A year ago on ICCD 2020 conf there was a presentation from Xilinx about XTA - Open Source eXtensible, Scalable and Adaptable Tensor Architecture for AI Acceleration, derived from TVM-VTA.
Does anybody know, are there any news on this solution? Presentation was really promising in terms of VTA development.
@ilyag-grovety , thanks for following up on the XTA, we have a tracking issue used for XTA related upstream, please refer this link https://github.com/apache/tvm/issues/8596 for the XTA upstream progress.
@hjiang , Is there any tutorial document available how I can try/benchmark/program current XTA implementation, just like what VTA has provided so far ?
Most of the items in the link for XTA upstream progress tracking are about pipeline executor and subgraph splitting, and the progress is convincing.
Are there any further information or updates about XTA JIT runtime, hardware abstraction and emulator part, I am eager to give an hand-on trial of XTA if it is available.