Created
May 8, 2019 23:01
-
-
Save syed-ahmed/4208c22c541f1d30ad6a9b1efc1d728f to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os, argparse, json, random, time | |
import torch | |
def int_list(s): | |
return [int(x) for x in s.split(',')] | |
parser = argparse.ArgumentParser() | |
parser.add_argument('--S', type=int_list, | |
default=[10, 100, 1000, 10000, 100000, 1000000, 10000000, 100000000]) | |
parser.add_argument('--num_trials', type=int, default=5) | |
parser.add_argument('--stats_json', default='stats.json') | |
dtype_list = [torch.float, torch.double, torch.half] | |
def main(args): | |
all_results = [] | |
for t, T in enumerate(dtype_list): | |
print('Running dtype = %s (value %d / %d)' % (str(T), t + 1, len(dtype_list))) | |
for s, S in enumerate(args.S): | |
print(' Running S = %d (value %d / %d)' % (S, s + 1, len(args.S))) | |
cur_results = { | |
'dtype': str(T), 'S': S, | |
'uniform_cuda_after': [], | |
'uniform_cuda_before': [], | |
} | |
for t in range(args.num_trials): | |
times = run_trial(S, T) | |
for key, time_ms in times.items(): | |
cur_results[key].append(time_ms) | |
all_results.append(cur_results) | |
with open(args.stats_json, 'w') as f: | |
json.dump(all_results, f) | |
def timeit(f, *args): | |
torch.cuda.synchronize() | |
t0 = time.time() | |
out = f(*args) | |
torch.cuda.synchronize() | |
t1 = time.time() | |
time_ms = (t1 - t0) * 1000.0 | |
return time_ms | |
def torch_uniform(tensor): | |
return tensor.uniform_() | |
def run_trial(sizes, dtype): | |
tensor_gpu = torch.zeros(sizes, dtype=dtype).cuda() | |
# We want to test torch on both cpu and gpu; randomize the order | |
# in which we call them to minimize any systematic effects of caching, etc | |
calls = [ | |
['uniform_cuda_after', torch_uniform, tensor_gpu], | |
['uniform_cuda_before', torch_uniform, tensor_gpu], | |
] | |
random.shuffle(calls) | |
results = {} | |
for key, f, args in calls: | |
results[key] = timeit(f, args) | |
return results | |
if __name__ == '__main__': | |
args = parser.parse_args() | |
main(args) | |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment