Koko-aio shader discussions and updates

[kokoko3k]

Ok, just merged.

---

Commit:
Implemented Warpsharp like algorithm that superseedes scanlines inflation. Since it can be applied to X dimension too, it also helps FXAA to stay sharp.

---

So, Inflation is gone and the Warped glow appeared under Glow/Blur section:

    Warped glow (X,Y):
        Embolden bright pixels near dark ones using a warpsharp like algorithm.
        This is a cheap way to emulate phosphor glowing.
        The Y parameter will also allow scanlines to be higher.
        It will also help (if coupled with) FXAA to stay sharp.

    Warped Dynamics:
        Change the amount of warpsharp applied based on the contrast between 
        nearby pixels, thereby altering their "Warped" shape.

[lpslucasps]

I’m trying and failing to get the dither/dedithering of checkerboard patterns heavily used on some DOS EGA games right. Here’s an example with Loom.

Loom EGA, Original image, no shaders:

Loom EGA, tv-NTSC-1-classic_take

The shader certainly improves things a lot, but there are still some noticeable checkerboard patterns on the rocks and houses. I tried messing with some parameters, specially “Dedither enable?”, but even after cranking the sensitivity to the max it didn’t make a difference. NTSC color artifacts also didn’t do anything in this case. Other presets gave me similar results.

However, with guest_r CRT shaders I was able to get the effect somewhat right, but with compromises. If I set the “NTSC Resolution Scale” parameter to 0.50 the dithering patterns will blend better. Of course, everything will be a little bit less sharp, since I’m halving the resolution.

Here’s an example with sonkun’s “10-1080p-crt-guest-advanced-ntsc-slot-mask-u-warm-composite-standard” preset, which is based on guest_r’s:

The “tvout+ntsc-320px-composite” also handles dithering well, but the final result is too blurry for my tastes:

Here’s another example, this time with Quest for Glory II:

Original:

Koko:

Sonkun:

Tvout:

---

I know koko-aio shaders can handle dither on most systems perfectly - they’re my shaders of choice when playing Sega Mega Drive games, which relied heavily on the technique. But for some reason I can’t get them to look right with DOS EGA games. Is there any parameter I can change to get better results?

[kokoko3k]

@lpslucasps

Is there any parameter I can change to get better results?

Probably/Maybe

The dedithering is mutually exclusive with NTSC artifacts, enabling the latter disables the former (docs are wrong, stating the opposite, will fix them asap).

So you can try to disable artifacts and see if dedithering works better (don’t expect miracles, it is a fast/dumb algorithm)

Btw, thanks for the detailed explaination, it will be certainly helpful, will check in a few days to tweak Ntsc artifacts parameters for EGA resolution.

Meanwhile, could you please post naked source resolution (1:1) EGA screens? I’ve no MsDos emulator setup nor games to try, so I’d rely on yours.

PS: Were EGA games supposed to be seen via NTSC composite signal at all?

[lpslucasps]

@kokoko3k

PS: Were EGA games supposed to be seen via NTSC composite signal at all?

That’s a good point I didn’t consider at all. I know CGA graphics used a composite connection, but it seems EGA was actually RGB, like the later VGA standard. Which begs the question, how did dithering work on games with EGA graphics?

[kokoko3k]

Did it? I Honestly ignore it.

[Cyber]

@kokoko3k

PS: Were EGA games supposed to be seen via NTSC composite signal at all?

Not to my knowledge at all. All my life gaming on MS-DOS computers dithering was there but never blended. Sharp pixels were par for the course.

@lpslucasps

but it seems EGA was actually RGB, like the later VGA standard.

This is what I know.

@lpslucasps

Which begs the question, how did dithering work on games with EGA graphics?

There’s a recent thread which talks about this in detail.

Apparently folks did some wild stuff with CGA graphics back in the day. Getting more colours and stuff through composite.

