1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -slp-vectorizer -instcombine -S -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 | FileCheck %s
3
4define float @dotf(<4 x float> %x, <4 x float> %y) {
5; CHECK-LABEL: @dotf(
6; CHECK-NEXT:  entry:
7; CHECK-NEXT:    [[TMP0:%.*]] = fmul fast <4 x float> [[X:%.*]], [[Y:%.*]]
8; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
9; CHECK-NEXT:    [[BIN_RDX:%.*]] = fadd fast <4 x float> [[TMP0]], [[RDX_SHUF]]
10; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x float> [[BIN_RDX]], <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
11; CHECK-NEXT:    [[BIN_RDX2:%.*]] = fadd fast <4 x float> [[BIN_RDX]], [[RDX_SHUF1]]
12; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[BIN_RDX2]], i32 0
13; CHECK-NEXT:    ret float [[TMP1]]
14;
15entry:
16  %vecext = extractelement <4 x float> %x, i32 0
17  %vecext1 = extractelement <4 x float> %y, i32 0
18  %mul = fmul fast float %vecext, %vecext1
19  %vecext.1 = extractelement <4 x float> %x, i32 1
20  %vecext1.1 = extractelement <4 x float> %y, i32 1
21  %mul.1 = fmul fast float %vecext.1, %vecext1.1
22  %add.1 = fadd fast float %mul.1, %mul
23  %vecext.2 = extractelement <4 x float> %x, i32 2
24  %vecext1.2 = extractelement <4 x float> %y, i32 2
25  %mul.2 = fmul fast float %vecext.2, %vecext1.2
26  %add.2 = fadd fast float %mul.2, %add.1
27  %vecext.3 = extractelement <4 x float> %x, i32 3
28  %vecext1.3 = extractelement <4 x float> %y, i32 3
29  %mul.3 = fmul fast float %vecext.3, %vecext1.3
30  %add.3 = fadd fast float %mul.3, %add.2
31  ret float %add.3
32}
33
34define double @dotd(<4 x double>* byval nocapture readonly align 32, <4 x double>* byval nocapture readonly align 32) {
35; CHECK-LABEL: @dotd(
36; CHECK-NEXT:  entry:
37; CHECK-NEXT:    [[X:%.*]] = load <4 x double>, <4 x double>* [[TMP0:%.*]], align 32
38; CHECK-NEXT:    [[Y:%.*]] = load <4 x double>, <4 x double>* [[TMP1:%.*]], align 32
39; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast <4 x double> [[X]], [[Y]]
40; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x double> [[TMP2]], <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
41; CHECK-NEXT:    [[BIN_RDX:%.*]] = fadd fast <4 x double> [[TMP2]], [[RDX_SHUF]]
42; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x double> [[BIN_RDX]], <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
43; CHECK-NEXT:    [[BIN_RDX2:%.*]] = fadd fast <4 x double> [[BIN_RDX]], [[RDX_SHUF1]]
44; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x double> [[BIN_RDX2]], i32 0
45; CHECK-NEXT:    ret double [[TMP3]]
46;
47entry:
48  %x = load <4 x double>, <4 x double>* %0, align 32
49  %y = load <4 x double>, <4 x double>* %1, align 32
50  %vecext = extractelement <4 x double> %x, i32 0
51  %vecext1 = extractelement <4 x double> %y, i32 0
52  %mul = fmul fast double %vecext, %vecext1
53  %vecext.1 = extractelement <4 x double> %x, i32 1
54  %vecext1.1 = extractelement <4 x double> %y, i32 1
55  %mul.1 = fmul fast double %vecext.1, %vecext1.1
56  %add.1 = fadd fast double %mul.1, %mul
57  %vecext.2 = extractelement <4 x double> %x, i32 2
58  %vecext1.2 = extractelement <4 x double> %y, i32 2
59  %mul.2 = fmul fast double %vecext.2, %vecext1.2
60  %add.2 = fadd fast double %mul.2, %add.1
61  %vecext.3 = extractelement <4 x double> %x, i32 3
62  %vecext1.3 = extractelement <4 x double> %y, i32 3
63  %mul.3 = fmul fast double %vecext.3, %vecext1.3
64  %add.3 = fadd fast double %mul.3, %add.2
65  ret double %add.3
66}
67
68define float @dotfq(<4 x float>* nocapture readonly %x, <4 x float>* nocapture readonly %y) {
69; CHECK-LABEL: @dotfq(
70; CHECK-NEXT:  entry:
71; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x float>, <4 x float>* [[X:%.*]], align 16
72; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[Y:%.*]], align 16
73; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast <4 x float> [[TMP1]], [[TMP0]]
74; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
75; CHECK-NEXT:    [[BIN_RDX:%.*]] = fadd fast <4 x float> [[TMP2]], [[RDX_SHUF]]
76; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x float> [[BIN_RDX]], <4 x float> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
77; CHECK-NEXT:    [[BIN_RDX2:%.*]] = fadd fast <4 x float> [[BIN_RDX]], [[RDX_SHUF1]]
78; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[BIN_RDX2]], i32 0
79; CHECK-NEXT:    ret float [[TMP3]]
80;
81entry:
82  %0 = load <4 x float>, <4 x float>* %x, align 16
83  %1 = load <4 x float>, <4 x float>* %y, align 16
84  %vecext = extractelement <4 x float> %0, i32 0
85  %vecext1 = extractelement <4 x float> %1, i32 0
86  %mul = fmul fast float %vecext1, %vecext
87  %vecext.1 = extractelement <4 x float> %0, i32 1
88  %vecext1.1 = extractelement <4 x float> %1, i32 1
89  %mul.1 = fmul fast float %vecext1.1, %vecext.1
90  %add.1 = fadd fast float %mul.1, %mul
91  %vecext.2 = extractelement <4 x float> %0, i32 2
92  %vecext1.2 = extractelement <4 x float> %1, i32 2
93  %mul.2 = fmul fast float %vecext1.2, %vecext.2
94  %add.2 = fadd fast float %mul.2, %add.1
95  %vecext.3 = extractelement <4 x float> %0, i32 3
96  %vecext1.3 = extractelement <4 x float> %1, i32 3
97  %mul.3 = fmul fast float %vecext1.3, %vecext.3
98  %add.3 = fadd fast float %mul.3, %add.2
99  ret float %add.3
100}
101
102define double @dotdq(<4 x double>* nocapture readonly %x, <4 x double>* nocapture readonly %y) {
103; CHECK-LABEL: @dotdq(
104; CHECK-NEXT:  entry:
105; CHECK-NEXT:    [[TMP0:%.*]] = load <4 x double>, <4 x double>* [[X:%.*]], align 32
106; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x double>, <4 x double>* [[Y:%.*]], align 32
107; CHECK-NEXT:    [[TMP2:%.*]] = fmul fast <4 x double> [[TMP1]], [[TMP0]]
108; CHECK-NEXT:    [[RDX_SHUF:%.*]] = shufflevector <4 x double> [[TMP2]], <4 x double> undef, <4 x i32> <i32 2, i32 3, i32 undef, i32 undef>
109; CHECK-NEXT:    [[BIN_RDX:%.*]] = fadd fast <4 x double> [[TMP2]], [[RDX_SHUF]]
110; CHECK-NEXT:    [[RDX_SHUF1:%.*]] = shufflevector <4 x double> [[BIN_RDX]], <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
111; CHECK-NEXT:    [[BIN_RDX2:%.*]] = fadd fast <4 x double> [[BIN_RDX]], [[RDX_SHUF1]]
112; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x double> [[BIN_RDX2]], i32 0
113; CHECK-NEXT:    ret double [[TMP3]]
114;
115entry:
116  %0 = load <4 x double>, <4 x double>* %x, align 32
117  %1 = load <4 x double>, <4 x double>* %y, align 32
118  %vecext = extractelement <4 x double> %0, i32 0
119  %vecext1 = extractelement <4 x double> %1, i32 0
120  %mul = fmul fast double %vecext1, %vecext
121  %vecext.1 = extractelement <4 x double> %0, i32 1
122  %vecext1.1 = extractelement <4 x double> %1, i32 1
123  %mul.1 = fmul fast double %vecext1.1, %vecext.1
124  %add.1 = fadd fast double %mul.1, %mul
125  %vecext.2 = extractelement <4 x double> %0, i32 2
126  %vecext1.2 = extractelement <4 x double> %1, i32 2
127  %mul.2 = fmul fast double %vecext1.2, %vecext.2
128  %add.2 = fadd fast double %mul.2, %add.1
129  %vecext.3 = extractelement <4 x double> %0, i32 3
130  %vecext1.3 = extractelement <4 x double> %1, i32 3
131  %mul.3 = fmul fast double %vecext1.3, %vecext.3
132  %add.3 = fadd fast double %mul.3, %add.2
133  ret double %add.3
134}
135