GL4ES / SDL+GL4ES / CLIB2 / Outdated NEWLIB libs and OnWards

@Sinan
Looks like some mess happens : you have errors like SDL1 and SDL2 swapped (i.e. when you try to compile SDL2 version, it call to undefs for SDL1), and when you try to compile SDL1 version, you have undefs like you use not clib2 version, but newlib one instead.

I think that happens because you use my makefile for examples which "../../lib/AmigaOS4/libIrrlicht.a" as you can see, mean it will use one you build (which you may have wrong version there , etc). So to build it correclty, you need to run manually like this:



Something like this to be sure (so you should have in clib2 directory in your SDK my libIrrlicht_SDL2.a)


@Petrol
Quote:

Both and SDL1 and SDL2 versions ? If so, then as Hans says INTERRUPT=YES should be

@Hans

I too have interrupt set to NO (due to random freezes otherwise) and i get ~160 FPS without any slowdowns (but with CPU at 100%).

What i meant was that i could do tests and report back with logs regarding those newly uncovered issues with the RX driver...i thought maybe because of the recent X1000 hack there might not have been much (if any) betatesting going on on the X1000 platform with this driver...

@kas1e

Sorry, but my case is kind of awkward to install a gfx board (especially the big HD 7950), so to not damage the hardware i'll stick with the RX driver/card combo for now.

@kas1e

Thanks...you are right :)

Ok I can compile now

putenv fixed in beta9 branch

@Petrol
Thinking more about i think you may have something running in background, which cause those pauses every few seconds.

@Raziel
Quote:
No worry, it's indeed better to not touch amiga hardware if there no real needs.

@Andrea
Thanks!

Tested on a Fresh OS install, no lags (need to see what is interfering on my other partition).

The results :
SDL1 : 109 FPS
SDL2 : 110 FPS

@kas1e

Where are the SDL2 tests in Irrlicht?
I can only find the SDL 1.2.15 inisde the archive bin/amigaos4 (at least that's what the output log tells me)

@Raziel
Quote:

SDL1 test: https://kas1e.mikendezign.com/a/irrlicht_quaketest_sdl1.zip
SDL2 test: https://kas1e.mikendezign.com/a/irrlicht_quaketest_sdl2.zip

@Petrol
Quote:

So mean SDL1/SDL2 version the same by speed, and that it was indeed something in background. My bet it will be some kind of docky (try firstly to disable whole amidock and/or xdock and then step by step enable one by one).

@All

So to summorize that what we have:

Sam460 + RadeonRX : ~34 FPS

X1000 + RadeonHD : ~110 FPS
X1000 + RadeonRX : ~160 FPS

X5000 + RadeonRX : ~260 FPS


Mean that:

1). SAM460 itself and/or GART disabling make things be bad and slow.
2). X1000 even with RadeonRX still slower much than X5000 for those tests (dunno why, maybe older memory ? The CPU is just 200mhz slower, it should't be 40% slower imho?).


This particular test, is one of the tests which strangely slow in compare with the same opengl tests on other platforms (linux/win on x86). Maybe this big differenes there can point us on something as well ..


Can some of you also test on X1000 with radeonHD and radeonRX , and on sam460 with RadeonHD and RadeonRX, from this archive:

http://os4depot.net/share/development ... ary/graphics/irrlicht.lha

That test:

bin/AmigaOS4/26.OcclusionQuer

with compositing enabled, and without compositing enabled.

That test is one with simple geometry draw ones, so we can see how this one reacts, and make a consclusion if those differences we have related to many draw calls, or in the whole to any drawing operations (that may help to diagnostic some problems on which Hans may works on)

@kas1e

bin/AmigaOS4/26.OcclusionQuer on SAM460ex & RX550:

composition off: ~580fps
composition on: ~550fps

@Javier
Quote:

OMG, Sam460 is really bad for 3D currently then. On x5000 for that "26.OcclusionQuer" test i have:

compositing off: ~3450 FPS
compositing on: ~3100 FPS


Interesting if it GART issue or general SAM's perfomance ..

@kas1e

The difference in performance predates the addition of GART to the drivers.

I recall seeing the difference in your Neverball ports.
Sam460ex never more than about 17 fps .

