.. index:: pair: page; Matmul Primitive Example .. _doxid-matmul_example_cpp: Matmul Primitive Example ======================== This C++ API example demonstrates how to create and execute a :ref:`MatMul ` primitive. Key optimizations included in this example: * Primitive attributes with fused post-ops. .. ref-code-block:: cpp /******************************************************************************* * Copyright 2020-2022 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 matmul_example(:ref:`dnnl::engine::kind ` engine_kind) { // Create execution dnnl::engine. :ref:`dnnl::engine ` :ref:`engine `(engine_kind, 0); // Create dnnl::stream. :ref:`dnnl::stream ` engine_stream(:ref:`engine `); // Tensor dimensions. const :ref:`memory::dim ` MB = 3, // batch size M = 128, K = 256, N = 512; // Source (src), weights, bias, and destination (dst) tensors dimensions. :ref:`memory::dims ` src_dims = {MB, M, K}; :ref:`memory::dims ` weights_dims = {MB, K, N}; :ref:`memory::dims ` bias_dims = {1, 1, N}; :ref:`memory::dims ` dst_dims = {MB, M, N}; // Allocate buffers. std::vector src_data(product(src_dims)); std::vector weights_data(product(weights_dims)); std::vector bias_data(product(bias_dims)); std::vector dst_data(product(dst_dims)); // Initialize src, weights, bias. std::generate(src_data.begin(), src_data.end(), []() { static int i = 0; return std::cos(i++ / 10.f); }); std::generate(weights_data.begin(), weights_data.end(), []() { static int i = 0; return std::sin(i++ * 2.f); }); std::generate(bias_data.begin(), bias_data.end(), []() { static int i = 0; return std::tanh(float(i++)); }); // Create memory descriptors and memory objects for src, weights, bias, and // dst. auto :ref:`src_md ` = :ref:`memory::desc `(src_dims, :ref:`dt::f32 `, tag::abc); auto :ref:`weights_md ` = :ref:`memory::desc `(weights_dims, :ref:`dt::f32 `, tag::abc); auto bias_md = :ref:`memory::desc `(bias_dims, :ref:`dt::f32 `, tag::abc); auto :ref:`dst_md ` = :ref:`memory::desc `(dst_dims, :ref:`dt::f32 `, tag::abc); auto src_mem = :ref:`memory `(src_md, :ref:`engine `); auto weights_mem = :ref:`memory `(weights_md, :ref:`engine `); auto bias_mem = :ref:`memory `(bias_md, :ref:`engine `); auto dst_mem = :ref:`memory `(dst_md, :ref:`engine `); // Write data to memory object's handles. write_to_dnnl_memory(src_data.data(), src_mem); write_to_dnnl_memory(weights_data.data(), weights_mem); write_to_dnnl_memory(bias_data.data(), bias_mem); // Create primitive post-ops (ReLU). const float alpha = 0.f; const float beta = 0.f; :ref:`post_ops ` matmul_ops; matmul_ops.:ref:`append_eltwise `(:ref:`algorithm::eltwise_relu `, alpha, beta); :ref:`primitive_attr ` matmul_attr; matmul_attr.:ref:`set_post_ops `(matmul_ops); // Create primitive descriptor. auto matmul_pd = :ref:`matmul::primitive_desc `( :ref:`engine `, src_md, weights_md, bias_md, dst_md, matmul_attr); // Create the primitive. auto matmul_prim = :ref:`matmul `(matmul_pd); // Primitive arguments. std::unordered_map matmul_args; matmul_args.insert({:ref:`DNNL_ARG_SRC `, src_mem}); matmul_args.insert({:ref:`DNNL_ARG_WEIGHTS `, weights_mem}); matmul_args.insert({:ref:`DNNL_ARG_BIAS `, bias_mem}); matmul_args.insert({:ref:`DNNL_ARG_DST `, dst_mem}); // Primitive execution: matrix multiplication with ReLU. matmul_prim.execute(engine_stream, matmul_args); // Wait for the computation to finalize. engine_stream.wait(); // Read data from memory object's handle. read_from_dnnl_memory(dst_data.data(), dst_mem); } int main(int argc, char **argv) { return handle_example_errors(matmul_example, parse_engine_kind(argc, argv)); }