Nyaaori/rockbox
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
rockbox/firmware/asm/mips/memset.S

240 lines
8 KiB
ArmAsm
Raw Normal View History

MIPS: * Add assembly optimised variants for memcpy, memset and find_first_set_bit * Add option to map_address in MMU to set caching algorithm git-svn-id: svn://svn.rockbox.org/rockbox/trunk@19920 a1c6a512-1295-4272-9138-f99709370657
2009-02-04 17:33:19 +00:00
/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* This file was originally part of the Linux/MIPS GNU C Library
* Copyright (C) 1998 by Ralf Baechle
* Adapted for Rockbox by Maurus Cuelenaere, 2009
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; either version 2
* of the License, or (at your option) any later version.
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
#include "config.h"
#include "mips.h"
#define FILL256(dst, offset, val) \
sw val, (offset + 0x00)(dst); \
sw val, (offset + 0x04)(dst); \
sw val, (offset + 0x08)(dst); \
sw val, (offset + 0x0c)(dst); \
sw val, (offset + 0x10)(dst); \
sw val, (offset + 0x14)(dst); \
sw val, (offset + 0x18)(dst); \
sw val, (offset + 0x1c)(dst); \
sw val, (offset + 0x20)(dst); \
sw val, (offset + 0x24)(dst); \
sw val, (offset + 0x28)(dst); \
sw val, (offset + 0x2c)(dst); \
sw val, (offset + 0x30)(dst); \
sw val, (offset + 0x34)(dst); \
sw val, (offset + 0x38)(dst); \
sw val, (offset + 0x3c)(dst); \
sw val, (offset + 0x40)(dst); \
sw val, (offset + 0x44)(dst); \
sw val, (offset + 0x48)(dst); \
sw val, (offset + 0x4c)(dst); \
sw val, (offset + 0x50)(dst); \
sw val, (offset + 0x54)(dst); \
sw val, (offset + 0x58)(dst); \
sw val, (offset + 0x5c)(dst); \
sw val, (offset + 0x60)(dst); \
sw val, (offset + 0x64)(dst); \
sw val, (offset + 0x68)(dst); \
sw val, (offset + 0x6c)(dst); \
sw val, (offset + 0x70)(dst); \
sw val, (offset + 0x74)(dst); \
sw val, (offset + 0x78)(dst); \
sw val, (offset + 0x7c)(dst); \
sw val, (offset + 0x80)(dst); \
sw val, (offset + 0x84)(dst); \
sw val, (offset + 0x88)(dst); \
sw val, (offset + 0x8c)(dst); \
sw val, (offset + 0x90)(dst); \
sw val, (offset + 0x94)(dst); \
sw val, (offset + 0x98)(dst); \
sw val, (offset + 0x9c)(dst); \
sw val, (offset + 0xa0)(dst); \
sw val, (offset + 0xa4)(dst); \
sw val, (offset + 0xa8)(dst); \
sw val, (offset + 0xac)(dst); \
sw val, (offset + 0xb0)(dst); \
sw val, (offset + 0xb4)(dst); \
sw val, (offset + 0xb8)(dst); \
sw val, (offset + 0xbc)(dst); \
sw val, (offset + 0xc0)(dst); \
sw val, (offset + 0xc4)(dst); \
sw val, (offset + 0xc8)(dst); \
sw val, (offset + 0xcc)(dst); \
sw val, (offset + 0xd0)(dst); \
sw val, (offset + 0xd4)(dst); \
sw val, (offset + 0xd8)(dst); \
sw val, (offset + 0xdc)(dst); \
sw val, (offset + 0xe0)(dst); \
sw val, (offset + 0xe4)(dst); \
sw val, (offset + 0xe8)(dst); \
sw val, (offset + 0xec)(dst); \
sw val, (offset + 0xf0)(dst); \
sw val, (offset + 0xf4)(dst); \
sw val, (offset + 0xf8)(dst); \
sw val, (offset + 0xfc)(dst);
#define FILL128(dst, offset, val) \
sw val, (offset + 0x00)(dst); \
sw val, (offset + 0x04)(dst); \
sw val, (offset + 0x08)(dst); \
sw val, (offset + 0x0c)(dst); \
sw val, (offset + 0x10)(dst); \
sw val, (offset + 0x14)(dst); \
sw val, (offset + 0x18)(dst); \
sw val, (offset + 0x1c)(dst); \
sw val, (offset + 0x20)(dst); \
sw val, (offset + 0x24)(dst); \
sw val, (offset + 0x28)(dst); \
sw val, (offset + 0x2c)(dst); \
sw val, (offset + 0x30)(dst); \
