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

Deprecate GNA_LIB_N_THREADS and SINGLE_THREAD parameters (#9486)

Browse Source
This commit is contained in:
Nadezhda Ageeva 2022-01-13 14:30:41 +03:00 committed by GitHub
parent 2b996c32ad
commit c06d4e7fb7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
10 changed files with 44 additions and 88 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
docs/IE_DG/supported_plugins/GNA.md
View File

@ -313,9 +313,6 @@ The plugin supports the configuration parameters listed below. The parameter nam
+----------------------------------+-------------------------+---------------+-----------------------------------------------------------------+
| ``KEY_PERF_COUNT`` | ``YES``, ``NO`` | ``NO`` | Turns on performance counters reporting. |
+----------------------------------+-------------------------+---------------+-----------------------------------------------------------------+
| ``KEY_GNA_LIB_N_THREADS`` | 1-127 integer number | 1 | Sets the number of GNA accelerator library worker threads used |
| | | | for inference computation in software modes. |
+----------------------------------+-------------------------+---------------+-----------------------------------------------------------------+
The parameters are passed as ``std::map<std::string, std::string>`` on ``InferenceEngine::Core::LoadNetwork`` or ``InferenceEngine::SetConfig``.
@ -352,9 +349,6 @@ The plugin supports the configuration parameters listed below. The parameter nam
+----------------------------------+-------------------------+---------------+-----------------------------------------------------------------+
| ``PERF_COUNT`` | ``YES``, ``NO`` | ``NO`` | Turns on performance counters reporting. |
+----------------------------------+-------------------------+---------------+-----------------------------------------------------------------+
| ``GNA_LIB_N_THREADS`` | 1-127 integer number | 1 | Sets the number of GNA accelerator library worker threads used |
| | | | for inference computation in software modes. |
+----------------------------------+-------------------------+---------------+-----------------------------------------------------------------+
The parameters are passed as strings to `IECore.load_network <api/ie_python_api/_autosummary/openvino.inference_engine.IECore.html#openvino.inference_engine.IECore.load_network>`_.
@ -402,25 +396,6 @@ Performance counters provided for the time being:
* Number of total cycles spent on scoring in hardware including compute and memory stall cycles
* Number of stall cycles spent in hardware
## Multithreading Support in GNA Plugin
The GNA plugin supports the following configuration parameters for multithreading management:
@sphinxdirective
.. tab:: C++
``KEY_GNA_LIB_N_THREADS``
.. tab:: Python
``GNA_LIB_N_THREADS``
@endsphinxdirective
By default, the GNA plugin uses one worker thread for inference computations. This parameter allows you to create up to 127 threads for software modes.
> **NOTE**: Multithreading mode does not guarantee the same computation order as the order of issuing. Additionally, in this case, software modes do not implement any serializations.
## Network Batch Size
Intel® GNA plugin supports the processing of context-windowed speech frames in batches of 1-8 frames in one

2
src/inference/include/ie/gna/gna_config.hpp
View File

@ -112,6 +112,7 @@ DECLARE_GNA_CONFIG_KEY(PWL_MAX_ERROR_PERCENT);
* Note that multithreading mode does not guarantee the same computation order as order
* of issuing. Additionally, in this case, software modes do not implement any serializations.
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_GNA_CONFIG_KEY(LIB_N_THREADS);
} // namespace GNAConfigParams
@ -131,6 +132,7 @@ namespace PluginConfigParams {
* It is passed to Core::SetConfig(), this option should be used with values:
* PluginConfigParams::YES or PluginConfigParams::NO
*/
INFERENCE_ENGINE_DEPRECATED("The config key will be removed")
DECLARE_CONFIG_KEY(SINGLE_THREAD);
} // namespace PluginConfigParams

