@NutsAboutAmiga

Quote:

It's kind of slow to have to jump into a library to do FPU instruction.

Taken to its extreme, this could apply to ANY algorithm ever, though. example: printf. I've started optimizing away many of my printf calls on AVR, replacing them with direct calls to my underlying print_number function which takes 1 uint32_t, a base, padding char for left side padding, two numbers to signify how many digits total and how many of them are decimals. A highly versatile function, with support for any sort of base I want (2,10,16 being the only sensible ones I can see right now), plus fixed-precision integers, which I use ALL the time, eg. when printing measurement values in %01.3dV and %01.2dC and even my oft-misspelled %01.3ddB (that second d in there is usually not present on any RC1 (oh and it helps that my version of puts doesn't output the newline, obviously)

but back to the topic at hand: It made sense to have libraries for floats because then the FPU instructions are available even if they shouldn't be. I forget, though. Was there any technical reason the "line 1111" emulator traps weren't used instead? I assume a trap is a bit slower than a library, but how much?

Last edited by olegil on 21-Sep-2015 at 08:18 AM.
Last edited by olegil on 21-Sep-2015 at 08:17 AM.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@OneTimer1

Actually, an FPGA-68k can implement whatever instructions it wants. The issue is the ACTUAL 68k soldered onto minimigs. This cannot be changed, it can just be clocked much higher than in 1985.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

As I followed Firebee it seems that is uses small 68kemu.prg for
EmuTOS to be able to run anything even Falcon compatible.It turns
out it's closer to NG amiga systems as only GEM written sw works
100 %.Okay , I didn't follow much after that.

With 040s cheap as dirt and plentiful to get by socket could
make sense.

BTW I think that it should be 300 Mhz Coldfire not 30 ???
What would be the point ?

Last edited by blizz1220 on 21-Sep-2015 at 06:46 PM.
Last edited by blizz1220 on 21-Sep-2015 at 01:22 PM.

@OneTimer1

Quote:

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.

The 68SEC000 in the Minimig is not on FPGA, it is a real physical CPU, and it is fast enough too. The 68000 has developed in newer and improved version all the time, just rarely used in amiga products.

Anyways - having the essential OS3.9 components rebuilt as 68000 compatible should not be a big issue. I am puzzled by how workbench.library became 020+ on rev 45.130 (I use 45.129 on my Minimig), without any word in the docs/changelog about why.

Last edited by kolla on 21-Sep-2015 at 03:44 PM.
Last edited by kolla on 21-Sep-2015 at 03:44 PM.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@olegil

Even mixing FPU and CPU instructions can be slow, the problem is that FPU registers and CPU registers are not the same, so whatever you do has to passed into RAM and back.

Something to think about when you write C code, only int's or only doubles.
yes doubles, PowerPC does not really have float, float's they are just truncated doubles.

Quote:

but back to the topic at hand: It made sense to have libraries for floats because then the FPU instructions are available even if they shouldn't be.

As far as I know there no TAN or SIN, COS instruction in Assembler, math will always need some kind of library, but anyway, most Amiga computer did not even have a FPU, so it made sense.

Quote:

I forget, though. Was there any technical reason the "line 1111" emulator traps weren't used instead? I assume a trap is a bit slower than a library, but how much?

A lot, that’s way OxyPatcher was made to prevent, illegal instruction hits.

I tried my own MMU hack in Basilisk II, it was interesting, to convert BGRA to ARGB, on the fly like that, but it was horribly slow, if every pixel on the screen was redrawn.
Yes I know not the same thing MMU hit and instruction hit, Anyway its interrupt based emulation.

Last edited by NutsAboutAmiga on 21-Sep-2015 at 06:17 PM.

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

@kolla

Quote:

The 68SEC000 in the Minimig is not on FPGA, it is a real physical CPU, and it is fast enough too. The 68000 has developed in newer and improved version all the time, just rarely used in amiga products.

