cjolivier01 commented on issue #8686: Kernel operator tuning URL: https://github.com/apache/incubator-mxnet/pull/8686#issuecomment-345419483 OMP overhead this run, ~8500 ns. unary and binary op times: 0.5-200 ns OperatorTuneBase::duration_t OperatorTuneBase::omp_overhead_ns_ = 8495; mshadow::op::identity time: 1.33008 ns mxnet::op::mshadow_op::identity time: 1.32959 ns mxnet::op::mshadow_op::identity_grad time: 1.22754 ns mxnet::op::mshadow_op::negation time: 1.56836 ns mxnet::op::mshadow_op::reciprocal time: 1.53467 ns mxnet::op::mshadow_op::reciprocal_grad time: 2.48975 ns mxnet::op::mshadow_op::sigmoid time: 29.3955 ns mxnet::op::mshadow_op::sigmoid_grad time: 2.18262 ns mxnet::op::mshadow_op::relu time: 1.77295 ns mxnet::op::mshadow_op::relu_grad time: 2.14844 ns mxnet::op::mshadow_op::tanh time: 38.1255 ns mxnet::op::mshadow_op::tanh_grad time: 2.25098 ns mxnet::op::mshadow_op::softrelu time: 51.7314 ns mxnet::op::mshadow_op::softrelu_grad time: 33.5557 ns mxnet::op::mshadow_op::exp time: 17.4258 ns mxnet::op::mshadow_op::exp time: 18.1758 ns mxnet::op::mshadow_op::expm1 time: 23.189 ns mxnet::op::mshadow_op::log time: 25.1323 ns mxnet::op::mshadow_op::log_grad time: 2.08008 ns mxnet::op::mshadow_op::log1p time: 24.5527 ns mxnet::op::mshadow_op::log1p_grad time: 2.52344 ns mxnet::op::mshadow_op::log2 time: 24.1777 ns mxnet::op::mshadow_op::log2_grad time: 2.0459 ns mxnet::op::mshadow_op::log10 time: 26.4624 ns mxnet::op::mshadow_op::log10_grad time: 2.08057 ns mxnet::op::mshadow_op::sin time: 12.6177 ns mxnet::op::mshadow_op::sin_grad time: 13.1626 ns mxnet::op::mshadow_op::sinh time: 41.6372 ns mxnet::op::mshadow_op::sinh_grad time: 29.5659 ns mxnet::op::mshadow_op::arcsin time: 16.2319 ns mxnet::op::mshadow_op::arcsin_grad time: 6.37695 ns mxnet::op::mshadow_op::arcsinh time: 38.6709 ns mxnet::op::mshadow_op::arcsinh_grad time: 15.4478 ns mxnet::op::mshadow_op::cos time: 10.5376 ns mxnet::op::mshadow_op::cos_grad time: 12.9243 ns mxnet::op::mshadow_op::cosh time: 27.4175 ns mxnet::op::mshadow_op::cosh_grad time: 41.7739 ns mxnet::op::mshadow_op::arccos time: 19.2676 ns mxnet::op::mshadow_op::arccos_grad time: 6.41064 ns mxnet::op::mshadow_op::arccosh time: 15.2773 ns mxnet::op::mshadow_op::arccosh_grad time: 19.4717 ns mxnet::op::mshadow_op::tan time: 26.4966 ns mxnet::op::mshadow_op::tan_grad time: 2.11426 ns mxnet::op::mshadow_op::arctan time: 19.2329 ns mxnet::op::mshadow_op::arctan_grad time: 3.00098 ns mxnet::op::mshadow_op::arctanh time: 38.6025 ns mxnet::op::mshadow_op::arctanh_grad time: 3.06934 ns mxnet::op::mshadow_op::square time: 1.50049 ns mxnet::op::mshadow_op::square_grad time: 1.90967 ns mxnet::op::mshadow_op::square_root time: 13.1631 ns mxnet::op::mshadow_op::square_root_grad time: 2.08008 ns mxnet::op::mshadow_op::reciprocal_square_root time: 13.7085 ns mxnet::op::mshadow_op::reciprocal_square_root_grad time: 16.062 ns mxnet::op::mshadow_op::cube_root time: 38.6367 ns mxnet::op::mshadow_op::cube_root_grad time: 3.00098 ns mxnet::op::mshadow_op::reciprocal_cube_root time: 38.1255 ns mxnet::op::mshadow_op::reciprocal_cube_root_grad time: 40.2393 ns mxnet::op::mshadow_op::abs time: 1.50049 ns mxnet::op::mshadow_op::sign