Good stuff here.

I had a hell of a time getting the ACube link to work until I changed the .com to .biz.

Hopefully, all the links work now.

____________
Moxee

_________________
Moxee
AmigaOne X1000
AmigaOne XE G4
I'd agree with you, but then we'd both be wrong.

Naughty feature... Interesting to see what good use can be found for it.

_________________
- KimmoK
// For freedom, for honor, for AMIGA
//
// Thing that I should find more time for: CC64 - 64bit Community Computer?

great job.

always nice to find out new features has been added.

_________________
GDG Entertainment

Cybersphere Page

Soooo geek! Thanks ACube

_________________
ikir

Intriguing. In essence, an L2 cache you can "malloc"

so not affected by context switches etc. There HAS to be a way to use this for video decompression. But knowing how much can be gained would require knowing a lot about how many cache misses one has already, in essence cache line profiling. Not sure how many amiga devs do that on their code.

Edit: Does AOS4 have the ability to count cache misses with some sort of debug setting? Does the 460 core have counters for it? I know QorIQ does, at least some of the cores, but I'm not that intimate with AMCC anymore.

Last edited by olegil on 28-Feb-2013 at 02:15 PM.

_________________
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.

Nice!

I've been thinking about this a LOT the last 24 hours. Many modern CPUs also have the ability to lock cache lines, this could be used to emulate the same effect on processors with more cache but no onboard generic RAM.

So this resource should also be portable to e500mc (nudge, nudge) and ppc440, possibly also PA6T.

Allthough, when used with cache line locking rather than dedicated SRAM, it'll cause an overall slowdown of all the programs who aren't using it. If you really need that extra few percents of performance, I think it could be well worth it.

_________________
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.

guys Chris over at amigans.net has made a benchmark to test out this new feature.

If you've got a 460 check it out at amigans thread

_________________
____________________________
c64-2sids, A1000, A1200T-060@50(finally working!),A4000-CSMKIII
! My Master Miggies- Amiga 1000 & AmigaOne X1000 !
mancave-ramblings
X1000 I BELIEVE

Quote:

I've been thinking about this a LOT the last 24 hours. Many modern CPUs also have the ability to lock cache lines, this could be used to emulate the same effect on processors with more cache but no onboard generic RAM.

Well, that way you could both:
1)Reserve really fast memory for a single instance of a single program and
2)Reduce the amount of "generic" cache available for all other programs. This could, of course, include libraries used by the program reserving the cache.

And even if you use only a few kilobytes? Well, how many such programs does it take to consume all cache?

I'd be WAY more worried about 2 than excited about any potential speedup by 1...

Also, what if a program that has reserved gets stuck / crashes, and the cache is never freed? Of course depends on how the related API's would be written.

And can't help thinking that this "Letting one program run a bit faster by giving it faster memory" isn't THAT far from "Why would anyone want to run more than one program simultaneously anyway?" used by Microsoft in the old days

It's not like anyone would force you to use it.

It's used a LOT in the embedded industry.

Personally I wouldn't allow more than one program from using this at a time, and unlocking it (if the program has crashed) is pretty easy as there's just one of the resource anyway. It wouldn't be 64k if using locked cache lines, but it would be enough to fit some registers in VM or often-used state variables (or even a few macroblocks of a h.264 decoder). Really anything small which would normally fit in the cache but which you REALLY don't want to risk getting switched out by a context switch.

But as I mentioned, to know if this is useful it would help to have a cache miss counter.

And as for the downside of other programs going slower, do you really care about the speed of your mail fetching while you're using an emulator or watching a movie? Wouldn't 5% more CPU power for the main task be worth a slight slowdown on background processes? There's an AWFUL lot of ground between this and throwing out the preemptive multitasker.

Last edited by olegil on 03-Mar-2013 at 08:54 AM.
Last edited by olegil on 02-Mar-2013 at 04:18 PM.

_________________
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.

Nice feature for Sam460 :)
will add support for this extra mem in the AmiDARK Engine for critical data :)

Regards,
AmiDARK