Nyaaori/rockbox
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity
rockbox/lib/rbcodec/dsp/dsp_cf.S

905 lines
42 KiB
ArmAsm
Raw Normal View History

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
/***************************************************************************
* __________ __ ___.
* Open \______ \ ____ ____ | | _\_ |__ _______ ___
* Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /
* Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <
* Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \
* \/ \/ \/ \/ \/
* $Id$
*
* Copyright (C) 2006 Thom Johansen
New crossfeed algorithm for Rockbox: "Meier" crossfeed Emulates the basic "Meier" crossfeed (2 capacitors, 3 resistors) as discussed in http://www.meier-audio.homepage.t-online.de/passivefilter.htm This crossfeed blends a bit of low-pass filtered L signal into the R signal (and vice versa) while adding about 300 us delay to the crossfed-signal. A difference with the crossfeed already present in rockbox, is that this algorithm keeps the total spectrum flat (the one currently in rockbox accentuates low-frequency signals, making it sound a bit muffled). This implementation is quite lightweight, just 3 multiplies per left-right pair of samples. Has a default C implementation and optimized assembly versions for ARM and Coldfire. The crossfeed effect is quite subtle and is noticeable mostly one albums that have very strong left-right separation (e.g. one instrument only on the left, another only on the right). In the user interface, the new crossfeed option appears as "Meier" and is not configureable. The existing crossfeed is renamed to "Custom" as it allows itself to be customised. There is no entry for the user manual yet. Change-Id: Iaa100616fe0fcd7e16f08cdb9a7f41501973eee1
2012-05-01 07:58:27 +00:00
* Copyright (C) 2007, 2012 Michael Sevakis
* Copyright (C) 2010 Bertrik Sikken
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
*
Updated our source code header to explicitly mention that we are GPL v2 or later. We still need to hunt down snippets used that are not. 1324 modified files... http://www.rockbox.org/mail/archive/rockbox-dev-archive-2008-06/0060.shtml git-svn-id: svn://svn.rockbox.org/rockbox/trunk@17847 a1c6a512-1295-4272-9138-f99709370657
2008-06-28 18:10:04 +00:00
* 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.
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
*
* This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY
* KIND, either express or implied.
*
****************************************************************************/
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
#include "config.h"
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
SWCODEC: DSP optimizations for conversion to internal format and resampling. Assembly resampling for Coldfire. Word has it ARM will get that soon. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12399 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 02:49:26 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void pga_process(struct dsp_proc_entry *this, struct dsp_buffer **buf_p)
SWCODEC: DSP optimizations for conversion to internal format and resampling. Assembly resampling for Coldfire. Word has it ARM will get that soon. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12399 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 02:49:26 +00:00
*/
Code police myself. Somehow got that full of tabs. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12401 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 03:29:37 +00:00
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.global pga_process
pga_process:
| input: 4(sp) = this, 8(sp) = buf_p
movem.l 4(%sp), %a0-%a1 | %a0 = this, %a1 = buf_p
move.l (%a0), %a0 | %a0 = this->data = &pga_data
move.l (%a0), %a0 | %a0 = data->gain
move.l (%a1), %a1 | %a1 = buf = *buf_p
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
lea.l -20(%sp), %sp | save registers
movem.l %d2-%d4/%a2-%a3, (%sp) |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
clr.l %d1 | %d1 = buf->format.num_channels
move.b 17(%a1), %d1 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
10: | channel loop |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
move.l (%a1), %d0 | %d0 = buf->remcount
move.l (%a1, %d1.l*4), %a2 | %a2 = s = buf->p32[ch-1]
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
move.l %a2, %a3 | %a3 = d = s
move.l (%a2)+, %d2 | %d2 = *s++,
mac.l %a0, %d2, (%a2)+, %d2, %acc0 | %acc0 = S(n)*gain, load S(n+1)
subq.l #1, %d0 | --count > 0 ? : effectively n++
ble.b 30f | loop done | no? finish up
20: | loop |
move.l %accext01, %d4 | fetch S(n-1)[7:0]
movclr.l %acc0, %d3 | fetch S(n-1)[40:8] in %d5[31:0]
asl.l #8, %d3 | *s++ = (S(n-1)[40:8] << 8) | S(n-1)[7:0]
mac.l %a0, %d2, (%a2)+, %d2, %acc0 | %acc0 = S(n)*gain, load S(n+1)
move.b %d4, %d3 |
move.l %d3, (%a3)+ |
subq.l #1, %d0 | --count > 0 ? : effectively n++
bgt.b 20b | loop | yes? do more samples
30: | loop done |
move.l %accext01, %d4 | fetch S(n-1)[7:0]
movclr.l %acc0, %d3 | fetch S(n-1)[40:8] in %d5[31:0]
asl.l #8, %d3 | *s = (S(n-1)[40:8] << 8) | S(n-1)[7:0]
move.b %d4, %d3 |
move.l %d3, (%a3) |
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
subq.l #1, %d1 | next channel
bgt.b 10b | channel loop |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
movem.l (%sp), %d2-%d4/%a2-%a3 | restore registers
lea.l 20(%sp), %sp | cleanup stack
rts |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.size pga_process, .-pga_process
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void crossfeed_process(struct dsp_proc_entry *this,
* struct dsp_buffer **buf_p)
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
*/
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.global crossfeed_process
crossfeed_process:
| input: 4(sp) = this, 8(sp) = buf_p
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
lea.l -44(%sp), %sp |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movem.l %d2-%d7/%a2-%a6, (%sp) | save all regs
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l 48(%sp), %a1/%a4 | %a1 = this, %a4 = buf_p
move.l (%a4), %a4 | %a4 = buf = *buf_p
movem.l (%a4), %d7/%a4-%a5 | %d7 = buf->remcount, %a4 = buf->p32[0],
| %a5 = buf->p32[1]
move.l (%a1), %a1 | %a1 = &crossfeed_state
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
move.l (%a1)+, %d6 | %d6 = direct gain
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movem.l 12(%a1), %d0-%d3 | fetch filter history samples
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
lea.l 132(%a1), %a6 | %a6 = delay line wrap limit
move.l (%a6), %a0 | fetch delay line address
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movem.l (%a1), %a1-%a3 | load filter coefs
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bra.b 20f | loop start | go to loop start point
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
/* Register usage in loop:
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
* %a0 = delay_p, %a1..%a3 = b0, b1, a1 (filter coefs),
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
* %a4 = buf[0], %a5 = buf[1],
* %a6 = delay line pointer wrap limit,
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
* %d0..%d3 = history
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
* %d4..%d5 = temp.
