Hi, For those of you (maybe 3 :)) who care, I have done a fair bit of performance work on bmovl2 over the past couple of days. I rewrote large sections of code and the performance gains are pretty significant.
For non-MMX systems, these performance improvements yield a 70% speed
increase. On my Athlon 1800 XP, blending a 800x600 image with alpha
took about 18500 usec. Now it takes 6000 usec.
MMX got an improvement too, although not as much: for MMX-capable
systems blending is now 40% faster. With MMX, that 800x600 image took
about 4700 usec, and now takes 2800 usec. This translates to about 5-7%
percent total CPU usage on my system.
You can download it at the usual place:
http://sault.org/mebox/downloads.php
Since a lot of code was rewritten, I may have introduced some bugs.
Please let me know if you find any.
Now Rob planted the idea in my head about a mevas canvas for ivtvosd, so
I at least have to try. :)
Jason.
signature.asc
Description: This is a digitally signed message part