AFAIK the 68SEC000 behaves more like an improved 68010 ( Supervisor / state register, no 16 bit misalign errors and maybe a longer prefetch queue) this might cause some minor troubles in very early games, games that would have problem handling expansions over the original A500.


Quote:

I am puzzled by how workbench.library became 020+ on rev 45.130 (I use 45.129 on my Minimig), without any word in the docs/changelog about why.

I don't know from where you got if, Kickstarts where always compiled for a specific Amiga, a A1200 Kickstart (Workbench ?) might cause troubles on a 68000, and hacks like AfA where only compiled for 68020 or higher ... I told the author about 68000 support but wont listen.

Last edited by OneTimer1 on 21-Sep-2015 at 09:18 PM.

@OneTimer1

Quote:

AFAIK the 68SEC000 behaves more like an improved 68010 ( Supervisor / state register, no 16 bit misalign errors and maybe a longer prefetch queue) this might cause some minor troubles in very early games, games that would have problem handling expansions over the original A500.

And this is relevant in this discussion how?

Quote:

I don't know from where you got if, Kickstarts where always compiled for a specific Amiga, a A1200 Kickstart (Workbench ?) might cause troubles on a 68000, and hacks like AfA where only compiled for 68020 or higher ... I told the author about 68000 support but wont listen.

Of course they were compiled for specific Amiga, but bits and pieces were rarely optimized for the different CPUs. For example, A1200 and A600 share a lot of hardware and hence use a lot of the same content in the kickstart, like scsi.device for the IDE controller, card.resource and carddisk.resource. Most parts are shared in all 3.1 kickstarts, only a few bits are uniq and yes some even optimized a little (utility and exec I think), but in general no.

AfA is irrelevant, it is not part of the OS, workbench.library (suddenly in 45.130) and resource.library are. I recall also some rather simple command in C: that suddenly was 020+ in OS3.9 for no obvious reason. Anyways, AfA is open source, you can compile it yourself for 68000 if you need it.

Last edited by kolla on 22-Sep-2015 at 12:11 PM.
Last edited by kolla on 22-Sep-2015 at 12:10 PM.

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@NutsAboutAmiga

What's wrong with fsin, fcos and ftan? They are software emulated by way of traps on 68040 and 060, but they exist in hardware on 68881 and 68882.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@kolla

Quote:

kolla wrote: AfA is irrelevant, it is not part of the OS, workbench.library (suddenly in 45.130) and resource.library are. I recall also some rather simple command in C: that suddenly was 020+ in OS3.9 for no obvious reason. Anyways, AfA is open source, you can compile it yourself for 68000 if you need it.

But OS 3.5 and 3.9 have the 020 listed in the minimum requirements, so it shouldn't be a big deal that certain components from them are compiled for the 020. It probably doesn't really need it, but I guess whatever optimisations are there for an 020+ might give it a slight performance boost over a plain 68000 version.

_________________
RobTheNerd.com | InstallerGen | SMBMounter | Atoms-X

@Daedalus

for no obvious reason usually rules out performance optimisations

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@Daedalus

Did I notice any speed gain between workbench.library 45.129 and 45.130? Nope.

When I have my Minimig again, I will find those 020+ binaries, and list them up. I suppose there are no disassembley tool around that can simply scan through a binary and determine if it is 68000 clean? :)

Would anyone be interested in "back patching" those 020+ bits of OS3.9 to 68000? Sort of the oppose route of that Cosmos is doing, haha!

_________________
B5D6A1D019D5D45BCC56F4782AC220D8B3E2A6CC

@thread

Just repeating these links in case it might answer to anything concerning plans or speak to the issue of who would communicate about what:

ACube and A-EON announce One Vision on future Amiga co-operation

and

A-EON Technology and ACube Systems strengthen strategic alliance

#6

_________________
This posting, in its entirety, represents solely the perspective of the author.
*Secrecy has served us so well*

@number6

"Large company and small company, both competing over the same shrinking market, strike a 'strategic alliance' deal for future cooperation"

Uh-oh. No wonder we haven't heard anything in two years. Thank heaven minimig is open source.

