Would rather have mathematical functions consistent from using
the same vector for input and output values point of view then
introducing questionable optimizations.
This implements simple function which collapses internal loops
caused by self-intersections into a singularity.
This loops can't be removed because rasterizer expects points
of feather be aligned with points from spline itself.
Issue was caused by threading conflict between compositor output node which
is freeing buffers used by render result image and image draw code which
could use buffers at the same time as compositor frees this buffers.
Solved by adding adding lock around viewer image invalidation and image
drawing.
Use renamed LOCK_PREVIEW mutex for this, which si not called LOCK_DRAW_IMAGE.
With new compositor locking for preview is not needed so it could be removed.
Added the same lock around viewer operation which also frees buffers used
by viewer image. It's actually quite difficult to check whether this is
indeed needed. This code seems to be using acquire/release technique, but
somehow acquiring ImBuf before invalidating it in compositor operation
doesn't resolve the issue, so probably it's not actually locking acquire
and things should be checked deeper.
- Fixed issue with black areas appearing when too many sites
are defined.
Currently tweak epsilon value for this, but probably actual
issue is somewhere else, can't see it yet.
- Fixed issue with bright pixels appearing in the sites, was
caused by accumulating color for pixels, which isn't needed.
Once color for pixel was set stop iterating via triangles.
Could give some speedup too.
- Ignore markers which are outside of frame bounds, they were
giving bad triangulation and they can't affect on gradient
due to color fir such sites is not known.
- Sites used to be created at position without track offset
taken into account.
When creating tile data include only triangles which have got intersection
with tile's rectangle only. This saves quite a lot of per-pixel iterations
through triangles which simply can not affect on current tile.
In fact, it's AABB check is used here. It could be improved further, but
it'll slowdown tile data generation with questionable speedup.
Another major slowdown is in fact caused by voronoi triangulation code.
Currently it's used naive algorithm which is O(N^2) where N is number
of edges. Added few euristics there and removed unused part of code, which
gave quite noticeable speedup already.
This could be improved further, but this node is not ment to be used for
lots of markers. It's also generates wrong triangulation when there're
many sites used. Need to be investigated further.
