Last time I talked about compiling and installing mxnet, this time I will talk about compiling and installing mxnet ( mxnet-mkl ) optimized for Intel CPU processors, which is also required for work. by almost 10 times).
First of all, download the source code first. I downloaded the latest release version mxnet-1.6.0 here. It is worth noting a little detail, mxnet-1.6.0 version is also the last version to support python2, after that it will no longer support python2.
wget https://github.com/apache/incubator-mxnet/archive/1.6.0.tar.gz |
The first step, environment preparation:
The second step is to switch the code installation path:
cd python pip install setup.py |
The third step is to verify whether the installation is correct
import mxnet as mx shape_x = (1, 10, 8) x_npy = np.random.normal(0, 1, shape_x) x = mx.sym.Variable('x') exe.forward(is_train=False) |
More detailed verification results:
|
output result
Numpy + Intel(R) MKL: THREADING LAYER: (null) Numpy + Intel(R) MKL: setting Intel(R) MKL to use INTEL OpenMP runtime Numpy + Intel(R) MKL: preloading libiomp5.so runtime MKL_VERBOSE Intel(R) MKL 2019.0 Update 3 Product build 20190125 for Intel(R) 64 architecture Intel(R) Advanced Vector Extensions 512 (Intel(R) AVX-512) enabled processors, Lnx 2.40GHz lp64 intel_thread NMICDev:0 MKL_VERBOSE SGEMM(T,N,12,10,8,0x7f7f927b1378,0x1bc2140,8,0x1ba8040,8,0x7f7f927b1380,0x7f7f7400a280,12) 8.93ms CNR:OFF Dyn:1 FastMM:1 TID:0 NThr:40 WDiv:HOST:+0.000 |
The fourth step is to optimize settings based on mkl reasoning
export MXNET_SUBGRAPH_BACKEND=MKLDNN |
For more MKL-DNN calculation graph optimization, please refer to https://cwiki.apache.org/confluence/display/MXNET/MXNet+Graph+Optimization+and+Quantization+based+on+subgraph+and+MKL-DNN
Finally, for calculation operator operations supported by mkl,
For more details, please see https://github.com/apache/incubator-mxnet/blob/v1.5.x/docs/tutorials/mkldnn/operator_list.md