* %d6 = direct gain,
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
* %d7 = count
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
*/
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
10: | loop |
movclr.l %acc0, %d4 | write outputs
move.l %d4, (%a4)+ | .
movclr.l %acc1, %d5 | .
move.l %d5, (%a5)+ | .
20: | loop start |
mac.l %a2, %d0, (%a0)+, %d0, %acc0 | %acc0 = b1*dl[n - 1], %d0 = dl[n]
mac.l %a1, %d0 , %acc0 | %acc0 += b0*dl[n]
mac.l %a3, %d1, (%a5), %d5, %acc0 | %acc0 += a1*y_r[n - 1], load R
mac.l %a2, %d2, (%a0)+, %d2, %acc1 | %acc1 = b1*dr[n - 1], %d2 = dr[n]
mac.l %a1, %d2 , %acc1 | %acc1 += b0*dr[n]
mac.l %a3, %d3, (%a4), %d4, %acc1 | %acc1 += a1*y_l[n - 1], load L
movem.l %d4-%d5, -8(%a0) | save left & right inputs to delay line
move.l %acc0, %d3 | get filtered delayed left sample (y_l[n])
move.l %acc1, %d1 | get filtered delayed right sample (y_r[n])
mac.l %d6, %d4, %acc0 | %acc0 += gain*x_l[n]
mac.l %d6, %d5, %acc1 | %acc1 += gain*x_r[n]
cmp.l %a6, %a0 | wrap %a0 if passed end
bhs.b 30f | wrap buffer |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
tpf.l | trap the buffer wrap
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
30: | wrap buffer | ...fwd taken branches more costly
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
lea.l -104(%a6), %a0 | wrap it up
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
subq.l #1, %d7 | --count > 0 ?
bgt.b 10b | loop | yes? do more
movclr.l %acc0, %d4 | write last outputs
move.l %d4, (%a4) | .
movclr.l %acc1, %d5 | .
move.l %d5, (%a5) | .
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l %d0-%d3, -120(%a6) | ...history
move.l %a0, (%a6) | ...delay_p
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movem.l (%sp), %d2-%d7/%a2-%a6 | restore all regs
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
lea.l 44(%sp), %sp |
rts |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.size crossfeed_process,.-crossfeed_process
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
New crossfeed algorithm for Rockbox: "Meier" crossfeed Emulates the basic "Meier" crossfeed (2 capacitors, 3 resistors) as discussed in http://www.meier-audio.homepage.t-online.de/passivefilter.htm This crossfeed blends a bit of low-pass filtered L signal into the R signal (and vice versa) while adding about 300 us delay to the crossfed-signal. A difference with the crossfeed already present in rockbox, is that this algorithm keeps the total spectrum flat (the one currently in rockbox accentuates low-frequency signals, making it sound a bit muffled). This implementation is quite lightweight, just 3 multiplies per left-right pair of samples. Has a default C implementation and optimized assembly versions for ARM and Coldfire. The crossfeed effect is quite subtle and is noticeable mostly one albums that have very strong left-right separation (e.g. one instrument only on the left, another only on the right). In the user interface, the new crossfeed option appears as "Meier" and is not configureable. The existing crossfeed is renamed to "Custom" as it allows itself to be customised. There is no entry for the user manual yet. Change-Id: Iaa100616fe0fcd7e16f08cdb9a7f41501973eee1
2012-05-01 07:58:27 +00:00
/****************************************************************************
* void crossfeed_meier_process(struct dsp_proc_entry *this,
* struct dsp_buffer **buf_p)
*/
.section .text
.global crossfeed_meier_process
crossfeed_meier_process:
| input: 4(sp) = this, 8(sp) = buf_p
movem.l 4(%sp), %a0-%a1 | %a0 = this, %a1 = buf_p
lea.l -24(%sp), %sp | save non-volatiles
movem.l %d2-%d6/%a2, (%sp) | .
move.l (%a0), %a0 | %a0 = &this->data = &crossfeed_state
move.l (%a1), %a1 | %a1 = buf = *buf_p
movem.l 16(%a0), %d1-%d5 | %d1 = vcl, %d2 = vcr, %d3 = vdiff,
| %d4 = coef1, %d5 = coef2
movem.l (%a1), %d0/%a1-%a2 | %d0 = count = buf->remcount
| %a1 = p32[0], %a2 = p32[1]
| Register usage in loop:
| %d0 = count, %d1 = vcl, %d2 = vcr, %d3 = vdiff/lout,
| %d4 = coef1, %d5 = coef2, %d6 = rout/scratch
| %a1 = p32[0], %a2 = p32[1]
10: | loop
mac.l %d5, %d3, %acc0 | %acc0 = common = coef2*vdiff
move.l %acc0, %acc1 | copy common
mac.l %d4, %d1, (%a1), %d3, %acc0 | %acc0 += coef1*vcl, %d3 = lout
msac.l %d4, %d2, (%a2), %d6, %acc1 | %acc1 -= coef1*vcr, %d6 = rout
add.l %d1, %d3 | lout += vcl
add.l %d2, %d6 | rout += vcr
move.l %d3, (%a1)+ | store left channel, pos inc
move.l %d6, (%a2)+ | store right channel, pos inc
sub.l %d6, %d3 | vdiff = lout - rout
movclr.l %acc0, %d6 | %d4 = fetch res1 in s0.31
sub.l %d6, %d1 | vcl -= res1
movclr.l %acc1, %d6 | %d5 = fetch -res2 in s0.31
add.l %d6, %d2 | vcr += -res2
subq.l #1, %d0 | count--
bgt 10b | loop | more samples?
|
movem.l %d1-%d3, 16(%a0) | save vcl, vcr, vdiff
movem.l (%sp), %d2-%d6/%a2 | restore non-volatiles
lea.l 24(%sp), %sp | .
