oneAPI device fails at load or kernels compilation on Archlinux #107618
Labels
No Label
Interest
Alembic
Interest
Animation & Rigging
Interest
Asset System
Interest
Audio
Interest
Automated Testing
Interest
Blender Asset Bundle
Interest
BlendFile
Interest
Collada
Interest
Compatibility
Interest
Compositing
Interest
Core
Interest
Cycles
Interest
Dependency Graph
Interest
Development Management
Interest
EEVEE
Interest
Freestyle
Interest
Geometry Nodes
Interest
Grease Pencil
Interest
ID Management
Interest
Images & Movies
Interest
Import Export
Interest
Line Art
Interest
Masking
Interest
Metal
Interest
Modeling
Interest
Modifiers
Interest
Motion Tracking
Interest
Nodes & Physics
Interest
OpenGL
Interest
Overlay
Interest
Overrides
Interest
Performance
Interest
Physics
Interest
Pipeline, Assets & IO
Interest
Platforms, Builds & Tests
Interest
Python API
Interest
Render & Cycles
Interest
Render Pipeline
Interest
Sculpt, Paint & Texture
Interest
Text Editor
Interest
Translations
Interest
Triaging
Interest
Undo
Interest
USD
Interest
User Interface
Interest
UV Editing
Interest
VFX & Video
Interest
Video Sequencer
Interest
Viewport & EEVEE
Interest
Virtual Reality
Interest
Vulkan
Interest
Wayland
Interest
Workbench
Interest: X11
Legacy
Asset Browser Project
Legacy
Blender 2.8 Project
Legacy
Milestone 1: Basic, Local Asset Browser
Legacy
OpenGL Error
Meta
Good First Issue
Meta
Papercut
Meta
Retrospective
Meta
Security
Module
Animation & Rigging
Module
Core
Module
Development Management
Module
Grease Pencil
Module
Modeling
Module
Nodes & Physics
Module
Pipeline, Assets & IO
Module
Platforms, Builds & Tests
Module
Python API
Module
Render & Cycles
Module
Sculpt, Paint & Texture
Module
Triaging
Module
User Interface
Module
VFX & Video
Module
Viewport & EEVEE
Platform
FreeBSD
Platform
Linux
Platform
macOS
Platform
Windows
Severity
High
Severity
Low
Severity
Normal
Severity
Unbreak Now!
Status
Archived
Status
Confirmed
Status
Duplicate
Status
Needs Info from Developers
Status
Needs Information from User
Status
Needs Triage
Status
Resolved
Type
Bug
Type
Design
Type
Known Issue
Type
Patch
Type
Report
Type
To Do
No Milestone
No project
No Assignees
5 Participants
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: blender/blender#107618
Loading…
Reference in New Issue
Block a user
No description provided.
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
System Information
Operating system: Linux-6.3.1-arch1-1-x86_64-with-glibc2.37 64 Bits, X11 UI
Graphics card: Mesa Intel(R) Arc(tm) A770 Graphics (DG2) Intel 4.6 (Core Profile) Mesa 23.0.3
Blender Version
Broken: version: 3.6.0 Alpha, branch: main, commit date: 2023-05-04 09:04, hash:
e674e32aa122
Short description of error
If I render a scene (it doesn't mind if the default scene or another one), Blender 3.6 ends in a crash.
It crashed also with the 6.2 kernel, that I have installed before.
I have attached the output of --debug-cycles. I hope that helps.
Exact steps for others to reproduce the error
Open Blender → render with cycles → select GPU rendering → render → blender crashed.
Thanks for the report!
I couldn't reproduce exactly your issue with
e674e32aa1
but it didn't run well on my setup either.I had success with latest alpha build (May 15th -
1ad4f67d78
) that fixed a lot of things with Embree 4.1 update, so it's likely it fixed your issue.Can you check again on your machine with such recent build?
