TaoLv commented on a change in pull request #14893: Integrating the MKL VML functions to MXNET to speed-up the (element-wised) mathematic computation URL: https://github.com/apache/incubator-mxnet/pull/14893#discussion_r283627743
########## File path: src/operator/mkl_functions-inl.h ########## @@ -0,0 +1,218 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, + * software distributed under the License is distributed on an + * "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY + * KIND, either express or implied. See the License for the + * specific language governing permissions and limitations + * under the License. + */ + +/*! + * Copyright (c) 2018 by Contributors + * \file mkl_functions-inl.h + * \brief + * \author +*/ +#ifndef MXNET_OPERATOR_MKL_FUNCTIONS_INL_H_ +#define MXNET_OPERATOR_MKL_FUNCTIONS_INL_H_ + +#if MSHADOW_USE_MKL == 1 +#include "mkl.h" + +namespace mxnet { +namespace op { +namespace mkl_func { + +MSHADOW_XINLINE +static bool check_size(const size_t n) { + const size_t MKL_INT_MAX = (sizeof(MKL_INT) == sizeof(int)) ? INT_MAX : LLONG_MAX; + return (n <= MKL_INT_MAX); +} + +MSHADOW_XINLINE +static bool check_type(const int t) { + return (t == mshadow::kFloat32 || t == mshadow::kFloat64); +} + +#define MXNET_MKL_UNARY_MATH_FUNC(name, func) \ +struct name { \ + MSHADOW_XINLINE static void Vectorize(const index_t n, const float *src, float *dst) { \ + vs##func(static_cast<MKL_INT>(n), src, dst); \ + } \ + MSHADOW_XINLINE static void Vectorize(const index_t n, const double *src, double *dst) { \ + vd##func(static_cast<MKL_INT>(n), src, dst); \ + } \ +}; + +#define MXNET_MKL_BINARY_MATH_FUNC(name, func) \ +struct name { \ + MSHADOW_XINLINE static void Vectorize(const index_t n, \ + const float *a, \ + const float *b, \ + float *c) { \ + vs##func(static_cast<MKL_INT>(n), a, b, c); \ + } \ + MSHADOW_XINLINE static void Vectorize(const index_t n, \ + const double *a, \ + const double *b, \ + double *c) { \ + vd##func(static_cast<MKL_INT>(n), a, b, c); \ + } \ +}; + +MXNET_MKL_UNARY_MATH_FUNC(erf, Erf); +MXNET_MKL_UNARY_MATH_FUNC(exp, Exp); +MXNET_MKL_UNARY_MATH_FUNC(exp2, Exp2); +MXNET_MKL_UNARY_MATH_FUNC(exp10, Exp10); +MXNET_MKL_UNARY_MATH_FUNC(expm1, Expm1); +MXNET_MKL_UNARY_MATH_FUNC(log, Ln); +MXNET_MKL_UNARY_MATH_FUNC(log2, Log2); +MXNET_MKL_UNARY_MATH_FUNC(log10, Log10); +MXNET_MKL_UNARY_MATH_FUNC(log1p, Log1p); + +MXNET_MKL_UNARY_MATH_FUNC(sin, Sin); +MXNET_MKL_UNARY_MATH_FUNC(cos, Cos); +MXNET_MKL_UNARY_MATH_FUNC(tan, Tan); +MXNET_MKL_UNARY_MATH_FUNC(asin, Asin); +MXNET_MKL_UNARY_MATH_FUNC(acos, Acos); +MXNET_MKL_UNARY_MATH_FUNC(atan, Atan); + +MXNET_MKL_UNARY_MATH_FUNC(sinh, Sinh); +MXNET_MKL_UNARY_MATH_FUNC(cosh, Cosh); +MXNET_MKL_UNARY_MATH_FUNC(tanh, Tanh); +MXNET_MKL_UNARY_MATH_FUNC(asinh, Asinh); +MXNET_MKL_UNARY_MATH_FUNC(acosh, Acosh); +MXNET_MKL_UNARY_MATH_FUNC(atanh, Atanh); + +MXNET_MKL_UNARY_MATH_FUNC(sqrt, Sqrt); +MXNET_MKL_UNARY_MATH_FUNC(abs, Abs); +MXNET_MKL_UNARY_MATH_FUNC(cbrt, Cbrt); +MXNET_MKL_UNARY_MATH_FUNC(round, Round); +MXNET_MKL_UNARY_MATH_FUNC(ceil, Ceil); +MXNET_MKL_UNARY_MATH_FUNC(floor, Floor); +MXNET_MKL_UNARY_MATH_FUNC(trunc, Trunc); + +MXNET_MKL_UNARY_MATH_FUNC(lgamma, LGamma); +MXNET_MKL_UNARY_MATH_FUNC(tgamma, TGamma); +MXNET_MKL_UNARY_MATH_FUNC(square, Sqr); + +MXNET_MKL_BINARY_MATH_FUNC(add, Add); +MXNET_MKL_BINARY_MATH_FUNC(sub, Sub); +MXNET_MKL_BINARY_MATH_FUNC(mul, Mul); +MXNET_MKL_BINARY_MATH_FUNC(pow, Pow); +MXNET_MKL_BINARY_MATH_FUNC(hypot, Hypot); Review comment: No. We just put all the VML functions here. We think these functions can be leveraged by MXNet in the future. But currently it need to change the registration of each operator to use these functions. In this PR we only optimized some operators which are used in BERT. We propose to optimize others when we face performance problems on them. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to 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
