xref: /llvm-project/llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse.ll (revision 4ad0fdd1631eeae432714c03ede01a10dc00025d) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; This is the loop in c++ being vectorize in this file with
3;vector.reverse
4;  #pragma clang loop vectorize_width(8, scalable) interleave_count(2)
5;  for (int i = N-1; i >= 0; --i)
6;    a[i] = b[i] + 1.0;
7
8; RUN: opt -passes=loop-vectorize,dce,instcombine -mtriple aarch64-linux-gnu -S \
9; RUN:   -prefer-predicate-over-epilogue=scalar-epilogue < %s | FileCheck %s
10
11define void @vector_reverse_f64(i64 %N, ptr noalias %a, ptr noalias %b) #0{
12; CHECK-LABEL: @vector_reverse_f64(
13; CHECK-NEXT:  entry:
14; CHECK-NEXT:    [[CMP7:%.*]] = icmp sgt i64 [[N:%.*]], 0
15; CHECK-NEXT:    br i1 [[CMP7]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
16; CHECK:       for.body.preheader:
17; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
18; CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[TMP0]], 4
19; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
20; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
21; CHECK:       vector.ph:
22; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
23; CHECK-NEXT:    [[TMP3:%.*]] = shl i64 [[TMP2]], 4
24; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
25; CHECK-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[N]], [[N_MOD_VF]]
26; CHECK-NEXT:    [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
27; CHECK-NEXT:    [[TMP5:%.*]] = shl i64 [[TMP4]], 3
28; CHECK-NEXT:    [[TMP6:%.*]] = shl i64 [[TMP4]], 4
29; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
30; CHECK:       vector.body:
31; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
32; CHECK-NEXT:    [[TMP7:%.*]] = xor i64 [[INDEX]], -1
33; CHECK-NEXT:    [[TMP8:%.*]] = add i64 [[N]], [[TMP7]]
34; CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds double, ptr [[B:%.*]], i64 [[TMP8]]
35; CHECK-NEXT:    [[TMP10:%.*]] = sub i64 1, [[TMP5]]
36; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds double, ptr [[TMP9]], i64 [[TMP10]]
37; CHECK-NEXT:    [[TMP12:%.*]] = sub i64 0, [[TMP5]]
38; CHECK-NEXT:    [[TMP13:%.*]] = sub i64 1, [[TMP5]]
39; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds double, ptr [[TMP9]], i64 [[TMP12]]
40; CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds double, ptr [[TMP14]], i64 [[TMP13]]
41; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 8 x double>, ptr [[TMP11]], align 8
42; CHECK-NEXT:    [[WIDE_LOAD1:%.*]] = load <vscale x 8 x double>, ptr [[TMP15]], align 8
43; CHECK-NEXT:    [[TMP16:%.*]] = fadd <vscale x 8 x double> [[WIDE_LOAD]], splat (double 1.000000e+00)
44; CHECK-NEXT:    [[TMP17:%.*]] = fadd <vscale x 8 x double> [[WIDE_LOAD1]], splat (double 1.000000e+00)
45; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds double, ptr [[A:%.*]], i64 [[TMP8]]
46; CHECK-NEXT:    [[TMP19:%.*]] = sub i64 1, [[TMP5]]
47; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds double, ptr [[TMP18]], i64 [[TMP19]]
48; CHECK-NEXT:    [[TMP21:%.*]] = sub i64 0, [[TMP5]]
49; CHECK-NEXT:    [[TMP22:%.*]] = sub i64 1, [[TMP5]]
50; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds double, ptr [[TMP18]], i64 [[TMP21]]
51; CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds double, ptr [[TMP23]], i64 [[TMP22]]
52; CHECK-NEXT:    store <vscale x 8 x double> [[TMP16]], ptr [[TMP20]], align 8
53; CHECK-NEXT:    store <vscale x 8 x double> [[TMP17]], ptr [[TMP24]], align 8
54; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP6]]
55; CHECK-NEXT:    [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
56; CHECK-NEXT:    br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
57; CHECK:       middle.block:
58; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
59; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
