2 Commits

Author SHA1 Message Date
Anthony Berg
e954b5b181 fix: missing bracket around multiple variables 2025-03-26 14:21:44 +01:00
Anthony Berg
277a6b4a3c fix: deprecated modules on LUMI 2025-03-26 14:20:49 +01:00
6 changed files with 82 additions and 91 deletions

View File

@@ -35,8 +35,6 @@ import gc
import netCDF4 import netCDF4
import json import json
from tqdm import tqdm
#import pycuda.compiler as cuda_compiler #import pycuda.compiler as cuda_compiler
#import pycuda.gpuarray #import pycuda.gpuarray
#import pycuda.driver as cuda #import pycuda.driver as cuda
@@ -180,11 +178,11 @@ def runSimulation(simulator, simulator_args, outfile, save_times, save_var_names
profiling_data_sim_runner["end"]["t_sim_init"] = time.time() profiling_data_sim_runner["end"]["t_sim_init"] = time.time()
#Start simulation loop #Start simulation loop
# progress_printer = ProgressPrinter(save_times[-1], print_every=10) progress_printer = ProgressPrinter(save_times[-1], print_every=10)
for k, t_step in tqdm(enumerate(t_steps), desc="Simulation Loop"): for k in range(len(save_times)):
#Get target time and step size there #Get target time and step size there
# t_step = t_steps[k] t_step = t_steps[k]
# t_end = save_times[k] t_end = save_times[k]
#Sanity check simulator #Sanity check simulator
try: try:
@@ -196,7 +194,7 @@ def runSimulation(simulator, simulator_args, outfile, save_times, save_var_names
profiling_data_sim_runner["start"]["t_full_step"] += time.time() profiling_data_sim_runner["start"]["t_full_step"] += time.time()
#Simulate #Simulate
if t_step > 0.0: if (t_step > 0.0):
sim.simulate(t_step, dt) sim.simulate(t_step, dt)
profiling_data_sim_runner["end"]["t_full_step"] += time.time() profiling_data_sim_runner["end"]["t_full_step"] += time.time()
@@ -213,11 +211,11 @@ def runSimulation(simulator, simulator_args, outfile, save_times, save_var_names
profiling_data_sim_runner["end"]["t_nc_write"] += time.time() profiling_data_sim_runner["end"]["t_nc_write"] += time.time()
#Write progress to screen #Write progress to screen
# print_string = progress_printer.getPrintString(t_end) print_string = progress_printer.getPrintString(t_end)
# if (print_string): if (print_string):
# logger.debug(print_string) logger.debug(print_string)
logger.debug("Simulated to t={:f} in {:d} timesteps (average dt={:f})".format(save_times[-1], sim.simSteps(), sim.simTime() / sim.simSteps())) logger.debug("Simulated to t={:f} in {:d} timesteps (average dt={:f})".format(t_end, sim.simSteps(), sim.simTime() / sim.simSteps()))
return outdata.filename, profiling_data_sim_runner, sim.profiling_data_mpi return outdata.filename, profiling_data_sim_runner, sim.profiling_data_mpi
#return outdata.filename #return outdata.filename
@@ -308,7 +306,7 @@ class IPEngine(object):
import ipyparallel import ipyparallel
self.cluster = ipyparallel.Client()#profile='mpi') self.cluster = ipyparallel.Client()#profile='mpi')
time.sleep(3) time.sleep(3)
while len(self.cluster.ids) != n_engines: while(len(self.cluster.ids) != n_engines):
time.sleep(0.5) time.sleep(0.5)
self.logger.info("Waiting for cluster...") self.logger.info("Waiting for cluster...")
self.cluster = ipyparallel.Client()#profile='mpi') self.cluster = ipyparallel.Client()#profile='mpi')
@@ -435,58 +433,58 @@ class DataDumper(object):
