[MPlayer-dev-eng] MPlayer licensing [and lotso other stuff as I'm typing...]

[Michael Niedermayer]

Hi

On Saturday 24 November 2001 18:26, pl wrote:
[...]
>
> We should also add around MMX/athlon/... asm optimization a :
>    if (gCpuCape.hasMMX) {...asm()...}
>
> I don't know to what extent it will cause a decrease in performance but
> I guess it's acceptable for current systems.  Michael, Nick ?
its not that easy because
1. lotsa macros inside asm blocks like PAVGB (for 3dnow -> pavgusb for mmx2 
-> pavgb)
2. these ifs would need to be called once per 8x8 block and that would mean a 
~1% performace loss perhaps

better would be, as arpi said some time ago to include the postprocessing 
functions several times with different #defines ... i wanted to do that but 
didnt do it yet, but i will do it :)
this would have no messureable performace effect as postProcess() is only 
called 90 times per secound an 30fps video

>
> Yet, if it's not, then there'll remain the standard current mecanism since
> we'll '#define gCpuCape.hasMMX  1' (and not as a variable). This way,
> gcc won't generate code for this always-true test (no performance loss).
>
> Not ideal but it looks like a step towards better runtime autodetection.

Michael