passthrough_dmas

Passthrough DMAs

This reference design can be run on a Ryzen™ AI NPU.

In the design data is brought from external memory to a compute tile and back, without modification from the tile, by using an implicit copy via the compute tile's Direct Memory Access (DMA). The data is read from and written to external memory through a shim tile.

The implicit copy is performed using the ObjectFifo forward() function that specifies how input data arriving via of_in should be sent further via of_out by leveraging the fowarding tile's DMA.

There are two versions of this design:

• passthrough_dmas.py
• passthrough_dmas_placed.py: This version of the design supports VCK500 and is written in a lower-level version of IRON. Instead of forward(), this version explicitly uses an object_fifo_link operation which is described in more depth in Section-2b of the programming guide.

To compile and run the design for NPU:

make
make run

To compile and run the placed design for NPU:

env use_placed=1 make
make run