| blob | 13bbe5e579812407a9adad4b1485b45e57272069 |
1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -reassociate -slp-vectorizer -slp-vectorize-hor -slp-vectorize-hor-store -S < %s -mtriple=x86_64-apple-macosx -mcpu=corei7-avx -mattr=+avx2 | FileCheck %s
4 define signext i8 @Foo(<32 x i8>* %__v) {
5 ; CHECK-LABEL: @Foo(
6 ; CHECK-NEXT: entry:
7 ; CHECK-NEXT: [[TMP0:%.*]] = load <32 x i8>, <32 x i8>* [[__V:%.*]], align 32
8 ; CHECK-NEXT: [[ADD_I_1_I:%.*]] = add i8 undef, undef
9 ; CHECK-NEXT: [[ADD_I_2_I:%.*]] = add i8 [[ADD_I_1_I]], undef
10 ; CHECK-NEXT: [[ADD_I_3_I:%.*]] = add i8 [[ADD_I_2_I]], undef
11 ; CHECK-NEXT: [[ADD_I_4_I:%.*]] = add i8 [[ADD_I_3_I]], undef
12 ; CHECK-NEXT: [[ADD_I_5_I:%.*]] = add i8 [[ADD_I_4_I]], undef
13 ; CHECK-NEXT: [[ADD_I_6_I:%.*]] = add i8 [[ADD_I_5_I]], undef
14 ; CHECK-NEXT: [[ADD_I_7_I:%.*]] = add i8 [[ADD_I_6_I]], undef
15 ; CHECK-NEXT: [[ADD_I_8_I:%.*]] = add i8 [[ADD_I_7_I]], undef
16 ; CHECK-NEXT: [[ADD_I_9_I:%.*]] = add i8 [[ADD_I_8_I]], undef
17 ; CHECK-NEXT: [[ADD_I_10_I:%.*]] = add i8 [[ADD_I_9_I]], undef
18 ; CHECK-NEXT: [[ADD_I_11_I:%.*]] = add i8 [[ADD_I_10_I]], undef
19 ; CHECK-NEXT: [[ADD_I_12_I:%.*]] = add i8 [[ADD_I_11_I]], undef
20 ; CHECK-NEXT: [[ADD_I_13_I:%.*]] = add i8 [[ADD_I_12_I]], undef
21 ; CHECK-NEXT: [[ADD_I_14_I:%.*]] = add i8 [[ADD_I_13_I]], undef
22 ; CHECK-NEXT: [[ADD_I_15_I:%.*]] = add i8 [[ADD_I_14_I]], undef
23 ; CHECK-NEXT: [[ADD_I_16_I:%.*]] = add i8 [[ADD_I_15_I]], undef
24 ; CHECK-NEXT: [[ADD_I_17_I:%.*]] = add i8 [[ADD_I_16_I]], undef
25 ; CHECK-NEXT: [[ADD_I_18_I:%.*]] = add i8 [[ADD_I_17_I]], undef
26 ; CHECK-NEXT: [[ADD_I_19_I:%.*]] = add i8 [[ADD_I_18_I]], undef
27 ; CHECK-NEXT: [[ADD_I_20_I:%.*]] = add i8 [[ADD_I_19_I]], undef
28 ; CHECK-NEXT: [[ADD_I_21_I:%.*]] = add i8 [[ADD_I_20_I]], undef
29 ; CHECK-NEXT: [[ADD_I_22_I:%.*]] = add i8 [[ADD_I_21_I]], undef
30 ; CHECK-NEXT: [[ADD_I_23_I:%.*]] = add i8 [[ADD_I_22_I]], undef
31 ; CHECK-NEXT: [[ADD_I_24_I:%.*]] = add i8 [[ADD_I_23_I]], undef
32 ; CHECK-NEXT: [[ADD_I_25_I:%.*]] = add i8 [[ADD_I_24_I]], undef
33 ; CHECK-NEXT: [[ADD_I_26_I:%.*]] = add i8 [[ADD_I_25_I]], undef
34 ; CHECK-NEXT: [[ADD_I_27_I:%.*]] = add i8 [[ADD_I_26_I]], undef
35 ; CHECK-NEXT: [[ADD_I_28_I:%.*]] = add i8 [[ADD_I_27_I]], undef
36 ; CHECK-NEXT: [[ADD_I_29_I:%.*]] = add i8 [[ADD_I_28_I]], undef
37 ; CHECK-NEXT: [[ADD_I_30_I:%.*]] = add i8 [[ADD_I_29_I]], undef
