Re: [AD] pink color conversion blit bug

[Javier Gonzalez <xaviergonz@xxxxxxxxxx>]

> blit() before my patch:
>
>              32->16   32->8   8->32
> no dithering 466.8    5.6     233.6
>    dithering 178.2    2.3     233.9
>
> blit() with my patch:
>
>              32->16   32->8   8->32
> no dithering 292.6    5.7     236.5
>    dithering 149.7    2.4     237.9
>
> blit() with my patch and with COLORCONV_KEEP_TRANS:
>
>              32->16   32->8   8->32
> no dithering 258.3    5.3     238.3
>    dithering 138.8    2.2     237.9
>
> blit() without doing conversions should not be affected at all of course.
>
> The 32->8 uses no map table so it doesn't say much, but i don't want to
run
> it again right now.. the 32->16 no dithering shows that after my patch
> conversions are only running at about 60% compared to before. I didn't
> expect the patch to have such a high impact on speed, it should only add
> one bit-flag test to every pixel.

Well, IMHO checking the flag for every pixel is not a good idea, but
check it only once before blitting would be better.

IOW, having other blitting functions for KEEP_TRANS instead of patching
existing ones and call those new special ones if the flag is set, making
the check only once and not loosing any speed in old ones

what do you think?