[AD] malloc for bitmaps alignment?

[aj <aj@xxxxxxxxxx>]

At 07:51 AM 20/05/2004 -0500, you wrote:
> > i have not suggested aligning
> > everything.
> > i have suggested aligning the malloc
> > of the data for memory bitmaps.
>
> Yes, sorry, I got carried away thinking
> of other places where SSE could be used :)
>
> Aligning bitmap data to 16 bytes is
> easily done, but would be useless if
> you do operations on sub bitmaps. I do
> a lot of that, say (in my GUI system)
> and almost never touch the master bitmap. Still, it's easily done, and,
> since you say you're doing your own
> SSE code for bitmaps, my last comment
> stands: have you got any measurement
> for speed, comparing aligned and non
> ailgned speeds, with small and large
> bitmaps ?


any hard data about performance gains would be influenced by  cache size, 
cache availability, cpu load, CPU model  etc..
the only thing i can go on is the available literature on the web, and the 
general expectance that aligned is faster than non-aligned.

i see the sub-bitmap  alignment  is a bit of an issue, but i see no 
solution for that.
as the data is malloced i dont see why mallocing from 16byte boundarys is 
much of an issue, we are not creating 'dead space' with any of allegro's 
structures..  any dead space would only be created by the memory manager 
used by the malloc, which is not our concern.

i see no reason why we should not use 16byte aligned malloc.

aj.