Christian Trott
Christian Trott
Basically the infinite recursion should only happen if the specialization (i.e. the Kokkos variant) isn't more special than the generic one which is completely unconstrained on the ExecPolicy see https://godbolt.org/z/7x5hhne8o...
We need [] for compatibility for C++ simd types
I am wondering about the constructors with a subset of elements. Since this is specifically for permute vectors, maybe a non-member function to create a permute vector would be better?...
Here is the test code: ```c++ #include #include extern "C" void dgemm_(const char*, const char*, int*, int*, int*, double*, double*, int*, double*, int*, double*, double*, int*); template void gemm(CT C,...
So @bradking partly the problem is I think we add stuff via add_compile_options which has quotes in the flags we need to pass to the compiler, and something then went...
So we actually discussed this recently. CMAKE_CXX_FLAGS set during Kokkos install are NOT propagated to downstream folks right now (when using Kokkos outside of Trilinos). We considered telling users to...
Ah @bradking: what I just wrote there is essentially the recommended way of handling this?
we propagate all kinds of stuff. In particular a lot of architecture specific and backend specific flags. Like all the "compile this for GPU" flags, "use these optimization flags", "enable...
> IIUC, Kokkos's purpose is in part to provide these flags. Yes. I mean its impossible for normal users to figure out which flags are needed on all these different...
Most likely miniFE just needs some updating ...