Blender Git Loki

Cycles: Fix atomics in split kernel after switching away from OpenCL atomics

Apparently Blender's atomics return the new value rather than the old value.

Fix T49743: Adding torus in edit mode local mode shows error

The 'local' layers were not correctly set when redoing 'add object'
addons using object_utils.py helper (we always want to restore layers
from view in local view, even if we set 'real' layers from operator
afterwards).

Fix T49829: Removal of custom icon previews during add-on unregister crashes Blender.

Issue was happening when removal of custom icons was done while they
were still being rendered by preview job.

Now add a 'deffered deletion' system, to prevent main thread to delete
preview image until loading thread is done with them.

Note that ideally, calling `ED_preview_kill_jobs()` on custom icon
removal would have been simpler, but we don't have easy access to
context here...

Cycles: More workarounds for weird crashes on AVX2

Oh man, is it a compiler bug? Is it something we do stupid?

For now more crap to prevent crashes. During the conference will talk to
Maxyn about how can we troubleshoot such weird issues.

Cycles: Another attempt to fix crashes on AVX2 processors

Basically don't use rcp() in areas which seems to be critical after
second look. Also disabled some multiplication operators, not sure
yet why they might be a problem.

Tomorrow will be setting up a full test with all cases which were
buggy in our farm to see if this fix is complete.

Cycles: Fix compilation error of AVX2 kernel without SSE math

Mantaflow: Use Path objects instead of os.path.join()

File paths are passed into these functions, so convert passed path to pathlib.Path object to avoid internal/external conflicts. This should be easier to maintain and visually cleaner.

Reviewers: sebbas

Reviewed By: sebbas

Differential Revision: https://developer.blender.org/D2317

Merge remote-tracking branch 'refs/remotes/origin/blender-v2.78-release' into fracture_modifier

Conflicts:
source/blender/makesrna/intern/rna_modifier.c

Merge branch 'master' into cycles_split_kernel

Cycles: Add CPU split kernels

This adds all split kernels for CPU devices besides the `data_init` kernel
which still needs to be written for CPU. While the kernels build and load
properly they don't do anything yet.

Viewport: Set outline color for mesh based on active/select state

(also, fix warning regarding const float being written)

You only see the color if you use the "modern" viewport option
(otherwise I believe Blender is drawing the old on top of the new outline).

That said, in the "modern" viewport we have unfreed mem. To be
investigated separately.

Cycles: Completely disable transform SSE for now

Was causing issues on another frame.

On a tight schedule, disabling for now so artists are happy.

Still looking into root of the issue!

RangeTree API rewrite

Rewrite the current range-tree API used by dyn-topo undo
to avoid inefficiencies from stdc++'s set use.

- every call to `take_any` (called for all verts & faces)
removed and added to the set.
- further range adjustment also took 2x btree edits.

This patch inlines a btree which is modified in-place,
so common resizing operations don't need to perform a remove & insert.
Ranges are stored in a list so `take_any` can access the first item
without a btree lookup.

Since range-tree isn't a bottleneck in sculpting, this only gives minor speedups.
Measured approx ~15% overall faster calculation for sculpting,
although this number time doesn't include GPU updates and depends on how
much edits fragment the range-tree.

Cycles: Fix crashes after recent optimization commits

There is some precision issues for big magnitude coordinates which started
to give weird behavior of release builds. Some weird memory usage in BVH
which is tricky to nail down because only happens in release builds and GDB
reports all variables as optimized out when trying to use RelWithDebInfo.

There are two things in this commit:

- Attempt to make vectorized code closer to original one, hoping that it'll
eliminate precision issue.
This seems to work for transform_point().
- Similar trick did not work for transform_direction() even tho absolute
error here is much smaller. For now disabled that function, need a more
careful look here.

BLI_bitmap_draw_2d: optimize polygon filling

Existing method was fine for basic polygons but didn't scale well
because its was checking all coordinates for every y-pixel.

Heres an optimized version.
Basic logic remains the same this just maintains an ordered list of intersections,
tracking in-out points, to avoid re-computing every row,
this means sorting is only done once when out of order segments are found,
the segments only need to be re-ordered if they cross each other.

Speedup isn't linear, test with full-screen complex lasso gave 11x speedup.

Cleanup: rename functions in BLI_bitmap_draw_2d

Cleanup: move bitmap drawing into its own module

Bitmap drawing is out-of-scope for a general math API,
move to BLI_bitmap_draw_2d.

OpenGL: follow up on mesh wire TODOs

- depth test on (affects scene depth for now)
- color variations
- object outline does not need GL_BLEND with latest shader

OpenGL: edge shader discards fragments of unused edges

Discarded edges should not update depth buffer or blend with color buffer.

Geometry shader version will make this obsolete.

Fixed glossy reflections and refractions for low roughness values and
cleaned up the code.

For low roughness values, the reflections had some strange behavior.