ZhennanQin commented on a change in pull request #13297: [MKLDNN]Add quantized concat URL: https://github.com/apache/incubator-mxnet/pull/13297#discussion_r235580692
########## File path: src/operator/quantization/mkldnn/mkldnn_quantized_concat.cc ########## @@ -0,0 +1,119 @@ +/* + * 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 quantized_concat.cc + * \brief + */ + +#if MXNET_USE_MKLDNN == 1 +#include "../../nn/mkldnn/mkldnn_concat-inl.h" +#include "../quantization_utils.h" + +namespace mxnet { +namespace op { + +namespace quantized_concat_enum { +enum QuantizedConcatOutputs { kOut, kMin, kMax }; +} + +static float GetScale(const NDArray& data, float min, float max) { + auto data_range = (data.dtype() == mshadow::kInt8) ? kInt8Range : kUint8Range; + return data_range / MaxAbs(min, max); +} + +static void MKLDNNQuantizedConcatForward(const nnvm::NodeAttrs& attrs, const OpContext& ctx, + const std::vector<NDArray>& in_data, + const std::vector<OpReqType>& req, + const std::vector<NDArray>& out_data) { + const ConcatParam& param_ = nnvm::get<ConcatParam>(attrs.parsed); + CHECK_EQ(in_data.size(), static_cast<size_t>(param_.num_args * 3)); + CHECK_EQ(out_data.size(), 3U); + // Collect data and output min/max + std::vector<float> data_min(param_.num_args); + std::vector<float> data_max(param_.num_args); + float output_min = 0.f; Review comment: I have to say that's what it suppose to do. Just not as your suppose. I didn't say it's going to get the minimal value from outputs, but it's going to get the minimal value not greater than 0. The 'min' in variable name and comments are refer to the output name:'min_output'. That's the `quantized_concat` op defined. Do you want me to rename the variable to "output_non_positive_min" ? ---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
