1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-vectorize -force-vector-width=2 -S %s | FileCheck %s
3
4@src = external global [32 x i16], align 1
5@dst = external global [32 x i16], align 1
6
7; The load in the loop does not need predication, because the accessed memory
8; is de-referenceable for all loop iterations.
9define void @single_incoming_phi_no_blend_mask(i64 %a, i64 %b) {
10; CHECK-LABEL: @single_incoming_phi_no_blend_mask(
11; CHECK-NEXT:  entry:
12; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
13; CHECK:       vector.ph:
14; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
15; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
16; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
17; CHECK:       vector.body:
18; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
19; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
20; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[INDEX]] to i16
21; CHECK-NEXT:    [[TMP3:%.*]] = add i16 [[TMP2]], 0
22; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
23; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP3]]
24; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i16, i16* [[TMP4]], i32 0
25; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i16* [[TMP5]] to <2 x i16>*
26; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP6]], align 1
27; CHECK-NEXT:    [[TMP7:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
28; CHECK-NEXT:    [[TMP8:%.*]] = xor <2 x i1> [[TMP7]], <i1 true, i1 true>
29; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP7]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[WIDE_LOAD]]
30; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
31; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds i16, i16* [[TMP9]], i32 0
32; CHECK-NEXT:    [[TMP11:%.*]] = bitcast i16* [[TMP10]] to <2 x i16>*
33; CHECK-NEXT:    store <2 x i16> [[PREDPHI]], <2 x i16>* [[TMP11]], align 2
34; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
35; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
36; CHECK-NEXT:    [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
37; CHECK-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]]
38; CHECK:       middle.block:
39; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
40; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
41; CHECK:       scalar.ph:
42; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
43; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
44; CHECK:       loop.header:
45; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
46; CHECK-NEXT:    [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i16
47; CHECK-NEXT:    br label [[LOOP_COND:%.*]]
48; CHECK:       loop.cond:
49; CHECK-NEXT:    [[BLEND:%.*]] = phi i16 [ [[IV_TRUNC]], [[LOOP_HEADER]] ]
50; CHECK-NEXT:    [[SRC_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[BLEND]]
51; CHECK-NEXT:    [[LV:%.*]] = load i16, i16* [[SRC_PTR]], align 1
52; CHECK-NEXT:    [[CMP_B:%.*]] = icmp sgt i64 [[IV]], [[A]]
53; CHECK-NEXT:    br i1 [[CMP_B]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
54; CHECK:       loop.next:
55; CHECK-NEXT:    br label [[LOOP_LATCH]]
56; CHECK:       loop.latch:
57; CHECK-NEXT:    [[RES:%.*]] = phi i16 [ [[LV]], [[LOOP_COND]] ], [ 1, [[LOOP_NEXT]] ]
58; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[IV]]
59; CHECK-NEXT:    store i16 [[RES]], i16* [[DST_PTR]], align 2
60; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
61; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
62; CHECK-NEXT:    br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], [[LOOP2:!llvm.loop !.*]]
63; CHECK:       exit:
64; CHECK-NEXT:    ret void
65;
66entry:
67  br label %loop.header
68
69loop.header:
70  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
71  %iv.trunc = trunc i64 %iv to i16
72  br label %loop.cond
73
74loop.cond:
75  %blend = phi i16 [ %iv.trunc, %loop.header ]
76  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
77  %lv = load i16, i16* %src.ptr, align 1
78  %cmp.b = icmp sgt i64 %iv, %a
79  br i1 %cmp.b, label %loop.next, label %loop.latch
80
81loop.next:
82  br label %loop.latch
83
84loop.latch:
85  %res = phi i16 [ %lv, %loop.cond ], [ 1, %loop.next ]
86  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
87  store i16 %res, i16* %dst.ptr
88  %iv.next = add nuw nsw i64 %iv, 1
89  %cmp439 = icmp ult i64 %iv, 31
90  br i1 %cmp439, label %loop.header, label %exit
91
92exit:
93  ret void
94}
95
96; The load in the loop does not need predication, because the accessed memory
97; is de-referenceable for all loop iterations.
98define void @single_incoming_phi_with_blend_mask(i64 %a, i64 %b) {
99; CHECK-LABEL: @single_incoming_phi_with_blend_mask(
100; CHECK-NEXT:  entry:
101; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
102; CHECK:       vector.ph:
103; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
104; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
105; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
106; CHECK:       vector.body:
107; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
108; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
109; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[INDEX]] to i16
110; CHECK-NEXT:    [[TMP3:%.*]] = add i16 [[TMP2]], 0
111; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
112; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
113; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP3]]
114; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr i16, i16* [[TMP5]], i32 0
115; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i16* [[TMP6]] to <2 x i16>*
116; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP7]], align 1
117; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
118; CHECK-NEXT:    [[TMP9:%.*]] = xor <2 x i1> [[TMP4]], <i1 true, i1 true>
119; CHECK-NEXT:    [[TMP10:%.*]] = xor <2 x i1> [[TMP8]], <i1 true, i1 true>
120; CHECK-NEXT:    [[TMP11:%.*]] = select <2 x i1> [[TMP4]], <2 x i1> [[TMP10]], <2 x i1> zeroinitializer
121; CHECK-NEXT:    [[TMP12:%.*]] = select <2 x i1> [[TMP4]], <2 x i1> [[TMP8]], <2 x i1> zeroinitializer
122; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP11]], <2 x i16> [[WIDE_LOAD]], <2 x i16> zeroinitializer
123; CHECK-NEXT:    [[PREDPHI1:%.*]] = select <2 x i1> [[TMP12]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[PREDPHI]]
124; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
125; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i16, i16* [[TMP13]], i32 0
126; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i16* [[TMP14]] to <2 x i16>*
127; CHECK-NEXT:    store <2 x i16> [[PREDPHI1]], <2 x i16>* [[TMP15]], align 2
128; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
129; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
130; CHECK-NEXT:    [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
131; CHECK-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
132; CHECK:       middle.block:
133; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
134; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
135;
136entry:
137  br label %loop.header
138
139loop.header:
140  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
141  %iv.trunc = trunc i64 %iv to i16
142  %cmp.a = icmp ugt i64 %iv, %a
143  br i1 %cmp.a, label %loop.cond, label %loop.latch
144
145loop.cond:
146  %blend = phi i16 [ %iv.trunc, %loop.header ]
147  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
148  %lv = load i16, i16* %src.ptr, align 1
149  %cmp.b = icmp sgt i64 %iv, %a
150  br i1 %cmp.b, label %loop.next, label %loop.latch
151
152loop.next:
153  br label %loop.latch
154
155loop.latch:
156  %res = phi i16 [ 0, %loop.header ], [ %lv, %loop.cond ], [ 1, %loop.next ]
157  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
158  store i16 %res, i16* %dst.ptr
159  %iv.next = add nuw nsw i64 %iv, 1
160  %cmp439 = icmp ult i64 %iv, 31
161  br i1 %cmp439, label %loop.header, label %exit
162
163exit:
164  ret void
165}
166
167define void @multiple_incoming_phi_with_blend_mask(i64 %a, i16* noalias %dst) {
168; CHECK-LABEL: @multiple_incoming_phi_with_blend_mask(
169; CHECK-NEXT:  entry:
170; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
171; CHECK:       vector.ph:
172; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
173; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
174; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
175; CHECK:       vector.body:
176; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
177; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
178; CHECK-NEXT:    [[VEC_IND1:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], [[VECTOR_BODY]] ]
179; CHECK-NEXT:    [[VEC_IND3:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT4:%.*]], [[VECTOR_BODY]] ]
180; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
181; CHECK-NEXT:    [[TMP2:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
182; CHECK-NEXT:    [[TMP3:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
183; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> [[VEC_IND3]], <2 x i16> [[VEC_IND1]]
184; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 0
185; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP4]]
186; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 1
187; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP6]]
188; CHECK-NEXT:    [[TMP8:%.*]] = load i16, i16* [[TMP5]], align 1
189; CHECK-NEXT:    [[TMP9:%.*]] = load i16, i16* [[TMP7]], align 1
190; CHECK-NEXT:    [[INS1:%.+]] = insertelement <2 x i16> poison, i16 [[TMP8]], i32 0
191; CHECK-NEXT:    [[INS2:%.+]] = insertelement <2 x i16> [[INS1]], i16 [[TMP9]], i32 1
192; CHECK-NEXT:    [[DST0:%.+]] = getelementptr inbounds i16, i16* %dst, i64 [[TMP0]]
193; CHECK-NEXT:    [[DST1:%.+]] = getelementptr inbounds i16, i16* [[DST0]], i32 0
194; CHECK-NEXT:    [[DST1_BC:%.+]] = bitcast i16* [[DST1]] to <2 x i16>*
195; CHECK-NEXT:    store <2 x i16> [[INS2]], <2 x i16>* [[DST1_BC]], align 2
196; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
197; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
198; CHECK-NEXT:    [[VEC_IND_NEXT2]] = add <2 x i16> [[VEC_IND1]], <i16 2, i16 2>
199; CHECK-NEXT:    [[VEC_IND_NEXT4]] = add <2 x i16> [[VEC_IND3]], <i16 2, i16 2>
200; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
201; CHECK-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP6:!llvm.loop !.*]]
202; CHECK:       middle.block:
203; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
204; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
205;
206entry:
207  br label %loop.header
208
209loop.header:
210  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
211  %iv.trunc = trunc i64 %iv to i16
212  %iv.trunc.2 = trunc i64 %iv to i16
213  %cmp.a = icmp ugt i64 %iv, %a
214  br i1 %cmp.a, label %loop.next, label %loop.latch
215
216loop.next:
217  br label %loop.latch
218
219loop.latch:
220  %blend = phi i16 [ %iv.trunc, %loop.header ], [ %iv.trunc.2, %loop.next ]
221  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
222  %lv = load i16, i16* %src.ptr, align 1
223  %dst.ptr = getelementptr inbounds i16, i16* %dst, i64 %iv
224  store i16 %lv, i16* %dst.ptr
225  %iv.next = add nuw nsw i64 %iv, 1
226  %cmp439 = icmp ult i64 %iv, 31
227  br i1 %cmp439, label %loop.header, label %exit
228
229exit:
230  ret void
231}
232
233; The load in the loop needs predication, because the accessed memory is not
234; de-referencable for all iterations of the loop.
235define void @single_incoming_needs_predication(i64 %a, i64 %b) {
236; CHECK-LABEL: @single_incoming_needs_predication(
237; CHECK-NEXT:  entry:
238; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
239; CHECK:       vector.ph:
240; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
241; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
242; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
243; CHECK:       vector.body:
244; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE4:%.*]] ]
245; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE4]] ]
246; CHECK-NEXT:    [[VEC_IND1:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], [[PRED_LOAD_CONTINUE4]] ]
247; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
248; CHECK-NEXT:    [[TMP2:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
249; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i32 0
250; CHECK-NEXT:    br i1 [[TMP3]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
251; CHECK:       pred.load.if:
252; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i16> [[VEC_IND1]], i32 0
253; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP4]]
254; CHECK-NEXT:    [[TMP6:%.*]] = load i16, i16* [[TMP5]], align 1
255; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i16> poison, i16 [[TMP6]], i32 0
256; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE]]
257; CHECK:       pred.load.continue:
258; CHECK-NEXT:    [[TMP8:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_LOAD_IF]] ]
259; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[TMP2]], i32 1
260; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4]]
261; CHECK:       pred.load.if3:
262; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x i16> [[VEC_IND1]], i32 1
263; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP10]]
264; CHECK-NEXT:    [[TMP12:%.*]] = load i16, i16* [[TMP11]], align 1
265; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x i16> [[TMP8]], i16 [[TMP12]], i32 1
266; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE4]]
