Loading...
1/*
2 * Normally compiler builtins are used, but sometimes the compiler calls out
3 * of line code. Based on asm-i386/string.h.
4 *
5 * This assembly file is re-written from memmove_64.c file.
6 * - Copyright 2011 Fenghua Yu <fenghua.yu@intel.com>
7 */
8#define _STRING_C
9#include <linux/linkage.h>
10#include <asm/dwarf2.h>
11#include <asm/cpufeature.h>
12#include <asm/alternative-asm.h>
13
14#undef memmove
15
16/*
17 * Implement memmove(). This can handle overlap between src and dst.
18 *
19 * Input:
20 * rdi: dest
21 * rsi: src
22 * rdx: count
23 *
24 * Output:
25 * rax: dest
26 */
27ENTRY(memmove)
28 CFI_STARTPROC
29
30 /* Handle more 32 bytes in loop */
31 mov %rdi, %rax
32 cmp $0x20, %rdx
33 jb 1f
34
35 /* Decide forward/backward copy mode */
36 cmp %rdi, %rsi
37 jge .Lmemmove_begin_forward
38 mov %rsi, %r8
39 add %rdx, %r8
40 cmp %rdi, %r8
41 jg 2f
42
43.Lmemmove_begin_forward:
44 /*
45 * movsq instruction have many startup latency
46 * so we handle small size by general register.
47 */
48 cmp $680, %rdx
49 jb 3f
50 /*
51 * movsq instruction is only good for aligned case.
52 */
53
54 cmpb %dil, %sil
55 je 4f
563:
57 sub $0x20, %rdx
58 /*
59 * We gobble 32 bytes forward in each loop.
60 */
615:
62 sub $0x20, %rdx
63 movq 0*8(%rsi), %r11
64 movq 1*8(%rsi), %r10
65 movq 2*8(%rsi), %r9
66 movq 3*8(%rsi), %r8
67 leaq 4*8(%rsi), %rsi
68
69 movq %r11, 0*8(%rdi)
70 movq %r10, 1*8(%rdi)
71 movq %r9, 2*8(%rdi)
72 movq %r8, 3*8(%rdi)
73 leaq 4*8(%rdi), %rdi
74 jae 5b
75 addq $0x20, %rdx
76 jmp 1f
77 /*
78 * Handle data forward by movsq.
79 */
80 .p2align 4
814:
82 movq %rdx, %rcx
83 movq -8(%rsi, %rdx), %r11
84 lea -8(%rdi, %rdx), %r10
85 shrq $3, %rcx
86 rep movsq
87 movq %r11, (%r10)
88 jmp 13f
89.Lmemmove_end_forward:
90
91 /*
92 * Handle data backward by movsq.
93 */
94 .p2align 4
957:
96 movq %rdx, %rcx
97 movq (%rsi), %r11
98 movq %rdi, %r10
99 leaq -8(%rsi, %rdx), %rsi
100 leaq -8(%rdi, %rdx), %rdi
101 shrq $3, %rcx
102 std
103 rep movsq
104 cld
105 movq %r11, (%r10)
106 jmp 13f
107
108 /*
109 * Start to prepare for backward copy.
110 */
111 .p2align 4
1122:
113 cmp $680, %rdx
114 jb 6f
115 cmp %dil, %sil
116 je 7b
1176:
118 /*
119 * Calculate copy position to tail.
120 */
121 addq %rdx, %rsi
122 addq %rdx, %rdi
123 subq $0x20, %rdx
124 /*
125 * We gobble 32 bytes backward in each loop.
126 */
1278:
128 subq $0x20, %rdx
129 movq -1*8(%rsi), %r11
130 movq -2*8(%rsi), %r10
131 movq -3*8(%rsi), %r9
132 movq -4*8(%rsi), %r8
133 leaq -4*8(%rsi), %rsi
134
135 movq %r11, -1*8(%rdi)
136 movq %r10, -2*8(%rdi)
137 movq %r9, -3*8(%rdi)
138 movq %r8, -4*8(%rdi)
139 leaq -4*8(%rdi), %rdi
140 jae 8b
141 /*
142 * Calculate copy position to head.
143 */
144 addq $0x20, %rdx
145 subq %rdx, %rsi
146 subq %rdx, %rdi
1471:
148 cmpq $16, %rdx
149 jb 9f
150 /*
151 * Move data from 16 bytes to 31 bytes.
152 */
153 movq 0*8(%rsi), %r11
154 movq 1*8(%rsi), %r10
155 movq -2*8(%rsi, %rdx), %r9
156 movq -1*8(%rsi, %rdx), %r8
157 movq %r11, 0*8(%rdi)
158 movq %r10, 1*8(%rdi)
159 movq %r9, -2*8(%rdi, %rdx)
160 movq %r8, -1*8(%rdi, %rdx)
161 jmp 13f
162 .p2align 4
1639:
164 cmpq $8, %rdx
165 jb 10f
166 /*
167 * Move data from 8 bytes to 15 bytes.
168 */
169 movq 0*8(%rsi), %r11
170 movq -1*8(%rsi, %rdx), %r10
171 movq %r11, 0*8(%rdi)
172 movq %r10, -1*8(%rdi, %rdx)
173 jmp 13f
17410:
175 cmpq $4, %rdx
176 jb 11f
177 /*
178 * Move data from 4 bytes to 7 bytes.
179 */
180 movl (%rsi), %r11d
181 movl -4(%rsi, %rdx), %r10d
182 movl %r11d, (%rdi)
183 movl %r10d, -4(%rdi, %rdx)
184 jmp 13f
18511:
186 cmp $2, %rdx
187 jb 12f
188 /*
189 * Move data from 2 bytes to 3 bytes.
190 */
191 movw (%rsi), %r11w
192 movw -2(%rsi, %rdx), %r10w
193 movw %r11w, (%rdi)
194 movw %r10w, -2(%rdi, %rdx)
195 jmp 13f
19612:
197 cmp $1, %rdx
198 jb 13f
199 /*
200 * Move data for 1 byte.
201 */
202 movb (%rsi), %r11b
203 movb %r11b, (%rdi)
20413:
205 retq
206 CFI_ENDPROC
207
208 .section .altinstr_replacement,"ax"
209.Lmemmove_begin_forward_efs:
210 /* Forward moving data. */
211 movq %rdx, %rcx
212 rep movsb
213 retq
214.Lmemmove_end_forward_efs:
215 .previous
216
217 .section .altinstructions,"a"
218 altinstruction_entry .Lmemmove_begin_forward, \
219 .Lmemmove_begin_forward_efs,X86_FEATURE_ERMS, \
220 .Lmemmove_end_forward-.Lmemmove_begin_forward, \
221 .Lmemmove_end_forward_efs-.Lmemmove_begin_forward_efs
222 .previous
223ENDPROC(memmove)
1/*
2 * Normally compiler builtins are used, but sometimes the compiler calls out
3 * of line code. Based on asm-i386/string.h.
4 *
5 * This assembly file is re-written from memmove_64.c file.
6 * - Copyright 2011 Fenghua Yu <fenghua.yu@intel.com>
7 */
8#include <linux/linkage.h>
9#include <asm/cpufeatures.h>
10#include <asm/alternative-asm.h>
11
12#undef memmove
13
14/*
15 * Implement memmove(). This can handle overlap between src and dst.
16 *
17 * Input:
18 * rdi: dest
19 * rsi: src
20 * rdx: count
21 *
22 * Output:
23 * rax: dest
24 */
25.weak memmove
26
27ENTRY(memmove)
28ENTRY(__memmove)
29
30 /* Handle more 32 bytes in loop */
31 mov %rdi, %rax
32 cmp $0x20, %rdx
33 jb 1f
34
35 /* Decide forward/backward copy mode */
36 cmp %rdi, %rsi
37 jge .Lmemmove_begin_forward
38 mov %rsi, %r8
39 add %rdx, %r8
40 cmp %rdi, %r8
41 jg 2f
42
43.Lmemmove_begin_forward:
44 ALTERNATIVE "", "movq %rdx, %rcx; rep movsb; retq", X86_FEATURE_ERMS
45
46 /*
47 * movsq instruction have many startup latency
48 * so we handle small size by general register.
49 */
50 cmp $680, %rdx
51 jb 3f
52 /*
53 * movsq instruction is only good for aligned case.
54 */
55
56 cmpb %dil, %sil
57 je 4f
583:
59 sub $0x20, %rdx
60 /*
61 * We gobble 32 bytes forward in each loop.
62 */
635:
64 sub $0x20, %rdx
65 movq 0*8(%rsi), %r11
66 movq 1*8(%rsi), %r10
67 movq 2*8(%rsi), %r9
68 movq 3*8(%rsi), %r8
69 leaq 4*8(%rsi), %rsi
70
71 movq %r11, 0*8(%rdi)
72 movq %r10, 1*8(%rdi)
73 movq %r9, 2*8(%rdi)
74 movq %r8, 3*8(%rdi)
75 leaq 4*8(%rdi), %rdi
76 jae 5b
77 addq $0x20, %rdx
78 jmp 1f
79 /*
80 * Handle data forward by movsq.
81 */
82 .p2align 4
834:
84 movq %rdx, %rcx
85 movq -8(%rsi, %rdx), %r11
86 lea -8(%rdi, %rdx), %r10
87 shrq $3, %rcx
88 rep movsq
89 movq %r11, (%r10)
90 jmp 13f
91.Lmemmove_end_forward:
92
93 /*
94 * Handle data backward by movsq.
95 */
96 .p2align 4
977:
98 movq %rdx, %rcx
99 movq (%rsi), %r11
100 movq %rdi, %r10
101 leaq -8(%rsi, %rdx), %rsi
102 leaq -8(%rdi, %rdx), %rdi
103 shrq $3, %rcx
104 std
105 rep movsq
106 cld
107 movq %r11, (%r10)
108 jmp 13f
109
110 /*
111 * Start to prepare for backward copy.
112 */
113 .p2align 4
1142:
115 cmp $680, %rdx
116 jb 6f
117 cmp %dil, %sil
118 je 7b
1196:
120 /*
121 * Calculate copy position to tail.
122 */
123 addq %rdx, %rsi
124 addq %rdx, %rdi
125 subq $0x20, %rdx
126 /*
127 * We gobble 32 bytes backward in each loop.
128 */
1298:
130 subq $0x20, %rdx
131 movq -1*8(%rsi), %r11
132 movq -2*8(%rsi), %r10
133 movq -3*8(%rsi), %r9
134 movq -4*8(%rsi), %r8
135 leaq -4*8(%rsi), %rsi
136
137 movq %r11, -1*8(%rdi)
138 movq %r10, -2*8(%rdi)
139 movq %r9, -3*8(%rdi)
140 movq %r8, -4*8(%rdi)
141 leaq -4*8(%rdi), %rdi
142 jae 8b
143 /*
144 * Calculate copy position to head.
145 */
146 addq $0x20, %rdx
147 subq %rdx, %rsi
148 subq %rdx, %rdi
1491:
150 cmpq $16, %rdx
151 jb 9f
152 /*
153 * Move data from 16 bytes to 31 bytes.
154 */
155 movq 0*8(%rsi), %r11
156 movq 1*8(%rsi), %r10
157 movq -2*8(%rsi, %rdx), %r9
158 movq -1*8(%rsi, %rdx), %r8
159 movq %r11, 0*8(%rdi)
160 movq %r10, 1*8(%rdi)
161 movq %r9, -2*8(%rdi, %rdx)
162 movq %r8, -1*8(%rdi, %rdx)
163 jmp 13f
164 .p2align 4
1659:
166 cmpq $8, %rdx
167 jb 10f
168 /*
169 * Move data from 8 bytes to 15 bytes.
170 */
171 movq 0*8(%rsi), %r11
172 movq -1*8(%rsi, %rdx), %r10
173 movq %r11, 0*8(%rdi)
174 movq %r10, -1*8(%rdi, %rdx)
175 jmp 13f
17610:
177 cmpq $4, %rdx
178 jb 11f
179 /*
180 * Move data from 4 bytes to 7 bytes.
181 */
182 movl (%rsi), %r11d
183 movl -4(%rsi, %rdx), %r10d
184 movl %r11d, (%rdi)
185 movl %r10d, -4(%rdi, %rdx)
186 jmp 13f
18711:
188 cmp $2, %rdx
189 jb 12f
190 /*
191 * Move data from 2 bytes to 3 bytes.
192 */
193 movw (%rsi), %r11w
194 movw -2(%rsi, %rdx), %r10w
195 movw %r11w, (%rdi)
196 movw %r10w, -2(%rdi, %rdx)
197 jmp 13f
19812:
199 cmp $1, %rdx
200 jb 13f
201 /*
202 * Move data for 1 byte.
203 */
204 movb (%rsi), %r11b
205 movb %r11b, (%rdi)
20613:
207 retq
208ENDPROC(__memmove)
209ENDPROC(memmove)