Shaderjoy 1.22

@All

Spend a few hours to parse shadertoy database and can say that we have many not working because of:

1). no iChannel implemented
2). compilation failed
3). error named "need XXX SGPRs, which exceeds the maximum of 102"
4). a bit of those ones which give some broken visuals (not very much, just 5-10 of ~400 shaders tested)

If takes 100% of all shaders, we can say that 60% of all shadertoy shaders want iChannel (so additional textures and stuff), then compilation failed are about 20%, then "need more SGRPS" about 10%, and other 10% works :)

There new ones which found interesting to look and which works:


myTest: https://www.shadertoy.com/view/ldX3Dn
china: https://www.shadertoy.com/view/4syyDw
Forever Fan Wall: https://www.shadertoy.com/view/4syXRw
Spiral Stretches: https://www.shadertoy.com/view/XsXBzS
Star Nest: https://www.shadertoy.com/view/XlfGRj
Tileable Water Caustic: https://www.shadertoy.com/view/MdlXz8
Bubbles: https://www.shadertoy.com/view/4dl3zn
Analytic Motionblur 2D: https://www.shadertoy.com/view/MdSGDm
[SIG2014] - Total Noob: https://www.shadertoy.com/view/XdlSDs
[SH2014] Cellular: https://www.shadertoy.com/view/Xs2GDd
Ball Of Fire: https://www.shadertoy.com/view/lsf3RH
Waves: https://www.shadertoy.com/view/4dsGzH
Transparent Cube Field: https://www.shadertoy.com/view/ll2SRy
The road to Hell: https://www.shadertoy.com/view/Mds3Rn


Through few of them are really top: "The road to Hell", "Ball Of Fire", "Star Nest", "Tileable Water Caustic" and "Transparent Cube Field".

Made a video if anyone in interest how those good ones look like and how the speed on x5000 with latest Radeon RX drivers:

https://youtu.be/yUubbKpkPQI

@kas1e

Quote:
Sounds like ShaderToy/Joy shaders are a good way to test the limits of our shader compiler. Please submit bug tickets for the Warp3D Nova compilation failures.

I wish we had a compiler expert who could help out with the shader compiler. The Linux drivers use LLVM, which is out of reach for us. And, writing compilers is an entire specialized field of its own.

Hans

@Hans

Maybe this can shed some light to improve usage of SGPR registers:

https://www.phoronix.com/scan.php?page ... rek-Even-Lower-SGPR-Usage

i was thinking that it was due to some hardware limitations of my own graphics card,so i ignored it on my own tests.

@Capehill

Ok raising icon stack to 100 000 and changing "Start from" to WORKBENCH (was set SHELL) works without problem.

So maybe is the SHELL<->WORKBENCH tag in icon that gives the crash/GR.

Setting in icon SEHLL I get the crash/GR.


EDIT: is the mix of changing in icon stack 100 000 and Start from:WORKBENCH that makes Shaderjoy work on my system.

GREAT!!!!

@jabirulo

Good. How much you have stack for Shell? I will add a stack cookie for the next version.

@kas1e

Nice video, as usual!

iChannel support is absolutely on the TODO list.

@Capehill

100 000 bytes

Using RANGER tool I can see Shadertoy uses ~42KB, so maybe setting to 100 000 is enough AND setting "Start from:" WORKBENCH.

BTW I'm not using RAWBInfo commoditiy.

@Hans
Quote:

Yep, i will do some more sorting so to make proper bug reports. But this ones happens pretty offten in the shaders which works: http://www.amiga.org/developer/bugreports/view.php?id=481

@Capehill
Quote:

Btw, i see that sometime it not only textures, but also sound. Do you think it possible to add it as well (over AHI maybe, etc?), why i ask , its because some shaders are pure "demos" , with nice music/sounds and stuff :)

@Capehill
Thank you!
Great app for start coding shaders and fast testing.

Edit: @Capehill @Ami603 @Daytona675x

Like Ami603 said - A live text editor would be killer!
Smash your smart brains together for this one please!

@kas1e

Sound shaders are probably possible, by letting sound shader render to a texture which then becomes a sample buffer.

Then there are also mic input and network features...

@Capehill
Through probabaly mic and network almost none will use, but after testing for more than 700 shaders now, textures and sound surely pretty common.

Btw, i do find a lot more shaders which produce noise we talk about on mantis, and it looks like that root cause will be surely the same.

I also found interting shader which produce "half working + half noise" effects, which can help to strip things down, check that one:

https://www.shadertoy.com/view/ltXSDB

Also after another few hours of shadertoy database testing, i can say that i didn't find any shader which crash (only one you report in mantis), also not that many fail to compile. Mostly non-working ones are limitations of our 102 SGRPs and 256 of VGRPs.

@kas1e

Quote:

Impressive! On my desktop PC, browsing Shadertoy is quite painful due to slowness and occasional WebGL crashes.

Quote:

That's good news.

Quote:

Yes, those seem pretty common. Have to admit that many Shadertoy shaders seem fairly complex.

X1000, Radon HD 7950, 4GB, ogles2 2.11, Nova 1.70:

default (blue window): 1005

capehill/amigaa: 1006
capehill/bars: 1018
capehill/gear: 1000
capehill/helloh: 1021
capehill/helloh2: 1020
capehill/mandelbrot: 80
capehill/nuts: 1019
capehill/ripple: 1000

shadertoy/dueling_mandelbulbs: 615
shadertoy/foot_torus: 22
shadertoy/kissing_donuts: 24
shadertoy/sunset: 1014
shadertoy/yetanother_torus: 39

The range of user reported FPS with foot torus shader is larger than expected. Lower on X1000 and higher on RRX X5000 - 17 through to 54. What is contributing to this? RX drivers?

@ddni
Its just shader done like this. It is slow even on my modern PC , check it on your setup:

https://www.shadertoy.com/view/4dKfDV

I have from 45 to 52 fps with that one on 2 years old PC notebook with more or less modern graphics card. You can speed it up to changing in the shader itself at top "#define AA 4 " , to " #define AA 3".

But on my RadeonRX it surely faster than on my RadeonHD (through i use latest beta of RadeonRX driver)

@kas1e

Thanks, on PC the browser makes a huge difference My Ryzen2600 Vega56 PC gets 3-4 fps using Chrome and 140 fps with firefox.

edit. Chrome wasn't using HW acceleration. It gives 140fps with that turned on.

@Capehill

Is "iResolution" are void in shaderjoy? Or it works already? Just interesting to know, maybe we had some issues because of that?

Btw, how you think to detect when iChannel is texture, and when are sound? Are there some tags/describe bytes at the beginning of iChannel's code?

Just in case you want to have something basic visually + working sound, there are 2 tasty shaders:

https://www.shadertoy.com/view/XsX3zS
https://www.shadertoy.com/view/XdfGRH

Also sometimes I can see that sound coming in iChannel (like in those 2 examples above), but sometimes sound come as another piece of code (sound shader?:) ), like there:

https://www.shadertoy.com/view/4sSSWz

Also, what I find a bit strange (but that not related to the ShaderJoy, it's related to all amigaos4 animation in the windows), that in the window we never have really fluid animation movement even if it shows 60 fps, while in fullscreen we have an exactly good look. While for example on the Win box, I can see in window modes the same fluid movement. On our side it always a bit jerky, even if it 60 fps and only real way to have it all looks as needed, to switch to full screen. Maybe that issue with all those Waitofs and co?


@ddni
Yep, even if on risen with all the stuff it give only 140 fps, it means that shader itself quite very heavy. But as no CPU should be involved, but only GPU, it means that we still have some issues with speed. And if there no DMA will help, then dunno what else.

@kas1e

iResolution should be working. Only open issue is that I don't know what the .z coordinate is supposed to be. IIRC I wrote 32 (bits) there. I haven't seen any shader to use iResolution.z though.

iChannel is always a texture when it's looked from a fragment shader perspective. It's sampler2D (could be samplerCube, too).

At first, I try to make texture channels working using image files. Then the procedural textures (multiple shaders).

@Capehill

Quote:

But how code will known when iChannels "texture" are real texture, or a sound ? I mean, there should be some header-check or something on the header of data files ?

@kas1e

Not sure how to implement it yet. One of the possibilities might be a project file (text, xml, json) which describes the channel usage and sources.

Or some file suffix, or a certain directory structure.