http://groups.google.com/groups?lr=&safe=off&ic=1&th=9b6e5d735187b950,5

A breakthrough?

Chris

I dunno if anyone else has noticed ths yet on the newsgroups but MD believe they have an XScale co-processor system running 32bit/26bit code on a 26bit RISC OS.

http://groups.google.com/groups?lr=&safe=off&ic=1&th=9b6e5d735187b950,5

A breakthrough?

Yes so we don't need 32 bit os just 32 bit apps.???

Chris

The fact that Omega can use the faster processors and run 26/32bit code is why I'm not that bothered about RISC OS 5 being pushed back; we need to have Omega there as a stepping stone between the two.

What if you attempt to change modes by writing to the low 2 bits of PC which switches you between USR and other modes (on 26 bit) but on xSCALE and ARM V5T processors (ARM9 and 10) could cause you to go into THUMB mode ?

Any part of the kernel that does this (which is 26 bit only) runs the risk of unceremoniously dumping you out into THUMB mode (on xScale) and then attempting to run ARM code as THUMB code (with obviously disasterous results).

I'd like to know how they get around this.

The clever Omega chip would think 'Hey, you don't want to go into thumb mode!' and switch the instruction with an xScale friendly one. Somehow.

I seem to remember RO Ltd. saying something about that a while back, cant remember what they said thou :/
wouldnt the mode change be performed on the 26bit SA though rather than the XScale??

If the code is given to the xScale to run, say a lines like

ORR PC,PC,#1 /sets bit 0 to '1'
ORR PC,PC,#2 /sets bit 1 to '1'
ORR PC,PC,#3 /sets both b1 and 0 to 1

What would xScale do ?

If you hit on the right one it would switch it to Thumb mode (I am not sure but I think its when you set bit 1 to 1 - anyone who knows better please enlighten us !). After all it would have no way of knowing that you did not intend to enter Thumb mode in a way that is legal to xScale and not just change modes in a way that's required by 26 bit code.

How could that be trapped ? Its a patently illegal instruction address (ARM PC can only address on WORD boundaries when in ARM mode or on half-word boundaries when in Thumb mode, where such a value WOULD be legal).

If the xScale has some sort of indicator of Thumb mode (has it a pin for this or some signal ?) then you could assume any digression into Thumb code was (probably) due to some 26 bit code attempting to fiddle with the two low bits of PC (a BX would be a possible second cause but that is not used in 26 bit code as (after all) the bulk of such code existed before and would be unaware of it or Thumb instructions), you could then recover from this by returning to ARM mode and setting the mode in a 32 bit friendly fashion.

It would be possible to trap for NZVC/IF flags (as they're at the top of PC in 26 bit mode) and you could trap any attempt to set or read them (by flagging a memory exception), then map the instruction to MSR or MRS (as appropriate). You could then continue with the instruction that followed the attempted (and now kludged) PSR set/read instruction.

[Edited by ams at 14:05, 4/6/2001]

[Edited by ams at 14:07, 4/6/2001]

I seem to remember RO Ltd. saying something about that a while back, cant remember what they said thou :/
wouldnt the mode change be performed on the 26bit SA though rather than the XScale??

If the code is given to the xScale to run, say a lines like

ORR PC,PC,#1 /sets bit 0 to '1'
ORR PC,PC,#2 /sets bit 1 to '1'
ORR PC,PC,#3 /sets both b1 and 0 to 1

What would xScale do ?

If you hit on the right one it would switch it to Thumb mode (I am not sure but I think its when you set bit 1 to 1 - anyone who knows better please enlighten us !). After all it would have no way of knowing that you did not intend to enter Thumb mode in a way that is legal to xScale and not just change modes in a way that's required by 26 bit code.

How could that be trapped ?

[snip]

It would be possible to trap for NZVC/IF flags (as they're at the top of PC in 26 bit mode) and you could trap any attempt to set or read them (by flagging a memory exception), then map the instruction to MSR or MRS (as appropriate). You could then continue with the instruction that followed the attempted (and now kludged) PSR set/read instruction.

HTH
JOhn

Fortunately, any code which isn't specified to be 32 bit will be running on the SA, only code which has explicitly declared itself to be 32 bit will be allowed onto the XSCale. If it;'s lying, then you'd expect it to go all wrong!

IIRC they can check how the code was compiled using a field in an AIF header when the AIF is loaded (so I guess they've intercepted whatever part of the OS loads and executes an app). I rather suspect that for any code it can't detemine the "bitness" of it'll run in 26bit mode.

Fortunately, any code which isn't specified to be 32 bit will be running on the SA, only code which has explicitly declared itself to be 32 bit will be allowed onto the XSCale. If it;'s lying, then you'd expect it to go all wrong!

IIRC they can check how the code was compiled using a field in an AIF header when the AIF is loaded (so I guess they've intercepted whatever part of the OS loads and executes an app). I rather suspect that for any code it can't detemine the "bitness" of it'll run in 26bit mode.

