Nyaaori/rockbox
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
rockbox/firmware/target/coldfire/ata-as-coldfire.S
Jens Arnold b774bae5fd Coldfire targets: Fixed performance-hitting bug for unaligned transfers. Now unaligned transfers are less than 10% slower than aligned transfers. 
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@11679 a1c6a512-1295-4272-9138-f99709370657
2006-12-06 18:44:45 +00:00

459 lines
13 KiB
ArmAsm
Executable file
Raw Blame History

/***************************************************************************
* __________ __ ___.
* 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.
*
****************************************************************************/
.section .icode,"ax",@progbits
.equ .ata_port, 0x20000020
.align 2
.global copy_read_sectors
.type copy_read_sectors,@function
/* Read a number of words from the ATA data port
*
* Utilises line bursts, assumes there is at least one full line to copy.
*
* Arguments:
* (4,%sp) - buffer address
* (8,%sp) - word count
*
* Register usage:
* %a0 - current address
* %a1 - end address
* %a2 - ata port
* %d0 - scratch
* %d1 - shift count
* %d2-%d6 - read buffers
*/
copy_read_sectors:
lea.l (-24, %sp), %sp
movem.l %d2-%d6/%a2, (%sp)
movem.l (28, %sp), %a0-%a1
add.l %a1, %a1
add.l %a0, %a1
lea.l .ata_port, %a2
move.l %a0, %d0
btst.l #0, %d0 /* 16-bit aligned? */
jeq .r_aligned /* yes, do word copy */
/* not 16-bit aligned */
subq.l #1, %a1 /* last byte is done unconditionally */
moveq.l #24, %d1 /* preload shift count */
move.w (%a2), %d2 /* load initial word */
move.l %d2, %d3
lsr.l #8, %d3
move.b %d3, (%a0)+ /* write high byte of it, aligns dest addr */
btst.l #1, %d0 /* longword aligned? (testing old d0 value!) */
bne.b .r_end_u_w1 /* yes, skip leading word handling */
swap %d2 /* move initial word up */
move.w (%a2), %d2 /* combine with second word */
move.l %d2, %d3
lsr.l #8, %d3
move.w %d3, (%a0)+ /* write bytes 2 and 3 as word */
.r_end_u_w1:
moveq.l #12, %d0
add.l %a0, %d0
and.l #0xFFFFFFF0,%d0 /* d0 == first line bound */
cmp.l %a0, %d0 /* any leading longwords? */
bls.b .r_end_u_l1 /* no: skip loop */
.r_loop_u_l1:
move.w (%a2), %d3 /* load first word */
swap %d3 /* move to upper 16 bit */
move.w (%a2), %d3 /* load second word */
move.l %d3, %d4
lsl.l %d1, %d2
lsr.l #8, %d3
or.l %d3, %d2 /* combine old low byte with new top 3 bytes */
move.l %d2, (%a0)+ /* store as long */
move.l %d4, %d2
cmp.l %a0, %d0 /* run up to first line bound */
bhi.b .r_loop_u_l1
.r_end_u_l1:
lea.l (-14, %a1), %a1 /* adjust end addr. to 16 bytes/pass */
.r_loop_u_line:
move.w (%a2), %d3 /* load 1st word */
swap %d3 /* move to upper 16 bit */
move.w (%a2), %d3 /* load 2nd word */
move.l %d3, %d0
lsl.l %d1, %d2
lsr.l #8, %d0
or.l %d0, %d2 /* combine old low byte with new top 3 bytes */
move.w (%a2), %d4 /* load 3rd word */
swap %d4 /* move to upper 16 bit */
move.w (%a2), %d4 /* load 4th word */
move.l %d4, %d0
lsl.l %d1, %d3
lsr.l #8, %d0
or.l %d0, %d3 /* combine old low byte with new top 3 bytes */
move.w (%a2), %d5 /* load 5th word */
swap %d5 /* move to upper 16 bit */
move.w (%a2), %d5 /* load 6th word */
move.l %d5, %d0
lsl.l %d1, %d4
lsr.l #8, %d0
or.l %d0, %d4 /* combine old low byte with new top 3 bytes */
move.w (%a2), %d6 /* load 7th word */
