[FFmpeg-devel] [PATCH/RFC] ARM9 optimization for ALT_BITSTREAMREADER_LE

[Siarhei Siamashka]

Hello,

The following patch adds ARM9 optimizations to the ALT_BITSTREAMREADER_LE.
It results in smaller code size and better performance on ARM9 over the
default code generated by the compiler. Also higher MIN_CACHE_BITS value
may be potentially taken advantage of later.

But it should be probably modified to be only enabled when FFmpeg is
configured for --cpu=arm9*

A similar patch can be done with instructions scheduling favourable for XScale
if anybody is still interested in improving performance for this
microarchitecture.

Benchmarked on arm926ej-s with wavpack decoding: 
./ffmpeg -benchmark -i $1 -f null /dev/null

-- before patch --
bench: utime=20.828s
bench: utime=20.828s
bench: utime=20.859s
bench: utime=20.805s
bench: utime=20.820s
-- after patch --
bench: utime=20.672s
bench: utime=20.672s
bench: utime=20.688s
bench: utime=20.672s
bench: utime=20.602s

Regression tests pass when run natively on ARM11 (with the default bitstream
reader set to ALT_BITSTREAM_READER).

-- 
Best regards,
Siarhei Siamashka
-------------- next part --------------
A non-text attachment was scrubbed...
Name: alt_bitstreamreader_le_arm9.diff
Type: text/x-diff
Size: 1154 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20081208/ba6f3e00/attachment.diff>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part.
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20081208/ba6f3e00/attachment.pgp>