@olegil

Quote:

olegil wrote: @cdimauro However, in HINDSIGHT, it would have been more convenient NOW if it was possible to just clock a '000 implementation extremely fast rather than having to add all the extras from the '020+.

68020 added also some useful stuff.
Quote:

Btw, you can run 3.1 on 68000.

It's good for 68000 users, but I'm greatly disappointed to hear that, because it means that the Kickstart 3.x wasn't optimized for 68020+ processors.

We can discuss about some questionable design decisions that Motorola did with that processor, which crippled in some way the processor family evolution, but once you have some features which help a lot the execution speed, not using them is a very stupid idea.

To be more clear, I don't say that the Kickstart 3.0 shouldn't run on a 68000, but that Amiga 1200 and 4000 (and 3000, via changing the ROMs or with SoftKick) should have used a properly 68020-optimized version of it.
Quote:

I still think any new HW recreations of amiga chipsets should be compatible with existing accelerator cards.

I don't think so. What is needed is an hardware support to RTG extensions, but this doesn't mean that you've to replicate the same graphic cards chipsets, which is overkill (and, again, another stupid decision).


@kolla

Quote:

kolla wrote: @cdimauro Quote: cdimauro wrote: I don't remember if version 3.0 supported only 68020. However it made sense to use it as a basis, since all AGA machines had at least a 68020. That was the point, and I agree with the decision they made. Look: we are talking about late 1992, not 1985. After 7 years continuing to stick with a 68000 is too much. IMO. No, we are not talking of late 1992 - we are talking year 1999, when OS3.5 was released by Haage&Partner, and 2000 when OS3.9 was release. All CBM incarnations of Amiga OS run fine on 68000. Before 3.1 was released, many friends of mine soft kicked A500 with 3.0 roms ripped from A1200 and installed 3.0. No problems. Also, many OS3.0 beta testers (and developers) did not have AGA systems (they were not available) and did their testing (and development) on systems with only 68000.

OK, but see my answer to olegil regarding this.
Quote:

The requirement of 68020 was some bright idea that came about when Haage&Partner did OS3.5, and 3.9. I have gone through all components of OS3.9 and the following boingbag packs to pick out the bits and parts that really require 68020 - they are rare and far between! - very little in OS3.9 is "optimized" for 68020, so little that I honestly do not see the point, the requirement is very artificial. My Minimig has a 680SEC00 processor that can run at 50MHz and it is still produced and cheaply available. For 68020 the situation is different. My Minimig runs OS3.9 kickstarts and OS3.9 Workbench. A few important components requires 68020, such as resource.library and I believe parts of Reaction. But Reaction is just ClassAct, for which there is 68000 version that runs fine - another artificial requirement. The biggest annoyance is that OS3.5+ resource.library is 68020+, and without it, the Prefs programs for example will not work. But for most part, one can use 3.1 Prefs programs, the only Prefs program that really changed was the one for Workbench.

I agree that only for such small changes it was a stupid idea to introduce 68020+ support for 3.5 & 3.9 versions of the Amiga o.s..

IMO at the time a 68020 should have been a minimum requirement, but only if the o.s. took HUGE advantage of the new processor features (read: not small changes, but a good recompilation and optmizations of o.s.).

@cdimauro

Quote:

It's good for 68000 users, but I'm greatly disappointed to hear that, because it means that the Kickstart 3.x wasn't optimized for 68020+ processors.

3.x is faster in some operations than 1.x - eg. GFX. Difference is noticeable on basic A500. I was surprised by speed of soft-kicked 3.x OS on my A500 in comparison to normal 1.x.

Quote:

I agree that only for such small changes it was a stupid idea to introduce 68020+ support for 3.5 & 3.9 versions of the Amiga o.s..