@kas1e

Here are updated results from X1000+RadeonRX(560) and CANDI turned off

02.Quake3Map

SDL1 - ~220 FPS (looking straight to the floor)
SDL2 - ~220 FPS (looking straight to the floor) - with slight higher top count

26.OcclusionQuer

~3000 FPS (with occluder)
~3600 FPS (w/o occluder)

@Spectre660

while doing those tests, does your CPU load go to 100%?
Because if so, then that will be the bottleneck

@Raziel
Oh, now its start to looks good and correct.

So, that mean that anyone who have CANDY on background and want proper speed of 3D , should disable it compeltely. But that kind of expected :)

All this mean X1000 with RadeonRX drivers are "ok".

Is it with interrupt=no or =yes ?

Quote:

It should be 100% , that ok. It's benchmark programm, will eat always 100% to hit maximum limits (as almost all SDL apps do). It is the same 100% on x5000 for me too when i do my tests.


@Petrol
Do you have Candy as well enabled ?

@All

Now what is interesting, is why Sam460 _THAT_ slower in compare with x1000 and x5000 ? I mean i expect it to be .. ok, on 50% slower than X1000, but not like now:

x1000 + rx = 220 FPS
x5000 + rx = 260 FPS
sam460 + rx = 34 FPS (!)

Strange!

@Spectre

Quote:

But why it so slow ? I mean, sam460ex : 1.1GHZ + more or less normal memory speed and stuff + RadeonRX on top of it, and it slower than x1000 for about 5-8 (!) times depends on tests !

026.OcclusionQuery

Compositing ON : 1456 FPS with occluder ON / 1790 FPS with occluder OFF
Compositing OFF : 1475 FPS with occluder ON / 1820 FPS with occluder OFF

No Candi installed on my system.

@Petrol
But something still wrong then on your setup. See values from Raziel mean you should have about the same if nothing bad in background running and if you have all latest drivers and libs.

EDIT: Ah, you on RadeonHD ? Not on RX ?

@kas1e

INTERRUPT set to NO here...might try it with YES setting now that my system is back to being rock stable.

@Petrol said in an earlier post he's using X1000 + R9 280X, so a Radeon HD.
I know that my RX (560) absolutely stomps my HD (7950) when it comes to pure grunt, so that might already be the reason...also, i read somewhere that the R9 series of cards were close to (if not a little inferior speed-wise) to the HD range of cards...

and wrt the SAMs...if the benchmarking occupies the cpu as well, shouldn't that explain the lower outcome already, even if the margin i *that* high?
I don't really know, but isn't the SAMs cpu inferior to the X1000/X5000 (by far?) and thus wouldn't gain as much speed from the benchmarks?

But maybe i'm misreading something?

@Raziel
Quote:

Yeah will be interesting to see if there will be difference.

Quote:

Of course CPU involved, sure. But _that_ low speed ? I mean, x1000 - 1.8ghz , x5000 - 2ghz, sam - 1.1ghz. Yes, we can say that sam have custom soc, yes memory slower, but _that_ much ? in 10 times ?:)

I expected it to be like this with radeonrx: x5000: 260 fps, x1000 : 220 fps, sam460: 100-120 fps. Like that to have "sane" difference. Not 34 fps :)

@kas1e

Another updated results from X1000
- RadeonRX(560)
- CANDI turned off
- INTERRUPT=Yes

02.Quake3Map

SDL1 - ~225 FPS (looking straight to the floor)
SDL2 - ~225 FPS (looking straight to the floor) - with slight higher top count

26.OcclusionQuer

~3000 FPS (with occluder)
~3600 FPS (w/o occluder)

@kas1e

At the moment the way to build the SDL2+GL4ES library is to pull down SDL2 and then apply the changes to that and then build. This is all described above.

What I wanted clarity on is on:




You authored those changes. Did you do these changes or someone else?

As well, we know that modifications need to be made after these changes applied to the base SDL2 library, e.g. the active window function because they are a little out of date.

My question is: is there a better method than making a diff file of 4d504983f3876bdf714af20a8bdfed757831b1d5, applying it to the very latest SDL2 branch (currently version 2.24) and then fixing up some of the differences.