xref: /llvm-project/llvm/test/Transforms/LoopVectorize/strict-fadd-interleave-only.ll (revision 82821254f532c1dbdfd5d985ef7130511efaaa83)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 3
2; RUN: opt -passes=loop-vectorize -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue -force-vector-interleave=2 -force-vector-width=1 -force-ordered-reductions -S %s | FileCheck %s
3; RUN: opt -passes=loop-vectorize -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue -force-vector-interleave=2 -force-vector-width=1 -force-ordered-reductions -force-tail-folding-style=data -S %s | FileCheck --check-prefix=CHECK-ALM %s
4
5define float @pr70988() {
6; CHECK-LABEL: define float @pr70988() {
7; CHECK-NEXT:  entry:
8; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
9; CHECK:       vector.ph:
10; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
11; CHECK:       vector.body:
12; CHECK-NEXT:    [[INDEX1:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VECTOR_BODY]] ]
13; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
14; CHECK-NEXT:    [[VEC_IV:%.*]] = add i32 [[INDEX1]], 0
15; CHECK-NEXT:    [[VEC_IV2:%.*]] = add i32 [[INDEX1]], 1
16; CHECK-NEXT:    [[TMP0:%.*]] = icmp ule i32 [[VEC_IV]], 1020
17; CHECK-NEXT:    [[TMP1:%.*]] = icmp ule i32 [[VEC_IV2]], 1020
18; CHECK-NEXT:    [[TMP2:%.*]] = select contract i1 [[TMP0]], float 1.000000e+00, float -0.000000e+00
19; CHECK-NEXT:    [[TMP3:%.*]] = fadd contract float [[VEC_PHI]], [[TMP2]]
20; CHECK-NEXT:    [[TMP4:%.*]] = select contract i1 [[TMP1]], float 1.000000e+00, float -0.000000e+00
21; CHECK-NEXT:    [[TMP5]] = fadd contract float [[TMP3]], [[TMP4]]
22; CHECK-NEXT:    [[INDEX_NEXT3]] = add nuw i32 [[INDEX1]], 2
23; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT3]], 1022
24; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
25; CHECK:       middle.block:
26; CHECK-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
27; CHECK:       scalar.ph:
28; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 1022, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
29; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
30; CHECK-NEXT:    br label [[LOOP:%.*]]
31; CHECK:       loop:
32; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDEX_NEXT:%.*]], [[LOOP]] ]
33; CHECK-NEXT:    [[RDX:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[RDX_NEXT:%.*]], [[LOOP]] ]
34; CHECK-NEXT:    [[RDX_NEXT]] = fadd contract float [[RDX]], 1.000000e+00
35; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw nsw i32 [[INDEX]], 1
36; CHECK-NEXT:    [[COND:%.*]] = icmp ult i32 [[INDEX_NEXT]], 1021
37; CHECK-NEXT:    br i1 [[COND]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]
38; CHECK:       exit:
39; CHECK-NEXT:    [[DOTLCSSA:%.*]] = phi float [ [[RDX_NEXT]], [[LOOP]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ]
40; CHECK-NEXT:    ret float [[DOTLCSSA]]
41;
42; CHECK-ALM-LABEL: define float @pr70988() {
43; CHECK-ALM-NEXT:  entry:
44; CHECK-ALM-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
45; CHECK-ALM:       vector.ph:
46; CHECK-ALM-NEXT:    br label [[VECTOR_BODY:%.*]]
47; CHECK-ALM:       vector.body:
48; CHECK-ALM-NEXT:    [[INDEX1:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT3:%.*]], [[VECTOR_BODY]] ]
49; CHECK-ALM-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
50; CHECK-ALM-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX1]], 0
51; CHECK-ALM-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX1]], 1
52; CHECK-ALM-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = icmp ult i32 [[TMP0]], 1021
53; CHECK-ALM-NEXT:    [[ACTIVE_LANE_MASK2:%.*]] = icmp ult i32 [[TMP1]], 1021
54; CHECK-ALM-NEXT:    [[TMP2:%.*]] = select contract i1 [[ACTIVE_LANE_MASK]], float 1.000000e+00, float -0.000000e+00
55; CHECK-ALM-NEXT:    [[TMP3:%.*]] = fadd contract float [[VEC_PHI]], [[TMP2]]
56; CHECK-ALM-NEXT:    [[TMP4:%.*]] = select contract i1 [[ACTIVE_LANE_MASK2]], float 1.000000e+00, float -0.000000e+00
57; CHECK-ALM-NEXT:    [[TMP5]] = fadd contract float [[TMP3]], [[TMP4]]
58; CHECK-ALM-NEXT:    [[INDEX_NEXT3]] = add nuw i32 [[INDEX1]], 2
59; CHECK-ALM-NEXT:    [[TMP6:%.*]] = icmp eq i32 [[INDEX_NEXT3]], 1022
60; CHECK-ALM-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
61; CHECK-ALM:       middle.block:
62; CHECK-ALM-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
63; CHECK-ALM:       scalar.ph:
64; CHECK-ALM-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 1022, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
65; CHECK-ALM-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
66; CHECK-ALM-NEXT:    br label [[LOOP:%.*]]
67; CHECK-ALM:       loop:
68; CHECK-ALM-NEXT:    [[INDEX:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDEX_NEXT:%.*]], [[LOOP]] ]
69; CHECK-ALM-NEXT:    [[RDX:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[RDX_NEXT:%.*]], [[LOOP]] ]
70; CHECK-ALM-NEXT:    [[RDX_NEXT]] = fadd contract float [[RDX]], 1.000000e+00
71; CHECK-ALM-NEXT:    [[INDEX_NEXT]] = add nuw nsw i32 [[INDEX]], 1
72; CHECK-ALM-NEXT:    [[COND:%.*]] = icmp ult i32 [[INDEX_NEXT]], 1021
73; CHECK-ALM-NEXT:    br i1 [[COND]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP3:![0-9]+]]
74; CHECK-ALM:       exit:
75; CHECK-ALM-NEXT:    [[DOTLCSSA:%.*]] = phi float [ [[RDX_NEXT]], [[LOOP]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ]
76; CHECK-ALM-NEXT:    ret float [[DOTLCSSA]]
77;
78entry:
79  br label %loop
80
81loop:
82  %index = phi i32 [ 0, %entry ], [ %index.next, %loop ]
83  %rdx = phi float [ 0.000000e+00, %entry ], [ %rdx.next, %loop ]
84  %rdx.next = fadd contract float %rdx, 1.000000e+00
85  %index.next = add nuw nsw i32 %index, 1
86  %cond = icmp ult i32 %index.next, 1021
87  br i1 %cond, label %loop, label %exit
88
89exit:
90  %.lcssa = phi float [ %rdx.next, %loop ]
91  ret float %.lcssa
92}
93
94define float @pr72720reduction_using_active_lane_mask(ptr %src) {
95; CHECK-LABEL: define float @pr72720reduction_using_active_lane_mask(
96; CHECK-SAME: ptr [[SRC:%.*]]) {
97; CHECK-NEXT:  entry:
98; CHECK-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
99; CHECK:       vector.ph:
100; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
101; CHECK:       vector.body:
102; CHECK-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE3:%.*]] ]
103; CHECK-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP13:%.*]], [[PRED_LOAD_CONTINUE3]] ]
104; CHECK-NEXT:    [[VEC_IV:%.*]] = add i32 [[INDEX]], 0
105; CHECK-NEXT:    [[VEC_IV1:%.*]] = add i32 [[INDEX]], 1
106; CHECK-NEXT:    [[TMP0:%.*]] = icmp ule i32 [[VEC_IV]], 14
107; CHECK-NEXT:    [[TMP1:%.*]] = icmp ule i32 [[VEC_IV1]], 14
108; CHECK-NEXT:    br i1 [[TMP0]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
109; CHECK:       pred.load.if:
110; CHECK-NEXT:    [[TMP2:%.*]] = add i32 [[INDEX]], 0
111; CHECK-NEXT:    [[TMP3:%.*]] = getelementptr float, ptr [[SRC]], i32 [[TMP2]]
