Home | History | Annotate | Line # | Download | only in arm32
bcopy_page.S revision 1.6 
      1 /*	$NetBSD: bcopy_page.S,v 1.6 2003/04/08 22:57:53 thorpej Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1995 Scott Stevens
      5  * All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by Scott Stevens.
     18  * 4. The name of the author may not be used to endorse or promote products
     19  *    derived from this software without specific prior written permission.
     20  *
     21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     31  *
     32  * RiscBSD kernel project
     33  *
     34  * bcopy_page.S
     35  *
     36  * page optimised bcopy and bzero routines
     37  *
     38  * Created      : 08/04/95
     39  */
     40 
     41 #include <machine/asm.h>
     42 
     43 #include "assym.h"
     44 
     45 /* #define BIG_LOOPS */
     46 
     47 /*
     48  * bcopy_page(src, dest)
     49  *
     50  * Optimised copy page routine.
     51  *
     52  * On entry:
     53  *   r0 - src address
     54  *   r1 - dest address
     55  *
     56  * Requires:
     57  *   number of bytes per page (PAGE_SIZE) is a multiple of 512 (BIG_LOOPS), 128
     58  *   otherwise.
     59  */
     60 
     61 #define	CHUNK_SIZE	32
     62 
     63 #ifdef __XSCALE__
     64 	/* Conveniently, the chunk size is the XScale cache line size. */
     65 #define	PREFETCH_FIRST_CHUNK	pld	[r0]
     66 #define	PREFETCH_NEXT_CHUNK	pld	[r0, #(CHUNK_SIZE)]
     67 #else
     68 #define	PREFETCH_FIRST_CHUNK	/* nothing */
     69 #define	PREFETCH_NEXT_CHUNK	/* nothing */
     70 #endif
     71 
     72 #ifndef COPY_CHUNK
     73 #define	COPY_CHUNK \
     74 	PREFETCH_NEXT_CHUNK ; \
     75 	ldmia	r0!, {r3-r8,ip,lr} ; \
     76 	stmia	r1!, {r3-r8,ip,lr}
     77 #endif /* ! COPY_CHUNK */
     78 
     79 #ifndef SAVE_REGS
     80 #define	SAVE_REGS	stmfd	sp!, {r4-r8, lr}
     81 #define	RESTORE_REGS	ldmfd	sp!, {r4-r8, pc}
     82 #endif
     83 
     84 ENTRY(bcopy_page)
     85 	PREFETCH_FIRST_CHUNK
     86 	SAVE_REGS
     87 #ifdef BIG_LOOPS
     88 	mov	r2, #(PAGE_SIZE >> 9)
     89 #else
     90 	mov	r2, #(PAGE_SIZE >> 7)
     91 #endif
     92 
     93 1:
     94 	COPY_CHUNK
     95 	COPY_CHUNK
     96 	COPY_CHUNK
     97 	COPY_CHUNK
     98 
     99 #ifdef BIG_LOOPS
    100 	/* There is little point making the loop any larger; unless we are
    101 	   running with the cache off, the load/store overheads will
    102 	   completely dominate this loop.  */
    103 	COPY_CHUNK
    104 	COPY_CHUNK
    105 	COPY_CHUNK
    106 	COPY_CHUNK
    107 
    108 	COPY_CHUNK
    109 	COPY_CHUNK
    110 	COPY_CHUNK
    111 	COPY_CHUNK
    112 
    113 	COPY_CHUNK
    114 	COPY_CHUNK
    115 	COPY_CHUNK
    116 	COPY_CHUNK
    117 #endif
    118 	subs	r2, r2, #1
    119 	bne	1b
    120 
    121 	RESTORE_REGS		/* ...and return. */
    122 
    123 /*
    124  * bzero_page(dest)
    125  *
    126  * Optimised zero page routine.
    127  *
    128  * On entry:
    129  *   r0 - dest address
    130  *
    131  * Requires:
    132  *   number of bytes per page (PAGE_SIZE) is a multiple of 512 (BIG_LOOPS), 128
    133  *   otherwise
    134  */
    135 
    136 ENTRY(bzero_page)
    137 	stmfd	sp!, {r4-r8, lr}
    138 #ifdef BIG_LOOPS
    139 	mov	r2, #(PAGE_SIZE >> 9)
    140 #else
    141 	mov	r2, #(PAGE_SIZE >> 7)
    142 #endif
    143 	mov	r3, #0
    144 	mov	r4, #0
    145 	mov	r5, #0
    146 	mov	r6, #0
    147 	mov	r7, #0
    148 	mov	r8, #0
    149 	mov	ip, #0
    150 	mov	lr, #0
    151 
    152 1:
    153 	stmia	r0!, {r3-r8,ip,lr}
    154 	stmia	r0!, {r3-r8,ip,lr}
    155 	stmia	r0!, {r3-r8,ip,lr}
    156 	stmia	r0!, {r3-r8,ip,lr}
    157 
    158 #ifdef BIG_LOOPS
    159 	/* There is little point making the loop any larger; unless we are
    160 	   running with the cache off, the load/store overheads will
    161 	   completely dominate this loop.  */
    162 	stmia	r0!, {r3-r8,ip,lr}
    163 	stmia	r0!, {r3-r8,ip,lr}
    164 	stmia	r0!, {r3-r8,ip,lr}
    165 	stmia	r0!, {r3-r8,ip,lr}
    166 
    167 	stmia	r0!, {r3-r8,ip,lr}
    168 	stmia	r0!, {r3-r8,ip,lr}
    169 	stmia	r0!, {r3-r8,ip,lr}
    170 	stmia	r0!, {r3-r8,ip,lr}
    171 
    172 	stmia	r0!, {r3-r8,ip,lr}
    173 	stmia	r0!, {r3-r8,ip,lr}
    174 	stmia	r0!, {r3-r8,ip,lr}
    175 	stmia	r0!, {r3-r8,ip,lr}
    176 
    177 #endif
    178 
    179 	subs	r2, r2, #1
    180 	bne	1b
    181 
    182 	ldmfd	sp!, {r4-r8, pc}
    183