mirror of
https://github.com/clearlinux/graphene.git
synced 2026-09-05 13:21:37 +00:00
Michał Kowalczyk
a12f96f49e
Non-PIE binaries support requires ELRANGE to start at low addresses, which on older SGX drivers required root permissions or reconfiguring the host, so this mapping strategy was made optional by us. This issue was fixed long time ago, so we can drop this option and always start enclaves at 0.
This example demonstrates how to run TensorFlow Lite v1.9. In particular, the
example runs label_image program on Graphene. It reads an input image
image.bmp from the current directory and uses TensorFlow Lite and the
Inception v3 model to label the image.
To install build dependencies on Ubuntu there is a convenience target invoked
with make install-dependencies-ubuntu. This also serves as a starting point to
figure out which packages to install on newer releases of Ubuntu.
To build TensorFlow Lite and Graphene artifacts:
- without SGX do
make - with SGX do
make SGX=1
To run the image labeling example:
- without Graphene do
make run-native - with Graphene do
make run-graphene - with Graphene-SGX do
make SGX=1 run-graphene