rts |
.size crossfeed_meier_process, .-crossfeed_meier_process
SWCODEC: DSP optimizations for conversion to internal format and resampling. Assembly resampling for Coldfire. Word has it ARM will get that soon. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12399 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 02:49:26 +00:00
/****************************************************************************
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
* int resample_hermite(struct resample_data *data, struct dsp_buffer *src,
* struct dsp_buffer *dst)
SWCODEC: DSP optimizations for conversion to internal format and resampling. Assembly resampling for Coldfire. Word has it ARM will get that soon. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12399 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 02:49:26 +00:00
*/
Code police myself. Somehow got that full of tabs. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12401 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 03:29:37 +00:00
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
.global resample_hermite
resample_hermite:
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
| input: 4(sp) = data, 8(sp) = src, 12(sp) = dst
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
lea.l -52(%sp), %sp | save non-volatiles, allocate temps
movem.l %d2-%d7/%a2-%a6, 8(%sp) |
movem.l 56(%sp), %a0-%a2 | %a0 = data
Code police myself. Somehow got that full of tabs. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12401 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 03:29:37 +00:00
| %a1 = src
| %a2 = dst
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
clr.l %d5 | %d5 = ch = src->format.num_channels
move.b 17(%a1), %d5 |
lea.l 8(%a0), %a5 | %a5 = h = history[ch]
moveq.l #16, %d7 | %d7 = shift val
.hrs_channel_loop: |
movem.l %d5/%a5, (%sp) | store ch, h
movem.l (%a0), %d1-%d2 | %d1 = delta = data->delta,
| %d2 = phase = data->phase
move.l (%a1), %d3 | %d3 = srcrem = src->remcount
move.l 12(%a2), %d4 | %d4 = dstrem = dst->bufcount
cmp.l #0x8000, %d3 | %d4 = MIN(srcrem, 0x8000)
ble.b 1f |
move.l #0x8000, %d3 |
1: |
move.l (%a1, %d5.l*4), %a1 | %a1 = s = src->p32[ch]
move.l (%a2, %d5.l*4), %a2 | %a2 = d = dst->p32[ch]
move.l %d2, %d0 | %d0 = pos = phase >> 16
lsr.l %d7, %d0 |
cmp.l %d3, %d0 | pos = MIN(pos, srcrem)
ble.b 1f |
move.l %d3, %d0 |
1:
lea.l (%a1, %d0.l*4), %a1 | %a1 = &s[pos]
cmp.l #3, %d0 |
bge.b 1f |
move.l %d0, %a0 |
lea.l (%a0, %a0.l*2), %a0 |
jmp 2(%pc, %a0.l*4) | 4b |
| 0
movem.l (%a5), %a3-%a5 | 4b | x3..x1 = h[0]..h[2]
bra.b 2f | 2b |
.dcb.w 3,0 | 6b | filler
| 1
movem.l 4(%a5), %a3-%a4 | 6b | x3..x2 = h[1]..h[2]
move.l -4(%a1), %a5 | 4b | x1 = s[0]
bra.b 2f | 2b |
| 2
move.l 8(%a5), %a3 | 4b | x3 = h[2]
movem.l -8(%a1), %a4-%a5 | 6b | x2..x1 = s[0]..s[1]
bra.b 2f | 2b |
1: | 3 +
movem.l -12(%a1), %a3-%a5 | x3...x1 = s[pos-3]..s[pos-1]
2:
cmp.l %d3, %d0 | pos past end?
bge.w .hrs_channel_done |
cmp.l #0x10000, %d1 | delta >= 1.0?
bhs.w .hrs_dsstart | yes? downsampling
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
|
/** Upsampling **/ |
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
sub.l %d3, %d0 | %d0 = pos - srcrem = -dte
lsl.l %d7, %d1 | move delta to bits 30..15
lsr.l #1, %d1 |
lsl.l %d7, %d2 | move phase to bits 30..15
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
lsr.l #1, %d2 |
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
|
| Register usage in loop:
| r0 = dte, d1 = delta, d2 = phase, d3 = srcrem, d4 = dstrem
| d5 = scratch, d6 = c3, d7 = scratch
| a0 = c2, a1 = &s[pos], a2 = d,
| a3 = x3, a4 = x2, a5 = x1, a6 = x0
|
| Try to avoid overflow as much as possible and at the same time preserve
| accuracy. Same formulas apply to downsampling but registers and
| instruction order differ due to specific constraints.
| c1 = -0.5*x3 + 0.5*x1
| = 0.5*(x1 - x3) <--
|
| v = x1 - x2, -v = x2 - x1
| c2 = x3 - 2.5*x2 + 2*x1 - 0.5*x0
| = x3 + 2*(x1 - x2) - 0.5*(x0 + x2)
| = x3 + 2*v - 0.5*(x0 + x2) <--
|
| c3 = -0.5*x3 + 1.5*x2 - 1.5*x1 + 0.5*x0
| = 0.5*x0 - 0.5*x3 + 0.5*(x2 - x1) + (x2 - x1)
| = 0.5*(x0 - x3 - v) - v <--
|
.hrs_usloop_carry:
move.l (%a1)+, %a6 | %a6 = s[pos]
move.l %a5, %d5 | v
sub.l %a4, %d5 |
move.l %a6, %d6 | c3
sub.l %a3, %d6 |
sub.l %d5, %d6 |
asr.l #1, %d6 |
sub.l %d5, %d6 |
lea.l (%a3, %d5.l*2), %a0 | c2
move.l %a6, %d5 |
add.l %a4, %d5 |
asr.l #1, %d5 |
sub.l %d5, %a0 |
.hrs_usloop_frac:
move.l %a0, %acc0 | %acc0 = frac * c3 + c2
mac.l %d2, %d6, %acc0 |
move.l %a5, %d5 | c1
sub.l %a3, %d5 |
asr.l #1, %d5 |
movclr.l %acc0, %d7 | %acc1 = frac * acc + c1
move.l %d5, %acc1 |
mac.l %d2, %d7, %acc1 |
move.l %a4, %acc0 | %acc0 = frac * acc + x2
movclr.l %acc1, %d5 |
mac.l %d2, %d5, %acc0 |
subq.l #1, %d4 | dstrem <= 0?
ble.b .hrs_usfull | yes? stop
movclr.l %acc0, %d5 | *d++ = d5 = result
move.l %d5, (%a2)+ |
add.l %d1, %d2 | phase += delta
bpl.b .hrs_usloop_frac | load next values?
move.l %a4, %a3 | x3 = x2
move.l %a5, %a4 | x2 = x1
move.l %a6, %a5 | x1 = x0
bclr.l #31, %d2 | clear sign bit
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
addq.l #1, %d0 | dte > 0?
