx.promoted = load i32, ptr %arrayidx, align 4
80 %0 = trunc i64 %indvars.iv25 to i32
83 for.body3: ; preds = %for.body3, %for.body3.lr.ph
84 %indvars.iv = phi i64 [ 0, %for.body3.lr.ph ], [ %indvars.iv.next, %for.body3 ]
85 %1 = phi i32 [ %arrayidx.promoted, %for.body3.lr.ph ], [ %add8, %for.body3 ]
86 %arrayidx5 = getelementptr inbounds [1024 x i32], ptr @B, i64 0, i64 %indvars.iv
87 %2 = load i32, ptr %arrayidx5, align 4
88 %add = add nsw i32 %2, %0
89 %add8 = add nsw i32 %add, %1
90 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
91 %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
92 br i1 %exitcond, label %for.cond1.for.inc9_crit_edge, label %for.body3
94 for.cond1.for.inc9_crit_edge: ; preds = %for.body3
95 store i32 %add8, ptr %arrayidx, align 4
98 for.inc9: ; preds = %for.cond1.for.inc9_crit_edge, %for.body
99 %indvars.iv.next26 = add nuw nsw i64 %indvars.iv25, 1
100 %exitcond28 = icmp eq i64 %indvars.iv.next26, %wide.trip.count27
101 br i1 %exitcond28, label %for.end11, label %for.body, !llvm.loop !1
103 for.end11: ; preds = %for.inc9, %entry
107 !1 = distinct !{!1, !2, !3}
108 !2 = !{!"llvm.loop.vectorize.width", i32 4}
109 !3 = !{!"llvm.loop.vectorize.enable", i1 true}