-
-
Notifications
You must be signed in to change notification settings - Fork 5.6k
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
Build failure on Linux/ARM64 #39877
Comments
There seem to be two separate things going on here:
This is a segfault while trying to invalidate the cache. I'm a bit surprised by this. This function hardly touches memory. It's possible of course for the cache invalidation instructions to generate a trap, but I don't think they do on other Aarch64 platforms.
This looks like an issue with openblas's architecture detection code. |
Is this on Julia master or the release-1.6 branch? |
After doing:
v1.6.0 now builds without errors. Julia master (after the same thorough pre-build clean) now fails in a different way than described above:
I've tried this with GCC versions 7 and 8, both with the same result. Happy to try any other suggested environment changes. |
Perhaps worth pinging @vchuravy |
I just tried on an aarch64 box, and master builds fine. Of course, not all arms are the same, but it would be helpful to get some more information here about the arm chip. |
I'm doing this on an NVIDIA AGX Xavier:
Apparently because of the way that NVIDIA chose to deal with the Linux kernel device tree, this system is stuck at Ubuntu 18.04, with kernel 4.9.201-tegra. I have another AGX Xavier that I'm going to try to update, after which I can try the build with an up-to-date kernel. I tried splitting up the loading of the Pkg commands (initialize
|
Ok this looks like a two part issue:
|
The only recent change to that file is gcc-mirror/gcc@761e6bb, e.g. GCC 10.1/10.2 |
Maybe report it to NVIDIA if there was a way to, I suppose. |
This isn't only a build failure, but also manifests with our binaries of course:
Given how ARMv8 on Linux is Tier 1, maybe we should put this on the milestone? I can have a look at backporting and testing the necessary patches next week. About the |
I'm actually stuck on the cache invalidation segfault trying to fix the OpenBLAS-related issue (I don't remember it actually getting in the way of compiling Julia on this system). Anyway, the full back-trace is as follows:
So this looks like something with our JIT's memory manager. I'm entirely unfamiliar with this code, any pointers @Keno? |
On what instruction are you segfaulting? What CPU specifically is it? |
Running on Ubuntu 18.04, on an NVIDIA AGX Xavier. Some of this may be related to #39612, but I think there's more than that going on.
The text was updated successfully, but these errors were encountered: