Script for benchmarking dot,batch_dot and fully_connected with OpPerf

benchmark_intel_mkl.py

import mxnet as mx
from mxnet import nd
from benchmark.opperf.utils.benchmark_utils import run_performance_test
import os
#export MKL_VERBOSE=1
#export MKLDNN_VERBOSE=1
os.environ["MKL_VERBOSE"]=1
os.environ["MKLDNN_VERBOSE"]=1
os.environ["OMP_NUM_THREADS"]="4"
#print version
print(mx.__version__)

#print env vars
from mxnet.runtime import feature_list
print(feature_list())

#dot_batch = run_performance_test([nd.dot,nd.batch_dot],run_backward=False, ctx=mx.cpu(), inputs=[{"lhs":(4, 512, 512),"rhs":(4, 512, 512),"transpose_b":True},{"lhs":(5, 512, 512),"rhs":(5, 512, 512),"transpose_b":True},{"lhs":(5, 512, 1536),"rhs":(5, 512, 1536),"transpose_b":True},{"lhs":(5, 512, 2048),"rhs":(5, 512, 2048),"transpose_b":True},{"lhs":(5, 2048, 512),"rhs":(5, 2048, 512),"transpose_b":True}])
#print(dot_batch)

fc = run_performance_test(nd.FullyConnected,run_backward=False, ctx=mx.cpu(), inputs=[{"data":(4, 512),"weight":(512,512),"no_bias":True,"num_hidden":512},{"data":(5, 512),"weight":(512,512),"no_bias":True,"num_hidden":512},{"data":(5, 512),"weight":(1536,512),"no_bias":True,"num_hidden":1536},{"data":(5, 512),"weight":(2048,512),"no_bias":True,"num_hidden":2048},{"data":(5, 2048),"weight":(512,2048),"no_bias":True,"num_hidden":512}])
print(fc)