Just the sort of thing we need, huh? I guess there would still be the dependencies on the rest of the hardware, but if ROL could just lisence dynamite...

An easy solution might be for simple 29bit-32bit translation and keep using ARM chips.

Ah well, will never happen, will it? I can but dream...

sheesh!

seriously, i'd rather like to have !draw on the office pox box.

[Edited by guy at 09:08, 14/6/2001]

The thing is the 26 to 32 bit step does not require ALL instructions to be translated/emulated (just ones that modify the PSR bits of R15 (PC)). In most instances perhaps 90% or more of the ARM code could be left untouched.

Dynamite might be a bit of overkill (and it would probably take a performance hit on ALL instructions not just the few that directly or indirectly modify the PSR bits in the PC).

Still if it were the only technology that would allow good old ROS to run on the latest 32 bit hardware it might be worth considering.

Dynamite might be a bit of overkill (and it would probably take a performance hit on ALL instructions not just the few that directly or indirectly modify the PSR bits in the PC).

No no, the thing about Dynamite is that it _accelerates_ the code. Translated code (eg. 29bit arm to 32bit arm) would run _quicker_ than, say, native 32bit arm code. So you would indeed do all instructions, and all would be good.

Sexy stuff, makes my mouth water. But will we get it?!

Dynamite can't make 26 bit code ANY faster on a 32 bit ARM because the code would already be running optimally (its only the PSR/PC stuff that's different and that WOULD take some sort of performance hit because the single instruction to set a flag AND return, say (ORRS PC,PC,#%1<<30) would need to be translated into (at least) 3 32 bit instructions. As most 32 bit hardware now doesn't go three times the clock rate of the 26 bit SA it follows that for THOSE instructions the Dynamite code MUST be no faster and would probably be slower).

As to running ARM code on a 1.4GHz Athlon there would be some non-trivial problems to be addressed, you would after all have to "emulate" a VIDC and IOMD in order to make a version of RISC OS run (and if you want 32 bit support you'd also have a further hit). Dynamite makes the code fast but a real Acorn doesn't need to emulate all its hardware before it can work !

I've seen Archie running on a fairly lowly Cyrix MX-2 (233MHz) and performing about 1/3 to 1/2 the speed of an original A-310 (4MHz). I suspect because Dynamite translates code this would give much better performance than that but I'd like to see the benchmarks first.

Dynamite can't make 26 bit code ANY faster on a 32 bit ARM because the code would already be running optimally

The thing is, it wouldn't be running optimally. Dynamite and other code morphers can speed up code by [doing what they do - I'm no programmer, check their website]. You could use dynamite to 'translate' 32bit to 32bit ARM and it would be faster. Do it 26bit to 32bit and you have the added benefit of allowing RISC OS and its apps to work with new ARM processors.

Their comment about speeding code up only applies to things like "collapsing loops" (basically optimising repeditive code). It (of course) has to analyse the code as it executes (a speed penalty) and then modify the code actually executed (which may result in increased speed on future runs).

This sort of scheme has been used before, by Digital Equipment Corporation. They developed code for the DEC Alpha that could emulate a Pentium. It, however, got faster the more the emulated code was used. How it worked was it analysed the instructions ACTUALLY used as they were executed (emulating them the first time through (slow) and then replacing the code with Alpha native code so that when the same code was tried on a subsequent run it would operate close to native Alpha speed).

That approach (if used by Dynamite) would result in emulated code which (over time) is gradually replaced by native code. It does impose a performance hit but a lot less extreme a one than full continuous emulation.

It is interesting and positive technology, it (however) holds one danger. It may become possible to emulate RISC OS on non-native hardware (say a high end Pentium or Athlon (say 1GHz or above)) and yield a performance similar to current 233MHz RISC PCs.

It's something they 'plan to... [demonstrate]'. Ie. it's a working product that they could actually liscence to someone, not merely a proposal.

http://www.eetimes.com/story/OEG20010611S0108

It's something they 'plan to... [demonstrate]'. Ie. it's a working product that they could actually liscence to someone, not merely a proposal.

"Transitive says it will show ..."

Hmm. I'll believe it when I see "Transitive have just demonstrated ..."

PlanetRiscOS have done a survey on what people are running RiscOS on:

vote - results
How do you use RISC OS?
With a RISC OS computer 45 %
By using an emulator on a PC 24 %
With both 25 %
I don't 4 %

Now I don't know how reliable those results are, and though it suggests that around 40% of installations are now on PC hardware, I expect this is somewhat skewed by their support for the emulators attracting such users.

Any ideas on how Red Squirrel and Archie might compare with Dynamite technology?

[Edited by guy at 11:43, 20/6/2001]

Now I don't know how reliable those results are, and though it suggests that around 40% of installations are now on PC hardware, I expect this is somewhat skewed by their support for the emulators attracting such users.

Any ideas on how Red Squirrel and Archie might compare with Dynamite technology?

Any ideas on how Red Squirrel and Archie might compare with Dynamite technology?

Since they are full emulators they differ in that they emulate all the extra hardware in the system - VIDC, IOMD, ...

The Dynamite technology just deals with the cpu, using some form of dynamic recompilation probably...

The emulators you mention deal with emulating the cpu by means of an interpretive system, which means you step through each instruction opcode and p[erform the relevant action each time. This is the way most emulators work. There are different techniques such as HLE (high level emulation) that have been most notably applied to N64 emulators - you sort of recreate functions (hence the reason UltraHLE performs so well on games created with the Nitendo libraries and less well on others). I don't believe either emulator uses this - I could be wrong of course. I know that work has been done to create a dynamic recompiling core that will work with RedSquirrel, which will improve speed significantly. I don't know if or when this will be added though.

Hmm. I'll believe it when I see "Transitive have just demonstrated ..."

My sentiments exactly.

Back to this Dynamite stuff, is everyone just assuming that it works? Someone commented to me the other day that it looked like a "proposal". I, of course, have no idea one way or the other...

hrmm & seeing as 1Ghz is growing towards "bottom end" we're going to have some speedy emulated SAs, which cost half the price

Sadly not quite true.

The speed of processors only scale when all the instructions are done in registers/on the chip. Anything with I/O (as most emulation will) will not scale exactly. A 1.4GHz Athlon will not emulate an Acorn 40% faster than a 1GHz one. It will all depend on the NUMBER of off chip calls (which more or less happen as fast for a 1GHz or 1.4GHz chip). Purely "in cache" code will run 40% faster. The real improvement will be less than the difference between the processor speeds.

In addition I said originally that code like dynamite might allow a 1GHz PC to emulate a 233MHz ARM (the only way to be certain is to go and do it (which I have not)). It may perform less well, it may only perform well with certain types of program..... as ever only believe the benchmarks you've run not the ones you have not.

Did you see the speed program with Jeremy Clarkson on BBC 1 on Sunday? That program where you need to neutralise blobs, do sums and match sequences of letters at the same time, in order to test your multitasking? When they did a close up of the keyboard you could see the front of a RISC PC at the top of the screen