# class ProgressPrinter(object): class ProgressPrinter(object):
# """ """
# Small helper class for Small helper class for
# """ """
# def __init__(self, total_steps, print_every=5): def __init__(self, total_steps, print_every=5):
# self.logger = logging.getLogger(__name__) self.logger = logging.getLogger(__name__)
# self.start = time.time() self.start = time.time()
# self.total_steps = total_steps self.total_steps = total_steps
# self.print_every = print_every self.print_every = print_every
# self.next_print_time = self.print_every self.next_print_time = self.print_every
# self.last_step = 0 self.last_step = 0
# self.secs_per_iter = None self.secs_per_iter = None
# def getPrintString(self, step): def getPrintString(self, step):
# elapsed = time.time() - self.start elapsed = time.time() - self.start
# if (elapsed > self.next_print_time): if (elapsed > self.next_print_time):
# dt = elapsed - (self.next_print_time - self.print_every) dt = elapsed - (self.next_print_time - self.print_every)
# dsteps = step - self.last_step dsteps = step - self.last_step
# steps_remaining = self.total_steps - step steps_remaining = self.total_steps - step
# if (dsteps == 0): if (dsteps == 0):
# return return
# self.last_step = step self.last_step = step
# self.next_print_time = elapsed + self.print_every self.next_print_time = elapsed + self.print_every
# if not self.secs_per_iter: if not self.secs_per_iter:
# self.secs_per_iter = dt / dsteps self.secs_per_iter = dt / dsteps
# self.secs_per_iter = 0.2*self.secs_per_iter + 0.8*(dt / dsteps) self.secs_per_iter = 0.2*self.secs_per_iter + 0.8*(dt / dsteps)
# remaining_time = steps_remaining * self.secs_per_iter remaining_time = steps_remaining * self.secs_per_iter
# return "{:s}. Total: {:s}, elapsed: {:s}, remaining: {:s}".format( return "{:s}. Total: {:s}, elapsed: {:s}, remaining: {:s}".format(
# ProgressPrinter.progressBar(step, self.total_steps), ProgressPrinter.progressBar(step, self.total_steps),
# ProgressPrinter.timeString(elapsed + remaining_time), ProgressPrinter.timeString(elapsed + remaining_time),
# ProgressPrinter.timeString(elapsed), ProgressPrinter.timeString(elapsed),
# ProgressPrinter.timeString(remaining_time)) ProgressPrinter.timeString(remaining_time))
# def timeString(seconds): def timeString(seconds):
# seconds = int(max(seconds, 1)) seconds = int(max(seconds, 1))
# minutes, seconds = divmod(seconds, 60) minutes, seconds = divmod(seconds, 60)
# hours, minutes = divmod(minutes, 60) hours, minutes = divmod(minutes, 60)
# periods = [('h', hours), ('m', minutes), ('s', seconds)] periods = [('h', hours), ('m', minutes), ('s', seconds)]
# time_string = ' '.join('{}{}'.format(value, name) time_string = ' '.join('{}{}'.format(value, name)
# for name, value in periods for name, value in periods
# if value) if value)
# return time_string return time_string
# def progressBar(step, total_steps, width=30): def progressBar(step, total_steps, width=30):
# progress = np.round(width * step / total_steps).astype(np.int32) progress = np.round(width * step / total_steps).astype(np.int32)
# progressbar = "0% [" + "#"*(progress) + "="*(width-progress) + "] 100%" progressbar = "0% [" + "#"*(progress) + "="*(width-progress) + "] 100%"
# return progressbar return progressbar
""" """

