What Software does 68K Amiga need?

[idrougge]

Originally Posted by Locutus

Have fun getting pthreads etc running on AmigaOS....

Why would you need pthreds? AFAIK Git is a program that checks files for changes, computes hashes and creates a config file.

[Marlon_]

A unified package manager platform for AOS/AROS would be great. Something like apt for Debian.

[Thorham]

Originally Posted by meynaf

Pure user software doesn't have to. But system tools have.

Sure, but can't the OS just supply the needed data? Give some examples of when not having access to system structures is a problem.

Originally Posted by meynaf

Who does that memory belong to ?

Two communicating tasks simply have their own message buffers, they don't share a single buffer.

Originally Posted by meynaf

For the others it's not visible. If you have full memory protection (paged) then even the address is wrong (may map anywhere in your process). And there are eventual pointers in the message itself.

All very true, but remember that the memory protection has to be optional, and that it should be possible to turn it on and off at will. You can't have each program have it's own address space like this.

Originally Posted by meynaf

I don't use WAV because i don't want to waste disk space.

I rather have low CPU usage to be honest.

Originally Posted by meynaf

Btw AIFF is more suited for the Amiga (no little endian to convert).

Not relevant if you use calibration. Just setup a calibration table for little endian samples.

Originally Posted by meynaf

Anyway you can't do that for video.

AAC isn't video Also, why not just use your peecee for video?

[meynaf]

Originally Posted by Thorham

I wish assemblers could do this

It's not their job. You have to write some code that does it for you. Just put debug checks everywhere.

Originally Posted by Thorham

Sure, but can't the OS just supply the needed data? Give some examples of when not having access to system structures is a problem.

That's for cases the OS doesn't plan anything.
Do "modern" OSes allow you to walk their task lists ? Their interrupt lists ? Their library lists ?
Maybe they do, but if they miss something - it then becomes impossible while on AOS it's still doable.

Originally Posted by Thorham

Two communicating tasks simply have their own message buffers, they don't share a single buffer.

If they don't share any buffer then they can't communicate. Both have to access the same memory area.

Originally Posted by Thorham

All very true, but remember that the memory protection has to be optional, and that it should be possible to turn it on and off at will. You can't have each program have it's own address space like this.

This implies some programs won't work with memory protection turned on - because the relevant logic to make them work in that case is simply not there.

Originally Posted by Thorham

I rather have low CPU usage to be honest.

No big deal for me, as long as there is enough. High cpu usage is temporary, high disk space usage is permanent - not to mention new files can be played directly.

Originally Posted by Thorham

Not relevant if you use calibration. Just setup a calibration table for little endian samples.

Not an easy thing to do. Calibration code isn't exactly straightforward.
Besides, it only goes one way (= not when you write samples to disk) and for the special case when you don't have to downsample.

Originally Posted by Thorham

AAC isn't video Also, why not just use your peecee for video?

'Course AAC isn't video (even though it's often embedded in it) but the point is that you can't save your video as raw data before playing it.
I could have said as well : why not just use your peecee for audio ?

[idrougge]

Originally Posted by meynaf

I could have said as well : why not just use your peecee for audio ?

Or, to get back on topic, why not use your pc as a transcoder? That's common for far more powerful hardware than the Amiga. All we need is a suitable client.

[DrBong]

Originally Posted by Thorham

A new operating system that's NOT compatible with AmigaOS.

Originally Posted by b0lt-thrower

You mean m68k BSD?

Originally Posted by Thorham

No, I don't. I mean an OS that does what AOS does, only better and faster with optional memory protection.

Originally Posted by ExiE

Do you remember QNX4? Posix compatible system with GUI and browser on one 3.5" floppy. This one was IMHO one of the best chances Amiga ever had and blow on software field.

Originally Posted by meynaf

QNX is a lot slower than AOS.

How about BeOS then (or Haiku as the open source version is known now)?

