DeepX Yocto Meta Layer

Using the meta-deepx layer

To add the meta-deepx layer to your build, run the following command in the build directory:

bitbake-layers add-layer ../meta-deepx

To use the meta-deepx layer, add the following to your local.conf:

IMAGE_INSTALL:append = " dx-rt"

To supress the QA warning, add the following to your local.conf:

LICENSE_PATH += "${TOPDIR}/../meta-deepx/custom-licenses"

Note that the dx-npu recipe requires DMA_ENGINE option to be enabled in the kernel. Use this command to modify the kernel configuration:

bitbake -c menuconfig virtual/kernel

This layer provides the following configuration options:

• DX_DEVICE - choose the device (m1, m1a, l1, l3)
• DX_PCIE - choose the PCIe (deepx, xilinx)
• PACKAGECONFIG[onnxruntime] - use ONNX Runtime for unsupported operations (requires onnxruntime recipe, version 1.12.0)
• PACKAGECONFIG[python] - enable Python bindings
• PACKAGECONFIG[service] - enable DX-RT service (support for multiple processes)
• PACKAGECONFIG[shared_dxrt_lib] - use shared DX-RT library

The default configuration is equivalent to the following:

DX_DEVICE = "m1a"
DX_PCIE = "deepx"
PACKAGECONFIG:pn-dx-rt = " python service shared_dxrt_lib"

Modify these variables in your local.conf to change the default behavior.