-
We have to figure out the best approach.
One approach is using our (to be implemented) MLIR backend (#4319) and possibly using an OpenMP MLIR dialect that AMD compilers might be able to offload.
…
-
Array ranges are not supported in some of the OpenMP and OpenACC clauses (e.g. in the data movement clauses) and this could be beneficial (and even mandatory for a C backend)
-
All checks we are doing now before emitting ISPC code are rather fragile. When redoing the code generation framework we should take care that for the various backends (ISPC, cuda, openacc, ...) there …
-
First steps:
- Nvidia GPUs support by writing a CUDA version of kernels packing and unpacking
- Add examples
-
-
### Summary
GCC is able to support OpenACC and OpenMP target offload on AMD GPUs starting from version 10.
https://gcc.gnu.org/wiki/OpenACC
https://gcc.gnu.org/wiki/Offloading
It would be nice i…
-
The overarching goal is to have `Kokkos::print_configuration` print something actually useful. At the moment it is clearly lacking important things like Kokkos commit hash, Desul bundled or external …
-
I followed the instructions here for building Flang:
`https://github.com/flang-compiler/flang/wiki/Building-Flang`
particular the OpenMP 4.5 version with the NVidia backend. However, if compile a …
-
`MDRangePolicy` has a `static_assert(rank > 1, "Use RangePolicy for rank 1")`, which hinders genericity. Christian and I agreed that constraint should be lifted
-
At some point we might want to input and/or output Python. It would be useful to have a straw-person implementation to check that there are no major issues with this.