Understandably not. I just wonder if you could make a v2.30.5-amigaos4 release with that fix?

@all
As a start, I have built a version of a cross-compiler and native-compiler; a pure GCC 6. There are no additional libraries right now.

Can someone test these. From there, I can add additional libraries. See post #1 for the links.

@MartinWQuote:

The A1222 doesn't need much cooling.
I have a beta system and it was supplied with a super tiny fan that was quite loud.

I can't remember what i replaced it with as it's been a few years since i did it but probably a 40 or 60mm noctua fan and a 3d printed fan adaptor.

I think that's fairly normal for the 5040. I stressed over it for a while but there's little I can do. The specs for the CPU say it's good for way more heat than that. The case it's in is already cavernous with chimney effect (vertical) cooling.

The only thing that would get it much better would be liquid cooling but then you're back into where do you get the right size block and mountings for a CPU that's designed to live in data centre router type equipment where noise is just not a concern.

The design is just poorly thought out in terms of where people that don't like jet engines under their desk (because I already have crappy tinnitus) are going to get cooling solutions from. I want to save what little decent hearing I have left for my guitar abuse

I have no idea what the A1222 is like, but I sincerely hope it's quieter or has better options.

@Tuvok

Amiga NG hardware +AOS4 is retro hardware... even if someone doesn't like this statement . I think almost everyone does it because they like a ‘have fun’

@walkero

Quote:

I checked a few things today but I didn't come across anything wrong with the SDL2 related code.
But I noticed the same thing as in your video what is occurring with me . The SDL2 version even when you do nothing eats a lot of CPU for you and me.
SDL1 in non-use mode eats about 1%CPU and SDL2 about 20-30%.
It is possible that the main loop or things related to the Event, the SDL_Delay setting in the program itself are causing the problem.

This sound all very encouraging for first time and wanna-be developers or people who want to "remember to have fun" developing for Amiga. Not.

Makes sense.

You merged that directly into the latest release?

@IamSONIC

Quote:

I tried two months ago to check something for the SPE.
At the moment the SPE topic is closed with me. I do not have an A1222, I only have QEMU. It is too time-consuming and I have not been able to test the programme at home.
I had to send something to the person who owns the A1222 to check it.

From what I remember it was
rebuilding a cross comlining environment with clib4 under SPE

..
git clone https://github.com/sba1/adtools.git
cd ~/adtools/gild
git submodule init
git submodule update
cd ~/adtools
guild list
- from the list, select
gild clone
gild checkout binutils 2.23.2
gild checkout coreutils 5.2
gild checkout gcc 6
...

Rebuilding clib4 under SPE
Change ‘GNUmakefile.os4’ options adding line ‘SPE=true’

After finishing the compilation, I had to completely rebuild ‘libgcc’ under clib4 and by changing the options in the makefile options related to ‘-msoft-float’ (and flags under SPE ) to avoid ‘hard float/ soft float’ warning.
You will get libgcc.a ..etc under soft float.
The base was ready.

Then rebuild zlib and compile SDL2. i think i disabled the extra GL stuff because it was not compiled under soft float.
Then you use SDL2 compiled with flags under SPE without GL.
Then you probably have to recompile all the extra libraries from scratch to build other things fully SPE-compatible i new base.

At least I think so, and I don't even know if I went the right way and it wasn't stupid
There wasn't a lot of material on the subject at the time.

@rjd324

Thanks to You for reporting the issue!

With my current understanding this doesn't require more implementation, in other words, filtering mode is set when texture is used and will be fed to CompositeTags() call. But this dummy function is required because SDL2 doesn't check whether function pointer is NULL or not (hence the ISI exception).

@Capehill
Thanks for quickly implementing something. Are you just stubbing out that function or are you planning on filling it in later? I ask wrt. commit https://github.com/AmigaPorts/SDL/comm ... 1e7d14451140fba6c4409d8b6

@rjd324

Quote:

I read that too and trying to compile simply everthing directly for SPE is also for me just a Test right now. If this way will work than we have a higher chance for more (faster) Apps on A1222. If every single App must be analyzed and tracked down in order to compile core floating point math operations for SPE and generic seperately then ...

@afxgroup

Quote:

Yes the mixing is odd but as you see in the SDL2 Thread using clib4 nonSPE it works without a problem. And not just the sdl2benchmark, till now everything. The main problem right now is that those libraries doesn't support SPE.

Does someone can provide those libs for GCC 6.4 built for SPE?

@IamSONIC

Quote:

That was to be expected @smarkusg has already ported some emulators via clib4 and it was simply faster. I also use a ScummVM-Clib4 version from @afxgroup and the result is the same, it seems to be a bit more stable than with newlib.

See: https://www.amigans.net/modules/newbb/ ... id=149822#forumpost149822. In this case, this is a successful demo where CLIB4 and SDL2 were not compiled with SPE flags at all.

It would be nice to good agreement or decisions as to whether or not
- we should be building C-libraries with the SPE flags in full, and whether
- we should be building SDL2 with SPE flags in full.

It would be nice to able to just do something like:

But as https://keasigmadelta.com/blog/taboriz ... imizing-for-the-e500-spe/ *bullet point 2* says, only critical code actually _using_ floats should be compiled with such flags in order to avoid the trashing of the ABI float parameters.

Presumably, we all need to be careful.

@rjd324

Thanks for starting this thread, there should be some clarity here and it could be of great value to developers and testers.

Some claim that it is not possible for Heretic2 to work on the A1222 neither in software rendering nor with Warp3d/gles. So let's just find out together.

@walkero

Quote:

Thanks for your support

@afxgroup

It works!

- clib4
- SDL2 against clib4
- the SDL2 Tests inside ../test against clib4

and just for comparison 2 benchmarks from X5000

clib4:


newlib:


clib4 Version beats newlib Version!

btw. Thanks for your great work with clib4 !!!

@rjd324

Quote:

No.

Quote:

It looks like this function implementation is missing in SDL2. Tests don't seem to use it so it wasn't noticed before :( I will try to fix this.

try to use clib4 without SPE and see if it works correctly

@walkero

Fantastic, look forward to it!