1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -loop-vectorize -force-vector-width=4 -S %s | FileCheck %s
3
4@dst = external global [32 x i16], align 1
5
6define void @blend_uniform_iv_trunc(i1 %c) {
7; CHECK-LABEL: @blend_uniform_iv_trunc(
8; CHECK-NEXT:  entry:
9; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
10; CHECK:       vector.ph:
11; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C:%.*]], i32 0
12; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer
13; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
14; CHECK:       vector.body:
15; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
16; CHECK-NEXT:    [[TMP0:%.*]] = trunc i64 [[INDEX]] to i16
17; CHECK-NEXT:    [[TMP1:%.*]] = add i16 [[TMP0]], 0
18; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i16> poison, i16 [[TMP1]], i32 0
19; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT1]], <4 x i16> poison, <4 x i32> zeroinitializer
20; CHECK-NEXT:    [[TMP2:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true>
21; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i16> [[BROADCAST_SPLAT2]], <4 x i16> undef
22; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i16> [[PREDPHI]], i32 0
23; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i16 [[TMP3]]
24; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr inbounds i16, i16* [[TMP4]], i32 0
25; CHECK-NEXT:    [[TMP6:%.*]] = bitcast i16* [[TMP5]] to <4 x i16>*
26; CHECK-NEXT:    store <4 x i16> zeroinitializer, <4 x i16>* [[TMP6]], align 2
27; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
28; CHECK-NEXT:    [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
29; CHECK-NEXT:    br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
30; CHECK:       middle.block:
31; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
32; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
33; CHECK:       scalar.ph:
34; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
35; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
36; CHECK:       loop.header:
37; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
38; CHECK-NEXT:    [[IV_TRUNC_2:%.*]] = trunc i64 [[IV]] to i16
39; CHECK-NEXT:    br i1 [[C]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
40; CHECK:       loop.next:
41; CHECK-NEXT:    br label [[LOOP_LATCH]]
42; CHECK:       loop.latch:
43; CHECK-NEXT:    [[BLEND:%.*]] = phi i16 [ undef, [[LOOP_HEADER]] ], [ [[IV_TRUNC_2]], [[LOOP_NEXT]] ]
44; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i16 [[BLEND]]
45; CHECK-NEXT:    store i16 0, i16* [[DST_PTR]], align 2
46; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
47; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
48; CHECK-NEXT:    br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], !llvm.loop [[LOOP2:![0-9]+]]
49; CHECK:       exit:
50; CHECK-NEXT:    ret void
51;
52
53entry:
54  br label %loop.header
55
56loop.header:                                      ; preds = %loop.latch, %entry
57  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
58  %iv.trunc.2 = trunc i64 %iv to i16
59  br i1 %c, label %loop.next, label %loop.latch
60
61loop.next:                                        ; preds = %loop.header
62  br label %loop.latch
63
64loop.latch:                                       ; preds = %loop.next, %loop.header
65  %blend = phi i16 [ undef, %loop.header ], [ %iv.trunc.2, %loop.next ]
66  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i16 %blend
67  store i16 0, i16* %dst.ptr
68  %iv.next = add nuw nsw i64 %iv, 1
69  %cmp439 = icmp ult i64 %iv, 31
70  br i1 %cmp439, label %loop.header, label %exit
71
72exit:                                             ; preds = %loop.latch
73  ret void
74}
75
76define void @blend_uniform_iv(i1 %c) {
77; CHECK-LABEL: @blend_uniform_iv(
78; CHECK-NEXT:  entry:
79; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
80; CHECK:       vector.ph:
81; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C:%.*]], i32 0
82; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer
83; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
84; CHECK:       vector.body:
85; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
86; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
87; CHECK-NEXT:    [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i64> poison, i64 [[TMP0]], i32 0
88; CHECK-NEXT:    [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i64> [[BROADCAST_SPLATINSERT1]], <4 x i64> poison, <4 x i32> zeroinitializer
89; CHECK-NEXT:    [[TMP1:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true>
90; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i64> [[BROADCAST_SPLAT2]], <4 x i64> undef
91; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x i64> [[PREDPHI]], i32 0
92; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP2]]
93; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds i16, i16* [[TMP3]], i32 0
94; CHECK-NEXT:    [[TMP5:%.*]] = bitcast i16* [[TMP4]] to <4 x i16>*
95; CHECK-NEXT:    store <4 x i16> zeroinitializer, <4 x i16>* [[TMP5]], align 2
96; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
97; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
98; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
99; CHECK:       middle.block:
100; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
101; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
102; CHECK:       scalar.ph:
103; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
104; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
105; CHECK:       loop.header:
106; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
107; CHECK-NEXT:    br i1 [[C]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
