First problem is that enabling the "in front" option made the wire object
test against an not updated stencil buffer.
Second problem is that the "in front" option was useless on wire objects.
To fix this we bypass the depth test and live with some sorting problem.
Full redesign of the cache system used for drawing strokes and handle derived frame data.
Before, the cache was saved in bGPdata and a hash was used to manage several objects with the same datablock.
Old design made the use of particles very inefficient and prone to bugs and segment faults, and especially when this was mixed with onion skinning and multiple objects using same datablock. Also, there were some conflicts with the depsgrah logic (the old design was done before despgraph was in place) that made the use of hash not working.
The new design saves the data in the object runtime struct and avoid the use of any hash to find the right data. This improves the speed and reduce a lot the complexity of the code, memory allocation, hash overload and adds full support for particles and reused datablocks.
The particles can reuse the modifiers and shader effects of the original grease pencil object.
This brings a big difference to meshes with edit cage adjusted for modifiers.
In my tests, the suzanne with subdivision modifier level 3 went from 4.80ms to 3.05ms.
By the tests I could only observe a considerable difference in the peformanse when the vertex size is 30.
Vertice 3 showed no difference in a suzzane with subdivision modifier level 3 + show-on-cage.
Point Size 30: 7.29ms vs 2.55ms
Reviewers: fclem
Reviewed By: fclem
Differential Revision: https://developer.blender.org/D3805
NOTE: There is a float imprecision near the focus plane
due to the current technique used for DOF. This makes the alpha channel
transparent on nearly in focus objects even when they should not.
This artifact should be fixed when the DOF will use scatter as gather for
low brightness areas.
Fix T57042 : Eevee does not render alpha when DOF is turned on
