Closed demarle closed 2 years ago
OIDN v0.7 is a pre-release alpha version, which should not be used in production code. All public releases (v0.8 and above) check for SSE4.1 and should not cause a crash when it’s not available. ParaView should upgrade to the latest version of OIDN.
Apologies, my mistake ParaView is on 0.8.1, which was the most recent version at the time we integrated it.
So my question is then do you know of any relevant changes after 0.8.1? Thanks.
I’m not aware of any relevant changes after 0.8.1. But I’m a bit confused about the issue. I looked up the ParaView discussion and it seems the machine where it crashes even has AVX, so SSE4.1 is fully supported. Or did it crash on a different machine? Looking at the gdb output, it seems to crash in an AVX-512 function, which should run only if AVX-512 is detected. I haven’t seen any false positive detections of AVX-512 so far, so this looks very strange. Could you please provide more info for reproducing the issue (e.g. machine specs)? Thanks!
Thanks. I'll dig in a bit then and pass along more information as I get it.
There seems to be a crash somewhere in library initialization after we improperly fail to detect SSE4.1 ISA. This was noticed with the OIDN 0.7 packaged in ParaView 5.7.0. Is this a known issue that has been fixed in more recent releases by chance?
I am keeping track of this in the ParaView issue tracker here: https://gitlab.kitware.com/paraview/paraview/issues/19424 I have not yet done any serious investigation or reproduced it myself.
thanks