xref: /linux/arch/x86/lib/memmove_64.S (revision f3a8b6645dc2e60d11f20c1c23afd964ff4e55ae)
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#include <asm/export.h>
12
13#undef memmove
14
15/*
16 * Implement memmove(). This can handle overlap between src and dst.
17 *
18 * Input:
19 * rdi: dest
20 * rsi: src
21 * rdx: count
22 *
23 * Output:
24 * rax: dest
25 */
26.weak memmove
27
28ENTRY(memmove)
29ENTRY(__memmove)
30
31	/* Handle more 32 bytes in loop */
32	mov %rdi, %rax
33	cmp $0x20, %rdx
34	jb	1f
35
36	/* Decide forward/backward copy mode */
37	cmp %rdi, %rsi
38	jge .Lmemmove_begin_forward
39	mov %rsi, %r8
40	add %rdx, %r8
41	cmp %rdi, %r8
42	jg 2f
43
44.Lmemmove_begin_forward:
45	ALTERNATIVE "", "movq %rdx, %rcx; rep movsb; retq", X86_FEATURE_ERMS
46
47	/*
48	 * movsq instruction have many startup latency
49	 * so we handle small size by general register.
50	 */
51	cmp  $680, %rdx
52	jb	3f
53	/*
54	 * movsq instruction is only good for aligned case.
55	 */
56
57	cmpb %dil, %sil
58	je 4f
593:
60	sub $0x20, %rdx
61	/*
62	 * We gobble 32 bytes forward in each loop.
63	 */
645:
65	sub $0x20, %rdx
66	movq 0*8(%rsi), %r11
67	movq 1*8(%rsi), %r10
68	movq 2*8(%rsi), %r9
69	movq 3*8(%rsi), %r8
70	leaq 4*8(%rsi), %rsi
71
72	movq %r11, 0*8(%rdi)
73	movq %r10, 1*8(%rdi)
74	movq %r9, 2*8(%rdi)
75	movq %r8, 3*8(%rdi)
76	leaq 4*8(%rdi), %rdi
77	jae 5b
78	addq $0x20, %rdx
79	jmp 1f
80	/*
81	 * Handle data forward by movsq.
82	 */
83	.p2align 4
844:
85	movq %rdx, %rcx
86	movq -8(%rsi, %rdx), %r11
87	lea -8(%rdi, %rdx), %r10
88	shrq $3, %rcx
89	rep movsq
90	movq %r11, (%r10)
91	jmp 13f
92.Lmemmove_end_forward:
93
94	/*
95	 * Handle data backward by movsq.
96	 */
97	.p2align 4
987:
99	movq %rdx, %rcx
100	movq (%rsi), %r11
101	movq %rdi, %r10
102	leaq -8(%rsi, %rdx), %rsi
103	leaq -8(%rdi, %rdx), %rdi
104	shrq $3, %rcx
105	std
106	rep movsq
107	cld
108	movq %r11, (%r10)
109	jmp 13f
110
111	/*
112	 * Start to prepare for backward copy.
113	 */
114	.p2align 4
1152:
116	cmp $680, %rdx
117	jb 6f
118	cmp %dil, %sil
119	je 7b
1206:
121	/*
122	 * Calculate copy position to tail.
123	 */
124	addq %rdx, %rsi
125	addq %rdx, %rdi
126	subq $0x20, %rdx
127	/*
128	 * We gobble 32 bytes backward in each loop.
129	 */
1308:
131	subq $0x20, %rdx
132	movq -1*8(%rsi), %r11
133	movq -2*8(%rsi), %r10
134	movq -3*8(%rsi), %r9
135	movq -4*8(%rsi), %r8
136	leaq -4*8(%rsi), %rsi
137
138	movq %r11, -1*8(%rdi)
139	movq %r10, -2*8(%rdi)
140	movq %r9, -3*8(%rdi)
141	movq %r8, -4*8(%rdi)
142	leaq -4*8(%rdi), %rdi
143	jae 8b
144	/*
145	 * Calculate copy position to head.
146	 */
147	addq $0x20, %rdx
148	subq %rdx, %rsi
149	subq %rdx, %rdi
1501:
151	cmpq $16, %rdx
152	jb 9f
153	/*
154	 * Move data from 16 bytes to 31 bytes.
155	 */
156	movq 0*8(%rsi), %r11
157	movq 1*8(%rsi), %r10
158	movq -2*8(%rsi, %rdx), %r9
159	movq -1*8(%rsi, %rdx), %r8
160	movq %r11, 0*8(%rdi)
161	movq %r10, 1*8(%rdi)
162	movq %r9, -2*8(%rdi, %rdx)
163	movq %r8, -1*8(%rdi, %rdx)
164	jmp 13f
165	.p2align 4
1669:
167	cmpq $8, %rdx
168	jb 10f
169	/*
170	 * Move data from 8 bytes to 15 bytes.
171	 */
172	movq 0*8(%rsi), %r11
173	movq -1*8(%rsi, %rdx), %r10
174	movq %r11, 0*8(%rdi)
175	movq %r10, -1*8(%rdi, %rdx)
176	jmp 13f
17710:
178	cmpq $4, %rdx
179	jb 11f
180	/*
181	 * Move data from 4 bytes to 7 bytes.
182	 */
183	movl (%rsi), %r11d
184	movl -4(%rsi, %rdx), %r10d
185	movl %r11d, (%rdi)
186	movl %r10d, -4(%rdi, %rdx)
187	jmp 13f
18811:
189	cmp $2, %rdx
190	jb 12f
191	/*
192	 * Move data from 2 bytes to 3 bytes.
193	 */
194	movw (%rsi), %r11w
195	movw -2(%rsi, %rdx), %r10w
196	movw %r11w, (%rdi)
197	movw %r10w, -2(%rdi, %rdx)
198	jmp 13f
19912:
200	cmp $1, %rdx
201	jb 13f
202	/*
203	 * Move data for 1 byte.
204	 */
205	movb (%rsi), %r11b
206	movb %r11b, (%rdi)
20713:
208	retq
209ENDPROC(__memmove)
210ENDPROC(memmove)
211EXPORT_SYMBOL(__memmove)
212EXPORT_SYMBOL(memmove)
213