oneAPI Deep Neural Network Library (oneDNN)  1.4.0
Performance library for Deep Learning
Softmax Primitive Example

This C++ API example demonstrates how to create and execute a Softmax primitive in forward training propagation mode.

Key optimizations included in this example:

  • In-place primitive execution;
  • Softmax along axis 1 (C) for 2D tensors.
/*******************************************************************************
* Copyright 2020 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 <algorithm>
#include <cmath>
#include <iostream>
#include <string>
#include <vector>
#include "dnnl.hpp"
#include "example_utils.hpp"
using namespace dnnl;
void softmax_example(dnnl::engine::kind engine_kind) {
// Create execution dnnl::engine.
dnnl::engine engine(engine_kind, 0);
// Create dnnl::stream.
dnnl::stream engine_stream(engine);
// Tensor dimensions.
const memory::dim N = 3, // batch size
IC = 1000; // channels
// Source (src) and destination (dst) tensors dimensions.
memory::dims src_dims = {N, IC};
// Allocate buffer.
std::vector<float> src_data(product(src_dims));
std::generate(src_data.begin(), src_data.end(), []() {
static int i = 0;
return std::cos(i++ / 10.f);
});
// Create src memory descriptor and memory object.
auto src_md = memory::desc(src_dims, dt::f32, tag::nc);
auto src_mem = memory(src_md, engine);
// Write data to memory object's handle.
write_to_dnnl_memory(src_data.data(), src_mem);
// Softmax axis.
const int axis = 1;
// Create operation descriptor.
auto softmax_d
// Create primitive descriptor.
auto softmax_pd = softmax_forward::primitive_desc(softmax_d, engine);
// Create the primitive.
auto softmax_prim = softmax_forward(softmax_pd);
// Primitive arguments. Set up in-place execution by assigning src as DST.
std::unordered_map<int, memory> softmax_args;
softmax_args.insert({DNNL_ARG_SRC, src_mem});
softmax_args.insert({DNNL_ARG_DST, src_mem});
// Primitive execution.
softmax_prim.execute(engine_stream, softmax_args);
// Wait for the computation to finalize.
engine_stream.wait();
// Read data from memory object's handle.
read_from_dnnl_memory(src_data.data(), src_mem);
}
int main(int argc, char **argv) {
return handle_example_errors(
softmax_example, parse_engine_kind(argc, argv));
}
dnnl::stream
An execution stream.
Definition: dnnl.hpp:1047
dnnl::engine
An execution engine.
Definition: dnnl.hpp:844
dnnl::prop_kind::forward_training
@ forward_training
Forward data propagation (training mode).
dnnl::engine::kind
kind
Kinds of engines.
Definition: dnnl.hpp:849
DNNL_ARG_DST
#define DNNL_ARG_DST
A special mnemonic for destination argument for primitives that have a single destination.
Definition: dnnl_types.h:1806
dnnl::softmax_forward::primitive_desc
Primitive descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:5738
dnnl.hpp
dnnl::softmax_forward
Softmax forward propagation primitive.
Definition: dnnl.hpp:5705
dnnl::memory::data_type
data_type
Data type specification.
Definition: dnnl.hpp:1208
dnnl::memory::format_tag
format_tag
Memory format tag specification.
Definition: dnnl.hpp:1282
dnnl::query::src_md
@ src_md
source memory desc
DNNL_ARG_SRC
#define DNNL_ARG_SRC
A special mnemonic for source argument for primitives that have a single source.
Definition: dnnl_types.h:1782
dnnl::memory::dim
dnnl_dim_t dim
Integer type for representing dimension sizes and indices.
Definition: dnnl.hpp:1190
dnnl::query::softmax_d
@ softmax_d
softmax descriptor
dnnl::memory
Memory object.
Definition: dnnl.hpp:1188
dnnl::memory::dims
std::vector< dim > dims
Vector of dimensions.
Definition: dnnl.hpp:1193
dnnl::memory::desc
A memory descriptor.
Definition: dnnl.hpp:1729
dnnl
oneDNN namespace
Definition: dnnl.hpp:81
dnnl::softmax_forward::desc
Descriptor for a softmax forward propagation primitive.
Definition: dnnl.hpp:5707