op_bench-py

performance benchmark for pytorch operators

the benchmark will run with TWO modes by defaults:
#1: use single socket
#2: use single core
./run.sh [xxx.py]

#the record will be generated under folers like r000ah
./profile.sh [xxx.py]

#you may collect on linux and review the record on windows
#but have to archive the record on linux first
./archive.sh [r000ah]

Name		Name	Last commit message	Last commit date
Latest commit History 41 Commits
README.md		README.md
adaptive_avg_pool2d.py		adaptive_avg_pool2d.py
adaptive_max_pool2d.py		adaptive_max_pool2d.py
archive.sh		archive.sh
avg_pool2d.py		avg_pool2d.py
batch_norm.py		batch_norm.py
cat.py		cat.py
group_norm.py		group_norm.py
index_select.py		index_select.py
layernorm.py		layernorm.py
masked_fill.py		masked_fill.py
masked_select.py		masked_select.py
matmul.py		matmul.py
max_pool2d.py		max_pool2d.py
max_unpool2d.py		max_unpool2d.py
profile.sh		profile.sh
rnn.py		rnn.py
run.sh		run.sh
test_batch_norm.py		test_batch_norm.py
test_mkldnn_conv_channles_last.py		test_mkldnn_conv_channles_last.py
test_mkldnn_deconv_channles_last.py		test_mkldnn_deconv_channles_last.py
test_mkldnn_rnn.py		test_mkldnn_rnn.py
uniform.py		uniform.py
upsample_bicubic.py		upsample_bicubic.py
upsample_linear.py		upsample_linear.py
upsample_nearest.py		upsample_nearest.py
weight_norm.py		weight_norm.py

Provide feedback