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
bmi.b .hrs_usloop_carry | yes? continue resampling
bra.b .hrs_usdone
.hrs_usfull:
movclr.l %acc0, %d5 | *d++ = d5 = result
move.l %d5, (%a2) |
add.l %d1, %d2 | do missed phase increment
bpl.b .hrs_usdone | was sign bit set?
move.l %a4, %a3 | do missed history update
move.l %a5, %a4 |
move.l %a6, %a5 |
addq.l #1, %d0 | do missed dte decrement
.hrs_usdone:
moveq.l #16, %d7 | restore shift
lsl.l #1, %d2 | frac -> phase
add.l %d3, %d0 | %d0 = -dte + srcrem = pos
or.l %d0, %d2 | restore phase
swap.w %d2 |
bra.w .hrs_channel_done |
/** Downsampling **/
|
| Register usage in loop:
| r0 = pos, d1 = delta, d2 = phase, d3 = srcrem, d4 = dstrem
| d5 = scratch, d6 = scratch, d7 = 16 (shift value)
| a0 = scratch, a1 = &s[pos], a2 = d,
| a3 = x3, a4 = x2, a5 = x1, a6 = x0
|
.hrs_dsloop:
movclr.l %acc0, %d5 | *d++ = acc
move.l %d5, (%a2)+ |
sub.l %d0, %a0 | %a0 = -shift = last_pos - pos
move.l %a0, %d5 |
asl.l #2, %d5 | -shift -> -bytes
sub.l %d5, %a1 | %a1 = s = s - -bytes
cmp.l #-4, %a0 | >= 4?
ble.b 1f |
add.l %d5, %a0 | %a0 = 5 * -shift
jmp 40(%pc, %a0.l*2) | 4b |
1: | +4 +
movem.l -12(%a1), %a3-%a5 | 6b | x3..x0 = s[pos-3]..s[pos-1]
bra.b 1f | 2b |
| +3
move.l %a6, %a3 | 2b | x3 = x0
movem.l -8(%a1), %a4-%a5 | 6b | x2..x0 = s[pos-2]..s[pos-1]
bra.b 1f | 2b | 10
| +2
move.l %a5, %a3 | 2b | x3 = x1
move.l %a6, %a4 | 2b | x2 = x0
move.l -4(%a1), %a5 | 4b | x1 = s[pos-1]
bra.b 1f | 2b | 10
| +1
move.l %a4, %a3 | 2b | x3 = x2 | expected loop destination
move.l %a5, %a4 | 2b | x2 = x1
move.l %a6, %a5 | 2b | x1 = x0
1:
subq.l #1, %d4 | 2b | dstrem <= 0?
ble.b .hrs_channel_done | 2b | yes? stop
cmp.l %d3, %d0 |
bge.b .hrs_channel_done |
.hrs_dsstart:
move.l (%a1), %a6 | %a6 = s[pos]
move.l %a5, %d5 | v
sub.l %a4, %d5 |
move.l %a6, %d6 | c3
sub.l %a3, %d6 |
sub.l %d5, %d6 |
asr.l #1, %d6 |
sub.l %d5, %d6 |
lea.l (%a3, %d5.l*2), %a0 | c2
move.l %a6, %d5 |
add.l %a4, %d5 |
asr.l #1, %d5 |
sub.l %d5, %a0 |
move.l %d2, %d5 | phase -> frac
lsl.l %d7, %d5 |
lsr.l #1, %d5 |
move.l %a0, %acc0 | %acc0 = frac * c3 + c2
mac.l %d5, %d6, %acc0 |
move.l %a5, %d6 | c1
sub.l %a3, %d6 |
asr.l #1, %d6 |
movclr.l %acc0, %a0 | %acc1 = frac * acc + c1
move.l %d6, %acc1 |
mac.l %d5, %a0, %acc1 |
move.l %d0, %a0 | %a0 = last_pos
add.l %d1, %d2 | phase += delta
move.l %d2, %d0 | pos = phase >> 16
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
lsr.l %d7, %d0 |
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
movclr.l %acc1, %d6 | %acc0 = frac * acc + x2
move.l %a4, %acc0 |
mac.l %d5, %d6, %acc0 |
cmp.l %d3, %d0 | %d0 = MIN(pos, srcrem)
ble.w .hrs_dsloop |
move.l %d3, %d0 |
bra.w .hrs_dsloop |
.hrs_channel_done: |
movem.l (%sp), %d5/%a0 | restore ch, h
movem.l %a3-%a5, (%a0) | h[0..2] = x3..x1
lea.l 12(%a0), %a5 | h++
movem.l 56(%sp), %a0-%a2 | load data, src, dst
subq.l #1, %d5 | ch > 0?
bgt.w .hrs_channel_loop | yes? process next channel
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
move.l 12(%a2), %d1 | %d1 = dst->bufcount
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
sub.l %d4, %d1 | written = dst->bufcount - dstrem
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
move.l %d1, (%a2) | dst->remcount = written
move.l %d0, %d1 | wrap phase to position in next frame
lsl.l %d7, %d1 | data->phase = phase - (pos << 16)
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
sub.l %d1, %d2 |
move.l %d2, 4(%a0) |
movem.l 8(%sp), %d2-%d7/%a2-%a6 | restore non-volatiles
lea.l 52(%sp), %sp | cleanup stack
Code police myself. Somehow got that full of tabs. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12401 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 03:29:37 +00:00
rts | buh-bye
SWCODEC: DSP optimizations for conversion to internal format and resampling. Assembly resampling for Coldfire. Word has it ARM will get that soon. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12399 a1c6a512-1295-4272-9138-f99709370657
2007-02-19 02:49:26 +00:00
Introduce new hermite polynomial resampler. Uses the Catmull-Rom case of Hermite cubic splines. Vastly improves the quality and accuracy of audio resampling with a rather minor additional overhead compared to the previous linear implementation. ARM and Coldfire assembly implementations included. Change-Id: Ic45d84bc66c5b312ef373198297a952167a4be26 Reviewed-on: http://gerrit.rockbox.org/304 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-05-07 07:12:56 +00:00
.size resample_hermite, .-resample_hermite
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void channel_mode_proc_mono(struct dsp_proc_entry *this,
* struct dsp_buffer **buf_p)
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
*
* Mix left and right channels 50/50 into a center channel.
