IntenseWebs/openvino
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[GPU] Removed legacy public gpu config and related processing (#14707)

Browse Source
This commit is contained in:
Vladimir Paramuzov 2022-12-19 11:19:07 +04:00 committed by GitHub
parent 787ba3de4f
commit ee256e801c
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
10 changed files with 11 additions and 294 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

130
src/inference/include/ie/cldnn/cldnn_config.hpp
View File

@ -1,130 +0,0 @@
// Copyright (C) 2018-2022 Intel Corporation
// SPDX-License-Identifier: Apache-2.0
//
/**
* @brief A header for advanced hardware related properties for clDNN plugin
* To use in SetConfig() method of plugins
*
* @file cldnn_config.hpp
*/
#pragma once
#include "gpu/gpu_config.hpp"
#include "ie_api.h"
#include "ie_plugin_config.hpp"
namespace InferenceEngine {
/**
* @brief GPU plugin configuration
*/
namespace CLDNNConfigParams {
/**
* @brief shortcut for defining configuration keys
*/
#define CLDNN_CONFIG_KEY(name) InferenceEngine::CLDNNConfigParams::_CONFIG_KEY(CLDNN_##name)
#define DECLARE_CLDNN_CONFIG_KEY(name) DECLARE_CONFIG_KEY(CLDNN_##name)
#define DECLARE_CLDNN_CONFIG_VALUE(name) DECLARE_CONFIG_VALUE(CLDNN_##name)
/**
* @brief This key instructs the clDNN plugin to use the OpenCL queue priority hint
* as defined in https://www.khronos.org/registry/OpenCL/specs/opencl-2.1-extensions.pdf
* this option should be used with an unsigned integer value (1 is lowest priority)
* 0 means no priority hint is set and default queue is created.
*/
INFERENCE_ENGINE_DEPRECATED("Use InferenceEngine::GPUConfigParams::GPU_PLUGIN_PRIORITY instead")
DECLARE_CLDNN_CONFIG_KEY(PLUGIN_PRIORITY);
/**
* @brief This key instructs the clDNN plugin to use throttle hints the OpenCL queue throttle hint
* as defined in https://www.khronos.org/registry/OpenCL/specs/opencl-2.1-extensions.pdf,
* chapter 9.19. This option should be used with an unsigned integer value (1 is lowest energy consumption)
* 0 means no throttle hint is set and default queue created.
*/
INFERENCE_ENGINE_DEPRECATED("Use InferenceEngine::GPUConfigParams::GPU_PLUGIN_THROTTLE instead")
DECLARE_CLDNN_CONFIG_KEY(PLUGIN_THROTTLE);
/**
* @brief This key controls clDNN memory pool optimization.
* Turned off by default.
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CLDNN_CONFIG_KEY(MEM_POOL);
/**
* @brief This key defines the directory name to which clDNN graph visualization will be dumped.
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CLDNN_CONFIG_KEY(GRAPH_DUMPS_DIR);
/**
* @brief This key defines the directory name to which full program sources will be dumped.
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CLDNN_CONFIG_KEY(SOURCES_DUMPS_DIR);
/**
* @brief This key enables FP16 precision for quantized models.
* By default the model is converted to FP32 precision before running LPT. If this key is enabled (default), then
* non-quantized layers will be converted back to FP16 after LPT, which might imrpove the performance if a model has a
* lot of compute operations in non-quantized path. This key has no effect if current device doesn't have INT8
* optimization capabilities.
*/
DECLARE_CLDNN_CONFIG_KEY(ENABLE_FP16_FOR_QUANTIZED_MODELS);
/**
* @brief This key should be set to correctly handle NV12 input without pre-processing.
* Turned off by default.
*/
INFERENCE_ENGINE_DEPRECATED("Use InferenceEngine::GPUConfigParams::GPU_NV12_TWO_INPUTS instead")
DECLARE_CLDNN_CONFIG_KEY(NV12_TWO_INPUTS);
} // namespace CLDNNConfigParams
namespace PluginConfigParams {
/**
* @brief This key enables dumping of the kernels used by the plugin for custom layers.
*
* This option should be used with values: PluginConfigParams::YES or PluginConfigParams::NO (default)
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CONFIG_KEY(DUMP_KERNELS);
/**
* @brief This key controls performance tuning done or used by the plugin.
*
* This option should be used with values:
* PluginConfigParams::TUNING_DISABLED (default)
* PluginConfigParams::TUNING_USE_EXISTING - use existing data from tuning file
* PluginConfigParams::TUNING_CREATE - create tuning data for parameters not present in tuning file
* PluginConfigParams::TUNING_UPDATE - perform non-tuning updates like removal of invalid/deprecated data
* PluginConfigParams::TUNING_RETUNE - create tuning data for all parameters, even if already present
*
* For values TUNING_CREATE and TUNING_RETUNE the file will be created if it does not exist.
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CONFIG_KEY(TUNING_MODE);
INFERENCE_ENGINE_DEPRECATED("The config value will be removed")
DECLARE_CONFIG_VALUE(TUNING_CREATE);
INFERENCE_ENGINE_DEPRECATED("The config value will be removed")
DECLARE_CONFIG_VALUE(TUNING_USE_EXISTING);
INFERENCE_ENGINE_DEPRECATED("The config value will be removed")
DECLARE_CONFIG_VALUE(TUNING_DISABLED);
INFERENCE_ENGINE_DEPRECATED("The config value will be removed")
DECLARE_CONFIG_VALUE(TUNING_UPDATE);
INFERENCE_ENGINE_DEPRECATED("The config value will be removed")
DECLARE_CONFIG_VALUE(TUNING_RETUNE);
/**
* @brief This key defines the tuning data filename to be created/used
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CONFIG_KEY(TUNING_FILE);
} // namespace PluginConfigParams
} // namespace InferenceEngine

