[Time tests] Add API 2.0 support (#9878)
* add performance hint to time infer * init commit - add api 2 support * change imInfo filling * change copyright dates * check hw positions to default * add debug info * fix mistake * add check layout funcs for api2 time infer * reformat code (2 -> 4) * upd with reshape api2 * upd with master * -- * fix fillTensors - set as template * fix common_utils.cpp after merge master
This commit is contained in:
Victor Kuznetsov authored and 
GitHub committed
1 parent
70f65bdb74
commit
0dd8d895a0
12 files changed
+532
-318
No files matched your search
@@ -4,9 +4,6 @@
|
||||
|
||||
#include "common_utils.h"
|
||||
|
||||
#include <inference_engine.hpp>
|
||||
#include <openvino/openvino.hpp>
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill InferRequest blobs with random values or image information
|
||||
@@ -15,6 +12,7 @@ void fillBlobs(InferenceEngine::InferRequest inferRequest,
|
||||
const InferenceEngine::ConstInputsDataMap &inputsInfo,
|
||||
const size_t &batchSize) {
|
||||
std::vector<std::pair<size_t, size_t>> input_image_sizes;
|
||||
|
||||
for (const InferenceEngine::ConstInputsDataMap::value_type &item: inputsInfo) {
|
||||
if (isImage(item.second))
|
||||
input_image_sizes.push_back(getTensorHeightWidth(item.second->getTensorDesc()));
|
||||
@@ -22,6 +20,7 @@ void fillBlobs(InferenceEngine::InferRequest inferRequest,
|
||||
|
||||
for (const InferenceEngine::ConstInputsDataMap::value_type &item: inputsInfo) {
|
||||
InferenceEngine::Blob::Ptr inputBlob = inferRequest.GetBlob(item.first);
|
||||
|
||||
if (isImageInfo(inputBlob) && (input_image_sizes.size() == 1)) {
|
||||
// Fill image information
|
||||
auto image_size = input_image_sizes.at(0);
|
||||
@@ -57,32 +56,4 @@ void fillBlobs(InferenceEngine::InferRequest inferRequest,
|
||||
throw std::logic_error("Input precision is not supported for " + item.first);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Fill infer_request tensors with random values or image information
|
||||
*/
|
||||
void fillTensors(ov::InferRequest &infer_request, const std::vector<ov::Output<ov::Node>> &inputs) {
|
||||
for (size_t i = 0; i < inputs.size(); ++i) {
|
||||
ov::Tensor input_tensor;
|
||||
if (inputs[i].get_element_type() == ov::element::f32) {
|
||||
input_tensor = fillTensorRandom<float>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::f16) {
|
||||
input_tensor = fillTensorRandom<short>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i32) {
|
||||
input_tensor = fillTensorRandom<int32_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::u8) {
|
||||
input_tensor = fillTensorRandom<uint8_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i8) {
|
||||
input_tensor = fillTensorRandom<int8_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::u16) {
|
||||
input_tensor = fillTensorRandom<uint16_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i16) {
|
||||
input_tensor = fillTensorRandom<int16_t>(inputs[i]);
|
||||
} else {
|
||||
throw std::logic_error(
|
||||
"Input precision is not supported for " + inputs[i].get_element_type().get_type_name());
|
||||
}
|
||||
infer_request.set_input_tensor(i, input_tensor);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -5,6 +5,7 @@
|
||||
#pragma once
|
||||
|
||||
#include <inference_engine.hpp>
|
||||
#include <openvino/openvino.hpp>
|
||||
#include <limits>
|
||||
|
||||
|
||||
@@ -67,7 +68,7 @@ inline std::pair<size_t, size_t> getTensorHeightWidth(const InferenceEngine::Ten
|
||||
*/
|
||||
template<typename T>
|
||||
void fillBlobRandom(InferenceEngine::Blob::Ptr &inputBlob) {
|
||||
auto minput = InferenceEngine::as<InferenceEngine::MemoryBlob>(inputBlob);
|
||||
InferenceEngine::MemoryBlob::Ptr minput = InferenceEngine::as<InferenceEngine::MemoryBlob>(inputBlob);
|
||||
// locked memory holder should be alive all time while access to its buffer happens
|
||||
auto minputHolder = minput->wmap();
|
||||
|
||||
@@ -78,23 +79,26 @@ void fillBlobRandom(InferenceEngine::Blob::Ptr &inputBlob) {
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill InferenceEngine tensor with random values (OV API 2.0)
|
||||
*/
|
||||
template<typename T>
|
||||
ov::Tensor fillTensorRandom(const ov::Output<ov::Node> &input) {
|
||||
template<typename T, typename U>
|
||||
ov::Tensor fillTensorRandom(T &input) {
|
||||
ov::Tensor tensor{input.get_element_type(), input.get_shape()};
|
||||
std::vector<T> values(ov::shape_size(input.get_shape()));
|
||||
for (size_t i = 0; i < values.size(); ++i) {
|
||||
values[i] = 1 + static_cast <T> (rand()) / (static_cast <T> (RAND_MAX / (std::numeric_limits<T>::max() - 1)));
|
||||
}
|
||||
std::memcpy(tensor.data(), values.data(), sizeof(T) * values.size());
|
||||
std::vector<U> values(ov::shape_size(input.get_shape()));
|
||||
|
||||
for (size_t i = 0; i < values.size(); ++i)
|
||||
values[i] = 1 + static_cast<U> (rand()) / (static_cast<U> (RAND_MAX / (std::numeric_limits<U>::max() - 1)));
|
||||
|
||||
std::memcpy(tensor.data(), values.data(), sizeof(U) * values.size());
|
||||
|
||||
return tensor;
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill InferenceEngine blob with image information (OV API 1.0)
|
||||
* @brief Fill InferenceEngine blob with image information (OV API 1)
|
||||
*/
|
||||
template<typename T>
|
||||
void fillBlobImInfo(InferenceEngine::Blob::Ptr &inputBlob,
|
||||
@@ -121,14 +125,39 @@ void fillBlobImInfo(InferenceEngine::Blob::Ptr &inputBlob,
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill InferRequest blobs with random values or image information
|
||||
* @brief Fill infer_request tensors with random values (OV API 2)
|
||||
*/
|
||||
template<typename T>
|
||||
void fillTensors(ov::InferRequest &infer_request, std::vector<T> &inputs) {
|
||||
for (size_t i = 0; i < inputs.size(); ++i) {
|
||||
ov::Tensor input_tensor;
|
||||
|
||||
if (inputs[i].get_element_type() == ov::element::f32) {
|
||||
input_tensor = fillTensorRandom<T, float>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::f16) {
|
||||
input_tensor = fillTensorRandom<T, short>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i32) {
