1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -O3 -S                                        | FileCheck %s
3; RUN: opt < %s -passes='default<O3>' -S | FileCheck %s
4
5; Test that IR is optimal after vectorization/unrolling/CSE/canonicalization.
6; In particular, there should be no fdivs inside loops because that is expensive.
7
8; TODO: There is a CSE opportunity to reduce the hoisted fdivs after vectorization/unrolling.
9; PR46115 - https://llvm.org/PR46115
10
11target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
12target triple = "x86_64-apple-macosx10.15.0"
13
14define void @vdiv(double* %x, double* %y, double %a, i32 %N) #0 {
15; CHECK-LABEL: @vdiv(
16; CHECK-NEXT:  entry:
17; CHECK-NEXT:    [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0
18; CHECK-NEXT:    br i1 [[CMP1]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
19; CHECK:       for.body.preheader:
20; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
21; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 16
22; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY_PREHEADER17:%.*]], label [[VECTOR_MEMCHECK:%.*]]
23; CHECK:       vector.memcheck:
24; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr double, double* [[X:%.*]], i64 [[WIDE_TRIP_COUNT]]
25; CHECK-NEXT:    [[SCEVGEP6:%.*]] = getelementptr double, double* [[Y:%.*]], i64 [[WIDE_TRIP_COUNT]]
26; CHECK-NEXT:    [[BOUND0:%.*]] = icmp ugt double* [[SCEVGEP6]], [[X]]
27; CHECK-NEXT:    [[BOUND1:%.*]] = icmp ugt double* [[SCEVGEP]], [[Y]]
28; CHECK-NEXT:    [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
29; CHECK-NEXT:    br i1 [[FOUND_CONFLICT]], label [[FOR_BODY_PREHEADER17]], label [[VECTOR_PH:%.*]]
30; CHECK:       vector.ph:
31; CHECK-NEXT:    [[N_VEC:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 4294967280
32; CHECK-NEXT:    [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x double> poison, double [[A:%.*]], i32 0
33; CHECK-NEXT:    [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x double> [[BROADCAST_SPLATINSERT]], <4 x double> poison, <4 x i32> zeroinitializer
34; CHECK-NEXT:    [[BROADCAST_SPLATINSERT11:%.*]] = insertelement <4 x double> poison, double [[A]], i32 0
35; CHECK-NEXT:    [[BROADCAST_SPLAT12:%.*]] = shufflevector <4 x double> [[BROADCAST_SPLATINSERT11]], <4 x double> poison, <4 x i32> zeroinitializer
36; CHECK-NEXT:    [[BROADCAST_SPLATINSERT13:%.*]] = insertelement <4 x double> poison, double [[A]], i32 0
37; CHECK-NEXT:    [[BROADCAST_SPLAT14:%.*]] = shufflevector <4 x double> [[BROADCAST_SPLATINSERT13]], <4 x double> poison, <4 x i32> zeroinitializer
38; CHECK-NEXT:    [[BROADCAST_SPLATINSERT15:%.*]] = insertelement <4 x double> poison, double [[A]], i32 0
39; CHECK-NEXT:    [[BROADCAST_SPLAT16:%.*]] = shufflevector <4 x double> [[BROADCAST_SPLATINSERT15]], <4 x double> poison, <4 x i32> zeroinitializer
40; CHECK-NEXT:    [[TMP0:%.*]] = fdiv fast <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, [[BROADCAST_SPLAT]]
41; CHECK-NEXT:    [[TMP1:%.*]] = fdiv fast <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, [[BROADCAST_SPLAT12]]
42; CHECK-NEXT:    [[TMP2:%.*]] = fdiv fast <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, [[BROADCAST_SPLAT14]]
43; CHECK-NEXT:    [[TMP3:%.*]] = fdiv fast <4 x double> <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00>, [[BROADCAST_SPLAT16]]
44; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
45; CHECK:       vector.body:
46; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
47; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDEX]]
48; CHECK-NEXT:    [[TMP5:%.*]] = bitcast double* [[TMP4]] to <4 x double>*
49; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x double>, <4 x double>* [[TMP5]], align 8, !tbaa [[TBAA3:![0-9]+]], !alias.scope !7
50; CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds double, double* [[TMP4]], i64 4
51; CHECK-NEXT:    [[TMP7:%.*]] = bitcast double* [[TMP6]] to <4 x double>*
52; CHECK-NEXT:    [[WIDE_LOAD8:%.*]] = load <4 x double>, <4 x double>* [[TMP7]], align 8, !tbaa [[TBAA3]], !alias.scope !7
53; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds double, double* [[TMP4]], i64 8
54; CHECK-NEXT:    [[TMP9:%.*]] = bitcast double* [[TMP8]] to <4 x double>*
55; CHECK-NEXT:    [[WIDE_LOAD9:%.*]] = load <4 x double>, <4 x double>* [[TMP9]], align 8, !tbaa [[TBAA3]], !alias.scope !7
56; CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds double, double* [[TMP4]], i64 12
57; CHECK-NEXT:    [[TMP11:%.*]] = bitcast double* [[TMP10]] to <4 x double>*
58; CHECK-NEXT:    [[WIDE_LOAD10:%.*]] = load <4 x double>, <4 x double>* [[TMP11]], align 8, !tbaa [[TBAA3]], !alias.scope !7
59; CHECK-NEXT:    [[TMP12:%.*]] = fmul fast <4 x double> [[WIDE_LOAD]], [[TMP0]]
60; CHECK-NEXT:    [[TMP13:%.*]] = fmul fast <4 x double> [[WIDE_LOAD8]], [[TMP1]]
61; CHECK-NEXT:    [[TMP14:%.*]] = fmul fast <4 x double> [[WIDE_LOAD9]], [[TMP2]]
62; CHECK-NEXT:    [[TMP15:%.*]] = fmul fast <4 x double> [[WIDE_LOAD10]], [[TMP3]]
63; CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDEX]]
64; CHECK-NEXT:    [[TMP17:%.*]] = bitcast double* [[TMP16]] to <4 x double>*
65; CHECK-NEXT:    store <4 x double> [[TMP12]], <4 x double>* [[TMP17]], align 8, !tbaa [[TBAA3]], !alias.scope !10, !noalias !7
66; CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds double, double* [[TMP16]], i64 4
67; CHECK-NEXT:    [[TMP19:%.*]] = bitcast double* [[TMP18]] to <4 x double>*
68; CHECK-NEXT:    store <4 x double> [[TMP13]], <4 x double>* [[TMP19]], align 8, !tbaa [[TBAA3]], !alias.scope !10, !noalias !7
69; CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds double, double* [[TMP16]], i64 8
70; CHECK-NEXT:    [[TMP21:%.*]] = bitcast double* [[TMP20]] to <4 x double>*
71; CHECK-NEXT:    store <4 x double> [[TMP14]], <4 x double>* [[TMP21]], align 8, !tbaa [[TBAA3]], !alias.scope !10, !noalias !7
72; CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds double, double* [[TMP16]], i64 12
73; CHECK-NEXT:    [[TMP23:%.*]] = bitcast double* [[TMP22]] to <4 x double>*
74; CHECK-NEXT:    store <4 x double> [[TMP15]], <4 x double>* [[TMP23]], align 8, !tbaa [[TBAA3]], !alias.scope !10, !noalias !7
75; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
76; CHECK-NEXT:    [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
77; CHECK-NEXT:    br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]]
78; CHECK:       middle.block:
79; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[WIDE_TRIP_COUNT]]
80; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_END]], label [[FOR_BODY_PREHEADER17]]
81; CHECK:       for.body.preheader17:
82; CHECK-NEXT:    [[INDVARS_IV_PH:%.*]] = phi i64 [ 0, [[VECTOR_MEMCHECK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
83; CHECK-NEXT:    [[TMP25:%.*]] = xor i64 [[INDVARS_IV_PH]], -1
84; CHECK-NEXT:    [[TMP26:%.*]] = add nsw i64 [[TMP25]], [[WIDE_TRIP_COUNT]]
85; CHECK-NEXT:    [[XTRAITER:%.*]] = and i64 [[WIDE_TRIP_COUNT]], 3
86; CHECK-NEXT:    [[LCMP_MOD_NOT:%.*]] = icmp eq i64 [[XTRAITER]], 0
87; CHECK-NEXT:    br i1 [[LCMP_MOD_NOT]], label [[FOR_BODY_PROL_LOOPEXIT:%.*]], label [[FOR_BODY_PROL_PREHEADER:%.*]]
88; CHECK:       for.body.prol.preheader:
89; CHECK-NEXT:    [[TMP27:%.*]] = fdiv fast double 1.000000e+00, [[A]]
90; CHECK-NEXT:    br label [[FOR_BODY_PROL:%.*]]
91; CHECK:       for.body.prol:
92; CHECK-NEXT:    [[INDVARS_IV_PROL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_PROL:%.*]], [[FOR_BODY_PROL]] ], [ [[INDVARS_IV_PH]], [[FOR_BODY_PROL_PREHEADER]] ]
93; CHECK-NEXT:    [[PROL_ITER:%.*]] = phi i64 [ [[PROL_ITER_SUB:%.*]], [[FOR_BODY_PROL]] ], [ [[XTRAITER]], [[FOR_BODY_PROL_PREHEADER]] ]
94; CHECK-NEXT:    [[ARRAYIDX_PROL:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDVARS_IV_PROL]]
95; CHECK-NEXT:    [[T0_PROL:%.*]] = load double, double* [[ARRAYIDX_PROL]], align 8, !tbaa [[TBAA3]]
96; CHECK-NEXT:    [[TMP28:%.*]] = fmul fast double [[T0_PROL]], [[TMP27]]
97; CHECK-NEXT:    [[ARRAYIDX2_PROL:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDVARS_IV_PROL]]
98; CHECK-NEXT:    store double [[TMP28]], double* [[ARRAYIDX2_PROL]], align 8, !tbaa [[TBAA3]]
99; CHECK-NEXT:    [[INDVARS_IV_NEXT_PROL]] = add nuw nsw i64 [[INDVARS_IV_PROL]], 1
100; CHECK-NEXT:    [[PROL_ITER_SUB]] = add i64 [[PROL_ITER]], -1
101; CHECK-NEXT:    [[PROL_ITER_CMP_NOT:%.*]] = icmp eq i64 [[PROL_ITER_SUB]], 0
102; CHECK-NEXT:    br i1 [[PROL_ITER_CMP_NOT]], label [[FOR_BODY_PROL_LOOPEXIT]], label [[FOR_BODY_PROL]], !llvm.loop [[LOOP14:![0-9]+]]
103; CHECK:       for.body.prol.loopexit:
104; CHECK-NEXT:    [[INDVARS_IV_UNR:%.*]] = phi i64 [ [[INDVARS_IV_PH]], [[FOR_BODY_PREHEADER17]] ], [ [[INDVARS_IV_NEXT_PROL]], [[FOR_BODY_PROL]] ]
105; CHECK-NEXT:    [[TMP29:%.*]] = icmp ult i64 [[TMP26]], 3
106; CHECK-NEXT:    br i1 [[TMP29]], label [[FOR_END]], label [[FOR_BODY_PREHEADER17_NEW:%.*]]
107; CHECK:       for.body.preheader17.new:
108; CHECK-NEXT:    [[TMP30:%.*]] = fdiv fast double 1.000000e+00, [[A]]
109; CHECK-NEXT:    [[TMP31:%.*]] = fdiv fast double 1.000000e+00, [[A]]
110; CHECK-NEXT:    [[TMP32:%.*]] = fdiv fast double 1.000000e+00, [[A]]
111; CHECK-NEXT:    [[TMP33:%.*]] = fdiv fast double 1.000000e+00, [[A]]
112; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
113; CHECK:       for.body:
114; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_UNR]], [[FOR_BODY_PREHEADER17_NEW]] ], [ [[INDVARS_IV_NEXT_3:%.*]], [[FOR_BODY]] ]
115; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDVARS_IV]]
116; CHECK-NEXT:    [[T0:%.*]] = load double, double* [[ARRAYIDX]], align 8, !tbaa [[TBAA3]]
117; CHECK-NEXT:    [[TMP34:%.*]] = fmul fast double [[T0]], [[TMP30]]
118; CHECK-NEXT:    [[ARRAYIDX2:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDVARS_IV]]
119; CHECK-NEXT:    store double [[TMP34]], double* [[ARRAYIDX2]], align 8, !tbaa [[TBAA3]]
120; CHECK-NEXT:    [[INDVARS_IV_NEXT:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 1
121; CHECK-NEXT:    [[ARRAYIDX_1:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDVARS_IV_NEXT]]
122; CHECK-NEXT:    [[T0_1:%.*]] = load double, double* [[ARRAYIDX_1]], align 8, !tbaa [[TBAA3]]
123; CHECK-NEXT:    [[TMP35:%.*]] = fmul fast double [[T0_1]], [[TMP31]]
124; CHECK-NEXT:    [[ARRAYIDX2_1:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDVARS_IV_NEXT]]
125; CHECK-NEXT:    store double [[TMP35]], double* [[ARRAYIDX2_1]], align 8, !tbaa [[TBAA3]]
126; CHECK-NEXT:    [[INDVARS_IV_NEXT_1:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 2
127; CHECK-NEXT:    [[ARRAYIDX_2:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDVARS_IV_NEXT_1]]
128; CHECK-NEXT:    [[T0_2:%.*]] = load double, double* [[ARRAYIDX_2]], align 8, !tbaa [[TBAA3]]
129; CHECK-NEXT:    [[TMP36:%.*]] = fmul fast double [[T0_2]], [[TMP32]]
130; CHECK-NEXT:    [[ARRAYIDX2_2:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDVARS_IV_NEXT_1]]
131; CHECK-NEXT:    store double [[TMP36]], double* [[ARRAYIDX2_2]], align 8, !tbaa [[TBAA3]]
132; CHECK-NEXT:    [[INDVARS_IV_NEXT_2:%.*]] = add nuw nsw i64 [[INDVARS_IV]], 3
133; CHECK-NEXT:    [[ARRAYIDX_3:%.*]] = getelementptr inbounds double, double* [[Y]], i64 [[INDVARS_IV_NEXT_2]]
134; CHECK-NEXT:    [[T0_3:%.*]] = load double, double* [[ARRAYIDX_3]], align 8, !tbaa [[TBAA3]]
135; CHECK-NEXT:    [[TMP37:%.*]] = fmul fast double [[T0_3]], [[TMP33]]
136; CHECK-NEXT:    [[ARRAYIDX2_3:%.*]] = getelementptr inbounds double, double* [[X]], i64 [[INDVARS_IV_NEXT_2]]
137; CHECK-NEXT:    store double [[TMP37]], double* [[ARRAYIDX2_3]], align 8, !tbaa [[TBAA3]]
138; CHECK-NEXT:    [[INDVARS_IV_NEXT_3]] = add nuw nsw i64 [[INDVARS_IV]], 4
139; CHECK-NEXT:    [[EXITCOND_NOT_3:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT_3]], [[WIDE_TRIP_COUNT]]
140; CHECK-NEXT:    br i1 [[EXITCOND_NOT_3]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]]
141; CHECK:       for.end:
142; CHECK-NEXT:    ret void
143;
144entry:
145  %div = fdiv fast double 1.0, %a
146  br label %for.cond
147
148for.cond:
149  %n.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
150  %cmp = icmp slt i32 %n.0, %N
151  br i1 %cmp, label %for.body, label %for.cond.cleanup
152
153for.cond.cleanup:
154  br label %for.end
155
156for.body:
157  %idxprom = sext i32 %n.0 to i64
158  %arrayidx = getelementptr inbounds double, double* %y, i64 %idxprom
159  %t0 = load double, double* %arrayidx, align 8, !tbaa !3
160  %mul = fmul fast double %t0, %div
161  %idxprom1 = sext i32 %n.0 to i64
162  %arrayidx2 = getelementptr inbounds double, double* %x, i64 %idxprom1
163  store double %mul, double* %arrayidx2, align 8, !tbaa !3
164  br label %for.inc
165
166for.inc:
167  %inc = add nsw i32 %n.0, 1
168  br label %for.cond
169
170for.end:
171  ret void
172}
173
174attributes #0 = { nounwind ssp uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "frame-pointer"="all" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-infs-fp-math"="true" "no-jump-tables"="false" "no-nans-fp-math"="true" "no-signed-zeros-fp-math"="true" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="skylake-avx512" "target-features"="+adx,+aes,+avx,+avx2,+avx512bw,+avx512cd,+avx512dq,+avx512f,+avx512vl,+bmi,+bmi2,+clflushopt,+clwb,+cx16,+cx8,+f16c,+fma,+fsgsbase,+fxsr,+invpcid,+lzcnt,+mmx,+movbe,+pclmul,+pku,+popcnt,+prfchw,+rdrnd,+rdseed,+sahf,+sse,+sse2,+sse3,+sse4.1,+sse4.2,+ssse3,+x87,+xsave,+xsavec,+xsaveopt,+xsaves" "unsafe-fp-math"="true" "use-soft-float"="false" }
175
176!llvm.module.flags = !{!0, !1}
177!llvm.ident = !{!2}
178
179!0 = !{i32 1, !"wchar_size", i32 4}
180!1 = !{i32 7, !"PIC Level", i32 2}
181!2 = !{!"clang version 11.0.0 (https://github.com/llvm/llvm-project.git 45ebe38ffc40bb7221fc587bfb4481cf7f53ebbc)"}
182!3 = !{!4, !4, i64 0}
183!4 = !{!"double", !5, i64 0}
184!5 = !{!"omnipotent char", !6, i64 0}
185!6 = !{!"Simple C/C++ TBAA"}
186
187