2
src/inference/include/ie/ie_plugin_config.hpp
View File

@ -475,9 +475,9 @@ DECLARE_CONFIG_KEY(FORCE_TBB_TERMINATE);
} // namespace InferenceEngine } // namespace InferenceEngine
#include "gpu/gpu_config.hpp"
#include "hetero/hetero_plugin_config.hpp" #include "hetero/hetero_plugin_config.hpp"
#include "multi-device/multi_device_config.hpp" #include "multi-device/multi_device_config.hpp"
// remove in 2022.1 major release // remove in 2022.1 major release
#include "cldnn/cldnn_config.hpp"
#include "gna/gna_config.hpp" #include "gna/gna_config.hpp"

10
src/plugins/intel_gpu/include/intel_gpu/plugin/device_config.hpp
View File

@ -24,18 +24,13 @@ struct Config {
useProfiling(false), useProfiling(false),
dumpCustomKernels(false), dumpCustomKernels(false),
exclusiveAsyncRequests(false), exclusiveAsyncRequests(false),
memory_pool_on(true),
enableDynamicBatch(false), enableDynamicBatch(false),
enableInt8(true), enableInt8(true),
nv12_two_inputs(false), nv12_two_inputs(false),
enable_fp16_for_quantized_models(true),
queuePriority(cldnn::priority_mode_types::med), queuePriority(cldnn::priority_mode_types::med),
queueThrottle(cldnn::throttle_mode_types::med), queueThrottle(cldnn::throttle_mode_types::med),
max_dynamic_batch(1), max_dynamic_batch(1),
customLayers({}), customLayers({}),
tuningConfig(),
graph_dumps_dir(""),
sources_dumps_dir(""),
kernels_cache_dir(""), kernels_cache_dir(""),
inference_precision(ov::element::undefined), inference_precision(ov::element::undefined),
task_exec_config({"GPU plugin internal task executor", // name task_exec_config({"GPU plugin internal task executor", // name
@ -70,18 +65,13 @@ struct Config {
bool useProfiling; bool useProfiling;
bool dumpCustomKernels; bool dumpCustomKernels;
bool exclusiveAsyncRequests; bool exclusiveAsyncRequests;
bool memory_pool_on;
bool enableDynamicBatch; bool enableDynamicBatch;
bool enableInt8; bool enableInt8;
bool nv12_two_inputs; bool nv12_two_inputs;
bool enable_fp16_for_quantized_models;
cldnn::priority_mode_types queuePriority; cldnn::priority_mode_types queuePriority;
cldnn::throttle_mode_types queueThrottle; cldnn::throttle_mode_types queueThrottle;
int max_dynamic_batch; int max_dynamic_batch;
CustomLayerMap customLayers; CustomLayerMap customLayers;
cldnn::tuning_config_options tuningConfig;
std::string graph_dumps_dir;
std::string sources_dumps_dir;
std::string kernels_cache_dir; std::string kernels_cache_dir;
ov::element::Type inference_precision; ov::element::Type inference_precision;
InferenceEngine::IStreamsExecutor::Config task_exec_config; InferenceEngine::IStreamsExecutor::Config task_exec_config;

