[GitHub] [incubator-mxnet] bartekkuncer commented on a change in pull request #20952: Restore quantized RNN to master

[GitBox <gi...@apache.org>]

bartekkuncer commented on a change in pull request #20952:
URL: https://github.com/apache/incubator-mxnet/pull/20952#discussion_r836261653



##########
File path: src/operator/quantization/dnnl/dnnl_quantized_rnn.cc
##########
@@ -0,0 +1,365 @@
+/*
+ * 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.
+ */
+
+/*!
+ * \file dnnl_quantized_rnn.cc
+ * \brief Common functions for quantized recurrent neural network
+ * \author Zixuan Wei
+ */
+
+#if MXNET_USE_ONEDNN == 1
+
+#include "operator/quantization/quantization_utils.h"
+#include "operator/quantization/dnnl/dnnl_quantized_rnn-inl.h"
+
+namespace mxnet {
+namespace op {
+
+std::vector<float> GetDNNLRnnWeightsQParams(const DNNLRnnFullParam& full_param, float* w_ptr) {
+  const int nthreads            = mxnet::engine::OpenMP::Get()->GetRecommendedOMPThreadCount();
+  const RNNParam& default_param = full_param.default_param;
+  const LayerParamVector& layer_params = full_param.layer_params;
+
+  const DNNLRnnLayerParam& layer_param0 = layer_params.at(0);
+  const size_t w_size0                  = layer_param0.single_w_size;
+  const size_t wx_size0                 = 4 * layer_param0.state_size * layer_param0.input_size;

Review comment:
       Magic number.

##########
File path: src/operator/nn/dnnl/dnnl_rnn.cc
##########
@@ -449,11 +482,19 @@ void DNNLRnnForward::SetNewDataMem(void* x,
   auto& cpu_engine      = CpuEngine::Get()->get_engine();
   dnnl_args_map_t& args = net_args_;
 
+  int src_dtype = dtype;
+  int dst_dtype = dtype;
+  if (param_.quantized) {
+    src_dtype = mshadow::kUint8;

Review comment:
       RNN does not support int8?

##########
File path: src/operator/quantization/quantized_rnn-inl.h
##########
@@ -0,0 +1,41 @@
+/*
+ * 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.
+ */
+
+/*!
+ * \file quantized_rnn-inl.h

Review comment:
       Is this file necessary?




-- 
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.

To unsubscribe, e-mail: commits-unsubscribe@mxnet.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org