Home | History | Annotate | Line # | Download | only in arm32
bcopy_page.S revision 1.2.6.3
      1  1.2.6.3  thorpej /*	$NetBSD: bcopy_page.S,v 1.2.6.3 2002/08/19 21:38:58 thorpej Exp $	*/
      2  1.2.6.2  nathanw 
      3  1.2.6.2  nathanw /*
      4  1.2.6.2  nathanw  * Copyright (c) 1995 Scott Stevens
      5  1.2.6.2  nathanw  * All rights reserved.
      6  1.2.6.2  nathanw  *
      7  1.2.6.2  nathanw  * Redistribution and use in source and binary forms, with or without
      8  1.2.6.2  nathanw  * modification, are permitted provided that the following conditions
      9  1.2.6.2  nathanw  * are met:
     10  1.2.6.2  nathanw  * 1. Redistributions of source code must retain the above copyright
     11  1.2.6.2  nathanw  *    notice, this list of conditions and the following disclaimer.
     12  1.2.6.2  nathanw  * 2. Redistributions in binary form must reproduce the above copyright
     13  1.2.6.2  nathanw  *    notice, this list of conditions and the following disclaimer in the
     14  1.2.6.2  nathanw  *    documentation and/or other materials provided with the distribution.
     15  1.2.6.2  nathanw  * 3. All advertising materials mentioning features or use of this software
     16  1.2.6.2  nathanw  *    must display the following acknowledgement:
     17  1.2.6.2  nathanw  *	This product includes software developed by Scott Stevens.
     18  1.2.6.2  nathanw  * 4. The name of the author may not be used to endorse or promote products
     19  1.2.6.2  nathanw  *    derived from this software without specific prior written permission.
     20  1.2.6.2  nathanw  *
     21  1.2.6.2  nathanw  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
     22  1.2.6.2  nathanw  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
     23  1.2.6.2  nathanw  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
     24  1.2.6.2  nathanw  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
     25  1.2.6.2  nathanw  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
     26  1.2.6.2  nathanw  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
     27  1.2.6.2  nathanw  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
     28  1.2.6.2  nathanw  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
     29  1.2.6.2  nathanw  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
     30  1.2.6.2  nathanw  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
     31  1.2.6.2  nathanw  *
     32  1.2.6.2  nathanw  * RiscBSD kernel project
     33  1.2.6.2  nathanw  *
     34  1.2.6.2  nathanw  * bcopy_page.S
     35  1.2.6.2  nathanw  *
     36  1.2.6.2  nathanw  * page optimised bcopy and bzero routines
     37  1.2.6.2  nathanw  *
     38  1.2.6.2  nathanw  * Created      : 08/04/95
     39  1.2.6.2  nathanw  */
     40  1.2.6.2  nathanw 
     41  1.2.6.2  nathanw #include <machine/param.h>
     42  1.2.6.2  nathanw #include <machine/asm.h>
     43  1.2.6.2  nathanw 
     44  1.2.6.2  nathanw /* #define BIG_LOOPS */
     45  1.2.6.2  nathanw 
     46  1.2.6.2  nathanw /*
     47  1.2.6.2  nathanw  * bcopy_page(src, dest)
     48  1.2.6.2  nathanw  *
     49  1.2.6.2  nathanw  * Optimised copy page routine.
     50  1.2.6.2  nathanw  *
     51  1.2.6.2  nathanw  * On entry:
     52  1.2.6.2  nathanw  *   r0 - src address
     53  1.2.6.2  nathanw  *   r1 - dest address
     54  1.2.6.2  nathanw  *
     55  1.2.6.2  nathanw  * Requires:
     56  1.2.6.2  nathanw  *   number of bytes per page (NBPG) is a multiple of 512 (BIG_LOOPS), 128
     57  1.2.6.2  nathanw  *   otherwise.
     58  1.2.6.2  nathanw  */
     59  1.2.6.2  nathanw 
     60  1.2.6.2  nathanw #define	CHUNK_SIZE	32
     61  1.2.6.2  nathanw 
     62  1.2.6.2  nathanw #ifdef __XSCALE__
     63  1.2.6.2  nathanw 	/* Conveniently, the chunk size is the XScale cache line size. */
     64  1.2.6.2  nathanw #define	PREFETCH_FIRST_CHUNK	pld	[r0]
     65  1.2.6.2  nathanw #define	PREFETCH_NEXT_CHUNK	pld	[r0, #(CHUNK_SIZE)]
     66  1.2.6.2  nathanw #else
     67  1.2.6.2  nathanw #define	PREFETCH_FIRST_CHUNK	/* nothing */
     68  1.2.6.2  nathanw #define	PREFETCH_NEXT_CHUNK	/* nothing */
     69  1.2.6.2  nathanw #endif
     70  1.2.6.2  nathanw 
     71  1.2.6.2  nathanw #ifndef COPY_CHUNK
     72  1.2.6.2  nathanw #define	COPY_CHUNK \
     73  1.2.6.2  nathanw 	PREFETCH_NEXT_CHUNK ; \
     74  1.2.6.2  nathanw 	ldmia	r0!, {r3-r8,ip,lr} ; \
     75  1.2.6.2  nathanw 	stmia	r1!, {r3-r8,ip,lr}
     76  1.2.6.2  nathanw #endif /* ! COPY_CHUNK */
     77  1.2.6.2  nathanw 
     78  1.2.6.2  nathanw #ifndef SAVE_REGS
