Added memset16() for filling memory regions with 16 bit values, needed for upcoming 16 bit LCD driver opts. ASM optimised for coldfire.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@8773 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Jens Arnold 2006-02-22 01:20:45 +00:00
parent a875703e6e
commit eb65f89f0e
4 changed files with 257 additions and 0 deletions

View file

@ -45,6 +45,11 @@ common/memcpy.c
common/memmove.c
common/memset.c
#endif
#ifdef CPU_COLDFIRE
common/memset16_a.S
#else
common/memset16.c
#endif
#ifdef HAVE_LCD_CHARCELLS
drivers/lcd-player-charset.c
drivers/lcd-player.c

79
firmware/common/memset16.c Executable file
View file

@ -0,0 +1,79 @@
/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2006 by Jens Arnold
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
#include <string.h>
#define LBLOCKSIZE (sizeof(long)/2)
#define UNALIGNED(X) ((long)X & (LBLOCKSIZE - 1))
#define TOO_SMALL(LEN) ((LEN) < LBLOCKSIZE)
void *memset16(void *dst, int val, size_t len)
{
#if defined(PREFER_SIZE_OVER_SPEED) || defined(__OPTIMIZE_SIZE__)
unsigned short *p = (unsigned short *)dst;
while (len--)
*p++ = val;
return dst;
#else
unsigned short *p = (unsigned short *)dst;
unsigned int i;
unsigned long buffer;
unsigned long *aligned_addr;
if (!TOO_SMALL(len) && !UNALIGNED(dst))
{
aligned_addr = (unsigned long *)dst;
val &= 0xffff;
if (LBLOCKSIZE == 2)
{
buffer = (val << 16) | val;
}
else
{
buffer = 0;
for (i = 0; i < LBLOCKSIZE; i++)
buffer = (buffer << 16) | val;
}
while (len >= LBLOCKSIZE*4)
{
*aligned_addr++ = buffer;
*aligned_addr++ = buffer;
*aligned_addr++ = buffer;
*aligned_addr++ = buffer;
len -= 4*LBLOCKSIZE;
}
while (len >= LBLOCKSIZE)
{
*aligned_addr++ = buffer;
len -= LBLOCKSIZE;
}
p = (unsigned short *)aligned_addr;
}
while (len--)
*p++ = val;
return dst;
#endif /* not PREFER_SIZE_OVER_SPEED */
}

146
firmware/common/memset16_a.S Executable file
View file

@ -0,0 +1,146 @@
/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2006 by Jens Arnold
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
#include "config.h"
.section .icode,"ax",@progbits
#ifdef CPU_COLDFIRE
.global memset16
.type memset16,@function
/* Fills a memory region with specified word value
* Start address must be word aligned, length is in words
* This version is optimized for speed
*
* arguments:
* (4,%sp) - start address
* (8,%sp) - data
* (12,%sp) - length
*
* return value:
* %d0 - start address
*
* register usage:
* %d0 - data (spread to both words when using long stores)
* %d1 - temporary / data (for burst transfer)
* %d2 - data (for burst transfer)
* %d3 - data (for burst transfer)
* %a0 - start address
* %a1 - current address (runs down from end to start)
*
* For maximum speed this routine uses both long stores and burst mode,
* storing whole lines with movem.l. The routine fills memory from end
* to start in order to ease returning the start address.
*/
memset16:
move.l (4,%sp),%a0 /* start address */
move.l (8,%sp),%d0 /* data */
move.l (12,%sp),%a1 /* length */
add.l %a1,%a1
add.l %a0,%a1 /* %a1 = end address */
move.l %a0,%d1
addq.l #6,%d1
and.l #0xFFFFFFFC,%d1 /* %d1 = first long bound + 4 */
cmp.l %d1,%a1 /* at least one aligned longword to fill? */
blo.b .no_longs /* no, jump directly to word loop */
and.l #0xFFFF,%d0 /* start: spread data to both words */
move.l %d0,%d1
swap %d1
or.l %d1,%d0 /* data now in both words */
move.l %a1,%d1
and.l #0xFFFFFFFC,%d1 /* %d1 = last long bound */
cmp.l %d1,%a1 /* one extra word? */
bls.b .end_w1 /* no: skip */
move.w %d0,-(%a1) /* set leading word */
.end_w1:
moveq.l #30,%d1
add.l %a0,%d1
and.l #0xFFFFFFF0,%d1 /* %d1 = first line bound + 16 */
cmp.l %d1,%a1 /* at least one full line to fill? */
blo.b .no_lines /* no, jump to longword loop */
mov.l %a1,%d1
and.l #0xFFFFFFF0,%d1 /* %d1 = last line bound */
cmp.l %d1,%a1 /* any longwords to set? */
bls.b .end_l1 /* no: skip longword loop */
/* leading longword loop: sets 0..3 longwords */
.loop_l1:
move.l %d0,-(%a1) /* store longword */
cmp.l %d1,%a1 /* runs %a1 down to last line bound */
bhi.b .loop_l1
.end_l1:
move.l %d2,-(%sp) /* free some registers */
move.l %d3,-(%sp)
move.l %d0,%d1 /* spread data to 4 data registers */
move.l %d0,%d2
move.l %d0,%d3
lea.l (14,%a0),%a0 /* start address += 14, acct. for trl. data */
/* main loop: set whole lines utilising burst mode */
.loop_line:
lea.l (-16,%a1),%a1 /* pre-decrement */
movem.l %d0-%d3,(%a1) /* store line */
cmp.l %a0,%a1 /* runs %a1 down to first line bound */
bhi.b .loop_line
lea.l (-14,%a0),%a0 /* correct start address */
move.l (%sp)+,%d3 /* restore registers */
move.l (%sp)+,%d2
move.l %a0,%d1 /* %d1 = start address ... */
addq.l #2,%d1 /* ... +2, account for possible trailing word */
cmp.l %d1,%a1 /* any longwords left */
bhi.b .loop_l2 /* yes: jump to longword loop */
bra.b .no_longs /* no: skip loop */
.no_lines:
move.l %a0,%d1 /* %d1 = start address ... */
addq.l #2,%d1 /* ... +2, account for possible trailing word */
/* trailing longword loop */
.loop_l2:
move.l %d0,-(%a1) /* store longword */
cmp.l %d1,%a1 /* runs %a1 down to first long bound */
bhi.b .loop_l2
.no_longs:
cmp.l %a0,%a1 /* any words left? */
bls.b .end_w2 /* no: skip loop */
/* trailing word loop */
.loop_w2:
move.w %d0,-(%a1) /* store word */
cmp.l %a0,%a1 /* runs %a1 down to start address */
bhi.b .loop_w2
.end_w2:
move.l %a0,%d0 /* return start address */
rts
.end:
.size memset16,.end-memset16
#endif

27
firmware/include/memory.h Executable file
View file

@ -0,0 +1,27 @@
/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2006 by Jens Arnold
*
* All files in this archive are subject to the GNU General Public License.
* See the file COPYING in the source tree root for full license agreement.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
#ifndef _MEMORY_H_
#define _MEMORY_H_
#include <sys/types.h>
void *memset16(void *dst, int val, size_t len);
#endif /* _MEMORY_H_ */