1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes=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;
42entry:
43  br label %loop.header
44
45loop.header:
46  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
47  %iv.trunc = trunc i64 %iv to i16
48  br label %loop.cond
49
50loop.cond:
51  %blend = phi i16 [ %iv.trunc, %loop.header ]
52  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
53  %lv = load i16, i16* %src.ptr, align 1
54  %cmp.b = icmp sgt i64 %iv, %a
55  br i1 %cmp.b, label %loop.next, label %loop.latch
56
57loop.next:
58  br label %loop.latch
59
60loop.latch:
61  %res = phi i16 [ %lv, %loop.cond ], [ 1, %loop.next ]
62  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
63  store i16 %res, i16* %dst.ptr
64  %iv.next = add nuw nsw i64 %iv, 1
65  %cmp439 = icmp ult i64 %iv, 31
66  br i1 %cmp439, label %loop.header, label %exit
67
68exit:
69  ret void
70}
71
72; The load in the loop does not need predication, because the accessed memory
73; is de-referenceable for all loop iterations.
74define void @single_incoming_phi_with_blend_mask(i64 %a, i64 %b) {
75; CHECK-LABEL: @single_incoming_phi_with_blend_mask(
76; CHECK-NEXT:  entry:
77; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
78; CHECK:       vector.ph:
79; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
80; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
81; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
82; CHECK:       vector.body:
83; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
84; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
85; CHECK-NEXT:    [[TMP2:%.*]] = trunc i64 [[INDEX]] to i16
86; CHECK-NEXT:    [[TMP3:%.*]] = add i16 [[TMP2]], 0
87; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
88; CHECK-NEXT:    [[TMP4:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
89; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP3]]
90; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr i16, i16* [[TMP5]], i32 0
91; CHECK-NEXT:    [[TMP7:%.*]] = bitcast i16* [[TMP6]] to <2 x i16>*
92; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP7]], align 1
93; CHECK-NEXT:    [[TMP8:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
94; CHECK-NEXT:    [[TMP9:%.*]] = xor <2 x i1> [[TMP4]], <i1 true, i1 true>
95; CHECK-NEXT:    [[TMP10:%.*]] = xor <2 x i1> [[TMP8]], <i1 true, i1 true>
96; CHECK-NEXT:    [[TMP11:%.*]] = select <2 x i1> [[TMP4]], <2 x i1> [[TMP10]], <2 x i1> zeroinitializer
97; CHECK-NEXT:    [[TMP12:%.*]] = select <2 x i1> [[TMP4]], <2 x i1> [[TMP8]], <2 x i1> zeroinitializer
98; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP11]], <2 x i16> [[WIDE_LOAD]], <2 x i16> zeroinitializer
99; CHECK-NEXT:    [[PREDPHI1:%.*]] = select <2 x i1> [[TMP12]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[PREDPHI]]
100; CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
101; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i16, i16* [[TMP13]], i32 0
102; CHECK-NEXT:    [[TMP15:%.*]] = bitcast i16* [[TMP14]] to <2 x i16>*
103; CHECK-NEXT:    store <2 x i16> [[PREDPHI1]], <2 x i16>* [[TMP15]], align 2
104; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
105; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
106; CHECK-NEXT:    [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
107; CHECK-NEXT:    br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
108; CHECK:       middle.block:
109; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
110; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
111;
112entry:
113  br label %loop.header
114
115loop.header:
116  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
117  %iv.trunc = trunc i64 %iv to i16
118  %cmp.a = icmp ugt i64 %iv, %a
119  br i1 %cmp.a, label %loop.cond, label %loop.latch
120
121loop.cond:
122  %blend = phi i16 [ %iv.trunc, %loop.header ]
123  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
124  %lv = load i16, i16* %src.ptr, align 1
125  %cmp.b = icmp sgt i64 %iv, %a
126  br i1 %cmp.b, label %loop.next, label %loop.latch
127
128loop.next:
129  br label %loop.latch
130
131loop.latch:
132  %res = phi i16 [ 0, %loop.header ], [ %lv, %loop.cond ], [ 1, %loop.next ]
133  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
134  store i16 %res, i16* %dst.ptr
135  %iv.next = add nuw nsw i64 %iv, 1
136  %cmp439 = icmp ult i64 %iv, 31
137  br i1 %cmp439, label %loop.header, label %exit
138
139exit:
140  ret void
141}
142
143define void @multiple_incoming_phi_with_blend_mask(i64 %a, i16* noalias %dst) {
144; CHECK-LABEL: @multiple_incoming_phi_with_blend_mask(
145; CHECK-NEXT:  entry:
146; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
147; CHECK:       vector.ph:
148; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
149; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
150; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
151; CHECK:       vector.body:
152; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
153; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
154; CHECK-NEXT:    [[VEC_IND1:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], [[VECTOR_BODY]] ]
155; CHECK-NEXT:    [[VEC_IND3:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT4:%.*]], [[VECTOR_BODY]] ]
156; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
157; CHECK-NEXT:    [[TMP2:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
158; CHECK-NEXT:    [[TMP3:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
159; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> [[VEC_IND3]], <2 x i16> [[VEC_IND1]]
160; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 0
161; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP4]]
162; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 1
163; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP6]]
164; CHECK-NEXT:    [[TMP8:%.*]] = load i16, i16* [[TMP5]], align 1
165; CHECK-NEXT:    [[TMP9:%.*]] = load i16, i16* [[TMP7]], align 1
166; CHECK-NEXT:    [[INS1:%.+]] = insertelement <2 x i16> poison, i16 [[TMP8]], i32 0
167; CHECK-NEXT:    [[INS2:%.+]] = insertelement <2 x i16> [[INS1]], i16 [[TMP9]], i32 1
168; CHECK-NEXT:    [[DST0:%.+]] = getelementptr inbounds i16, i16* %dst, i64 [[TMP0]]
169; CHECK-NEXT:    [[DST1:%.+]] = getelementptr inbounds i16, i16* [[DST0]], i32 0
170; CHECK-NEXT:    [[DST1_BC:%.+]] = bitcast i16* [[DST1]] to <2 x i16>*
171; CHECK-NEXT:    store <2 x i16> [[INS2]], <2 x i16>* [[DST1_BC]], align 2
172; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
173; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
174; CHECK-NEXT:    [[VEC_IND_NEXT2]] = add <2 x i16> [[VEC_IND1]], <i16 2, i16 2>
175; CHECK-NEXT:    [[VEC_IND_NEXT4]] = add <2 x i16> [[VEC_IND3]], <i16 2, i16 2>
176; CHECK-NEXT:    [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
177; CHECK-NEXT:    br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP6:!llvm.loop !.*]]
178; CHECK:       middle.block:
179; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
180; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
181;
182entry:
183  br label %loop.header
184
185loop.header:
186  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
187  %iv.trunc = trunc i64 %iv to i16
188  %iv.trunc.2 = trunc i64 %iv to i16
189  %cmp.a = icmp ugt i64 %iv, %a
190  br i1 %cmp.a, label %loop.next, label %loop.latch
191
192loop.next:
193  br label %loop.latch
194
195loop.latch:
196  %blend = phi i16 [ %iv.trunc, %loop.header ], [ %iv.trunc.2, %loop.next ]
197  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
198  %lv = load i16, i16* %src.ptr, align 1
199  %dst.ptr = getelementptr inbounds i16, i16* %dst, i64 %iv
200  store i16 %lv, i16* %dst.ptr
201  %iv.next = add nuw nsw i64 %iv, 1
202  %cmp439 = icmp ult i64 %iv, 31
203  br i1 %cmp439, label %loop.header, label %exit
204
205exit:
206  ret void
207}
208
209; The load in the loop needs predication, because the accessed memory is not
210; de-referencable for all iterations of the loop.
211define void @single_incoming_needs_predication(i64 %a, i64 %b) {
212; CHECK-LABEL: @single_incoming_needs_predication(
213; CHECK-NEXT:  entry:
214; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
215; CHECK:       vector.ph:
216; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
217; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
218; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
219; CHECK:       vector.body:
220; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE2:%.*]] ]
221; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE2]] ]
222; CHECK-NEXT:    [[TMP0:%.*]] = trunc i64 [[INDEX]] to i16
223; CHECK-NEXT:    [[TMP1:%.*]] = add i64 [[INDEX]], 0
224; CHECK-NEXT:    [[TMP2:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
225; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i32 0
226; CHECK-NEXT:    br i1 [[TMP3]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
227; CHECK:       pred.load.if:
228; CHECK-NEXT:    [[TMP4:%.*]] = add i16 [[TMP0]], 0
229; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP4]]
230; CHECK-NEXT:    [[TMP6:%.*]] = load i16, i16* [[TMP5]], align 1
231; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <2 x i16> poison, i16 [[TMP6]], i32 0
232; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE]]
233; CHECK:       pred.load.continue:
234; CHECK-NEXT:    [[TMP8:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_LOAD_IF]] ]
235; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <2 x i1> [[TMP2]], i32 1
236; CHECK-NEXT:    br i1 [[TMP9]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2]]
237; CHECK:       pred.load.if1:
