Blender Git Loki

Vertex Paint: apply when cursor isn't over faces

This behavior makes more sense for sculpt, less so for painting.
Restores non PBVH behavior, adding `BKE_pbvh_find_nearest_to_ray` -
similar to ray-cast except it finds the closest point on the surface.

Cleanup: rename dist -> depth

Prepare to add code that stores distance to the ray, avoid confusion.

Cleanup: remove unused struct member

Merged with soc-2016-pbvh-painting, no longer needed.

Disable cursor drawing while navigating

Was performing ray-casts in sculpt mode on every update.

Fix sculpt secondary color missing some brushes

D2841 by @uvwxyz w/ edits

Cleanup: use bool for brush checks

Fix brush reset (missing notifier)

D2843 by @uvwxyz

CMake: use restrict w/ gcc, not clang

PyAPI: fast keyword parsing for bpy modules

No functional changes.

CMake: add -Wrestrict for GCC

Fix passing the same argument twice to BLI_strncpy

Fix setting the operator name in Py operator API

Fix incorrect CUDA remaining time estimate after previous commit.

Cycles: CUDA faster rendering of small tiles, using multiple samples like OpenCL.

The work size is still very conservative, and this doesn't help for progressive
refine. For that we will need to render multiple tiles at the same time. But this
should already help for denoising renders that require too much memory with big
tiles, and just generally soften the performance dropoff with small tiles.

Differential Revision: https://developer.blender.org/D2856

Fix use of uninitialized memory in Cycles normal baking.

Code refactor: zero render buffers outside of kernel.

This was originally done with the first sample in the kernel for better
performance, but it doesn't work anymore with atomics. Any benefit was
very minor anyway, too small to measure it seems.

Code refactor: use split variance calculation for mega kernels too.

There is no significant difference in denoised benchmark scenes and
denoising ctests, so might as well make it all consistent.

Code refactor: remove rng_state buffer and compute hash on the fly.

A little faster on some benchmark scenes, a little slower on others, seems
about performance neutral on average and saves a little memory.

Code refactor: add WorkTile struct for passing work to kernel.

This makes sharing some code between mega/split in following commits a bit
easier, and also paves the way for rendering multiple tiles later.

Fix T52645, T52645: AMD OpenCL compiler crash with recent drivers.

Work around the bug by reshuffling code.