3 commits

Author	SHA1	Message	Date
Dave Chapman	f52c178517	Revert r14050 (apart from the non-related whitespace changes) and fix the S3C2440 (Gigabeat F/X) problems introduced by my r14041 commit in a slightly different way: Introduce a mad_iram.h file and centralise all IRAM-related decisions in there (similar to how Tremor does it). Now checks for CPU type (S3C2440) instead of target (TOSHIBA_GIGABEAT_F) and there is also a .ibss variable in syth_full_arm.S which shouldn't go in IRAM on S3C2440 targets. Closes FS#7500 ... git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14057 a1c6a512-1295-4272-9138-f99709370657	2007-07-29 16:40:11 +00:00
Michael Giacomelli	f68ab0d497	Fix mp3 playback on the gigabeat by not putting code in IRAM. Someone actually familar with the gigabeat port should check this. ... git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14050 a1c6a512-1295-4272-9138-f99709370657	2007-07-29 05:21:20 +00:00
Dave Chapman	66b51909c0	FS #6705 - ARM optimisations for libmad by Tomasz Malesinski. Modified slightly by me to not put code in IRAM for PP502x (it's slower), and for the mpegplayer version of libmad for PP5002 (there isn't enough room). On my ipod Color, it increases a 320kbps MP3 test file from 169% realtime to 188% realtime. Reported speedup on the ipod 3G was from 118% to 155% realtime for a 192kbps MP3. ... git-svn-id: svn://svn.rockbox.org/rockbox/trunk@14041 a1c6a512-1295-4272-9138-f99709370657	2007-07-28 15:21:25 +00:00