saymrwulf/pytorch
1
0
Fork 0
mirror of https://github.com/saymrwulf/pytorch.git synced 2026-05-14 20:57:59 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Set number of threads for operator_benchmarks (#27010)

Browse source 
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/27010

Setting OMP_NUM_THREADS programmatically doesn't do the right thing because initialization is already done. Fixing this by calling torch.set_num_threads explicitly.

Passing --omp_num_threads works as expected now.

In dir benchmarks/operator_benchmark/

python -m pt.qconv_test --tag_filter resnext101_32x4 --wipe_cache --test_name QConv2d_N1_IC64_OC128_H56_W56_G1_kernel1_stride1_pad0 --omp_num_threads 1
```
# ----------------------------------------
# PyTorch/Caffe2 Operator Micro-benchmarks
# ----------------------------------------
# Tag : None

# Benchmarking PyTorch: QConv2d
# Mode: Eager
# Name: QConv2d_N1_IC64_OC128_H56_W56_G1_kernel1_stride1_pad0
# Input: N: 1, IC: 64, OC: 128, H: 56, W: 56, G: 1, kernel: 1, stride: 1, pad: 0
Forward Execution Time (us) : 509.965

# Benchmarking PyTorch: QConv2d
# Mode: Eager
# Name: QConv2d_N1_IC64_OC128_H56_W56_G1_kernel1_stride1_pad0
# Input: N: 1, IC: 64, OC: 128, H: 56, W: 56, G: 1, kernel: 1, stride: 1, pad: 0
Forward Execution Time (us) : 576.007
```

python -m pt.qconv_test --tag_filter resnext101_32x4 --wipe_cache --test_name QConv2d_N1_IC64_OC128_H56_W56_G1_kernel1_stride1_pad0 --omp_num_threads 4

```
# ----------------------------------------
# PyTorch/Caffe2 Operator Micro-benchmarks
# ----------------------------------------
# Tag : None

# Benchmarking PyTorch: QConv2d
# Mode: Eager
# Name: QConv2d_N1_IC64_OC128_H56_W56_G1_kernel1_stride1_pad0
# Input: N: 1, IC: 64, OC: 128, H: 56, W: 56, G: 1, kernel: 1, stride: 1, pad: 0
Forward Execution Time (us) : 195.002

# Benchmarking PyTorch: QConv2d
# Mode: Eager
# Name: QConv2d_N1_IC64_OC128_H56_W56_G1_kernel1_stride1_pad0
# Input: N: 1, IC: 64, OC: 128, H: 56, W: 56, G: 1, kernel: 1, stride: 1, pad: 0
Forward Execution Time (us) : 189.788
```
ghstack-source-id: 91050434

Test Plan: See summary

Differential Revision: D17647391

fbshipit-source-id: e00de1151902291ed94fd34446995ea1f0199d14
This commit is contained in:
Daya Khudia 2019-09-30 17:00:31 -07:00 committed by Facebook Github Bot
parent 5e7549c3b5
commit bf7ebc5a53
1 changed files with 12 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

12
benchmarks/operator_benchmark/benchmark_runner.py
View file

@ -6,6 +6,7 @@ from __future__ import unicode_literals
import argparse
from caffe2.python import workspace
import torch
import benchmark_core
import benchmark_utils
@ -127,7 +128,18 @@ def main():
workspace.GlobalInit(['caffe2', '--caffe2_log_level=0'])
workspace.ClearGlobalNetObserver()
if args.omp_num_threads:
# benchmark_utils.set_omp_threads sets the env variable OMP_NUM_THREADS
# which doesn't have any impact as C2 init logic has already been called
# before setting the env var.
# In general, OMP_NUM_THREADS (and other OMP env variables) needs to be set
# before the program is started.
# From Chapter 4 in OMP standard: https://www.openmp.org/wp-content/uploads/openmp-4.5.pdf
# "Modifications to the environment variables after the program has started,
# even if modified by the program itself, are ignored by the OpenMP implementation"
benchmark_utils.set_omp_threads(args.omp_num_threads)
if benchmark_utils.is_pytorch_enabled(args.framework):
torch.set_num_threads(args.omp_num_threads)
if args.mkl_num_threads:
benchmark_utils.set_mkl_threads(args.mkl_num_threads)