Fix #22, add_barrier before time + fix IO and compute time order #23
Conversation
|
Perhaps, we need to change also the output to show that we have found the bug? |
| os << "[RESULT] Compute: " << obj.time_spent_in_io.count() << " s" << std::endl; | ||
|
|
||
| os << "[RESULT] I/O: " << obj.time_spent_in_io.count() << " s" << std::endl; | ||
| os << "[RESULT] Compute: " << obj.time_spent_in_compute.count() << " s" << std::endl; |
There was a problem hiding this comment.
I suggest that we also output the wall time of the whole execution. Thus, we can see the difference between wall_time and io_time + compute_time, which gives us the time for initialization, etc.
There was a problem hiding this comment.
The Wall time correspond to the time spent in the solver (method EngineNd::run() without the time in the first pdiExposeData).
It doen't take into account the time of initialization of kokkos for example.
Here, Wall time \simeq I/O + Compute
Yes |
add kokkos::fence() to have better understanding of times