New release BlendLuxCore Wheels v2.9a2 - bugs, quirks and feedback

[epileptrolytics]

First, big thanks to @howetuft and @odil24 for updating and improving BlendLuxCore!

I want to give some feedback concerning the Mac OS ARM version I have installed on my MacBook Pro M3 Pro, Sonoma.
I'm unsing Blender 4.3.2 which is more stable than 4.2 LTS which crashes a lot even with Cycles.

Comparing with BlendLuxCore 2.61 on Blender 3.1 Intel/Rosetta rendering with PathOCL on the GPU is much slower especially at the render start, while BiDir speed has improved on the new ARM release. But the "post12" version I tested previously with Blender 4.2 was even faster with BiDir, while Path OCL was only 1/2 speed of v2.9a2 and 1/4 of 2.61 (not in the image above).

OIDN works faster as well.

I opened and tested various files and mostly everything works, the render speed gain respectively loss is also similar, so the OpenCL implementation for Apple Silicon needs some improvement.

I will continue testing and post here when something comes up.

[howetuft]

Hello,
I've activated a few optimizations in the wheels, including:

• SIMD use in embree
• Intel TBB use in embree
• SIMD use in openvdb
• General optimization for speed in compilation (O2/O3) all over the code

It's in the last version of wheels (post13)

It is expected to speed up rendering: --> to check it, would you be so kind to rerun your above benchmarks, please?

Normally, the last version of the wheel should be installed automatically by BlendLuxCore, but it may be safer to uninstall/reinstall BLC.
Many thanks in advance!

[epileptrolytics]

Hi, thanks for your improvements!

Now I'm an artist, not a coder and don't understand this new wheels installation stuff.
I'm only on GitHub because when I found a bug the old LuxCore Devs told me: Get on GitHub and open an issue!

Now I tried to deactivate the BLC extension which caused this error:

Then I deleted BLC and installed anew from the official 2.9.a2 wheels release.
Then I found the refresh button I should have used.

Anyways, the speed test with the food scene shows no change at all, 775 k/s for BiDir, PathOCL ~800k/s.

I went searching for this post13 release and found a bunch of entries, trying the 13-11 one it wouldn't install with this error:

Sorry for being of no help!

A noob like me needs a valid download link.
Also it would help when the BlendLuxCore Add-on preference setting would show the current version like: 2.9.a13 so that there were a feedback if the runtime refresh actually succeeded.

[odil24]

Hi, thanks for your improvements!

Now I'm an artist, not a coder and don't understand this new wheels installation stuff.
I'm only on GitHub because when I found a bug the old LuxCore Devs told me: Get on GitHub and open an issue!

Now I tried to deactivate the BLC extension which caused this error:

Then I deleted BLC and installed anew from the official 2.9.a2 wheels release.
Then I found the refresh button I should have used.

Anyways, the speed test with the food scene shows no change at all, 775 k/s for BiDir, PathOCL ~800k/s.

I went searching for this post13 release and found a bunch of entries, trying the 13-11 one it wouldn't install with this error:

Sorry for being of no help!

A noob like me needs a valid download link.
Also it would help when the BlendLuxCore Add-on preference setting would show the current version like: 2.9.a13 so that there were a feedback if the runtime refresh actually succeeded.

@howetuft It seems-like that bug appeared which I fixed before (are you remember my commit about reg/unreg?).

@epileptrolytics It is because "blender_manifest.toml" file has not been included in the zip archive (@howetuft will review it again)

[howetuft]

Hello @epileptrolytics,

First of all, thank you for your voluntary help!
In the latest version of BlendLuxCore Wheels, you're not supposed to fetch the wheel yourself: the extension takes care of this itself - it must have done so silently.
Moreover, user installation of the wheel is not provided for, so the error message you got when trying to do so (“Missing manifest”) is logical.
On the other hand, I agree with you that the wheel version should be displayed on the extension page. I'll see what I can do.

@odil24 regarding the other “Missing bl_rna attribute” message, perhaps the repairing commit got lost in the episodes of erroneous merges and rollbacks? Can you please check and, if necessary, push back a PR (being careful to circumscribe it to this commit only)?

Anyways, the speed test with the food scene shows no change at all, 775 k/s for BiDir, PathOCL ~800k/s.

That's the most annoying thing.

[epileptrolytics]

There is a new BLCw v2.9a3 release online, tested on MacBook M3 Pro MacOS 14.7 on Blender 4.2.5 LTS ARM:
No change of the results in my first post with CPU and GPU speed in the food scene.

Still the development is very encouraging and much appreciated, keep up the good work guys!

[CodeFHD]

I've started testing with my Archlinux notebook again. There is an issue with geometry it seems.

System info:
Blender 4.2.5 LTS and 4.3.1
BlendLuxCore: v2.9a3 and blc-build from source
CPU: Ryzen 8845HS
GPU: RTX4060 Mobile

Description to reproduce:

• Load default scene
• Activate Viewport render
• Blender crashes, see log attached

Variation 1: If the cube is deleted, viewport doesn't crash. Lights and Camera seem ok.
Variation 2: Final Render (Press F12) doesn't crash, but the cube is not visible

blender.crash.txt

Edit: Just to be clear, no issue on Windows (with different hardware though)

[CodeFHD]

Possibly related Core issue:
LuxCoreRender/LuxCore#626

Shows a similar backtrace by someone else earlier in 2024, so might not be related to the wheels build

[howetuft]

There is an issue with geometry it seems.
Load default scene
Activate Viewport render
Blender crashes, see log attached

Hello,

I actually confirm the issue in post13 for Linux (not in post12 and lower, and not in Windows).

Curiously, the bug does not occur :

• during the automatic tests that are run when the wheel is built
• during tests that can be run on the command line after the wheel has been installed (pyluxcoretest)
• during rendering of the file generated by BlendLuxCore, using pyluxcoreconsole.
  This only happens in the context of Blender's Python interpreter.

After investigation, it is linked to an optimization triggered in post13 which compiles embree3 with TBB activated. This also implies using embree3 as a shared library, not a static one like in previous version. To me, it smells like an alignment problem, but without certainty.

In any case, I've rollbacked the embree / TBB optimisation and made a new release of the wheel: post14 version. This should restore normal operation.

However, it raises the question of why precisely this bug appears in the circumstances described above and how TBB can be activated in embree without it occurring; and also the question of introducing Blender-based tests into the construction of the wheel.

[odil24]

Anyways, the speed test with the food scene shows no change at all, 775 k/s for BiDir, PathOCL ~800k/s.

@howetuft @epileptrolytics
Here my results on Windows 10:

[odil24]

...made a new release of the wheel: post14 version

@howetuft sorry, but I do not understand what does mean that "postxx"

[howetuft]

This is the wheel version (or sub-version exactly). The version is 2.9.alpha1, which is LuxCore version which the wheels are based on, but when I release a new version of the wheel, tweaking or fixing the build, I increase the "post".
So, the complete version for the wheel is: 2.9.alpha1.postXX
By the way, nice render!