1; RUN: opt -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck %s
2
3@dst = external global [32 x i16], align 1
4
5define void @blend_uniform_iv_trunc(i1 %c) {
6; CHECK-LABEL: @blend_uniform_iv_trunc(
7; CHECK:       vector.ph:
8; CHECK-NEXT:    [[MASK0:%.*]] = insertelement <4 x i1> poison, i1 %c, i32 0
9; CHECK-NEXT:    [[MASK1:%.*]] = shufflevector <4 x i1> [[MASK0]], <4 x i1> poison, <4 x i32> zeroinitializer
10
11; CHECK:       vector.body:
12; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
13; CHECK-NEXT:    [[TMP1:%.*]] = trunc i64 [[INDEX]] to i16
14; CHECK-NEXT:    [[TMP2:%.*]] = add i16 [[TMP1]], 0
15; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i16> poison, i16 [[TMP2]], i32 0
16; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT1]], <4 x i16> poison, <4 x i32> zeroinitializer
17; CHECK-NEXT:    [[TMP3:%.*]] = xor <4 x i1> [[MASK1]], <i1 true, i1 true, i1 true, i1 true>
18; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[MASK1]], <4 x i16> [[BROADCAST_SPLAT2]], <4 x i16> undef
19; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <4 x i16> [[PREDPHI]], i32 0
20; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i16 [[TMP4]]
21; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds i16, i16* [[TMP5]], i32 0
22; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i16* [[TMP6]] to <4 x i16>*
23; CHECK-NEXT:    store <4 x i16> zeroinitializer, <4 x i16>* [[TMP7]], align 2
24; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
25; CHECK-NEXT:    [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
26; CHECK-NEXT:    br i1 [[TMP8]], label %middle.block, label %vector.body
27;
28entry:
29  br label %loop.header
30
31loop.header:                                      ; preds = %loop.latch, %entry
32  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
33  %iv.trunc.2 = trunc i64 %iv to i16
34  br i1 %c, label %loop.next, label %loop.latch
35
36loop.next:                                        ; preds = %loop.header
37  br label %loop.latch
38
39loop.latch:                                       ; preds = %loop.next, %loop.header
40  %blend = phi i16 [ undef, %loop.header ], [ %iv.trunc.2, %loop.next ]
41  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i16 %blend
42  store i16 0, i16* %dst.ptr
43  %iv.next = add nuw nsw i64 %iv, 1
44  %cmp439 = icmp ult i64 %iv, 31
45  br i1 %cmp439, label %loop.header, label %exit
46
47exit:                                             ; preds = %loop.latch
48  ret void
49}
50
51define void @blend_uniform_iv(i1 %c) {
52; CHECK-LABEL: @blend_uniform_iv(
53; CHECK:       vector.ph:
54; CHECK-NEXT:    [[MASK0:%.*]] = insertelement <4 x i1> poison, i1 %c, i32 0
55; CHECK-NEXT:    [[MASK1:%.*]] = shufflevector <4 x i1> [[MASK0]], <4 x i1> poison, <4 x i32> zeroinitializer
56
57; CHECK:       vector.body:
58; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
59; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
60; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i64> poison, i64 [[TMP0]], i32 0
61; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT1]], <4 x i64> poison, <4 x i32> zeroinitializer
62; CHECK-NEXT:    [[TMP1:%.*]] = xor <4 x i1> [[MASK1]], <i1 true, i1 true, i1 true, i1 true>
63; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[MASK1]], <4 x i64> [[BROADCAST_SPLAT2]], <4 x i64> undef
64; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x i64> [[PREDPHI]], i32 0
65; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP2]]
66; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i16, i16* [[TMP3]], i32 0
67; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16* [[TMP4]] to <4 x i16>*
68; CHECK-NEXT:    store <4 x i16> zeroinitializer, <4 x i16>* [[TMP5]], align 2
69; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
70; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
71; CHECK-NEXT:    br i1 [[TMP6]], label %middle.block, label %vector.body
72;
73entry:
74  br label %loop.header
75
