New dilation code for bake filtering

Hello everyone,

Morten Mikkelsen (in fact, not only him) noticed really bad things with Blender's bake filtering (this was discussed in our mailing list).

Morten provided new code which is supposed to replace our filtering. I've tried to blender-ize it so it should fit our code conventions.

Also this patch increases maximal margin from 32 to 64. It's needed because baker is now used for really large images.

[nazgul, 2011-07-23 15:49:01]

Fixed possible memory overflow issue. Pointed by Morten. It was my own
regression..

[ideasman42, 2011-07-24 09:46:09]

Aside from the minor concerns this patch looks good to commnit, tested with
alpha, with/without clear option enabled, and making sure overlaps are handled
properly.

http://codereview.appspot.com/4822045/diff/3002/source/blender/imbuf/intern/f...
File source/blender/imbuf/intern/filter.c (right):

http://codereview.appspot.com/4822045/diff/3002/source/blender/imbuf/intern/f...
source/blender/imbuf/intern/filter.c:370: float weight[100];
This seems arbitrarily large, 25 is enough?

http://codereview.appspot.com/4822045/diff/3002/source/blender/imbuf/intern/f...
source/blender/imbuf/intern/filter.c:381: const int is_float =
ibuf->rect_float!=NULL ? 1 : 0;
minor -
can just read:
const int is_float = ibuf->rect_float!=NULL;

and can be at the start of the function.

http://codereview.appspot.com/4822045/diff/3002/source/blender/imbuf/intern/f...
source/blender/imbuf/intern/filter.c:435: ((unsigned char *)
dstbuf)[depth*index+c]= FTOCHAR(acc[c]/255.0f);
Seems unnecessary to /255, then *255 to convert to a char again.

[nazgul, 2011-07-24 10:00:36]

Fixed issues. Ternary operator seems to be easier than re-using different
macroses =\