At 02:22 PM 6/28/99 +1000, Micke Cheng wrote:
...
>3.2 LAME
>
>For the LAME project [7] the MDCT was unrolled by hand (by
>[EMAIL PROTECTED]).
>By exploiting the symmetry of the trig factors involved in the calculation,
>successive multiplication operations by the same factor were replaced
>with additions. This bought the computation count down to 244 multiplies
>and 324 additions and cut the MDCT() time by about 70%.
>
I'm not sure, haven't checked this much, but when I looked at the lame-mdct
code last summer, I found your MDCT-code much like the "Lee's Fast DCT
algorithm" I first found on
http://pentium.intel.com/drg/mmx/appnotes/ap528.htm
Just wanted to mention this, because someone might want to take a look at
MMX-optimizing the MDCT, and that link has some code. (For decoding though..)
I also found this MDCT-link mentioning lee's algorithm too:
http://mpeg.openprojects.net/mpeg2/src/mpeg2dec/AP528.txt
...
/Jan Peman (not good at MDCT:s at all..)
____________________________________________________________________________
Jan Peman, Fyrplus Mekatronik AB Tel: +46 (0)54-175369 (Work)
Kvarnbergsgatan 38, 652 29 Karlstad +46 (0)54-150117 (Fax)
MailTo:[EMAIL PROTECTED] +46 (0)708-380629 (Cellular)
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