106
src/plugins/intel_gpu/src/plugin/device_config.cpp
View File

@ -7,7 +7,6 @@
#include <ie_system_conf.h> #include <ie_system_conf.h>
#include <sys/stat.h> #include <sys/stat.h>
#include <cldnn/cldnn_config.hpp>
#include <gpu/gpu_config.hpp> #include <gpu/gpu_config.hpp>
#include <thread> #include <thread>
@ -92,16 +91,7 @@ void Config::UpdateFromMap(const std::map<std::string, std::string>& configMap,
} else { } else {
IE_THROW(NotFound) << "Unsupported property value by plugin: " << val; IE_THROW(NotFound) << "Unsupported property value by plugin: " << val;
} }
} else if (key.compare(PluginConfigParams::KEY_DUMP_KERNELS) == 0) { } else if (key.compare(GPUConfigParams::KEY_GPU_PLUGIN_PRIORITY) == 0) {
if (val.compare(PluginConfigParams::YES) == 0) {
dumpCustomKernels = true;
} else if (val.compare(PluginConfigParams::NO) == 0) {
dumpCustomKernels = false;
} else {
IE_THROW(NotFound) << "Unsupported property value by plugin: " << val;
}
} else if (key.compare(GPUConfigParams::KEY_GPU_PLUGIN_PRIORITY) == 0 ||
key.compare(CLDNNConfigParams::KEY_CLDNN_PLUGIN_PRIORITY) == 0) {
std::stringstream ss(val); std::stringstream ss(val);
uint32_t uVal(0); uint32_t uVal(0);
ss >> uVal; ss >> uVal;
@ -160,8 +150,7 @@ void Config::UpdateFromMap(const std::map<std::string, std::string>& configMap,
task_exec_config._streams = task_exec_config._streams =
std::min(task_exec_config._streams, static_cast<int>(std::thread::hardware_concurrency())); std::min(task_exec_config._streams, static_cast<int>(std::thread::hardware_concurrency()));
} }
} else if (key.compare(GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE) == 0 || } else if (key.compare(GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE) == 0) {
key.compare(CLDNNConfigParams::KEY_CLDNN_PLUGIN_THROTTLE) == 0) {
std::stringstream ss(val); std::stringstream ss(val);
uint32_t uVal(0); uint32_t uVal(0);
ss >> uVal; ss >> uVal;
@ -200,45 +189,11 @@ void Config::UpdateFromMap(const std::map<std::string, std::string>& configMap,
for (auto& file : configFiles) { for (auto& file : configFiles) {
CustomLayer::LoadFromFile(file, customLayers); CustomLayer::LoadFromFile(file, customLayers);
} }
} else if (key.compare(PluginConfigParams::KEY_TUNING_MODE) == 0) {
if (val.compare(PluginConfigParams::TUNING_DISABLED) == 0) {
tuningConfig.mode = cldnn::tuning_mode::tuning_disabled;
} else if (val.compare(PluginConfigParams::TUNING_CREATE) == 0) {
tuningConfig.mode = cldnn::tuning_mode::tuning_tune_and_cache;
} else if (val.compare(PluginConfigParams::TUNING_USE_EXISTING) == 0) {
tuningConfig.mode = cldnn::tuning_mode::tuning_use_cache;
} else if (val.compare(PluginConfigParams::TUNING_UPDATE) == 0) {
tuningConfig.mode = cldnn::tuning_mode::tuning_use_and_update;
} else if (val.compare(PluginConfigParams::TUNING_RETUNE) == 0) {
tuningConfig.mode = cldnn::tuning_mode::tuning_retune_and_cache;
} else {
IE_THROW(NotFound) << "Unsupported tuning mode value by plugin: " << val;
}
} else if (key.compare(PluginConfigParams::KEY_TUNING_FILE) == 0) {
tuningConfig.cache_file_path = val;
} else if (key.compare(CLDNNConfigParams::KEY_CLDNN_MEM_POOL) == 0) {
if (val.compare(PluginConfigParams::YES) == 0) {
memory_pool_on = true;
} else if (val.compare(PluginConfigParams::NO) == 0) {
memory_pool_on = false;
} else {
IE_THROW(NotFound) << "Unsupported memory pool flag value: " << val;
}
} else if (key.compare(CLDNNConfigParams::KEY_CLDNN_GRAPH_DUMPS_DIR) == 0) {
if (!val.empty()) {
graph_dumps_dir = val;
createDirectory(graph_dumps_dir);
}
} else if (key.compare(PluginConfigParams::KEY_CACHE_DIR) == 0 || key == ov::cache_dir) { } else if (key.compare(PluginConfigParams::KEY_CACHE_DIR) == 0 || key == ov::cache_dir) {
if (!val.empty()) { if (!val.empty()) {
kernels_cache_dir = val; kernels_cache_dir = val;
createDirectory(kernels_cache_dir); createDirectory(kernels_cache_dir);
} }
} else if (key.compare(CLDNNConfigParams::KEY_CLDNN_SOURCES_DUMPS_DIR) == 0) {
