[Openmp-dev] [RFC] Clarify the absence of API stability for the *device runtime* (aka. libomptarget-nvptx-sm_XX.bc)

[Jonas Hahnfeld via Openmp-dev]

Am Donnerstag, den 09.07.2020, 21:27 -0500 schrieb Ye Luo:
> > The host runtimes on the other side are linked dynamically and IMO it's
> > one of the strengths of libomp to mix-and-match versions. Unless
> > there's a compelling reason, I'll make another plea to keep
> > compatibility.
> 
>  
> I also value allowing mix-and-match versions a critical feature long term.
> If I wrote a library and compiled it based on one version while the application is compiled and linked by another version, we hope everything works without recompiling the library.
> Unfortunately, this mix-and-match scheme is quite brittle at the current stage. I still remember when I built libomptarget with CUDA9, CUDA 10.2 refused to link some device binary code.

I think you're once again mixing the different libraries: CUDA9 vs CUDA
10.2 makes no difference for the generic libomptarget host runtime, and
probably not even for the CUDA libomptarget plugin. It does for the
device runtime, which this thread was about, but my statement above is
about the host runtimes.

Jonas
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: This is a digitally signed message part
URL: <http://lists.llvm.org/pipermail/openmp-dev/attachments/20200710/783b2fee/attachment-0001.sig>