I presume then most of the kernel, ADFS and the like will be 26bit only and therefore ONLY run on the SA and not the xScale. In effect any disk I/O and most graphics will NOT be run to any effective benefit by the xScale ?

Mind you some computationally intensive tasks (which are unlikely to be kernel) could be made 32 bit to take advantage (MP3 springs to mind).

I presume then most of the kernel, ADFS and the like will be 26bit only and therefore ONLY run on the SA and not the xScale. In effect any disk I/O and most graphics will NOT be run to any effective benefit by the xScale ?

Mind you some computationally intensive tasks (which are unlikely to be kernel) could be made 32 bit to take advantage (MP3 springs to mind).

This is how I understand it - basically anything that jumps into RISC OS is going to have to run on the 26bit processor but I don't know if Omega can do something like run a 32bit app on an XScale and switch to the SA if a SWI is called. I guess only MD really know.

Chris (all IMO and all not strictly true)

I dunno if anyone else has noticed ths yet on the newsgroups but MD believe they have an XScale co-processor system running 32bit/26bit code on a 26bit RISC OS.

http://groups.google.com/groups?lr=&safe=off&ic=1&th=9b6e5d735187b950,5

A breakthrough?

Chris

Yes so we don't need 32 bit os just 32 bit apps.???

Looks like it, go to Microdigital's Newsroom to read an short explaination of their technology. Seems that 26bit applications will continue to work even when you have a 32bit only RISC OS. Omega is beginning to look even better now :-)
Hope RiscStation or Castle will be able to keep up so we'll have at least two players on the market in the future.

I'd like the see a SA Evolution that could take an XScale upgrade. I'm not overly bothered if it can do the processing switching tricks of Omega. When a 32bit OS appears I know I'm going to lose a lot of software, but then I only really use software that is still supported these days so it's less of a problem for me.

That's a good policy - wish more people were willing to move to software that's still being supported.

Has anyone seen the current poll on www.riscos.net where Acorn Browse and ANT Fresco still show up with some following and Webster and ArcWeb have at least one vote each (although it seems not many have bothered to vote). This is quite remarkable for the RISC OS market, a product gets used for years even if there hasn't been any support for it for a long time. I'm willing to bet there are numerous people still using Impression out there and all the while Ovation suffers with lack of sales.

With WebsterXL costing UKP 25 and Oregano UKP 49+VAT it's hardly a huge investment and I'm quite sure that both are somewhat more up to date than the other browsers and these two are the only ones still being developed.

That's a good policy - wish more people were willing to move to software that's still being supported.

Yes but OTOH I only use my RISC OS box for coding and EasiWriter (although I may look into some accounts stuff) so I don't have much in the way of legacy apps. My Windows laptop is currently used for all net access, and will probably stay this way even when my RiscStation turns up.

Has anyone seen the current poll on www.riscos.net where Acorn Browse and ANT Fresco still show up with some following and Webster and ArcWeb have at least one vote each (although it seems not many have bothered to vote). This is quite remarkable for the RISC OS market, a product gets used for years even if there hasn't been any support for it for a long time. I'm willing to bet there are numerous people still using Impression out there and all the while Ovation suffers with lack of sales.

I know, it's crazy, especially given that Ovation Pro will import Impression files now. Ok so it's extra investment and the time to learn a new package but surely it has to be worth it in the long run.

With WebsterXL costing UKP 25 and Oregano UKP 49+VAT it's hardly a huge investment and I'm quite sure that both are somewhat more up to date than the other browsers and these two are the only ones still being developed.

Well I can't really comment as I've never used either. I think the problem is that they're not free and people expect them to be. Personally I have no qualms about paying for any software that I want if it's worth the cash. If I only had a RISC OS box then I'd buy one for web browsing. Having recently been a full-time student I understand it can be difficult for them but I'm under the impression that some of the people who moan are actually on fair incomes.

Note I don't advocate buying software just to keep it alive - it has to be worth the cash but, given the RISC OS market, I am sometimes willing to pay a slight premium knowning that the software often cannot be subsidised by other income.

Doubt it because the MD stuff is custom hardware. AIUI what has been holding Evolution back are the VIDC and IOMD dependencies, although John Kortnik has shown that the VIDC dependencies aren't that difficult (but not trivial) to work around.

Hope RiscStation or Castle will be able to keep up so we'll have at least two players on the market in the future.

I doubt that we have to worry about RiscStation keeping up. Certainly from a conversation I (and one other ) had with Roy Heslop at Wakefield, it would seem that Evolution has actually been sitting on the shelf for a very, very long time waiting for RISC OS to catch up. Given that the board was actually on show at Wakefield two years ago...

Hope RiscStation or Castle will be able to keep up so we'll have at least two players on the market in the future.

I doubt that we have to worry about RiscStation keeping up. Certainly from a conversation I (and one other ) had with Roy Heslop at Wakefield, it would seem that Evolution has actually been sitting on the shelf for a very, very long time waiting for RISC OS to catch up. Given that the board was actually on show at Wakefield two years ago...

