Does TVM 0.4.0 support RK3288?

Questions
1

Hi all, i’m trying build tvm on RK3288 chip. That i can build tvm done, but when i import tvm, it has errors.
Error log: OSError: /usr/local/lib/python2.7/dist-packages/tvm-0.4.0-py2.7-linux-armv7l.egg/tvm/libtvm.so: wrong ELF class: ELFCLASS64. I checked RK3288 board, it’s 32bit processor, maybe it don’t fit with tvm.
So, i want ask that Does TVM support RK3288?
Thank so much!

2

Please check the installation guide the mobile section. You DO NOT need to build tvm on the board, instead, you only have to build the tvm runtime and run things through rpc

1 Like
3

Link: https://docs.tvm.ai/tutorials/nnvm/deploy_model_on_rasp.html#deploy-the-pretrained-model-on-raspberry-pi

1 Like
4

I’m using Ubuntu 16.04 on rk3288 board, it’s armv7l (32 bits). I can build tvm on this board, but since i import tvm from example then there is error: OSError: /usr/local/lib/python2.7/dist-packages/tvm-0.4.0-py2.7-linux-armv7l.egg/tvm/libtvm.so: wrong ELF class: ELFCLASS64. Seem, tvm is 64 bits and rk3288 is 32 bits, they conflict.
So, I want clear: “Does TVM support armv7l (32 bits)?”
Thank you!

5

TVM supports 32bits runtime system.

Try out the following steps:

  • Remove your old installation of tvm(remove the folder), you do not need entire tvm on your board.
  • Type make runtime in the tvm (so you get the runtime library) on your board
  • Install tvm on a desktop server(not your board)
  • Follow https://docs.tvm.ai/tutorials/nnvm/deploy_model_on_rasp.html to do cross compile and rpc

See also https://docs.tvm.ai/deploy/index.html#deploy-and-integration

1 Like
6

I try with your guideline and i can import tvm, nnvm from examples for RK3288.
Thank you for supporting from you!