*/
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.global channel_mode_proc_mono
channel_mode_proc_mono:
| input: 4(sp) = this, 8(sp) = buf_p
move.l 8(%sp), %a0 | %a0 = buf_p
move.l (%a0), %a0 | %a0 = buf = *buf_p
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
lea.l -20(%sp), %sp | save registers
movem.l %d2-%d4/%a2-%a3, (%sp) |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l (%a0), %d0/%a0-%a1 | %d0 = buf->remcount, %a0 = buf->p32[0],
| %a1 = buf->p32[1]
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
move.l %a0, %a2 | use separate dst pointers since read
move.l %a1, %a3 | pointers run one ahead of write
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l #0x40000000, %d3 | %d3 = 0.5
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
move.l (%a0)+, %d1 | prime the input registers
move.l (%a1)+, %d2 |
mac.l %d1, %d3, (%a0)+, %d1, %acc0 |
mac.l %d2, %d3, (%a1)+, %d2, %acc0 |
subq.l #1, %d0 |
ble.s 20f | loop done |
10: | loop |
movclr.l %acc0, %d4 | L = R = l/2 + r/2
mac.l %d1, %d3, (%a0)+, %d1, %acc0 |
mac.l %d2, %d3, (%a1)+, %d2, %acc0 |
move.l %d4, (%a2)+ | output to original buffer
move.l %d4, (%a3)+ |
subq.l #1, %d0 |
bgt.s 10b | loop |
20: | loop done |
movclr.l %acc0, %d4 | output last sample
move.l %d4, (%a2) |
move.l %d4, (%a3) |
movem.l (%sp), %d2-%d4/%a2-%a3 | restore registers
lea.l 20(%sp), %sp | cleanup
rts |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.size channel_mode_proc_mono, .-channel_mode_proc_mono
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void channel_mode_proc_custom(struct dsp_proc_entry *this,
* struct dsp_buffer **buf_p)
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
*
* Apply stereo width (narrowing/expanding) effect.
*/
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.global channel_mode_proc_custom
channel_mode_proc_custom:
| input: 4(sp) = this, 8(sp) = buf_p
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
lea.l -28(%sp), %sp | save registers
movem.l %d2-%d6/%a2-%a3, (%sp) |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l 32(%sp), %a0-%a1 | %a0 = this, %a1 = buf_p
move.l (%a1), %a1 | %a1 = buf = *buf_p
move.l (%a0), %a2 | %a2 = this->data = &channel_mode_data
movem.l (%a1), %d0/%a0-%a1 | %d0 = buf->remcount, %a0 = buf->p32[0],
| %a1 = buf->p32[1]
movem.l (%a2), %d3-%d4 | %d3 = sw_gain, %d4 = sw_cross
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
move.l %a0, %a2 | use separate dst pointers since read
move.l %a1, %a3 | pointers run one ahead of write
move.l (%a0)+, %d1 | prime the input registers
move.l (%a1)+, %d2 |
mac.l %d1, %d3 , %acc0 | L = l*gain + r*cross
mac.l %d1, %d4, (%a0)+, %d1, %acc1 | R = r*gain + l*cross
mac.l %d2, %d4 , %acc0 |
mac.l %d2, %d3, (%a1)+, %d2, %acc1 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
subq.l #1, %d0 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
ble.b 20f | loop done |
10: | loop |
movclr.l %acc0, %d5 |
movclr.l %acc1, %d6 |
mac.l %d1, %d3 , %acc0 | L = l*gain + r*cross
mac.l %d1, %d4, (%a0)+, %d1, %acc1 | R = r*gain + l*cross
mac.l %d2, %d4 , %acc0 |
mac.l %d2, %d3, (%a1)+, %d2, %acc1 |
move.l %d5, (%a2)+ |
move.l %d6, (%a3)+ |
subq.l #1, %d0 |
bgt.s 10b | loop |
20: | loop done |
movclr.l %acc0, %d5 | output last sample
movclr.l %acc1, %d6 |
move.l %d5, (%a2) |
move.l %d6, (%a3) |
movem.l (%sp), %d2-%d6/%a2-%a3 | restore registers
lea.l 28(%sp), %sp | cleanup
rts |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.size channel_mode_proc_custom, .-channel_mode_proc_custom
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void channel_mode_proc_karaoke(struct dsp_proc_entry *this,
* struct dsp_buffer **buf_p)
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
*
* Separate channels into side channels.
