|
Gromacs
2025.4
|
#include <gromacs/timing/include/gromacs/timing/gpu_timing.h>
Collaboration diagram for gmx_wallclock_gpu_nbnxn_t:GPU NB timings for kernels and H2d/D2H transfers.
Public Attributes | |
| gmx_kernel_timing_data_t | ktime [2][2] |
| table containing the timings of the four versions of the nonbonded kernels: force-only, force+energy, force+pruning, and force+energy+pruning | |
| gmx_kernel_timing_data_t | pruneTime |
| table containing the timings of the 1st pass prune-only kernels | |
| gmx_kernel_timing_data_t | dynamicPruneTime |
| table containing the timings of dynamic prune-only kernels | |
| double | nb_h2d_t |
| host to device transfer time in nb calculation | |
| double | nb_d2h_t |
| device to host transfer time in nb calculation | |
| int | nb_c |
| total call count of the nonbonded gpu operations | |
| double | pl_h2d_t |
| pair search step host to device transfer time | |
| int | pl_h2d_c |
| pair search step host to device transfer call count | |
1.8.5