- curve map insert point had a nested loop which used the same value to index different arrays.
- dynamic paint used ternary operator where both outcomes were the same.
- Force libsamplerate linking statically
- Remove global compiler flags for sse2 and sse.
Tests of release builds didn't show big difference in time of
tracking using SAD algorithm or time of solving,
so this shouldn't introduce big slowdown.
Currently this stuff is used for 2d tracking job only, but the same
thing would necessary for camera solver job, so moved it into more
general utility stuff.
The problem was in fact in the changes made to blend_ramp() func at rev42164. The checks on green pointer (if(g) {...) were checking the existance of a valid value (i.e. if input color was rgb or only shades of grey in r pointer), but as we get now a full float[3] pointer, this test is no more possible. And doing if(r_col[1]) {... was wrong, as it only broke things in case green was 0.0 (which was the case with the generated glare color of bug report).
So, just removed all those tests!
Transparent rendering can't be implemented in way it gives accurate shading in
3d viewport. It's acceptable for object mode, but when you're in edit mode
it gives more problems than benefits of seeing transparent faces.
This is possible fix for for weird polygons in [#29255] Object invisible and weird polygons appearance
"Just a simple check for non power of two support (GLEW_ARB_texture_non_power_of_two), to avoid what seems to be a very slow, and very unnecessary gluScaleImage call."
This is the only part of the VideoTexture that does the POT calculation, so the check seems good.
It would be interesting if some opengl guru would like to benchmark the use of this in other areas of Blender (e.g. 2D Filters, and all GLSL materials).
Note that mipmap should also be supported automatically by this extension, so it's not something to worry about.
I've written a convenient function that returns the sibling of a node in the
red-black tree implementation originally implemented by Joshua Leung.
I want to use this get_sibling() function in the future to implement the missing
removal function of the red-black tree implementation.
For now the get_sibling() function just simplifies the get_uncle() function.
Just like the rest of the red-black tree implementation this diff is based on
Wikipedia: http://en.wikipedia.org/wiki/Red%E2%80%93black_tree#Removal
This issue it totally related on issue with changing object datablock.
For curves it used to guess object type from curve datablock based on
count of control points in V direction.
This quess fails in case when SurfCircle datablock is trying to be reused
by another surface object or as another sample empty surface databouck used
to be treated as curve.
Store type in Curve when creating new Curve datablock which is used in
this object type quessing function.
Note: Previously saved files wouldn't change behavior at all.
If enter is pressed outside of any button in menu, generate two menu return values:
- UI_RETURN_CANCEL so button wouldn't be executed
- UI_RETURN_POPUP_OK so callback associated with popup block would be executed
There were some issues with PBVH which prevented working it for meshes without faces.
Discussed with Brecht, for benefits of dynamic-topology-sculpting and so better to
make PBVH survive such things.
Added some extra NULL-pointer checks for this.
- setting the color channels on save as would ignore the channel - BW/RGB/RGBA setting.
now its used when available but still need to hide BW for formats blender can only write as color.