If it isn't too heavy for 68K (perhaps a lite version if it is?), then there would be a reasonable foundation of software ready to use if someone managed an optimised port to Amiga. Also, it's probably one of the most clean Amiga-like operating systems that has been developed since AmigaOS.

@ExiE
Are you the guy who used to run Czech Amiga News?

[Thorham]

Originally Posted by meynaf

It's not their job. You have to write some code that does it for you. Just put debug checks everywhere.

I know, but it may be possible with a 68k VM with JIT. Don't know if it would be overly useful...

Originally Posted by meynaf

That's for cases the OS doesn't plan anything.
Do "modern" OSes allow you to walk their task lists ? Their interrupt lists ? Their library lists ?
Maybe they do, but if they miss something - it then becomes impossible while on AOS it's still doable.

Actually, I don't see a problem with allowing reading the lists. In fact, any actions that can't screw up the system should just be allowed.

Originally Posted by meynaf

If they don't share any buffer then they can't communicate. Both have to access the same memory area.

I didn't explain that very well. They share each others message buffer, but they can only read message buffers that they receive pointers to. That way, tasks can't alter each others memory, but they can read it for memory blocks where that's relevant.

Originally Posted by meynaf

This implies some programs won't work with memory protection turned on - because the relevant logic to make them work in that case is simply not there.

Such as?

Originally Posted by meynaf

No big deal for me, as long as there is enough. High cpu usage is temporary

Yes, and while the CPU is almost maxed out you can't do much else with your machine. On my 68030 I can only playback mp3s in crappy quality, and the CPU is almost maxed out. What use is that?

Originally Posted by meynaf

high disk space usage is permanent

I have a 500GB HD on my Amiga, not a problem.

Originally Posted by meynaf

Not an easy thing to do. Calibration code isn't exactly straightforward.

Isn't it just reading a table?

Originally Posted by meynaf

Besides, it only goes one way (= not when you write samples to disk) and for the special case when you don't have to downsample.

I just checked, and WAV supports big endian.

Originally Posted by meynaf

I could have said as well : why not just use your peecee for audio ?

Fair enough.

[Thorham]

Originally Posted by DrBong

How about BeOS then (or Haiku as the open source version is known now)?

If it isn't too heavy for 68K (perhaps a lite version if it is?), then there would be a reasonable foundation of software ready to use if someone managed an optimised port to Amiga. Also, it's probably one of the most clean Amiga-like operating systems that has been developed since AmigaOS.

That's worth taking a look at.

[meynaf]

Originally Posted by Thorham

I know, but it may be possible with a 68k VM with JIT. Don't know if it would be overly useful...

No it's not possible. As long as the memory belongs to your process, nothing will identify out-of-range accesses. So you can still happily overflow array bounds.

Originally Posted by Thorham

Actually, I don't see a problem with allowing reading the lists. In fact, any actions that can't screw up the system should just be allowed.

For reading the lists you need to disable multitasking - else they can change while you read. Forget to reenable it and you screw up the system.

Originally Posted by Thorham

I didn't explain that very well. They share each others message buffer, but they can only read message buffers that they receive pointers to. That way, tasks can't alter each others memory, but they can read it for memory blocks where that's relevant.

So everyone's read only for the other ?

Originally Posted by Thorham

Such as?

You speak about optional memory protection, but it has nothing to do with the fact a program will work or not. If a program doesn't have the added overhead memory protection implies, then it won't work when it's active.

Originally Posted by Thorham

Yes, and while the CPU is almost maxed out you can't do much else with your machine. On my 68030 I can only playback mp3s in crappy quality, and the CPU is almost maxed out. What use is that?

Switch to a better cpu
Anyway you can still code (= type text in an editor).

Originally Posted by Thorham

I have a 500GB HD on my Amiga, not a problem.

Not everyone has that. I have a 15GB CF.

Originally Posted by Thorham

Isn't it just reading a table?

