Re: [AD] Allegro x86 clear and blit optimizations - update

[tom st denis <tomstdenis@xxxxxxxxxx>]

--- Esaelon@xxxxxxxxxx wrote:
> The main platform the clears and blitters were optimized for was
> PPlain/PMMX. 
>  For PPro+, there's not a lot that can be done for the masked blits 
> (expensive branch mispredictions). Virtually no changes were done to
> the 
> 16-bit code (only a few slight changes to the blit loop in blit.inc)
> - 
> iblit16.s was not changed, 16-bit clears were not changed at all
> (though it's 
> hard to tell from that -3%???). I considered working on the 16-bit
> code, but 
> was waiting for a reply on the alignment issue in the clears.

I was wondering, what about special clear routines that assume you are
clearing something aligned already and it's a multiple of 8 bytes? 
That way you can setup an MMX register and do repeated 8byte stores :-)
 In 8bpp modes you can clear 8 pixels at once, etc... no need to do
align code...

Tom

__________________________________________________
Do You Yahoo!?
Yahoo! Auctions - Buy the things you want at great prices! http://auctions.yahoo.com/