sw val, (offset + 0x34)(dst); \
sw val, (offset + 0x38)(dst); \
sw val, (offset + 0x3c)(dst); \
sw val, (offset + 0x40)(dst); \
sw val, (offset + 0x44)(dst); \
sw val, (offset + 0x48)(dst); \
sw val, (offset + 0x4c)(dst); \
sw val, (offset + 0x50)(dst); \
sw val, (offset + 0x54)(dst); \
sw val, (offset + 0x58)(dst); \
sw val, (offset + 0x5c)(dst); \
sw val, (offset + 0x60)(dst); \
sw val, (offset + 0x64)(dst); \
sw val, (offset + 0x68)(dst); \
sw val, (offset + 0x6c)(dst); \
sw val, (offset + 0x70)(dst); \
sw val, (offset + 0x74)(dst); \
sw val, (offset + 0x78)(dst); \
sw val, (offset + 0x7c)(dst);
#define FILL64(dst, offset, val) \
sw val, (offset + 0x00)(dst); \
sw val, (offset + 0x04)(dst); \
sw val, (offset + 0x08)(dst); \
sw val, (offset + 0x0c)(dst); \
sw val, (offset + 0x10)(dst); \
sw val, (offset + 0x14)(dst); \
sw val, (offset + 0x18)(dst); \
sw val, (offset + 0x1c)(dst); \
sw val, (offset + 0x20)(dst); \
sw val, (offset + 0x24)(dst); \
sw val, (offset + 0x28)(dst); \
sw val, (offset + 0x2c)(dst); \
sw val, (offset + 0x30)(dst); \
sw val, (offset + 0x34)(dst); \
sw val, (offset + 0x38)(dst); \
sw val, (offset + 0x3c)(dst);
#define FILL32(dst, offset, val) \
sw val, (offset + 0x00)(dst); \
sw val, (offset + 0x04)(dst); \
sw val, (offset + 0x08)(dst); \
sw val, (offset + 0x0c)(dst); \
sw val, (offset + 0x10)(dst); \
sw val, (offset + 0x14)(dst); \
sw val, (offset + 0x18)(dst); \
sw val, (offset + 0x1c)(dst);
MIPS memset: use a bigger buffer (yields more performance) git-svn-id: svn://svn.rockbox.org/rockbox/trunk@19922 a1c6a512-1295-4272-9138-f99709370657
2009-02-04 20:18:30 +00:00
#define FILL 64
#define F_FILL FILL64
MIPS: * Add assembly optimised variants for memcpy, memset and find_first_set_bit * Add option to map_address in MMU to set caching algorithm git-svn-id: svn://svn.rockbox.org/rockbox/trunk@19920 a1c6a512-1295-4272-9138-f99709370657
2009-02-04 17:33:19 +00:00
#ifdef ROCKBOX_BIG_ENDIAN
# define SWHI swl /* high part is left in big-endian */
#else
# define SWHI swr /* high part is right in little-endian */
#endif
/*
* memset(void *s, int c, size_t n)
*
* a0: start of area to clear
* a1: char to fill with
* a2: size of area to clear
*/
.section .icode, "ax", %progbits
.global memset
.type memset, %function
.set noreorder
.align 5
memset:
beqz a1, 1f
move v0, a0 /* result */
andi a1, 0xff /* spread fillword */
sll t1, a1, 8
or a1, t1
sll t1, a1, 16
or a1, t1
1:
sltiu t0, a2, 4 /* very small region? */
bnez t0, small_memset
andi t0, a0, 3 /* aligned? */
beqz t0, 1f
subu t0, 4 /* alignment in bytes */
SWHI a1, (a0) /* make word aligned */
subu a0, t0 /* word align ptr */
addu a2, t0 /* correct size */
1: ori t1, a2, (FILL-1) /* # of full blocks */
xori t1, (FILL-1)
beqz t1, memset_partial /* no block to fill */
andi t0, a2, (FILL-4)
addu t1, a0 /* end address */
.set reorder
1: addiu a0, FILL
F_FILL( a0, -FILL, a1 )
bne t1, a0, 1b
.set noreorder
memset_partial:
la t1, 2f /* where to start */
subu t1, t0
jr t1
addu a0, t0 /* dest ptr */
F_FILL( a0, -FILL, a1 ) /* ... but first do words ... */
2: andi a2, 3 /* 0 <= n <= 3 to go */
beqz a2, 1f
addu a0, a2 /* What's left */
SWHI a1, -1(a0)
1: jr ra
move a2, zero
small_memset:
beqz a2, 2f
addu t1, a0, a2
1: addiu a0, 1 /* fill bytewise */
bne t1, a0, 1b
sb a1, -1(a0)
2: jr ra /* done */
move a2, zero
.set reorder
Reference in a new issue Copy permalink