60; CHECK:       scalar.ph:
61; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_MOD_VF]], [[MIDDLE_BLOCK]] ], [ [[N]], [[FOR_BODY_PREHEADER]] ]
62; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
63; CHECK:       for.cond.cleanup.loopexit:
64; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
65; CHECK:       for.cond.cleanup:
66; CHECK-NEXT:    ret void
67; CHECK:       for.body:
68; CHECK-NEXT:    [[I_08_IN:%.*]] = phi i64 [ [[I_08:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
69; CHECK-NEXT:    [[I_08]] = add nsw i64 [[I_08_IN]], -1
70; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[B]], i64 [[I_08]]
71; CHECK-NEXT:    [[TMP26:%.*]] = load double, ptr [[ARRAYIDX]], align 8
72; CHECK-NEXT:    [[ADD:%.*]] = fadd double [[TMP26]], 1.000000e+00
73; CHECK-NEXT:    [[ARRAYIDX1:%.*]] = getelementptr inbounds double, ptr [[A]], i64 [[I_08]]
74; CHECK-NEXT:    store double [[ADD]], ptr [[ARRAYIDX1]], align 8
75; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i64 [[I_08_IN]], 1
76; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP4:![0-9]+]]
77;
78entry:
79  %cmp7 = icmp sgt i64 %N, 0
80  br i1 %cmp7, label %for.body, label %for.cond.cleanup
81
82for.cond.cleanup:                                 ; preds = %for.body
83  ret void
84
85for.body:                                         ; preds = %entry, %for.body
86  %i.08.in = phi i64 [ %i.08, %for.body ], [ %N, %entry ]
87  %i.08 = add nsw i64 %i.08.in, -1
88  %arrayidx = getelementptr inbounds double, ptr %b, i64 %i.08
89  %0 = load double, ptr %arrayidx, align 8
90  %add = fadd double %0, 1.000000e+00
91  %arrayidx1 = getelementptr inbounds double, ptr %a, i64 %i.08
92  store double %add, ptr %arrayidx1, align 8
93  %cmp = icmp sgt i64 %i.08.in, 1
94  br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
95}
96
97
98define void @vector_reverse_i64(i64 %N, ptr %a, ptr %b) #0 {
99; CHECK-LABEL: @vector_reverse_i64(
100; CHECK-NEXT:  entry:
101; CHECK-NEXT:    [[A2:%.*]] = ptrtoint ptr [[A:%.*]] to i64
102; CHECK-NEXT:    [[B1:%.*]] = ptrtoint ptr [[B:%.*]] to i64
103; CHECK-NEXT:    [[CMP8:%.*]] = icmp sgt i64 [[N:%.*]], 0
104; CHECK-NEXT:    br i1 [[CMP8]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
105; CHECK:       for.body.preheader:
106; CHECK-NEXT:    [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
107; CHECK-NEXT:    [[TMP1:%.*]] = shl i64 [[TMP0]], 4
108; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
109; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
110; CHECK:       vector.memcheck:
111; CHECK-NEXT:    [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
112; CHECK-NEXT:    [[TMP3:%.*]] = shl i64 [[TMP2]], 7
113; CHECK-NEXT:    [[TMP4:%.*]] = sub i64 [[B1]], [[A2]]
114; CHECK-NEXT:    [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP4]], [[TMP3]]
115; CHECK-NEXT:    br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
116; CHECK:       vector.ph:
117; CHECK-NEXT:    [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
118; CHECK-NEXT:    [[TMP6:%.*]] = shl i64 [[TMP5]], 4
119; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP6]]
120; CHECK-NEXT:    [[N_VEC:%.*]] = sub nsw i64 [[N]], [[N_MOD_VF]]
121; CHECK-NEXT:    [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
122; CHECK-NEXT:    [[TMP8:%.*]] = shl i64 [[TMP7]], 3
123; CHECK-NEXT:    [[TMP9:%.*]] = shl i64 [[TMP7]], 4
124; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
125; CHECK:       vector.body:
126; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
127; CHECK-NEXT:    [[TMP10:%.*]] = xor i64 [[INDEX]], -1
128; CHECK-NEXT:    [[TMP11:%.*]] = add i64 [[N]], [[TMP10]]
129; CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP11]]
130; CHECK-NEXT:    [[TMP13:%.*]] = sub i64 1, [[TMP8]]
131; CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds i64, ptr [[TMP12]], i64 [[TMP13]]