I was speaking about the code that actually writes the table.

Originally Posted by Thorham

I just checked, and WAV supports big endian.

Where did you check ?
All docs i've read say little endian.
https://en.wikipedia.org/wiki/WAV

[Locutus]

Originally Posted by idrougge

Why would you need pthreds? AFAIK Git is a program that checks files for changes, computes hashes and creates a config file.

I know exactly what Git is.....

Theres loads reasons to use threads and can assume that you run on a multiprocessor machine where overlapping I/O requests improve throughput, you can just check the code yourself to see ;-)

And if you grep and you see loads of #ifdef NO_THREADS, please actually read the codepaths given there....

[Thorham]

Originally Posted by meynaf

No it's not possible. As long as the memory belongs to your process, nothing will identify out-of-range accesses. So you can still happily overflow array bounds.

You can just handle assembly language like a script language. The question is whether it's useful or not. Probably not

Originally Posted by meynaf

For reading the lists you need to disable multitasking

That's very true. I find this kind of thing quite dirty, though.

Originally Posted by meynaf

So everyone's read only for the other ?

In the message buffer case, yes. Don't see the problem with it. Tasks simply use their own buffer to replay to a message instead of reusing the received buffer.

Originally Posted by meynaf

You speak about optional memory protection, but it has nothing to do with the fact a program will work or not. If a program doesn't have the added overhead memory protection implies, then it won't work when it's active.

But where does most of the overhead come from? I don't see how a simple memory protection scheme is going to generate much overhead except for the MMU, but I might be wrong.

Originally Posted by meynaf

Switch to a better cpu

Well, I have a peecee...

Originally Posted by meynaf

Anyway you can still code (= type text in an editor).

Unworkable. The whole system slows down when most of the CPU is in use. It's just not practical.

Originally Posted by meynaf

Not everyone has that. I have a 15GB CF.

That's true, but when you're on a 68030, I don't see the usefulness of sacrificing sound quality, just so that you can almost grind the system to a halt.

Originally Posted by meynaf

I was speaking about the code that actually writes the table.

Just count through all 16 bit combinations, endian convert them, look them up in the calibration table as if they're normal samples, and write them out to a new table. I don't see how that's very hard.

Originally Posted by meynaf

Where did you check ?
All docs i've read say little endian.
https://en.wikipedia.org/wiki/WAV

The Sox documentations says that. There are two formats. RIFF and RIFX. They're identical except for the endian and the identifier in the header.

[Samurai_Crow]

AIFF is big endian.

[meynaf]

Originally Posted by Thorham

You can just handle assembly language like a script language. The question is whether it's useful or not. Probably not

Ok. Let's say you've found the instruction move.l (a0)+,d0. Now how to determine it's valid or not ?
No, really, it's simply not possible.

Originally Posted by Thorham

But where does most of the overhead come from? I don't see how a simple memory protection scheme is going to generate much overhead except for the MMU, but I might be wrong.

How do you handle the memory lists ? Remember, for memory protection you have to know which areas belong to whom. Might be very funny in case of heavy memory fragmentation.

Originally Posted by Thorham

Well, I have a peecee...

Oh yeah ? Didn't know

Originally Posted by Thorham

Unworkable. The whole system slows down when most of the CPU is in use. It's just not practical.

I don't see any slow down when just typing.
Perhaps you're just using the wrong texteditor+soundplayer combination

Originally Posted by Thorham

That's true, but when you're on a 68030, I don't see the usefulness of sacrificing sound quality, just so that you can almost grind the system to a halt.

Aren't you a little bit exaggerating ?
90% cpu use isn't grinding to a halt.
And for many files quality remains acceptable.

Originally Posted by Thorham

Just count through all 16 bit combinations, endian convert them, look them up in the calibration table as if they're normal samples, and write them out to a new table. I don't see how that's very hard.

