Blender Git Loki

Cycles kernel split: Enable some features for split intel and nvidia kernels

There still some differences between split kernel and megakernel and this is
to be investigated further.

Merge branch 'master' into asset-experiments

Fix type

fix kernel names some more

move to ccl_fetch for kernel_path.h

Cycles kernel split: Move requested tile size from render tile to device task

It is semantically more correct since it's used for only device-specific
calculations.

Cycles kernel split: Make sure task pool is cancelled in device destructor

Also reverted whitespace change compared to master.

Cycles kernel split: De-duplicte device task logic

Move it to the base class and make thread-run virtual function.
This way we don't need to keep task-related logic in all the
derived classes.

Cycles kernel split: Get rid of tile-split related logic in RenderTile

It's only used by OpenCL device and needs to be done in some more generic way,
likely involving Blender querying the maximum/best render settings from the
render engine.

Merge branch 'master' into ui-preview-buttons

Move again GPU init code before UI_init one (since icons use OGL, I assue they need it initiated first).

Oddly enough this did not seem to be an issue, but...

Fix bad alpha - PreviewImage are assumed to be premultiplied...

Cycles kernel split: Finish kernel files rename

Cycles kernel split: Style cleanup in new classes from device_opencl

Cycles kernel split: Make device_opencl even closer to master

use ccl_fetch in kernel_path_surdace

Rename opencl entry points

Formatting fix

replace sc_fetch with ccl_fetch_array

Merge branch 'master' into cycles_kernel_split