When using OpenCL with Cycles the rendering time increased substantial.
After doing some tests the bottleneck was found in 4d voronoi and 2d and
3d smooth voronoi.
This change will hide these behind a specific compile directive so the
speed will improve.
AMD RX480 + BMW scene
2.80 (3:10)
2.81 (5:48)
2.81 excluding 4d voronoi+2d/3d smooth (3:50)
Reviewed By: sergey
Differential Revision: https://developer.blender.org/D6231
The root of the issue goes to the discontinuity between the way how
mesh_calc_modifiers() and BKE_sculpt_multires_active() works.
At some point detection of original data usage by a modifier got
broken: the mesh_final based check is unreliable because deform-only
modifiers will create mesh_final for the connectivity information.
This made it so modifier stack evaluation would skip multires
evaluation, but the sculpt code will assume the multires is properly
applied.
This change makes it an explicit check about whether there are any
non-deform-only modifiers applied.
Pair programming and review together with Bastien, thanks!
There was a discontinuity between how deform-only modifiers are applied
for the case when result deform mesh is requested and when it is not.
Namely, the input mesh will always be guaranteed to present in the
former case, but not in the latter.
This change makes it so input mesh to deform-only modifiers is always
at consistent state.
Pair programming and review together with Bastien, thanks!
This changes integrates code signing steps into a buildbot worker
process.
The configuration requires having a separate machine running with
a shared folder access between the signing machine and worker machine.
Actual signing is happening as a "POST-INSTALL" script run by CMake,
which allows to sign any binary which ends up in the final bundle.
Additionally, such way allows to avoid signing binaries in the build
folder (if we were signing as a built process, which iwas another
alternative).
Such complexity is needed on platforms which are using CPack to
generate final bundle: CPack runs INSTALL target into its own location,
so it is useless to run signing on a folder which is considered INSTALL
by the buildbot worker.
There is a signing script which can be used as a standalone tool,
making it possible to hook up signing for macOS's bundler.
There is a dummy Linux signer implementation, which can be activated
by returning True from mock_codesign in linux_code_signer.py.
Main purpose of this signer is to give an ability to develop the
scripts on Linux environment, without going to Windows VM.
The code is based on D6036 from Nathan Letwory.
Differential Revision: https://developer.blender.org/D6216
The heap on windows is single threaded causing it to lag behind linux in performance in allocation heavy multithreaded scenarios, BVH building is a prime example.
See https://developer.blender.org/D6218 for benchmark results
for testing with the allocator enabled/disabled you can set the environment variable TBB_MALLOC_DISABLE_REPLACEMENT=1 to disable the TBB allocator.
Reviewed By: @sergey
Differential Revision: https://developer.blender.org/D6218
Makes it so compilation doesn't fail when the SVN updating
stumbles upon checkout which doesn't have correspondence in
a tag, but which isn't so risky as previous change.
This reverts commit 8e9e58895b.
The change broke behavior when typing `make update` from the root of
the sources: tests folder wouldn't be updated anymore.
Getting quite close to release now, so will revert to a safer change.
The issue was rooting to the fact that the script was iterating into
every directory inside of blender.git/../lib/ and attempted to switch
them to the desired path. This doesn't work in an environment where
both master and release branch are built (or any environment where
non-needed SVN directories are not automatically removed).
This change makes it so script explicitly generates a list of
directories which are required for the build. For example, the script
now stores an exact folder with ABI such as win64_vc14.
Only those explicitly listed directories will be updated.
This allows to:
- Solve compilation failure of 2.81 branch after checkout for
win64_vc15 libraries has been created.
- Fail compilation if actually expected tag is missing (for example,
when trying to build release branch prior to libraries tag).
Now, there was a confusing logic about possible .svn folder in
lib_dirpath (effectively, blender.git/../lib/.svn) which is not
something what is supposed to happen with the setup of buildbot we are
using for quite some time now. This logic has been removed now.
This change includes old-style string format(), mainly because it is
not know that the buidlbot scripts are run using python3 on CentOS
builder.
Differential Revision: https://developer.blender.org/D6230
For `Particle Properties -> Viewport Display -> Display As` set to
circle/cross/axis, particle instances are associated with a single
resource handle (and, in particular, a single model matrix), so define
`IN_PLACE_INSTANCES` to get the right index for `ModelMatrix` and
`ModelInverseMatrix` in the shader.
Differential Revision: https://developer.blender.org/D6220
Open the file browser sidebar by default when adding Movie/Sound/Image Sequence Strips, to show the operator options.
Differential Revision: https://developer.blender.org/D6212
Reviewed by: William Reynish, Julian Eisel
This caused T71483.
Having timeline and dopesheet merged in 2.8 makes using S/E keys in timeline not work well.
2.7x keymap needs to fit with 2.8x internals, so probably 2.7x keymap will have to live with Ctrl+Home / Ctrl+End here.
This reverts commit 4fbcbbfb96.
As correctly pointed out by a comment in the code, adding
new springs wasn't thread safe, and caused crashes.
Fix by buffering new springs in intermediate thread-local
arrays, which are flushed on the main thread. This is valid
because the new springs are not used until the next sim step.
Differential Revision: https://developer.blender.org/D6133
Fixes alignment issues on Graph Editor menus used to insert keyframes.
Differential Revision: https://developer.blender.org/D6213
Reviewed by Campbell Barton
Very stupid mistake in own new generic ID lib_link function, that would try
to link ID pointers for all data-blocks, not only those actually needing it.
