Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

dma + axi memory #53

Open
juanma-rm opened this issue May 24, 2024 · 0 comments
Open

dma + axi memory #53

juanma-rm opened this issue May 24, 2024 · 0 comments

Comments

@juanma-rm
Copy link

Hi Alex! First of all, congrats and thanks for all your great work with verilog_* repositories and corundum.

I have been testing verilog-pcie in an Alveo U25 platform (using AU200 as baseline) successfully communicating host (a workstation running Linux) and card (the AU25 connected via PCIe to the host). In these tests, I configure the pcie dma from the host side via the driver running on the Linux host and move data in both directions (host -> card and card -> host). In particular, the data in the card side is stored in the psdpram (embedded in the fpga). All this using the "standard fpga" version (not the fpga_axi one).

However, I find myself now in the following situation: I need to move data host -> fpga, do some processing on the fpga side and then move the results back to the host, being the size of the data in the order several MBs (> 10 MB). I cannot work with a dma psdpram embedded in the fpga now because of fpga resource limitation. My plan is to use the PS DDR (AXI) as the dma memory instead. Then I wonder:

  • Is it the best option for my scenario to use fpga_axi instead? This way the dma could interface the PS DDR directly via its AXI ports (HP0 for instance). I know you mention in some issues that it is legacy and less performant because of AXI working principles (in fact I think it has been removed from the repo), but in my case I have no choice than using AXI interface. Is it still a bad idea to use fpga_axi?
  • I also see an axi ram connected to bar2, but I think I cannot use the dma in that case and would have to move data with iowrite/ioread word by word from the host (which does not seem feasible in performance).
  • Is there any alternative that you would suggest? I was thinking of building a bridge psdp <-> axi, but I guess I would end up having the same performance than directly using fpga_axi version, right?

Many thanks and best regards!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant