Valentin Churavy

``` vchuravy@odin ~> pacman -Qe | grep amd amd-ucode 20250408.c1a774f3-1 amdmemorytweak-git 40.9a64ff1-1 amdvlk 2025.Q1.3-1 hip-runtime-amd 6.3.3-1 xf86-video-amdgpu 23.0.0-2 vchuravy@odin ~> pacman -Qe | grep roc python-zeroconf 0.146.1-1 rocm-hip-sdk 6.3.3-1 roctracer...

`UCX_ERROR_SIGNALS` is only applicable to MPI based libraries that use the UCX library. I agree that the error looks similar to that, I assume this is Julia 1.11? The backtrace...

@pchintalapudi I am confused that this didn't cause crashes.

IIUC this will trigger an error in user code when we merge this we do error checking before calling LLVM. Do we need a minor version? But this is a...

> or the general idea of a reduction abstraction to be added to KA.jl? That is still a goal of mine, but I am prioritizing getting the CPU backend to...

My strategy would be a debug build of Julia and then using `rr` to record a crashing execution and reverse follow the pointer to see what object is getting lost....

Do have a standalone reproducer? Likely something with a "simulated" low memory environment, sometimes spurious sprinkling of `GC.gc()` helps.

Only that https://github.com/JuliaLang/julia/issues/58475 is hopefully the same

> To clarify, I haven't been able to reproduce it locally yet. Without a reproducer, we are all just staring into the fog, trying to see the shape of the...

> I wonder if this is related to https://github.com/JuliaLang/julia/issues/57333 which was closed after being fixed in https://github.com/JuliaLang/julia/pull/57392 on Julia 1.12. Perhaps a backport be needed on Julia 1.11 Grasping a...