Thanks guys, glad you found the article interesting. Replying to a few points:
fxgogo wrote: ↑Mon Apr 18, 2022 1:11 pm
I know you said PAL pixels are almost square that we can consider them as such, but I just can’t do that. I think it is my job as a designer that sees such imperfections, he he.
Well, Amiga OS itself reports PAL as 44/44 pixel aspect ratio. Pixels are *supposed* to be square in PAL, at least when using a monitor. Because of the PAL TV standard, when you hooked up your Amiga with an RF adapter to a properly calibrated TV, the pixel aspect ratio was a tiny little bit off. Like 1.03 or something instead of 1.0. But who cares about TVs, really...
I think you might be referring to stretching of the 5/4 PAL image to fill the 3/4 screen, for which you had to stretch the image a tiny bit horizontally. I'd say that's incorrect, and even back in the day I drew a large 200x200 square or something in DPaint and adjusted my monitor stretch controls until the sides were the same length when measured with a ruler. That makes perfect sense for PAL games, as squares will be square, circles will be circles, and so on. But yeah, I guess quite a few users just did the full horizontal stretch thing... the evil what men do
You can easily adjust my settings for that: keep the vertical scale factor at x3 etc. and only adjust the horizontal scale factor the the desired "squash" (I published the formulas to use, just do the math yourself). Don't do it the other way around (keep horizontal stretch constant and adjust vertical stretch) because that could introduce scanline-aliasing.
Yeah, some of the low-res single-scanned example look decent or even quite good, but the double-scanned VGA stuff is completely wrong. There is *no way* you're gonna emulate the ~0.25 triad dot mask at 1080p, it simply won't happen, it's a physical impossibility. Even 4k isn't enough for that, 8k would be about right (and at fullscreen only, probably).
For state-of-the-art CRT shaders, check out this LibRetro thread. This is Guest's new work (sadly can't be backported to WinUAE as it doesn't have the advanced shader architecture that RetroArch has):
https://forums.libretro.com/t/new-crt-s ... 25444/2131
iljitsch wrote: ↑Tue Apr 19, 2022 5:19 am
Anyway, interesting post, but about the NTSC stretch: my A1200 tells me that this is 26:22 = 1.181818... so a hair less than a factor 1.2.
Yeah, that's what DPaint V reports, 44/52 (=22/26) pixel aspect ratio for NTSC low-res, and it gets it from the OS.
iljitsch wrote: ↑Tue Apr 19, 2022 5:19 am
We did discuss the stretch thing a while back on this forum, and it's obvious that this is not just an intended for PAL / intended for NTSC thing.
Sorry I can't parse that sentence, what's obvious and why?
As I see it, this is all purely theorethical, bit of a much ado about nothing situation:
- For PAL, the OS reported PAR is 1:1, and that's what the more discerning users did back in the day (adjusted their monitors for 100% square pixels)
- For NTSC, all the different reported, calculated, etc. values don't matter one bit. In practice people, adjusted their monitors so the 320x200 image fills the whole 4/3 screen, which gives 1:1.2 PAR. Exact same deal as with VGA.
Of course, you have my formulas, you can do whatever easily. But I didn't want to confuse my readers with all that unnecessary detail that ultimately doesn't give you "better" or "more authentic" results. Plus these are the exact settings I'm using; I wasted a day trying all these esoteric ratios, but my conclusion was it's not worth the effort, the differences are so minimal, and not even desirable.