smyalygames/FiniteVolumeGPU
1
0
Fork 0
mirror of https://github.com/smyalygames/FiniteVolumeGPU.git synced 2025-05-18 06:24:13 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

More time counters and separation of experiments.

Browse Source
This commit is contained in:
Martin Lilleeng Sætra 2022-04-11 13:57:33 +00:00
parent 42aff3e095
commit dcef56a1b9
3 changed files with 170 additions and 50 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

176
Figures.ipynb
View File

File diff suppressed because one or more lines are too long

42
GPUSimulators/MPISimulator.py
View File

@ -204,6 +204,12 @@ class MPISimulator(Simulator.BaseSimulator):
self.profiling_data_mpi = { 'start': {}, 'end': {} }
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange"] = 0
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange"] = 0
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_download"] = 0
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_download"] = 0
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_upload"] = 0
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_upload"] = 0
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_sendreceive"] = 0
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_sendreceive"] = 0
self.profiling_data_mpi["start"]["t_step_mpi"] = 0
self.profiling_data_mpi["end"]["t_step_mpi"] = 0
self.profiling_data_mpi["n_time_steps"] = 0
@ -343,7 +349,7 @@ class MPISimulator(Simulator.BaseSimulator):
y1 = y0 + height
return [x0, x1, y0, y1]
def exchange(self):
def exchange(self):
####
# FIXME: This function can be optimized using persitent communications.
# Also by overlapping some of the communications north/south and east/west of GPU and intra-node
@ -355,6 +361,9 @@ class MPISimulator(Simulator.BaseSimulator):
####
#Download from the GPU
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_download"] += time.time()
if self.north is not None:
for k in range(self.nvars):
self.sim.u0[k].download(self.sim.stream, cpu_data=self.out_n[k,:,:], asynch=True, extent=self.read_n)
@ -364,6 +373,10 @@ class MPISimulator(Simulator.BaseSimulator):
self.sim.stream.synchronize()
#Send/receive to north/south neighbours
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_download"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
comm_send = []
comm_recv = []
if self.north is not None:
@ -378,6 +391,10 @@ class MPISimulator(Simulator.BaseSimulator):
comm.wait()
#Upload to the GPU
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_upload"] += time.time()
if self.north is not None:
for k in range(self.nvars):
self.sim.u0[k].upload(self.sim.stream, self.in_n[k,:,:], extent=self.write_n)
@ -386,6 +403,10 @@ class MPISimulator(Simulator.BaseSimulator):
self.sim.u0[k].upload(self.sim.stream, self.in_s[k,:,:], extent=self.write_s)
#Wait for sending to complete
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_upload"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
for comm in comm_send:
comm.wait()
@ -396,6 +417,10 @@ class MPISimulator(Simulator.BaseSimulator):
####
#Download from the GPU
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_download"] += time.time()
if self.east is not None:
for k in range(self.nvars):
self.sim.u0[k].download(self.sim.stream, cpu_data=self.out_e[k,:,:], asynch=True, extent=self.read_e)
@ -405,6 +430,10 @@ class MPISimulator(Simulator.BaseSimulator):
self.sim.stream.synchronize()
#Send/receive to east/west neighbours
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_download"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
comm_send = []
comm_recv = []
if self.east is not None:
@ -420,6 +449,10 @@ class MPISimulator(Simulator.BaseSimulator):
comm.wait()
#Upload to the GPU
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_upload"] += time.time()
if self.east is not None:
for k in range(self.nvars):
self.sim.u0[k].upload(self.sim.stream, self.in_e[k,:,:], extent=self.write_e)
@ -428,6 +461,13 @@ class MPISimulator(Simulator.BaseSimulator):
self.sim.u0[k].upload(self.sim.stream, self.in_w[k,:,:], extent=self.write_w)
#Wait for sending to complete
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_upload"] += time.time()
self.profiling_data_mpi["start"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()
for comm in comm_send:
comm.wait()
if self.profiling_data_mpi["n_time_steps"] > 0:
self.profiling_data_mpi["end"]["t_step_mpi_halo_exchange_sendreceive"] += time.time()

2
seymour_strong_scaling_benchmark.sh
View File

@ -36,4 +36,4 @@ mpiexec -n 8 python mpiTesting.py -nx 4096 -ny 512 --profile &&
mkdir -p output_seymour/$NOW/8_proc &&
mv *.log output_seymour/$NOW/8_proc/ && mv *.nc output_seymour/$NOW/8_proc/ &&
for filename in *.json; do mv "$filename" "MPI_${NOW}_${filename#????}"; done;
for filename in *.json; do mv "$filename" "output_seymour/$NOW/MPI_${NOW}_${filename#????}"; done;