*/
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.global channel_mode_proc_karaoke
channel_mode_proc_karaoke:
| input: 4(sp) = this, 8(sp) = buf_p
move.l 8(%sp), %a0 | %a0 = buf_p
move.l (%a0), %a0 | %a0 = buf = *buf_p
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
lea.l -20(%sp), %sp | save registers
movem.l %d2-%d4/%a2-%a3, (%sp) |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l (%a0), %d0/%a0-%a1 | %d0 = buf->remcount, %a0 = buf->p32[0],
| %a1 = buf->p32[1]
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
move.l %a0, %a2 | use separate dst pointers since read
move.l %a1, %a3 | pointers run one ahead of write
move.l #0x40000000, %d3 | %d3 = 0.5
move.l (%a0)+, %d1 | prime the input registers
move.l (%a1)+, %d2 |
mac.l %d1, %d3, (%a0)+, %d1, %acc0 | L = l/2 - r/2
msac.l %d2, %d3, (%a1)+, %d2, %acc0 |
subq.l #1, %d0 |
ble.b 20f | loop done |
10: | loop |
movclr.l %acc0, %d4 |
mac.l %d1, %d3, (%a0)+, %d1, %acc0 | L = l/2 - r/2
msac.l %d2, %d3, (%a1)+, %d2, %acc0 |
move.l %d4, (%a2)+ |
neg.l %d4 | R = -L = -(l/2 - r/2) = r/2 - l/2
move.l %d4, (%a3)+ |
subq.l #1, %d0 |
bgt.s 10b | loop |
20: | loop done |
movclr.l %acc0, %d4 | output last sample
move.l %d4, (%a2) |
neg.l %d4 | R = -L = -(l/2 - r/2) = r/2 - l/2
move.l %d4, (%a3) |
movem.l (%sp), %d2-%d4/%a2-%a3 | restore registers
lea.l 20(%sp), %sp | cleanup
rts |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
.size channel_mode_proc_karaoke, .-channel_mode_proc_karaoke
/****************************************************************************
* void filter_process(struct dsp_filter *f, int32_t *buf[], int count,
* unsigned int channels)
*
* define HIGH_PRECISION as '1' to make filtering calculate lower bits after
* shifting. without this, "shift" - 1 of the lower bits will be lost here.
*/
#define HIGH_PRECISION 0
.text
.global filter_process
filter_process:
| input: 4(sp) = f, 8(sp) = buf, 12(sp) = count, 16(sp) = channels
lea.l -44(%sp), %sp | save clobbered regs
#if HIGH_PRECISION
movem.l %d2-%d7/%a2-%a6, (%sp) | .
#else
movem.l %d2-%d6/%a2-%a6, (%sp) |
#endif
move.l 48(%sp), %a5 | fetch filter structure address
clr.l %d6 | load shift count
move.b 52(%a5), %d6 | .
subq.l #1, %d6 | EMAC gives us one free shift
#if HIGH_PRECISION
moveq.l #8, %d7
sub.l %d6, %d7 | shift for lower part of accumulator
#endif
movem.l (%a5), %a0-%a4 | load coefs
lea.l 20(%a5), %a5 | point to filter history
10: | channel loop
move.l 52(%sp), %a6 | load input channel pointer
addq.l #4, 52(%sp) | point x to next channel
move.l (%a6), %a6 |
move.l 56(%sp), %d5 | number of samples
movem.l (%a5), %d0-%d3 | load filter history
| d0-d3 = history, d4 = temp, d5 = sample count, d6 = upper shift amount,
| d7 = lower shift amount,a0-a4 = coefs, a5 = history pointer, a6 = buf[ch]
20: | loop
| Direct form 1 filtering code. We assume DSP has put EMAC in frac mode.
| y[n] = b0*x[i] + b1*x[i - 1] + b2*x[i - 2] + a1*y[i - 1] + a2*y[i - 2],
| where y[] is output and x[] is input. This is performed out of order
| to do parallel load of input value.
mac.l %a2, %d1, %acc0 | acc = b2*x[i - 2]
move.l %d0, %d1 | fix input history
mac.l %a1, %d0, (%a6), %d0, %acc0 | acc += b1*x[i - 1], x[i] -> d0
mac.l %a0, %d0, %acc0 | acc += b0*x[i]
mac.l %a3, %d2, %acc0 | acc += a1*y[i - 1]
mac.l %a4, %d3, %acc0 | acc += a2*y[i - 2]
move.l %d2, %d3 | fix output history
#if HIGH_PRECISION
move.l %accext01, %d2 | fetch lower part of accumulator
move.b %d2, %d4 | clear upper three bytes
lsr.l %d7, %d4 | shift lower bits
#endif
movclr.l %acc0, %d2 | fetch upper part of result
asl.l %d6, %d2 | restore fixed point format
#if HIGH_PRECISION
or.l %d2, %d4 | combine lower and upper parts
#endif
move.l %d2, (%a6)+ | save result
subq.l #1, %d5 | are we done with this channel?
bgt 20b | loop
movem.l %d0-%d3, (%a5) | save history back to struct
lea.l 16(%a5), %a5 | point to next channel's history
subq.l #1, 60(%sp) | have we processed both channels?
bhi 10b | channel loop
#if HIGH_PRECISION
movem.l (%sp), %d2-%d7/%a2-%a6
#else
movem.l (%sp), %d2-%d6/%a2-%a6
#endif
lea.l 44(%sp), %sp
rts
.size filter_process, .-filter_process
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void sample_output_stereo(struct sample_io_data *this,
* struct dsp_buffer *src,
* struct dsp_buffer *dst)
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
*
* Framework based on the ubiquitous Rockbox line transfer logic for
* Coldfire CPUs.
*
* Does emac clamping and scaling (which proved faster than the usual
* checks and branches - even single test clamping) and writes using
* line burst transfers. Also better than writing a single L-R pair per
* loop but a good deal more code.
*
* Attemping bursting during reads is rather futile since the source and
* destination alignments rarely agree and too much complication will
* slow us up. The parallel loads seem to do a bit better at least until
* a pcm buffer can always give line aligned chunk and then aligning the
* dest can then imply the source is aligned if the source buffers are.
* For now longword alignment is assumed of both the source and dest.