View File

@@ -25,7 +25,6 @@ import numpy as np
import math import math
import logging import logging
from enum import IntEnum from enum import IntEnum
from tqdm import tqdm
#import pycuda.compiler as cuda_compiler #import pycuda.compiler as cuda_compiler
#import pycuda.gpuarray #import pycuda.gpuarray
@@ -157,7 +156,7 @@ class BaseSimulator(object):
self.num_substeps = num_substeps self.num_substeps = num_substeps
#Handle autotuning block size #Handle autotuning block size
if self.context.autotuner: if (self.context.autotuner):
peak_configuration = self.context.autotuner.get_peak_performance(self.__class__) peak_configuration = self.context.autotuner.get_peak_performance(self.__class__)
block_width = int(peak_configuration["block_width"]) block_width = int(peak_configuration["block_width"])
block_height = int(peak_configuration["block_height"]) block_height = int(peak_configuration["block_height"])
@@ -196,45 +195,42 @@ class BaseSimulator(object):
Requires that the step() function is implemented in the subclasses Requires that the step() function is implemented in the subclasses
""" """
# printer = Common.ProgressPrinter(t) printer = Common.ProgressPrinter(t)
t_start = self.simTime() t_start = self.simTime()
t_end = t_start + t t_end = t_start + t
update_dt = True update_dt = True
if dt is not None: if (dt is not None):
update_dt = False update_dt = False
self.dt = dt self.dt = dt
for _ in tqdm(range(math.ceil((t_end - t_start) / self.dt)), desc="Simulation"): while(self.simTime() < t_end):
# Update dt every 100 timesteps and cross your fingers it works # Update dt every 100 timesteps and cross your fingers it works
# for the next 100 # for the next 100
# TODO this is probably broken now after fixing the "infinite" loop if (update_dt and (self.simSteps() % 100 == 0)):
if update_dt and (self.simSteps() % 100 == 0):
self.dt = self.computeDt()*self.cfl_scale self.dt = self.computeDt()*self.cfl_scale
# Compute timestep for "this" iteration (i.e., shorten last timestep) # Compute timestep for "this" iteration (i.e., shorten last timestep)
current_dt = np.float32(min(self.dt, t_end-self.simTime())) current_dt = np.float32(min(self.dt, t_end-self.simTime()))
# Stop if end reached (should not happen) # Stop if end reached (should not happen)
if current_dt <= 0.0: if (current_dt <= 0.0):
self.logger.warning("Timestep size {:d} is less than or equal to zero!".format(self.simSteps())) self.logger.warning("Timestep size {:d} is less than or equal to zero!".format(self.simSteps()))
break break
# Step forward in time # Step forward in time
self.step(current_dt) self.step(current_dt)
#Print info #Print info
# print_string = printer.getPrintString(self.simTime() - t_start) print_string = printer.getPrintString(self.simTime() - t_start)
# if (print_string): if (print_string):
# self.logger.info("%s: %s", self, print_string) self.logger.info("%s: %s", self, print_string)
# try: try:
# self.check() self.check()
# except AssertionError as e: except AssertionError as e:
# e.args += ("Step={:d}, time={:f}".format(self.simSteps(), self.simTime()),) e.args += ("Step={:d}, time={:f}".format(self.simSteps(), self.simTime()),)
# raise raise
print("Done")
def step(self, dt): def step(self, dt):

View File

@@ -22,7 +22,7 @@ Myapplication=${Mydir}/FiniteVolumeGPU_hip/mpiTesting.py
ml LUMI/24.03 partition/G ml LUMI/24.03 partition/G
ml lumi-container-wrapper ml lumi-container-wrapper
ml cray-python/3.11.7 ml cray-python/3.11.7
ml rocm/6.2.2 ml rocm/5.4.6
ml craype-accel-amd-gfx90a ml craype-accel-amd-gfx90a
ml cray-mpich/8.1.29 ml cray-mpich/8.1.29

View File

@@ -13,9 +13,8 @@ dependencies:
- pytools - pytools
- netcdf4 - netcdf4
- scipy - scipy
- tqdm
- pip: - pip:
- hip-python==6.2.0.499.16 - hip-python==5.4.3.470.16
- -i https://test.pypi.org/simple/ - -i https://test.pypi.org/simple/

