On Feb 28, 2004, at 11:45 AM, Rafael Ávila de Espíndola wrote:

Is there any reason for not using the Szu-Wei Lee's algorithm for the short IMDCT (size 12)?

Annexed is a patch witch implements it.

Thanks for this patch.

It's not clear there's a real benefit to using Szu-Wei Lee's algorithm for short blocks. It only saves one multiplication per window, and adds additional complexity. I wasn't able to observe any performance improvement.

If you found otherwise, I'd be interested in your analysis.

Cheers,