-
Notifications
You must be signed in to change notification settings - Fork 343
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
clinfo segfaults with rocm-2.7 #880
Comments
This is the output if I answer no to the gdb library breakpoint question:
|
Have you previously had any success with this device? It's a gfx802 based chip and not listed as supported. All of the supported GPUs are gfx803. Tonga is another gfx802 based device, this is known not to be supported and is listed as such. Unfortunately rocminfo and some other utilities still run with unsupported GPUs, creating a false impression that there is another problem elsewhere. |
Apologies, I thought Tonga was fully supported (without actually checking) Are there any plans to add support in future? I'm assuming the AMDGPU-Pro isn't open source, I've been testing out the new ebuilds in Gentoo and I'm not that keen on using blobs when I don't need to |
Would it be possible to have the driver spit out that there isn't a compatible device rather than segfaulting? |
The issue I linked had some discussion, the conclusion was that the devs had planned to add it but it got set back and other events have presumably overtaken it. I highly doubt it's going to be added now. The support for gfx803 is apparently no longer being actively developed and has become maintenance only.
I'm not an AMD developer but I would have thought so. From the looks of it the failure occurs when an attempt is made to compile the OpenCL kernel (which fails as the compiler has no support for the GPU arch). I don't know if anything could be actually be printed to stdout, it would depend on how the driver is implemented. |
And something similar with Raven :
Reading the doc you linked to, this should probably work but isn't officially supported, again I don't think clinfo should segfault |
And the rocminfo for that machine:
|
This is in the dmesg
The queue bit didn't happen on Tonga |
It seems that it depends on the motherboard configuration. The detailed (but now outdated) hardware compatibility page states this regarding APUs: You could try tweaking the settings or having a look in the manual for your motherboard.
Ideally not, but it seems that there is no infrastructure in place to catch these problems, I'm doubtful that AMD will do anything about this. |
Seems related to ROCm/ROCR-Runtime#68 ? |
Is this still an issue? If not, can we please close it? Thanks! |
Original ticket is more than a year old and the person that opened ticket originally has not responded to the latest request. If this is still an issue, please file a new ticket and we will happy to investigate it. Thanks! |
I think I still saw this issue the last time I tested, I'll check |
No sorry that was a different issue, clinfo is working fine |
Great! Thanks for checking. |
Running clinfo segfaults:
This is the output of rocminfo:
lspci -nn:
The text was updated successfully, but these errors were encountered: