100Base-FX

Comments

• rogloh Posts: 5,170

  2021-03-09 22:49 edited 2021-03-09 22:59

  Are you not planning to use a PHY for this? Eg, are you trying to get direct differential serial output to modulate a SFP at 125Mbps?

  If so, you could get the 4 to 5 bit translation done in the pin streamer using a LUT mode. However adding NRZI will be trickier because it requires a different sequence depending on the state of the line on the prior bit. You might be able to select which 1 bit "palette" is to be used based on the prior ending bit which can create the two outcomes from an initial state. This means some more COG intervention per nibble and you lose the efficient packed modes where nibbles are read per long and streamed directly from hub. You'll probably need to read data and extract nibbles and send them in immediate mode to the streamer. But given you only need to transfer 25M nibbles/s then you have spare cycles for doing this with the FIFO. Eg. 5 instructions per nibble (250MHz P2) just gets there in a REP loop, something like:

  REP     #9, bytecount
  RFBYTE  data
  GETNIB  nibble, data #0
  BITC    palette, #16 ' uses prior C flag
  RDLUT   serialdata, nibble wc ' set C flag for next time (C=last bit high or low)
  XCONT   palette, serialdata ' stream 5 bits
  GETNIB  nibble, data #1
  BITC    palette, #16
  RDLUT   serialdata, nibble wc ' set C flag for next time
  XCONT   palette, serialdata
  

  You'd also need to prime the streamer because of the asymmetry in the XCONT timing (11,9,11,9 clock spacing).

  Receive will be more challenging at 125Mbps to decode.
• Ramon Posts: 484

  2021-03-10 16:22

  Thank you for your reply, I will definitely need help from someone with knowledge.

  At first, I was looking how to send from COG RAM (to avoid HUB synchronization mistakes), and drive directly without PHY or SFP.
  (Not even thinking about receiving yet.)

  I was planning to drive directly an electro-optical converter (HP 83402C) But I can consider other options if they are easier to achieve. Will need time to digest the code above. No idea how the instructions work, is there any debugger out there to help me check the registers execution step by step?

  I have equipment to test any rate (10M,100M/1000M and 10G). But have zero P2 programming skills. No idea on how to start coding.
  
• rogloh Posts: 5,170

  2021-03-11 01:52

  The code above is an idea only not final code and needs further work etc to figure out the streamer commands and setup lookup table etc. There is a debugger that @ozpropdev developed that may help you at some point. You will certainly need to learn quite a lot of PASM2 code for this project. It is challenging, particularly RX. Probably a little easier to start with using MII I think. Then it's only 25M nibbles per second to process then and easier for the P2 to keep up rather than push it to the limit right away to decode received data serially bit by bit (but perhaps a serial smart pin can clock up to 125Mbps and you could try to capture raw unframed and encoded data for later decoding/processing, don't know).

  By the way, I found that you can probably send out entire encoded bytes at a time via the streamer so the code above simplifies to something like this (uses up 256 LUT entries for a table). This saves clocks and you only need 11 P2 clocks per byte sent now to keep up instead of the 20 above.

       REP     #5, bytecount
       RFBYTE  data
       RDLUT   data, data ' map 256 bytes (2 nibbles) into 10 bit sequence (2x5 bits)
  if_c NOT     data ' invert data output if last bit sent was a 1
       TESTB   data, #9 wc ' test top bit for next time
       XCONT   streamer, data  ' streamer is setup to transmit 10 bits of immediate output from data