[Bf-cycles] Volumetric patch update

[storm]

Minor bi-dir black pixel cleanup (clamp weights to [0. +inf) ), and
finally i make proper volume sampling pdf support, result not that great
as i expected, but noticeable (only homogeneous volume take advantage
because it perfectly fit importance sampling).

After all that, i surprised how much less fireflies in many test scenes,
maybe i miss some major core update?

At last I get rid of old stupid sample_number_buffer, it was wrong turn.

Unfortunately i still in doubt if weighting work as intended, i must to
play with test scenes more before clean all rest commented out nightmare
in kernel_mlt.h

Found new bug that look dangerous, on one scene there are many bright
colored random pixels not far from transparent surface, as if background
(air) volume shader present, but it not. Looks like there is rare
condition when volume step can run beyond face. That seen only with
insane number of bounces.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: blender_cycles_volume_mlt_45900.patch.bz2
Type: application/x-bzip
Size: 43045 bytes
Desc: not available
Url : http://lists.blender.org/pipermail/bf-cycles/attachments/20120424/c9d0cb1e/attachment.bin