12
src/plugins/intel_gna/gna_device.cpp
View File

@ -441,7 +441,7 @@ void GNADeviceHelper::updateGnaDeviceVersion() {
checkGna2Status(status, "Gna2DeviceGetVersion");
}
void GNADeviceHelper::open(uint8_t n_threads) {
void GNADeviceHelper::open() {
std::unique_lock<std::mutex> lockGnaCalls{ acrossPluginsSync };
updateGnaDeviceVersion();
const auto gnaExecTarget = parseTarget(executionTarget);
@ -457,10 +457,6 @@ void GNADeviceHelper::open(uint8_t n_threads) {
const auto status = Gna2DeviceOpen(nGnaDeviceIndex);
checkGna2Status(status, "Gna2DeviceOpen");
}
// TODO: GNA2: uncomment when scratchpad repaired
// status = Gna2DeviceSetNumberOfThreads(nGnaDeviceIndex, n_threads);
// checkGna2Status(status);
deviceOpened = true;
}
@ -483,12 +479,6 @@ void GNADeviceHelper::close() {
deviceOpened = false;
}
void GNADeviceHelper::setOMPThreads(uint8_t const n_threads) {
std::unique_lock<std::mutex> lockGnaCalls{ acrossPluginsSync };
const auto status = Gna2DeviceSetNumberOfThreads(nGnaDeviceIndex, n_threads);
checkGna2Status(status, "Gna2DeviceSetNumberOfThreads");
}
void GNADeviceHelper::updateGnaPerfCounters() {
if (!isPerformanceMeasuring)
return;

12
src/plugins/intel_gna/gna_device.hpp
View File

@ -65,8 +65,6 @@ public:
explicit GNADeviceHelper(std::string executionTargetIn = "",
std::string compileTargetIn = "",
bool swExactModeIn = false,
uint8_t lib_async_n_threads = 1,
bool use_openmp = false,
bool isPerformanceMeasuring = false,
bool deviceEmbedded = false,
int deviceVersionParsed = 0) :
@ -77,7 +75,7 @@ public:
nGnaDeviceIndex{selectGnaDevice()},
useDeviceEmbeddedExport(deviceEmbedded),
exportGeneration(static_cast<Gna2DeviceVersion>(deviceVersionParsed)) {
open(lib_async_n_threads);
open();
initGnaPerfCounters();
// check GNA Library version
@ -88,11 +86,6 @@ public:
if (gnaLibVersion.rfind("3.0", 0) == 0) {
isGnaLibVersion3_0 = true;
}
if (use_openmp) {
uint8_t num_cores = std::thread::hardware_concurrency();
setOMPThreads((num_cores != 0) ? num_cores : 1);
}
}
GNADeviceHelper(const GNADeviceHelper&) = delete;
@ -149,7 +142,7 @@ public:
std::string getEffectiveGnaCompileTarget() const;
private:
void open(uint8_t const n_threads);
void open();
void close();
static std::string getGnaLibraryVersionPrivate();
@ -167,7 +160,6 @@ public:
void createVirtualDevice(Gna2DeviceVersion devVersion, std::string purpose = "");
void updateGnaDeviceVersion();
void setOMPThreads(uint8_t const n_threads);
void initGnaPerfCounters() {
std::unique_lock<std::mutex> lockGnaCalls{ acrossPluginsSync };

2
src/plugins/intel_gna/gna_plugin.cpp
View File

@ -358,8 +358,6 @@ void GNAPlugin::InitGNADevice() {
gnadevice = std::make_shared<GNADeviceHelper>(config.gnaExecTarget,
config.gnaCompileTarget,
config.swExactMode,
gnaFlags->gna_lib_async_threads_num,
gnaFlags->gna_openmp_multithreading,
gnaFlags->performance_counting,
!config.dumpXNNPath.empty(),
GetDeviceVersionFromString(config.dumpXNNGeneration));

4
src/plugins/intel_gna/gna_plugin_config.cpp
View File

@ -29,6 +29,7 @@ static const std::set<std::string> supportedTargets = {
""
};
OPENVINO_SUPPRESS_DEPRECATED_START
void Config::UpdateFromMap(const std::map<std::string, std::string>& config) {
for (auto&& item : config) {
auto key = item.first;
@ -210,6 +211,7 @@ void Config::UpdateFromMap(const std::map<std::string, std::string>& config) {
AdjustKeyMapValues();
}
OPENVINO_SUPPRESS_DEPRECATED_END
void Config::AdjustKeyMapValues() {
std::lock_guard<std::mutex> lockGuard{ mtx4keyConfigMap };
@ -251,9 +253,11 @@ void Config::AdjustKeyMapValues() {
keyConfigMap[GNA_CONFIG_KEY(PWL_MAX_ERROR_PERCENT)] = std::to_string(gnaFlags.pwlMaxErrorPercent);
keyConfigMap[CONFIG_KEY(PERF_COUNT)] =
gnaFlags.performance_counting ? PluginConfigParams::YES: PluginConfigParams::NO;
OPENVINO_SUPPRESS_DEPRECATED_START
keyConfigMap[GNA_CONFIG_KEY(LIB_N_THREADS)] = std::to_string(gnaFlags.gna_lib_async_threads_num);
keyConfigMap[CONFIG_KEY(SINGLE_THREAD)] =
gnaFlags.gna_openmp_multithreading ? PluginConfigParams::NO: PluginConfigParams::YES;
OPENVINO_SUPPRESS_DEPRECATED_END
keyConfigMap[CONFIG_KEY(LOG_LEVEL)] = gnaFlags.log_level;
}