swap %d6 /* move to upper 16 bit */
move.w (%a2), %d6 /* load 8th word */
move.l %d6, %d0
lsl.l %d1, %d5
lsr.l #8, %d0
or.l %d0, %d5 /* combine old low byte with new top 3 bytes */
movem.l %d2-%d5, (%a0) /* store line */
lea.l (16, %a0), %a0
move.l %d6, %d2
cmp.l %a0, %a1 /* run up to last line bound */
bhi.b .r_loop_u_line
lea.l (12, %a1), %a1 /* readjust for longword loop */
cmp.l %a0, %a1 /* any trailing longwords? */
bls.b .r_end_u_l2 /* no: skip loop */
.r_loop_u_l2:
move.w (%a2), %d3 /* load first word */
swap %d3 /* move to upper 16 bit */
move.w (%a2), %d3 /* load second word */
move.l %d3, %d4
lsl.l %d1, %d2
lsr.l #8, %d3
or.l %d3, %d2 /* combine old low byte with new top 3 bytes */
move.l %d2, (%a0)+ /* store as long */
move.l %d4, %d2
cmp.l %a0, %a1 /* run up to last long bound */
bhi.b .r_loop_u_l2
.r_end_u_l2:
addq.l #2, %a1 /* back to final end address */
cmp.l %a0, %a1 /* one word left? */
bls.b .r_end_u_w2
swap %d2 /* move old word to upper 16 bits */
move.w (%a2), %d2 /* load final word */
move.l %d2, %d3
lsr.l #8, %d3
move.w %d3, (%a0)+ /* write bytes 2 and 3 as word */
.r_end_u_w2:
move.b %d2, (%a0)+ /* store final byte */
bra.b .r_exit
/* 16-bit aligned */
.r_aligned:
btst.l #1, %d0 /* longword aligned? */
beq.b .r_end_a_w1 /* yes, skip leading word handling */
move.w (%a2), (%a0)+ /* copy initial word */
.r_end_a_w1:
moveq.l #12, %d0
add.l %a0, %d0
and.l #0xFFFFFFF0,%d0 /* d0 == first line bound */
cmp.l %a0, %d0 /* any leading longwords? */
bls.b .r_end_a_l1 /* no: skip loop */
.r_loop_a_l1:
move.w (%a2), %d1 /* load first word */
swap %d1 /* move it to upper 16 bits */
move.w (%a2), %d1 /* load second word */
move.l %d1, (%a0)+ /* store as long */
cmp.l %a0, %d0 /* run up to first line bound */
bhi.b .r_loop_a_l1
.r_end_a_l1:
lea.l (-14, %a1), %a1 /* adjust end addr. to 16 bytes/pass */
.r_loop_a_line:
move.w (%a2), %d0 /* load 1st word */
swap %d0 /* move it to upper 16 bits */
move.w (%a2), %d0 /* load 2nd word */
move.w (%a2), %d1 /* load 3rd word */
swap %d1 /* move it to upper 16 bits */
move.w (%a2), %d1 /* load 4th word */
move.w (%a2), %d2 /* load 5th word */
swap %d2 /* move it to upper 16 bits */
move.w (%a2), %d2 /* load 6th word */
move.w (%a2), %d3 /* load 7th word */
swap %d3 /* move it to upper 16 bits */
move.w (%a2), %d3 /* load 8th word */
movem.l %d0-%d3, (%a0) /* store line */
lea.l (16, %a0), %a0
cmp.l %a0, %a1 /* run up to last line bound */
bhi.b .r_loop_a_line
lea.l (12, %a1), %a1 /* readjust for longword loop */
cmp.l %a0, %a1 /* any trailing longwords? */
bls.b .r_end_a_l2 /* no: skip loop */
.r_loop_a_l2:
move.w (%a2), %d1 /* read first word */
swap %d1 /* move it to upper 16 bits */
move.w (%a2), %d1 /* read second word */
move.l %d1, (%a0)+ /* store as long */
cmp.l %a0, %a1 /* run up to last long bound */
bhi.b .r_loop_a_l2
.r_end_a_l2:
addq.l #2, %a1 /* back to final end address */
cmp.l %a0, %a1 /* one word left? */
bls.b .r_end_a_w2
move.w (%a2), (%a0)+ /* copy final word */
.r_end_a_w2:
.r_exit:
movem.l (%sp), %d2-%d6/%a2
lea.l (24, %sp), %sp
rts
.r_end:
.size copy_read_sectors,.r_end-copy_read_sectors
.align 2
.global copy_write_sectors
.type copy_write_sectors,@function
/* Write a number of words to the ATA data port
*
* Utilises line bursts, assumes there is at least one full line to copy.
