Skip to content

Latest commit

 

History

History
104 lines (75 loc) · 6.66 KB

0157-2023-10-11.md

File metadata and controls

104 lines (75 loc) · 6.66 KB

11 Oct 2023

Previous journal: Next journal:
0156-2023-10-10.md 0158-2023-10-13.md

tt05-vga-spi-rom and cocotb stuff

Accomplishments

  • I spent a lot of the morning tidying up where test files are. Now there's src/test for everything except the Makefile (because TT05's test GHA template still expects to run make in there for tests). This included using PYTHONPATH in my Makefile so that it was possible to run make from the src/ dir even though src/test/test.py is the main MODULE file that cocotb is told to use.

  • In the afternoon I tried to rewrite the code to be better-structured, simpler, easier to read, etc. There were some issues with this and it took up too much time. Changes are in the cleanup branch. Latest commit is here, where it works and currently looks like this: Latest output from tt05-vga-spi-rom

    It currently only uses RGB111 colour, because I wanted to prove that would work OK, before having versions that work up to RGB333.

    NOTE: Blue channel now shows /CS, and I shifted the stored-mode range to be fully visible (for the time being).

Notes

  • Can't do the following, because the incoming state is contradictory on different phases (or so it appears to the synth) even though they are separated from each other by hpos/state: 
    // In stored mode, buffer-to-screen display cycle on POSEDGE of clk:
always @(posedge clk) begin
    if (stored_mode) begin
        // Shift buffer bits at the right time. This will make them appear on screen.
        if (hpos > PREAMBLE_LEN && hpos <= STREAM_LEN) begin
            data_buffer <= {data_buffer[BUFFER_DEPTH-2:0], 1'b0};
        end
    end
end
// In stored mode, shift MISO into buffer on NEGEDGE of clk:
always @(negedge clk) begin
    if (stored_mode) begin
        // Load MISO bits into buffer at the right time:
        if (state > PREAMBLE_LEN && state <= STREAM_LEN) begin
            data_buffer <= {data_buffer[BUFFER_DEPTH-2:0], spi_miso};
        end
    end
end
//SMELL: Should above pair of 'always' blocks just be combined into @(clk) ?
  • I think it would make more sense to just let all states and screen output be delayed by 1, then we don't have to worry about ugly "what comes next" logic. Just comment the code enough to show that on negedges, hpos has already advanced. CHECK: Will this be more or LESS compatible with setting registers shift state changes?
  • If things get a bit tricky with state tracking, especially during 799=>0 wrap-around, we could instead just use a separate counter register to count states. Then we can set it to 0 at 799 or whatever our 'start' register is programmed to.
  • Could SPI SCLK follow CLK directly (positive, instead of inverted)? Then, we just prep data to clock in on negedge of CLK? Hmm... Might just be the same problem anyway.
  • Would it help to register our outputs, then we don't have to worry about case/clock/state mismatches?

Goals

  • Fix X signals in tt05-vga-spi-rom: Probably just need reset for all other regs?
  • Review yesterday's (0156).
  • Proving that certain signals remain known/stable during test running, e.g. uio_oe: cocotb coroutine to run complex test sequencing, with an await on any edge in the main thread?

cocotb tests

  • Tidied things up by moving all test files into src/test/, including test.py, which cocotb (actually Python) is instructed to find via PYTHONPATH. The tests workin GHA.

RegNext

https://discord.com/channels/1009193568256135208/1016726411400200292/1161155587573305374

VCD to image

  • Eldritch conversation kicked off this idea.
  • I want to make it happen by default in make (for test GHA), with an ENV option to disable it.
  • GTKWave Automation (to save a PDF) using Tcl scripts.
    • NOTE: Given script doesn't work as-is on Windows because of missing ps2pdf util. We could install this (Ghostscript?) or just use PS output instead (which Adobe Illustrator can open).
    • I also changed paper size to: Legal (14" x 8.5")
    • See also: GTKWave User Guide - Appendix E: Tcl Command Syntax
    • There's also 'File' => 'Grab to File' which just saves a screenshot of the current window as a PNG.
    • See here for my experiments with this. I then moved these files around, so when you want to use it, do so like this: 
      gtkwave -S utils/vcd2png.tcl tb.vcd utils/vcd2png.gtkw utils/vcd2png.ini
      Note:
      • Tcl script directs GTKWave to save the view to both PS and PNG.
      • .gtkw file describes how to display the waveform.
      • .ini file specifies other visual rendering parameters.
  • Try also, VCD conversion to wavedrom:

VGA-DAC PCBs

My VGA-DAC PCBs (see journal 0146) have arrived from JLCPCB 14 days after submission, and they look great:

VGA-DAC PCB front and back, straight VGA-DAC PCB front and back, angled

The silkscreen quality (actually, all quality) seems excellent. The footprint of the VGA connector's 15 pins doesn't precisely match the two connectors I bought from DigiKey (see 0146) -- in KiCad they seem to be equally spaced on X and Y (but middle row just offset by half a pin), but the actual connectors seem to space the rows more than the individual pins. I will still able to get one of the connectors to fit by slightly bending its pins inwards towards the middle row. Interestingly, this error (or something similar) also shows up in the KiCad 3D render.

Notes