*
*/
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
.global sample_output_stereo
sample_output_stereo:
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
| input: 4(sp) = count, 8(sp) = src, 12(sp) = dst
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
lea.l -48(%sp), %sp | save registers
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l %macsr, %d1 | do it now as at many lines will
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
movem.l %d1-%d7/%a2-%a6, (%sp) | be the far more common condition
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l #0x80, %macsr | put emac unit in signed int mode
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l 52(%sp), %a0-%a2 | %a0 = this, %a1 = src, %a2 = dst
move.l (%a0), %a0 | %a0 = this->outcount
move.l 4(%a2), %a4 | %a4 = dst->p16out
lea.l (%a4, %a0.l*4), %a0 | %a0 = count -> end address
movem.l 4(%a1), %a2-%a3 | %a2 = src->p32[0], %a3 = src->p32[1]
clr.l %d1 | %a1 = multiplier: (1 << (16 - scale))
move.b 19(%a1), %d1 | %d1 = src->format.output_scale
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
sub.l #16, %d1 |
neg.l %d1 |
Hrm. move.q? I want moveq.l :) git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12474 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:34:18 +00:00
moveq.l #1, %d0 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
asl.l %d1, %d0 |
move.l %d0, %a1 |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l #0x8000, %a6 | %a6 = rounding term
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
moveq.l #28, %d0 | %d0 = second line bound
add.l %a4, %d0 |
and.l #0xfffffff0, %d0 |
Coldfire DSP Assembly: Fix a bug in the resampler that would only affect one frame on the transition from up to down. Fix a bug in sample output that could be the cause of reported FLAC glitching. Output routines now pass all alignment and size combination tests. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12733 a1c6a512-1295-4272-9138-f99709370657
2007-03-12 08:36:42 +00:00
cmp.l %a0, %d0 | at least a full line?
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.w 40f | long loop 1 start | no? do as trailing longwords
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
sub.l #16, %d0 | %d1 = first line bound
cmp.l %a4, %d0 | any leading longwords?
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bls.b 20f | line loop start | no? start line loop
10: | long loop 0 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l (%a2)+, %d1 | read longword from L and R
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a6, %acc0 |
move.l %acc0, %acc1 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
mac.l %d1, %a1, (%a3)+, %d2, %acc0 | shift L to high word
mac.l %d2, %a1, %acc1 | shift R to high word
movclr.l %acc0, %d1 | get possibly saturated results
movclr.l %acc1, %d2 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d2 | move R to low word
Some whitespace correction in dsp_*.S Change-Id: I6ee14c0adc9dd456c8a2d171952cbaedb3752428
2012-04-27 20:55:16 +00:00
move.w %d2, %d1 | interleave MS 16 bits of each
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l %d1, (%a4)+ | ...and write both
cmp.l %a4, %d0 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.b 10b | long loop 0 |
20: | line loop start |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
lea.l -12(%a0), %a5 | %a5 = at or just before last line bound
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
30: | line loop |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l (%a3)+, %d4 | get next 4 R samples and scale
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a6, %acc0 |
move.l %acc0, %acc1 |
move.l %acc1, %acc2 |
move.l %acc2, %acc3 |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
mac.l %d4, %a1, (%a3)+, %d5, %acc0 | with saturation
mac.l %d5, %a1, (%a3)+, %d6, %acc1 |
mac.l %d6, %a1, (%a3)+, %d7, %acc2 |
mac.l %d7, %a1, (%a2)+, %d0, %acc3 |
lea.l 16(%a4), %a4 | increment dest here, mitigate stalls
movclr.l %acc0, %d4 | obtain R results
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
movclr.l %acc1, %d5 |
movclr.l %acc2, %d6 |
movclr.l %acc3, %d7 |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a6, %acc0 |
move.l %acc0, %acc1 |
move.l %acc1, %acc2 |
move.l %acc2, %acc3 |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
mac.l %d0, %a1, (%a2)+, %d1, %acc0 | get next 4 L samples and scale
mac.l %d1, %a1, (%a2)+, %d2, %acc1 | with saturation
mac.l %d2, %a1, (%a2)+, %d3, %acc2 |
mac.l %d3, %a1 , %acc3 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d4 | a) interleave most significant...
swap.w %d5 |
swap.w %d6 |
swap.w %d7 |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movclr.l %acc0, %d0 | obtain L results
movclr.l %acc1, %d1 |
movclr.l %acc2, %d2 |
movclr.l %acc3, %d3 |
move.w %d4, %d0 | a) ... 16 bits of L and R
move.w %d5, %d1 |
move.w %d6, %d2 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d7, %d3 |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movem.l %d0-%d3, -16(%a4) | write four stereo samples
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
cmp.l %a4, %a5 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.b 30b | line loop |
40: | long loop 1 start |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
cmp.l %a4, %a0 | any longwords left?
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bls.b 60f | output end | no? stop
50: | long loop 1 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l (%a2)+, %d1 | handle trailing longwords
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a6, %acc0 |
move.l %acc0, %acc1 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
mac.l %d1, %a1, (%a3)+, %d2, %acc0 | the same way as leading ones
mac.l %d2, %a1, %acc1 |
movclr.l %acc0, %d1 |
movclr.l %acc1, %d2 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d2 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d2, %d1 |
move.l %d1, (%a4)+ |
cmp.l %a4, %a0 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.b 50b | long loop 1
60: | output end |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
movem.l (%sp), %d1-%d7/%a2-%a6 | restore registers
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l %d1, %macsr |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
lea.l 48(%sp), %sp | cleanup
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
rts |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
.size sample_output_stereo, .-sample_output_stereo
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
/****************************************************************************
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
* void sample_output_mono(struct sample_io_data *this,
* struct dsp_buffer *src,
* struct dsp_buffer *dst)
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
*
* Same treatment as sample_output_stereo but for one channel.
*/
.section .text
Argh. Get the tabs out the asm again. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12923 a1c6a512-1295-4272-9138-f99709370657
2007-03-26 15:20:44 +00:00
.align 2
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
.global sample_output_mono
sample_output_mono:
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
| input: 4(sp) = count, 8(sp) = src, 12(sp) = dst
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
lea.l -32(%sp), %sp | save registers
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l %macsr, %d1 | do it now as at many lines will
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
movem.l %d1-%d5/%a2-%a4, (%sp) | be the far more common condition
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l #0x80, %macsr | put emac unit in signed int mode
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
movem.l 36(%sp), %a0-%a2 | %a0 = this, %a1 = src, %a2 = dst
move.l (%a0), %a0 | %a0 = this->outcount
move.l 4(%a2), %a3 | %a3 = dst->p16out
movem.l 4(%a1), %a2 | %a2 = src->p32[0]
lea.l (%a3, %a0.l*4), %a0 | %a0 = count -> end address
clr.l %d1 | %d5 = multiplier: (1 << (16 - scale))
move.b 19(%a1), %d1 | %d1 = src->format.output_scale
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
sub.l #16, %d1 |
neg.l %d1 |
Hrm. move.q? I want moveq.l :) git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12474 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:34:18 +00:00
moveq.l #1, %d5 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
asl.l %d1, %d5 |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l #0x8000, %a4 | %a4 = rounding term
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
moveq.l #28, %d0 | %d0 = second line bound
add.l %a3, %d0 |
and.l #0xfffffff0, %d0 |
Coldfire DSP Assembly: Fix a bug in the resampler that would only affect one frame on the transition from up to down. Fix a bug in sample output that could be the cause of reported FLAC glitching. Output routines now pass all alignment and size combination tests. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12733 a1c6a512-1295-4272-9138-f99709370657
2007-03-12 08:36:42 +00:00
cmp.l %a0, %d0 | at least a full line?
