Damien L-G comments

Results 491 comments of


                                            Damien L-G

Deprecate `CudaUVMSpace::available()`

Sounds good

`std::hash`, `operator<` and `operator==` for execution spaces

I think we should make execution spaces be regular (that is add `operator==`) I am not convinced we ~~need to~~ should support anything beyond that.

Try reenabling OpenMPTarget+ROCm CI

Really? You choose the moment where we are down to one machine with AMD GPUs to re-enable... On a side note it does not seem like we had an issue...

Nightly build failures with intel/19, "missing return statement at end of non-void function"

Same issue as #5426

Nightly build failures with intel/19, "missing return statement at end of non-void function"

https://godbolt.org/z/q7vE9EGYs

Kokkos allow HIP compilers that aren't called hipcc

It also means you will need to repeat the flags downstream

Missing Target for offloading after successful target offloading. The program will stop.

The compile line does not look right to compile OpenMPTarget code. How do you configure to build against your kokkos install?

Missing Target for offloading after successful target offloading. The program will stop.

The error message you posted is not from building Kokkos is it? From our perspective, it looks like the error is emitted from user code, the ``` In file included...

Missing Target for offloading after successful target offloading. The program will stop.

You are missing all the compile options that are needed to build against that Kokkos install. You can look in the `/usr/workspace/li50/kokkos-git/kokkos/install_lassen_omptarget/lib64/cmake/Kokkos` directory to see what these options might be.

Trilinos test failures in Tpetra with cuda/11.2 : `Error: Kokkos::finalize() has already been called.`

The error message is self-explanatory. The "it is not legal to call Kokkos::finalize() twice" is enforced since #5425 Are the Tpetra not able to handle it?