If you still want to experiment you can try some Blargg Video Filter presets as well. Included with RetroArch are some “Blargg_NTSC_SNES_Custom_pseudo…” presets and I have a more comprehensive and updated filter preset pack called CyberLab Custom Blargg NTSC Video Filter Presets on my thread.

[Jamirus]

Those patterns essentially get lost in the mask/scanline interaction. You still see them on a VGA monitor /31 kHz. Alternatively, you could use TV-Out, but this is non-standard for EGA and probably not common on typical VGA cards used with DOS either. Common feature ofc with later Windows cards, but then, you’d probably mostly use it with S-Video.

[lpslucasps]

@kokoko3k

The dedithering is mutually exclusive with NTSC artifacts, enabling the latter disables the former (docs are wrong, stating the opposite, will fix them asap).

So you can try to disable artifacts and see if dedithering works better (don’t expect miracles, it is a fast/dumb algorithm)

I already tried that, didn’t make a difference. But I think I found the problem, which was of my own making: I was using the ScummVM core, which for some reason or another doesn’t play nice with the dithering effect. Even trying to use dedicated dithering shaders like mdapt didn’t do a thing. But now I switched to the DOSBOX Pure core and. suddenly, everything works flawlessly. The results are stunning. Here’s an example with Loom.

Raw, naked source resolution (PAR), with uncorrected aspect ratio:

Monitor-New_aperturegrille_gm

Monitor-FXAA_sharp-aperturegrille

Monitor-VGA-DoubleScan

In all examples, I enabled dedithering and changed the strength to 1.

@kokoko3k

Meanwhile, could you please post naked source resolution (1:1) EGA screens? I’ve no MsDos emulator setup nor games to try, so I’d rely on yours.

Sure thing!

@kokoko3k

@lpslucasps

Which begs the question, how did dithering work on games with EGA graphics?

Did it? I Honestly ignore it.

The checkerboard patterns suit some games, but specially for a lot of Sierra adventures I’d rather the colors blended.

@Cyber

Apparently folks did some wild stuff with CGA graphics back in the day. Getting more colours and stuff through composite.

They sure did, and I was able to emulate a lot of the crazy things they did with shaders. Here’s the raw CGA input of Agent USA vs what I was able to accomplish with the artifact-colors shader:

Koko’s shaders are actually quite capable of replicating some of these effects. Here’s what I was able to do with the B&W version of Borrowed Time (Apple II) after fiddling with the NTSC Color Artifacts parameters a bit:

[Hari-82]

@lpslucasps

But I think I found the problem, which was of my own causing: I was using the ScummVM core, which for some reason or another doesn’t play nice with the dithering effect

Not an expert, but I think the Scummvm core always output 480 and not the original resolution, so that might be the cause of “problems” with shaders (and scanlines).

off topic: in the scummvm core, Loom for amiga does not work for me (dos version does), is it working for you guys?

[kokoko3k]

That’s great, I asked you for 1:1 res screens to try to do what you achieved by your own.

Indeed, like @Hari-82 suggested, probably a 2x res ruined everything.

Btw, great work, care to share your findings in regard of cga to color parameter settings?

[lpslucasps]

@kokoko3k

Btw, great work, care to share your findings in regard of cga to color parameter settings?

Of course! Here’s how I’ve got the B&W Borrowed Time to look like that:

#reference "shaders_slang/bezel/koko-aio/Presets-4.1/tv-NTSC-1-classic_take.slangp"
NTSC_MIX = "3.000000"
NTSC_FILTER_SCF = "3.75"
DO_SPOT = "0.000000"

The Subcarrier Frequency Mul (NTSC_FILTER_SCF) parameter is the key. As you increase or decrease it the colors will change wildly, until you find a sweet spot where they get “stabilized”.