76loop.header:                                      ; preds = %loop.latch, %entry
77  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
78  br i1 %c, label %loop.next, label %loop.latch
79
80loop.next:                                        ; preds = %loop.header
81  br label %loop.latch
82
83loop.latch:                                       ; preds = %loop.next, %loop.header
84  %blend = phi i64 [ undef, %loop.header ], [ %iv, %loop.next ]
85  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %blend
86  store i16 0, i16* %dst.ptr
87  %iv.next = add nuw nsw i64 %iv, 1
88  %cmp439 = icmp ult i64 %iv, 31
89  br i1 %cmp439, label %loop.header, label %exit
90
91exit:                                             ; preds = %loop.latch
92  ret void
93}
94
95define void @blend_chain_iv(i1 %c) {
96; CHECK-LABEL: @blend_chain_iv(
97; CHECK:       vector.ph:
98; CHECK-NEXT:    [[MASK0:%.*]] = insertelement <4 x i1> poison, i1 %c, i32 0
99; CHECK-NEXT:    [[MASK1:%.*]] = shufflevector <4 x i1> [[MASK0]], <4 x i1> poison, <4 x i32> zeroinitializer
100
101; CHECK:       vector.body:
102; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %vector.ph ], [ [[INDEX_NEXT:%.*]], %vector.body ]
103; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %vector.ph ], [ [[VEC_IND_NEXT:%.*]], %vector.body ]
104; CHECK-NEXT:    [[TMP4:%.*]] = xor <4 x i1> [[MASK1]], <i1 true, i1 true, i1 true, i1 true>
105; CHECK-NEXT:    [[TMP5:%.*]] = select <4 x i1> [[MASK1]], <4 x i1> [[TMP4]], <4 x i1> zeroinitializer
106; CHECK-NEXT:    [[TMP6:%.*]] = select <4 x i1> [[MASK1]], <4 x i1> [[MASK1]], <4 x i1> zeroinitializer
107; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP6]], <4 x i64> [[VEC_IND]], <4 x i64> undef
108; CHECK-NEXT:    [[TMP7:%.*]] = xor <4 x i1> [[MASK1]], <i1 true, i1 true, i1 true, i1 true>
109; CHECK-NEXT:    [[TMP8:%.*]] = or <4 x i1> [[TMP6]], [[TMP5]]
110; CHECK-NEXT:    [[PREDPHI1:%.*]] = select <4 x i1> [[TMP8]], <4 x i64> [[PREDPHI]], <4 x i64> undef
111; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 0
112; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP9]]
113; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 1
114; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP11]]
115; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 2
116; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP13]]
117; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 3
118; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP15]]
119; CHECK-NEXT:    store i16 0, i16* [[TMP10]], align 2
120; CHECK-NEXT:    store i16 0, i16* [[TMP12]], align 2
121; CHECK-NEXT:    store i16 0, i16* [[TMP14]], align 2
122; CHECK-NEXT:    store i16 0, i16* [[TMP16]], align 2
123; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
124; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], <i64 4, i64 4, i64 4, i64 4>
125; CHECK-NEXT:    [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
126; CHECK-NEXT:    br i1 [[TMP17]], label %middle.block, label %vector.body
127;
128entry:
129  br label %loop.header
130
131loop.header:                                      ; preds = %loop.latch, %entry
132  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
133  br i1 %c, label %loop.next, label %loop.latch
134
135loop.next:                                        ; preds = %loop.header
136  br i1 %c, label %loop.next.2, label %loop.next.3
137
138loop.next.2:
139  br label %loop.next.3
140
141loop.next.3:
142  %blend.1 = phi i64 [ undef, %loop.next ], [ %iv, %loop.next.2 ]
143  br label %loop.latch
144
145loop.latch:                                       ; preds = %loop.next, %loop.header
146  %blend = phi i64 [ undef, %loop.header ], [ %blend.1, %loop.next.3 ]
147  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %blend
148  store i16 0, i16* %dst.ptr
149  %iv.next = add nuw nsw i64 %iv, 1
150  %cmp439 = icmp ult i64 %iv, 31
151  br i1 %cmp439, label %loop.header, label %exit
152
153exit:                                             ; preds = %loop.latch
154  ret void
155}
156