commutativity.ll
9.78 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.11.0 -slp-vectorizer -S -mattr=+sse2 | FileCheck %s --check-prefixes=CHECK,SSE
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.11.0 -slp-vectorizer -S -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX,AVX1
; RUN: opt < %s -mtriple=x86_64-apple-macosx10.11.0 -slp-vectorizer -S -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX,AVX2
; Verify that the SLP vectorizer is able to figure out that commutativity
; offers the possibility to splat/broadcast %c and thus make it profitable
; to vectorize this case
@cle = external unnamed_addr global [32 x i8], align 16
@cle32 = external unnamed_addr global [32 x i32], align 16
; Check that we correctly detect a splat/broadcast by leveraging the
; commutativity property of `xor`.
define void @splat(i8 %a, i8 %b, i8 %c) {
; SSE-LABEL: @splat(
; SSE-NEXT: [[TMP1:%.*]] = xor i8 [[C:%.*]], [[A:%.*]]
; SSE-NEXT: store i8 [[TMP1]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 0), align 16
; SSE-NEXT: [[TMP2:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP2]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 1)
; SSE-NEXT: [[TMP3:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP3]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 2)
; SSE-NEXT: [[TMP4:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP4]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 3)
; SSE-NEXT: [[TMP5:%.*]] = xor i8 [[C]], [[A]]
; SSE-NEXT: store i8 [[TMP5]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 4)
; SSE-NEXT: [[TMP6:%.*]] = xor i8 [[C]], [[B:%.*]]
; SSE-NEXT: store i8 [[TMP6]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 5)
; SSE-NEXT: [[TMP7:%.*]] = xor i8 [[C]], [[A]]
; SSE-NEXT: store i8 [[TMP7]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 6)
; SSE-NEXT: [[TMP8:%.*]] = xor i8 [[C]], [[B]]
; SSE-NEXT: store i8 [[TMP8]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 7)
; SSE-NEXT: [[TMP9:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP9]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 8)
; SSE-NEXT: [[TMP10:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP10]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 9)
; SSE-NEXT: [[TMP11:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP11]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 10)
; SSE-NEXT: [[TMP12:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP12]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 11)
; SSE-NEXT: [[TMP13:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP13]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 12)
; SSE-NEXT: [[TMP14:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP14]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 13)
; SSE-NEXT: [[TMP15:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP15]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 14)
; SSE-NEXT: [[TMP16:%.*]] = xor i8 [[A]], [[C]]
; SSE-NEXT: store i8 [[TMP16]], i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 15)
; SSE-NEXT: ret void
;
; AVX-LABEL: @splat(
; AVX-NEXT: [[TMP1:%.*]] = insertelement <16 x i8> undef, i8 [[C:%.*]], i32 0
; AVX-NEXT: [[TMP2:%.*]] = insertelement <16 x i8> [[TMP1]], i8 [[C]], i32 1
; AVX-NEXT: [[TMP3:%.*]] = insertelement <16 x i8> [[TMP2]], i8 [[C]], i32 2
; AVX-NEXT: [[TMP4:%.*]] = insertelement <16 x i8> [[TMP3]], i8 [[C]], i32 3
; AVX-NEXT: [[TMP5:%.*]] = insertelement <16 x i8> [[TMP4]], i8 [[C]], i32 4
; AVX-NEXT: [[TMP6:%.*]] = insertelement <16 x i8> [[TMP5]], i8 [[C]], i32 5
; AVX-NEXT: [[TMP7:%.*]] = insertelement <16 x i8> [[TMP6]], i8 [[C]], i32 6
; AVX-NEXT: [[TMP8:%.*]] = insertelement <16 x i8> [[TMP7]], i8 [[C]], i32 7
; AVX-NEXT: [[TMP9:%.*]] = insertelement <16 x i8> [[TMP8]], i8 [[C]], i32 8
; AVX-NEXT: [[TMP10:%.*]] = insertelement <16 x i8> [[TMP9]], i8 [[C]], i32 9
; AVX-NEXT: [[TMP11:%.*]] = insertelement <16 x i8> [[TMP10]], i8 [[C]], i32 10
; AVX-NEXT: [[TMP12:%.*]] = insertelement <16 x i8> [[TMP11]], i8 [[C]], i32 11
; AVX-NEXT: [[TMP13:%.*]] = insertelement <16 x i8> [[TMP12]], i8 [[C]], i32 12
; AVX-NEXT: [[TMP14:%.*]] = insertelement <16 x i8> [[TMP13]], i8 [[C]], i32 13
; AVX-NEXT: [[TMP15:%.*]] = insertelement <16 x i8> [[TMP14]], i8 [[C]], i32 14
; AVX-NEXT: [[TMP16:%.*]] = insertelement <16 x i8> [[TMP15]], i8 [[C]], i32 15
; AVX-NEXT: [[TMP17:%.*]] = insertelement <2 x i8> undef, i8 [[A:%.*]], i32 0
; AVX-NEXT: [[TMP18:%.*]] = insertelement <2 x i8> [[TMP17]], i8 [[B:%.*]], i32 1