Cool, another 128kb table to allocate

This will not bring you far. If you get a 44100hz wav to be played on a machine that's not in productivity mode, you're out. Better endian convert in all cases.

Originally Posted by Thorham

The Sox documentations says that. There are two formats. RIFF and RIFX. They're identical except for the endian and the identifier in the header.

Yea maybe. But 99.99% of existing wav files are little endian.

[idrougge]

Originally Posted by Locutus

Theres loads reasons to use threads and can assume that you run on a multiprocessor machine where overlapping I/O requests improve throughput, you can just check the code yourself to see ;-)

And if you grep and you see loads of #ifdef NO_THREADS, please actually read the codepaths given there....

No-one said they wanted a port of the official Git client.

[Thorham]

Originally Posted by meynaf

Ok. Let's say you've found the instruction move.l (a0)+,d0. Now how to determine it's valid or not ?

Just check against some bounds. Bounds can be checked by keeping track of memory allocations. With a special assembler, you could even track dcb.x and ds.x directives. You can check if custom chip registers are read or written. It won't be perfect, but you can certainly do more than just nothing at all.

Originally Posted by meynaf

How do you handle the memory lists ? Remember, for memory protection you have to know which areas belong to whom.

The OS keeps track of that. Not exactly difficult.

Originally Posted by meynaf

Might be very funny in case of heavy memory fragmentation.

That probably won't be better than AOS: RESET There may be some possibilities, but I don't know if they're practical.

Originally Posted by meynaf

Perhaps you're just using the wrong texteditor+soundplayer combination

I use FrexxEd which is a heavyish editor, but I want a little more functionality than other text editors and have spent a lot of time overhauling the thing. FrexxEd ain't going nowhere

Originally Posted by meynaf

90% cpu use isn't grinding to a halt.
And for many files quality remains acceptable.

What?

~90% -> crappy mp3 at 22khz mono
~20% -> original quality wav/aiff at 28khz stereo (20% is a guess)

That's NOT a choice

Originally Posted by meynaf

Cool, another 128kb table to allocate

I didn't say it was a good idea, although it's ten times better than 90% CPU usage But, it's not really necessary if you use a file with the right endian, of course, so it's not important.

Originally Posted by meynaf

This will not bring you far. If you get a 44100hz wav to be played on a machine that's not in productivity mode, you're out. Better endian convert in all cases.

What does the sample rate have to do with endian conversion? Also, 44.1khz would have to be down sampled in real time, which is far worse. Better if the user just uses a properly down sampled 28khz file. Sounds better, and uses less CPU.

Originally Posted by meynaf

Yea maybe. But 99.99% of existing wav files are little endian.

I make my own WAV files, so it's of no concern to me. A proper WAV codec has to be able to handle both anyway, and Samurai_Crow just gave AIFF as a little endian option. Just tried AIFF with Hippo, works fine.

[meynaf]

Originally Posted by Thorham

Just check against some bounds. Bounds can be checked by keeping track of memory allocations. With a special assembler, you could even track dcb.x and ds.x directives. You can check if custom chip registers are read or written. It won't be perfect, but you can certainly do more than just nothing at all.

And you will catch many false errors.

Originally Posted by Thorham

The OS keeps track of that. Not exactly difficult.

Well, everything is easy when you don't have to do it, right ?

Originally Posted by Thorham

What?

~90% -> crappy mp3 at 22khz mono
~20% -> original quality wav/aiff at 28khz stereo (20% is a guess)

That's NOT a choice

What ?

3Mb -> mp3 at 22khz mono that's preloaded in memory (and doesn't play that bad)
30Mb -> huge file that has to do permanent disk access while playing (making HD access noise on top of the music)

That IS a choice

Anyway, what about :
~66% -> 28khz stereo FLAC

Originally Posted by Thorham

What does the sample rate have to do with endian conversion? Also, 44.1khz would have to be down sampled in real time, which is far worse. Better if the user just uses a properly down sampled 28khz file. Sounds better, and uses less CPU.

