xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/redux-feed-buildvector.ll (revision 38fffa630ee80163dc65e759392ad29798905679) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -mtriple=x86_64 -passes=slp-vectorizer -S -mcpu=skylake-avx512 | FileCheck %s
3
4; The test represents the case with multiple vectorization possibilities
5; but the most effective way to vectorize it is to match both 8-way reductions
6; feeding the insertelement vector build sequence.
7
8declare void @llvm.masked.scatter.v2f64.v2p0(<2 x double>, <2 x ptr>, i32 immarg, <2 x i1>)
9
10define void @test(ptr nocapture readonly %arg, ptr nocapture readonly %arg1, ptr nocapture %arg2) {
11; CHECK-LABEL: @test(
12; CHECK-NEXT:  entry:
13; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <8 x ptr> poison, ptr [[ARG:%.*]], i32 0
14; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <8 x ptr> [[TMP0]], <8 x ptr> poison, <8 x i32> zeroinitializer
15; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr double, <8 x ptr> [[TMP1]], <8 x i64> <i64 1, i64 3, i64 5, i64 7, i64 9, i64 11, i64 13, i64 15>
16; CHECK-NEXT:    [[GEP2_0:%.*]] = getelementptr inbounds double, ptr [[ARG1:%.*]], i64 16
17; CHECK-NEXT:    [[TMP3:%.*]] = call <8 x double> @llvm.masked.gather.v8f64.v8p0(<8 x ptr> [[TMP2]], i32 8, <8 x i1> splat (i1 true), <8 x double> poison)
18; CHECK-NEXT:    [[TMP4:%.*]] = load <8 x double>, ptr [[GEP2_0]], align 8
19; CHECK-NEXT:    [[TMP5:%.*]] = fmul fast <8 x double> [[TMP4]], [[TMP3]]
20; CHECK-NEXT:    [[TMP6:%.*]] = load <8 x double>, ptr [[ARG1]], align 8
21; CHECK-NEXT:    [[TMP7:%.*]] = fmul fast <8 x double> [[TMP6]], [[TMP3]]
22; CHECK-NEXT:    [[TMP8:%.*]] = call fast double @llvm.vector.reduce.fadd.v8f64(double 0.000000e+00, <8 x double> [[TMP7]])
23; CHECK-NEXT:    [[TMP9:%.*]] = call fast double @llvm.vector.reduce.fadd.v8f64(double 0.000000e+00, <8 x double> [[TMP5]])
24; CHECK-NEXT:    [[I142:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i64 0
25; CHECK-NEXT:    [[I143:%.*]] = insertelement <2 x double> [[I142]], double [[TMP9]], i64 1
26; CHECK-NEXT:    [[P:%.*]] = getelementptr inbounds double, ptr [[ARG2:%.*]], <2 x i64> <i64 0, i64 16>
27; CHECK-NEXT:    call void @llvm.masked.scatter.v2f64.v2p0(<2 x double> [[I143]], <2 x ptr> [[P]], i32 8, <2 x i1> splat (i1 true))
28; CHECK-NEXT:    ret void
29;
30entry:
31  %gep1.0 = getelementptr inbounds double, ptr %arg, i64 1
32  %ld1.0 = load double, ptr %gep1.0, align 8
33  %ld0.0 = load double, ptr %arg1, align 8
34  %mul1.0 = fmul fast double %ld0.0, %ld1.0
35  %gep2.0 = getelementptr inbounds double, ptr %arg1, i64 16
36  %ld2.0 = load double, ptr %gep2.0, align 8
37  %mul2.0 = fmul fast double %ld2.0, %ld1.0
38  %gep1.1 = getelementptr inbounds double, ptr %arg, i64 3
39  %ld1.1 = load double, ptr %gep1.1, align 8
40  %gep0.1 = getelementptr inbounds double, ptr %arg1, i64 1
41  %ld0.1 = load double, ptr %gep0.1, align 8
42  %mul1.1 = fmul fast double %ld0.1, %ld1.1
43  %rdx1.0 = fadd fast double %mul1.0, %mul1.1
44  %gep2.1 = getelementptr inbounds double, ptr %arg1, i64 17
45  %ld2.1 = load double, ptr %gep2.1, align 8
46  %mul2.1 = fmul fast double %ld2.1, %ld1.1
47  %rdx2.0 = fadd fast double %mul2.0, %mul2.1
48  %gep1.2 = getelementptr inbounds double, ptr %arg, i64 5
49  %ld1.2 = load double, ptr %gep1.2, align 8
