Subversion Repositories c0or1k

/*
 * Copyright 2010 B Labs.Ltd.
 *
 * Author: Prem Mallappa  <prem.mallappa@b-labs.co.uk>
 *
 * Description: Optimized memcpy for ARM
 *
 */

#include  INC_ARCH(asm.h)
/*
void*
_memcpy(void *dst, const void *src, register uint len)
*/
BEGIN_PROC(_memcpy)
                push {r0, r4-r11, lr}
        loop32:
                cmp r2, #32
                blt loop16
                ldmia r1!, {r4 - r11}
                stmia r0!, {r4 - r11}
                sub r2, r2, #32
                b loop32

        loop16:
                cmp r2, #16
                blt loop8
                ldmia   r1!, {r4 - r7}
                stmia   r0!, {r4 - r7}
                sub r2, r2, #16
                b loop16

        loop8:
                cmp r2, #8
                blt loop4
                ldmia   r1!, {r4, r5}
                stmia   r0!, {r4, r5}
                sub r2, r2, #8
                b loop8

        loop4:
                cmp r2, #4
                blt end
                ldmia   r1!, {r4}
                stmia   r0!, {r4}
                sub r2, r2, #4
                b loop4
        end:
        last:
                teq   r2, #0
                ldrgtb r4, [r1]
                strneb r4, [r0]         // V7 supports strneb <rt>, [<rb>, +/-<index>] !, with write back
                lsrne r4, r4, #8
                subne r2, r2, #1        // Can be reduced to 1 LDR, but has a catch if it is end of memory
                addne r0, r0, #1        // we dont want to fetch 1 byte extra to end up in abort
                addne r1, r1, #1        // so, playing safe, worst case 3 LDRs
                bne last
        1:
        pop {r0, r4 - r11, pc}
END_PROC(_memcpy)