* added missed layout to string conversion
* added an unit test for layout_from_string
* added missing spaces around =
* replaced 95 with SCALAR
* moved layout_to_string into a new file
* changed to parametrized test
* change AUTO default hint to Latency
* Change the comment tput to latency according to wangyang opinion
* fix testcase for MULTI, the hint default value returned by MULTI is throughput
* Remove the redundant testcase and modify the name of the testcase that returns the default value of hint
* Code optimization according to bell opinion, add comments to testcase
* Correct the comments of testcase
* When user sets num_streams, AUTO/MULTI does not set the default hint to HW plugin
* Fix the problem that smoke_AUTO_MULTI_ReturnDefaultHintTest fails to run
* add num_streams and default hint mock testcase
* add auto default perf hint mock testcase
* temp resolution to support model path for CPU in auto
Signed-off-by: fishbell <bell.song@intel.com>
* disable batch when load through model path
Signed-off-by: fishbell <bell.song@intel.com>
* add mark for future release
Signed-off-by: fishbell <bell.song@intel.com>
* implement step1: donotparse batch config if user not set explictly
Signed-off-by: fishbell <bell.song@intel.com>
* correct typo in case
Signed-off-by: fishbell <bell.song@intel.com>
Signed-off-by: fishbell <bell.song@intel.com>
* Use openvino pass graph_rewrite
* Use openvino pass pattern matcher
* Remove ngraph opsets
* Remove ngraph.hpp
* Remove ngraph includes
* Remove ngraph includes
* Use transformations API
* Remove ngraph includes
* Remove ngraph rt_info
* Remove unused ngraph includes
* Replace ngraph:: scope with ov:: here and there
* Remove serialize proxy header
* Remove proxy include
* Bring back a file for vpu-plugin
* Fix after upstream merge
* Remove nested namespace conflict
Co-authored-by: Andrei Kochin <andrei.kochin@intel.com>
Co-authored-by: Ivan Tikhonov <ivan.tikhonov@intel.com>
Co-authored-by: Mateusz Tabaka <mateusz.tabaka@intel.com>
* [Python API][Tools] Top up upper bound version for dependencies: NumPy, TensorFlow, NetworkX
Signed-off-by: Kazantsev, Roman <roman.kazantsev@intel.com>
* Remove upper-bound for TensorFlow and NumPy
* Revert "Remove upper-bound for TensorFlow and NumPy"
This reverts commit 662085df2e.
* Remove upper-bound for NumPy for default installation
Signed-off-by: Kazantsev, Roman <roman.kazantsev@intel.com>
* [LPT] FQ Decomposition modified to create FQ without constants with several consumers
* [LPT] Added subgraph test
* CPUTestsBase: Added the ability to check only fusing results
* [CPU] Added subgraph test
* LPT review comments applied
* CPUTestsUtils: added special string 'anytype' for selectedType check ignore
* Added BenchmarkLayerTest class for simplifing of benchmarking
* Added example of using BenchmarkLayerTest for GenerateProposals operator
* Added report to file and also updated validate methods
* Switch from using json to xml for BenchmarkLayerTestReporter
* Remove smoke prefix from BenchmarkLayerTest usage example
Co-authored-by: Oleh Kurachenko <oleh.kurachenko@capgemini.com>
Co-authored-by: Irina Efode <irina.efode@intel.com>
ConvolutionBackpropData weights have following shape {input channels, output_channels, ...},
so unlike forward convolution, the output channels are on the second dimension.
* Update.
Signed-off-by: Wang, Yang <yang4.wang@intel.com>
* 1. Remove part of AUTO/MULTI test case since they has been coverred in another PR[13738].
2. Move the rest part of AUTO/MULTI test cases into CPU function test suit.
Signed-off-by: Wang, Yang <yang4.wang@intel.com>
* Update.
Signed-off-by: Wang, Yang <yang4.wang@intel.com>
Signed-off-by: Wang, Yang <yang4.wang@intel.com>
* change gpunum to 3
* hold threads for GPU for MULTI:GPU,CPU
* need to first check if there is a CPU in the device list
* use getNumberOfCPUCores to get CPU cores
* load GPU first
* assign the correct value to multiSContext->_devicePriorities
* load GPU first and load CPU last and set numa for CPU
* MULTI set CPU affinity to “NUMA” during load network
* Load the CPU last while maintaining the original device priority
* not using vector for CPU
* There is no user setting affinity in MULTI, and NUMA is set for the CPU
* pass key ENABLE_HYPER_THREAD to CPU plugin and merge xiaoxia PR
* set ENABLE_HYPER_THREAD to NO
* modify log
* Modify the code according to xiaoxia and wanglei comments
* Modify the code according to bell comments
* fix custom setting nstreams doesn't work due to the logic changed in ApplyPerformanceHints
* add same limitation for latency mode
* change bind core method to binding full physical core first, then logical core
* update custom setting streams and threads
* fix corner case when threads_per_stream is 1
* fix with comments
* add condition of big core stream is zero when binding core
Co-authored-by: Shen, Wanglei <wanglei.shen@intel.com>
Co-authored-by: Chen Peter <peter.chen@intel.com>