50  %gep0.2 = getelementptr inbounds double, ptr %arg1, i64 2
51  %ld0.2 = load double, ptr %gep0.2, align 8
52  %mul1.2 = fmul fast double %ld0.2, %ld1.2
53  %rdx1.1 = fadd fast double %rdx1.0, %mul1.2
54  %gep2.2 = getelementptr inbounds double, ptr %arg1, i64 18
55  %ld2.2 = load double, ptr %gep2.2, align 8
56  %mul2.2 = fmul fast double %ld2.2, %ld1.2
57  %rdx2.1 = fadd fast double %rdx2.0, %mul2.2
58  %gep1.3 = getelementptr inbounds double, ptr %arg, i64 7
59  %ld1.3 = load double, ptr %gep1.3, align 8
60  %gep0.3 = getelementptr inbounds double, ptr %arg1, i64 3
61  %ld0.3 = load double, ptr %gep0.3, align 8
62  %mul1.3 = fmul fast double %ld0.3, %ld1.3
63  %rdx1.2 = fadd fast double %rdx1.1, %mul1.3
64  %gep2.3 = getelementptr inbounds double, ptr %arg1, i64 19
65  %ld2.3 = load double, ptr %gep2.3, align 8
66  %mul2.3 = fmul fast double %ld2.3, %ld1.3
67  %rdx2.2 = fadd fast double %rdx2.1, %mul2.3
68  %gep1.4 = getelementptr inbounds double, ptr %arg, i64 9
69  %ld1.4 = load double, ptr %gep1.4, align 8
70  %gep0.4 = getelementptr inbounds double, ptr %arg1, i64 4
71  %ld0.4 = load double, ptr %gep0.4, align 8
72  %mul1.4 = fmul fast double %ld0.4, %ld1.4
73  %rdx1.3 = fadd fast double %rdx1.2, %mul1.4
74  %gep2.4 = getelementptr inbounds double, ptr %arg1, i64 20
75  %ld2.4 = load double, ptr %gep2.4, align 8
76  %mul2.4 = fmul fast double %ld2.4, %ld1.4
77  %rdx2.3 = fadd fast double %rdx2.2, %mul2.4
78  %gep1.5 = getelementptr inbounds double, ptr %arg, i64 11
79  %ld1.5 = load double, ptr %gep1.5, align 8
80  %gep0.5 = getelementptr inbounds double, ptr %arg1, i64 5
81  %ld0.5 = load double, ptr %gep0.5, align 8
82  %mul1.5 = fmul fast double %ld0.5, %ld1.5
83  %rdx1.4 = fadd fast double %rdx1.3, %mul1.5
84  %gep2.5 = getelementptr inbounds double, ptr %arg1, i64 21
85  %ld2.5 = load double, ptr %gep2.5, align 8
86  %mul2.5 = fmul fast double %ld2.5, %ld1.5
87  %rdx2.4 = fadd fast double %rdx2.3, %mul2.5
88  %gep1.6 = getelementptr inbounds double, ptr %arg, i64 13
89  %ld1.6 = load double, ptr %gep1.6, align 8
90  %gep0.6 = getelementptr inbounds double, ptr %arg1, i64 6
91  %ld0.6 = load double, ptr %gep0.6, align 8
92  %mul1.6 = fmul fast double %ld0.6, %ld1.6
93  %rdx1.5 = fadd fast double %rdx1.4, %mul1.6
94  %gep2.6 = getelementptr inbounds double, ptr %arg1, i64 22
95  %ld2.6 = load double, ptr %gep2.6, align 8
96  %mul2.6 = fmul fast double %ld2.6, %ld1.6
97  %rdx2.5 = fadd fast double %rdx2.4, %mul2.6
98  %gep1.7 = getelementptr inbounds double, ptr %arg, i64 15
99  %ld1.7 = load double, ptr %gep1.7, align 8
100  %gep0.7 = getelementptr inbounds double, ptr %arg1, i64 7
101  %ld0.7 = load double, ptr %gep0.7, align 8
102  %mul1.7 = fmul fast double %ld0.7, %ld1.7
103  %rdx1 = fadd fast double %rdx1.5, %mul1.7
104  %gep2.7 = getelementptr inbounds double, ptr %arg1, i64 23
105  %ld2.7 = load double, ptr %gep2.7, align 8
106  %mul2.7 = fmul fast double %ld2.7, %ld1.7
107  %rdx2 = fadd fast double %rdx2.5, %mul2.7
108  %i142 = insertelement <2 x double> poison, double %rdx1, i64 0
109  %i143 = insertelement <2 x double> %i142, double %rdx2, i64 1
110  %p = getelementptr inbounds double, ptr %arg2, <2 x i64> <i64 0, i64 16>
111  call void @llvm.masked.scatter.v2f64.v2p0(<2 x double> %i143, <2 x ptr> %p, i32 8, <2 x i1> <i1 true, i1 true>)
112  ret void
113}
114