TVM in Intel SGX Example
This application demonstrates the use of a simple TVM model in the Intel SGX trusted computing environment.
- The TVM premade Docker image
- A GNU/Linux environment
- TVM compiled with LLVM and SGX; and the
- The Linux SGX SDK link to pre-built libraries
- The rust-sgx-sdk
Check out the
/tvm/install/ubuntu_install_sgx.sh for the commands to get these dependencies.
Running the example
If using Docker, start by running
git clone --recursive https://github.com/dmlc/tvm.git docker run --rm -it -v $(pwd)/tvm:/mnt tvmai/ci-cpu /bin/bash
then, in the container
cd /mnt mkdir build && cd build cmake .. -DUSE_LLVM=ON -DUSE_SGX=/opt/sgxsdk -DRUST_SGX_SDK=/opt/rust-sgx-sdk make -j4 cd .. pip install -e python -e topi/python -e nnvm/python cd apps/sgx
Once TVM is build and installed, just
If everything goes well, you should see a lot of build messages and below them
First of all, it helps to think of an SGX enclave as a library that can be called to perform trusted computation. In this library, one can use other libraries like TVM.
Building this example performs the following steps:
- Creates a simple TVM module that computes
x + 1and save it as a system library.
- Builds a TVM runtime that links the module and allows running it using the TVM Python runtime.
- Packages the bundle into an SGX enclave
- Runs the enclave using the usual TVM Python
For more information on building, please refer to the
For more information on the TVM module, please refer to
For more in formation on SGX enclaves, please refer to the SGX Enclave Demo