xfcp VS corundum

This is exactly what I created https://github.com/alexforencich/xfcp for - Ethernet and serial to multiple internal components, with the ability to enumerate said components.

This is basically what I made XFCP for: https://github.com/alexforencich/xfcp

Simple interface framework for connecting Python to FPGA designs over a serial port or over Ethernet: https://github.com/alexforencich/xfcp .

I'll use my XFCP project to interface with the FPGA from Python via a USB serial chip. This provides access to the I2C bus on the board, for configuring the PLL chips and interfacing with the QSFP28 optical transceivers. Additionally, it connects to the dynamic reconfiguration ports (DRP) on the GTY transceivers, and I'll use that for performing BER measurements at 25 Gbps through a handful of QSFP28 cables and optical modules. It looks like I might also have to do some fine-tuning of some of the analog parameters on the transceivers (namely pre-emphasis).

I recently acquired a pair of rather large FPGA boards that have a bunch of high-speed IO. I figure it might be interesting to show the process for bringing them up in terms of the reference clock generation and distribution components on the board for the high-speed serializers, as well as performing some simple sanity checks (BER testing) on all of the interfaces to make sure everything is operational. I'll use my XFCP project to interface with the FPGA from Python for configuring the clocking components over I2C and for performing the BER measurements on the GTY transceivers via DRP.

The GPU uses this: https://github.com/alexforencich/verilog-pcie . And there is an open-source 100G NIC here, including open source 10G/25G MACs: https://github.com/corundum/corundum

Well, I did build this: https://github.com/corundum/corundum

I would look into 10GbE PCS/MAC packet processors implemented under AXI Stream interfaces for example. There are open source examples https://github.com/corundum/corundum and https://netfpga.org/ .

I built my own 100 Gbps capable NIC, along with driver: https://github.com/corundum/corundum. You're welcome to ask if you have any questions, though it is quite a different animal from a 100 Mbps NIC you might have on a microcontroller.

If you're looking for resources, here's one rather comprehensive example of a high-performance FPGA design with a fully custom DMA engine and driver, that runs on both Xilinx and Intel FPGAs: https://github.com/corundum/corundum

I have no idea, but you're welcome to build the design and look at it yourself: https://github.com/corundum/corundum/tree/master/fpga/mqnic/NetFPGA_SUME/fpga. The barrel shifters are in the DMA engine, both the read DMA and write DMA engines have wide barrel shifters.

Dive right into the slack channel and introduce yourself. There is also a new contributor guide. /u/alexforencich/ is on these reddits and he may be able to chime in and give you more concrete suggestions.