38 ; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <32 x i8> [[TMP0]], <32 x i8> undef, <32 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
39 ; CHECK-NEXT: [[BIN_RDX:%.*]] = add <32 x i8> [[TMP0]], [[RDX_SHUF]]
40 ; CHECK-NEXT: [[RDX_SHUF1:%.*]] = shufflevector <32 x i8> [[BIN_RDX]], <32 x i8> undef, <32 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
41 ; CHECK-NEXT: [[BIN_RDX2:%.*]] = add <32 x i8> [[BIN_RDX]], [[RDX_SHUF1]]
42 ; CHECK-NEXT: [[RDX_SHUF3:%.*]] = shufflevector <32 x i8> [[BIN_RDX2]], <32 x i8> undef, <32 x i32> <i32 4, i32 5, i32 6, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
43 ; CHECK-NEXT: [[BIN_RDX4:%.*]] = add <32 x i8> [[BIN_RDX2]], [[RDX_SHUF3]]
44 ; CHECK-NEXT: [[RDX_SHUF5:%.*]] = shufflevector <32 x i8> [[BIN_RDX4]], <32 x i8> undef, <32 x i32> <i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
45 ; CHECK-NEXT: [[BIN_RDX6:%.*]] = add <32 x i8> [[BIN_RDX4]], [[RDX_SHUF5]]
46 ; CHECK-NEXT: [[RDX_SHUF7:%.*]] = shufflevector <32 x i8> [[BIN_RDX6]], <32 x i8> undef, <32 x i32> <i32 1, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef, i32 undef>
47 ; CHECK-NEXT: [[BIN_RDX8:%.*]] = add <32 x i8> [[BIN_RDX6]], [[RDX_SHUF7]]
48 ; CHECK-NEXT: [[TMP1:%.*]] = extractelement <32 x i8> [[BIN_RDX8]], i32 0
49 ; CHECK-NEXT: [[ADD_I_31_I:%.*]] = add i8 [[ADD_I_30_I]], undef
50 ; CHECK-NEXT: ret i8 [[TMP1]]
51 ;
52 entry:
53 %0 = load <32 x i8>, <32 x i8>* %__v, align 32
54 %vecext.i.i.i = extractelement <32 x i8> %0, i64 0
55 %vecext.i.i.1.i = extractelement <32 x i8> %0, i64 1
56 %add.i.1.i = add i8 %vecext.i.i.1.i, %vecext.i.i.i
57 %vecext.i.i.2.i = extractelement <32 x i8> %0, i64 2
58 %add.i.2.i = add i8 %vecext.i.i.2.i, %add.i.1.i
59 %vecext.i.i.3.i = extractelement <32 x i8> %0, i64 3
60 %add.i.3.i = add i8 %vecext.i.i.3.i, %add.i.2.i
61 %vecext.i.i.4.i = extractelement <32 x i8> %0, i64 4
62 %add.i.4.i = add i8 %vecext.i.i.4.i, %add.i.3.i
63 %vecext.i.i.5.i = extractelement <32 x i8> %0, i64 5
64 %add.i.5.i = add i8 %vecext.i.i.5.i, %add.i.4.i
65 %vecext.i.i.6.i = extractelement <32 x i8> %0, i64 6
66 %add.i.6.i = add i8 %vecext.i.i.6.i, %add.i.5.i
67 %vecext.i.i.7.i = extractelement <32 x i8> %0, i64 7
68 %add.i.7.i = add i8 %vecext.i.i.7.i, %add.i.6.i
69 %vecext.i.i.8.i = extractelement <32 x i8> %0, i64 8
70 %add.i.8.i = add i8 %vecext.i.i.8.i, %add.i.7.i
71 %vecext.i.i.9.i = extractelement <32 x i8> %0, i64 9
72 %add.i.9.i = add i8 %vecext.i.i.9.i, %add.i.8.i
73 %vecext.i.i.10.i = extractelement <32 x i8> %0, i64 10
