ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Adding Kokkos warning

Browse Source 
git-svn-id: svn://svn.icms.temple.edu/lammps-ro/trunk@15505 f3b2605a-c512-4ea7-a41b-209d697bcdaa
This commit is contained in:
stamoor
2016-08-26 18:17:16 +00:00
parent 8a7fecbd91
commit 56628fe2b6
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/finish.cpp
View File

@ -33,6 +33,7 @@
#include "neigh_request.h" #include "neigh_request.h"
#include "output.h" #include "output.h"
#include "memory.h" #include "memory.h"
#include "error.h"
#ifdef LMP_USER_OMP #ifdef LMP_USER_OMP
#include "modify.h" #include "modify.h"
@ -515,6 +516,13 @@ void Finish::end(int flag)
} }
#endif #endif
if (lmp->kokkos && lmp->kokkos->ngpu > 0)
if (const char* env_clb = std::getenv("CUDA_LAUNCH_BLOCKING"))
if (!(strcmp(env_clb,"1") == 0)) {
error->warning(FLERR,"Timing breakdown may not be accurate since GPU/CPU overlap is enabled. "
"Using 'export CUDA_LAUNCH_BLOCKING=1' will give an accurate timing breakdown but will reduce performance");
}
// FFT timing statistics // FFT timing statistics
// time3d,time1d = total time during run for 3d and 1d FFTs // time3d,time1d = total time during run for 3d and 1d FFTs
// loop on timing() until nsample FFTs require at least 1.0 CPU sec // loop on timing() until nsample FFTs require at least 1.0 CPU sec