147 lines
6.4 KiB
ArmAsm
147 lines
6.4 KiB
ArmAsm
/*
|
|
* (c) Copyright 1986 HEWLETT-PACKARD COMPANY
|
|
*
|
|
* To anyone who acknowledges that this file is provided "AS IS"
|
|
* without any express or implied warranty:
|
|
* permission to use, copy, modify, and distribute this file
|
|
* for any purpose is hereby granted without fee, provided that
|
|
* the above copyright notice and this notice appears in all
|
|
* copies, and that the name of Hewlett-Packard Company not be
|
|
* used in advertising or publicity pertaining to distribution
|
|
* of the software without specific, written prior permission.
|
|
* Hewlett-Packard Company makes no representations about the
|
|
* suitability of this software for any purpose.
|
|
*/
|
|
|
|
/* HPUX_ID: @(#) $Revision$ */
|
|
/*
|
|
* memcpy(s1, s2, n)
|
|
*
|
|
* Copy n characters from s2 to s1; returns s1.
|
|
*/
|
|
|
|
#define d_addr arg0
|
|
#define s_addr arg1
|
|
#define count arg2
|
|
#define tmp5 arg3
|
|
#define tmp1 r19
|
|
#define tmp2 r20
|
|
#define tmp3 r21
|
|
#define tmp4 r22
|
|
#define tmp6 r31
|
|
|
|
#include "DEFS.h"
|
|
|
|
ENTRY(memcpy)
|
|
comib,>= 5,count,byteloop /* If count is <= 6 don't get fancy.*/
|
|
movb,=,n d_addr,ret0,done /* The return value is defined to be the value of d_addr. DELAY SLOT */
|
|
/* if d_addr is null then exit */
|
|
extru s_addr,31,2,tmp1 /* Extract the low two bits of the source address. */
|
|
extru d_addr,31,2,tmp2 /* Extract the low two bits of the destination address. */
|
|
add count,tmp2,count /* pre increment the count to adjust for alignment of s1 */
|
|
comb,<> tmp2,tmp1,not_aligned /* see if s1 is aligned w.r.t. s2. */
|
|
dep 0,31,2,s_addr /* Compute the word address of the source. DELAY SLOT. */
|
|
|
|
/* aligned */
|
|
|
|
/* We will now begin the 16 byte at a time word move if count >= 16 ! */
|
|
/* Else we will branch to the 4 byte-at-a time word move ! */
|
|
|
|
addibt,<,n -16,count,chekchunk /* If count < 16 then we can't move 16 byte chunks ! */
|
|
/* actually we can legally move 13 or more bytes on the first loop. */
|
|
/* These loads and stores are done so as to prevent processor interlock. */
|
|
chunks:
|
|
ldwm 16(0,s_addr),tmp1 /* tmp1 = *s_addr s_addr += 16 */
|
|
ldw -12(0,s_addr),tmp2 /* tmp2 = 2nd word */
|
|
ldw -8(0,s_addr),tmp3 /* tmp3 = 3rd word */
|
|
ldw -4(0,s_addr),tmp4 /* tmp4 = 4th word */
|
|
/* Now store the results ! */
|
|
stbys,b,m tmp1,4(0,d_addr) /* tmp1 = 1st word stored d_addr += 16 also take care of front porch. */
|
|
stwm tmp2,4(0,d_addr) /* tmp2 = 2nd word stored. */
|
|
stwm tmp3,4(0,d_addr) /* tmp3 = 3rd word stored. */
|
|
addibf,< -16,count,chunks /* If count is still >= 16 do another loop. */
|
|
stwm tmp4,4(0,d_addr) /* tmp4 = 4th word stored. DELAY SLOT */
|
|
|
|
chekchunk:
|
|
addibt,<,n 12,count,back_porch /* since the count is already decremented by -16 we're testing */
|
|
/* to see if there are at least 4 bytes left ? */
|
|
subchunk:
|
|
ldws,ma 4(s_addr),tmp1 /* tmp1 = *s_addr++ */
|
|
