iiinova — where your digital self keeps evolving

Using MDFourier results I tweaked the values a little bit further to get something that matched more closely. Everything is within 3dB of the original hardware and the relative volumes of the PSG and FM data.

1163

Aurally speaking, the FM would have sounded "tinnier." On the board the FM audio goes through a series of low pass filters. I have the Final Blow schematics, but instead of analyzing it myself I decided to ask ChatGPT. I gave it this image and asked, "What does this do?"

160

The yellow line is PSG audio from the YM2610, this matches pretty closely because it has no filtering on the board. The slight differences are due to the power amp (and maybe the supergun). The green line is the FM audio, which does have filtering and you can see larger differences on the high end.

172

What does it sound like? Amazing. All the beep, bloops and csssschks you could hope for. I captured this audio through the supergun and ran the same test on the core on the DE-10 and captured it's output using a HDMI capture device, then compared them using MDFourier.

1132

To do this I needed to output an audio test pattern. I was able to port this from the Neo Geo 240p Test Suite pretty easily. Just needed to adapt addresses and ports. Luckily, since PicoROMs can be powered over USB, they can work as 28-pin ROMs as long as you have the space.

192

Having your own code running on hardware and the simulator allows narrow it down to the bare essentials. Really useful when you just want to see that first image on screen to keep yourself motivated.

2190

I damaged a Final Blow board almost two years ago and that's what motivated me to develop the PicoROM. I used it through out development to build small isolated test cases that I could run on real hardware and in the simulator and compare the results.

1140

Robert Peip uses save states extensively so I knew they would be useful, but I never appreciated just how useful. Simulation is slow, about 1fps, so being able to jump to a particular point in time is a huge time saving. Being able to save a state on the DE-10 and load it in the sim is massive!

1110

Maybe most important of all it allows me to work on my laptop, which is where I did 90% of the development. I built this sprite memory viewer while on a flight and I implemented most of the sprite system while on vacation.

1120

I took inspiration from @mrjimmystones.bsky.social and built my own verilator-based simulator. This allows for significantly faster compile times, 10 seconds vs 13 minutes. It makes it possible to trace the entire state of the system at any time.

1140

Martin Donlon @wickerwaka.com · 02/03/2025

A month ago I started working on an FPGA core for Taito's Final Blow, the first game on their F2 arcade system. Yesterday I had it running on the #MiSTerFPGA with all major systems in place. Lots of bugs left to fix, but the hardest parts are done.

715529

She’s 8,000 miles away, working her butt off and still finds time to show her love. She is the greatest gift.

1170

Martin Donlon @wickerwaka.com · 10/02/2025

I like that Windows Terminal includes this font preview. But no uppercase 'I' or 'O' and no 0 (zero)?!

110

Martin Donlon @wickerwaka.com · 20/01/2025

So I decided to make some adaptors that would allow me to connect these CPUs directly to a DE-10 Nano. With these I should be able to both drive and analyze the CPU behavior without any additional hardware.

1153

Martin Donlon @wickerwaka.com · 20/01/2025

Since the beginning I've been analyzing CPU behavior directly on arcade boards. It's convenient because it's already there and working, but hooking up probes is time consuming and fragile. Plus you are limited by what signals are used by the arcade board.

161

Martin Donlon @wickerwaka.com · 14/12/2024

The amount of CAD/CAM accessible to any idiot these days is fascinating. I was having trouble soldering this single pin neatly. So I used the amazing 3d model that my *free* PCB design software created to design a jig and 3D printed it. Went through 3 iterations on a Friday evening.

0130

Martin Donlon @wickerwaka.com · 13/12/2024

I've never used PC manipulation or sideset on the RP2040 PIO before. Two days ago while watching a music recital at my kids school I realized I could use it to do a really fast OR. It could be used for any kind of LUT with up to 5 inputs and outputs.

.program or_input
.origin 0
.side_set 1
mov pc, pins side 0
mov pc, pins side 1
mov pc, pins side 1
mov pc, pins side 1

150

Martin Donlon @wickerwaka.com · 11/12/2024

Main change with this latest version is that they can be powered directly from the board without a USB connection.

020

Martin Donlon @wickerwaka.com · 11/12/2024

PicoROMs! PicoROMs! PicoROMs!

4231

Martin Donlon @wickerwaka.com · 17/11/2024

I’m starting my career as a men’s lifestyle influencer by recommending this product.

160

And with that single byte fix we get a nice wobbly title screen, just like original hardware.

2221

In this case 0x37 should not map to 0x91 (xch aw,cw) it should map to 0x98 (cvtbw). cvtbw sign extends the byte from the lookup table to a 16-bit word allowing it to be safely used for the multiply. Without that the upper 8-bits were basically random.

180

It's loading a byte from a lookup table, multiplying it by some value and dividing it my a constant. There is always the chance of some weird hardware-specific edge case in something like a multiply/divide but that wasn't the problem here.

140

However, when I ran the same code on real hardware it showed the exact same problem. So the issue wasn't with how the scroll data was being interpreted, the issue was with the data itself. Using a watchpoint I tracked down the code that was generating the scroll data and I stared at it for a while.

150

So I captured the row scroll data from MAME and used it in my test rom. MAME and MiSTer both show the same broken scrolling. It should look like a nice sine wave, instead it is a disjointed mess.

140