/
copy_amd64.s
153 lines (135 loc) · 2.51 KB
/
copy_amd64.s
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
// Code generated by command: go run copy_asm.go -pkg mem -out ../mem/copy_amd64.s -stubs ../mem/copy_amd64.go. DO NOT EDIT.
//go:build !purego
#include "textflag.h"
// func Copy(dst []byte, src []byte) int
// Requires: AVX, CMOV, SSE2
TEXT ·Copy(SB), NOSPLIT, $0-56
MOVQ dst_base+0(FP), AX
MOVQ src_base+24(FP), CX
MOVQ dst_len+8(FP), DX
MOVQ src_len+32(FP), BX
CMPQ BX, DX
CMOVQLT BX, DX
MOVQ DX, ret+48(FP)
tail:
CMPQ DX, $0x00
JE done
CMPQ DX, $0x01
JE handle1
CMPQ DX, $0x03
JBE handle2to3
CMPQ DX, $0x04
JE handle4
CMPQ DX, $0x08
JB handle5to7
JE handle8
CMPQ DX, $0x10
JBE handle9to16
CMPQ DX, $0x20
JBE handle17to32
CMPQ DX, $0x40
JBE handle33to64
BTL $0x08, github·com∕segmentio∕asm∕cpu·X86+0(SB)
JCC generic
CMPQ DX, $0x00000080
JB avx2_tail
JMP avx2
generic:
MOVOU (CX), X0
MOVOU 16(CX), X1
MOVOU 32(CX), X2
MOVOU 48(CX), X3
MOVOU X0, (AX)
MOVOU X1, 16(AX)
MOVOU X2, 32(AX)
MOVOU X3, 48(AX)
ADDQ $0x40, CX
ADDQ $0x40, AX
SUBQ $0x40, DX
CMPQ DX, $0x40
JBE tail
JMP generic
done:
RET
handle1:
MOVB (CX), CL
MOVB CL, (AX)
RET
handle2to3:
MOVW (CX), BX
MOVW -2(CX)(DX*1), CX
MOVW BX, (AX)
MOVW CX, -2(AX)(DX*1)
RET
handle4:
MOVL (CX), CX
MOVL CX, (AX)
RET
handle5to7:
MOVL (CX), BX
MOVL -4(CX)(DX*1), CX
MOVL BX, (AX)
MOVL CX, -4(AX)(DX*1)
RET
handle8:
MOVQ (CX), CX
MOVQ CX, (AX)
RET
handle9to16:
MOVQ (CX), BX
MOVQ -8(CX)(DX*1), CX
MOVQ BX, (AX)
MOVQ CX, -8(AX)(DX*1)
RET
handle17to32:
MOVOU (CX), X0
MOVOU -16(CX)(DX*1), X1
MOVOU X0, (AX)
MOVOU X1, -16(AX)(DX*1)
RET
handle33to64:
MOVOU (CX), X0
MOVOU 16(CX), X1
MOVOU -32(CX)(DX*1), X2
MOVOU -16(CX)(DX*1), X3
MOVOU X0, (AX)
MOVOU X1, 16(AX)
MOVOU X2, -32(AX)(DX*1)
MOVOU X3, -16(AX)(DX*1)
RET
// AVX optimized version for medium to large size inputs.
avx2:
VMOVDQU (CX), Y0
VMOVDQU 32(CX), Y1
VMOVDQU 64(CX), Y2
VMOVDQU 96(CX), Y3
VMOVDQU Y0, (AX)
VMOVDQU Y1, 32(AX)
VMOVDQU Y2, 64(AX)
VMOVDQU Y3, 96(AX)
ADDQ $0x00000080, CX
ADDQ $0x00000080, AX
SUBQ $0x00000080, DX
JZ avx2_done
CMPQ DX, $0x00000080
JAE avx2
avx2_tail:
CMPQ DX, $0x40
JBE avx2_tail_1to64
VMOVDQU (CX), Y0
VMOVDQU 32(CX), Y1
VMOVDQU -64(CX)(DX*1), Y2
VMOVDQU -32(CX)(DX*1), Y3
VMOVDQU Y0, (AX)
VMOVDQU Y1, 32(AX)
VMOVDQU Y2, -64(AX)(DX*1)
VMOVDQU Y3, -32(AX)(DX*1)
JMP avx2_done
avx2_tail_1to64:
VMOVDQU -64(CX)(DX*1), Y0
VMOVDQU -32(CX)(DX*1), Y1
VMOVDQU Y0, -64(AX)(DX*1)
VMOVDQU Y1, -32(AX)(DX*1)
avx2_done:
VZEROUPPER
RET