Reaction would be sloooow on 68000 (it is not that fast even on 68020), OCS is too weak to display GlowIcons (yes, it is possible, but 16 colour 3.5/3.9 WB doesn´t look good). In 1999, there weren´t any FPGA boards comparable to 68030 class performance. Market for 3.5/3.9 was clear: AGA machines and expanded (CPU, GFX card) A2000/3000 computers. I don´t think anybody even imagined, that after 15 years someone will require 68000 support, because his FPGA hardware implementation is not compatible enough with 68020+.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: It's good for 68000 users, but I'm greatly disappointed to hear that, because it means that the Kickstart 3.x wasn't optimized for 68020+ processors. 3.x is faster in some operations than 1.x - eg. GFX. Difference is noticeable on basic A500. I was surprised by speed of soft-kicked 3.x OS on my A500 in comparison to normal 1.x.

That's because the genius which developed the o.s. used the Blitter even to draw a single pixel on the screen.

Then with o.s. 2.0 or 3.0 such graphic routines were optimized, and the CPU was used instead of the Blitter for some tasks like this.

Anyway, a 68020 version of the kernel would have benefited by the new features.
Quote:

Quote: I agree that only for such small changes it was a stupid idea to introduce 68020+ support for 3.5 & 3.9 versions of the Amiga o.s.. Reaction would be sloooow on 68000 (it is not that fast even on 68020), OCS is too weak to display GlowIcons (yes, it is possible, but 16 colour 3.5/3.9 WB doesn´t look good). In 1999, there weren´t any FPGA boards comparable to 68030 class performance. Market for 3.5/3.9 was clear: AGA machines and expanded (CPU, GFX card) A2000/3000 computers.

Exactly. But requiring a 68020 only for a few things was a mistake.
Quote:

I don´t think anybody even imagined, that after 15 years someone will require 68000 support, because his FPGA hardware implementation is not compatible enough with 68020+.

Let's say this: such o.s. versions aren't meat neither fish.

Consider that you have an A2000 with a 68060 card and something fails and part of debugging procedure is to remove the acc card - how much would it not suck if you couldn't even load kickstart and boot into shell and access your data using the 68000.

020 optimization brings marginal speed improvements in the OS components, just look at the micro optimization done by Cosmos. It is a nice exercise, but rather pointless.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

Besides - how hard is it for code to detect a 020 and optimize internally in those corner cases when it is worth it?

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@pavlor

ClassAct runs fine on 68000, just keep the settings simple. That said, I would love to see alternative CA/Reaction free alternative Prefs programs for OS3.9. I don't know how well documented the prefs files are, for my Minimig I made some presets on another machine and copied them over, throwing them in env:sys and envarc:sys from a little GUI I scripted together.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@cdimauro

Quote:

But requiring a 68020 only for a few things was a mistake.

They simply didn´t care back then, I think. Why they should?

@kolla

I hope it is only question of time until "68020 compatible enough" FPGA CPU core is there and our current topic will be forgotten.

Latest dragonball cpus has a gpu integrated. That would ne cool to have picasso96 driver for that.

@kolla

Quote:

kolla wrote: Consider that you have an A2000 with a 68060 card and something fails and part of debugging procedure is to remove the acc card - how much would it not suck if you couldn't even load kickstart and boot into shell and access your data using the 68000.

I understand this, but at the time (late 1992) a 68020 is the bare minimum for a new Amiga hardware. You cannot think about an Amiga 2000 that has to run the Kickstart 3.0: it's a 5 years older machine!

What would have been made is a different Kickstart compilation/optimization per processor: one for 68000, and another for 68020+.
Quote:

020 optimization brings marginal speed improvements in the OS components, just look at the micro optimization done by Cosmos. It is a nice exercise, but rather pointless.

I can assure you that a 68020 allowed A LOT of optimizations, which made sensible performance advantage.

The 80186 emulator which I was working on required at least a 68020 because of that. I remember that I used a HUGE amount of double memory indirect addressing modes, scaled index addressing modes, and bit fields, which GREATLY pumped the performences.

@kolla

Quote:

kolla wrote: Besides - how hard is it for code to detect a 020 and optimize internally in those corner cases when it is worth it?

For small optimizations it's trivial, and the o.s. already did it.

@pavlor

Quote:

pavlor wrote: @cdimauro Quote: But requiring a 68020 only for a few things was a mistake. They simply didn´t care back then, I think. Why they should?

I agree, but once you raise the limit... USE the new processor features!

@utri007

Quote:

Latest dragonball cpus has a gpu integrated. That would ne cool to have picasso96 driver for that.

640x480 16 bit is not bad, but it is probably only simple display controller without much hardware acceleration.

@pavlor

The Minimig will never have a 020, it is "stuck" with a 50MHz 68SEC000.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@cdimauro

So you are saying that OS3.1, that was specifically made to support _all_ Amiga models, should have dropped support for 68000? No 3.1 in A600, no 3.1 in A1000, no 3.1 in A2000 or A500 unless you have acc card.

The entire point of 3.1 was to bring _all_ systems up to same lever OS wise, so developers would not have to support ancient OS releases specifically.

Did you ever notice that there never ever was any 020+ code in a kickstart? Would you have noticed? No. So why come up with this now, 22 years later? It is utter nonsense.

Last edited by kolla on 20-Sep-2015 at 03:28 AM.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@kolla: wake-up and re-read my comments because you completely misunderstood what I've said, which I've also repeated several times (see my reply to olegil too).

And yes, there's 68020 specific code (a very small amount) which is transparently (this is the key point) executed by the Kickstart, depending on the processor on which it's running.

@cdimauro

Quote:

Quote: 020 optimization brings marginal speed improvements in the OS components, just look at the micro optimization done by Cosmos. It is a nice exercise, but rather pointless. I can assure you that a 68020 allowed A LOT of optimizations, which made sensible performance advantage. The 80186 emulator which I was working on required at least a 68020 because of that. I remember that I used a HUGE amount of double memory indirect addressing modes, scaled index addressing modes, and bit fields, which GREATLY pumped the performences.

As far as Kickstart 3.1 is concerned, I doubt that. It is relatively simplistic operating system with very thin layer between user applications and hardware. It was always issue with the hardware that could not get on par with CPU upgrades.

_________________
Amiga Developer
Amiga 500, Efika, Mac Mini and PowerBook

@itix: at least the mathieee* libraries would have greatly benefited of 68020-specific versions.

Anyway, it's something that Commodore should have done at its time. Now it doesn't make sense anymore.

@cdimauro

In theory they would, but at that time this kind of software was already optimized for real FPU.

Last edited by itix on 20-Sep-2015 at 03:02 PM.

_________________
Amiga Developer
Amiga 500, Efika, Mac Mini and PowerBook

@cdimauro

It's kind of slow to have to jump into a library to do FPU instruction.
It most likely makes more sense to have two versions of same binary, with FPU and without FPU.

_________________
http://lifeofliveforit.blogspot.no/
Facebook::LiveForIt Software for AmigaOS

@cdimauro

Quote:

cdimauro wrote: @itix: at least the mathieee* libraries would have greatly benefited of 68020-specific versions.

But they don't need to be in kickstart. There are third party math libraries, optimized for all the CPUs with and without FPUs, you can knock yourself out already.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@cdimauro

Quote:

cdimauro wrote: It's good for 68000 users, but I'm greatly disappointed to hear that, because it means that the Kickstart 3.x wasn't optimized for 68020+ processors. IMO at the time a 68020 should have been a minimum requirement, ...

I don't know if AOS3.x for A1200/A4000 has 68020 optimisation or not, but many software for AGA was never made for 68000, you will find some software that was never compiled for 68000 because the Authors thought a 68000 CPU with 8MHz RAM will never be enough for this software.

Therefore 68020 command set on AGA is a must.

But when it comes to performance the FPGA-68k is just some kind of 68000 on speed, a 68k variant with more commands per clock as the original 68k and with clock rates never available on a naked 68000.