If the crash is still here, please try with HWRT enabled and disabled and report what are your versions of
ls -l /lib/x86_64-linux-gnu/libze*
ls -l /lib/x86_64-linux-gnu/libigc*
I have try it with the latest alpha build from builder.blender.org. (4.0.0 Alpha, branch: main, commit date: 2023-05-17 23:34, hash:
9fbadc98cd64
) and it was the same problem.The output of
ls -l /lib/x86_64-linux-gnu/libze*
was not possible, because the directory /lib/x86_94-linux-gnu wasn't present. I have tried the following command:ls -l /lib/libze*
This was the Output:
lrwxrwxrwx 1 root root 20 2. Mai 03:11 /lib/libze_intel_gpu.so -> libze_intel_gpu.so.1 lrwxrwxrwx 1 root root 28 2. Mai 03:11 /lib/libze_intel_gpu.so.1 -> libze_intel_gpu.so.1.3.25812 -rwxr-xr-x 1 root root 19542576 2. Mai 03:11 /lib/libze_intel_gpu.so.1.3.25812 lrwxrwxrwx 1 root root 17 2. Mai 03:09 /lib/libze_loader.so -> libze_loader.so.1 lrwxrwxrwx 1 root root 21 2. Mai 03:09 /lib/libze_loader.so.1 -> libze_loader.so.1.9.4 -rwxr-xr-x 1 root root 300976 2. Mai 03:09 /lib/libze_loader.so.1.9.4 lrwxrwxrwx 1 root root 24 2. Mai 03:09 /lib/libze_tracing_layer.so -> libze_tracing_layer.so.1 lrwxrwxrwx 1 root root 28 2. Mai 03:09 /lib/libze_tracing_layer.so.1 -> libze_tracing_layer.so.1.9.4 -rwxr-xr-x 1 root root 469040 2. Mai 03:09 /lib/libze_tracing_layer.so.1.9.4 lrwxrwxrwx 1 root root 27 2. Mai 03:09 /lib/libze_validation_layer.so -> libze_validation_layer.so.1 lrwxrwxrwx 1 root root 31 2. Mai 03:09 /lib/libze_validation_layer.so.1 -> libze_validation_layer.so.1.9.4 -rwxr-xr-x 1 root root 63296 2. Mai 03:09 /lib/libze_validation_layer.so.1.9.4
This was the Output of
ls -l /lib/libigc*
:lrwxrwxrwx 1 root root 11 2. Mai 02:59 /lib/libigc.so -> libigc.so.1 lrwxrwxrwx 1 root root 15 2. Mai 02:59 /lib/libigc.so.1 -> libigc.so.1.0.1 -rwxr-xr-x 1 root root 35792864 2. Mai 02:59 /lib/libigc.so.1.0.1
I hope it helps.
IGC version isn't correctly reported in this build of IGC but if you're using latest arch packages:
https://github.com/archlinux/svntogit-community/blob/packages/intel-graphics-compiler/trunk/PKGBUILD it's IGC 1.0.13463.18 and compute-runtime 23.09.25812.14 which I'd expect to work.
blender.org builds are tested with IGC 1.0.13064.7 and compute-runtime 23.05.25593.18 at the moment.
I'll try next week to reproduce with your slightly newer versions.
I've compiled with same version of IGC and compute-runtime without reproducing the issue, crash may be coming from somewhere else.
From your logs I've noticed you were using GPU+CPU, I've fixed it recently as there were rendering corruptions
23de320878
but prior behavior shouldn't have been an invalid free().Sorry to ask you for more but as I can't reproduce the issue, I need you to do a few additional debugging steps using latest beta and GPU only:
gdb
andbt
gives it.I have run the
embree-vars.sh
script from embree 4.1 and rebooted: Blender crashes with the same error.after that I have run the script a second time and start Blender. Now it is the following error:
/home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/blender: symbol lookup error: /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/blender: undefined symbol: rtcIsSYCLDeviceSupported
sadly it doesn't write a blender.crash.txt. The only thing I have found is from
systemd-coredum
:Process 2891 (blender) of user 1000 dumped core. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_serialization.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_serialization.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_locale.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_locale.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_atomic.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_wave.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_wave.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_regex.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_regex.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libcycles_kernel_oneapi_aot.so without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libcycles_kernel_oneapi_aot.so Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_chrono.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_chrono.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_system.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_iostreams.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_iostreams.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_date_time.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_filesystem.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_filesystem.