Building Desktop binaries with XNNPack Delegate

Jump to bottom Edit New page

Amish Garg edited this page Jun 14, 2021 · 4 revisions

Examples with Desktop Support

Instructions to build desktop binaries

Clone the TensorFlow repository

git clone https://github.com/tensorflow/tensorflow.git
cd tensorflow
git checkout r2.5

Checkout to any tensorflow version using git checkout r<version>

Build using bazel

These steps will generate tensorflow binary based on platform you are using to build.
To install bazel refer this guide.
Execute the build command in tensorflow directory root.

bazel build -c opt //tensorflow/lite/c:tensorflowlite_c --define tflite_with_xnnpack=true
The generated shared library will be available in your bazel-bin/tensorflow/lite/c directory.

The exact output library name is platform dependent:

Linux: libtensorflowlite_c.so
Mac: libtensorflowlite_c.dylib
Windows: tensorflowlite_c.dll

Rename it to:

Linux: libtensorflowlite_c-linux.so
Mac: libtensorflowlite_c-mac.dylib
Windows: libtensorflowlite_c-win.dll

Place the renamed binary in <your-project-root>/blobs/