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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
| ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=CHECK --check-prefix=AVX
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=AVX2
; Check constant loads of every 128-bit and 256-bit vector type
; for size optimization using splat ops available with AVX and AVX2.
; There is no AVX broadcast from double to 128-bit vector because movddup has been around since SSE3 (grrr).
define <2 x double> @splat_v2f64(<2 x double> %x) #0 {
; CHECK-LABEL: splat_v2f64:
; CHECK: # %bb.0:
; CHECK-NEXT: vmovddup {{.*#+}} xmm1 = [1.0E+0,1.0E+0]
; CHECK-NEXT: # xmm1 = mem[0,0]
; CHECK-NEXT: vaddpd %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%add = fadd <2 x double> %x, <double 1.0, double 1.0>
ret <2 x double> %add
}
define <4 x double> @splat_v4f64(<4 x double> %x) #1 {
; CHECK-LABEL: splat_v4f64:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastsd {{.*#+}} ymm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; CHECK-NEXT: vaddpd %ymm1, %ymm0, %ymm0
; CHECK-NEXT: retq
%add = fadd <4 x double> %x, <double 1.0, double 1.0, double 1.0, double 1.0>
ret <4 x double> %add
}
define <4 x float> @splat_v4f32(<4 x float> %x) #0 {
; CHECK-LABEL: splat_v4f32:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastss {{.*#+}} xmm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; CHECK-NEXT: vaddps %xmm1, %xmm0, %xmm0
; CHECK-NEXT: retq
%add = fadd <4 x float> %x, <float 1.0, float 1.0, float 1.0, float 1.0>
ret <4 x float> %add
}
define <8 x float> @splat_v8f32(<8 x float> %x) #1 {
; CHECK-LABEL: splat_v8f32:
; CHECK: # %bb.0:
; CHECK-NEXT: vbroadcastss {{.*#+}} ymm1 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0,1.0E+0]
; CHECK-NEXT: vaddps %ymm1, %ymm0, %ymm0
; CHECK-NEXT: retq
%add = fadd <8 x float> %x, <float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0, float 1.0>
ret <8 x float> %add
}
; AVX can't do integer splats, so fake it: use vmovddup to splat 64-bit value.
; We also generate vmovddup for AVX2 because it's one byte smaller than vpbroadcastq.
define <2 x i64> @splat_v2i64(<2 x i64> %x) #1 {
; AVX-LABEL: splat_v2i64:
; AVX: # %bb.0:
; AVX-NEXT: vmovddup {{.*#+}} xmm1 = [2,2]
; AVX-NEXT: # xmm1 = mem[0,0]
; AVX-NEXT: vpaddq %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v2i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm1 = [2,2]
; AVX2-NEXT: vpaddq %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
%add = add <2 x i64> %x, <i64 2, i64 2>
ret <2 x i64> %add
}
; AVX can't do 256-bit integer ops, so we split this into two 128-bit vectors,
; and then we fake it: use vmovddup to splat 64-bit value.
define <4 x i64> @splat_v4i64(<4 x i64> %x) #0 {
; AVX-LABEL: splat_v4i64:
; AVX: # %bb.0:
; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX-NEXT: vmovddup {{.*#+}} xmm2 = [2,2]
; AVX-NEXT: # xmm2 = mem[0,0]
; AVX-NEXT: vpaddq %xmm2, %xmm1, %xmm1
; AVX-NEXT: vpaddq %xmm2, %xmm0, %xmm0
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v4i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastq {{.*#+}} ymm1 = [2,2,2,2]
; AVX2-NEXT: vpaddq %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
%add = add <4 x i64> %x, <i64 2, i64 2, i64 2, i64 2>
ret <4 x i64> %add
}
; AVX can't do integer splats, so fake it: use vbroadcastss to splat 32-bit value.
define <4 x i32> @splat_v4i32(<4 x i32> %x) #1 {
; AVX-LABEL: splat_v4i32:
; AVX: # %bb.0:
; AVX-NEXT: vbroadcastss {{.*#+}} xmm1 = [2,2,2,2]
; AVX-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v4i32:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastd {{.*#+}} xmm1 = [2,2,2,2]
; AVX2-NEXT: vpaddd %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
%add = add <4 x i32> %x, <i32 2, i32 2, i32 2, i32 2>
ret <4 x i32> %add
}
; AVX can't do integer splats, so fake it: use vbroadcastss to splat 32-bit value.
define <8 x i32> @splat_v8i32(<8 x i32> %x) #0 {
; AVX-LABEL: splat_v8i32:
; AVX: # %bb.0:
; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX-NEXT: vbroadcastss {{.*#+}} xmm2 = [2,2,2,2]
; AVX-NEXT: vpaddd %xmm2, %xmm1, %xmm1
; AVX-NEXT: vpaddd %xmm2, %xmm0, %xmm0
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v8i32:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastd {{.*#+}} ymm1 = [2,2,2,2,2,2,2,2]
; AVX2-NEXT: vpaddd %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
%add = add <8 x i32> %x, <i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2, i32 2>
ret <8 x i32> %add
}
; AVX can't do integer splats, and there's no broadcast fakery for 16-bit. Could use pshuflw, etc?
define <8 x i16> @splat_v8i16(<8 x i16> %x) #1 {
; AVX-LABEL: splat_v8i16:
; AVX: # %bb.0:
; AVX-NEXT: vpaddw {{.*}}(%rip), %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v8i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastw {{.*#+}} xmm1 = [2,2,2,2,2,2,2,2]
; AVX2-NEXT: vpaddw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
%add = add <8 x i16> %x, <i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2>
ret <8 x i16> %add
}
; AVX can't do integer splats, and there's no broadcast fakery for 16-bit. Could use pshuflw, etc?
define <16 x i16> @splat_v16i16(<16 x i16> %x) #0 {
; AVX-LABEL: splat_v16i16:
; AVX: # %bb.0:
; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [2,2,2,2,2,2,2,2]
; AVX-NEXT: vpaddw %xmm2, %xmm1, %xmm1
; AVX-NEXT: vpaddw %xmm2, %xmm0, %xmm0
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v16i16:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastw {{.*#+}} ymm1 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; AVX2-NEXT: vpaddw %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
%add = add <16 x i16> %x, <i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2, i16 2>
ret <16 x i16> %add
}
; AVX can't do integer splats, and there's no broadcast fakery for 8-bit. Could use pshufb, etc?
define <16 x i8> @splat_v16i8(<16 x i8> %x) #1 {
; AVX-LABEL: splat_v16i8:
; AVX: # %bb.0:
; AVX-NEXT: vpaddb {{.*}}(%rip), %xmm0, %xmm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v16i8:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastb {{.*#+}} xmm1 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; AVX2-NEXT: vpaddb %xmm1, %xmm0, %xmm0
; AVX2-NEXT: retq
%add = add <16 x i8> %x, <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>
ret <16 x i8> %add
}
; AVX can't do integer splats, and there's no broadcast fakery for 8-bit. Could use pshufb, etc?
define <32 x i8> @splat_v32i8(<32 x i8> %x) #0 {
; AVX-LABEL: splat_v32i8:
; AVX: # %bb.0:
; AVX-NEXT: vextractf128 $1, %ymm0, %xmm1
; AVX-NEXT: vmovdqa {{.*#+}} xmm2 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; AVX-NEXT: vpaddb %xmm2, %xmm1, %xmm1
; AVX-NEXT: vpaddb %xmm2, %xmm0, %xmm0
; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX-NEXT: retq
;
; AVX2-LABEL: splat_v32i8:
; AVX2: # %bb.0:
; AVX2-NEXT: vpbroadcastb {{.*#+}} ymm1 = [2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2]
; AVX2-NEXT: vpaddb %ymm1, %ymm0, %ymm0
; AVX2-NEXT: retq
%add = add <32 x i8> %x, <i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2, i8 2>
ret <32 x i8> %add
}
; PR23259: Verify that ISel doesn't crash with a 'fatal error in backend'
; due to a missing AVX pattern to select a v2i64 X86ISD::BROADCAST of a
; loadi64 with multiple uses.
@A = common global <3 x i64> zeroinitializer, align 32
define <8 x i64> @pr23259() #1 {
; AVX-LABEL: pr23259:
; AVX: # %bb.0: # %entry
; AVX-NEXT: pushq $1
; AVX-NEXT: .cfi_adjust_cfa_offset 8
; AVX-NEXT: popq %rax
; AVX-NEXT: .cfi_adjust_cfa_offset -8
; AVX-NEXT: vmovq %rax, %xmm0
; AVX-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX-NEXT: vperm2f128 {{.*#+}} ymm0 = mem[2,3],ymm0[0,1]
; AVX-NEXT: vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; AVX-NEXT: vbroadcastsd {{.*#+}} ymm1 = [1,1,1,1]
; AVX-NEXT: retq
;
; AVX2-LABEL: pr23259:
; AVX2: # %bb.0: # %entry
; AVX2-NEXT: vmovdqa {{.*}}(%rip), %ymm0
; AVX2-NEXT: pushq $1
; AVX2-NEXT: .cfi_adjust_cfa_offset 8
; AVX2-NEXT: popq %rax
; AVX2-NEXT: .cfi_adjust_cfa_offset -8
; AVX2-NEXT: vmovq %rax, %xmm1
; AVX2-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; AVX2-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,1,1]
; AVX2-NEXT: vbroadcastsd {{.*#+}} ymm1 = [1,1,1,1]
; AVX2-NEXT: retq
entry:
%0 = load <4 x i64>, <4 x i64>* bitcast (<3 x i64>* @A to <4 x i64>*), align 32
%1 = shufflevector <4 x i64> %0, <4 x i64> undef, <3 x i32> <i32 undef, i32 undef, i32 2>
%shuffle = shufflevector <3 x i64> <i64 1, i64 undef, i64 undef>, <3 x i64> %1, <8 x i32> <i32 5, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
ret <8 x i64> %shuffle
}
attributes #0 = { optsize }
attributes #1 = { minsize }
|