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libembree4.so.4 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libembree4.so.4 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_thread.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_thread.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_python310.so.1.80.0 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libboost_python310.so.1.80.0 Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libtbb.so.2 without build-id. Module /home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libtbb.so.2 Stack trace of thread 3003: #0 0x00007fa81b49f26c n/a (libc.so.6 + 0x8926c) #1 0x00007fa81b44fa08 raise (libc.so.6 + 0x39a08) #2 0x00007fa81b4385c2 abort (libc.so.6 + 0x225c2) #3 0x00007fa81b4392db n/a (libc.so.6 + 0x232db) #4 0x00007fa81b4a91b7 n/a (libc.so.6 + 0x931b7) #5 0x00007fa81b4ab048 n/a (libc.so.6 + 0x95048) #6 0x00007fa81b4adcb3 __libc_free (libc.so.6 + 0x97cb3) #7 0x00007fa7ba02e23a n/a (libigc.so.1 + 0x22e23a) #8 0x00007fa7ba7f507d n/a (libigc.so.1 + 0x9f507d) #9 0x00007fa7ba03220f n/a (libigc.so.1 + 0x23220f) #10 0x00007fa7ba1271c4 n/a (libigc.so.1 + 0x3271c4) #11 0x00007fa7cce7dd51 n/a (libze_intel_gpu.so.1 + 0x47dd51) #12 0x00007fa7ccbd1de7 n/a (libze_intel_gpu.so.1 + 0x1d1de7) #13 0x00007fa7ccbd45fd n/a (libze_intel_gpu.so.1 + 0x1d45fd) #14 0x00007fa7ccbd9c09 n/a (libze_intel_gpu.so.1 + 0x1d9c09) #15 0x00007fa7ccbdacca n/a (libze_intel_gpu.so.1 + 0x1dacca) #16 0x00007fa7ccbae535 n/a (libze_intel_gpu.so.1 + 0x1ae535) #17 0x00007fa7f2f94aa7 n/a (/home/ephraim/Downloads/blender-3.6.0-beta+v36.2533576802f5-linux.x86_64-release/lib/libpi_level_zero.so + 0x27aa7) ELF object binary architecture: AMD x86-64
maybe that helps:)
Ah I haven't been clear with the first step.
embree_pathtracer
is a binary for a embree tutorial that's precompiled, that'd be good to know if you can actually run it standalone on your machine, not to try its libs with Blender.The stack trace is useful as it points to libigc.so (https://github.com/intel/intel-graphics-compiler) triggering the crash.
Latest archlinux IGC update (
8a8f13692e
) dropped a patch that was needed (d6bb93ee1a
) for the specific spirv-llvm-translator version that was used in archlinux, maybe it's related, I'll reproduce with archlinux exact stack and not just same IGC version.Hi, a new version of IGC (https://archlinux.org/packages/extra-testing/x86_64/intel-graphics-compiler/ - 1.0.13700.14) and compute-runtime (https://archlinux.org/packages/extra-testing/x86_64/intel-compute-runtime/ - 23.13.26032.30-1) got pushed to extra-testing archlinux repository a few minutes ago, with the missing patch included, can you give it a try on your system?
Sadly I'm not at my PC in the next two or three weeks because I'm not at home, so I can't test it.
I try it as fast as I can.
Sorry for it.
No worries - we've been able to test with archlinux maintainers and they could reproduce the crash with test versions as well, it's a new issue.
It's by no means specific to Blender though, it crashes during the graphics-compiler static vars initialization. Investigation is ongoing.
I'm now back at my PC and using the new versions of ICG and compute-runtime.
Sadly it is still the same error (
free(): invalid pointer
), but now it crashes, if i select the cycles gpu rendering or start the preferences (really weird).I hope this infos are usable.:)
I am using Arch-based, and there was some Intel-related updates today. I installed them and rebooted the PC. Now even Blender 3.5 crashes at the start-up. I guess this is because OneAPI has been enabled. I downloaded Blender 3.6 RC, cleared settings for 3.6, opened Preferences, clicked System, and if I click the OneAPI tab, Blender crashes. with that
message. Below are the package versions
A770 here.
I am also encountering crashes with the 3.6 stable release.
I'm my case, I'm using live view and tweaking materials (subsurface and reflection) on a new project. I love the performance, if only it wouldn't crash so much. Event viewer said the fault was with ze_intel_gpu64.dll.
Using:
Blender 3.6 (stable)
Intel A770 driver v31.0.101.4502
Win11 pro
Archlinux package maintainers were able to have everything working yesterday with new packages of intel-graphics-compiler, intel-compute-runtime, intel-opencl-clang based on LLVM 14 instead of 15, please wait a bit more so they become available.
@Slippy a fault in ze_intel_gpu64.dll is a different problem as the one investigated here, I believe #109282 is what you're facing, please subsribe here to get updates or open a new bug if you are running into something else.