238; CHECK-NEXT:    [[TMP10:%.*]] = add i16 [[TMP0]], 1
239; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP10]]
240; CHECK-NEXT:    [[TMP12:%.*]] = load i16, i16* [[TMP11]], align 1
241; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <2 x i16> [[TMP8]], i16 [[TMP12]], i32 1
242; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE2]]
243; CHECK:       pred.load.continue2:
244; CHECK-NEXT:    [[TMP14:%.*]] = phi <2 x i16> [ [[TMP8]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP13]], [[PRED_LOAD_IF1]] ]
245; CHECK-NEXT:    [[TMP15:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
246; CHECK-NEXT:    [[TMP16:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
247; CHECK-NEXT:    [[TMP17:%.*]] = xor <2 x i1> [[TMP15]], <i1 true, i1 true>
248; CHECK-NEXT:    [[TMP18:%.*]] = select <2 x i1> [[TMP2]], <2 x i1> [[TMP17]], <2 x i1> zeroinitializer
249; CHECK-NEXT:    [[TMP19:%.*]] = select <2 x i1> [[TMP2]], <2 x i1> [[TMP15]], <2 x i1> zeroinitializer
250; CHECK-NEXT:    [[PREDPHI:%.*]] = select <2 x i1> [[TMP18]], <2 x i16> [[TMP14]], <2 x i16> zeroinitializer
251; CHECK-NEXT:    [[PREDPHI3:%.*]] = select <2 x i1> [[TMP19]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[PREDPHI]]
252; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP1]]
253; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i16, i16* [[TMP20]], i32 0
254; CHECK-NEXT:    [[TMP22:%.*]] = bitcast i16* [[TMP21]] to <2 x i16>*
255; CHECK-NEXT:    store <2 x i16> [[PREDPHI3]], <2 x i16>* [[TMP22]], align 2
256; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
257; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
258; CHECK-NEXT:    [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], 64
259; CHECK-NEXT:    br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
260; CHECK:       middle.block:
261; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 64, 64
262; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
263;
264entry:
265  br label %loop.header
266
267loop.header:
268  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
269  %iv.trunc = trunc i64 %iv to i16
270  %cmp.a = icmp ugt i64 %iv, %a
271  br i1 %cmp.a, label %loop.cond, label %loop.latch
272
273loop.cond:
274  %blend = phi i16 [ %iv.trunc, %loop.header ]
275  %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
276  %lv = load i16, i16* %src.ptr, align 1
277  %cmp.b = icmp sgt i64 %iv, %a
278  br i1 %cmp.b, label %loop.next, label %loop.latch
279
280loop.next:
281  br label %loop.latch
282
283loop.latch:
284  %res = phi i16 [ 0, %loop.header ], [ %lv, %loop.cond ], [ 1, %loop.next ]
285  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
286  store i16 %res, i16* %dst.ptr
287  %iv.next = add nuw nsw i64 %iv, 1
288  %cmp439 = icmp ult i64 %iv, 63
289  br i1 %cmp439, label %loop.header, label %exit
290
291exit:
292  ret void
293}
294
295; Test case for PR44800.
296define void @duplicated_incoming_blocks_blend(i32 %x, i32* %ptr) {
297; CHECK-LABEL: @duplicated_incoming_blocks_blend(
298; CHECK-NEXT:  entry:
299; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
300; CHECK:       vector.ph:
301; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
302; CHECK:       vector.body:
303; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
304; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <2 x i32> [ <i32 0, i32 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
305; CHECK-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
306; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr i32, i32* [[PTR:%.*]], i32 [[TMP0]]
307; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr i32, i32* [[TMP1]], i32 0
308; CHECK-NEXT:    [[TMP3:%.*]] = bitcast i32* [[TMP2]] to <2 x i32>*
309; CHECK-NEXT:    store <2 x i32> [[VEC_IND]], <2 x i32>* [[TMP3]], align 4
310; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
311; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <2 x i32> [[VEC_IND]], <i32 2, i32 2>
312; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000
313; CHECK-NEXT:    br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]]
314; CHECK:       middle.block:
315; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i32 1000, 1000
316; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
317;
318entry:
319  br label %loop.header
320
321loop.header:
322  %iv = phi i32 [ 0 , %entry ], [ %add.i, %loop.latch ]
323  %c.0 = icmp ugt i32 %iv, %x
324  br i1 %c.0, label %loop.latch, label %loop.latch
325
326loop.latch:
327  %p = phi i32 [ %iv, %loop.header ], [ %iv, %loop.header ]
328  %gep.ptr = getelementptr i32, i32* %ptr, i32 %p
329  store i32 %p, i32* %gep.ptr
330  %add.i = add nsw i32 %p, 1
331  %cmp = icmp slt i32 %add.i, 1000
332  br i1 %cmp, label %loop.header, label %exit
333
334exit:
335  ret void
336}
337