Here we have The Bard’s Tale (Apple II), in B&W:

Now, the same game with the “tv-NTSC-1-classic_take.slangp” preset and the parameters specified above:

If I change the “Subcarrier Frequency Mul” from 3.75 to 3.74, I get this mess instead:

There are other spots where the colors get stable, though, although the palette gets quite different. I’d guess that for some other games/systems you’d have to fiddle with this value until things get right.

Here’s what I get with a value of 5.25:

And here’s the rather cursed/cool 6.00:

For comparison’s sake, here’s the raw image of the game in color mode:

[lpslucasps]

Oh, btw, switching gears a bit. Do your handheld shaders come with color-correction and LCD ghosting effect built in? I’m trying to find if I should use the options on the emulator or if I could rely on the shader for that.

[kokoko3k]

No palette or lut based corrections. The Gameboy mono preset is using hue shifting, gba relies on core corrections.

Ghosting is possible, but used by gameboy mono preset only.

[lpslucasps]

I see. So, if I understand correctly, I don’t need to use the emulator’s LCD ghosting effect simulation with the gameboy mono preset?

[kokoko3k]

Yes, already active in all gbmono presets because it was so evident.

I left it disabled in other handhelds, because most lcd already emulates that …by design

[kokoko3k]

Exploring interactions between warped glow, glow light sharpeness and fxaa.

The antialiasing works by emboldening brightest areas. but seem to respect enough the original shapes; also the tipical fxaa “the dog licked the screen” effect si mostly gone :

The settings used are fxaa enabled, glow to blur bias 0.0, glow light gain: 4.0, glow light sharpness: 1.74, sharp x,y: -1.0, warped glow x,y:0.75 and warped glow dynamics: 0.44

[ROBMARK85]

Exceptional, congratulations! PS in these screens which preset did you start from?

[kokoko3k]

It’s ng/monitor-screen-hmask-screen-slotmask-taller

[kokoko3k]

Just tweaked VGA doublescan presets and added Green and amber variants.

-edit-

Also, new release 1.9.30 merge request submitted to Retroarch shader Repo, should be available soon:

• REMOVED: Removed support for Retroarch versions prior to to 1.16 due to the use of the rotation uniform.
• REMOVED: Removed dot matrix debug parameter
• REMOVED: Y-Tilt FOV option gone.
• NEW: Adjust content/bezel size by also taking width into account
• NEW: Add presets for PSP
• NEW: New parameter: “Interlaced scanlines brighness push” used to lower input gamma When emulating interlaced screens: Useless for fast response screens such as OLeds
• NEW: Scanline inflation turned into a generic Warped glow setting which can be applied to x axis too.
• NEW: Add a new subfolder “Presets_HiresGames_Fast” containing speedy presets
• NEW: VGA Dobulescan monochrome green and amber presets
• CHANGE: Corners now have a 4:3 aspect
• CHANGE: Y-Tilt does not require you to tweak Bezel tilting
• CHANGE: Lowered default flicker power across the presets (but the flicker one).
• CHANGE: General Presets tuning
• CHANGE: Docs update
• FIX: Keep Deconvergenge constant across input resolutions
• FIX: CVBS color bleed: keep the size constant across input resolutions intended for Hi resolution content (640x480+) that don’t use 2x internal processing
• FIX: Back/Foreground image: correctly scale offset so that it is in “sync” with content position
• FIX: Fully commented config-user.txt, sync’d with config.inc
• FIX: Usual work to enable specific workarounds for D3D11,12
• FIX: Lowered RESSWITCH_SAMPLING_POINT because too near to the screen border
• OPTIMIZED: Slightly optimized fn_border
• OPTIMIZED: Optimized halo vs scanlines blending
• OPTIMIZED: Skip unneded alpha channel assigningments to FragColor
• OPTIMIZED: Debanding: spare useless calls to scale_to_range()
• Overall performance vs previous release lowered from 111 to 108 fps measured by my standard benchmark on Haswell. this is mainly due to new warped glow feature.

[lpslucasps]

How much faster are the “HiresGames_Fast” presets? And what are the compromises?