M7350/kernel/arch/blackfin/lib/memcpy.S
2024-09-09 08:52:07 +00:00

125 lines
2.6 KiB
ArmAsm

/*
* internal version of memcpy(), issued by the compiler to copy blocks of
* data around. This is really memmove() - it has to be able to deal with
* possible overlaps, because that ambiguity is when the compiler gives up
* and calls a function. We have our own, internal version so that we get
* something we trust, even if the user has redefined the normal symbol.
*
* Copyright 2004-2009 Analog Devices Inc.
*
* Licensed under the ADI BSD license or the GPL-2 (or later)
*/
#include <linux/linkage.h>
/* void *memcpy(void *dest, const void *src, size_t n);
* R0 = To Address (dest) (leave unchanged to form result)
* R1 = From Address (src)
* R2 = count
*
* Note: Favours word alignment
*/
#ifdef CONFIG_MEMCPY_L1
.section .l1.text
#else
.text
#endif
.align 2
ENTRY(_memcpy)
CC = R2 <= 0; /* length not positive? */
IF CC JUMP .L_P1L2147483647; /* Nothing to do */
P0 = R0 ; /* dst*/
P1 = R1 ; /* src*/
P2 = R2 ; /* length */
/* check for overlapping data */
CC = R1 < R0; /* src < dst */
IF !CC JUMP .Lno_overlap;
R3 = R1 + R2;
CC = R0 < R3; /* and dst < src+len */
IF CC JUMP .Lhas_overlap;
.Lno_overlap:
/* Check for aligned data.*/
R3 = R1 | R0;
R1 = 0x3;
R3 = R3 & R1;
CC = R3; /* low bits set on either address? */
IF CC JUMP .Lnot_aligned;
/* Both addresses are word-aligned, so we can copy
at least part of the data using word copies.*/
P2 = P2 >> 2;
CC = P2 <= 2;
IF !CC JUMP .Lmore_than_seven;
/* less than eight bytes... */
P2 = R2;
LSETUP(.Lthree_start, .Lthree_end) LC0=P2;
.Lthree_start:
R3 = B[P1++] (X);
.Lthree_end:
B[P0++] = R3;
RTS;
.Lmore_than_seven:
/* There's at least eight bytes to copy. */
P2 += -1; /* because we unroll one iteration */
LSETUP(.Lword_loops, .Lword_loope) LC0=P2;
I1 = P1;
R3 = [I1++];
#if ANOMALY_05000202
.Lword_loops:
[P0++] = R3;
.Lword_loope:
R3 = [I1++];
#else
.Lword_loops:
.Lword_loope:
MNOP || [P0++] = R3 || R3 = [I1++];
#endif
[P0++] = R3;
/* Any remaining bytes to copy? */
R3 = 0x3;
R3 = R2 & R3;
CC = R3 == 0;
P1 = I1; /* in case there's something left, */
IF !CC JUMP .Lbytes_left;
RTS;
.Lbytes_left: P2 = R3;
.Lnot_aligned:
/* From here, we're copying byte-by-byte. */
LSETUP (.Lbyte_start, .Lbyte_end) LC0=P2;
.Lbyte_start:
R1 = B[P1++] (X);
.Lbyte_end:
B[P0++] = R1;
.L_P1L2147483647:
RTS;
.Lhas_overlap:
/* Need to reverse the copying, because the
* dst would clobber the src.
* Don't bother to work out alignment for
* the reverse case.
*/
P0 = P0 + P2;
P0 += -1;
P1 = P1 + P2;
P1 += -1;
LSETUP(.Lover_start, .Lover_end) LC0=P2;
.Lover_start:
R1 = B[P1--] (X);
.Lover_end:
B[P0--] = R1;
RTS;
ENDPROC(_memcpy)