openvino/inference-engine/tools/compile_tool/main.cpp

523 lines
21 KiB
C++

// Copyright (C) 2018-2020 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
#include <cstdlib>
#include <iostream>
#include <fstream>
#include <algorithm>
#include <chrono>
#include <unordered_map>
#include <map>
#include <vector>
#include <string>
#include <gflags/gflags.h>
#include "inference_engine.hpp"
#include <vpu/vpu_plugin_config.hpp>
#include <vpu/private_plugin_config.hpp>
#include <vpu/utils/string.hpp>
#include "samples/common.hpp"
static constexpr char help_message[] =
"Optional. Print the usage message.";
static constexpr char model_message[] =
"Required. Path to the XML model.";
static constexpr char targetDeviceMessage[] =
"Required. Specify a target device for which executable network will be compiled.\n"
" Use \"-d HETERO:<comma-separated_devices_list>\" format to specify HETERO plugin.\n"
" Use \"-d MULTI:<comma-separated_devices_list>\" format to specify MULTI plugin.\n"
" The application looks for a suitable plugin for the specified device.";
static constexpr char output_message[] =
"Optional. Path to the output file. Default value: \"<model_xml_file>.blob\".";
static constexpr char config_message[] =
"Optional. Path to the configuration file.";
static constexpr char inputs_precision_message[] =
"Optional. Specifies precision for all input layers of the network.";
static constexpr char outputs_precision_message[] =
"Optional. Specifies precision for all output layers of the network.";
static constexpr char iop_message[] =
"Optional. Specifies precision for input and output layers by name.\n"
" Example: -iop \"input:FP16, output:FP16\".\n"
" Notice that quotes are required.\n"
" Overwrites precision from ip and op options for specified layers.";
static constexpr char inputs_layout_message[] =
"Optional. Specifies layout for all input layers of the network.";
static constexpr char outputs_layout_message[] =
"Optional. Specifies layout for all input layers of the network.";
static constexpr char iol_message[] =
"Optional. Specifies layout for input and output layers by name.\n"
" Example: -iol \"input:NCHW, output:NHWC\".\n"
" Notice that quotes are required.\n"
" Overwrites layout from il and ol options for specified layers.";
// MYRIAD-specific
static constexpr char number_of_shaves_message[] =
"Optional. Specifies number of shaves.\n"
" Should be set with \"VPU_NUMBER_OF_CMX_SLICES\".\n"
" Overwrites value from config.\n";
static constexpr char number_of_cmx_slices_message[] =
"Optional. Specifies number of CMX slices.\n"
" Should be set with \"VPU_NUMBER_OF_SHAVES\".\n"
" Overwrites value from config.";
static constexpr char tiling_cmx_limit_message[] =
"Optional. Specifies CMX limit for data tiling.\n"
" Value should be equal or greater than -1.\n"
" Overwrites value from config.";
// FPGA-specific
static constexpr char dla_arch_name[] =
"Optional. Specify architecture name used to compile executable network for FPGA device.";
DEFINE_bool(h, false, help_message);
DEFINE_string(m, "", model_message);
DEFINE_string(d, "", targetDeviceMessage);
DEFINE_string(o, "", output_message);
DEFINE_string(c, "", config_message);
DEFINE_string(ip, "", inputs_precision_message);
DEFINE_string(op, "", outputs_precision_message);
DEFINE_string(iop, "", iop_message);
DEFINE_string(il, "", inputs_layout_message);
DEFINE_string(ol, "", outputs_layout_message);
DEFINE_string(iol, "", iol_message);
DEFINE_string(VPU_NUMBER_OF_SHAVES, "", number_of_shaves_message);
DEFINE_string(VPU_NUMBER_OF_CMX_SLICES, "", number_of_cmx_slices_message);
DEFINE_string(VPU_TILING_CMX_LIMIT_KB, "", tiling_cmx_limit_message);
DEFINE_string(DLA_ARCH_NAME, "", dla_arch_name);
static void showUsage() {
std::cout << "compile_tool [OPTIONS]" << std::endl;
std::cout << std::endl;
std::cout << " Common options: " << std::endl;
std::cout << " -h " << help_message << std::endl;
std::cout << " -m <value> " << model_message << std::endl;
std::cout << " -d <value> " << targetDeviceMessage << std::endl;
std::cout << " -o <value> " << output_message << std::endl;
std::cout << " -c <value> " << config_message << std::endl;