@CodeSmith

which is that "larger" company?

@number6

Yes. And?

@OlafS25

Well, it COULD (to someone who knows neither ACube nor A-Eon) be seen as a way to stop competing and start screwing us instead. But I doubt that. I think it's a good thing, actually.

Allthough, it sure would have been nice if ACube had made something new now and then

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

I remember someone asking one of the Frieden brothers to compile a new version of Warp3D for 68k. Answer was "I could but I do not. Buy 4.X". I think that propably shows the attitude of Hyperion regarding 68k. For them 68k community is a resource of potential 4.X buyers, not a target group to make money. That is different to a degree for A-eon and Amigakit but the OS is owned and controlled by Hyperion. So in my view no chance.

@wawa

Hm, you're right. Make that "two small companies..." and it would be more correct. Point in case is, there isn't enough profit in competing over minimigs when you can be cooperating and selling PPC boards for a lot more profit - AEon covers the high end market, ACube the low end market and that's everything tidily sewn up. It's also a lot easier to get people on the upgrade treadmill when your users aren't running software designed for < 66MHz CPUs. Why do you think Hyperion and their partners are so keen on getting everyone running PPC, even buying out other companies to make sure their PPC software stays relevant? They don't want 68K, it's bad for business.

Last edited by CodeSmith on 25-Sep-2015 at 04:55 AM.

@CodeSmith

Unfortunately hobbyists like majsta can do an accelerator using 68k technology for his A600 and get software to run on it, but if someone were to do something similar with a PPC, who knows how you would get any version of Amiga OS on it.

Given that we have hobbyists (and small businesses like Jens) who make hardware, Hyperion could at least have a plan for how to support more hardware.

At a minimum I would have hoped for a commitment to support any Amiga PPC accelerator designed that uses a core which is already supported by the OS, enough memory to be useful, and which has UBoot running. plus possibly additional rules about not being a dick when you draw your schematics, noone likes hardware that exploits loop holes in circuit theory. Example: writing by addressing (address bits wired as data, using chip selects to address). Yes, I know people who do things like this.*

Only official statement we've had from Hyperion is still the old "give us ALL your money and we'll consider selling CDs to your customers" (exaggeration to absurdity for effect).

I know it would take some effort to support more platforms, but given strict enough rules it should be within the realm of the possible.

If it's possible to take an FPGA, some RAM, some flash and some interface glue logic and make an accelerator for an A600 then it's PERFECTLY plausible that you could do the same with a PPC/RAM/flash/glue logic. But only one of these gets done, because only one of these has an OS.

Edit:
*: This isn't a very good (or bad, as it was) example of being a dick, but it is annoying to support things like this just because some guy wanted to cut corners. Abusing logic chips for analog effects is much worse, see http://www.macaronimusic.com/handmade-circuits/ and http://electro-music.com/forum/forum-160.html for strange examples. Allthough those are actually USEFUL (for some weird value of useful) examples, so not very bad

Last edited by olegil on 25-Sep-2015 at 06:59 AM.

_________________
This weeks pet peeve:
Using "voltage" instead of "potential", which leads to inventing new words like "amperage" instead of "current" (I, measured in A) or possible "charge" (amperehours, Ah or Coulomb, C). Sometimes I don't even know what people mean.

@olegil

Steve Wozniak was famous for doing that sort of thing. The engineer in me recoils in horror at the thought of the poor SOB who had to maintain Woz's "clever" designs once he moved on to a different project.

On the main topic, I'm quite convinced that the future of the amiga is in virtualized 68K in FPGAs by similar projects to this and AROS or some other OSS for the OS.

Last edited by CodeSmith on 25-Sep-2015 at 07:43 AM.