|
||||
input_tensor = fillTensorRandom<T, int32_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::u8) {
|
||||
input_tensor = fillTensorRandom<T, uint8_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i8) {
|
||||
input_tensor = fillTensorRandom<T, int8_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::u16) {
|
||||
input_tensor = fillTensorRandom<T, uint16_t>(inputs[i]);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i16) {
|
||||
input_tensor = fillTensorRandom<T, int16_t>(inputs[i]);
|
||||
} else {
|
||||
throw std::logic_error(
|
||||
"Input precision is not supported for " + inputs[i].get_element_type().get_type_name());
|
||||
}
|
||||
infer_request.set_input_tensor(i, input_tensor);
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill InferRequest blobs with random values or image information (OV API 1)
|
||||
*/
|
||||
void fillBlobs(InferenceEngine::InferRequest inferRequest,
|
||||
const InferenceEngine::ConstInputsDataMap &inputsInfo,
|
||||
const size_t &batchSize);
|
||||
|
||||
/**
|
||||
* @brief Fill InferRequest tensors with random values or image information
|
||||
*/
|
||||
void fillTensors(ov::InferRequest &infer_request,
|
||||
const std::vector<ov::Output<ov::Node>> &inputs);
|
||||
const size_t &batchSize);
|
||||
+120
-55
@@ -1,70 +1,76 @@
|
||||
// Copyright (C) 2018-2021 Intel Corporation
|
||||
// Copyright (C) 2018-2022 Intel Corporation
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
|
||||
#include "common_utils.h"
|
||||
#include "reshape_utils.h"
|
||||
#include <inference_engine.hpp>
|
||||
|
||||
using namespace InferenceEngine;
|
||||
|
||||
|
||||
/**
|
||||
* @brief Parse data shapes for model
|
||||
*/
|
||||
std::map<std::string, std::vector<size_t>> parseDataShapes(const std::string& shapeString) {
|
||||
std::map<std::string, std::vector<size_t>> data_shapes;
|
||||
// Parse input parameter string
|
||||
std::vector<std::string> inputsShapes = split(shapeString, '&');
|
||||
std::map<std::string, std::vector<size_t>> parseDataShapes(const std::string &shapeString) {
|
||||
std::map<std::string, std::vector<size_t>> data_shapes;
|
||||
// Parse input parameter string
|
||||
std::vector<std::string> inputsShapes = split(shapeString, '&');
|
||||
|
||||
for (int i = 0; i < inputsShapes.size(); i++) {
|
||||
std::vector<std::string> curLayout = split(inputsShapes[i], ':');
|
||||
for (int i = 0; i < inputsShapes.size(); i++) {
|
||||
std::vector<std::string> curLayout = split(inputsShapes[i], '*');
|
||||
|
||||
std::string curLayoutName = curLayout.at(0);
|
||||
std::vector<size_t> shape;
|
||||
std::string curLayoutName = curLayout.at(0);
|
||||
std::vector<size_t> shape;
|
||||
|
||||
for (auto& dim : split(curLayout.at(1), ','))
|
||||
shape.emplace_back(std::stoi(dim));
|
||||
|
||||
data_shapes[curLayoutName] = shape;
|
||||
}
|
||||
return data_shapes;
|
||||
try {
|
||||
for (auto &dim: split(curLayout.at(1), ','))
|
||||
shape.emplace_back(std::stoi(dim));
|
||||
} catch (const std::exception &ex) {
|
||||
std::cerr << "Parsing data shapes failed with exception:\n"
|
||||
<< ex.what() << "\n";
|
||||
}
|
||||
data_shapes[curLayoutName] = shape;
|
||||
}
|
||||
return data_shapes;
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Parse input shapes for model reshape
|
||||
*/
|
||||
std::map<std::string, ov::PartialShape> parseReshapeShapes(const std::string& shapeString) {
|
||||
std::map<std::string, ov::PartialShape> reshape_info;
|
||||
// Parse input parameter string
|
||||
std::vector<std::string> inputsShapes = split(shapeString, '&');
|
||||
std::map<std::string, ov::PartialShape> parseReshapeShapes(const std::string &shapeString) {
|
||||
std::map<std::string, ov::PartialShape> reshape_info;
|
||||
// Parse input parameter string
|
||||
std::vector<std::string> inputsShapes = split(shapeString, '&');
|
||||
|
||||
for (int i = 0; i < inputsShapes.size(); i++) {
|
||||
std::vector<std::string> curLayout = split(inputsShapes[i], ':');
|
||||
for (int i = 0; i < inputsShapes.size(); i++) {
|
||||
std::vector<std::string> curLayout = split(inputsShapes[i], '*');
|
||||
|
||||
std::string curLayoutName = curLayout.at(0);
|
||||
std::vector<ov::Dimension> shape;
|
||||
std::string curLayoutName = curLayout.at(0);
|
||||
std::vector<ov::Dimension> shape;
|
||||
|
||||
for (auto& dim : split(curLayout.at(1), ',')) {
|
||||
if (dim == "?" || dim == "-1") {
|
||||
shape.emplace_back(ov::Dimension::dynamic());
|
||||
}
|
||||
else {
|
||||
const std::string range_divider = "..";
|
||||
size_t range_index = dim.find(range_divider);
|
||||
if (range_index != std::string::npos) {
|
||||
std::string min = dim.substr(0, range_index);
|
||||
std::string max = dim.substr(range_index + range_divider.length());
|
||||
shape.emplace_back(ov::Dimension(std::stoi(min), std::stoi(max)));
|
||||
} else {
|
||||
shape.emplace_back(ov::Dimension(std::stoi(dim)));
|
||||
for (auto& dim : split(curLayout.at(1), ',')) {
|
||||
try {
|
||||
if (dim == "?" || dim == "-1") {
|
||||
shape.emplace_back(ov::Dimension::dynamic());
|
||||
}
|
||||
else {
|
||||
const std::string range_divider = "..";
|
||||
size_t range_index = dim.find(range_divider);
|
||||
if (range_index != std::string::npos) {
|
||||
std::string min = dim.substr(0, range_index);
|
||||
std::string max = dim.substr(range_index + range_divider.length());
|
||||
shape.emplace_back(ov::Dimension(std::stoi(min), std::stoi(max)));
|
||||
} else {
|
||||
shape.emplace_back(ov::Dimension(std::stoi(dim)));
|
||||
}
|
||||
}
|
||||
} catch (const std::exception &ex) {
|
||||
std::cerr << "Parsing reshape shapes failed with exception:\n"
|
||||
<< ex.what() << "\n";
|
||||
}
|
||||
}
|
||||
}
|
||||
reshape_info[curLayoutName] = ov::PartialShape(shape);
|
||||
}
|
||||
reshape_info[curLayoutName] = ov::PartialShape(shape);
|
||||
}
|
||||
return reshape_info;
|
||||
return reshape_info;
|
||||
}
|
||||
|
||||
|
||||
@@ -72,7 +78,7 @@ std::map<std::string, ov::PartialShape> parseReshapeShapes(const std::string& sh
|
||||
* @brief Split input string using specified delimiter.