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.w 40f | long loop 1 start | no? do as trailing longwords
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
sub.l #16, %d0 | %d1 = first line bound
cmp.l %a3, %d0 | any leading longwords?
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bls.b 20f | line loop start | no? start line loop
10: | long loop 0 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l (%a2)+, %d1 | read longword from L and R
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a4, %acc0 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
mac.l %d1, %d5, %acc0 | shift L to high word
movclr.l %acc0, %d1 | get possibly saturated results
move.l %d1, %d2 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d2 | move R to low word
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d2, %d1 | duplicate single channel into
move.l %d1, (%a3)+ | L and R
cmp.l %a3, %d0 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.b 10b | long loop 0 |
20: | line loop start |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
lea.l -12(%a0), %a1 | %a1 = at or just before last line bound
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
30: | line loop |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l (%a2)+, %d0 | get next 4 L samples and scale
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a4, %acc0 |
move.l %acc0, %acc1 |
move.l %acc1, %acc2 |
move.l %acc2, %acc3 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
mac.l %d0, %d5, (%a2)+, %d1, %acc0 | with saturation
mac.l %d1, %d5, (%a2)+, %d2, %acc1 |
mac.l %d2, %d5, (%a2)+, %d3, %acc2 |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
mac.l %d3, %d5 , %acc3 |
lea.l 16(%a3), %a3 | increment dest here, mitigate stalls
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
movclr.l %acc0, %d0 | obtain results
movclr.l %acc1, %d1 |
movclr.l %acc2, %d2 |
movclr.l %acc3, %d3 |
move.l %d0, %d4 | duplicate single channel
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d4 | into L and R
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d4, %d0 |
move.l %d1, %d4 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d4 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d4, %d1 |
move.l %d2, %d4 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d4 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d4, %d2 |
move.l %d3, %d4 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d4 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d4, %d3 |
SWCODEC: Tighten up coldfire assembly a little bit more. Cleanup to make differing parameters between ARM and Coldfire halfway clean. Hopefully those differences can be reconciled soon. A tiny bit of C optimizing for karaoke channel mode. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12505 a1c6a512-1295-4272-9138-f99709370657
2007-02-27 14:25:36 +00:00
movem.l %d0-%d3, -16(%a3) | write four stereo samples
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
cmp.l %a3, %a1 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.b 30b | line loop |
40: | long loop 1 start |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
cmp.l %a3, %a0 | any longwords left?
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bls.b 60f | output end | no? stop
50: | loop loop 1 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l (%a2)+, %d1 | handle trailing longwords
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
move.l %a4, %acc0 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
mac.l %d1, %d5, %acc0 | the same way as leading ones
movclr.l %acc0, %d1 |
move.l %d1, %d2 |
Fundamentally rewrite much of the audio DSP. Creates a standard buffer passing, local data passing and messaging system for processing stages. Stages can be moved to their own source files to reduce clutter and ease assimilation of new ones. dsp.c becomes dsp_core.c which supports an engine and framework for effects. Formats and change notifications are passed along with the buffer so that they arrive at the correct time at each stage in the chain regardless of the internal delays of a particular one. Removes restrictions on the number of samples that can be processed at a time and it pays attention to destination buffer size restrictions without having to limit input count, which also allows pcmbuf to remain fuller and safely set its own buffer limits as it sees fit. There is no longer a need to query input/output counts given a certain number of input samples; just give it the sizes of the source and destination buffers. Works in harmony with stages that are not deterministic in terms of sample input/output ratio (like both resamplers but most notably the timestretch). As a result it fixes quirks with timestretch hanging up with certain settings and it now operates properly throughout its full settings range. Change-Id: Ib206ec78f6f6c79259c5af9009fe021d68be9734 Reviewed-on: http://gerrit.rockbox.org/200 Reviewed-by: Michael Sevakis <jethead71@rockbox.org> Tested-by: Michael Sevakis <jethead71@rockbox.org>
2012-03-27 23:52:15 +00:00
swap.w %d2 |
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.w %d2, %d1 |
move.l %d1, (%a3)+ |
cmp.l %a3, %a0 |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
bhi.b 50b | long loop 1 |
60: | output end |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
movem.l (%sp), %d1-%d5/%a2-%a4 | restore registers
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
move.l %d1, %macsr |
FS #7286. Do correct rounding of final 16 bit samples before sending to DAC, for you golden-eared people. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14514 a1c6a512-1295-4272-9138-f99709370657
2007-08-29 14:32:52 +00:00
lea.l 32(%sp), %sp | cleanup
SWCODEC: Dsp speed optimizations. Changes for more modularity. Removal of some usless stuff. Some assembly routines for Coldfire with speed in mind over size for the outputs but the channel modes remain compact. Miscellaneous coldfire asm updates to accomodate the changes. Codec API structure version has to increase so do a full update. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12472 a1c6a512-1295-4272-9138-f99709370657
2007-02-24 17:06:36 +00:00
rts |
SWCODEC & Coldfire: Do some more DSP straigntening out. Do as much Coldfire optimizing as seems reasonably possible by jumping through some hoops to avoid stalls. Further boost reduction will just be fractional points if taken to extremes-- not worth it. Wrap up the ASM for awhile. git-svn-id: svn://svn.rockbox.org/rockbox/trunk@12905 a1c6a512-1295-4272-9138-f99709370657
2007-03-25 04:03:44 +00:00
.size sample_output_mono, .-sample_output_mono
Reference in a new issue Copy permalink