- Fixed incorrect detection of AVX-512 on macOS (sometimes causing a crash)
- Fixed inconsistent performance and costly initialization for AVX-512
- Fixed JIT’ed AVX-512 kernels not showing up correctly in VTune
The first bullet point might fix the issue report in T79929. I don’t have the necessary hardware to test it.
Looking at OIDN’s commit log, the change was in mkl-dnn. OIDN uses it’s own fork of mkl-dnn. I checked to see if the same fix is in (the renamed) oneAPI oneDNN. It isn’t.
Do any of the other Intel libraries Blender uses aka tbb or embree have the math kernel library - deep neural network as a dependency?
If so, does this explain the MacOS 2.90 crashes in T78793 and T79707? If the fix was added upstream, can tbb (assuming it is the cause) even be upgraded do to Blender aligning to the VFX platform?