112; CHECK-NEXT:    [[TMP4:%.*]] = load float, ptr [[TMP3]], align 4
113; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE]]
114; CHECK:       pred.load.continue:
115; CHECK-NEXT:    [[TMP5:%.*]] = phi float [ poison, [[VECTOR_BODY]] ], [ [[TMP4]], [[PRED_LOAD_IF]] ]
116; CHECK-NEXT:    br i1 [[TMP1]], label [[PRED_LOAD_IF2:%.*]], label [[PRED_LOAD_CONTINUE3]]
117; CHECK:       pred.load.if2:
118; CHECK-NEXT:    [[TMP6:%.*]] = add i32 [[INDEX]], 1
119; CHECK-NEXT:    [[TMP7:%.*]] = getelementptr float, ptr [[SRC]], i32 [[TMP6]]
120; CHECK-NEXT:    [[TMP8:%.*]] = load float, ptr [[TMP7]], align 4
121; CHECK-NEXT:    br label [[PRED_LOAD_CONTINUE3]]
122; CHECK:       pred.load.continue3:
123; CHECK-NEXT:    [[TMP9:%.*]] = phi float [ poison, [[PRED_LOAD_CONTINUE]] ], [ [[TMP8]], [[PRED_LOAD_IF2]] ]
124; CHECK-NEXT:    [[TMP10:%.*]] = select contract i1 [[TMP0]], float [[TMP5]], float -0.000000e+00
125; CHECK-NEXT:    [[TMP11:%.*]] = fadd contract float [[VEC_PHI]], [[TMP10]]
126; CHECK-NEXT:    [[TMP12:%.*]] = select contract i1 [[TMP1]], float [[TMP9]], float -0.000000e+00
127; CHECK-NEXT:    [[TMP13]] = fadd contract float [[TMP11]], [[TMP12]]
128; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
129; CHECK-NEXT:    [[TMP14:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16
130; CHECK-NEXT:    br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
131; CHECK:       middle.block:
132; CHECK-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
133; CHECK:       scalar.ph:
134; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 16, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
135; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP13]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
136; CHECK-NEXT:    br label [[LOOP:%.*]]
137; CHECK:       loop:
138; CHECK-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[NARROW:%.*]], [[LOOP]] ]
139; CHECK-NEXT:    [[RDX:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[RDX_NEXT:%.*]], [[LOOP]] ]
140; CHECK-NEXT:    [[NARROW]] = add nuw nsw i32 [[IV]], 1
141; CHECK-NEXT:    [[GEP:%.*]] = getelementptr float, ptr [[SRC]], i32 [[IV]]
142; CHECK-NEXT:    [[L:%.*]] = load float, ptr [[GEP]], align 4
143; CHECK-NEXT:    [[RDX_NEXT]] = fadd contract float [[RDX]], [[L]]
144; CHECK-NEXT:    [[EC:%.*]] = icmp ult i32 [[NARROW]], 15
145; CHECK-NEXT:    br i1 [[EC]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP5:![0-9]+]]
146; CHECK:       exit:
147; CHECK-NEXT:    [[DOTLCSSA:%.*]] = phi float [ [[RDX_NEXT]], [[LOOP]] ], [ [[TMP13]], [[MIDDLE_BLOCK]] ]
148; CHECK-NEXT:    ret float [[DOTLCSSA]]
149;
150; CHECK-ALM-LABEL: define float @pr72720reduction_using_active_lane_mask(
151; CHECK-ALM-SAME: ptr [[SRC:%.*]]) {
152; CHECK-ALM-NEXT:  entry:
153; CHECK-ALM-NEXT:    br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
154; CHECK-ALM:       vector.ph:
155; CHECK-ALM-NEXT:    br label [[VECTOR_BODY:%.*]]
156; CHECK-ALM:       vector.body:
157; CHECK-ALM-NEXT:    [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE3:%.*]] ]
158; CHECK-ALM-NEXT:    [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[PRED_LOAD_CONTINUE3]] ]
159; CHECK-ALM-NEXT:    [[TMP0:%.*]] = add i32 [[INDEX]], 0
160; CHECK-ALM-NEXT:    [[TMP1:%.*]] = add i32 [[INDEX]], 1
161; CHECK-ALM-NEXT:    [[ACTIVE_LANE_MASK:%.*]] = icmp ult i32 [[TMP0]], 15
162; CHECK-ALM-NEXT:    [[ACTIVE_LANE_MASK1:%.*]] = icmp ult i32 [[TMP1]], 15
163; CHECK-ALM-NEXT:    br i1 [[ACTIVE_LANE_MASK]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
