Shaderjoy 1.25

@Daniel
So.. If all is ok from code side, then how it possible that such a "simple" effects , give so small fps numbers ? I mean, why for example mandelbrot shader give so small fps. While , for example, the same kind of Maldenbrot fractals (fractalnova.lha on os4depot) give pretty decent results.

The more i see stuff which use shaders, the more it looks strange for me : everything which use shaders, usually slower ,than the same stuff done on pure CPU :) Why ?:) That not related to amigaos4 only, i mean the same on win32 for example.

Or it just related to the code of shaders it self , like if they bad, then they slow, but if they good, then they fast ?

@kas1e

Fractal Nova seems to use 100 iterations per fragment while Shaderjoy version uses 1000. Try changing that and then it will be faster.

I get a GR/crash when launching Shadertoy on my SAM460ex+RX550 (ogles2.library 2.11):
...
[_impl_RemTask] Removing 0x57d091f0 (self) = windowfade.task
[_impl_InitResident] Initializing native autoinit ogles2.library V2, priority 0
[_impl_InitResident] ogles2.library V2 initialized
[_impl_InitResident] Initializing native autoinit W3DN_SI.library V1, priority 0
[vaa_DataFaultHandler] Task hit stack guard page
[HAL_Alert] ALERT! 0x80000003, IP = 0x7ea480f4
kernel 54.21 (25.4.2020) Sam460 debug
Machine model: 7 (Sam460ex)
Dump of context at 0xff992ba0
Trap type: DSI exception
Machine State (raw): 0x2f030
Machine State (verbose): [ExtInt on] [User] [FPU on] [IAT on] [DAT on]
Temporary stack trace:
#0: 0x7ea480f4
#1: 0x7eac45ac
#2: 0x7eac4c7c
#3: 0x7eac5790
#4: 0x7eac6658
#5: 0x7ea47670
#6: 0x7ea36fc8
#7: 0x7ebb280c
#8: 0x7ebb1eec
#9: 0x7ebad6dc
#10: in module Kickstart/newlib.library.kmod+0x260c (0x1b5428c)
#11: in module Kickstart/newlib.library.kmod+0x3338 (0x1b54fb8)
#12: in module Kickstart/newlib.library.kmod+0x385c (0x1b554dc)
#13: 0x7ebad288
#14: in module Kickstart/dos.library.kmod+0x2a41c (0x1982a1c)
#15: in module Kickstart/kernel.debug+0x7fb70 (0x187fb70)
#16: in module Kickstart/kernel.debug+0x7fbe8 (0x187fbe8)
#17: 0x0

Crashed process: Shaderjoy (0x58e8d630)
0: ffff7820 541cce60 00000000 53e66018 5be25024 00000000 00000013 00000000
8: 00001be0 00000000 54190820 01b5b1c8 541cce60 5886faa8 00000001 00000008
16: 00000136 541cd7ec 541cd254 5cecc02c 5ce90e98 541cd4a0 541cd24c 541cd22c
24: 541cd214 541cd1ec 00000001 541cceb4 541cce9c 53e66018 541cd168 53e66018
CR: 22444482 XER: 20000004 CTR: 7eaf8d64 LR: 7ea8bc28
ESR: 00800000
DEAR: 541c4680
mcsrr0: 0x0
csrr0: 0x0

Disassembly of crash site:
7ea480e4: 4bfffe38 b 0x7EA47F1C
7ea480e8: 3c00ffff lis r0,-1
7ea480ec: 7c2c0b78 mr r12,r1
7ea480f0: 60007820 ori r0,r0,30752
>7ea480f4: 7c21016e stwux r1,r1,r0
7ea480f8: 7c0802a6 mflr r0
7ea480fc: 38a10008 addi r5,r1,8
7ea48100: bdccffb8 stmw r14,-72(r12)
7ea48104: 3fe50001 addis r31,r5,1
7ea48108: 3dc10001 addis r14,r1,1
Fault caused by store operation

Kernel command line: debuglevel=4 SERIAL MUNGE

Registers pointing to code:
r10: module LIBS:ogles2.library at 0x54190820 (section 1 @ 0x2A7FC)
r11: native kernel module Kickstart/newlib.library.kmod+0x00009548
r14: module LIBS:ogles2.library at 0x00000001 (section 0 @ 0xFFFFFFDC)
r26: module LIBS:ogles2.library at 0x00000001 (section 0 @ 0xFFFFFFDC)
ip : module LIBS:ogles2.library at 0x7EA480F4 (section 0 @ 0x2C0D0)
lr : module LIBS:ogles2.library at 0x7EA8BC28 (section 0 @ 0x6FC04)
ctr: module LIBS:ogles2.library at 0x7EAF8D64 (section 0 @ 0xDCD40)