132; CHECK-NEXT:    [[TMP15:%.*]] = sub i64 0, [[TMP8]]
133; CHECK-NEXT:    [[TMP16:%.*]] = sub i64 1, [[TMP8]]
134; CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds i64, ptr [[TMP12]], i64 [[TMP15]]
135; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds i64, ptr [[TMP17]], i64 [[TMP16]]
136; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 8 x i64>, ptr [[TMP14]], align 8
137; CHECK-NEXT:    [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i64>, ptr [[TMP18]], align 8
138; CHECK-NEXT:    [[TMP19:%.*]] = add <vscale x 8 x i64> [[WIDE_LOAD]], splat (i64 1)
139; CHECK-NEXT:    [[TMP20:%.*]] = add <vscale x 8 x i64> [[WIDE_LOAD3]], splat (i64 1)
140; CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP11]]
141; CHECK-NEXT:    [[TMP22:%.*]] = sub i64 1, [[TMP8]]
142; CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds i64, ptr [[TMP21]], i64 [[TMP22]]
143; CHECK-NEXT:    [[TMP24:%.*]] = sub i64 0, [[TMP8]]
144; CHECK-NEXT:    [[TMP25:%.*]] = sub i64 1, [[TMP8]]
145; CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds i64, ptr [[TMP21]], i64 [[TMP24]]
146; CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds i64, ptr [[TMP26]], i64 [[TMP25]]
147; CHECK-NEXT:    store <vscale x 8 x i64> [[TMP19]], ptr [[TMP23]], align 8
148; CHECK-NEXT:    store <vscale x 8 x i64> [[TMP20]], ptr [[TMP27]], align 8
149; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP9]]
150; CHECK-NEXT:    [[TMP28:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
151; CHECK-NEXT:    br i1 [[TMP28]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
152; CHECK:       middle.block:
153; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
154; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
155; CHECK:       scalar.ph:
156; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_MOD_VF]], [[MIDDLE_BLOCK]] ], [ [[N]], [[VECTOR_MEMCHECK]] ], [ [[N]], [[FOR_BODY_PREHEADER]] ]
157; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
158; CHECK:       for.cond.cleanup.loopexit:
159; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
160; CHECK:       for.cond.cleanup:
161; CHECK-NEXT:    ret void
162; CHECK:       for.body:
163; CHECK-NEXT:    [[I_09_IN:%.*]] = phi i64 [ [[I_09:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
164; CHECK-NEXT:    [[I_09]] = add nsw i64 [[I_09_IN]], -1
165; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[I_09]]
166; CHECK-NEXT:    [[TMP29:%.*]] = load i64, ptr [[ARRAYIDX]], align 8
167; CHECK-NEXT:    [[ADD:%.*]] = add i64 [[TMP29]], 1
168; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[I_09]]
169; CHECK-NEXT:    store i64 [[ADD]], ptr [[ARRAYIDX2]], align 8
170; CHECK-NEXT:    [[CMP:%.*]] = icmp sgt i64 [[I_09_IN]], 1
171; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP6:![0-9]+]]
172;
173entry:
174  %cmp8 = icmp sgt i64 %N, 0
175  br i1 %cmp8, label %for.body, label %for.cond.cleanup
176
177for.cond.cleanup:                                 ; preds = %for.body
178  ret void
179
180for.body:                                         ; preds = %entry, %for.body
181  %i.09.in = phi i64 [ %i.09, %for.body ], [ %N, %entry ]
182  %i.09 = add nsw i64 %i.09.in, -1
183  %arrayidx = getelementptr inbounds i64, ptr %b, i64 %i.09
184  %0 = load i64, ptr %arrayidx, align 8
185  %add = add i64 %0, 1
186  %arrayidx2 = getelementptr inbounds i64, ptr %a, i64 %i.09
187  store i64 %add, ptr %arrayidx2, align 8
188  %cmp = icmp sgt i64 %i.09.in, 1
189  br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
190}
191
192attributes #0 = { "target-cpu"="generic" "target-features"="+neon,+sve" }
193
194!0 = distinct !{!0, !1, !2, !3, !4, !5}
195!1 = !{!"llvm.loop.mustprogress"}
196!2 = !{!"llvm.loop.vectorize.width", i32 8}
197!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
198!4 = !{!"llvm.loop.vectorize.enable", i1 true}
199!5 = !{!"llvm.loop.interleave.count", i32 2}
200
201