I.MX8MP NPU inference source code

[mbrundler]

Hi,

We have built a disk image and an SDK for an I/MX8M Plus target, using yocto from a imx-5.10.35-2.0.0.xml manifest.

We have developped an inference sw based on the 'label_image' example, that executes inferences targeting the NPU (VIP8000Nano), using the tflite / NNAPI sw stack.

Information from the I.MX_Machine_learning _UG lists the sw layers involved in NPU inference and we would like to see the source code of those layers (tflite, NNAPI delegate, NNAPI, nn_runtime,  OVXLIB and OpenVX driver).

For the components built by yocto/bitbake from sources, the source files should be left sowewhere in the yocto work tree, but I actually found only the tflite sources.

For  the other layers, are the sources availaible and where (codeaurora, a github, ...) ?

Thanks in advance

[joanxie]

you can search from here, for example tensorflow:

"https://source.codeaurora.org/external/imx/tensorflow-imx/tree/?h=lf-5.10.72_2.2.0"

or onnxruntime

"https://source.codeaurora.org/external/imx/onnxruntime-imx/tree/?h=lf-5.15.32_2.0.0"

[mbrundler]

I agree the high level software sources are easy to find (I mentionned tflite is OK). It is the low level software sources I'm trying to find : NNRT, OVXLIB and the OpenVX driver.

Thanks

[joanxie]

I don't think we can share this GPU driver in this public community