Intel Embree HWRT ends in Crashto oneAPI device fails at load or kernels compilation on ArchlinuxCan't wait until this actually materializes somewhere so I can install it.
I was getting tired of waiting for Arch updates, so today, I installed Windows 10 on a spare SSD on the same computer. I updated Windows and then installed the latest Arc driver and Blender 3.6. I started Blender 3.6 afresh without importing preferences and changed only the System -> Cycles device and the render sample/denoise settings.
Blender crashed multiple times randomly when I tried to resize a plane (the viewport only had one cube and one plane) in rendered view mode. Obviously it is not the same as #109282 Maybe Blender 3.6 isn't just stable with Arc A770 on Windows.
I've tried to reproduce and it worked fine on my A750 with driver 101.4369 - if the crash signature is different than in #109282, can you open a new bug so we have a chance to fix it?
My driver version is 31.0.101.4502. I checked blender.crash, and it seems EXCEPTION_ACCESS_VIOLATION in embree4.dll.
I created a new issue: #109771
I am sorry to ask this, but do you happen to know when those maintainers are going to release those "everything working" packages? I thought it would only take a few days, but there have been no updates and I cannot use GPU rendering at all. It would be less tiring if I knew how long I have to wait.
The requisite intel-opencl-clang based on clang14 and llvm-libs14, and the newly packaged spirv-llvm-translator14, are both in Extra-Staging as of the 14th. I have no idea what else is needed for them to be moved to Extra-Testing.
Edit: I just tested these two packages, and while they do seem to eliminate the crashing, they do not present the Embree checkbox, so something is still off.
Edit 2: Yup, looks like they have at least a week or two of porting yet more packages over to llvm14 before this will work.
Well, if it is just a few more weeks, I think I can wait. Besides, as far as I know, there are at least two Blender-crashing bugs in Embree under Windows (the 3-minute crash, the static variable release crash), so assuming that Linux Embree is using the same code and has the same problems as Windows, Embree should be unusable under Linux at the moment anyway.
The 3-minute crash is specific to the driver in a Windows only path. The static var crash is from embree, cross-OS but kind of easy to avoid until the fix gets into a release.
I have now switched to the extra-testing repo, so i use intel-compute-runtime 23.22.26516.18-3, intel-graphics-compiler 1:1.0.14062.11-3 and blender 3.6.2. Now it works!
Thanks for working on this!
I think this issue can now be closed?
Thanks for the confirmation !
If everything works, why don't they release it and kept it for "testing" for two months? I don't want to switch to "testing", because it probably will mess up the system.
various things got tested over the last two months, it hasn't been working for that long :) last update is from a week ago. If you want to switch early I believe you can add the testing repos with low priority and only pull what you want, but it shouldn't take much longer now.
Well, I checked updates today, and all Intel/Blender stuff were just updated, so I installed them all. Blender doesn't crash anymore, but I am not sure if the rendering problem is actually fixed.
Under Linux, unlike under Windows, I do not see the "Embree on GPU" checkbox. Embree is installed on my Linux PC, of course. I ran Blender benchmark and it seems that I get about 130 points less score under Linux than under Windows, on the same computer.
Is it like that the Arch developers solved this issue by disabling Blender 3.6's new feature, hardware-accelerated ray tracing for Intel A770?
Embree on Archlinux repo hasn't yet GPU support built, so that's expected that the Blender build from arch lacks the checkbox, it's not directly related.
The issue here in this thread is caused by the failure of JIT compilation from the graphics driver, which is required when using a Blender build with Embree on GPU. Now it's fixed, you should be able to use the build from blender.org, that has the checkbox.
The blender benchmark app uses blender.org build and worked, you'd get a bigger gap than 130 points if embree on GPU wasn't used.
3.6.2 from blender.org is still crashing for me on some scenes when building the kernels. On one of these scenes, it takes over 20 minutes of churning away before it finally decides to crash. Still sitting on the kernel build step.
3.6.3-RC from the blender-3.6-bin daily package can't even render any simpler scenes without crashing on that kernel build step.
https://gist.github.com/kode54/af85ddff1318082bcffd84c7e49ae767
Do I still need packages from testing at this point?
I downloaded that 3.6.3 RC version and rendered some files (e.g.,
blender_35_splash_nicole_morena.blend
) and they were rendered without crashing.I can also confirm much increased stability with Blender v3.6.3 RC using Intel driver v4644.
😁❤️ 😍