On Sat, Apr 19, 2025 at 1:33 PM Santiago Vila <sanv...@debian.org> wrote: > > El 19/4/25 a las 10:48, Roman Lebedev escribió: > > Does the bug reproduce if the whole build is rerun from scratch? > > I tried six times before reporting: > > Status: failed halide_19.0.0-6_amd64-20250418T100010.385Z > Status: failed halide_19.0.0-6_amd64-20250418T100011.126Z > Status: failed halide_19.0.0-6_amd64-20250418T100012.500Z > Status: failed halide_19.0.0-6_amd64-20250418T100014.400Z > Status: failed halide_19.0.0-6_amd64-20250418T222003.977Z > Status: failed halide_19.0.0-6_amd64-20250418T222004.030Z > > and each time it was made "from scratch". This is not a hand-made > build with dpkg-buildpackage, this was a mass rebuild using sbuild. > > > Looks like it is failing with SIGILL "illegal instruction", > > what is the CPU there? /proc/cpuinfo? > > In this case they are c7i and m7i machines from AWS, so they have > a 4th Generation Intel Xeon Scalable processor. > > The cpuinfo is attached. Right, so that CPU does claim to have AMX support. But is AMX supported in such a virtualized environment? https://community.intel.com/t5/Intel-Tiber-Developer-Cloud/Can-t-run-AMX-Instructions-on-Intel-Xeon-4th-Gen-Scalable/m-p/1532428 suggests not. Could you please file an issue with AWS? I really feel like the issue is on their end...
> (If you need a VM to reproduce, please contact me privately) > > Thanks. Thank you. Roman.
