2 * Normally compiler builtins are used, but sometimes the compiler calls out
3 * of line code. Based on asm-i386/string.h.
5 * This assembly file is re-written from memmove_64.c file.
6 * - Copyright 2011 Fenghua Yu <fenghua.yu@intel.com>
8 #include <linux/linkage.h>
9 #include <asm/cpufeatures.h>
10 #include <asm/alternative-asm.h>
11 #include <asm/export.h>
16 * Implement memmove(). This can handle overlap between src and dst.
31 /* Handle more 32 bytes in loop */
36 /* Decide forward/backward copy mode */
38 jge .Lmemmove_begin_forward
44 .Lmemmove_begin_forward:
45 ALTERNATIVE "", "movq %rdx, %rcx; rep movsb; retq", X86_FEATURE_ERMS
48 * movsq instruction have many startup latency
49 * so we handle small size by general register.
54 * movsq instruction is only good for aligned case.
62 * We gobble 32 bytes forward in each loop.
81 * Handle data forward by movsq.
86 movq -8(%rsi, %rdx), %r11
87 lea -8(%rdi, %rdx), %r10
92 .Lmemmove_end_forward:
95 * Handle data backward by movsq.
102 leaq -8(%rsi, %rdx), %rsi
103 leaq -8(%rdi, %rdx), %rdi
112 * Start to prepare for backward copy.
122 * Calculate copy position to tail.
128 * We gobble 32 bytes backward in each loop.
132 movq -1*8(%rsi), %r11
133 movq -2*8(%rsi), %r10
136 leaq -4*8(%rsi), %rsi
138 movq %r11, -1*8(%rdi)
139 movq %r10, -2*8(%rdi)
142 leaq -4*8(%rdi), %rdi
145 * Calculate copy position to head.
154 * Move data from 16 bytes to 31 bytes.
158 movq -2*8(%rsi, %rdx), %r9
159 movq -1*8(%rsi, %rdx), %r8
162 movq %r9, -2*8(%rdi, %rdx)
163 movq %r8, -1*8(%rdi, %rdx)
170 * Move data from 8 bytes to 15 bytes.
173 movq -1*8(%rsi, %rdx), %r10
175 movq %r10, -1*8(%rdi, %rdx)
181 * Move data from 4 bytes to 7 bytes.
184 movl -4(%rsi, %rdx), %r10d
186 movl %r10d, -4(%rdi, %rdx)
192 * Move data from 2 bytes to 3 bytes.
195 movw -2(%rsi, %rdx), %r10w
197 movw %r10w, -2(%rdi, %rdx)
203 * Move data for 1 byte.
211 EXPORT_SYMBOL(__memmove)
212 EXPORT_SYMBOL(memmove)