Skip to content

Latest commit

 

History

History
72 lines (57 loc) · 2.97 KB

README.md

File metadata and controls

72 lines (57 loc) · 2.97 KB
Raw

ibench

Benchmarks for Intel Distribution for Python.

Install

# Since you want to benchmark numpy/scipy, manually install the one you want
# instead of letting pip install one
# cython is needed to build native extensions in ibench_native
conda install scipy cython scikit-learn
pip install -v --upgrade .

Native versions

If icc and cython are available during the build, they will be used to build native benchmarks. To specify a different compiler, specify one in the environment variable CXX.

Run

# basic command
python -m ibench run -b all --size large --runs 3 --file all.out

Specifying benchmarks

  • To run one or multiple benchmarks, pass the -b BENCHMARKS... option. Benchmarks can be specified individually, or in predefined groups (e.g. native contains all native benchmarks)
  • To specify the problem size, use the --size option. This selects from a list of predefined problem sizes.

Configuring output

ibench supports both JSON (default) and CSV output with collection of environment information. When using CSV format, ibench will prefix each line of environment information with a comment character (default @). Use the following arguments to the run subparser to configure its output:

  • -f,--format FORMAT - use the specified format (json or csv)
  • --no-get-env-info - don't collect environment or machine info
  • --env-info-prefix - prefix character to use before environment info

Running benchmarks by domain

Linear Algebra

  • To run python benchmarks: python -m ibench run -b linalg --size large --runs 3 --file linalg.out
  • To run native benchmarks*: python -m ibench run -b native --size large --runs 3 --file native.out

* Currently, native benchmarks are only available for det, dot, lu, and inv.

Machine Learning

  • To run python benchmarks: python -m ibench run -b sklearn --size large --runs 3 --file sklearn.out
  • For comparable python and native benchmarks, see scikit-learn_bench.

Fast Fourier Transforms

  • To run python benchmarks: python -m ibench run -b fft --size large --runs 3 --file fft.out
  • For comparable python and native benchmarks, see fft_benchmark.

Random Number Generation

  • To run python benchmarks: python -m ibench run -b rng --size large --runs 3 --file rng.out
  • For comparable python and native benchmarks, see optimizations_bench.

Black-Scholes Formula

  • To run python Black-Scholes benchmark: python -m ibench run -b blacksch --size large --runs 3 --file blacksch.out
  • For comparable python and native benchmarks, see BlackScholes_bench.

UMath

Help

python -m ibench --help
python -m ibench run --help