Google Git
Sign in
apache / tvm-vta / 57db5a718c74a788c98120ebbe1230797be698c8
commit57db5a718c74a788c98120ebbe1230797be698c8[log] [tgz]
authorDaniel Steger <dstegs@gmail.com>Wed Dec 09 19:34:15 2020 -0800
committerGitHub <noreply@github.com>Thu Dec 10 11:34:15 2020 +0800
tree99f836abad4792c005fa7fe77709fed3134b2b7c
parent6096ca0a88191f0d2b31d2131c95c491a01e7f62 [diff]
Enable Supported Xilinx target ZCU104 with Hardware Preset (#20)

* targets: Added zcu104 vta target

This commit adds support for the Xilinx zcu104 development board.
Currently, TVM-VTA does not support a production board thus this
commit. Leveraging a supported board which is integrated into vivado
provides board presets properly configuring peripherals and IO for
development. This means that the project can be used for further
bsp development using the xsa.

* scripts: Enable applying board preset in vivado

Enable fpga_board and fpga_board_ver properties for
Xilinx Platforms. This enables the hardware project to
produce a usable xsa which contains the board presets.
The board preset contains board specific config such as
IO/PS/DDR settings.

The end goal of this commit is to allow the output products
of TVM-VTA to be used in bsp creation and.. produce more
meaningful output products.
4 files changed
tree: 99f836abad4792c005fa7fe77709fed3134b2b7c
  1. apps/
  2. config/
  3. hardware/
  4. include/
  5. src/
  6. tests/
  7. .asf.yaml
  8. .gitignore
  9. Jenkinsfile
  10. LICENSE
  11. NOTICE
  12. README.md
README.md

VTA Hardware Design Stack

Build Status

VTA (versatile tensor accelerator) is an open-source deep learning accelerator complemented with an end-to-end TVM-based compiler stack.

The key features of VTA include:

  • Generic, modular, open-source hardware
    • Streamlined workflow to deploy to FPGAs.
    • Simulator support to prototype compilation passes on regular workstations.
  • Driver and JIT runtime for both simulator and FPGA hardware back-end.
  • End-to-end TVM stack integration
    • Direct optimization and deployment of models from deep learning frameworks via TVM.
    • Customized and extensible TVM compiler back-end.
    • Flexible RPC support to ease deployment, and program FPGAs with the convenience of Python.