267; CHECK:       pred.load.continue4:
268; CHECK-NEXT:    [[TMP14:%.*]] = phi <2 x i16> [ [[TMP8]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP13]], [[PRED_LOAD_IF3]] ]
269; CHECK-NEXT:    [[TMP15:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
270; CHECK-NEXT:    [[TMP16:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
271; CHECK-NEXT:    [[TMP17:%.*]] = xor <2 x i1> [[TMP15]], <i1 true, i1 true>
272; CHECK-NEXT:    [[TMP18:%.*]] = select <2 x i1> [[TMP2]], <2 x i1> [[TMP17]], <2 x i1> zeroinitializer
273; CHECK-NEXT:    [[TMP19:%.*]] = select <2 x i1> [[TMP2]], <2 x i1> [[TMP15]], <2 x i1> zeroinitializer
274; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP18]], <2 x i16> [[TMP14]], <2 x i16> zeroinitializer
275; CHECK-NEXT:    [[PREDPHI5:%.*]] = select <2 x i1> [[TMP19]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[PREDPHI]]
276; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
277; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i16, i16* [[TMP20]], i32 0
278; CHECK-NEXT:    [[TMP22:%.*]] = bitcast i16* [[TMP21]] to <2 x i16>*
279; CHECK-NEXT:    store <2 x i16> [[PREDPHI5]], <2 x i16>* [[TMP22]], align 2
280; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
281; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
282; CHECK-NEXT:    [[VEC_IND_NEXT2]] = add <2 x i16> [[VEC_IND1]], <i16 2, i16 2>
283; CHECK-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], 64
284; CHECK-NEXT:    br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP8:!llvm.loop !.*]]
285; CHECK:       middle.block:
286; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 64, 64
287; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
288;
289entry:
290  br label %loop.header
291
292loop.header:
293  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
294  %iv.trunc = trunc i64 %iv to i16
295  %cmp.a = icmp ugt i64 %iv, %a
296  br i1 %cmp.a, label %loop.cond, label %loop.latch
297
298loop.cond:
299  %blend = phi i16 [ %iv.trunc, %loop.header ]
300  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
301  %lv = load i16, i16* %src.ptr, align 1
302  %cmp.b = icmp sgt i64 %iv, %a
303  br i1 %cmp.b, label %loop.next, label %loop.latch
304
305loop.next:
306  br label %loop.latch
307
308loop.latch:
309  %res = phi i16 [ 0, %loop.header ], [ %lv, %loop.cond ], [ 1, %loop.next ]
310  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
311  store i16 %res, i16* %dst.ptr
312  %iv.next = add nuw nsw i64 %iv, 1
313  %cmp439 = icmp ult i64 %iv, 63
314  br i1 %cmp439, label %loop.header, label %exit
315
316exit:
317  ret void
318}
319
320; Test case for PR44800.
321define void @duplicated_incoming_blocks_blend(i32 %x, i32* %ptr) {
322; CHECK-LABEL: @duplicated_incoming_blocks_blend(
323; CHECK-NEXT:  entry:
324; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
325; CHECK:       vector.ph:
326; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
327; CHECK:       vector.body:
328; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
329; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
330; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <2 x i32> [[VEC_IND]], i32 0
331; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i32, i32* [[PTR:%.*]], i32 [[TMP1]]
332; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr i32, i32* [[TMP2]], i32 0
333; CHECK-NEXT:    [[TMP4:%.*]] = bitcast i32* [[TMP3]] to <2 x i32>*
334; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP4]], align 4
335; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
336; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
337; CHECK-NEXT:    [[TMP5:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
338; CHECK-NEXT:    br i1 [[TMP5]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP10:!llvm.loop !.*]]
339; CHECK:       middle.block:
340; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 1000, 1000
341; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
342;
343entry:
344  br label %loop.header
345
346loop.header:
347  %iv = phi i32 [ 0 , %entry ], [ %add.i, %loop.latch ]
348  %c.0 = icmp ugt i32 %iv, %x
349  br i1 %c.0, label %loop.latch, label %loop.latch
350
351loop.latch:
352  %p = phi i32 [ %iv, %loop.header ], [ %iv, %loop.header ]
353  %gep.ptr = getelementptr i32, i32* %ptr, i32 %p
354  store i32 %p, i32* %gep.ptr
355  %add.i = add nsw i32 %p, 1
356  %cmp = icmp slt i32 %add.i, 1000
357  br i1 %cmp, label %loop.header, label %exit
358
359exit:
360  ret void
361}
362