; AVX-NEXT: [[SHUFFLE:%.*]] = shufflevector <2 x i8> [[TMP18]], <2 x i8> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0, i32 1, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
; AVX-NEXT: [[TMP19:%.*]] = xor <16 x i8> [[TMP16]], [[SHUFFLE]]
; AVX-NEXT: store <16 x i8> [[TMP19]], <16 x i8>* bitcast ([32 x i8]* @cle to <16 x i8>*), align 16
; AVX-NEXT: ret void
;
%1 = xor i8 %c, %a
store i8 %1, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 0), align 16
%2 = xor i8 %a, %c
store i8 %2, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 1)
%3 = xor i8 %a, %c
store i8 %3, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 2)
%4 = xor i8 %a, %c
store i8 %4, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 3)
%5 = xor i8 %c, %a
store i8 %5, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 4)
%6 = xor i8 %c, %b
store i8 %6, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 5)
%7 = xor i8 %c, %a
store i8 %7, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 6)
%8 = xor i8 %c, %b
store i8 %8, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 7)
%9 = xor i8 %a, %c
store i8 %9, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 8)
%10 = xor i8 %a, %c
store i8 %10, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 9)
%11 = xor i8 %a, %c
store i8 %11, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 10)
%12 = xor i8 %a, %c
store i8 %12, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 11)
%13 = xor i8 %a, %c
store i8 %13, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 12)
%14 = xor i8 %a, %c
store i8 %14, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 13)
%15 = xor i8 %a, %c
store i8 %15, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 14)
%16 = xor i8 %a, %c
store i8 %16, i8* getelementptr inbounds ([32 x i8], [32 x i8]* @cle, i64 0, i64 15)
ret void
}
; Check that we correctly detect that we can have the same opcode on one side by
; leveraging the commutativity property of `xor`.
define void @same_opcode_on_one_side(i32 %a, i32 %b, i32 %c) {
; SSE-LABEL: @same_opcode_on_one_side(
; SSE-NEXT: [[ADD1:%.*]] = add i32 [[C:%.*]], [[A:%.*]]
; SSE-NEXT: [[ADD2:%.*]] = add i32 [[C]], [[A]]
; SSE-NEXT: [[ADD3:%.*]] = add i32 [[A]], [[C]]
; SSE-NEXT: [[ADD4:%.*]] = add i32 [[C]], [[A]]
; SSE-NEXT: [[TMP1:%.*]] = xor i32 [[ADD1]], [[A]]
; SSE-NEXT: store i32 [[TMP1]], i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 0), align 16
; SSE-NEXT: [[TMP2:%.*]] = xor i32 [[B:%.*]], [[ADD2]]
; SSE-NEXT: store i32 [[TMP2]], i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 1)
; SSE-NEXT: [[TMP3:%.*]] = xor i32 [[C]], [[ADD3]]
; SSE-NEXT: store i32 [[TMP3]], i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 2)
; SSE-NEXT: [[TMP4:%.*]] = xor i32 [[A]], [[ADD4]]
; SSE-NEXT: store i32 [[TMP4]], i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 3)
; SSE-NEXT: ret void
;
; AVX-LABEL: @same_opcode_on_one_side(
; AVX-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> undef, i32 [[C:%.*]], i32 0
; AVX-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> [[TMP1]], i32 [[C]], i32 1
; AVX-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> [[TMP2]], i32 [[C]], i32 2
; AVX-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> [[TMP3]], i32 [[C]], i32 3
; AVX-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> undef, i32 [[A:%.*]], i32 0
; AVX-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> [[TMP5]], i32 [[A]], i32 1
; AVX-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> [[TMP6]], i32 [[A]], i32 2
; AVX-NEXT: [[TMP8:%.*]] = insertelement <4 x i32> [[TMP7]], i32 [[A]], i32 3
; AVX-NEXT: [[TMP9:%.*]] = add <4 x i32> [[TMP4]], [[TMP8]]
; AVX-NEXT: [[TMP10:%.*]] = insertelement <4 x i32> [[TMP5]], i32 [[B:%.*]], i32 1
; AVX-NEXT: [[TMP11:%.*]] = insertelement <4 x i32> [[TMP10]], i32 [[C]], i32 2
; AVX-NEXT: [[TMP12:%.*]] = insertelement <4 x i32> [[TMP11]], i32 [[A]], i32 3
; AVX-NEXT: [[TMP13:%.*]] = xor <4 x i32> [[TMP9]], [[TMP12]]
; AVX-NEXT: store <4 x i32> [[TMP13]], <4 x i32>* bitcast ([32 x i32]* @cle32 to <4 x i32>*), align 16
; AVX-NEXT: ret void
;
%add1 = add i32 %c, %a
%add2 = add i32 %c, %a
%add3 = add i32 %a, %c
%add4 = add i32 %c, %a
%1 = xor i32 %add1, %a
store i32 %1, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 0), align 16
%2 = xor i32 %b, %add2
store i32 %2, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 1)
%3 = xor i32 %c, %add3
store i32 %3, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 2)
%4 = xor i32 %a, %add4
store i32 %4, i32* getelementptr inbounds ([32 x i32], [32 x i32]* @cle32, i64 0, i64 3)
ret void
}