Nicolas Miller
Nicolas Miller
Maybe, we just ran into this for cuBLAS, I'll see if the other ones need it too
@bader this is a proof of concept for moving C++ library handling from libdevice code into headers. It allows us to remove the hack blocking LLVM intrinsic generation for standard...
> Interesting... I thought that clang only adds path to the clang headers at the beginning of the search paths list to make sure that clang wrapper header is included...
It turns out the C++ standard library also provides overloads that promote integers to double, or that promotes floating point types if they're mixed. To support that we've had to...
ping @intel/llvm-reviewers-runtime @intel/dpcpp-cfe-reviewers this is ready to review
> * A change to diagnose declarations that have both the sycl_device and sycl_device_only attributes as errors. > * A change to diagnose overload sets that contain both sycl_device and...
> > > It would be good to know if this would be interesting for non-AOT targets as well, there's a lot of logic in the driver to conditionally link...
No, there's no definite plans at the moment to support HIPRTC, we've had some discussions about it previously and it would be good to have but it's a significant amount...
Closing this after further discussions offline
What @Rbiessy linked is correct, upstream LLVM moved to the newer version of AMD device ABI, it now defaults to 6 which is only supported by ROCm 6.3+. You can...