Improved the console and log output layout and messages.
Change-Id: I0f81cbe9f74ada5a40b2da0c8b2d7dd405bfa7e5
}
else
{
- sprintf(sbuf, "%d GPU%s %sselected to be used for this run: ",
+ sprintf(sbuf, "%d GPU%s %sselected for this run: ",
ngpu, (ngpu > 1) ? "s" : "",
gpu_info->bUserSet ? "user-" : "auto-");
for (i = 0; i < ngpu; i++)
#endif
/* inform the user about the settings */
- if (SIMMASTER(cr) && bOMP)
+ if (bOMP)
{
#ifdef GMX_THREAD_MPI
const char *mpi_str="per tMPI thread";
/* for group scheme we print PME threads info only */
if (bFullOmpSupport)
{
- fprintf(stderr, "Using %d OpenMP thread%s %s\n",
- modth.gnth,modth.gnth > 1 ? "s" : "",
- cr->nnodes > 1 ? mpi_str : "");
+ md_print_info(cr, fplog, "Using %d OpenMP thread%s %s\n",
+ modth.gnth,modth.gnth > 1 ? "s" : "",
+ cr->nnodes > 1 ? mpi_str : "");
}
if (bSepPME && modth.gnth_pme != modth.gnth)
{
- fprintf(stderr, "Using %d OpenMP thread%s %s for PME\n",
- modth.gnth_pme,modth.gnth_pme > 1 ? "s" : "",
- cr->nnodes > 1 ? mpi_str : "");
+ md_print_info(cr, fplog, "Using %d OpenMP thread%s %s for PME\n",
+ modth.gnth_pme,modth.gnth_pme > 1 ? "s" : "",
+ cr->nnodes > 1 ? mpi_str : "");
}
}
mda->deviceOptions=deviceOptions;
mda->Flags=Flags;
- fprintf(stderr, "Starting %d tMPI threads\n",hw_opt->nthreads_tmpi);
- fflush(stderr);
/* now spawn new threads that start mdrunner_start_fn(), while
the main thread returns */
ret=tMPI_Init_fn(TRUE, hw_opt->nthreads_tmpi,
cr->nnodes==1 ? "process" : "processes"
#endif
);
+ fflush(stderr);
#endif
gmx_omp_nthreads_init(fplog, cr,
if (lossf >= DD_PERF_LOSS)
{
sprintf(buf,
- "NOTE: %.1f %% performance was lost due to load imbalance\n"
+ "NOTE: %.1f %% of the available CPU time was lost due to load imbalance\n"
" in the domain decomposition.\n",lossf*100);
if (!comm->bDynLoadBal)
{
fprintf(fplog, "%s\n", hline);
gpu_cpu_ratio = tot_gpu/tot_cpu_overlap;
- fprintf(fplog, "\n Force evaluation time GPU/CPU: %.3f ms/%.3f ms = %.3f\n",
+ fprintf(fplog, "\nForce evaluation time GPU/CPU: %.3f ms/%.3f ms = %.3f\n",
tot_gpu/gpu_t->nb_c, tot_cpu_overlap/wc->wcc[ewcFORCE].n,
gpu_cpu_ratio);
* but we currently can't check that here.
*/
md_print_warn(NULL,fplog,
- "NOTE: The GPU has >25%% less load than the CPU. This imbalance causes\n"
+ "\nNOTE: The GPU has >25%% less load than the CPU. This imbalance causes\n"
" performance loss. Maybe the domain decomposition limits the PME tuning.\n"
- " In that case, try setting the DD grid manually (-dd) or lowering -dds.\n");
+ " In that case, try setting the DD grid manually (-dd) or lowering -dds.");
}
else
{
* too small for increasing the cut-off for PME tuning.
*/
md_print_warn(NULL,fplog,
- "NOTE: The GPU has >25%% less load than the CPU. This imbalance causes\n"
- " performance loss.\n");
+ "\nNOTE: The GPU has >25%% less load than the CPU. This imbalance causes\n"
+ " performance loss.");
}
}
if (gpu_cpu_ratio > 1.2)
{
md_print_warn(NULL,fplog,
- "NOTE: The GPU has >20%% more load than the CPU. This imbalance causes\n"
- " performance loss, consider using a shorter cut-off and a finer PME grid.\n");
+ "\nNOTE: The GPU has >20%% more load than the CPU. This imbalance causes\n"
+ " performance loss, consider using a shorter cut-off and a finer PME grid.");
}
}
}