*
* Arguments:
* (4,%sp) - buffer address
* (8,%sp) - word count
*
* Register usage:
* %a0 - current address
* %a1 - end address
* %a2 - ata port
* %d0 - scratch
* %d1 - shift count
* %d2-%d6 - read buffers
*/
copy_write_sectors:
lea.l (-24, %sp), %sp
movem.l %d2-%d6/%a2, (%sp)
movem.l (28, %sp), %a0-%a1
add.l %a1, %a1
add.l %a0, %a1
lea.l .ata_port, %a2
move.l %a0, %d0
btst.l #0, %d0 /* 16-bit aligned? */
jeq .w_aligned /* yes, do word copy */
/* not 16-bit aligned */
subq.l #1, %a1 /* last byte is done unconditionally */
moveq.l #24, %d1 /* preload shift count */
move.b (%a0)+, %d2
btst.l #1, %d0 /* longword aligned? (testing old d0 value!) */
bne.b .w_end_u_w1 /* yes, skip leading word handling */
swap %d2
move.w (%a0)+, %d2
move.l %d2, %d3
lsr.l #8, %d3
move.w %d3, (%a2)
.w_end_u_w1:
moveq.l #12, %d0
add.l %a0, %d0
and.l #0xFFFFFFF0,%d0 /* d0 == first line bound */
cmp.l %a0, %d0 /* any leading longwords? */
bls.b .w_end_u_l1 /* no: skip loop */
.w_loop_u_l1:
move.l (%a0)+, %d3
move.l %d3, %d4
lsl.l %d1, %d2
lsr.l #8, %d3
or.l %d3, %d2
swap %d2
move.w %d2, (%a2)
swap %d2
move.w %d2, (%a2)
move.l %d4, %d2
cmp.l %a0, %d0 /* run up to first line bound */
bhi.b .w_loop_u_l1
.w_end_u_l1:
lea.l (-14, %a1), %a1 /* adjust end addr. to 16 bytes/pass */
.w_loop_u_line:
movem.l (%a0), %d3-%d6
lea.l (16, %a0), %a0
move.l %d3, %d0
lsl.l %d1, %d2
lsr.l #8, %d0
or.l %d0, %d2
swap %d2
move.w %d2, (%a2)
swap %d2
move.w %d2, (%a2)
move.l %d4, %d0
lsl.l %d1, %d3
lsr.l #8, %d0
or.l %d0, %d3
swap %d3
move.w %d3, (%a2)
swap %d3
move.w %d3, (%a2)
move.l %d5, %d0
lsl.l %d1, %d4
lsr.l #8, %d0
or.l %d0, %d4
swap %d4
move.w %d4, (%a2)
swap %d4
move.w %d4, (%a2)
move.l %d6, %d0
lsl.l %d1, %d5
lsr.l #8, %d0
or.l %d0, %d5
swap %d5
move.w %d5, (%a2)
swap %d5
move.w %d5, (%a2)
move.l %d6, %d2
cmp.l %a0, %a1 /* run up to last line bound */
bhi.b .w_loop_u_line
lea.l (12, %a1), %a1 /* readjust for longword loop */
cmp.l %a0, %a1 /* any trailing longwords? */
bls.b .w_end_u_l2 /* no: skip loop */
.w_loop_u_l2:
move.l (%a0)+, %d3
move.l %d3, %d4
lsl.l %d1, %d2
lsr.l #8, %d3
or.l %d3, %d2
swap %d2
move.w %d2, (%a2)
swap %d2
move.w %d2, (%a2)
move.l %d4, %d2
cmp.l %a0, %a1 /* run up to first line bound */
bhi.b .w_loop_u_l2
.w_end_u_l2:
addq.l #2, %a1 /* back to final end address */
cmp.l %a0, %a1 /* one word left? */
bls.b .w_end_u_w2
swap %d2
move.w (%a0)+, %d2
move.l %d2, %d3
lsr.l #8, %d3
move.w %d3, (%a2)
.w_end_u_w2:
lsl.l #8, %d2
move.b (%a0)+, %d2
move.w %d2, (%a2)
bra.b .w_exit
/* 16-bit aligned */
.w_aligned:
btst.l #1, %d0
beq.b .w_end_a_w1
move.w (%a0)+, (%a2) /* copy initial word */
.w_end_a_w1:
moveq.l #12, %d0
add.l %a0, %d0
and.l #0xFFFFFFF0,%d0 /* d0 == first line bound */
cmp.l %a0, %d0 /* any leading longwords? */
bls.b .w_end_a_l1 /* no: skip loop */
.w_loop_a_l1:
move.l (%a0)+, %d1
swap %d1
move.w %d1, (%a2)
swap %d1
move.w %d1, (%a2)
cmp.l %a0, %d0 /* run up to first line bound */
bhi.b .w_loop_a_l1
.w_end_a_l1:
lea.l (-14, %a1), %a1 /* adjust end addr. to 16 bytes/pass */
.w_loop_a_line:
movem.l (%a0), %d0-%d3
lea.l (16, %a0), %a0
swap %d0
move.w %d0, (%a2)
swap %d0
move.w %d0, (%a2)
swap %d1
move.w %d1, (%a2)
swap %d1
move.w %d1, (%a2)
swap %d2
move.w %d2, (%a2)
swap %d2
move.w %d2, (%a2)
swap %d3
move.w %d3, (%a2)
swap %d3
move.w %d3, (%a2)
cmp.l %a0, %a1 /* run up to last line bound */
bhi.b .w_loop_a_line
lea.l (12, %a1), %a1 /* readjust for longword loop */
cmp.l %a0, %a1 /* any trailing longwords? */
bls.b .w_end_a_l2 /* no: skip loop */
.w_loop_a_l2:
move.l (%a0)+, %d1
swap %d1
move.w %d1, (%a2)
swap %d1
move.w %d1, (%a2)
cmp.l %a0, %a1 /* run up to first line bound */
bhi.b .w_loop_a_l2
.w_end_a_l2:
addq.l #2, %a1 /* back to final end address */
cmp.l %a0, %a1 /* one word left? */
bls.b .w_end_a_w2
move.w (%a0)+, (%a2) /* copy final word */
.w_end_a_w2:
.w_exit:
movem.l (%sp), %d2-%d6/%a2
lea.l (24, %sp), %sp
rts
.w_end:
.size copy_write_sectors,.w_end-copy_write_sectors
Reference in a new issue View git blame Copy permalink