Christian Trott
Christian Trott
They were right after I submitted. But I fixed it in under 2 minutes after creating the issue. BOUNDS_CHECK=OFF gives shorter compile time. Note all of these are compile times.
I added NVC++23.3 data. Looks ok too.
This doesn't make much sense. 5s and 10s is just something I can't see for this test. I executed on a local machine with everything coming from SSDs.
Hm are any of the kernels getting slower? Or is it more time on the host. And are we sure you get the right answer without leaking memory or anything...
Hm interesting, it should already be compile time unmanaged. We should check why this stuff is at all invoked with tracking.
Just to make that clear the ifpack two fix used there is definitely illegal, and may in the future break in debug mode. Its a precondition violation to construct a...
> @masterleinad Does it affect any of the > > ``` > #ifdef KOKKOS_COMPILER_NVHPC > ``` > > directives in the code, and should some/all be removed? There some in...
Retest this please!
We are starting to work on this for real. Couple things we need to change: * filenames/includes: currently we include files like "pack.hpp" right from the global include scope *...
I have to think about this a bit.