     79  1.2.6.2  nathanw #define	SAVE_REGS	stmfd	sp!, {r4-r8, lr}
     80  1.2.6.2  nathanw #define	RESTORE_REGS	ldmfd	sp!, {r4-r8, pc}
     81  1.2.6.2  nathanw #endif
     82  1.2.6.2  nathanw 
     83  1.2.6.2  nathanw ENTRY(bcopy_page)
     84  1.2.6.2  nathanw 	PREFETCH_FIRST_CHUNK
     85  1.2.6.2  nathanw 	SAVE_REGS
     86  1.2.6.2  nathanw #ifdef BIG_LOOPS
     87  1.2.6.2  nathanw 	mov	r2, #(NBPG >> 9)
     88  1.2.6.2  nathanw #else
     89  1.2.6.2  nathanw 	mov	r2, #(NBPG >> 7)
     90  1.2.6.2  nathanw #endif
     91  1.2.6.2  nathanw 
     92  1.2.6.3  thorpej 1:
     93  1.2.6.2  nathanw 	COPY_CHUNK
     94  1.2.6.2  nathanw 	COPY_CHUNK
     95  1.2.6.2  nathanw 	COPY_CHUNK
     96  1.2.6.2  nathanw 	COPY_CHUNK
     97  1.2.6.2  nathanw 
     98  1.2.6.2  nathanw #ifdef BIG_LOOPS
     99  1.2.6.2  nathanw 	/* There is little point making the loop any larger; unless we are
    100  1.2.6.2  nathanw 	   running with the cache off, the load/store overheads will
    101  1.2.6.2  nathanw 	   completely dominate this loop.  */
    102  1.2.6.2  nathanw 	COPY_CHUNK
    103  1.2.6.2  nathanw 	COPY_CHUNK
    104  1.2.6.2  nathanw 	COPY_CHUNK
    105  1.2.6.2  nathanw 	COPY_CHUNK
    106  1.2.6.2  nathanw 
    107  1.2.6.2  nathanw 	COPY_CHUNK
    108  1.2.6.2  nathanw 	COPY_CHUNK
    109  1.2.6.2  nathanw 	COPY_CHUNK
    110  1.2.6.2  nathanw 	COPY_CHUNK
    111  1.2.6.2  nathanw 
    112  1.2.6.2  nathanw 	COPY_CHUNK
    113  1.2.6.2  nathanw 	COPY_CHUNK
    114  1.2.6.2  nathanw 	COPY_CHUNK
    115  1.2.6.2  nathanw 	COPY_CHUNK
    116  1.2.6.2  nathanw #endif
    117  1.2.6.2  nathanw 	subs	r2, r2, #1
    118  1.2.6.3  thorpej 	bne	1b
    119  1.2.6.2  nathanw 
    120  1.2.6.2  nathanw 	RESTORE_REGS		/* ...and return. */
    121  1.2.6.2  nathanw 
    122  1.2.6.2  nathanw /*
    123  1.2.6.2  nathanw  * bzero_page(dest)
    124  1.2.6.2  nathanw  *
    125  1.2.6.2  nathanw  * Optimised zero page routine.
    126  1.2.6.2  nathanw  *
    127  1.2.6.2  nathanw  * On entry:
    128  1.2.6.2  nathanw  *   r0 - dest address
    129  1.2.6.2  nathanw  *
    130  1.2.6.2  nathanw  * Requires:
    131  1.2.6.2  nathanw  *   number of bytes per page (NBPG) is a multiple of 512 (BIG_LOOPS), 128
    132  1.2.6.2  nathanw  *   otherwise
    133  1.2.6.2  nathanw  */
    134  1.2.6.2  nathanw 
    135  1.2.6.2  nathanw ENTRY(bzero_page)
    136  1.2.6.2  nathanw 	stmfd	sp!, {r4-r8, lr}
    137  1.2.6.2  nathanw #ifdef BIG_LOOPS
    138  1.2.6.2  nathanw 	mov	r2, #(NBPG >> 9)
    139  1.2.6.2  nathanw #else
    140  1.2.6.2  nathanw 	mov	r2, #(NBPG >> 7)
    141  1.2.6.2  nathanw #endif
    142  1.2.6.2  nathanw 	mov	r3, #0
    143  1.2.6.2  nathanw 	mov	r4, #0
    144  1.2.6.2  nathanw 	mov	r5, #0
    145  1.2.6.2  nathanw 	mov	r6, #0
    146  1.2.6.2  nathanw 	mov	r7, #0
    147  1.2.6.2  nathanw 	mov	r8, #0
    148  1.2.6.2  nathanw 	mov	ip, #0
    149  1.2.6.2  nathanw 	mov	lr, #0
    150  1.2.6.2  nathanw 
    151  1.2.6.3  thorpej 1:
    152  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    153  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    154  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    155  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    156  1.2.6.2  nathanw 
    157  1.2.6.2  nathanw #ifdef BIG_LOOPS
    158  1.2.6.2  nathanw 	/* There is little point making the loop any larger; unless we are
    159  1.2.6.2  nathanw 	   running with the cache off, the load/store overheads will
    160  1.2.6.2  nathanw 	   completely dominate this loop.  */
    161  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    162  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    163  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    164  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    165  1.2.6.2  nathanw 
    166  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    167  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    168  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    169  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    170  1.2.6.2  nathanw 
    171  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    172  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    173  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    174  1.2.6.2  nathanw 	stmia	r0!, {r3-r8,ip,lr}
    175  1.2.6.2  nathanw 
    176  1.2.6.2  nathanw #endif
    177  1.2.6.2  nathanw 
    178  1.2.6.2  nathanw 	subs	r2, r2, #1
    179  1.2.6.3  thorpej 	bne	1b
    180  1.2.6.2  nathanw 
    181  1.2.6.2  nathanw 	ldmfd	sp!, {r4-r8, pc}
    182