[sldev] [patch] Vectorized DWT in OpenJPEG

[John Hurliman]

Dzonatas wrote:
> Hello,
>
> Attached is the patch to move the DWT into vectorized code. It works 
> with floating point, so I only enabled it for the 9-7 decode.
>
> Speed wise, it is just slightly slower than KDU. One my reference file 
> "bird.jp2," KDU decodes in 2 seconds DWT time, and this patch will 
> make OpenJPEG decode the same in under 4 seconds in DWT time.
>
> Actual speed-up will vary overall, but it ranges from 2x to 20x in 
> just DWT time.
>
> Enjoy. =)

Forgive my ignorance, but will this patch have any effect (negative or 
positive) on compilers that aren't GCC4? For example if I apply this and 
compile with Visual Studio 2005 should I expect the code to run at the 
same speed as previously, or will it have an impact? And what about 
Intel's compiler?

John Hurliman