Skip to content

Latest commit

 

History

History
61 lines (48 loc) · 2.84 KB

backward_compatibility.md

File metadata and controls

61 lines (48 loc) · 2.84 KB

Enabling backward compatibility

This section describes how to use an updated tools version (Say from SDK 10.1) with a lower SDK version (e.g. SDK 10.0). The tools version will clearly call out that if this is possible in the Version Compatibility Table.

NOTE: This is an experimental feature which has gone through limited validation

Setup on Target Device

  • Follow the steps below on the target device to patch the firmware, libraries, wheels, shared object files.

  • UPDATE_OSRT_COMPONENTS and UPDATE_FIRMWARE_AND_LIB env variables can be set before running the script to control what to update.

  • UPDATE_OSRT_COMPONENTS replaces the following components on the target device filesystem and is set to 1 by default i.e it will replace the components:

    • onnxruntime, tflite_runtime and dlr python wheels
    • /usr/include/itidl_rt.h, /usr/include/itvm_rt.h, /usr/include/tensorflow, /usr/include/onnxruntime
    • /usr/lib/tflite_2.21, /usr/lib/libonnxruntime.so, /usr/lib/libtensorflow-lite.a

  • UPDATE_FIRMWARE_AND_LIB replaces the following components on the target device filesystem and is set to 0 by default i.e it will not replace the components unless explicitly set:

    • C7X firmwares under /lib/firmware
    • /usr/lib/libtivision_apps.so, /usr/lib/libvx_tidl_rt.so, /usr/lib/libtidl_onnxrt_EP.so, /usr/lib/libtidl_tfl_delegate.so

NOTE: Make sure you have stable internet connection on the EVM to run this script.

Example usage for updating OSRT components (This assumes you are manually updating C7x firmwares with a patched SDK)

export SOC=*soc*                    // [am62a,am68a,am68pa,am69a,am67a]
export TISDK_IMAGE=*adas or edgeai* // [adas for evm boards, edgeai for sk boards]
export UPDATE_OSRT_COMPONENTS=1
./setup_target_device.sh

Example usage for updating OSRT components and C7x firmwares

export SOC=*soc*                    // [am62a,am68a,am68pa,am69a,am67a]
export TISDK_IMAGE=*adas or edgeai* // [adas for evm boards, edgeai for sk boards]
export UPDATE_OSRT_COMPONENTS=1
export UPDATE_FIRMWARE_AND_LIB=1
./setup_target_device.sh

Compile and Validate on Target Device

  • Once the setup is done, follow the steps below to build CPP application on EVM
mkdir build
cd build
cmake ../examples
make -j
cd ../
  • Compile the models on X86_PC using the latest tidl-tools and copy over the artifacts to target device file system at ./edgeai-tidl-tools/
  • Execute below to run inference on target device with both Python and CPP APIs
# scp -r <pc>/edgeai-tidl-tools/model-artifacts/  <dev board>/edgeai-tidl-tool/
# scp -r <pc>/edgeai-tidl-tools/models/  <dev board>/edgeai-tidl-tool/
python3 ./scripts/gen_test_report.py
  • The execution of above step will generate output images at ./edgeai-tidl-tools/output_images.