time: 3.54639 ns mxnet::op::mshadow_op::sign time: 4.09229 ns mxnet::op::mshadow_op::sign_grad time: 1.53467 ns mxnet::op::mshadow_op::round time: 7.74121 ns mxnet::op::mshadow_op::floor time: 5.25146 ns mxnet::op::mshadow_op::trunc time: 6.20654 ns mxnet::op::mshadow_op::rint time: 9.10498 ns mxnet::op::mshadow_op::fix time: 10.3667 ns mxnet::op::mshadow_op::gamma time: 107.555 ns mxnet::op::mshadow_op::gamma_grad time: 208.938 ns mxnet::op::mshadow_op::gammaln time: 74.5112 ns mxnet::op::mshadow_op::gammaln_grad time: 95.3125 ns mxnet::op::mshadow_op::ceil time: 4.43311 ns mxnet::op::mshadow_op::degrees time: 1.50049 ns mxnet::op::mshadow_op::degrees_grad time: 1.50049 ns mxnet::op::mshadow_op::radians time: 1.50049 ns mxnet::op::mshadow_op::radians_grad time: 1.50049 ns mshadow::op::plus time: 1.53467 ns mshadow::op::minus time: 1.53467 ns mshadow::op::mul time: 1.53467 ns mshadow::op::div time: 1.63672 ns mshadow::op::right time: 1.32959 ns mxnet::op::mshadow_op::rminus time: 1.80762 ns mxnet::op::mshadow_op::rdiv time: 1.84131 ns mxnet::op::mshadow_op::div_grad time: 1.53467 ns mxnet::op::mshadow_op::div_grad time: 1.90967 ns mxnet::op::mshadow_op::div_rgrad time: 2.28467 ns mxnet::op::mshadow_op::div_rgrad time: 2.83057 ns mxnet::op::mshadow_op::rdiv_grad time: 2.62598 ns mxnet::op::mshadow_op::mod time: 41.8423 ns mxnet::op::mshadow_op::mod_grad time: 1.26172 ns mxnet::op::mshadow_op::mod_rgrad time: 4.97852 ns mxnet::op::mshadow_op::rmod time: 42.5581 ns mxnet::op::mshadow_op::rmod_grad time: 5.0127 ns mxnet::op::mshadow_op::left time: 1.19336 ns mxnet::op::mshadow_op::left time: 1.53418 ns mxnet::op::mshadow_op::right time: 1.22803 ns mxnet::op::mshadow_op::right time: 1.50049 ns mxnet::op::mshadow_op::power time: 71.7827 ns mxnet::op::mshadow_op::rpower time: 68.4067 ns mxnet::op::mshadow_op::power_grad time: 70.9302 ns mxnet::op::mshadow_op::rpower_grad time: 22.0293 ns mxnet::op::mshadow_op::power_rgrad time: 91.1865 ns mxnet::op::mshadow_op::maximum time: 1.56836 ns mxnet::op::mshadow_op::minimum time: 1.53467 ns mxnet::op::mshadow_op::hypot time: 25.5757 ns mxnet::op::mshadow_op::hypot_grad_left time: 19.3013 ns mxnet::op::mshadow_op::hypot_grad_left time: 13.8789 ns mxnet::op::mshadow_op::hypot_grad_right time: 13.1973 ns mxnet::op::mshadow_op::hypot_grad_right time: 13.231 ns mxnet::op::mshadow_op::lt time: 1.77295 ns mxnet::op::mshadow_op::lt time: 2.5918 ns mxnet::op::mshadow_op::le time: 1.73926 ns mxnet::op::mshadow_op::le time: 2.62598 ns mxnet::op::mshadow_op::gt time: 2.5918 ns mxnet::op::mshadow_op::gt time: 2.08008 ns mxnet::op::mshadow_op::ge time: 2.01172 ns mxnet::op::mshadow_op::ge time: 2.55762 ns mxnet::op::mshadow_op::ne time: 2.11426 ns mxnet::op::mshadow_op::ne time: 2.62598 ns mxnet::op::mshadow_op::eq time: 2.11426 ns mxnet::op::mshadow_op::eq time: 2.62598 ns mxnet::op::mshadow_op::smooth_l1_loss time: 4.29639 ns mxnet::op::mshadow_op::smooth_l1_gradient time: 3.85352 ns mxnet::op::mxnet_op::set_to_int<0> time: 1.09131 ns mxnet::op::mxnet_op::set_to_int<1> time: 0.647949 ns mxnet::op::PopulateFullIdxRspKernel time: 0.647949 ns
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services