Re: [AD] [4.9] pitch in memory bitmap

[AJ <aj@xxxxxxxxxx>]

what are the reasons for a :

8 bit alignment.
32 bit alignment.
128 bit alignment.



8..    pro: saves memory      con:  prevents tricks like using uint32_t  
to move 4 at a time,   fails SSE alignment rules.
32:   pro:  average alignment for many cases at many pixel 
depths.            con:  small waste of space;   fails SSE alignment rules.
128:   pro:   SSE aligned (non aligned data can actually crash if (some) 
SSE instructions are issued).        con:  waste of space.



128bit alignments worst case, is lots of space wasted..  how much?    
lets say a 1025 x 768 surface..  (why anyone would have a 102[5], i dont 
know!!, anyway)..   1025 would have 15 bytes of slop per line,  15*768 
=  11500 bytes

mmm... 11k is not completely trivial, but its not outrageous either.

however, real use-case suggests  anyone wanting surfaces of this size 
would be  making 102[4] x ..  not  102[5].   
more reality use-case suggests anyone make a bazillion  33x33 surfaces 
for sprites,  isn't going to use SSE anyway (the overhead is not worth 
it).   therefore doesn't need 128bit alignment.


The use-case really determines the alignment requirements;  it has to be 
left upto the user to specify if they want other than the default.
Anyone aware of alignment issues and has alignment requirements is going 
to set the alignment they want regardless of what the default is..
By that i mean, document it to say "the default",  never say 32, 8, 128  
whatever, dont be specific;  as long as the docs always say default, 
then the user must take responsibility to set it to their requirements.
That can be made quite clear in the docs,  "If you want an alignment 
other than default, you must specify it before creating the bitmap".


Speaking as a user, if the API for this gets screwed up, i'll be 
re-writing my own allegro version to give me 128bit alignment were i 
need it.  For me, the only con for 128bit is the amount of waste, which 
i find acceptable given the amount of memory modern PCs have, and the 
majority of use-cases aligns naturally.




Peter Wang wrote:
> On 2008-02-21, Peter Hull <peterhull90@xxxxxxxxxx> wrote:
>   
> > The actual value should be transparent to the user - it's only
> > relevant if the bitmap is locked for direct access and in that case
> > the pitch is given by the ALLEGRO_LOCKED_REGION structure. The default
> > should be chosen by Allegro based on the platform, is there any reason
> > why the user would need to change it?
> >     
>
> I'm not sure that Allegro will always be able to pick an alignment that
> suits all uses, and when that happens the user would be SOL.  Not sure
> how likely that is though.
>
> Peter
>
>
> -------------------------------------------------------------------------
> This SF.net email is sponsored by: Microsoft
> Defy all challenges. Microsoft(R) Visual Studio 2008.
> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
>