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
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
|
; RUN: llc < %s -march=x86-64 -mcpu=core2 -mattr=+sse2 | FileCheck %s
; RUN: llc < %s -march=x86-64 -mcpu=core2 -mattr=+sse2 -x86-experimental-vector-widening-legalization | FileCheck %s --check-prefix=CHECK-WIDE
define double @test1(double %A) {
%1 = bitcast double %A to <2 x i32>
%add = add <2 x i32> %1, <i32 3, i32 5>
%2 = bitcast <2 x i32> %add to double
ret double %2
}
; FIXME: Ideally we should be able to fold the entire body of @test1 into a
; single paddd instruction. At the moment we produce the sequence
; pshufd+paddq+pshufd. This is fixed with the widening legalization.
;
; CHECK-LABEL: test1
; CHECK-NOT: movsd
; CHECK: pshufd
; CHECK-NEXT: paddd
; CHECK-NEXT: pshufd
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test1
; CHECK-WIDE-NOT: movsd
; CHECK-WIDE: paddd
; CHECK-WIDE-NEXT: ret
define double @test2(double %A, double %B) {
%1 = bitcast double %A to <2 x i32>
%2 = bitcast double %B to <2 x i32>
%add = add <2 x i32> %1, %2
%3 = bitcast <2 x i32> %add to double
ret double %3
}
; CHECK-LABEL: test2
; CHECK-NOT: movsd
; CHECK: paddd
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test2
; CHECK-WIDE-NOT: movsd
; CHECK-WIDE: paddd
; CHECK-WIDE-NEXT: ret
define i64 @test3(i64 %A) {
%1 = bitcast i64 %A to <2 x float>
%add = fadd <2 x float> %1, <float 3.0, float 5.0>
%2 = bitcast <2 x float> %add to i64
ret i64 %2
}
; CHECK-LABEL: test3
; CHECK-NOT: pshufd
; CHECK: addps
; CHECK-NOT: pshufd
; CHECK: ret
;
; CHECK-WIDE-LABEL: test3
; CHECK-WIDE-NOT: pshufd
; CHECK-WIDE: addps
; CHECK-WIDE-NOT: pshufd
; CHECK-WIDE: ret
define i64 @test4(i64 %A) {
%1 = bitcast i64 %A to <2 x i32>
%add = add <2 x i32> %1, <i32 3, i32 5>
%2 = bitcast <2 x i32> %add to i64
ret i64 %2
}
; FIXME: At the moment we still produce the sequence pshufd+paddq+pshufd.
; Ideally, we should fold that sequence into a single paddd. This is fixed with
; the widening legalization.
;
; CHECK-LABEL: test4
; CHECK: pshufd
; CHECK-NEXT: paddq
; CHECK-NEXT: pshufd
; CHECK: ret
;
; CHECK-WIDE-LABEL: test4
; CHECK-WIDE: movd %{{rdi|rcx}},
; CHECK-WIDE-NEXT: paddd
; CHECK-WIDE-NEXT: movd {{.*}}, %rax
; CHECK-WIDE: ret
define double @test5(double %A) {
%1 = bitcast double %A to <2 x float>
%add = fadd <2 x float> %1, <float 3.0, float 5.0>
%2 = bitcast <2 x float> %add to double
ret double %2
}
; CHECK-LABEL: test5
; CHECK: addps
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test5
; CHECK-WIDE: addps
; CHECK-WIDE-NEXT: ret
define double @test6(double %A) {
%1 = bitcast double %A to <4 x i16>
%add = add <4 x i16> %1, <i16 3, i16 4, i16 5, i16 6>
%2 = bitcast <4 x i16> %add to double
ret double %2
}
; FIXME: Ideally we should be able to fold the entire body of @test6 into a
; single paddw instruction. This is fixed with the widening legalization.
;
; CHECK-LABEL: test6
; CHECK-NOT: movsd
; CHECK: punpcklwd
; CHECK-NEXT: paddw
; CHECK-NEXT: pshufb
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test6
; CHECK-WIDE-NOT: mov
; CHECK-WIDE-NOT: punpcklwd
; CHECK-WIDE: paddw
; CHECK-WIDE-NEXT: ret
define double @test7(double %A, double %B) {
%1 = bitcast double %A to <4 x i16>
%2 = bitcast double %B to <4 x i16>
%add = add <4 x i16> %1, %2
%3 = bitcast <4 x i16> %add to double
ret double %3
}
; CHECK-LABEL: test7
; CHECK-NOT: movsd
; CHECK-NOT: punpcklwd
; CHECK: paddw
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test7
; CHECK-WIDE-NOT: movsd
; CHECK-WIDE-NOT: punpcklwd
; CHECK-WIDE: paddw
; CHECK-WIDE-NEXT: ret
define double @test8(double %A) {
%1 = bitcast double %A to <8 x i8>
%add = add <8 x i8> %1, <i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10>
%2 = bitcast <8 x i8> %add to double
ret double %2
}
; FIXME: Ideally we should be able to fold the entire body of @test8 into a
; single paddb instruction. At the moment we produce the sequence
; pshufd+paddw+pshufd. This is fixed with the widening legalization.
;
; CHECK-LABEL: test8
; CHECK-NOT: movsd
; CHECK: punpcklbw
; CHECK-NEXT: paddb
; CHECK-NEXT: pshufb
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test8
; CHECK-WIDE-NOT: movsd
; CHECK-WIDE-NOT: punpcklbw
; CHECK-WIDE: paddb
; CHECK-WIDE-NEXT: ret
define double @test9(double %A, double %B) {
%1 = bitcast double %A to <8 x i8>
%2 = bitcast double %B to <8 x i8>
%add = add <8 x i8> %1, %2
%3 = bitcast <8 x i8> %add to double
ret double %3
}
; CHECK-LABEL: test9
; CHECK-NOT: movsd
; CHECK-NOT: punpcklbw
; CHECK: paddb
; CHECK-NEXT: ret
;
; CHECK-WIDE-LABEL: test9
; CHECK-WIDE-NOT: movsd
; CHECK-WIDE-NOT: punpcklbw
; CHECK-WIDE: paddb
; CHECK-WIDE-NEXT: ret
|
