Find a file
Michael Giacomelli d514ee9282 opus: ARMv4 assembly for the backward MDCT inner loops
Cuts realtime decode on the Sansa e200v1 from 52.1 MHz to 50.8 MHz.
clt_mdct_backward was the largest remaining item at 13.5% of decode.

Only the three inner loops move to assembly.  The setup stays in C, so
mdct.c remains readable and the assembly needs no knowledge of
mdct_lookup.

What the compiled loops lose is registers.  Each needs more live values
than gcc can hold, so it spills the loop-invariant pointers, strides and
limits and reloads them every pass: five stack accesses per iteration in
the post-rotation alone.  Holding the twiddle as a 16-bit value and
accumulating the product pair with smull/smlal is what makes the
bookkeeping fit, needing seven live registers where the shifted
MULT16_32_Q15 form needs nine.

ldm/stm helps only where the addressing allows.  The post-rotation walks
the buffer from both ends and so reads and writes contiguous pairs.  The
pre-rotation reads the spectrum through a runtime stride and writes
through the bitrev table, so only its 8-byte output pair merges, and the
TDAC mirror merges nothing.

Over 160 ms of stereo music, traced under qemu:

  clt_mdct_backward  1,037,962 ->   900,982   -13.2%
  whole decode       7,695,876 -> 7,558,896    -1.8%
  loads                650,157 ->   611,667    -5.9%
  stores               350,605 ->   323,605    -7.7%
  multiplies           337,493 ->   337,493   unchanged

Accuracy improves substantially, because all three loops keep 32 bits of
each Q15 product where MULT16_32_Q15_armv4 drops the low bit, and the
backward MDCT applies three such rounds per sample.  The rounding SNR of
the backward transform rises about 9.5 dB, and its worst case error falls
from 708 to 186.  Decoded output differs from the previous build in 90 of
15,360 samples, each by one LSB.

Build with OPUS_ARM_NO_MDCT_ASM to select the C loops instead.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Change-Id: I3c4404b4dbe581d8bcf1f357266a658a068fdb50
2026-09-22 07:26:56 -04:00
android chore: Get rid of *all* vestigal CVS '$Id:$' tags 2026-06-01 16:01:18 -04:00
apps ipodnano3g: power, RTC, backlight, battery and audio 2026-09-20 08:53:53 -04:00
backdrops ap80max: New hosted port 2026-08-13 17:55:06 -07:00
bootloader ipodnano3g: NAND check image for validating other chips 2026-09-19 21:26:45 -04:00
docs ipod6g: add hardware H.264 video playback 2026-09-13 08:07:59 -04:00
firmware ipodnano3g: mark the 4-CE B614D5EC row checked 2026-09-21 09:16:33 -04:00
fonts chore: Get rid of *all* vestigal CVS '$Id:$' tags 2026-06-01 16:01:18 -04:00
icons chore: Get rid of *all* vestigal CVS '$Id:$' tags 2026-06-01 16:01:18 -04:00
lib opus: ARMv4 assembly for the backward MDCT inner loops 2026-09-22 07:26:56 -04:00
manual manual: add the iPod Nano 3G 2026-09-20 08:58:09 -04:00
packaging 3ds: Port refactor. 2026-09-01 08:39:07 -04:00
tools build: move '-mthumb-interwork' into the configure script 2026-09-19 21:18:30 -04:00
uisimulator Add "rbfs" prefix to native filesystem functions 2026-09-17 16:22:31 -04:00
utils ipodnano3g: mark the 4-CE B614D5EC row checked 2026-09-21 09:16:33 -04:00
wps ap80max: New hosted port 2026-08-13 17:55:06 -07:00
.gitattributes Add a gitattributes file for the migration. 2011-12-01 14:14:59 +00:00
.gitignore misc: Update gitignore to cover a bunch of utils/* binaries 2026-06-17 13:56:29 -04:00
.gitreview misc: Add a .gitreview file for better gerrit integration 2021-06-24 07:22:19 -04:00

               __________               __   ___.
     Open      \______   \ ____   ____ |  | _\_ |__   _______  ___
     Source     |       _//  _ \_/ ___\|  |/ /| __ \ /  _ \  \/  /
     Jukebox    |    |   (  <_> )  \___|    < | \_\ (  <_> > <  <
     Firmware   |____|_  /\____/ \___  >__|_ \|___  /\____/__/\_ \
                       \/            \/     \/    \/            \/

Build Your Own Rockbox

1. Clone 'rockbox' from git (or extract a downloaded archive).

   $ git clone git://git.rockbox.org/rockbox

     or

   $ tar xJf rockbox.tar.xz

2. Create a build directory, preferably in the same directory as the firmware/
   and apps/ directories. This is where all generated files will be written.

   $ cd rockbox
   $ mkdir build
   $ cd build

3. Make sure you have mips/m68k/arm-elf-gcc and siblings in the PATH. Make sure
   that you have 'perl' in your PATH too. Your gcc cross compiler needs to be
   a particular version depending on what player you are compiling for. These
   can be generated using the rockboxdev.sh script in the /tools/ folder of the
   source.

   $ which arm-elf-eabi-gcc
   $ which perl

4. In your build directory, run the 'tools/configure' script and enter what
   target you want to build for and if you want a debug version or not (and a
   few more questions). It'll prompt you. The debug version is for making a
   gdb version out of it. It is only useful if you run gdb towards your target
   Archos.

   $ ../tools/configure

5. *ploink*. Now you have got a Makefile generated for you.

6. Run 'make' and soon the necessary pieces from the firmware and the apps
   directories have been compiled, linked and scrambled for you.

   $ make
   $ make zip

7. unzip the rockbox.zip on your music player, reboot it and
   *smile*.

If you want to build for more than one target, just create several build
directories and create a setup for each target:

   $ mkdir build-fuzeplus
   $ cd build-fuzeplus
   $ ../tools/configure

   $ mkdir build-xduoox3
   $ cd build-xduoox3
   $ ../tools/configure

Questions anyone? Ask on the mailing list or on IRC. We'll be happy to help you!