Then ROLs claim that "only one company has dedicated themselves to 32bit processors" just doesn't hold at all since Paul named Microdigital (if I remember correctly) in his news comment here on IconBar. That didn't ever hold anyway knowing that Millipede was developing Imago with XScale in mind!

Too bad Evolution will be gathering dust for at least another year! Maybe they'll get this magical software from Microdigital to run on Evolution?

I doubt that we have to worry about RiscStation keeping up. Certainly from a conversation I (and one other ) had with Roy Heslop at Wakefield, it would seem that Evolution has actually been sitting on the shelf for a very, very long time waiting for RISC OS to catch up. Given that the board was actually on show at Wakefield two years ago...

Bear in mind that the original Evolution spec was for a SA processor running on, what seemed to be, an enhanced CATs board, so 32bit wasn't necessary. When I spoke to Mark Gillard at Epsom 2000 he seemed to imply that Evolution had gone back to the drawing board during the wait to make it even better.

Then ROLs claim that "only one company has dedicated themselves to 32bit processors" just doesn't hold at all since Paul named Microdigital (if I remember correctly) in his news comment here on IconBar. That didn't ever hold anyway knowing that Millipede was developing Imago with XScale in mind!

Certainly Mark Gillard mentioned XScale in the same breath as Evolution. Before people get their hopes up I should point out we were having a somewhat less than serious conversation that started with me saying "So how does Evolution compare to Omega?"

Too bad Evolution will be gathering dust for at least another year! Maybe they'll get this magical software from Microdigital to run on Evolution?

Doubt it because the MD stuff is custom hardware. AIUI what has been holding Evolution back are the VIDC and IOMD dependencies, although John Kortnik has shown that the VIDC dependencies aren't that difficult (but not trivial) to work around.

I'd like the see a SA Evolution that could take an XScale upgrade. I'm not overly bothered if it can do the processing switching tricks of Omega. When a 32bit OS appears I know I'm going to lose a lot of software, but then I only really use software that is still supported these days so it's less of a problem for me.

But there needs to be a reason (beyond charity) for people to buy something. If you have !Publisher which does what you want, it's hard to justify buying more DTP Software (unless you have cash to throw around, of course). I don't.

And now a thinly disguised effort to keep on topic... seems to me that Evolution (aside from looking like a rather amusing new David Duchovny film out soon) and Omega (aside from being a very troubling greek letter used to torment engineering students) are both as "unlikely" as each other. Pah.

Gulli, you bet correctly.

But there needs to be a reason (beyond charity) for people to buy something. If you have !Publisher which does what you want, it's hard to justify buying more DTP Software (unless you have cash to throw around, of course). I don't.

You're absolutely right, if you don't need it don't buy it unless you've stacks of cash just lying around taking up valuable space. The trouble is, I think, too many people have been hanging onto their old outdated software and moaned about it lacking features other software has and is still being supported. Ovation Pro isn't the cheapest piece of software around but if you want a DTP app that's being updated - there's no other choice for RISC OS.
I really hope many people will make a switch from ArtWorks to Vantage when it's released because ArtWorks, despite good efforts from several people, is at full stop and will continue to be so.
As for still using Fresco, Browse, ArcWeb and Webster - all of these were good in their time but 25 pounds to move to WebsterXL is hardly going to break the bank. By moving you get a more up-to-date browser that will enhance your internet experience and at least let you see more of some websites. (Note I'm only using WebsterXL as an example, Oregano and 49+VAT could be replaced).

And now a thinly disguised effort to keep on topic... seems to me that Evolution (aside from looking like a rather amusing new David Duchovny film out soon) and Omega (aside from being a very troubling greek letter used to torment engineering students) are both as "unlikely" as each other. Pah.

Why do you think that? Omega seems to be getting closer although more than 6 months behind schedule already Somehow most of the info about Evolution has passed me by!

And now a thinly disguised effort to keep on topic... seems to me that Evolution (aside from looking like a rather amusing new David Duchovny film out soon) and Omega (aside from being a very troubling greek letter used to torment engineering students) are both as "unlikely" as each other. Pah.

Why do you think that? Omega seems to be getting closer although more than 6 months behind schedule already Somehow most of the info about Evolution has passed me by!

Most of the information I have about Evolution has been gleaned from conversations with people at RiscStation - Roy and Mark have been pretty forthcoming in the past about the status of things (barring NDAs) you only have to go to a show and ask.

Roy actually offered me a working Evolution board if I wanted one. As he said, it doesn't run RISC OS though but BSD is fine, and possibly WinCE as well.

Why do you think that? Omega seems to be getting closer although more than 6 months behind schedule already Somehow most of the info about Evolution has passed me by!

Most of the information I have about Evolution has been gleaned from conversations with people at RiscStation - Roy and Mark have been pretty forthcoming in the past about the status of things (barring NDAs) you only have to go to a show and ask.

Pretty expensive to buy a trip from Iceland to get that little piece of information - I'll wait for a news release

Pretty expensive to buy a trip from Iceland to get that little piece of information - I'll wait for a news release

Heh, yeah sorry, I often forget that I'm talking to people who don't live in the UK