if (!val.empty()) {
sources_dumps_dir = val;
createDirectory(sources_dumps_dir);
}
} else if (key.compare(PluginConfigParams::KEY_EXCLUSIVE_ASYNC_REQUESTS) == 0) { } else if (key.compare(PluginConfigParams::KEY_EXCLUSIVE_ASYNC_REQUESTS) == 0) {
if (val.compare(PluginConfigParams::YES) == 0) { if (val.compare(PluginConfigParams::YES) == 0) {
exclusiveAsyncRequests = true; exclusiveAsyncRequests = true;
@ -282,8 +237,7 @@ void Config::UpdateFromMap(const std::map<std::string, std::string>& configMap,
} else { } else {
IE_THROW(NotFound) << "Unsupported property value by plugin: " << val; IE_THROW(NotFound) << "Unsupported property value by plugin: " << val;
} }
} else if (key.compare(GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS) == 0 || } else if (key.compare(GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS) == 0) {
key.compare(CLDNNConfigParams::KEY_CLDNN_NV12_TWO_INPUTS) == 0) {
if (val.compare(PluginConfigParams::YES) == 0) { if (val.compare(PluginConfigParams::YES) == 0) {
nv12_two_inputs = true; nv12_two_inputs = true;
} else if (val.compare(PluginConfigParams::NO) == 0) { } else if (val.compare(PluginConfigParams::NO) == 0) {
@ -291,14 +245,6 @@ void Config::UpdateFromMap(const std::map<std::string, std::string>& configMap,
} else { } else {
IE_THROW(NotFound) << "Unsupported NV12 flag value: " << val; IE_THROW(NotFound) << "Unsupported NV12 flag value: " << val;
} }
} else if (key.compare(CLDNNConfigParams::KEY_CLDNN_ENABLE_FP16_FOR_QUANTIZED_MODELS) == 0) {
if (val.compare(PluginConfigParams::YES) == 0) {
enable_fp16_for_quantized_models = true;
} else if (val.compare(PluginConfigParams::NO) == 0) {
enable_fp16_for_quantized_models = false;
} else {
IE_THROW(NotFound) << "Unsupported KEY_CLDNN_ENABLE_FP16_FOR_QUANTIZED_MODELS flag value: " << val;
}
} else if (key.compare(GPUConfigParams::KEY_GPU_MAX_NUM_THREADS) == 0 || key == ov::compilation_num_threads) { } else if (key.compare(GPUConfigParams::KEY_GPU_MAX_NUM_THREADS) == 0 || key == ov::compilation_num_threads) {
int max_threads = std::max(1, static_cast<int>(std::thread::hardware_concurrency())); int max_threads = std::max(1, static_cast<int>(std::thread::hardware_concurrency()));
try { try {
@ -353,39 +299,22 @@ void Config::adjustKeyMapValues() {
key_config_map[ov::enable_profiling.name()] = PluginConfigParams::NO; key_config_map[ov::enable_profiling.name()] = PluginConfigParams::NO;
} }
if (dumpCustomKernels)
key_config_map[PluginConfigParams::KEY_DUMP_KERNELS] = PluginConfigParams::YES;
else
key_config_map[PluginConfigParams::KEY_DUMP_KERNELS] = PluginConfigParams::NO;
if (exclusiveAsyncRequests) if (exclusiveAsyncRequests)
key_config_map[PluginConfigParams::KEY_EXCLUSIVE_ASYNC_REQUESTS] = PluginConfigParams::YES; key_config_map[PluginConfigParams::KEY_EXCLUSIVE_ASYNC_REQUESTS] = PluginConfigParams::YES;
else else
key_config_map[PluginConfigParams::KEY_EXCLUSIVE_ASYNC_REQUESTS] = PluginConfigParams::NO; key_config_map[PluginConfigParams::KEY_EXCLUSIVE_ASYNC_REQUESTS] = PluginConfigParams::NO;
if (memory_pool_on)
key_config_map[CLDNNConfigParams::KEY_CLDNN_MEM_POOL] = PluginConfigParams::YES;
else
key_config_map[CLDNNConfigParams::KEY_CLDNN_MEM_POOL] = PluginConfigParams::NO;
if (enableDynamicBatch) if (enableDynamicBatch)
key_config_map[PluginConfigParams::KEY_DYN_BATCH_ENABLED] = PluginConfigParams::YES; key_config_map[PluginConfigParams::KEY_DYN_BATCH_ENABLED] = PluginConfigParams::YES;
else else
key_config_map[PluginConfigParams::KEY_DYN_BATCH_ENABLED] = PluginConfigParams::NO; key_config_map[PluginConfigParams::KEY_DYN_BATCH_ENABLED] = PluginConfigParams::NO;
if (nv12_two_inputs) { if (nv12_two_inputs) {
key_config_map[CLDNNConfigParams::KEY_CLDNN_NV12_TWO_INPUTS] = PluginConfigParams::YES;
key_config_map[GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS] = PluginConfigParams::YES; key_config_map[GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS] = PluginConfigParams::YES;
