mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
8286ce1e3a
Summary: Closes https://github.com/caffe2/caffe2/pull/1260 Differential Revision: D5906739 Pulled By: Yangqing fbshipit-source-id: e482ba9ba60b5337d9165f28f7ec68d4518a0902
75 lines
2.9 KiB
Python
75 lines
2.9 KiB
Python
# Copyright (c) 2016-present, Facebook, Inc.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
##############################################################################
|
|
|
|
from __future__ import absolute_import
|
|
from __future__ import division
|
|
from __future__ import print_function
|
|
from __future__ import unicode_literals
|
|
from caffe2.proto import caffe2_pb2
|
|
from caffe2.python import workspace, core, lstm_benchmark, utils
|
|
from copy import copy
|
|
|
|
@utils.debug
|
|
def Compare(args):
|
|
results = []
|
|
num_iters = 1000
|
|
args.gpu = True
|
|
with core.DeviceScope(core.DeviceOption(caffe2_pb2.CUDA, 0)):
|
|
for batch_size in [64, 128, 256]:
|
|
for seq_length in [20, 100]:
|
|
for hidden_dim in [40, 100, 400, 800]:
|
|
args.batch_size = batch_size
|
|
args.seq_length = seq_length
|
|
args.hidden_dim = hidden_dim
|
|
args.data_size = batch_size * seq_length * num_iters
|
|
args.iters_to_report = num_iters // 3
|
|
|
|
args.implementation = 'own'
|
|
t_own = lstm_benchmark.Benchmark(args)
|
|
workspace.ResetWorkspace()
|
|
args.implementation = 'cudnn'
|
|
t_cudnn = lstm_benchmark.Benchmark(args)
|
|
workspace.ResetWorkspace()
|
|
results.append((copy(args), float(t_own), float(t_cudnn)))
|
|
print(args)
|
|
print("t_cudnn / t_own: {}".format(t_cudnn / t_own))
|
|
|
|
for args, t_own, t_cudnn in results:
|
|
print("{}: cudnn time: {}, own time: {}, ratio: {}".format(
|
|
str(args), t_cudnn, t_own, t_cudnn / t_own))
|
|
|
|
ratio_sum = 0
|
|
for args, t_own, t_cudnn in results:
|
|
ratio = float(t_cudnn) / t_own
|
|
ratio_sum += ratio
|
|
print("hidden_dim: {}, seq_lengths: {}, batch_size: {}, num_layers: {}:"
|
|
" cudnn time: {}, own time: {}, ratio: {}".format(
|
|
args.hidden_dim, args.seq_length, args.batch_size,
|
|
args.num_layers, t_cudnn, t_own, ratio))
|
|
|
|
print("Ratio average: {}".format(ratio_sum / len(results)))
|
|
|
|
|
|
if __name__ == '__main__':
|
|
args = lstm_benchmark.GetArgumentParser().parse_args()
|
|
|
|
workspace.GlobalInit([
|
|
'caffe2',
|
|
'--caffe2_log_level=0',
|
|
'--caffe2_print_blob_sizes_at_exit=0',
|
|
'--caffe2_gpu_memory_tracking=1'])
|
|
|
|
Compare(args)
|