xref: /llvm-project/llvm/test/Transforms/LoopVectorize/uniform-blend.ll (revision f0d5104c944b329c479802788571ed6df41e0e86) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
2; RUN: opt -passes=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: define void @blend_uniform_iv_trunc(
8; CHECK-SAME: i1 [[C:%.*]]) {
9; CHECK-NEXT:  [[ENTRY:.*]]:
10; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
11; CHECK:       [[VECTOR_PH]]:
12; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
13; CHECK:       [[VECTOR_BODY]]:
14; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
15; CHECK-NEXT:    [[TMP0:%.*]] = trunc i64 [[INDEX]] to i16
16; CHECK-NEXT:    [[TMP1:%.*]] = add i16 [[TMP0]], 0
17; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[C]], i16 [[TMP1]], i16 poison
18; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i16 [[TMP6]]
19; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr inbounds i16, ptr [[TMP7]], i32 0
20; CHECK-NEXT:    store <4 x i16> zeroinitializer, ptr [[TMP3]], align 2
21; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
22; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
23; CHECK-NEXT:    br i1 [[TMP4]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
24; CHECK:       [[MIDDLE_BLOCK]]:
25; CHECK-NEXT:    br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]]
26; CHECK:       [[SCALAR_PH]]:
27; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
28; CHECK-NEXT:    br label %[[LOOP_HEADER:.*]]
29; CHECK:       [[LOOP_HEADER]]:
30; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
31; CHECK-NEXT:    [[IV_TRUNC_2:%.*]] = trunc i64 [[IV]] to i16
32; CHECK-NEXT:    br i1 [[C]], label %[[LOOP_NEXT:.*]], label %[[LOOP_LATCH]]
33; CHECK:       [[LOOP_NEXT]]:
34; CHECK-NEXT:    br label %[[LOOP_LATCH]]
35; CHECK:       [[LOOP_LATCH]]:
36; CHECK-NEXT:    [[BLEND:%.*]] = phi i16 [ poison, %[[LOOP_HEADER]] ], [ [[IV_TRUNC_2]], %[[LOOP_NEXT]] ]
37; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i16 [[BLEND]]
38; CHECK-NEXT:    store i16 0, ptr [[DST_PTR]], align 2
39; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
40; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
41; CHECK-NEXT:    br i1 [[CMP439]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]
42; CHECK:       [[EXIT]]:
43; CHECK-NEXT:    ret void
44;
45entry:
46  br label %loop.header
47
48loop.header:                                      ; preds = %loop.latch, %entry
49  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
50  %iv.trunc.2 = trunc i64 %iv to i16
51  br i1 %c, label %loop.next, label %loop.latch
52
53loop.next:                                        ; preds = %loop.header
54  br label %loop.latch
55
56loop.latch:                                       ; preds = %loop.next, %loop.header
57  %blend = phi i16 [ poison, %loop.header ], [ %iv.trunc.2, %loop.next ]
58  %dst.ptr = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i16 %blend
59  store i16 0, ptr %dst.ptr
60  %iv.next = add nuw nsw i64 %iv, 1
61  %cmp439 = icmp ult i64 %iv, 31
62  br i1 %cmp439, label %loop.header, label %exit
63
64exit:                                             ; preds = %loop.latch
65  ret void
66}
67
68define void @blend_uniform_iv(i1 %c) {
69; CHECK-LABEL: define void @blend_uniform_iv(
70; CHECK-SAME: i1 [[C:%.*]]) {
71; CHECK-NEXT:  [[ENTRY:.*]]:
72; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
73; CHECK:       [[VECTOR_PH]]:
74; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
75; CHECK:       [[VECTOR_BODY]]:
76; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
77; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
78; CHECK-NEXT:    [[TMP6:%.*]] = select i1 [[C]], i64 [[TMP0]], i64 poison
79; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP6]]
80; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds i16, ptr [[TMP7]], i32 0
81; CHECK-NEXT:    store <4 x i16> zeroinitializer, ptr [[TMP2]], align 2
82; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
83; CHECK-NEXT:    [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
84; CHECK-NEXT:    br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
85; CHECK:       [[MIDDLE_BLOCK]]:
86; CHECK-NEXT:    br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]]
87; CHECK:       [[SCALAR_PH]]:
88; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
89; CHECK-NEXT:    br label %[[LOOP_HEADER:.*]]
90; CHECK:       [[LOOP_HEADER]]:
91; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
92; CHECK-NEXT:    br i1 [[C]], label %[[LOOP_NEXT:.*]], label %[[LOOP_LATCH]]
93; CHECK:       [[LOOP_NEXT]]:
94; CHECK-NEXT:    br label %[[LOOP_LATCH]]
95; CHECK:       [[LOOP_LATCH]]:
96; CHECK-NEXT:    [[BLEND:%.*]] = phi i64 [ poison, %[[LOOP_HEADER]] ], [ [[IV]], %[[LOOP_NEXT]] ]
97; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[BLEND]]
98; CHECK-NEXT:    store i16 0, ptr [[DST_PTR]], align 2
99; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
100; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
101; CHECK-NEXT:    br i1 [[CMP439]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP5:![0-9]+]]
102; CHECK:       [[EXIT]]:
103; CHECK-NEXT:    ret void
104;
105
106entry:
107  br label %loop.header
108
109loop.header:                                      ; preds = %loop.latch, %entry
110  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
111  br i1 %c, label %loop.next, label %loop.latch
112
113loop.next:                                        ; preds = %loop.header
114  br label %loop.latch
115
116loop.latch:                                       ; preds = %loop.next, %loop.header
117  %blend = phi i64 [ poison, %loop.header ], [ %iv, %loop.next ]
118  %dst.ptr = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 %blend
119  store i16 0, ptr %dst.ptr
120  %iv.next = add nuw nsw i64 %iv, 1
121  %cmp439 = icmp ult i64 %iv, 31
122  br i1 %cmp439, label %loop.header, label %exit
123
124exit:                                             ; preds = %loop.latch
125  ret void
126}
127
128define void @blend_chain_iv(i1 %c) {
129; CHECK-LABEL: define void @blend_chain_iv(
130; CHECK-SAME: i1 [[C:%.*]]) {
131; CHECK-NEXT:  [[ENTRY:.*]]:
132; CHECK-NEXT:    br i1 false, label %[[SCALAR_PH:.*]], label %[[VECTOR_PH:.*]]
133; CHECK:       [[VECTOR_PH]]:
134; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i1> poison, i1 [[C]], i64 0
135; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i1> [[BROADCAST_SPLATINSERT]], <4 x i1> poison, <4 x i32> zeroinitializer
136; CHECK-NEXT:    [[TMP0:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> [[BROADCAST_SPLAT]], <4 x i1> zeroinitializer
137; CHECK-NEXT:    br label %[[VECTOR_BODY:.*]]
138; CHECK:       [[VECTOR_BODY]]:
139; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
140; CHECK-NEXT:    [[VEC_IND:%.*]] = phi <4 x i64> [ <i64 0, i64 1, i64 2, i64 3>, %[[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], %[[VECTOR_BODY]] ]
141; CHECK-NEXT:    [[PREDPHI:%.*]] = select <4 x i1> [[TMP0]], <4 x i64> [[VEC_IND]], <4 x i64> undef
142; CHECK-NEXT:    [[PREDPHI1:%.*]] = select <4 x i1> [[BROADCAST_SPLAT]], <4 x i64> [[PREDPHI]], <4 x i64> undef
143; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 0
144; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP1]]
145; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 1
146; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP3]]
147; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 2
148; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP5]]
149; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x i64> [[PREDPHI1]], i32 3
150; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[TMP7]]
151; CHECK-NEXT:    store i16 0, ptr [[TMP2]], align 2
152; CHECK-NEXT:    store i16 0, ptr [[TMP4]], align 2
153; CHECK-NEXT:    store i16 0, ptr [[TMP6]], align 2
154; CHECK-NEXT:    store i16 0, ptr [[TMP8]], align 2
155; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
156; CHECK-NEXT:    [[VEC_IND_NEXT]] = add <4 x i64> [[VEC_IND]], splat (i64 4)
157; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
158; CHECK-NEXT:    br i1 [[TMP9]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
159; CHECK:       [[MIDDLE_BLOCK]]:
160; CHECK-NEXT:    br i1 true, label %[[EXIT:.*]], label %[[SCALAR_PH]]
161; CHECK:       [[SCALAR_PH]]:
162; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, %[[MIDDLE_BLOCK]] ], [ 0, %[[ENTRY]] ]
163; CHECK-NEXT:    br label %[[LOOP_HEADER:.*]]
164; CHECK:       [[LOOP_HEADER]]:
165; CHECK-NEXT:    [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP_LATCH:.*]] ]
166; CHECK-NEXT:    br i1 [[C]], label %[[LOOP_NEXT:.*]], label %[[LOOP_LATCH]]
167; CHECK:       [[LOOP_NEXT]]:
168; CHECK-NEXT:    br i1 [[C]], label %[[LOOP_NEXT_2:.*]], label %[[LOOP_NEXT_3:.*]]
169; CHECK:       [[LOOP_NEXT_2]]:
170; CHECK-NEXT:    br label %[[LOOP_NEXT_3]]
171; CHECK:       [[LOOP_NEXT_3]]:
172; CHECK-NEXT:    [[BLEND_1:%.*]] = phi i64 [ undef, %[[LOOP_NEXT]] ], [ [[IV]], %[[LOOP_NEXT_2]] ]
173; CHECK-NEXT:    br label %[[LOOP_LATCH]]
174; CHECK:       [[LOOP_LATCH]]:
175; CHECK-NEXT:    [[BLEND:%.*]] = phi i64 [ undef, %[[LOOP_HEADER]] ], [ [[BLEND_1]], %[[LOOP_NEXT_3]] ]
176; CHECK-NEXT:    [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 [[BLEND]]
177; CHECK-NEXT:    store i16 0, ptr [[DST_PTR]], align 2
178; CHECK-NEXT:    [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
179; CHECK-NEXT:    [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
180; CHECK-NEXT:    br i1 [[CMP439]], label %[[LOOP_HEADER]], label %[[EXIT]], !llvm.loop [[LOOP7:![0-9]+]]
181; CHECK:       [[EXIT]]:
182; CHECK-NEXT:    ret void
183;
184
185entry:
186  br label %loop.header
187
188loop.header:                                      ; preds = %loop.latch, %entry
189  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
190  br i1 %c, label %loop.next, label %loop.latch
191
192loop.next:                                        ; preds = %loop.header
193  br i1 %c, label %loop.next.2, label %loop.next.3
194
195loop.next.2:
196  br label %loop.next.3
197
198loop.next.3:
199  %blend.1 = phi i64 [ undef, %loop.next ], [ %iv, %loop.next.2 ]
200  br label %loop.latch
201
202loop.latch:                                       ; preds = %loop.next, %loop.header
203  %blend = phi i64 [ undef, %loop.header ], [ %blend.1, %loop.next.3 ]
204  %dst.ptr = getelementptr inbounds [32 x i16], ptr @dst, i16 0, i64 %blend
205  store i16 0, ptr %dst.ptr
206  %iv.next = add nuw nsw i64 %iv, 1
207  %cmp439 = icmp ult i64 %iv, 31
208  br i1 %cmp439, label %loop.header, label %exit
209
210exit:                                             ; preds = %loop.latch
211  ret void
212}
213;.
214; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
215; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
216; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
217; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
218; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
219; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
220; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
221; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META2]], [[META1]]}
222;.
223