4
src/tests/functional/plugin/gna/shared_tests_instances/behavior/ov_infer_request/multithreading.cpp
View File

@ -7,10 +7,14 @@
using namespace ov::test::behavior;
namespace {
OPENVINO_SUPPRESS_DEPRECATED_START
const std::vector<std::map<std::string, std::string>> configs = {
{{GNA_CONFIG_KEY(LIB_N_THREADS), "3"}}
};
OPENVINO_SUPPRESS_DEPRECATED_END
INSTANTIATE_TEST_SUITE_P(smoke_BehaviorTests, OVInferRequestMultithreadingTests,
::testing::Combine(
::testing::Values(CommonTestUtils::DEVICE_GNA),

6
src/tests/unit/gna/gna_plugin_config_test.cpp
View File

@ -11,6 +11,7 @@
using namespace InferenceEngine;
using namespace GNAPluginNS;
IE_SUPPRESS_DEPRECATED_START
const std::map<std::string, std::string> supportedConfigKeysWithDefaults = {
{GNA_CONFIG_KEY(SCALE_FACTOR), "1.000000"},
{GNA_CONFIG_KEY(SCALE_FACTOR) + std::string("_0"), "1.000000"},
@ -28,6 +29,7 @@ const std::map<std::string, std::string> supportedConfigKeysWithDefaults = {
{CONFIG_KEY(SINGLE_THREAD), CONFIG_VALUE(YES)},
{CONFIG_KEY(LOG_LEVEL), PluginConfigParams::LOG_NONE}
};
IE_SUPPRESS_DEPRECATED_END
class GNAPluginConfigTest : public ::testing::Test {
protected:
@ -159,6 +161,8 @@ TEST_F(GNAPluginConfigTest, GnaConfigPerfCountTest) {
config.gnaFlags.performance_counting);
}
IE_SUPPRESS_DEPRECATED_START
TEST_F(GNAPluginConfigTest, GnaConfigLibNThreadsTest) {
SetAndCompare(GNA_CONFIG_KEY(LIB_N_THREADS), "2");
EXPECT_EQ(config.gnaFlags.gna_lib_async_threads_num, 2);
@ -176,6 +180,8 @@ TEST_F(GNAPluginConfigTest, GnaConfigSingleThreadTest) {
true);
}
IE_SUPPRESS_DEPRECATED_END
TEST_F(GNAPluginConfigTest, GnaConfigGnaExecTargetTest) {
SetAndCompare(GNA_CONFIG_KEY(EXEC_TARGET), "GNA_TARGET_2_0");
EXPECT_EQ(config.gnaExecTarget, "GNA_TARGET_2_0");

9
src/tests_deprecated/unit/engines/gna/configuration_test.cpp
View File

@ -61,15 +61,6 @@ TEST_F(GNAConfigTest, canNOTMatchWith128AsyncThreads) {
.throws();
}
TEST_F(GNAConfigTest, canMatchWithSingleMultipleOMPThreads) {
assert_that()
.onInferModel(GNATestIRs::Fc2DOutputModel())
.inNotCompactMode()
.withGNAConfig(GNA_CONFIG_KEY(SCALE_FACTOR), 1.0f)
.enable_omp_multithreading()
.gna().propagate_forward().called_without().pwl_inserted_into_nnet();
}
TEST_F(GNAConfigTest, failToCreatePluginWithDifferentInputScaleFactors) {
assert_that().creating().gna_plugin()
.withGNAConfig(std::string(GNA_CONFIG_KEY(SCALE_FACTOR))+"_0", 1000)

6
src/tests_deprecated/unit/engines/gna/gna_matcher.hpp
View File

@ -183,12 +183,6 @@ class GNATestConfigurability : public GNATestBase{
_env.config[CONFIG_KEY(PERF_COUNT)] = InferenceEngine::PluginConfigParams::YES;
return *dynamic_cast<T*>(this);
}
T & enable_omp_multithreading() {
_env.is_setup_of_omp_theads_expected = true;
_env.config[CONFIG_KEY(SINGLE_THREAD)] = InferenceEngine::PluginConfigParams::NO;
return *dynamic_cast<T*>(this);
}
};
/**