std::cout << " -ip <value> " << inputs_precision_message << std::endl;
std::cout << " -op <value> " << outputs_precision_message << std::endl;
std::cout << " -iop \"<value>\" " << iop_message << std::endl;
std::cout << " -il <value> " << inputs_layout_message << std::endl;
std::cout << " -ol <value> " << outputs_layout_message << std::endl;
std::cout << " -iol \"<value>\" " << iol_message << std::endl;
std::cout << std::endl;
std::cout << " MYRIAD-specific options: " << std::endl;
std::cout << " -VPU_NUMBER_OF_SHAVES <value> " << number_of_shaves_message << std::endl;
std::cout << " -VPU_NUMBER_OF_CMX_SLICES <value> " << number_of_cmx_slices_message << std::endl;
std::cout << " -VPU_TILING_CMX_LIMIT_KB <value> " << tiling_cmx_limit_message << std::endl;
std::cout << std::endl;
std::cout << " FPGA-specific options: " << std::endl;
std::cout << " -DLA_ARCH_NAME <value> " << dla_arch_name << std::endl;
std::cout << std::endl;
}
static bool parseCommandLine(int* argc, char*** argv, InferenceEngine::Core& ie) {
gflags::ParseCommandLineNonHelpFlags(argc, argv, true);
if (FLAGS_h) {
showUsage();
return false;
}
if (FLAGS_m.empty()) {
throw std::invalid_argument("Path to model xml file is required");
}
if (FLAGS_d.empty()) {
throw std::invalid_argument("Target device name is required");
}
if (1 < *argc) {
std::stringstream message;
message << "Unknown arguments: ";
for (auto arg = 1; arg < *argc; arg++) {
message << argv[arg];
if (arg < *argc) {
message << " ";
}
}
throw std::invalid_argument(message.str());
}
return true;
}
static std::map<std::string, std::string> parseConfigFile(char comment = '#') {
std::map<std::string, std::string> config;
std::ifstream file(FLAGS_c);
if (file.is_open()) {
std::string key, value;
while (file >> key >> value) {
if (key.empty() || key[0] == comment) {
continue;
}
config[key] = value;
}
}
return config;
}
static std::map<std::string, std::string> configure() {
const bool isMYRIAD = FLAGS_d.find("MYRIAD") != std::string::npos;
const bool isFPGA = FLAGS_d.find("FPGA") != std::string::npos;
auto config = parseConfigFile();
if (isMYRIAD) {
IE_SUPPRESS_DEPRECATED_START
config[VPU_MYRIAD_CONFIG_KEY(PLATFORM)] = "VPU_MYRIAD_2480";
IE_SUPPRESS_DEPRECATED_END
if (!FLAGS_VPU_NUMBER_OF_SHAVES.empty()) {
config[InferenceEngine::MYRIAD_NUMBER_OF_SHAVES] = FLAGS_VPU_NUMBER_OF_SHAVES;
}
if (!FLAGS_VPU_NUMBER_OF_CMX_SLICES.empty()) {
config[InferenceEngine::MYRIAD_NUMBER_OF_CMX_SLICES] = FLAGS_VPU_NUMBER_OF_CMX_SLICES;
}
if (!FLAGS_VPU_TILING_CMX_LIMIT_KB.empty()) {
config[InferenceEngine::MYRIAD_TILING_CMX_LIMIT_KB] = FLAGS_VPU_TILING_CMX_LIMIT_KB;
}
}
if (isFPGA) {
if (!FLAGS_DLA_ARCH_NAME.empty()) {
config["DLIA_ARCH_NAME"] = FLAGS_DLA_ARCH_NAME;
}
}
return config;
}
static std::map<std::string, std::string> parseArgMap(std::string argMap) {
argMap.erase(std::remove_if(argMap.begin(), argMap.end(), ::isspace), argMap.end());
std::vector<std::string> pairs;
vpu::splitStringList(argMap, pairs, ',');
std::map<std::string, std::string> parsedMap;
for (auto&& pair : pairs) {
std::vector<std::string> keyValue;
vpu::splitStringList(pair, keyValue, ':');
if (keyValue.size() != 2) {
throw std::invalid_argument("Invalid key/value pair " + pair + ". Expected <layer_name>:<value>");
}
parsedMap[keyValue[0]] = keyValue[1];
}
return parsedMap;
}
using supported_precisions_t = std::unordered_map<std::string, InferenceEngine::Precision>;
using supported_layouts_t = std::unordered_map<std::string, InferenceEngine::Layout>;
using matchLayoutToDims_t = std::unordered_map<size_t, size_t>;
static InferenceEngine::Layout getLayout(std::string value,
const supported_layouts_t& supported_layouts) {
std::transform(value.begin(), value.end(), value.begin(), ::toupper);
const auto layout = supported_layouts.find(value);
if (layout == supported_layouts.end()) {
throw std::logic_error("\"" + value + "\"" + " is not a valid layout");
}
return layout->second;
}
static InferenceEngine::Layout getLayout(const std::string& value) {
static const supported_layouts_t supported_layouts = {
{ "NCDHW", InferenceEngine::Layout::NCDHW },