} else { } else {
key_config_map[CLDNNConfigParams::KEY_CLDNN_NV12_TWO_INPUTS] = PluginConfigParams::NO;
key_config_map[GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS] = PluginConfigParams::NO; key_config_map[GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS] = PluginConfigParams::NO;
} }
if (enable_fp16_for_quantized_models)
key_config_map[CLDNNConfigParams::KEY_CLDNN_ENABLE_FP16_FOR_QUANTIZED_MODELS] = PluginConfigParams::YES;
else
key_config_map[CLDNNConfigParams::KEY_CLDNN_ENABLE_FP16_FOR_QUANTIZED_MODELS] = PluginConfigParams::NO;
key_config_map[ov::hint::inference_precision.name()] = inference_precision.get_type_name(); key_config_map[ov::hint::inference_precision.name()] = inference_precision.get_type_name();
{ {
@ -420,7 +349,6 @@ void Config::adjustKeyMapValues() {
default: default:
break; break;
} }
key_config_map[CLDNNConfigParams::KEY_CLDNN_PLUGIN_PRIORITY] = qp;
key_config_map[GPUConfigParams::KEY_GPU_PLUGIN_PRIORITY] = qp; key_config_map[GPUConfigParams::KEY_GPU_PLUGIN_PRIORITY] = qp;
} }
{ {
@ -448,7 +376,6 @@ void Config::adjustKeyMapValues() {
default: default:
break; break;
} }
key_config_map[CLDNNConfigParams::KEY_CLDNN_PLUGIN_THROTTLE] = qt;
key_config_map[GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE] = qt; key_config_map[GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE] = qt;
} }
{ {
@ -471,30 +398,7 @@ void Config::adjustKeyMapValues() {
hostTaskPriority = ov::util::to_string(ov::hint::Priority::MEDIUM); hostTaskPriority = ov::util::to_string(ov::hint::Priority::MEDIUM);
key_config_map[ov::intel_gpu::hint::host_task_priority.name()] = hostTaskPriority; key_config_map[ov::intel_gpu::hint::host_task_priority.name()] = hostTaskPriority;
} }
{
std::string tm = PluginConfigParams::TUNING_DISABLED;
switch (tuningConfig.mode) {
case cldnn::tuning_mode::tuning_tune_and_cache:
tm = PluginConfigParams::TUNING_CREATE;
break;
case cldnn::tuning_mode::tuning_use_cache:
tm = PluginConfigParams::TUNING_USE_EXISTING;
break;
case cldnn::tuning_mode::tuning_use_and_update:
tm = PluginConfigParams::TUNING_UPDATE;
break;
case cldnn::tuning_mode::tuning_retune_and_cache:
tm = PluginConfigParams::TUNING_RETUNE;
break;
default:
break;
}
key_config_map[PluginConfigParams::KEY_TUNING_MODE] = tm;
key_config_map[PluginConfigParams::KEY_TUNING_FILE] = tuningConfig.cache_file_path;
}
key_config_map[CLDNNConfigParams::KEY_CLDNN_GRAPH_DUMPS_DIR] = graph_dumps_dir;
key_config_map[CLDNNConfigParams::KEY_CLDNN_SOURCES_DUMPS_DIR] = sources_dumps_dir;
key_config_map[PluginConfigParams::KEY_CACHE_DIR] = kernels_cache_dir; key_config_map[PluginConfigParams::KEY_CACHE_DIR] = kernels_cache_dir;
key_config_map[ov::cache_dir.name()] = kernels_cache_dir; key_config_map[ov::cache_dir.name()] = kernels_cache_dir;
@ -560,12 +464,8 @@ bool Config::CanShareContextWith(const Config& other) const {
return this->throughput_streams == other.throughput_streams && return this->throughput_streams == other.throughput_streams &&
this->useProfiling == other.useProfiling && this->useProfiling == other.useProfiling &&
this->dumpCustomKernels == other.dumpCustomKernels && this->dumpCustomKernels == other.dumpCustomKernels &&
this->memory_pool_on == other.memory_pool_on &&
this->queueThrottle == other.queueThrottle && this->queueThrottle == other.queueThrottle &&
this->queuePriority == other.queuePriority && this->queuePriority == other.queuePriority &&
this->sources_dumps_dir == other.sources_dumps_dir &&
this->tuningConfig.mode == other.tuningConfig.mode &&
this->tuningConfig.cache_file_path == other.tuningConfig.cache_file_path &&
this->kernels_cache_dir == other.kernels_cache_dir && this->kernels_cache_dir == other.kernels_cache_dir &&
this->device_id == other.device_id && this->device_id == other.device_id &&
this->task_exec_config._streams == other.task_exec_config._streams && this->task_exec_config._streams == other.task_exec_config._streams &&

