[MPlayer-dev-eng] Bugfix (sig11 if sdl, rgb and resize)

[Felix Buenemann]

On Wednesday, 26. September 2001 19:59, Michael Niedermayer wrote:
> On Wednesday 26 September 2001 13:06, Felix Buenemann wrote:
> > On Tuesday, 25. September 2001 22:16, Michael Niedermayer wrote:
> > > > ok, it'll be changed in cvs with the next commiting of local changes
> > > > (have other unfinished changes...).
> > >
> > > the priv->dblit=0 variant?
> > > it does an extra copy pass, and is slower (x2)!
> > > the second percent number (video_out cpu usage according to the faq)
> > > increased from 4-5% to 8% on my test movie.
> >
> > yes, but I think it's endian independent then.
>
> more bugs ...
> 1. video_out cpu usage doesnt count the time spent in flip_page
> 2. rgbsurface isnt aligned correctly
yea, but the problem is inside SDL where I can't change it, cause the malloc 
is done by SDL_CreateRGBSurface, otherwise I could speed it up using 
memalign() for malloc.
>
> i did a few more speed tests ...
> memcpy to priv->surface needs 4-5%
> SDL_CreateRGBSurfaceFrom & SDL_BlitSurface 8%
> memcpy to priv->rgbsurface needs 8%
> memcpy to priv->rgbsurface+24 needs 4% (neeeds to be aligned at 32byte?)
> SDL_BlitSurface from rgbsurface to surface needs 12%
yes I know that it's really slow, that's why I added the direct blitting 
stuff which is enabled with dblit=1.
So I should take a look at your patch again.
Btw. if you remove SDL_ANYFORMAT flag it'll be veery slow :)

>
> the best solution IMHO for rgb would be to do a memcpy if possible
> (completly identical formats) and if thats not possible than
> SDL_CreateRGBSurfaceFrom & SDL_BlitSurface is next best, and could be used
> for rgb format conversation too.
>
yes.
> Michael

-- 
Best Regards,
	Atmos

_____
RTFM
	- Gabucino