{ "NDHWC", InferenceEngine::Layout::NDHWC },
{ "NCHW", InferenceEngine::Layout::NCHW },
{ "NHWC", InferenceEngine::Layout::NHWC },
{ "CHW", InferenceEngine::Layout::CHW },
{ "NC", InferenceEngine::Layout::NC },
{ "C", InferenceEngine::Layout::C },
};
return getLayout(value, supported_layouts);
}
static bool isMatchLayoutToDims(InferenceEngine::Layout layout, size_t dimension) {
static const matchLayoutToDims_t matchLayoutToDims = {
{static_cast<size_t>(InferenceEngine::Layout::NCDHW), 5 },
{static_cast<size_t>(InferenceEngine::Layout::NDHWC), 5 },
{static_cast<size_t>(InferenceEngine::Layout::NCHW), 4 },
{static_cast<size_t>(InferenceEngine::Layout::NHWC), 4 },
{static_cast<size_t>(InferenceEngine::Layout::CHW), 3 },
{static_cast<size_t>(InferenceEngine::Layout::NC), 2 },
{static_cast<size_t>(InferenceEngine::Layout::C), 1 }
};
const auto dims = matchLayoutToDims.find(static_cast<size_t>(layout));
if (dims == matchLayoutToDims.end()) {
throw std::logic_error("Layout is not valid.");
}
return dimension == dims->second;
}
static InferenceEngine::Precision getPrecision(std::string value,
const supported_precisions_t& supported_precisions) {
std::transform(value.begin(), value.end(), value.begin(), ::toupper);
const auto precision = supported_precisions.find(value);
if (precision == supported_precisions.end()) {
throw std::logic_error("\"" + value + "\"" + " is not a valid precision");
}
return precision->second;
}
static InferenceEngine::Precision getPrecision(const std::string& value) {
static const supported_precisions_t supported_precisions = {
{ "FP32", InferenceEngine::Precision::FP32 },
{ "FP16", InferenceEngine::Precision::FP16 },
{ "BF16", InferenceEngine::Precision::BF16 },
{ "U64", InferenceEngine::Precision::U64 },
{ "I64", InferenceEngine::Precision::I64 },
{ "U32", InferenceEngine::Precision::U32 },
{ "I32", InferenceEngine::Precision::I32 },
{ "U16", InferenceEngine::Precision::U16 },
{ "I16", InferenceEngine::Precision::I16 },
{ "U8", InferenceEngine::Precision::U8 },
{ "I8", InferenceEngine::Precision::I8 },
{ "BOOL", InferenceEngine::Precision::BOOL },
};
return getPrecision(value, supported_precisions);
}
bool isFP16(InferenceEngine::Precision precision) {
return precision == InferenceEngine::Precision::FP16;
}
bool isFP32(InferenceEngine::Precision precision) {
return precision == InferenceEngine::Precision::FP32;
}
bool isFloat(InferenceEngine::Precision precision) {
return isFP16(precision) || isFP32(precision);
}
static void setPrecisions(const InferenceEngine::CNNNetwork& network) {
const auto user_precisions_map = parseArgMap(FLAGS_iop);
auto inputs = network.getInputsInfo();
auto outputs = network.getOutputsInfo();
for (auto&& item : user_precisions_map) {
const auto& layer_name = item.first;
const auto& user_precision = item.second;
const auto input = inputs.find(layer_name);
const auto output = outputs.find(layer_name);
if (input != inputs.end()) {
input->second->setPrecision(getPrecision(user_precision));
} else if (output != outputs.end()) {
output->second->setPrecision(getPrecision(user_precision));
} else {
throw std::logic_error(layer_name + " is not an input neither output");
}
}
}
static void setDefaultIO(InferenceEngine::CNNNetwork& network) {
const bool isMYRIAD = FLAGS_d.find("MYRIAD") != std::string::npos;
const bool isVPUX = FLAGS_d.find("VPUX") != std::string::npos;
if (isMYRIAD) {
const InferenceEngine::Precision fp16 = InferenceEngine::Precision::FP16;
for (auto&& layer : network.getInputsInfo()) {
if (isFloat(layer.second->getPrecision())) {
layer.second->setPrecision(fp16);
}
}
for (auto&& layer : network.getOutputsInfo()) {
if (isFloat(layer.second->getPrecision())) {
layer.second->setPrecision(fp16);
}
}
}
if (isVPUX) {
const InferenceEngine::Precision u8 = InferenceEngine::Precision::U8;
const InferenceEngine::Precision fp32 = InferenceEngine::Precision::FP32;
for (auto&& layer : network.getInputsInfo()) {
layer.second->setPrecision(u8);
}
for (auto&& layer : network.getOutputsInfo()) {
layer.second->setPrecision(fp32);
}
}
}
static void processPrecisions(InferenceEngine::CNNNetwork& network) {
