rockbox/apps/dsp_cf.S
Thom Johansen cc94ae45a7 Keep assembler versions of DSP routines in dsp_arch.S files. This also
solves the annoying register allocator problem when not using
-fomit-frame-pointer for the Coldfire assembler optimised crossfeed
routine.


git-svn-id: svn://svn.rockbox.org/rockbox/trunk@9215 a1c6a512-1295-4272-9138-f99709370657
2006-03-23 19:59:52 +00:00

104 lines
4.2 KiB
ArmAsm

/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2006 Thom Johansen
*
* 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 .idata,"aw",@progbits
crossfeed_coefs:
.long 0x4CCCCCCD | LOW
.long 0x33333333 | LOW_COMP
.long -0x66666666 | HIGH_NEG
.long 0x66666666 | HIGH_COMP
.long 0x0CCCCCCD | ATT
.long 0x73333333 | ATT_COMP
.section .text
.global apply_crossfeed
apply_crossfeed:
lea.l (-44, %sp), %sp
movem.l %d2-%d7/%a2-%a6, (%sp) | save all regs
move.l (44+4, %sp), %a4
movem.l (%a4), %a4-%a5 | a4 = src[0], a5 = src[1]
move.l (44+8, %sp), %d7 | d7 = count
lea.l crossfeed_data, %a1
lea.l crossfeed_coefs, %a6
lea.l (16, %a1), %a0 | a0 = &delay[0][0]
movem.l (%a1), %d0-%d3 | fetch filter history samples
move.l (120, %a1), %d4 | fetch delay line index
move.l (%a4), %d5 | d5 = left sample
move.l (%a5), %d6 | d6 = right sample
move.l (%a6)+, %a1 | a1 = LOW value
move.l (%a6)+, %a2 | a2 = LOW_COMP value
/* Register usage in loop:
* a0 = &delay[0][0], a1 & a2 = coefs, a3 = temp storage,
* a4 = src[0], a5 = src[1], a6 = &crossfeed_coefs[0],
* d0 = low_left, d1 = low_right,
* d2 = high_left, d3 = high_right,
* d4 = delay line index,
* d5 = src[0][i], d6 = src[1][i].
* d7 = count
*/
.cfloop:
| LOW*low_left + LOW_COMP*left
mac.l %a1, %d0, %acc0
mac.l %a2, %d5, %acc0
| LOW*low_right + LOW_COMP*right
mac.l %a1, %d1, (%a6)+, %a1, %acc1 | a1 = HIGH_NEG
mac.l %a2, %d6, (%a6)+, %a2, %acc1 | a2 = HIGH_COMP
movclr.l %acc0, %d0 | get low_left
movclr.l %acc1, %d1 | get low_right
| HIGH_NEG*high_left + HIGH_COMP*left
mac.l %a1, %d2, %acc0
mac.l %a2, %d5, %acc0
| HIGH_NEG*high_right + HIGH_COMP*right
mac.l %a1, %d3, (%a6)+, %a1, %acc1 | a1 = ATT
mac.l %a2, %d6, (%a6)+, %a2, %acc1 | a2 = ATT_COMP
lea.l (-6*4, %a6), %a6 | coef = &coefs[0]
move.l (%a0, %d4*4), %a3 | a3 = delay[0][idx]
move.l (52, %a0, %d4*4), %d5 | d5 = delay[1][idx]
movclr.l %acc0, %d2 | get high_left
movclr.l %acc1, %d3 | get high_right
| ATT*delay_r + ATT_COMP*high_left
mac.l %a1, %d5, (4, %a4), %d5, %acc0 | d5 = src[0][i+1]
mac.l %a2, %d2, (4, %a5), %d6, %acc0 | d6 = src[1][i+1]
| ATT*delay_l + ATT_COMP*high_right
mac.l %a1, %a3, (%a6)+, %a1, %acc1 | a1 = LOW
mac.l %a2, %d3, (%a6)+, %a2, %acc1 | a2 = LOW_COMP
| save crossfed samples to output
movclr.l %acc0, %a3
move.l %a3, (%a4)+ | src[0][i++] = out_l
movclr.l %acc1, %a3
move.l %a3, (%a5)+ | src[1][i++] = out_r
move.l %d0, (%a0, %d4*4) | delay[0][index] = low_left
move.l %d1, (52, %a0, %d4*4) | delay[1][index] = low_right */
addq.l #1, %d4 | index++ */
cmp.l #13, %d4 | if (index >= 13) {
jlt .nowrap
clr.l %d4 | index = 0
.nowrap: | }
subq.l #1, %d7
jne .cfloop
| save data back to struct
lea.l crossfeed_data, %a1
movem.l %d0-%d3, (%a1)
move.l %d4, (120, %a1)
movem.l (%sp), %d2-%d7/%a2-%a6
lea.l (44, %sp), %sp
rts