Nodes which are not connected to output no longer tags tree as updated. This is done to avoid unnecessary re-renders when artists are working on an unconnected parts of the shading tree.
Bastien, are you looking into applying this?
Don't think it's really handy to have just a one line per library. It's important to have exact version, possible local modifications and such.
Please do a little more research before opening bug reports like this.
Please attach .blend file which clearly demonstrates the issue. We can not deal with issues we can't reproduce or for which it's only known that they are happening sometimes. Try isolating the issue.
I don't really see regression in behavior here, but it depends on active render engine, material settings and so. So please always attach .blend file which demonstrates the issue.
Closing as invalid.
@George (hisforever) please try http://blenderartists.org/forum/forumdisplay.php?11-Python-Support for issue's with your addon.
It's always a good idea to attach exact .blend with exact audio file which demonstrates the issue.
Thanks for the report, but it's one of the known limitations of BI renderer. The idea now is to retire this engine and start working on a new one OpenGL+GLSL based as a part of Blender2.8 project.
Something is wrong with indentation. Other than that seems fine.
Bastien replied there and GHOST is maintained by us.
sigh… Next time, please mention what your patch is about (what piece of code it affects) in the report… Sorry, but this report looked totally the same as T48491, so no, did not think about looking at patch file itself. GHOST is maintained by us indeed, and totally eligible to patches submission.
It looks, according to T48499, like the code for GHOST is off-limits for patches as well...
Are you saying the GHOST code is off limits for patches as well?
This is what I get with the latest patch, slight less memory usage still. About the same render times as the original patch on my GTX 960.
Ok, the memory usage is normal but wouldn't expect that much of a slowdown. So this is a multi GPU render, with the default koro_gpu.blend?
Update for latest master, try to avoid pointer indirection.
Hi, check again with latest master and got slightly higher memory usage and slowdown on most test files.
Different textures are using different math. It might be fast or slow, depending on particular videocard you're using.
Since it's a dev feature, how about this then? if the environment is setup correctly it'll work, but we don't go out of our way to set it up for them (the same way we don't really check if gcc 5 is installed on linux) It'll either work or it won't
I think it's clear that this is a developer tool. It needs to be enabled via environment variable or Blenders Debug Menu (value 256). Don't mind to add this for Windows as well, but I would not add so much code for it. We also dont check for a supported gcc on Linux. Hardcode some common paths (default for MSVC on Windows), assume it is installed among the CUDA Toolkit, and then it works or doesnt.
I think it would be reasonable to have this as a developer feature on Windows (I'm doing CUDA development mostly on Windows at the moment and would probably use this). I would also ensure it's clear this is a developer feature on Linux too. The choice of NVRTC or MSVC for me mostly depends on code complexity, whatever is simpler.
As a development tool that could be handy, but it wouldn't be something which we'll advertice to artists to use.
Thanks for the tests, there indeed seems to be something I have to fix here. Testing with a GTX 960 I also saw 1-3% speedups rather than slowdowns, on multiple benchmark scenes.
If i were to go the NVRTC route (looked further into it, can probably make it work for us), bringing the requirement down to just needing the cuda toolkit, will you consider it? or am I wasting my time here?