View File

@@ -70,7 +70,7 @@ def hip_check(call_result):
args = parser.parse_args() args = parser.parse_args()
if args.profile: if(args.profile):
profiling_data = {} profiling_data = {}
# profiling: total run time # profiling: total run time
t_total_start = time.time() t_total_start = time.time()
@@ -79,8 +79,6 @@ if args.profile:
# Get MPI COMM to use # Get MPI COMM to use
comm = MPI.COMM_WORLD comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()
#### ####
@@ -88,7 +86,7 @@ rank = comm.Get_rank()
#### ####
log_level_console = 20 log_level_console = 20
log_level_file = 10 log_level_file = 10
log_filename = 'mpi_' + str(rank) + '.log' log_filename = 'mpi_' + str(comm.rank) + '.log'
logger = logging.getLogger('GPUSimulators') logger = logging.getLogger('GPUSimulators')
logger.setLevel(min(log_level_console, log_level_file)) logger.setLevel(min(log_level_console, log_level_file))
@@ -112,7 +110,7 @@ logger.info("File logger using level %s to %s",
# Initialize MPI grid etc # Initialize MPI grid etc
#### ####
logger.info("Creating MPI grid") logger.info("Creating MPI grid")
grid = MPISimulator.MPIGrid(comm) grid = MPISimulator.MPIGrid(MPI.COMM_WORLD)
""" """
job_id = int(os.environ["SLURM_JOB_ID"]) job_id = int(os.environ["SLURM_JOB_ID"])
@@ -154,7 +152,7 @@ gamma = 1.4
#save_times = np.linspace(0, 0.000099, 11) #save_times = np.linspace(0, 0.000099, 11)
#save_times = np.linspace(0, 0.000099, 2) #save_times = np.linspace(0, 0.000099, 2)
save_times = np.linspace(0, 0.0000999, 2) save_times = np.linspace(0, 0.0000999, 2)
outfile = "mpi_out_" + str(rank) + ".nc" outfile = "mpi_out_" + str(MPI.COMM_WORLD.rank) + ".nc"
save_var_names = ['rho', 'rho_u', 'rho_v', 'E'] save_var_names = ['rho', 'rho_u', 'rho_v', 'E']
arguments = IC.genKelvinHelmholtz(nx, ny, gamma, grid=grid) arguments = IC.genKelvinHelmholtz(nx, ny, gamma, grid=grid)
@@ -162,7 +160,7 @@ arguments['context'] = cuda_context
arguments['theta'] = 1.2 arguments['theta'] = 1.2
arguments['grid'] = grid arguments['grid'] = grid
if args.profile: if(args.profile):
t_init_end = time.time() t_init_end = time.time()
t_init = t_init_end - t_init_start t_init = t_init_end - t_init_start
profiling_data["t_init"] = t_init profiling_data["t_init"] = t_init
@@ -183,14 +181,14 @@ def genSim(grid, **kwargs):
(outfile, sim_runner_profiling_data, sim_profiling_data) = Common.runSimulation( (outfile, sim_runner_profiling_data, sim_profiling_data) = Common.runSimulation(
genSim, arguments, outfile, save_times, save_var_names, dt) genSim, arguments, outfile, save_times, save_var_names, dt)
if args.profile: if(args.profile):
t_total_end = time.time() t_total_end = time.time()
t_total = t_total_end - t_total_start t_total = t_total_end - t_total_start
profiling_data["t_total"] = t_total profiling_data["t_total"] = t_total
print("Total run time on rank " + str(rank) + " is " + str(t_total) + " s") print("Total run time on rank " + str(MPI.COMM_WORLD.rank) + " is " + str(t_total) + " s")
# write profiling to json file # write profiling to json file
if args.profile and rank == 0: if(args.profile and MPI.COMM_WORLD.rank == 0):
job_id = "" job_id = ""
if "SLURM_JOB_ID" in os.environ: if "SLURM_JOB_ID" in os.environ:
job_id = int(os.environ["SLURM_JOB_ID"]) job_id = int(os.environ["SLURM_JOB_ID"])
@@ -201,7 +199,7 @@ if args.profile and rank == 0:
str(job_id) + "_" + str(allocated_nodes) + "_nodes_and_" + str(allocated_gpus) + "_GPUs_profiling.json" str(job_id) + "_" + str(allocated_nodes) + "_nodes_and_" + str(allocated_gpus) + "_GPUs_profiling.json"
profiling_data["outfile"] = outfile profiling_data["outfile"] = outfile
else: else:
profiling_file = "MPI_" + str(size) + "_procs_and_" + str(num_cuda_devices) + "_GPUs_profiling.json" profiling_file = "MPI_" + str(MPI.COMM_WORLD.size) + "_procs_and_" + str(num_cuda_devices) + "_GPUs_profiling.json"
for stage in sim_runner_profiling_data["start"].keys(): for stage in sim_runner_profiling_data["start"].keys():
profiling_data[stage] = sim_runner_profiling_data["end"][stage] - sim_runner_profiling_data["start"][stage] profiling_data[stage] = sim_runner_profiling_data["end"][stage] - sim_runner_profiling_data["start"][stage]
@@ -216,7 +214,7 @@ if args.profile and rank == 0:
profiling_data["slurm_job_id"] = job_id profiling_data["slurm_job_id"] = job_id
profiling_data["n_cuda_devices"] = str(num_cuda_devices) profiling_data["n_cuda_devices"] = str(num_cuda_devices)
profiling_data["n_processes"] = str(size) profiling_data["n_processes"] = str(MPI.COMM_WORLD.size)
profiling_data["git_hash"] = Common.getGitHash() profiling_data["git_hash"] = Common.getGitHash()
profiling_data["git_status"] = Common.getGitStatus() profiling_data["git_status"] = Common.getGitStatus()