Offload problems on Nvidia Jetson AGX Thor (sm_110)

[pauljurczak]

I tried the example from Using device ids in target regions on Nvidia Jetson AGX Thor (Blackwell architecture, sm_110) with Python 3.14.4 and pyomp 0.5.1. It produces this error:

'sm_110' is not a recognized processor for this target (ignoring processor)
'sm_110' is not a recognized processor for this target (ignoring processor)
'sm_110' is not a recognized processor for this target (ignoring processor)
'sm_110' is not a recognized processor for this target (ignoring processor)
'sm_110' is not a recognized processor for this target (ignoring processor)
'sm_110' is not a recognized processor for this target (ignoring processor)
LLVM ERROR: Cannot select: intrinsic %llvm.nvvm.barrier.sync

If this is not pyomp problem, where should I report it?

[ggeorgakoudis]

This is an upstream LLVM issue:

llvm/llvm-project#156096

Jetson AGX Thor is identified as sm_110 starting with CUDA 13. It was previously exposed as sm_101, and older LLVM versions do not recognize the new sm_110 target name.

We plan to update PyOMP to work with newer Numba/llvmlite versions based on LLVM 22, which include sm_110 support and should resolve this issue..