74 %add.i.10.i = add i8 %vecext.i.i.10.i, %add.i.9.i
75 %vecext.i.i.11.i = extractelement <32 x i8> %0, i64 11
76 %add.i.11.i = add i8 %vecext.i.i.11.i, %add.i.10.i
77 %vecext.i.i.12.i = extractelement <32 x i8> %0, i64 12
78 %add.i.12.i = add i8 %vecext.i.i.12.i, %add.i.11.i
79 %vecext.i.i.13.i = extractelement <32 x i8> %0, i64 13
80 %add.i.13.i = add i8 %vecext.i.i.13.i, %add.i.12.i
81 %vecext.i.i.14.i = extractelement <32 x i8> %0, i64 14
82 %add.i.14.i = add i8 %vecext.i.i.14.i, %add.i.13.i
83 %vecext.i.i.15.i = extractelement <32 x i8> %0, i64 15
84 %add.i.15.i = add i8 %vecext.i.i.15.i, %add.i.14.i
85 %vecext.i.i.16.i = extractelement <32 x i8> %0, i64 16
86 %add.i.16.i = add i8 %vecext.i.i.16.i, %add.i.15.i
87 %vecext.i.i.17.i = extractelement <32 x i8> %0, i64 17
88 %add.i.17.i = add i8 %vecext.i.i.17.i, %add.i.16.i
89 %vecext.i.i.18.i = extractelement <32 x i8> %0, i64 18
90 %add.i.18.i = add i8 %vecext.i.i.18.i, %add.i.17.i
91 %vecext.i.i.19.i = extractelement <32 x i8> %0, i64 19
92 %add.i.19.i = add i8 %vecext.i.i.19.i, %add.i.18.i
93 %vecext.i.i.20.i = extractelement <32 x i8> %0, i64 20
94 %add.i.20.i = add i8 %vecext.i.i.20.i, %add.i.19.i
95 %vecext.i.i.21.i = extractelement <32 x i8> %0, i64 21
96 %add.i.21.i = add i8 %vecext.i.i.21.i, %add.i.20.i
97 %vecext.i.i.22.i = extractelement <32 x i8> %0, i64 22
98 %add.i.22.i = add i8 %vecext.i.i.22.i, %add.i.21.i
99 %vecext.i.i.23.i = extractelement <32 x i8> %0, i64 23
100 %add.i.23.i = add i8 %vecext.i.i.23.i, %add.i.22.i
101 %vecext.i.i.24.i = extractelement <32 x i8> %0, i64 24
102 %add.i.24.i = add i8 %vecext.i.i.24.i, %add.i.23.i
103 %vecext.i.i.25.i = extractelement <32 x i8> %0, i64 25
104 %add.i.25.i = add i8 %vecext.i.i.25.i, %add.i.24.i
105 %vecext.i.i.26.i = extractelement <32 x i8> %0, i64 26
106 %add.i.26.i = add i8 %vecext.i.i.26.i, %add.i.25.i
107 %vecext.i.i.27.i = extractelement <32 x i8> %0, i64 27
108 %add.i.27.i = add i8 %vecext.i.i.27.i, %add.i.26.i
109 %vecext.i.i.28.i = extractelement <32 x i8> %0, i64 28
110 %add.i.28.i = add i8 %vecext.i.i.28.i, %add.i.27.i
111 %vecext.i.i.29.i = extractelement <32 x i8> %0, i64 29
112 %add.i.29.i = add i8 %vecext.i.i.29.i, %add.i.28.i
113 %vecext.i.i.30.i = extractelement <32 x i8> %0, i64 30
114 %add.i.30.i = add i8 %vecext.i.i.30.i, %add.i.29.i
115 %vecext.i.i.31.i = extractelement <32 x i8> %0, i64 31
116 %add.i.31.i = add i8 %vecext.i.i.31.i, %add.i.30.i
117 ret i8 %add.i.31.i
118 }