NVDLA Accelerator Inference Simulation

[Hassan313]

How can we use OpenPiton to do inference simulation with NVDLA accelerator? Is it possible to do it without FPGA?

[Jbalkind]

At present we don't have a great simulation-only environment for this. For FPGA, we can make use of AXI-based infrastructure to hook up NVDLA on a crossbar to memory, but to do this in simulation we don't presently have an AXI-based fake memory. It don't think it would be too hard to extend our fake memory to do something like this, though.

Another approach some students took was to translate from the AXI memory port of NVDLA to P-Mesh messages. However, this effort wasn't implemented to completion.