Stack trace:
(0x541cce60) module LIBS:ogles2.library at 0x7EA480F4 (section 0 @ 0x2C0D0)
(0x541cce70) module LIBS:ogles2.library at 0x7EA8BC28 (section 0 @ 0x6FC04)
(0x541cd490) module LIBS:ogles2.library at 0x7EAC4C7C (section 0 @ 0xA8C58)
(0x541cd550) module LIBS:ogles2.library at 0x7EAC5790 (section 0 @ 0xA976C)
(0x541cd890) module LIBS:ogles2.library at 0x7EAC6658 (section 0 @ 0xAA634)
(0x541cd940) module LIBS:ogles2.library at 0x7EA47670 (section 0 @ 0x2B64C)
(0x541cd9c0) module LIBS:ogles2.library at 0x7EA36FC8 (section 0 @ 0x1AFA4)
(0x541cd9f0) module Shaderjoy at 0x7EBB280C (section 0 @ 0x57E8)
(0x541cda80) module Shaderjoy at 0x7EBB1EEC (section 0 @ 0x4EC8)
(0x541cdae0) module Shaderjoy at 0x7EBAD6DC (section 0 @ 0x6B8)
(0x541cdd10) native kernel module Kickstart/newlib.library.kmod+0x0000260c
(0x541cdd60) native kernel module Kickstart/newlib.library.kmod+0x00003338
(0x541cdf10) native kernel module Kickstart/newlib.library.kmod+0x0000385c
(0x541cdf40) Shaderjoy:_start()+0x1e0 (section 1 @ 0x284)
(0x541cdf90) native kernel module Kickstart/dos.library.kmod+0x0002a41c
(0x541cdfc0) native kernel module Kickstart/kernel.debug+0x0007fb70
(0x541cdfd0) native kernel module Kickstart/kernel.debug+0x0007fbe8

Disassembly of crash site:
7ea480e4: 4bfffe38 b 0x7EA47F1C
7ea480e8: 3c00ffff lis r0,-1
7ea480ec: 7c2c0b78 mr r12,r1
7ea480f0: 60007820 ori r0,r0,30752
>7ea480f4: 7c21016e stwux r1,r1,r0
7ea480f8: 7c0802a6 mflr r0
7ea480fc: 38a10008 addi r5,r1,8
7ea48100: bdccffb8 stmw r14,-72(r12)
7ea48104: 3fe50001 addis r31,r5,1
7ea48108: 3dc10001 addis r14,r1,1
Stack pointer (0x541cce60) is inside bounds
Redzone is OK (4)

68k register dump
DATA: a3dbd300 00000000 00000000 00000000 00000000 00000000 00000000 00000000
ADDR: 6ffa6000 a34b0f00 00000000 00000000 00000000 00000000 00000000 541cd3b0
Page information:
Page 0xff8c4540:
Virtual Address: 0x541c4000
Physical Address: 0x0
Lock count: 0
Flags (0x800): (Guard)
Protection bits (0x0): (super state only)
Page is assigned to VMArea primary heap
...


SAM460ex + RadeonRX 550 (Poalris 12)

@Jabirulo
Is 3D works for you at all with that RadeonRX ? Also do you have ogles 2.11 and warp3dnova 1.68, as well as latest RadeonrX drivers ?

@Capehill
Quote:

Yeah, chning 1000 on 100 in the mandelbrot fractal shader give me a lot better fps, around 900 :)

Tried to see if someting like this can be done in foot_torus : and yeah, there some #define AA, which comment says can be set to 1 , but set to 1 give visual noice, set to 2 black screen, but to 3 speed it up a bit in compare with current value 4 for about 30-40%. I.e. 90 fps for "3", and 52 fps for "4".

For kissing_donuts didn't find ability to speed this up.

ps. Tried about 10 random shaderyoy's shader and most of them fail because of iChannel. Maybe have some "filter" in mind by which can parse shadertoy's shaders so they may work on our setups ?

@Capehill

First working shader found which works on our setup:

https://www.shadertoy.com/view/4scBz2

Give 1200 fps on me

@kas1e

Max 1365 fps for me .

@Spektre
Tested after reboot : 1930 :) Seems after i tried lots of shaders which fail, it somehow made impact on the speed of working ones.

Another working one:

https://www.shadertoy.com/view/llV3zz

Through a bit buggy. On os4 there a bit of noise + left side is wrong a little.

@kas1e

Well, I launch W3DNLogo, it shows/animates properly and get this on shell:
Loaded model: Models/Warp3D_text.obj (8202 vertices, 2734 triangles)
Loaded model: Models/BoingBall.obj (480 vertices, 224 triangles)
Loaded model: Models/Nova_Box.obj (396 vertices, 132 triangles)
Loaded model: Models/N_text.obj (492 vertices, 164 triangles)
Loaded model: Models/O_text.obj (1200 vertices, 400 triangles)
Loaded model: Models/V_text.obj (156 vertices, 52 triangles)
Loaded model: Models/A_text.obj (1008 vertices, 336 triangles)
Loaded model: Models/Glass_Twister.obj (864 vertices, 1296 triangles)
#


Warp3DNova.library 1.70 (27-01-2020)
ogles2.library 2.11 (11-11-2019)
RadeonRX.chip 1.17 ( 6-02-2020)

@jabirulo


working for me on Sam460ex with same versions .

Warp3DNova.library 1.70
ogles2.library 2.11
RadeonRX.chip 1.17

Here are my results (X1000 + R7 250X + VSYNC disabled + default Window size):

default (blue window) : 1391

capehill/amigaa : 1374
capehill/bars: 1272
capehill/gear: 1375
capehill/helloh: 1379
capehill/helloh2 : 1369
capehill/mandelbrot: 68
capehill/nuts: 1386
capehill/ripple: 1390

shadertoy/dueling_mandelbulbs : 582
shadertoy/foot_torus: 18
shadertoy/kissing_donuts: 19
shadertoy/sunset: 1315
shadertoy/yetanother_torus: 30

@Spectre660

yep, I "saw" your post, so I tried on my SAM460ex, but...

will try with public OS4.1FE (as I tested under beta system)

thx

@Jabirulo
Some RadeonRx have known issue called "working 2d, compositing, but not 3D (while some 3d still may runs)". To be 100% sure that 3d works on your setup, try some gl4es games firstly, like those:

http://os4depot.net/share/game/fps/night_of_the_zombies.lha
http://os4depot.net/share/game/driving/supertuxkart.lha
http://os4depot.net/share/game/shmup/prototype.lha
http://os4depot.net/share/game/misc/neverball.lha
http://os4depot.net/share/game/shmup/frikingshark.lha
http://os4depot.net/share/game/board/foobillardplus.lha
etc

As well as some pure ogles2/warp3dnova:

http://os4depot.net/share/game/platform/spencertrial.lha
http://os4depot.net/share/game/misc/amicraft.lha

If all works, then it can be fail of shaderjoy, but if not, then you hit the "non working 3d" issue with your radeonrx :)

ps. i do test and on public and on beta too, works everywhere.

@kas1e

Heve not heard of any 3D issues with any RX550 or RX560 models thus far.
Only RX570 and RX580 models.

@jabirulo

According to addr2line, crash happened in glShaderSource. Does it crash always? Is it a certain shader or all of them? Could you enable verbose logging too? And how much stack you have use?

Question

Why do the (Fragment) Shaders coming with Shaderjoy have a varying CPU demand?

In case of Shaderjoy the only (static) Geometry we have is the (window)rectangle which shows (texture mapped to that rectangle) us the output, the procedurally textures created using the fragment shaders.

Why does "dueling_mandelbulbs.frag" has lower CPU requirements than "kissing_donuts.frag" for example?

kissing_donuts.frag causes about 80% of CPU load. Shouldn't that be 0% because of the fact that the shader is being excuted entirely on the GPU ???

@kas1e

Quote:

You can reduce I_MAX to loop less, or increase E value from 0.001 to 0.01, for some speedup. Of course the visual quality may reduce, too.

@Capehill

yep, everytime I double clikc its icon I get a crash/GR (raised stack icon to 300KB, but no luck)


@kas1e
I tried/have some of those games and they work fine. Zombies works, frikinshark works,....haven't test all off them (or I don`t remember)
Spencer trial works.

@IamSONIC

Interesting question. I checked glSnoop PROFILE and in both cases most time is spent in glSwapBuffers and in Nova's WaitDone function. I don't know how waiting is done inside the driver.

@IamSonic,Capehill

I also checked cpu loading of all shaders too, and result are:

default: 62% cpu loading

capehill/amigaaa: 62% cpu loading
capehill/bars: 62% cpu loading
capehill/gears: 62% cpu loading
capehill/helloh: 62% cpu loading
capehill/helloh2: 62% cpu loading
capehill/mandelbrot: 35% cpu loading
capehill/nuts: 62% cpu loading
capehill/ripple: 62% cpu loading

shadertoy/dueling_mandelbubs: 47% cpu loading
shadertoy/foot_torus: 0% cpu loading
shadertoy/kissing_donuts : 100% cpu loading
shadertoy/sunset : 62% cpu loading
shadertoy/yet_another_torus: 100% cpu loading.


I may talking bool, but can't it be that some fragment's functions implemented in fragment shaders in the nova, just "cpu based" for compatibility reassons ?

Also what is interesting, is that if i grab a window when kissing_donuts works, and start to move it : then fps still the same, but while i move window, cpu loading are 3-4% , once i realise mouse pointer, cpu back to 100%. But when i move window by mouse and cpu drops, speed still the same, no changes.