this is our final 912x262 video signal bitmap. The graphics don't actually need to be centered, since we're not trying to sync to VSYNC - we're producing VSYNC, so it can happen whenever. In this case its easier to translate CRTC parameters keeping the active display area origin at 0,0.

Now we just need to save it as a raw binary.

"RAW binary" isn't one of Aseprite's export options, unfortunately, but we have BMP. The resulting BMP file is 240,022 bytes. That's an additional 1,078 bytes. That just so happens to be:

We have a picture. It isn't perfect - there's some flickery bits - I don't know if dot crawl is the right word. Not sure what's going on there.

Driving sync signals with a single image buffer was untenable - it's really difficult to have seamless, continuous DMA on the Pico. Every time you restart DMA there's a discontinuity, and if the thing you're DMA'ing represents every single clock cycle in a video field, inserting extra time messes up your picture.

So separate PIO programs generate hsync and vsync, and we restart our now image-only DMA in vsync, just like the OG Glyphblaster does, which we already know works.

I think the crawling may be that the color output is not perfectly aligned with the dot clock.

I'm starting to think a pico is not the appropriate thing to build a video card with.

all my fun ideas always end up with me concluding i should use an FPGA.

FPGAs are like the crabs of electronics projects. everything wants to turn into an FPGA if you give it enough time.

okay this isn't that confusing

for example, these are the snake pins. you can choose from left snakes or right snakes

you know, there's a very good reason NOT to use an FPGA

when you actually use the proper tools for a job, it stops becoming a hack.

you're no longer a hacker, you're some kind of engineer or something. ugh.

Okay, lets see if I can get a 14.31818MHz clock out of this thing.

We have a 50MHz clock source, on a pin called FPGA_CLK1_50.

How do we get 14.31818MHz out of that?

By attaching a PLL to it, hopefully.

pll_inst is a PLL instantiation. But notice it says nothing about like dividers or anything. We have to go into something confusingly called the IP (Incendiary Pickle) Catalog to actually configure the PLL.

This is far as a I get because doing this causes Quartus to hang.

Okay, if I did all this right, i should have a 14.31818MHz clock on GPIO pin 1.

Let's find out! To the workbench!

Hot damn!

looks just like the Pico clock - maybe it is my probe, lol.

So, the cool thing about FPGAs, is that this clock signal now just ... lives inside the vast and mysterious gate goo within the chip. We connected it to GPIO1, but we can connect it to just about anything, internally or externally, and we can trigger other logic on it.

It's like real legit design shit. It makes the Pico's PIO mode feel like a bit of a toy.

So remember that digital logic simulation I made of the CGA card in Digital?

Digital can export your entire simulation to Verilog.

In theory, I just need to wire up the OSC pin to this 14.31818Mhz PLL clock, and wire the simulation's output pins to some GPIOs, route them out through a 244, and I'll have a picture on screen.

surely it can't be that simple?

that would be a neat party trick but i wouldn't really learn anything and i'm sure the resulting Verilog would be spaghetti.

let's actually build this thing intentionally. Generating a clock was fun but i'm actually going to let the Pico continue to do that, and we'll treat a GPIO on the FPGA as the OSC input pin.

We'll divide it by 8, feed it to my Verilog MC6845 pre-configured for 80-column text mode, and wire up the HSYNC and VSYNC outputs to two more GPIOs.

We should get a 15.7kHz HSYNC and a 59.9kHz VSYNC out of that.

This is incredibly cool.

This is a spot-on, 15.7kHz horizontal sync pulse generated the Cyclone V FPGA running my (very simplified) Motorola 6845 Verilog implementation!

The DE-10 Nano has four user switches. The CGA outputs 4 bit color. Very convenient. Since I don't have any video memory yet, I'm just going to assign each switch to a color component, and AND it with display enable from the 6845. This should give us a well defined display rectangle on screen.

That looks something like this

Well, it took me a bit to figure out that the 47pF capacitors right at the D-SUB connector are not optional, specifically on HSYNC. This explains why the RGB2HDMI would sync to things the real monitor wouldn't.

But hey look, FPGA CGA video!

I didn't properly reset the green switch which is why we got cyan instead of blue. lol. But they do all work.

The FPGA is being fed the 14.31818 clock from the Pico to simulate the ISA bus' OSC pin as an input to some future card. I'm going simulate the ISA bus with Picos until I have some idea how I'm going to connect all this business to an actual PC's ISA bus.

It's a bit silly to talk about building an FPGA-based CGA card without mentioning the Graphics Gremlin, by @tubetime

https://github.com/schlae/graphics-gremlin

He's already made such a thing.

I'm interested in writing my own Verilog implementation, and of course I want to stick a Pico 2 W on it so we can stream video to it, but it seems sensible to fork the Gremlin instead of re-inventing the entire wheel.

@tubetime

There are some downsides, apparently this is a somewhat expensive card to build, just looking through forum discussions. But it's been five years, maybe there are better/cheaper FPGA options available today?

The Gremlin uses a Lattice iCE40 HX FPGA that costs about $20. That's not outrageous.

It has 7680 LUTs.

The Cyclone V I'm using has 110,000, but then again this FPGA is massive overkill. It can emulate a Nintendo 64, after all.

A lot has changed in the last 5 years since the Gremlin was specc'd out.

This is a Tang Nano 9k. It's about $22 on AliExpress.

It has 8640 LUTs, onboard SPI flash, and an HDMI port (!)

It could be socketed so you could take it out and use it for something else if you got bored with your ISA GlyphBlaster.

There's even a Tang Nano 20K now that has 8MB (yes MB) of SRAM on-board, and an SD-card reader. It's $45.

That's like your whole ass GlyphBlaster right there, just add bus buffers and a de-9 port.

Turns out the 244 buffer I pulled out of my cheapo amazon special Box o' Components was just bad, or maybe I inadvertently killed it.

In any case, I replaced it with one from the Chameleon and most of my signal issues cleared right up.

Translating my digital simulation into Verilog proceeds slowly. It's nice that I can copy and paste parts of the sim and export them to Verilog in isolation, then it's a a matter of simplifying things and making connections back to the main file.

I've got the font ROM in place and the chargen serializer. To test it, I hardcoded the font ROM's upper 8 address, then I AND the serializer output directly with the color switches.

What does that all mean? We have smiley faces!

Now I get to translate this hot mess into Verilog.

Yay.

I found a pretty decent way of checking my work against the digital simulation.

We can just run the card in a test harness in iVerilog, and export a CSV of signal states, so we can do a direct comparison against a CSV dumped from Digital.

This is a whole lot easier than using the oscilloscope lol.