Skip to content

Instantly share code, notes, and snippets.

@yuxuanzhuang

yuxuanzhuang/benchmark_on_thread_effect_of_numpy.ipynb Secret

Last active October 14, 2020 15:27
Show Gist options
  • Save yuxuanzhuang/82e1e7b57d0cda80ac964d1cd138f618 to your computer and use it in GitHub Desktop.
Save yuxuanzhuang/82e1e7b57d0cda80ac964d1cd138f618 to your computer and use it in GitHub Desktop.
Download ZIP
Display the source blob
Display the rendered blob
Raw
benchmark_on_thread_effect_of_numpy.ipynb
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
@yuxuanzhuang
Copy link
Author

yuxuanzhuang commented Oct 14, 2020 •
edited
Loading

As a sum up for the code performance after numpy.dot (Line 6)

Threads Time
version_1 version_2 version_3
without numpy.dot 1 37070 37606 36344
2 37132 37527 36547
6 37139 37747 36638
12 37205 37493 36684 (not really using 12 threads with MKL)
with numpy.dot 1 37360 36285 36369
2 38269 36373 37200
6 38038 39142 36863
12 72670 67931 36739 (not really using 12 threads with MKL)
non numpy operation after numpy.dot 1 201422 208773 213470
2 200999 208948 212878
6 428546 216324 213259
12 440734 463529 213133 (not really using 12 threads with MKL)

Another interesting observation is that non-numpy operation is already performing pretty bad with 6 threads.

The test was down with

  • 6-core CPU (hyperthreading enabled)

version_1

  • numpy 1.19.1
  • python 3.8
threadpool_info()
[{'filepath': '/home/scottzhuang/anaconda3/envs/gsoc/lib/python3.8/site-packages/numpy.libs/libopenblasp-r0-ae94cfde.3.9.dev.so',
  'prefix': 'libopenblas',
  'user_api': 'blas',
  'internal_api': 'openblas',
  'version': '0.3.9.dev',
  'num_threads': 12,
  'threading_layer': 'pthreads'},
 {'filepath': '/home/scottzhuang/anaconda3/envs/gsoc/lib/libgomp.so.1.0.0',
  'prefix': 'libgomp',
  'user_api': 'openmp',
  'internal_api': 'openmp',
  'version': None,
  'num_threads': 12},
 {'filepath': '/home/scottzhuang/anaconda3/envs/gsoc/lib/libmkl_rt.so',
  'prefix': 'libmkl_rt',
  'user_api': 'blas',
  'internal_api': 'mkl',
  'version': '2020.0.1',
  'num_threads': 6,
  'threading_layer': 'intel'},
 {'filepath': '/home/scottzhuang/anaconda3/envs/gsoc/lib/libiomp5.so',
  'prefix': 'libiomp',
  'user_api': 'openmp',
  'internal_api': 'openmp',
  'version': None,
  'num_threads': 12}]

version_2

  • numpy 1.20.0.dev0+4ccfbe6
  • python 3.8
threadpool_info()
[{'filepath': '/opt/OpenBLAS/lib/libopenblas_haswellp-r0.3.10.dev.so',
  'prefix': 'libopenblas',
  'user_api': 'blas',
  'internal_api': 'openblas',
  'version': '0.3.10.dev',
  'num_threads': 12,
  'threading_layer': 'pthreads'}]

version_3

  • numpy 1.19
  • python 3.8
threadpool_info()
[{'filepath': '/home/scottzhuang/anaconda3/lib/libmkl_rt.so',
  'prefix': 'libmkl_rt',
  'user_api': 'blas',
  'internal_api': 'mkl',
  'version': '2020.0.0',
  'num_threads': 6,
  'threading_layer': 'intel'},
 {'filepath': '/home/scottzhuang/anaconda3/lib/libiomp5.so',
  'prefix': 'libiomp',
  'user_api': 'openmp',
  'internal_api': 'openmp',
  'version': None,
  'num_threads': 12}]

Will check if it's version related.
EDIT: Update with other clean-installed version of numpy (OpenBlas, MKL).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment