- User Since
- Nov 28 2009, 10:22 PM (520 w, 4 d)
@Philipp Oeser (lichtwerk) you read from evaluated. If you need to modify stuff you write to original.
If the issue is in FFTW then i would expect locking to happen closer to it.
Tue, Nov 19
Looks good to me now.
@Sybren A. Stüvel (sybren) this isn't a place for such kind of nagging. You issue has nothing to do with this exact change.
@Stefan Werner (swerner), is this still relevant?
This could very well be a limitation of cloth/physics.
Comes from the fact that OpenSubdiv tries to find a smoother surface approximation for triangular (any non-quad) faces. Since the model is rather high-poly and all triangulated this results in much higher memory footprint during subdivision process.
Some inlined comments which I've potted while compiling this patch.
Mon, Nov 18
After having some discussion in blender-chat and having a second look i actually think this is a right way to go: we must not have spin lock while we are reading file from disk.
Sun, Nov 17
Fri, Nov 15
The root of the issue is the lack of smooth propagation from lower levels to higher levels. This applies to sculpting and to Apply Base. And to Subdivide to certain extend, and is leaving to either spiky or facety artifacts on a higher levels.
Thu, Nov 14
Same root cause than T64793
@Ha Hyung-jin (robobeg), i am confused a bit. Do you still need help figuring out the regression tests?
That is exciting to see movement on this front!
Wed, Nov 13
@Alan Ihsan (AlRedd), the proper fix will be started to be worked on as soon as 2.81 tasks are tackled.
@Brecht Van Lommel (brecht), thanks for the review!
Compiles without warning, and regression tests are passing. But I can not test OpenCL on my workstation.
Tue, Nov 12
If there are precompiled libraries in SVN we should enable it on Windows, yes.
Eeh, spammy fix for spelling of CPack :)
Round of mainly small fixes.
When using OpenCL with Cycles the rendering performance became more.
In future should also use pathlib here.
Enabled in the depsgraph
So, the Badges have to be created manually in Phabricator.
That's a great find!
Fri, Nov 8
At this state the code is so called "seems to work". Tested it in a local environment of both buildbot server and worker VMs.
If there is a way which avoids use of environment variable i'd choose that way.
Thu, Nov 7
Wed, Nov 6
Tue, Nov 5
It seems that i am actually having 2x slowdown from this change.