This reverts commit 7c7cb01aa5.
The smoothing weights will cause shifting in the image if not calculated
as they were in the branch. Solution for better performance will be to force-clip
kernel to a square of size two for projective painting.
Now we build 2 .cubins per architecture (e.g. kernel_sm_21.cubin, kernel_experimental_sm_21.cubin).
The experimental kernel can be used by switching to the Experimental Feature Set: http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Cycles/Experimental_Features
This enables Subsurface Scattering and Correlated Multi Jitter Sampling on GPU, while keeping the stability and performance of the regular kernel.
Differential Revision: https://developer.blender.org/D762
Patch by Sergey and myself.
Developer / Builder Note:
CUDA Toolkit 6.5 is highly recommended for this, also note that building the experimental kernel requires a lot of system memory (~7-8GB).
Just have a default 'Misc' category (harmless for panels in non-cat context).
In case we would still want a panel to show in all tabs (rather unlikely),
just explicitely give an empty string to its bl_category property.
Note I choose 'Misc' because it's much shorter than 'Uncategorized' (space
is an issue here), it's a one-liner to change it anyway if UI Mafia does not like it!
whenever enter edge bevel weight editing.
This is what happens when one makes an edge sharp, for instance.
The edge bevel weight display is not on by default for performance
reasons, but seems reasonable to enable it when user indicates
interest by editing bevel weight values.
Turned out there were several issues in handling of scale parameter by numinput.
Fixed that by factorizing more some code in common with 'usual' numbuttons eval code
(new `bUnit_getScaleUnit()` helper will return valid scaled value, depending on
given system and type).
Now, numinput behaves as expected - using default unit amended by scale in case no unit is given
(i.e. entering '20' with a scale of 0.01 will give you 20cm, and '20cm' as well!).
Also stumbled uppon 'move marker' code, was needing a bunch of fixes, cleanup and simplification,
and added a candy feature - now you will enter numinput values in seconds when editor is
in 'time' mode, instead of frames!
Turned out to be a clean/fix up of modal bevel tool, percentage mode handling was broken,
numinput handling was broken, etc.
Also added a way to switch between bevel types (modes) with M key, and tweaked a bit
numinput code to return early in case of 'char' event with ctrl modifier.
Also, tweaked versionning for lnors color - we can add immediately versionning code,
using future next subversion, even without actually switching to this subversion now.
Avoids 'empty' versionning blocks floating around, and often forgotten when actually
raising version numbers!
This is more like a workaround actually, we use a fixed 'margin' for height in case of search menus,
instead of using shadow width (which gave the bug with low values, and insane margins with big ones).
Note root of the issue is that if 'top' margin is too small, the first entry of the search menu
gets activated before the 'opening' click is released. This means that button will get the
KM_RELEASE event, and immediately quit (see interface_handlers.c:7945, ui_handle_menu_button()).
The sunbeams node was clamping the range of influence to start at 1
pixel distance from the source. This was a poor fix for artifacts caused
by an off set in buffer coordinates. Since the u coordinate starts at
ceil(umax) the v coordinate also has to use ceil. This also fixes some
discontinuities that became visible when the source point is close to
a sharp line in the input image.
Scene replacement with invalid scene name was crashing blender,
now it's a no-op.
KS_Scene.replace() to return a boolean to indicate if the scene
is valid and is scheduled for replacement. This allows more
robust game management.