if (!FLAGS_ip.empty()) {
const auto user_precision = getPrecision(FLAGS_ip);
for (auto&& layer : network.getInputsInfo()) {
layer.second->setPrecision(user_precision);
}
}
if (!FLAGS_op.empty()) {
auto user_precision = getPrecision(FLAGS_op);
for (auto&& layer : network.getOutputsInfo()) {
layer.second->setPrecision(user_precision);
}
}
if (!FLAGS_iop.empty()) {
setPrecisions(network);
}
}
static void setLayouts(const InferenceEngine::CNNNetwork& network) {
const auto user_layouts_map = parseArgMap(FLAGS_iol);
auto inputs = network.getInputsInfo();
auto outputs = network.getOutputsInfo();
for (auto&& item : user_layouts_map) {
const auto& layer_name = item.first;
const auto& user_layout = getLayout(item.second);
const auto input = inputs.find(layer_name);
const auto output = outputs.find(layer_name);
if (input != inputs.end()) {
if (!isMatchLayoutToDims(user_layout, input->second->getTensorDesc().getDims().size())) {
throw std::logic_error(item.second + " layout is not applicable to " + layer_name);
}
input->second->setLayout(user_layout);
} else if (output != outputs.end()) {
if (!isMatchLayoutToDims(user_layout, output->second->getTensorDesc().getDims().size())) {
throw std::logic_error(item.second + " layout is not applicable to " + layer_name);
}
output->second->setLayout(user_layout);
} else {
throw std::logic_error(layer_name + " is not an input neither output");
}
}
}
static void processLayout(InferenceEngine::CNNNetwork& network) {
if (!FLAGS_il.empty()) {
const auto layout = getLayout(FLAGS_il);
for (auto&& layer : network.getInputsInfo()) {
if (isMatchLayoutToDims(layout, layer.second->getTensorDesc().getDims().size())) {
layer.second->setLayout(layout);
}
}
}
if (!FLAGS_ol.empty()) {
const auto layout = getLayout(FLAGS_ol);
for (auto&& layer : network.getOutputsInfo()) {
if (isMatchLayoutToDims(layout, layer.second->getTensorDesc().getDims().size())) {
layer.second->setLayout(layout);
}
}
}
if (!FLAGS_iol.empty()) {
setLayouts(network);
}
}
std::string getFileNameFromPath(const std::string& path,
#if defined(_WIN32)
const std::string& sep = "\\") {
#else
const std::string& sep = "/") {
#endif
const auto pos = path.rfind(sep);
if (std::string::npos == pos) {
return path;
} else {
return path.substr(pos + 1);
}
}
using TimeDiff = std::chrono::milliseconds;
int main(int argc, char* argv[]) {
TimeDiff loadNetworkTimeElapsed {0};
try {
std::cout << "Inference Engine: " << InferenceEngine::GetInferenceEngineVersion() << std::endl;
std::cout << std::endl;
InferenceEngine::Core ie;
if (!parseCommandLine(&argc, &argv, ie)) {
return EXIT_SUCCESS;
}
auto network = ie.ReadNetwork(FLAGS_m);
setDefaultIO(network);
processPrecisions(network);
processLayout(network);
std::cout << "Network inputs:" << std::endl;
for (auto&& layer : network.getInputsInfo()) {
std::cout << " " << layer.first << " : " << layer.second->getPrecision() << " / " << layer.second->getLayout() << std::endl;
}
std::cout << "Network outputs:" << std::endl;
for (auto&& layer : network.getOutputsInfo()) {
std::cout << " " << layer.first << " : " << layer.second->getPrecision() << " / " << layer.second->getLayout() << std::endl;
}
std::cout << std::endl;
auto timeBeforeLoadNetwork = std::chrono::steady_clock::now();
auto executableNetwork = ie.LoadNetwork(network, FLAGS_d, configure());
loadNetworkTimeElapsed = std::chrono::duration_cast<TimeDiff>(std::chrono::steady_clock::now() - timeBeforeLoadNetwork);
std::string outputName = FLAGS_o;
if (outputName.empty()) {
outputName = getFileNameFromPath(fileNameNoExt(FLAGS_m)) + ".blob";
}
std::ofstream outputFile{outputName};
if (!outputFile.is_open()) {
std::cout << "Output file " << outputName << " can't be opened for writing" << std::endl;
return EXIT_FAILURE;
} else {
executableNetwork.Export(outputFile);
}
} catch (const std::exception& error) {
std::cerr << error.what() << std::endl;
return EXIT_FAILURE;
} catch (...) {
std::cerr << "Unknown/internal exception happened." << std::endl;
return EXIT_FAILURE;
}
std::cout << "Done. LoadNetwork time elapsed: " << loadNetworkTimeElapsed.count() << " ms" << std::endl;
return EXIT_SUCCESS;
}