164; CHECK-ALM:       pred.load.if:
165; CHECK-ALM-NEXT:    [[TMP2:%.*]] = getelementptr float, ptr [[SRC]], i32 [[TMP0]]
166; CHECK-ALM-NEXT:    [[TMP3:%.*]] = load float, ptr [[TMP2]], align 4
167; CHECK-ALM-NEXT:    br label [[PRED_LOAD_CONTINUE]]
168; CHECK-ALM:       pred.load.continue:
169; CHECK-ALM-NEXT:    [[TMP4:%.*]] = phi float [ poison, [[VECTOR_BODY]] ], [ [[TMP3]], [[PRED_LOAD_IF]] ]
170; CHECK-ALM-NEXT:    br i1 [[ACTIVE_LANE_MASK1]], label [[PRED_LOAD_IF2:%.*]], label [[PRED_LOAD_CONTINUE3]]
171; CHECK-ALM:       pred.load.if2:
172; CHECK-ALM-NEXT:    [[TMP5:%.*]] = getelementptr float, ptr [[SRC]], i32 [[TMP1]]
173; CHECK-ALM-NEXT:    [[TMP6:%.*]] = load float, ptr [[TMP5]], align 4
174; CHECK-ALM-NEXT:    br label [[PRED_LOAD_CONTINUE3]]
175; CHECK-ALM:       pred.load.continue3:
176; CHECK-ALM-NEXT:    [[TMP7:%.*]] = phi float [ poison, [[PRED_LOAD_CONTINUE]] ], [ [[TMP6]], [[PRED_LOAD_IF2]] ]
177; CHECK-ALM-NEXT:    [[TMP8:%.*]] = select contract i1 [[ACTIVE_LANE_MASK]], float [[TMP4]], float -0.000000e+00
178; CHECK-ALM-NEXT:    [[TMP9:%.*]] = fadd contract float [[VEC_PHI]], [[TMP8]]
179; CHECK-ALM-NEXT:    [[TMP10:%.*]] = select contract i1 [[ACTIVE_LANE_MASK1]], float [[TMP7]], float -0.000000e+00
180; CHECK-ALM-NEXT:    [[TMP11]] = fadd contract float [[TMP9]], [[TMP10]]
181; CHECK-ALM-NEXT:    [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2
182; CHECK-ALM-NEXT:    [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], 16
183; CHECK-ALM-NEXT:    br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
184; CHECK-ALM:       middle.block:
185; CHECK-ALM-NEXT:    br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
186; CHECK-ALM:       scalar.ph:
187; CHECK-ALM-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i32 [ 16, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
188; CHECK-ALM-NEXT:    [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP11]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ]
189; CHECK-ALM-NEXT:    br label [[LOOP:%.*]]
190; CHECK-ALM:       loop:
191; CHECK-ALM-NEXT:    [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[NARROW:%.*]], [[LOOP]] ]
192; CHECK-ALM-NEXT:    [[RDX:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[RDX_NEXT:%.*]], [[LOOP]] ]
193; CHECK-ALM-NEXT:    [[NARROW]] = add nuw nsw i32 [[IV]], 1
194; CHECK-ALM-NEXT:    [[GEP:%.*]] = getelementptr float, ptr [[SRC]], i32 [[IV]]
195; CHECK-ALM-NEXT:    [[L:%.*]] = load float, ptr [[GEP]], align 4
196; CHECK-ALM-NEXT:    [[RDX_NEXT]] = fadd contract float [[RDX]], [[L]]
197; CHECK-ALM-NEXT:    [[EC:%.*]] = icmp ult i32 [[NARROW]], 15
198; CHECK-ALM-NEXT:    br i1 [[EC]], label [[LOOP]], label [[EXIT]], !llvm.loop [[LOOP5:![0-9]+]]
199; CHECK-ALM:       exit:
200; CHECK-ALM-NEXT:    [[DOTLCSSA:%.*]] = phi float [ [[RDX_NEXT]], [[LOOP]] ], [ [[TMP11]], [[MIDDLE_BLOCK]] ]
201; CHECK-ALM-NEXT:    ret float [[DOTLCSSA]]
202;
203entry:
204  br label %loop
205
206loop:
207  %iv = phi i32 [ 0, %entry ], [ %narrow, %loop ]
208  %rdx = phi float [ 0.0, %entry ], [ %rdx.next, %loop ]
209  %narrow = add nuw nsw i32 %iv, 1
210  %gep = getelementptr float, ptr %src, i32 %iv
211  %l = load float, ptr %gep, align 4
212  %rdx.next = fadd contract float %rdx, %l
213  %ec = icmp ult i32 %narrow, 15
214  br i1 %ec, label %loop, label %exit
215
216exit:
217  %.lcssa = phi float [ %rdx.next, %loop ]
218  ret float %.lcssa
219}
220