108; CHECK:       loop.next:
109; CHECK-NEXT:    br label [[LOOP_LATCH]]
110; CHECK:       loop.latch:
111; CHECK-NEXT:    [[BLEND:%.*]] = phi i64 [ undef, [[LOOP_HEADER]] ], [ [[IV]], [[LOOP_NEXT]] ]
112; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[BLEND]]
113; CHECK-NEXT:    store i16 0, i16* [[DST_PTR]], align 2
114; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
115; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
116; CHECK-NEXT:    br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], !llvm.loop [[LOOP5:![0-9]+]]
117; CHECK:       exit:
118; CHECK-NEXT:    ret void
119;
120
121entry:
122  br label %loop.header
123
124loop.header:                                      ; preds = %loop.latch, %entry
125  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
126  br i1 %c, label %loop.next, label %loop.latch
127
128loop.next:                                        ; preds = %loop.header
129  br label %loop.latch
130
131loop.latch:                                       ; preds = %loop.next, %loop.header
132  %blend = phi i64 [ undef, %loop.header ], [ %iv, %loop.next ]
133  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %blend
134  store i16 0, 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:                                             ; preds = %loop.latch
140  ret void
141}
142
143define void @blend_chain_iv(i1 %c) {
144; CHECK-LABEL: @blend_chain_iv(
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 <4 x i1> poison, i1 [[C:%.*]], i32 0
149; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 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 <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
154; CHECK-NEXT:    [[TMP0:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true>
155; CHECK-NEXT:    [[TMP1:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> [[TMP0]], <4 x i1> zeroinitializer
156; CHECK-NEXT:    [[TMP2:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> zeroinitializer
157; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP2]], <4 x i64> [[VEC_IND]], <4 x i64> undef
158; CHECK-NEXT:    [[TMP3:%.*]] = xor <4 x i1> [[BROADCAST_SPLAT]], <i1 true, i1 true, i1 true, i1 true>
159; CHECK-NEXT:    [[TMP4:%.*]] = or <4 x i1> [[TMP2]], [[TMP1]]
160; CHECK-NEXT:    [[PREDPHI1:%.*]] = select <4 x i1> [[TMP4]], <4 x i64> [[PREDPHI]], <4 x i64> undef
161; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 0
162; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP5]]
163; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 1
164; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP7]]
165; CHECK-NEXT:    [[TMP9:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 2
166; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP9]]
167; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 3
168; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP11]]
169; CHECK-NEXT:    store i16 0, i16* [[TMP6]], align 2
170; CHECK-NEXT:    store i16 0, i16* [[TMP8]], align 2
171; CHECK-NEXT:    store i16 0, i16* [[TMP10]], align 2
172; CHECK-NEXT:    store i16 0, i16* [[TMP12]], align 2
173; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
174; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], <i64 4, i64 4, i64 4, i64 4>
175; CHECK-NEXT:    [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
176; CHECK-NEXT:    br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
177; CHECK:       middle.block:
178; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 32, 32
179; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
180; CHECK:       scalar.ph:
181; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
182; CHECK-NEXT:    br label [[LOOP_HEADER:%.*]]
183; CHECK:       loop.header:
184; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
185; CHECK-NEXT:    br i1 [[C]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
186; CHECK:       loop.next:
187; CHECK-NEXT:    br i1 [[C]], label [[LOOP_NEXT_2:%.*]], label [[LOOP_NEXT_3:%.*]]
188; CHECK:       loop.next.2:
189; CHECK-NEXT:    br label [[LOOP_NEXT_3]]
190; CHECK:       loop.next.3:
191; CHECK-NEXT:    [[BLEND_1:%.*]] = phi i64 [ undef, [[LOOP_NEXT]] ], [ [[IV]], [[LOOP_NEXT_2]] ]
192; CHECK-NEXT:    br label [[LOOP_LATCH]]
193; CHECK:       loop.latch:
194; CHECK-NEXT:    [[BLEND:%.*]] = phi i64 [ undef, [[LOOP_HEADER]] ], [ [[BLEND_1]], [[LOOP_NEXT_3]] ]
195; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[BLEND]]
196; CHECK-NEXT:    store i16 0, i16* [[DST_PTR]], align 2
197; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
198; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
199; CHECK-NEXT:    br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], !llvm.loop [[LOOP7:![0-9]+]]
200; CHECK:       exit:
201; CHECK-NEXT:    ret void
202;
203
204entry:
205  br label %loop.header
206
207loop.header:                                      ; preds = %loop.latch, %entry
208  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
209  br i1 %c, label %loop.next, label %loop.latch
210
211loop.next:                                        ; preds = %loop.header
212  br i1 %c, label %loop.next.2, label %loop.next.3
213
214loop.next.2:
215  br label %loop.next.3
216
217loop.next.3:
218  %blend.1 = phi i64 [ undef, %loop.next ], [ %iv, %loop.next.2 ]
219  br label %loop.latch
220
221loop.latch:                                       ; preds = %loop.next, %loop.header
222  %blend = phi i64 [ undef, %loop.header ], [ %blend.1, %loop.next.3 ]
223  %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %blend
224  store i16 0, 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:                                             ; preds = %loop.latch
230  ret void
231}
232