addibf,< -4,count,subchunk /* count -= 4 */
|
|
stbys,b,m tmp1,4(d_addr) /* *d_addr++ = tmp1 */
|
|
|
|
|
|
back_porch:
|
|
addibt,=,n 4,count,done /* if count = 0 we're, of course, done ! */
|
|
ldws 0(s_addr),tmp1 /* load up the back_porch */
|
|
add d_addr,count,d_addr/* final store address is +1 too high ! */
|
|
bv 0(r2) /* return--were done. */
|
|
stbys,e tmp1,0(d_addr) /* kerplunk! whew ! */
|
|
|
|
/* Begin non_aligned code. (no refrence to politics) */
|
|
not_aligned:
|
|
sub,>= tmp2,tmp1,tmp3 /* compute the shift quantity again and skip the load if tmp2 > tmp1. */
|
|
ldwm 4(0,s_addr),tmp1 /* load up the first word from the source. tmp1 = *s_addr++ */
|
|
zdep tmp3,28,29,tmp4 /* compute the number of bits to shift based on the number of bytes above. */
|
|
mtctl tmp4,11 /* load the shift count into cr11 = shift count register. */
|
|
|
|
addibt,<,n -16,count,chkchnk2 /* first step in pre adjustment of count for looping. */
|
|
|
|
chunk2:
|
|
ldwm 16(0,s_addr),tmp2 /* get either first or second word . tmp2 = *s_addr++ */
|
|
ldw -12(s_addr),tmp3
|
|
ldw -8(s_addr),tmp4
|
|
ldw -4(s_addr),tmp5
|
|
vshd tmp1,tmp2,tmp6 /* position data ! */
|
|
stbys,b,m tmp6,4(0,d_addr) /* store ! */
|
|
|
|
vshd tmp2,tmp3,tmp6 /* position data ! */
|
|
stwm tmp6,4(0,d_addr) /* store ! */
|
|
|
|
vshd tmp3,tmp4,tmp6 /* position data ! */
|
|
stwm tmp6,4(0,d_addr) /* store ! */
|
|
|
|
vshd tmp4,tmp5,tmp6 /* position data ! */
|
|
stwm tmp6,4(0,d_addr) /* store the data ! */
|
|
addibf,< -16,count,chunk2 /* If count is still >= 16 do another loop. */
|
|
copy tmp5,tmp1
|
|
|
|
|
|
chkchnk2:
|
|
addibt,<,n 12,count,bp_0 /* if we don't have 4 bytes left then do the back porch (bp_0) */
|
|
|
|
subchnk2:
|
|
ldwm 4(0,s_addr),tmp2 /* get next word ! */
|
|
vshd tmp1,tmp2,tmp3 /* position data ! */
|
|
addibt,< -4,count,bp_1 /* decrement count and when count < 4 goto back_porch (bp_1) */
|
|
stbys,b,m tmp3,4(0,d_addr) /* store ! */
|
|
|
|
ldwm 4(0,s_addr),tmp1 /* get 4th word ! */
|
|
vshd tmp2,tmp1,tmp3 /* position data ! */
|
|
addib,>= -4,count,subchnk2 /* decrement count and when count <= 4 go to back porch (bp_2) */
|
|
stbys,b,m tmp3,4(0,d_addr) /* store the data ! */
|
|
|
|
bp_0: copy tmp1,tmp2 /* switch registers used in the shift process. */
|
|
bp_1: addibt,<=,n 4,count,done /* if count = -4 this implies that count = 0 -> done */
|
|
add d_addr,count,d_addr /* bump destination address to be +1 too high ! */
|
|
mfctl sar,tmp3 /* suppress final ldwm unless result used */
|
|
extru tmp3,28,2,tmp3 /* convert bitshift to byteshift */
|
|
sub,<= count,tmp3,r0 /* bytes unused if (count-byteshift <= 0*/
|
|
|
|
ldwm 4(0,s_addr),tmp1 /* get final word ! */
|
|
vshd tmp2,tmp1,tmp3 /* position data ! */
|
|
bv 0(r2) /* return */
|
|
stbys,e tmp3,0(0,d_addr) /* store the data ! */
|
|
|
|
/* here we do ye old byte-at-a-time moves. */
|
|
byteloop:
|
|
comb,>=,n 0,count,done
|
|
|
|
encore:
|
|
ldbs,ma 1(s_addr),tmp1
|
|
addibf,= -1,count,encore
|
|
stbs,ma tmp1,1(d_addr)
|
|
|
|
done:
|
|
EXIT(memcpy)
|