|
||||
Return vector with input tensor information
|
||||
*/
|
||||
std::vector<std::string> split(const std::string& s, char delim) {
|
||||
std::vector<std::string> split(const std::string &s, char delim) {
|
||||
std::vector<std::string> result;
|
||||
std::stringstream ss(s);
|
||||
std::string item;
|
||||
@@ -87,18 +93,77 @@ std::vector<std::string> split(const std::string& s, char delim) {
|
||||
/**
|
||||
* @brief Reshape blobs with dynamic shapes with static information from data shape
|
||||
*/
|
||||
void setStaticShapesBlobs(InferenceEngine::InferRequest inferRequest,
|
||||
const InferenceEngine::ConstInputsDataMap& inputsInfo,
|
||||
void setBlobsStaticShapes(InferenceEngine::InferRequest inferRequest,
|
||||
const InferenceEngine::ConstInputsDataMap &inputsInfo,
|
||||
std::map<std::string, std::vector<size_t>> dataShape) {
|
||||
for (const ConstInputsDataMap::value_type& item : inputsInfo) {
|
||||
Blob::Ptr inputBlob = inferRequest.GetBlob(item.first);
|
||||
for (const InferenceEngine::ConstInputsDataMap::value_type &item : inputsInfo) {
|
||||
InferenceEngine::Blob::Ptr inputBlob = inferRequest.GetBlob(item.first);
|
||||
|
||||
if (dataShape.count(item.first)) {
|
||||
SizeVector newInputShape;
|
||||
for (size_t i = 0; i < dataShape[item.first].size(); i++) {
|
||||
newInputShape.emplace_back(dataShape[item.first][i]);
|
||||
}
|
||||
inputBlob->setShape(newInputShape);
|
||||
if (dataShape.count(item.first)) {
|
||||
InferenceEngine::SizeVector newInputShape;
|
||||
for (size_t i = 0; i < dataShape[item.first].size(); i++) {
|
||||
newInputShape.emplace_back(dataShape[item.first][i]);
|
||||
}
|
||||
inputBlob->setShape(newInputShape);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Getting tensor shapes. If tensor is dynamic, static shape from data info will be returned.
|
||||
*/
|
||||
ov::Shape getTensorStaticShape(ov::Output<const ov::Node> &input,
|
||||
std::map<std::string, std::vector<size_t>> dataShape) {
|
||||
std::string name;
|
||||
try {
|
||||
name = input.get_any_name();
|
||||
} catch (const ov::Exception &iex) {
|
||||
// Attempt to get a name for a Tensor without names
|
||||
}
|
||||
|
||||
ov::Shape inputShape;
|
||||
|
||||
if (!input.get_partial_shape().is_dynamic()) {
|
||||
return input.get_shape();
|
||||
}
|
||||
else if (dataShape.count(name)) {
|
||||
for (size_t j = 0; j < dataShape[name].size(); j++)
|
||||
inputShape.emplace_back(dataShape[name][j]);
|
||||
}
|
||||
else {
|
||||
throw std::logic_error("Please provide static shape for " + name + "input using -data_shapes argument!");
|
||||
}
|
||||
return inputShape;
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill infer_request tensors with random values. The model shape is set separately. (OV API 2)
|
||||
*/
|
||||
void fillTensorsWithSpecifiedShape(ov::InferRequest& infer_request, std::vector<ov::Output<const ov::Node>> &inputs,
|
||||
std::map<std::string, std::vector<size_t>> dataShape) {
|
||||
for (size_t i = 0; i < inputs.size(); i++) {
|
||||
ov::Tensor input_tensor;
|
||||
ov::Shape inputShape = getTensorStaticShape(inputs[i], dataShape);
|
||||
|
||||
if (inputs[i].get_element_type() == ov::element::f32) {
|
||||
input_tensor = fillTensorRandomDynamic<float>(inputs[i], inputShape);
|
||||
} else if (inputs[i].get_element_type() == ov::element::f16) {
|
||||
input_tensor = fillTensorRandomDynamic<short>(inputs[i], inputShape);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i32) {
|
||||
input_tensor = fillTensorRandomDynamic<int32_t>(inputs[i], inputShape);
|
||||
} else if (inputs[i].get_element_type() == ov::element::u8) {
|
||||
input_tensor = fillTensorRandomDynamic<uint8_t>(inputs[i], inputShape);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i8) {
|
||||
input_tensor = fillTensorRandomDynamic<int8_t>(inputs[i], inputShape);
|
||||
} else if (inputs[i].get_element_type() == ov::element::u16) {
|
||||
input_tensor = fillTensorRandomDynamic<uint16_t>(inputs[i], inputShape);
|
||||
} else if (inputs[i].get_element_type() == ov::element::i16) {
|
||||
input_tensor = fillTensorRandomDynamic<int16_t>(inputs[i], inputShape);
|
||||
} else {
|
||||
throw std::logic_error("Input precision is not supported for " + inputs[i].get_element_type().get_type_name());
|
||||
}
|
||||
infer_request.set_input_tensor(i, input_tensor);
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -1,36 +1,66 @@
|
||||
// Copyright (C) 2018-2021 Intel Corporation
|
||||
// Copyright (C) 2018-2022 Intel Corporation
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <inference_engine.hpp>
|
||||
#include <openvino/openvino.hpp>
|
||||
|
||||
using namespace InferenceEngine;
|
||||
|
||||
/**
|
||||
* @brief Fill InferenceEngine tensor with random values. The model shape is set separately.
|
||||
*/
|
||||
template<typename T>
|
||||
ov::Tensor fillTensorRandomDynamic(ov::Output<const ov::Node> &input, ov::Shape shape) {
|
||||
ov::Tensor tensor {input.get_element_type(), shape};
|
||||
std::vector<T>values(ov::shape_size(shape));
|
||||
|
||||
for (size_t i = 0; i < values.size(); ++i) {
|
||||
values[i] = 1 + static_cast<T>(rand()) / (static_cast<T>(RAND_MAX / (std::numeric_limits<T>::max() - 1)));
|
||||
}
|
||||
std::memcpy(tensor.data(), values.data(), sizeof(T) * values.size());
|
||||
|
||||
return tensor;
|
||||
}
|
||||
|
||||
|
||||
/**
|
||||
* @brief Split input string using specified delimiter.