You can't down sample with the wrong endian, that's the point.
If i release a sound player and tell the users : oh, by the way, downsample your 44.1 prior to playing - NO GOOD.

Originally Posted by Thorham

A proper WAV codec has to be able to handle both anyway,

Will you stop using Hippo if it doesn't ? Perhaps you have to try.
Me, i won't support anything without having real files using it. If Sox can really produce them (i.e. if it's not an internal, unused option), then i'd like to have a few (i won't install Sox or anything else just for this, m'kay).

Originally Posted by Thorham

and Samurai_Crow just gave AIFF as a little endian option.

He said AIFF is big endian. AIFF has no little endian option.
Only stupid thing with AIFF is the way it stores the replay frequency.


Btw. What software does 68k Amiga need ? A new sound-player that takes over what Delitracker did. And i'm working on it.

[Thorham]

Originally Posted by meynaf

And you will catch many false errors.

Perhaps.

Originally Posted by meynaf

Well, everything is easy when you don't have to do it, right ?

What's so hard about keeping track of which task allocated which memory? None of this is rocket science.

Originally Posted by meynaf

doesn't play that bad

It sounds crappy to me

Originally Posted by meynaf

making HD access noise on top of the music

Can't hear it with my headphones

Originally Posted by meynaf

~66% -> 28khz stereo FLAC

That's a lot better.

Originally Posted by meynaf

You can't down sample with the wrong endian, that's the point.

Yeah, I indeed missed that

Originally Posted by meynaf

If i release a sound player and tell the users : oh, by the way, downsample your 44.1 prior to playing - NO GOOD.

You don't have to, but I refuse to use real time down sampling because Sox sounds much better.

Originally Posted by meynaf

Will you stop using Hippo if it doesn't ? Perhaps you have to try.

No, I won't It doesn't support RIFX, though.

Originally Posted by meynaf

i won't install Sox or anything else just for this, m'kay

It's just a folder with some files, no need to install anything. Runs from the command line. If you're serious about writing that player, then I suggest that you use it, because it can produce more formats than WAV and AIFF. Can't hurt to take a look.

Originally Posted by meynaf

He said AIFF is big endian. AIFF has no little endian option.

Yeah, that's what I meant to write, actually

[meynaf]

Originally Posted by Thorham

What's so hard about keeping track of which task allocated which memory? None of this is rocket science.

It becomes fun when you have tasks allocating many small memory chunks, that's all.

Originally Posted by Thorham

It sounds crappy to me

MP3 always does, doesn't it ?

Originally Posted by Thorham

Can't hear it with my headphones

Cruel guy. You don't even hear your HD grumbling about being abused

Originally Posted by Thorham

It's just a folder with some files, no need to install anything. Runs from the command line. If you're serious about writing that player, then I suggest that you use it, because it can produce more formats than WAV and AIFF. Can't hurt to take a look.

It can hurt, if i end up with thousands of nearly never used format variations (especially some that are poorly documented).
I would like to know if there are really RIFX WAVE files out there. Where's the point in supporting something nobody will ever use ?

[Ratteler]

I'm dreaming here, but I would love OpenOffice and Blender3D.
Gimp, Inkscape, and FreeCAD wouldn't hurt either.

If I had to choose just one, it would be Blender3D for Blender Game Engine compatibility.
If BGE games could be played on the Amigas, we would have a world of new games to choose from.

[Zetr0]

+1 for Blender3D

it would appear I am a sucker for any 3D really

If the Vampire / Apollo project has shown anything to the community at large, it would be that FPGA's will get faster and cheaper - in my exuberance of this I can unequivocally say that eventually all working Amiga's will have one. Maybe not to day or tomorrow, but most assuredly eventually.

So Raytraced and or Texture mapped 3D please, and then I want the Unity Engine if you still have some steam for that