Use the windowing approach from ffmpeg in tremor, does the mdct doubling, windowing and overlap add in one go.
Also uses less memory so all the processing buffers fit in iram on targets with small iram for the common blocksizes (256/2048) now.
Speeds up decoding of vorbis files by 3MHz for 256/2048 and 20MHz for 512/4096 files on h300.
Speeds up decoding of vorbis files by 3MHz for 256/2048 and 4.5MHz for 512/4096 on the beast.
Speeds up decoding of vorbis files by 0.3MHz for 256/2048 and 1MHz for 512/4096 on c200v1.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@30580 a1c6a512-1295-4272-9138-f99709370657
Such targets would previously default to using the configuration for targets with small iram which uses an extra memcpy per block.
This saves 2MHz decoding a 128kbps vorbis file on the Gigabeat S and saves a bit of codec buffer.
Patch from FS#11268, also replaces patch from FS#12147.
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@29976 a1c6a512-1295-4272-9138-f99709370657
