Failed to compile concrete-ml models on Ubuntu about concrete-ml HOT 8 CLOSED

@weimingma an Illegal instruction could mean your CPU isn't exposing aes cpu instructions to the VM. Could you run lscpu or cpuid from the host and the vm and grep -i aes. If the instructions are available on the host and not on the VM, then you need to make them available through your hypervisor

@youben11 Followed your comments, I have confirmed that the compiling failure was due to lack of aes flag on the VM cpu. This issue is resolved after adding aes to my VM. I am closing this issue. Really apprecite your help~

from concrete-ml.

Hello again @weimingma , I discussed with my colleagues but we can't find an obvious explanation from the information we have now.

Could you please check your glibc version with ldd --version? Also we might need a full traceback if there is more information there and the hardware specifications of the system you are using to run the VM (any detail about your configuration would help 🙏🏼 ).

We never tested Concrete in a VM afaik so that would allow us to replicate your issue and in the future add a test for it in our CI.

@fd0r The ldd version is: ldd (Ubuntu GLIBC 2.31-0ubuntu9.14) 2.31. I am trying to get HW related info and fulltrace and will get back to you later. Thanks!

from concrete-ml.

Hello @weimingma , sorry to hear your troubles using Concrete ML.
When you say:

Ubuntu 22.04 (virtual environment)

Do you mean a VM or docker container? Also are on an ARM system or x86?

from concrete-ml.

Hello @weimingma , sorry to hear your troubles using Concrete ML. When you say:

Ubuntu 22.04 (virtual environment)

Do you mean a VM or docker container? Also are on an ARM system or x86?

Hi @fd0r , Appreciate for your prompt reply. It's a VM and x86. Let me know if need more info. Thanks.

from concrete-ml.

Hello again @weimingma , I discussed with my colleagues but we can't find an obvious explanation from the information we have now.

Could you please check your glibc version with ldd --version? Also we might need a full traceback if there is more information there and the hardware specifications of the system you are using to run the VM (any detail about your configuration would help 🙏🏼 ).

We never tested Concrete in a VM afaik so that would allow us to replicate your issue and in the future add a test for it in our CI.

from concrete-ml.

Alright so that's probably not an issue with glibc, could you link us or send directly the full traceback please? 🙏🏼

from concrete-ml.

Alright so that's probably not an issue with glibc, could you link us or send directly the full traceback please? 🙏🏼

Hi @fd0r I have tried several times in the case. Unfortunately, there is no Python traceback shown. There is only llvm logs shown after the model is compiled in my initial post. Also, the following shown at the end of the logs:

PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace. Illegal instruction (core dumped)

Please let me know if you were requesting any other logs.
BTW, FYI. the CPU info: CPU: 16 x AMD EPYC 7232P 8-Core Processor (1 Socket).
Thanks.

from concrete-ml.

@weimingma an Illegal instruction could mean your CPU isn't exposing aes cpu instructions to the VM. Could you run lscpu or cpuid from the host and the vm and grep -i aes. If the instructions are available on the host and not on the VM, then you need to make them available through your hypervisor

from concrete-ml.