|
||||
Return vector with input tensor information
|
||||
*/
|
||||
std::vector<std::string> split(const std::string& s, char delim);
|
||||
std::vector<std::string> split(const std::string &s, char delim);
|
||||
|
||||
|
||||
/**
|
||||
* @brief Parse input shapes for model reshape
|
||||
*/
|
||||
std::map<std::string, ov::PartialShape> parseReshapeShapes(const std::string& shapeString);
|
||||
std::map<std::string, ov::PartialShape> parseReshapeShapes(const std::string &shapeString);
|
||||
|
||||
|
||||
/**
|
||||
* @brief Parse data shapes for model
|
||||
*/
|
||||
std::map<std::string, std::vector<size_t>> parseDataShapes(const std::string& shapeString);
|
||||
std::map<std::string, std::vector<size_t>> parseDataShapes(const std::string &shapeString);
|
||||
|
||||
|
||||
/**
|
||||
* @brief Getting tensor shapes. If tensor is dynamic, static shape from data info will be returned.
|
||||
*/
|
||||
ov::Shape getTensorStaticShape(ov::Output<const ov::Node> &input,
|
||||
std::map<std::string, std::vector<size_t>> dataShape);
|
||||
|
||||
|
||||
/**
|
||||
* @brief Reshape blobs with dynamic shapes with static information from data shape
|
||||
*/
|
||||
void setStaticShapesBlobs(InferenceEngine::InferRequest inferRequest,
|
||||
const InferenceEngine::ConstInputsDataMap& inputsInfo,
|
||||
std::map<std::string, std::vector<size_t>> dataShape);
|
||||
void setBlobsStaticShapes(InferenceEngine::InferRequest inferRequest,
|
||||
const InferenceEngine::ConstInputsDataMap &inputsInfo,
|
||||
std::map<std::string, std::vector<size_t>> dataShape);
|
||||
|
||||
|
||||
/**
|
||||
* @brief Fill infer_request tensors with random values. The model shape is set separately. (OV API 2)
|
||||
*/
|
||||
void fillTensorsWithSpecifiedShape(ov::InferRequest& infer_request, std::vector<ov::Output<const ov::Node>> &inputs,
|
||||
std::map<std::string, std::vector<size_t>> dataShape);
|
||||
@@ -8,24 +8,24 @@
|
||||
#include <string>
|
||||
|
||||
namespace TimeTest {
|
||||
using time_point = std::chrono::high_resolution_clock::time_point;
|
||||
using time_point = std::chrono::high_resolution_clock::time_point;
|
||||
|
||||
/** Encapsulate time measurements.
|
||||
Object of a class measures time at start and finish of object's life cycle.
|
||||
When destroyed, reports duration.
|
||||
*/
|
||||
class Timer {
|
||||
private:
|
||||
std::string name;
|
||||
time_point start_time;
|
||||
class Timer {
|
||||
private:
|
||||
std::string name;
|
||||
time_point start_time;
|
||||
|
||||
public:
|
||||
/// Constructs Timer object and measures start time.
|
||||
Timer(const std::string &timer_name);
|
||||
public:
|
||||
/// Constructs Timer object and measures start time.
|
||||
Timer(const std::string &timer_name);
|
||||
|
||||
/// Destructs Timer object, measures duration and reports it.
|
||||
~Timer();
|
||||
};
|
||||
/// Destructs Timer object, measures duration and reports it.
|
||||
~Timer();
|
||||
};
|
||||
|
||||
#define SCOPED_TIMER(timer_name) TimeTest::Timer timer_name(#timer_name);
|
||||
|
||||
|
||||
@@ -5,6 +5,7 @@
|
||||
#pragma once
|
||||
|
||||
#include <inference_engine.hpp>
|
||||
#include <openvino/openvino.hpp>
|
||||
#include <ie_plugin_config.hpp>
|
||||
|
||||
#include <string>
|
||||
@@ -22,17 +23,32 @@ std::string fileExt(const std::string &filename) {
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Function that enables Latency performance hint for specified device.
|
||||
* @brief Function that enables Latency performance hint for specified device (OV API 1)
|
||||
*/
|
||||
void setPerformanceConfig(InferenceEngine::Core ie, const std::string &device) {
|
||||
std::vector<std::string> supported_config_keys = ie.GetMetric(device, METRIC_KEY(SUPPORTED_CONFIG_KEYS));
|
||||
std::vector<std::string> supported_config_keys = ie.GetMetric(device, METRIC_KEY(SUPPORTED_CONFIG_KEYS));
|
||||
|
||||
if (std::find(supported_config_keys.begin(), supported_config_keys.end(), "PERFORMANCE_HINT") ==
|
||||
supported_config_keys.end()) {
|
||||
std::cerr << "Device " << device << " doesn't support config key 'PERFORMANCE_HINT'!\n"
|
||||
<< "Performance config was not set.";
|
||||
}
|
||||
else
|
||||
ie.SetConfig({{CONFIG_KEY(PERFORMANCE_HINT), CONFIG_VALUE(LATENCY)}}, device);
|
||||
if (std::find(supported_config_keys.begin(), supported_config_keys.end(), "PERFORMANCE_HINT") ==
|
||||
supported_config_keys.end()) {
|
||||
std::cerr << "Device " << device << " doesn't support config key 'PERFORMANCE_HINT'!\n"
|
||||
<< "Performance config was not set.";
|
||||
}
|
||||
else
|
||||
ie.SetConfig({{CONFIG_KEY(PERFORMANCE_HINT), CONFIG_VALUE(LATENCY)}}, device);
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Function that enables Latency performance hint for specified device (OV API 2)
|
||||
*/
|
||||
void setPerformanceConfig(ov::Core ie, const std::string &device) {
|
||||
std::vector<std::string> supported_config_keys = ie.get_property(device, METRIC_KEY(SUPPORTED_CONFIG_KEYS));
|
||||
|
||||
if (std::find(supported_config_keys.begin(), supported_config_keys.end(), "PERFORMANCE_HINT") ==
|
||||
supported_config_keys.end()) {
|
||||
std::cerr << "Device " << device << " doesn't support config key 'PERFORMANCE_HINT'!\n"
|
||||
<< "Performance config was not set.";
|
||||
}
|
||||
else
|
||||
ie.set_property(device, {{CONFIG_KEY(PERFORMANCE_HINT), CONFIG_VALUE(LATENCY)}});
|
||||
}
|
||||
}
|
||||
@@ -2,7 +2,6 @@
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
|
||||
#include <inference_engine.hpp>
|
||||
#include <iostream>
|
||||
|
||||
#include "common_utils.h"
|
||||
@@ -10,8 +9,6 @@
|
||||
#include "timetests_helper/timer.h"
|
||||
#include "timetests_helper/utils.h"
|
||||
|
||||
using namespace InferenceEngine;
|
||||
|
||||
|
||||
/**
|
||||
* @brief Function that contain executable pipeline which will be called from
|
||||
@@ -21,96 +18,94 @@ using namespace InferenceEngine;
|
||||
int runPipeline(const std::string &model, const std::string &device, const bool isCacheEnabled,
|
||||
std::map<std::string, ov::PartialShape> reshapeShapes,
|
||||
std::map<std::string, std::vector<size_t>> dataShapes) {
|
||||
auto pipeline = [](const std::string &model, const std::string &device, const bool isCacheEnabled,
|
||||
std::map<std::string, ov::PartialShape> reshapeShapes,
|
||||
std::map<std::string, std::vector<size_t>> dataShapes) {
|
||||
Core ie;
|
||||
CNNNetwork cnnNetwork;
|
||||
ExecutableNetwork exeNetwork;
|
||||
InferRequest inferRequest;
|
||||
size_t batchSize = 0;
|
||||
auto pipeline = [](const std::string &model, const std::string &device, const bool isCacheEnabled,
|
||||
std::map<std::string, ov::PartialShape> reshapeShapes,
|
||||
std::map<std::string, std::vector<size_t>> dataShapes) {
|
||||
InferenceEngine::Core ie;
|
||||
InferenceEngine::CNNNetwork cnnNetwork;
|
||||
InferenceEngine::ExecutableNetwork exeNetwork;
|
||||
InferenceEngine::InferRequest inferRequest;
|
||||
size_t batchSize = 0;
|
||||
|
||||
bool reshape = false;
|
||||
if (!reshapeShapes.empty()) {
|
||||
reshape = true;
|
||||
}
|
||||
bool reshape = false;
|
||||
if (!reshapeShapes.empty()) {
|
||||
reshape = true;
|
||||
}
|
||||
|
||||
// first_inference_latency = time_to_inference + first_inference
|
||||
{
|
||||
SCOPED_TIMER(time_to_inference);
|
||||
{
|
||||
SCOPED_TIMER(load_plugin);
|
||||
ie.GetVersions(device);
|
||||
}
|
||||
// Set performance and model cache configs
|
||||
TimeTest::setPerformanceConfig(ie, device);
|
||||
if (isCacheEnabled)
|
||||
ie.SetConfig({{CONFIG_KEY(CACHE_DIR), "models_cache"}});
|
||||
{
|
||||
SCOPED_TIMER(create_exenetwork);
|
||||
if (!isCacheEnabled) {
|
||||
if (TimeTest::fileExt(model) == "blob") {
|
||||
SCOPED_TIMER(import_network);
|
||||
exeNetwork = ie.ImportNetwork(model, device);
|
||||
}
|
||||
else {
|
||||
// first_inference_latency = time_to_inference + first_inference
|
||||
{
|
||||
SCOPED_TIMER(time_to_inference);
|
||||
{
|
||||
SCOPED_TIMER(read_network);
|
||||
cnnNetwork = ie.ReadNetwork(model);
|
||||
batchSize = cnnNetwork.getBatchSize();
|
||||
}
|
||||
if (reshape) {
|
||||
{
|
||||
SCOPED_TIMER(reshape);
|
||||
cnnNetwork.reshape(reshapeShapes);
|
||||
}
|
||||
SCOPED_TIMER(load_plugin);
|
||||
TimeTest::setPerformanceConfig(ie, device);
|
||||
ie.GetVersions(device);
|
||||
|
||||
if (isCacheEnabled)
|
||||
ie.SetConfig({{CONFIG_KEY(CACHE_DIR), "models_cache"}});
|
||||
}
|
||||
{
|
||||
SCOPED_TIMER(load_network);
|
||||
exeNetwork = ie.LoadNetwork(cnnNetwork, device);
|
||||
SCOPED_TIMER(create_exenetwork);
|
||||
if (!isCacheEnabled) {
|
||||
if (TimeTest::fileExt(model) == "blob") {
|
||||
SCOPED_TIMER(import_network);
|
||||
exeNetwork = ie.ImportNetwork(model, device);
|
||||
}
|
||||
else {
|
||||
{
|
||||
SCOPED_TIMER(read_network);
|
||||
cnnNetwork = ie.ReadNetwork(model);
|
||||
batchSize = cnnNetwork.getBatchSize();
|
||||
}
|
||||
if (reshape) {
|
||||
{
|
||||
SCOPED_TIMER(reshape);
|
||||
cnnNetwork.reshape(reshapeShapes);
|
||||
}
|
||||
}
|
||||
{
|
||||
SCOPED_TIMER(load_network);
|
||||
exeNetwork = ie.LoadNetwork(cnnNetwork, device);
|
||||
}
|
||||
}
|
||||
}
|
||||
else {
|
||||
SCOPED_TIMER(load_network_cache);
|
||||
exeNetwork = ie.LoadNetwork(model, device);
|
||||
}
|
||||
}
|
||||
}
|
||||
inferRequest = exeNetwork.CreateInferRequest();
|
||||
}
|
||||
else {
|
||||
SCOPED_TIMER(load_network_cache);
|
||||
exeNetwork = ie.LoadNetwork(model, device);
|
||||
{
|
||||
SCOPED_TIMER(first_inference);
|
||||
{
|
||||
SCOPED_TIMER(fill_inputs);
|
||||
const InferenceEngine::ConstInputsDataMap inputsInfo(exeNetwork.GetInputsInfo());
|
||||
batchSize = batchSize != 0 ? batchSize : 1;
|
||||
if (reshape) {
|
||||
setBlobsStaticShapes(inferRequest, inputsInfo, dataShapes);
|
||||
fillBlobs(inferRequest, inputsInfo, batchSize);
|
||||
} else {
|
||||
fillBlobs(inferRequest, inputsInfo, batchSize);
|
||||
}
|
||||
}
|
||||
inferRequest.Infer();
|
||||
}
|
||||
}
|
||||
inferRequest = exeNetwork.CreateInferRequest();
|
||||
};
|
||||
|
||||
try {
|
||||
pipeline(model, device, isCacheEnabled, reshapeShapes, dataShapes);
|
||||
} catch (const InferenceEngine::Exception &iex) {
|
||||
std::cerr
|
||||
<< "Inference Engine pipeline failed with Inference Engine exception:\n"
|
||||
<< iex.what();
|
||||
return 1;
|
||||
} catch (const std::exception &ex) {
|
||||
std::cerr << "Inference Engine pipeline failed with exception:\n"
|
||||
<< ex.what();
|
||||
return 2;
|
||||
} catch (...) {
|
||||
std::cerr << "Inference Engine pipeline failed\n";
|
||||
return 3;
|
||||
}
|
||||
{
|
||||
SCOPED_TIMER(first_inference);
|
||||
{
|
||||
SCOPED_TIMER(fill_inputs);
|
||||
|
||||
const InferenceEngine::ConstInputsDataMap inputsInfo(exeNetwork.GetInputsInfo());
|
||||
batchSize = batchSize != 0 ? batchSize : 1;
|
||||
|
||||
if (reshape) {
|
||||
setStaticShapesBlobs(inferRequest, inputsInfo, dataShapes);
|
||||
fillBlobs(inferRequest, inputsInfo, batchSize);
|
||||
} else {
|
||||
fillBlobs(inferRequest, inputsInfo, batchSize);
|
||||
}
|
||||
}
|
||||
inferRequest.Infer();
|
||||
}
|
||||
};
|
||||
|
||||
try {
|
||||
pipeline(model, device, isCacheEnabled, reshapeShapes, dataShapes);
|
||||
} catch (const InferenceEngine::Exception &iex) {
|
||||
std::cerr
|
||||
<< "Inference Engine pipeline failed with Inference Engine exception:\n"
|
||||
<< iex.what();
|
||||
return 1;
|
||||
} catch (const std::exception &ex) {
|
||||
std::cerr << "Inference Engine pipeline failed with exception:\n"
|
||||
<< ex.what();
|
||||
return 2;
|
||||
} catch (...) {
|
||||
std::cerr << "Inference Engine pipeline failed\n";
|
||||
return 3;
|
||||
}
|
||||
return 0;
|
||||
return 0;
|
||||
}
|
||||
@@ -0,0 +1,109 @@
|
||||
// Copyright (C) 2018-2021 Intel Corporation
|
||||
// SPDX-License-Identifier: Apache-2.0
|
||||
//
|
||||
#include <openvino/runtime/core.hpp>
|
||||
|
||||
#include <fstream>
|
||||
|
||||
#include "common_utils.h"
|
||||
#include "reshape_utils.h"
|
||||
#include "timetests_helper/timer.h"
|
||||
#include "timetests_helper/utils.h"
|
||||
|
||||
|
||||
/**
|
||||
* @brief Function that contain executable pipeline which will be called from
|
||||
* main(). The function should not throw any exceptions and responsible for
|
||||
* handling it by itself.
|
||||
*/
|
||||
int runPipeline(const std::string &model, const std::string &device, const bool isCacheEnabled,
|
||||
std::map<std::string, ov::PartialShape> reshapeShapes,
|
||||
std::map<std::string, std::vector<size_t>> dataShapes) {
|
||||
auto pipeline = [](const std::string &model, const std::string &device, const bool isCacheEnabled,
|
||||
std::map<std::string, ov::PartialShape> reshapeShapes,
|
||||
std::map<std::string, std::vector<size_t>> dataShapes) {
|
||||
ov::Core ie;
|
||||
std::shared_ptr<ov::Model> cnnNetwork;
|
||||
ov::CompiledModel exeNetwork;
|
||||
ov::InferRequest inferRequest;
|
||||
|
||||
bool reshape = false;
|
||||
if (!reshapeShapes.empty()) {
|
||||
reshape = true;
|
||||
}
|
||||
|
||||
// first_inference_latency = time_to_inference + first_inference
|
||||
{
|
||||
SCOPED_TIMER(time_to_inference);
|
||||
{
|
||||
SCOPED_TIMER(load_plugin);
|
||||
TimeTest::setPerformanceConfig(ie, device);
|
||||
ie.get_versions(device);
|
||||
|
||||
if (isCacheEnabled)
|
||||
ie.set_property({{CONFIG_KEY(CACHE_DIR), "models_cache"}});
|
||||
}
|
||||
{
|
||||
SCOPED_TIMER(create_exenetwork);
|
||||
if (!isCacheEnabled) {
|
||||
if (TimeTest::fileExt(model) == "blob") {
|
||||
SCOPED_TIMER(import_network);
|
||||
std::ifstream streamModel{model};
|
||||
exeNetwork = ie.import_model(streamModel, device);
|
||||
}
|
||||
else {
|
||||
{
|
||||
SCOPED_TIMER(read_network);
|
||||
cnnNetwork = ie.read_model(model);
|
||||
}
|
||||
if (reshape) {
|
||||
{
|
||||
SCOPED_TIMER(reshape);
|
||||
cnnNetwork->reshape(reshapeShapes);
|
||||
}
|
||||
}
|
||||
{
|
||||
SCOPED_TIMER(load_network);
|
||||
exeNetwork = ie.compile_model(cnnNetwork, device);
|
||||
}
|
||||
}
|
||||
}
|
||||
else {
|
||||
SCOPED_TIMER(load_network_cache);
|
||||
exeNetwork = ie.compile_model(model, device);
|
||||
}
|
||||
}
|
||||
inferRequest = exeNetwork.create_infer_request();
|
||||
}
|
||||
{
|
||||
SCOPED_TIMER(first_inference);
|
||||
{
|
||||
SCOPED_TIMER(fill_inputs);
|
||||
std::vector<ov::Output<const ov::Node>> inputs = exeNetwork.inputs();
|
||||
if (reshape) {
|
||||
fillTensorsWithSpecifiedShape(inferRequest, inputs,dataShapes);
|
||||
} else {
|
||||
fillTensors(inferRequest, inputs);
|
||||
}
|
||||
}
|
||||
inferRequest.infer();
|
||||
}
|
||||
};
|
||||
|
||||
try {
|
||||
pipeline(model, device, isCacheEnabled, reshapeShapes, dataShapes);
|
||||
} catch (const ov::Exception &iex) {
|
||||
std::cerr
|
||||
<< "Inference Engine pipeline failed with Inference Engine exception:\n"
|
||||
<< iex.what();
|
||||
return 1;
|
||||
} catch (const std::exception &ex) {
|
||||
std::cerr << "Inference Engine pipeline failed with exception:\n"
|
||||
<< ex.what();
|
||||
return 2;
|
||||
} catch (...) {
|
||||
std::cerr << "Inference Engine pipeline failed\n";
|
||||
return 3;
|
||||
}
|
||||
return 0;
|
||||
}
|
||||
@@ -11,40 +11,40 @@
|
||||
|
||||
/// @brief message for help argument
|
||||
static const char help_message[] =
|
||||
"Print a usage message.";
|
||||
"Print a usage message.";
|
||||
|
||||
/// @brief message for model argument
|
||||
static const char model_message[] =
|
||||
"Required. Path to an .xml/.onnx file with a trained model or to "
|
||||
"a .blob files with a trained compiled model.";
|
||||
"Required. Path to an .xml/.onnx file with a trained model or to "
|
||||
"a .blob files with a trained compiled model.";
|
||||
|
||||
/// @brief message for target device argument
|
||||
static const char target_device_message[] =
|
||||
"Required. Specify a target device to infer on. \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.";
|
||||
"Required. Specify a target device to infer on. \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.";
|
||||
|
||||
/// @brief message for cache argument
|
||||
static const char model_cache_message[] =
|
||||
"Not required. Use this key to run timetests with models caching. \n"
|
||||
"TimeInfer executable should be run twice - the second run will use cache prepared from first run.";
|
||||
"Not required. Use this key to run timetests with models caching. \n"
|
||||
"TimeInfer executable should be run twice - the second run will use cache prepared from first run.";
|
||||
|
||||
/// @brief message for shapes argument
|
||||
static const char reshape_shapes_message[] =
|
||||
"Not required. Use this key to run timetests with reshape. \n"
|
||||
"Example: 'input:1..2 3 100 100'. Use '&' delimiter for several inputs. Example: 'input1:1..2 100&input2:1..2 100' ";
|
||||
"Not required. Use this key to run timetests with reshape. \n"
|
||||
"Example: 'input*1..2 3 100 100'. Use '&' delimiter for several inputs. Example: 'input1*1..2 100&input2*1..2 100' ";
|
||||
|
||||
/// @brief message for shapes argument
|
||||
static const char data_shapes_message[] =
|
||||
"Not required. Use this key to run timetests with reshape. Used with 'reshape_shapes' arg. \n"
|
||||
"Only static shapes for data. Example: 'input:1 3 100 100'. Use '&' delimiter for several inputs. Example: 'input1:1 100&input2:1 100' ";
|
||||
"Not required. Use this key to run timetests with reshape. Used with 'reshape_shapes' arg. \n"
|
||||
"Only static shapes for data. Example: 'input*1 3 100 100'. Use '&' delimiter for several inputs. Example: 'input1*1 100&input2*1 100' ";
|
||||
|
||||
/// @brief message for statistics path argument
|
||||
static const char statistics_path_message[] =
|
||||
"Required. Path to a file to write statistics.";
|
||||
"Required. Path to a file to write statistics.";
|
||||
|
||||
/// @brief Define flag for showing help message <br>
|
||||
DEFINE_bool(h, false, help_message);
|
||||
@@ -80,15 +80,15 @@ DEFINE_string(s, "", statistics_path_message);
|
||||
* @brief This function show a help message
|
||||
*/
|
||||
static void showUsage() {
|
||||
std::cout << std::endl;
|
||||
std::cout << "TimeInfer [OPTION]" << std::endl;
|
||||
std::cout << "Options:" << std::endl;
|
||||
std::cout << std::endl;
|
||||
std::cout << " -h, --help " << help_message << std::endl;
|
||||
std::cout << " -m \"<path>\" " << model_message << std::endl;
|
||||
std::cout << " -d \"<device>\" " << target_device_message << std::endl;
|
||||
std::cout << " -s \"<path>\" " << statistics_path_message << std::endl;
|
||||
std::cout << " -c " << model_cache_message << std::endl;
|
||||
std::cout << " -reshape_shapes " << reshape_shapes_message << std::endl;
|
||||
std::cout << " -data_shapes " << data_shapes_message << std::endl;
|
||||
std::cout << std::endl;
|
||||
std::cout << "TimeInfer [OPTION]" << std::endl;
|
||||
std::cout << "Options:" << std::endl;
|
||||
std::cout << std::endl;
|
||||
std::cout << " -h, --help " << help_message << std::endl;
|
||||
std::cout << " -m \"<path>\" " << model_message << std::endl;
|
||||
std::cout << " -d \"<device>\" " << target_device_message << std::endl;
|
||||
std::cout << " -s \"<path>\" " << statistics_path_message << std::endl;
|
||||
std::cout << " -c " << model_cache_message << std::endl;
|
||||
std::cout << " -reshape_shapes " << reshape_shapes_message << std::endl;
|
||||
std::cout << " -data_shapes " << data_shapes_message << std::endl;
|
||||
}
|
||||
@@ -18,29 +18,29 @@ int runPipeline(const std::string &model, const std::string &device, const bool
|
||||
* @brief Parses command line and check required arguments
|
||||
*/
|
||||
bool parseAndCheckCommandLine(int argc, char **argv) {
|
||||
gflags::ParseCommandLineNonHelpFlags(&argc, &argv, true);
|
||||
if (FLAGS_help || FLAGS_h) {
|
||||
showUsage();
|
||||
return false;
|
||||
}
|
||||
gflags::ParseCommandLineNonHelpFlags(&argc, &argv, true);
|
||||
if (FLAGS_help || FLAGS_h) {
|
||||
showUsage();
|
||||
return false;
|
||||
}
|
||||
|
||||
if (FLAGS_m.empty())
|
||||
throw std::logic_error(
|
||||
"Model is required but not set. Please set -m option.");
|
||||
if (FLAGS_m.empty())
|
||||
throw std::logic_error(
|
||||
"Model is required but not set. Please set -m option.");
|
||||
|
||||
if (FLAGS_d.empty())
|
||||
throw std::logic_error(
|
||||
"Device is required but not set. Please set -d option.");
|
||||
if (FLAGS_d.empty())
|
||||
throw std::logic_error(
|
||||
"Device is required but not set. Please set -d option.");
|
||||
|
||||
if (FLAGS_s.empty())
|
||||
throw std::logic_error(
|
||||
"Statistics file path is required but not set. Please set -s option.");
|
||||
if (FLAGS_s.empty())
|
||||
throw std::logic_error(
|
||||
"Statistics file path is required but not set. Please set -s option.");
|
||||
|
||||
if (!FLAGS_reshape_shapes.empty() && FLAGS_data_shapes.empty())
|
||||
throw std::logic_error(
|
||||
"Data shapes is required for reshape shapes argument. Please set -data_shapes option.");
|
||||
if (!FLAGS_reshape_shapes.empty() && FLAGS_data_shapes.empty())
|
||||
throw std::logic_error(
|
||||
"Data shapes is required for reshape shapes argument. Please set -data_shapes option.");
|
||||
|
||||
return true;
|
||||
return true;
|
||||
}
|
||||
|
||||
/**
|
||||
@@ -48,22 +48,22 @@ bool parseAndCheckCommandLine(int argc, char **argv) {
|
||||
*/
|
||||
int _runPipeline(std::map<std::string, ov::PartialShape> dynamicShapes,
|
||||
std::map<std::string, std::vector<size_t>> staticShapes) {
|
||||
SCOPED_TIMER(full_run);
|
||||
return runPipeline(FLAGS_m, FLAGS_d, FLAGS_c, dynamicShapes, staticShapes);
|
||||
SCOPED_TIMER(full_run);
|
||||
return runPipeline(FLAGS_m, FLAGS_d, FLAGS_c, dynamicShapes, staticShapes);
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Main entry point
|
||||
*/
|
||||
int main(int argc, char **argv) {
|
||||
if (!parseAndCheckCommandLine(argc, argv))
|
||||
return -1;
|
||||
if (!parseAndCheckCommandLine(argc, argv))
|
||||
return -1;
|
||||
|
||||
auto dynamicShapes = parseReshapeShapes(FLAGS_reshape_shapes);
|
||||
auto staticShapes = parseDataShapes(FLAGS_data_shapes);
|
||||
auto dynamicShapes = parseReshapeShapes(FLAGS_reshape_shapes);
|
||||
auto staticShapes = parseDataShapes(FLAGS_data_shapes);
|
||||
|
||||
auto status = _runPipeline(dynamicShapes, staticShapes);
|
||||
StatisticsWriter::Instance().setFile(FLAGS_s);
|
||||
StatisticsWriter::Instance().write();
|
||||
return status;
|
||||
auto status = _runPipeline(dynamicShapes, staticShapes);
|
||||
StatisticsWriter::Instance().setFile(FLAGS_s);
|
||||
StatisticsWriter::Instance().write();
|
||||
return status;
|
||||
}
|
||||
@@ -22,62 +22,62 @@
|
||||
*/
|
||||
class StatisticsWriter {
|
||||
private:
|
||||
std::ofstream statistics_file;
|
||||
std::ofstream statistics_file;
|
||||
|
||||
std::map<std::string, std::pair<int, float>> time_structure; // timer_name, <tab number, duration>
|
||||
std::vector<std::string> time_struct_order;
|
||||
int tab_count = 0;
|
||||
std::map<std::string, std::pair<int, float>> time_structure; // timer_name, <tab number, duration>
|
||||
std::vector<std::string> time_struct_order;
|
||||
int tab_count = 0;
|
||||
|
||||
StatisticsWriter() = default;
|
||||
StatisticsWriter(const StatisticsWriter &) = delete;
|
||||
StatisticsWriter &operator=(const StatisticsWriter &) = delete;
|
||||
StatisticsWriter() = default;
|
||||
StatisticsWriter(const StatisticsWriter &) = delete;
|
||||
StatisticsWriter &operator=(const StatisticsWriter &) = delete;
|
||||
|
||||
public:
|
||||
/**
|
||||
* @brief Creates StatisticsWriter singleton object
|
||||
*/
|
||||
static StatisticsWriter &Instance() {
|
||||
/**
|
||||
* @brief Creates StatisticsWriter singleton object
|
||||
*/
|
||||
static StatisticsWriter &Instance() {
|
||||
static StatisticsWriter writer;
|
||||
return writer;
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Specifies, opens and validates statistics path for writing
|
||||
*/
|
||||
void setFile(const std::string &statistics_path) {
|
||||
/**
|
||||
* @brief Specifies, opens and validates statistics path for writing
|
||||
*/
|
||||
void setFile(const std::string &statistics_path) {
|
||||
statistics_file.open(statistics_path);
|
||||
if (!statistics_file.good()) {
|
||||
std::stringstream err;
|
||||
err << "Statistic file \"" << statistics_path
|
||||
<< "\" can't be used for writing";
|
||||
throw std::runtime_error(err.str());
|
||||
std::stringstream err;
|
||||
err << "Statistic file \"" << statistics_path
|
||||
<< "\" can't be used for writing";
|
||||
throw std::runtime_error(err.str());
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Compute order for statistics operations.
|
||||
*/
|
||||
void addTimer(const std::string name) {
|
||||
/**
|
||||
* @brief Compute order for statistics operations.
|
||||
*/
|
||||
void addTimer(const std::string name) {
|
||||
time_struct_order.push_back(name);
|
||||
tab_count++;
|
||||
}
|
||||
}
|
||||
|
||||
void deleteTimer(const std::pair<std::string, float> &record) {
|
||||
void deleteTimer(const std::pair<std::string, float> &record) {
|
||||
tab_count--;
|
||||
time_structure[record.first] = std::make_pair(tab_count, record.second);
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* @brief Writes provided statistics in YAML format.
|
||||
*/
|
||||
void write() {
|
||||
/**
|
||||
* @brief Writes provided statistics in YAML format.
|
||||
*/
|
||||
void write() {
|
||||
if (!statistics_file)
|
||||
throw std::runtime_error("Statistic file path isn't set");
|
||||
throw std::runtime_error("Statistic file path isn't set");
|
||||
for (auto& timer: time_struct_order) {
|
||||
std::string tabs = std::string(TAB* time_structure[timer].first, ' ');
|
||||
statistics_file << tabs << "- " << timer << ":" << '\n'
|
||||
<< tabs << " " << "- " << time_structure[timer].second << '\n';
|
||||
std::string tabs = std::string(TAB* time_structure[timer].first, ' ');
|
||||
statistics_file << tabs << "- " << timer << ":" << '\n'
|
||||
<< tabs << " " << "- " << time_structure[timer].second << '\n';
|
||||
}
|
||||
statistics_file << "---" << '\n' << "measurement_unit: microsecs";
|
||||
}
|
||||
}
|
||||
};
|
||||
@@ -14,18 +14,17 @@ using time_point = std::chrono::high_resolution_clock::time_point;
|
||||
|
||||
namespace TimeTest {
|
||||
|
||||
|
||||
Timer::Timer(const std::string &timer_name) {
|
||||
name = timer_name;
|
||||
start_time = std::chrono::high_resolution_clock::now();
|
||||
StatisticsWriter::Instance().addTimer(name);
|
||||
name = timer_name;
|
||||
start_time = std::chrono::high_resolution_clock::now();
|
||||
StatisticsWriter::Instance().addTimer(name);
|
||||
}
|
||||
|
||||
Timer::~Timer() {
|
||||
float duration = std::chrono::duration_cast<std::chrono::microseconds>(
|
||||
std::chrono::high_resolution_clock::now() - start_time)
|
||||
.count();
|
||||
StatisticsWriter::Instance().deleteTimer({name, duration});
|
||||
float duration = std::chrono::duration_cast<std::chrono::microseconds>(
|
||||
std::chrono::high_resolution_clock::now() - start_time)
|
||||
.count();
|
||||
StatisticsWriter::Instance().deleteTimer({name, duration});
|
||||
}
|
||||
|
||||
} // namespace TimeTest
|
||||
Reference in new issue
Block a user