For those who are trying vta and having troubles now

here is a guide for you to run the current version of vta(i’ve successed the GEMM test and rpc test this way on pynq)

  1. do a brand new recursive git clone of tvm on board
  2. for pynq follow https://tvm.apache.org/docs/vta/install.html to the cmake … step
  3. perform a full build of tvm by typing make -j2 instead of make runtime vta -j2 because ir.RegisterOpLowerIntrinsic seems not built for runtime.this step takes roughly about 4 hrs
  4. copy the lib*.so(3 lib files) in /build to your host computer because we need these file after
  5. go back to the /build on pynq and delete the lib*.so
  6. make runtime vta -j2 on pynq
  7. paste the libtvm.so you’ve just backed up back to the /build on pynq.this lib should be enough, but check for the file size. you can back up these library files to prevent building all of these again
  8. edit the vta/python/vta/exec.py and python/tvm/rpc/init.py on pynq Test_benchmark_topi_conv2d.py failure caused by AttributeError: Module has no function ‘tvm.contrib.vta.init’
  9. start the server and perform the test on host PC if you have the cloudpickle issue see Problem - RPC server on ZCU104