Open8

dpnpの速度ベンチマーク

fate_shelledfate_shelled

検証環境

  • CPU: Intel Core i5 11500
  • iGPU: Intel UHD Graphics 750
  • OS: Ubuntu22.04
  • Python: Intel oneAPI Python(2023.1.0-46399) の仮想環境 (Python3.9)

numpyとdpnpとpytorchで比較

fate_shelledfate_shelled

NumPy

intel MKLが有効になっている。

>>> numpy.__version__
'1.23.5'

>>> numpy.show_config()
blas_armpl_info:
  NOT AVAILABLE
blas_mkl_info:
    libraries = ['mkl_rt', 'pthread']
    library_dirs = ['/opt/intel/oneapi/intelpython/latest/lib']
    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
    include_dirs = ['/opt/intel/oneapi/intelpython/latest/include']
blas_opt_info:
    libraries = ['mkl_rt', 'pthread']
    library_dirs = ['/opt/intel/oneapi/intelpython/latest/lib']
    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
    include_dirs = ['/opt/intel/oneapi/intelpython/latest/include']
lapack_armpl_info:
  NOT AVAILABLE
lapack_mkl_info:
    libraries = ['mkl_rt', 'pthread']
    library_dirs = ['/opt/intel/oneapi/intelpython/latest/lib']
    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
    include_dirs = ['/opt/intel/oneapi/intelpython/latest/include']
lapack_opt_info:
    libraries = ['mkl_rt', 'pthread']
    library_dirs = ['/opt/intel/oneapi/intelpython/latest/lib']
    define_macros = [('SCIPY_MKL_H', None), ('HAVE_CBLAS', None)]
    include_dirs = ['/opt/intel/oneapi/intelpython/latest/include']
Supported SIMD extensions in this NumPy install:
    baseline = SSE,SSE2,SSE3,SSSE3,SSE41,POPCNT,SSE42
    found = AVX512_ICL
    not found =
fate_shelledfate_shelled

torch (CPU版)

pipからinstallしていますが、intel MKLとMKL DNNが有効になっています。

pip3 install torch  --index-url https://download.pytorch.org/whl/cpu

>>> torch.__version__
'2.0.1+cpu'

>>> print(torch.__config__.show())
PyTorch built with:
  - GCC 9.3
  - C++ Version: 201703
  - Intel(R) oneAPI Math Kernel Library Version 2023.1-Product Build 20230303 for Intel(R) 64 architecture applications
  - Intel(R) MKL-DNN v2.7.3 (Git Hash 6dbeffbae1f23cbbeae17adb7b5b13f1f37c080e)
  - OpenMP 201511 (a.k.a. OpenMP 4.5)
  - LAPACK is enabled (usually provided by MKL)
  - NNPACK is enabled
  - CPU capability usage: AVX2
  - Build settings: BLAS_INFO=mkl, BUILD_TYPE=Release, CXX_COMPILER=/opt/rh/devtoolset-9/root/usr/bin/c++, CXX_FLAGS= -D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow, LAPACK_INFO=mkl, PERF_WITH_AVX=1, PERF_WITH_AVX2=1, PERF_WITH_AVX512=1, TORCH_DISABLE_GPU_ASSERTS=ON, TORCH_VERSION=2.0.1, USE_CUDA=0, USE_CUDNN=OFF, USE_EXCEPTION_PTR=1, USE_GFLAGS=OFF, USE_GLOG=OFF, USE_MKL=ON, USE_MKLDNN=ON, USE_MPI=OFF, USE_NCCL=OFF, USE_NNPACK=ON, USE_OPENMP=ON, USE_ROCM=OFF,
fate_shelledfate_shelled
  • dpnpがfloat64の計算ができなかったのでfloat32, int32, int64で検証
  • dpnpリポジトリのexampleを修正して、pytorchも比較する。
  • 速度は10回計算の平均値とする。単位はsec/回。
fate_shelledfate_shelled

matmul

  • shape: 16x16, 32x32, 64x64, 128x128, 256x256, 512x512, 1024x1024

float32

NumPy/IntelMKL dpnp/CPU dpnp/iGPU torch/CPU,MKL,MKLDNN
16 0.000001 0.000090 0.000167 0.000003
32 0.000002 0.000096 0.000175 0.000005
64 0.000006 0.000141 0.000209 0.000006
128 0.000017 0.000165 0.000191 0.000018
256 0.000104 0.000258 0.000325 0.000108
512 0.000705 0.000959 0.001019 0.000716
1024 0.005930 0.006764 0.005296 0.007296

int64

NumPy/IntelMKL dpnp/CPU dpnp/iGPU torch/CPU,MKL,MKLDNN
16 0.000003 0.000168 0.000172 0.000004
32 0.000021 0.000206 0.000195 0.000010
64 0.000153 0.000189 0.000222 0.000028
128 0.001168 0.000189 0.000505 0.000208
256 0.011949 0.000561 0.002539 0.001641
512 0.164101 0.005005 0.021120 0.014062
1024 1.732274 0.054449 0.127659 0.110528

int32

NumPy/IntelMKL dpnp/CPU dpnp/iGPU torch/CPU,MKL,MKLDNN
16 0.000003 0.000143 0.000173 0.000004
32 0.000021 0.000159 0.000183 0.000014
64 0.000144 0.000174 0.000195 0.000094
128 0.001139 0.000200 0.000382 0.000831
256 0.010917 0.000314 0.001492 0.008805
512 0.160692 0.001904 0.010477 0.054832
1024 1.266861 0.015211 0.088180 0.431325