7
src/plugins/intel_gpu/src/plugin/graph.cpp
View File

@ -13,7 +13,6 @@
#include "intel_gpu/plugin/graph.hpp" #include "intel_gpu/plugin/graph.hpp"
#include "intel_gpu/plugin/simple_math.hpp" #include "intel_gpu/plugin/simple_math.hpp"
#include <cldnn/cldnn_config.hpp>
#include "intel_gpu/plugin/infer_request.hpp" #include "intel_gpu/plugin/infer_request.hpp"
#include "intel_gpu/plugin/itt.hpp" #include "intel_gpu/plugin/itt.hpp"
@ -136,14 +135,14 @@ std::shared_ptr<cldnn::network> Graph::BuildNetwork(std::shared_ptr<cldnn::progr
network = std::make_shared<cldnn::network>(program, m_stream_id); network = std::make_shared<cldnn::network>(program, m_stream_id);
} }
GPU_DEBUG_GET_INSTANCE(debug_config);
if (!m_config.graph_dumps_dir.empty() && m_stream_id == 0) { GPU_DEBUG_IF(!debug_config->dump_graphs.empty() && m_stream_id == 0) {
static int net_id = 0; static int net_id = 0;
auto steps_info = network->get_optimizer_passes_info(); auto steps_info = network->get_optimizer_passes_info();
size_t step_idx = 0; size_t step_idx = 0;
for (auto& step : steps_info) { for (auto& step : steps_info) {
CNNNetwork net(GetExecGraphInfoByPrimitivesInfo(step.second, true)); CNNNetwork net(GetExecGraphInfoByPrimitivesInfo(step.second, true));
net.serialize(m_config.graph_dumps_dir + std::to_string(net_id) + "_" + net.serialize(debug_config->dump_graphs + std::to_string(net_id) + "_" +
std::to_string(step_idx) + "_" + step.first + "_graph.xml"); std::to_string(step_idx) + "_" + step.first + "_graph.xml");
step_idx++; step_idx++;
} }

3
src/plugins/intel_gpu/src/plugin/plugin.cpp
View File

@ -353,7 +353,6 @@ void Plugin::SetConfig(const std::map<std::string, std::string> &config) {
streamsSet = config.find(PluginConfigParams::KEY_GPU_THROUGHPUT_STREAMS) != config.end() || streamsSet = config.find(PluginConfigParams::KEY_GPU_THROUGHPUT_STREAMS) != config.end() ||
config.find(ov::num_streams.name()) != config.end(); config.find(ov::num_streams.name()) != config.end();
throttlingSet = config.find(GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE) != config.end() || throttlingSet = config.find(GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE) != config.end() ||
config.find(CLDNNConfigParams::KEY_CLDNN_PLUGIN_THROTTLE) != config.end() ||
config.find(ov::intel_gpu::hint::queue_throttle.name()) != config.end(); config.find(ov::intel_gpu::hint::queue_throttle.name()) != config.end();
std::string device_id; std::string device_id;
cldnn::device_info device_info = device_map.begin()->second->get_info(); cldnn::device_info device_info = device_map.begin()->second->get_info();
@ -882,7 +881,7 @@ Parameter Plugin::GetMetric(const std::string& name, const std::map<std::string,
auto engine_params = Plugin::GetParams(config, device, nullptr); auto engine_params = Plugin::GetParams(config, device, nullptr);
auto engine = cldnn::engine::create(engine_params.engine_type, engine_params.runtime_type, device, auto engine = cldnn::engine::create(engine_params.engine_type, engine_params.runtime_type, device,
cldnn::engine_configuration(false, engine_params.queue_type, std::string(), cldnn::engine_configuration(false, engine_params.queue_type, std::string(),
config.queuePriority, config.queueThrottle, config.memory_pool_on, config.queuePriority, config.queueThrottle, true,
engine_params.use_unified_shared_memory, std::string(), config.throughput_streams), engine_params.use_unified_shared_memory, std::string(), config.throughput_streams),
engine_params.task_executor); engine_params.task_executor);

4
src/plugins/intel_gpu/src/plugin/program.cpp
View File

@ -314,9 +314,6 @@ std::shared_ptr<cldnn::program> Program::BuildProgram(const std::vector<std::sha
OV_ITT_SCOPED_TASK(itt::domains::intel_gpu_plugin, "Program::BuildProgram"); OV_ITT_SCOPED_TASK(itt::domains::intel_gpu_plugin, "Program::BuildProgram");
cldnn::build_options options; cldnn::build_options options;
if (!m_config.graph_dumps_dir.empty()) {
options.set_option(cldnn::build_option::graph_dumps_dir(m_config.graph_dumps_dir));
}
for (const auto& op : ops) { for (const auto& op : ops) {
if (op->is_dynamic()) { if (op->is_dynamic()) {
allow_new_shape_infer = true; allow_new_shape_infer = true;
@ -326,7 +323,6 @@ std::shared_ptr<cldnn::program> Program::BuildProgram(const std::vector<std::sha
options.set_option(cldnn::build_option::allow_new_shape_infer(allow_new_shape_infer)); options.set_option(cldnn::build_option::allow_new_shape_infer(allow_new_shape_infer));
options.set_option(cldnn::build_option::optimize_data(true)); options.set_option(cldnn::build_option::optimize_data(true));
options.set_option(cldnn::build_option::tuning_config(m_config.tuningConfig));
if (partialBuild) { if (partialBuild) {
options.set_option(cldnn::build_option::partial_build_program(true)); options.set_option(cldnn::build_option::partial_build_program(true));
} }

10
src/plugins/intel_gpu/src/plugin/remote_context.cpp
View File

@ -336,19 +336,15 @@ ExecutionContextImpl::ExecutionContextImpl(const std::shared_ptr<IInferencePlugi
iter = device_map.begin(); iter = device_map.begin();
auto& dev = iter->second; auto& dev = iter->second;
bool enable_profiling = (m_config.useProfiling ||
(m_config.tuningConfig.mode == cldnn::tuning_mode::tuning_tune_and_cache) ||
(m_config.tuningConfig.mode == cldnn::tuning_mode::tuning_retune_and_cache));
auto engine_params = Plugin::GetParams(m_config, dev, m_external_queue); auto engine_params = Plugin::GetParams(m_config, dev, m_external_queue);
m_engine = cldnn::engine::create(engine_params.engine_type, m_engine = cldnn::engine::create(engine_params.engine_type,
engine_params.runtime_type, dev, engine_params.runtime_type, dev,
cldnn::engine_configuration(enable_profiling, cldnn::engine_configuration(m_config.useProfiling,
engine_params.queue_type, engine_params.queue_type,
m_config.sources_dumps_dir, std::string(),
m_config.queuePriority, m_config.queuePriority,
m_config.queueThrottle, m_config.queueThrottle,
m_config.memory_pool_on, true,
engine_params.use_unified_shared_memory, engine_params.use_unified_shared_memory,
m_config.kernels_cache_dir, m_config.kernels_cache_dir,
m_config.throughput_streams), m_config.throughput_streams),

8
src/plugins/intel_gpu/src/plugin/transformations_pipeline.cpp
View File

@ -424,14 +424,6 @@ void TransformationsPipeline::apply(std::shared_ptr<ov::Model> func) {
OV_ITT_SCOPED_TASK(itt::domains::intel_gpu_plugin, "TransformationsPipeline::apply::lpt"); OV_ITT_SCOPED_TASK(itt::domains::intel_gpu_plugin, "TransformationsPipeline::apply::lpt");
using namespace ngraph::pass::low_precision; using namespace ngraph::pass::low_precision;
// Conversion to FP32 might be needed for quantized models that face any fp16 related issues (e.g. overflow) for non-quantized layers
// With this key users can work-around such issues
if (!config.enable_fp16_for_quantized_models) {
ngraph::pass::Manager manager;
manager.register_pass<ngraph::pass::ConvertPrecision>(precisions_array {{ ngraph::element::f16, ngraph::element::f32 }});
manager.run_passes(func);
}
auto supportedPrecisions = std::vector<PrecisionsRestriction>({ auto supportedPrecisions = std::vector<PrecisionsRestriction>({
PrecisionsRestriction::create<ngraph::opset1::Convolution>({ PrecisionsRestriction::create<ngraph::opset1::Convolution>({
{{0}, {ngraph::element::u8, ngraph::element::i8}}, {{0}, {ngraph::element::u8, ngraph::element::i8}},

25
src/tests/functional/plugin/gpu/shared_tests_instances/behavior/plugin/configuration_tests.cpp
View File

@ -3,7 +3,6 @@
// //
#include "behavior/plugin/configuration_tests.hpp" #include "behavior/plugin/configuration_tests.hpp"
#include "cldnn/cldnn_config.hpp"
#include "gpu/gpu_config.hpp" #include "gpu/gpu_config.hpp"
using namespace BehaviorTestsDefinitions; using namespace BehaviorTestsDefinitions;
@ -28,8 +27,6 @@ namespace {
{{InferenceEngine::PluginConfigParams::KEY_GPU_THROUGHPUT_STREAMS, "OFF"}}, {{InferenceEngine::PluginConfigParams::KEY_GPU_THROUGHPUT_STREAMS, "OFF"}},
{{InferenceEngine::PluginConfigParams::KEY_PERF_COUNT, "ON"}}, {{InferenceEngine::PluginConfigParams::KEY_PERF_COUNT, "ON"}},
{{InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}}, {{InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}},
{{InferenceEngine::PluginConfigParams::KEY_DUMP_KERNELS, "ON"}},
{{InferenceEngine::PluginConfigParams::KEY_TUNING_MODE, "TUNING_UNKNOWN_MODE"}},
{{InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}}}; {{InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}}};
}; };
@ -55,10 +52,6 @@ namespace {
{InferenceEngine::PluginConfigParams::KEY_PERF_COUNT, "ON"}}, {InferenceEngine::PluginConfigParams::KEY_PERF_COUNT, "ON"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_GPU}, {{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}}, {InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_DUMP_KERNELS, "ON"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_TUNING_MODE, "TUNING_UNKNOWN_MODE"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_GPU}, {{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}}, {InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_CPU}, {{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, CommonTestUtils::DEVICE_CPU},
@ -80,12 +73,6 @@ namespace {
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, {{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES,
CommonTestUtils::DEVICE_GPU + std::string(",") + CommonTestUtils::DEVICE_CPU}, CommonTestUtils::DEVICE_GPU + std::string(",") + CommonTestUtils::DEVICE_CPU},
{InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}}, {InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES,
CommonTestUtils::DEVICE_GPU + std::string(",") + CommonTestUtils::DEVICE_CPU},
{InferenceEngine::PluginConfigParams::KEY_DUMP_KERNELS, "ON"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES,
CommonTestUtils::DEVICE_GPU + std::string(",") + CommonTestUtils::DEVICE_CPU},
{InferenceEngine::PluginConfigParams::KEY_TUNING_MODE, "TUNING_UNKNOWN_MODE"}},
{{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES, {{InferenceEngine::MultiDeviceConfigParams::KEY_MULTI_DEVICE_PRIORITIES,
CommonTestUtils::DEVICE_GPU + std::string(",") + CommonTestUtils::DEVICE_CPU}, CommonTestUtils::DEVICE_GPU + std::string(",") + CommonTestUtils::DEVICE_CPU},
{InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}}, {InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}},
@ -113,10 +100,6 @@ namespace {
{InferenceEngine::PluginConfigParams::KEY_PERF_COUNT, "ON"}}, {InferenceEngine::PluginConfigParams::KEY_PERF_COUNT, "ON"}},
{{CONFIG_KEY(AUTO_BATCH_DEVICE_CONFIG), CommonTestUtils::DEVICE_GPU}, {{CONFIG_KEY(AUTO_BATCH_DEVICE_CONFIG), CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}}, {InferenceEngine::PluginConfigParams::KEY_CONFIG_FILE, "unknown_file"}},
{{CONFIG_KEY(AUTO_BATCH_DEVICE_CONFIG), CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_DUMP_KERNELS, "ON"}},
{{CONFIG_KEY(AUTO_BATCH_DEVICE_CONFIG), CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_TUNING_MODE, "TUNING_UNKNOWN_MODE"}},
{{CONFIG_KEY(AUTO_BATCH_DEVICE_CONFIG), CommonTestUtils::DEVICE_GPU}, {{CONFIG_KEY(AUTO_BATCH_DEVICE_CONFIG), CommonTestUtils::DEVICE_GPU},
{InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}}}; {InferenceEngine::PluginConfigParams::KEY_DEVICE_ID, "DEVICE_UNKNOWN"}}};
}; };
@ -155,14 +138,6 @@ namespace {
IE_SUPPRESS_DEPRECATED_START IE_SUPPRESS_DEPRECATED_START
auto conf_gpu = []() { auto conf_gpu = []() {
return std::vector<std::map<std::string, std::string>>{ return std::vector<std::map<std::string, std::string>>{
// Deprecated
{{InferenceEngine::CLDNNConfigParams::KEY_CLDNN_NV12_TWO_INPUTS, InferenceEngine::PluginConfigParams::YES}},
{{InferenceEngine::CLDNNConfigParams::KEY_CLDNN_NV12_TWO_INPUTS, InferenceEngine::PluginConfigParams::NO}},
{{InferenceEngine::CLDNNConfigParams::KEY_CLDNN_PLUGIN_THROTTLE, "0"}},
{{InferenceEngine::CLDNNConfigParams::KEY_CLDNN_PLUGIN_THROTTLE, "1"}},
{{InferenceEngine::CLDNNConfigParams::KEY_CLDNN_PLUGIN_PRIORITY, "0"}},
{{InferenceEngine::CLDNNConfigParams::KEY_CLDNN_PLUGIN_PRIORITY, "1"}},
{{InferenceEngine::GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS, InferenceEngine::PluginConfigParams::YES}}, {{InferenceEngine::GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS, InferenceEngine::PluginConfigParams::YES}},
{{InferenceEngine::GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS, InferenceEngine::PluginConfigParams::NO}}, {{InferenceEngine::GPUConfigParams::KEY_GPU_NV12_TWO_INPUTS, InferenceEngine::PluginConfigParams::NO}},
{{InferenceEngine::GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE, "0"}}, {{InferenceEngine::GPUConfigParams::KEY_GPU_PLUGIN_THROTTLE, "0"}},