.. index:: pair: page; Layer Normalization Primitive Example .. _doxid-layer_normalization_example_cpp: Layer Normalization Primitive Example ===================================== This C++ API example demonstrates how to create and execute a :ref:`Layer normalization ` primitive in forward propagation mode. Key optimizations included in this example: * In-place primitive execution; * Creation of memory objects using the primitive descriptor. .. ref-code-block:: cpp /******************************************************************************* * Copyright 2020-2023 Intel Corporation * * Licensed 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. *******************************************************************************/ #include #include #include #include #include #include "example_utils.hpp" #include "oneapi/dnnl/dnnl.hpp" using namespace :ref:`dnnl `; using :ref:`tag ` = :ref:`memory::format_tag `; using :ref:`dt ` = :ref:`memory::data_type `; void layer_normalization_example(:ref:`dnnl::engine::kind ` engine_kind) { :ref:`dnnl::engine ` :ref:`engine `(engine_kind, 0); // Create dnnl::stream. :ref:`dnnl::stream ` engine_stream(:ref:`engine `); // Tensor dimensions. const :ref:`memory::dim ` T = 12, // time steps N = 3, // batch C = 227; // channels // Source (src) and destination (dst) tensors dimensions. const :ref:`memory::dims ` src_dims = {T, N, C}; // Scale/shift tensor dimensions. :ref:`memory::dims ` scaleshift_dims = {C}; // Allocate buffer. std::vector src_data(product(src_dims)); std::vector scale_data(product(scaleshift_dims)); std::vector shift_data(product(scaleshift_dims)); // Initialize src tensor. std::generate(src_data.begin(), src_data.end(), []() { static int i = 0; return std::cos(i++ / 10.f); }); // Initialize scale. std::generate(scale_data.begin(), scale_data.end(), []() { static int i = 0; return std::sin(i++ * 2.f); }); // Initialize shift. std::generate(shift_data.begin(), shift_data.end(), []() { static int i = 0; return std::tan(float(i++)); }); // Create src memory descriptor and memory objects. auto :ref:`src_md ` = :ref:`memory::desc `(src_dims, :ref:`dt::f32 `, tag::tnc); auto :ref:`dst_md ` = :ref:`memory::desc `(src_dims, :ref:`dt::f32 `, tag::tnc); auto scaleshift_md = :ref:`memory::desc `(scaleshift_dims, :ref:`dt::f32 `, tag::x); auto src_mem = :ref:`memory `(src_md, :ref:`engine `); auto scale_mem = :ref:`memory `(scaleshift_md, :ref:`engine `); auto shift_mem = :ref:`memory `(scaleshift_md, :ref:`engine `); // Write data to memory object's handle. write_to_dnnl_memory(src_data.data(), src_mem); write_to_dnnl_memory(scale_data.data(), scale_mem); write_to_dnnl_memory(shift_data.data(), shift_mem); // Create primitive descriptor. const float epsilon = 1.e-10f; auto lnorm_pd = :ref:`layer_normalization_forward::primitive_desc `(:ref:`engine `, :ref:`prop_kind::forward_training `, src_md, dst_md, :ref:`dt::f32 `, epsilon, :ref:`normalization_flags::use_scale ` | :ref:`normalization_flags::use_shift `); // Use the memory descriptors from the primitive to create memory objects // required for the primitive: mean, variance, scale/shift. auto mean_mem = :ref:`memory `(lnorm_pd.mean_desc(), :ref:`engine `); auto variance_mem = :ref:`memory `(lnorm_pd.variance_desc(), :ref:`engine `); // Create the primitive. auto lnorm_prim = :ref:`layer_normalization_forward `(lnorm_pd); // Primitive arguments. Set up in-place execution by assigning src as DST. std::unordered_map lnorm_args; lnorm_args.insert({:ref:`DNNL_ARG_SRC `, src_mem}); lnorm_args.insert({:ref:`DNNL_ARG_MEAN `, mean_mem}); lnorm_args.insert({:ref:`DNNL_ARG_VARIANCE `, variance_mem}); lnorm_args.insert({:ref:`DNNL_ARG_SCALE `, scale_mem}); lnorm_args.insert({:ref:`DNNL_ARG_SHIFT `, shift_mem}); lnorm_args.insert({:ref:`DNNL_ARG_DST `, src_mem}); // Primitive execution: layer normalization. lnorm_prim.execute(engine_stream, lnorm_args); // Wait for the computation to finalize. engine_stream.wait(); // Read data from memory object's handle.s read_from_dnnl_memory(src_data.data(), src_mem); } int main(int argc, char **argv) { return handle_example_errors( layer_normalization_example, parse_engine_kind(argc, argv)); }