Port of linux version to Android
#11
ARM recompilers means rewriting the whole PCSX2 core that took a decade to come up with.
Reply

Sponsored links

#12
Yet if you try your hand on arm its not as if the pcsx2 core will be lost (it would still be fully operational on x86).

Also you don't need the level of fidelity you've managed on x86, at first (in 3 years time) only few games will play (the lowest of the low), in 10 years time you'd have everything play (by then all phones , even the low ends, would have the power for ARMSX2 anyway)... so there's no rush from this point of view, but one has to start thinking about arm, by now already, don't you think?



Edit: And on rather relevant news, TSMC already taped out -6 months before schedule- the first A57 based socs: http://www.tomshardware.com/news/ARM-TSM...21804.html ...

That means that by mid 2015 we are going to have the equivalent performance of a quad Intel @ 2.66ghz on phones. Things are actually moving even faster than how I implied in my initial post... That means that by 2015 already (and not 2016) arm would have the hardware to play a handful of pc2 games already, I cannot even imagine what 2016-2017 will bring (3-4 years from now).

As you can see, though, even as we reach Intel's performance there is no sign of slowing down, it would be veeery hard for Intel to keep x86 relevant ... Interesting years are coming ahead, and I recommend you to be prepared so that much of your work may not be discredited (as hard and as important it may already be)...
Reply
#13
First, it isn't a soc but only a cpu that a very big difference, actually it is only a test chip. Second, they only did the taped out not any serious production. ARM core are good candidate to test silicon behavior, see how you need to tune the process and what you can get. They didn't even have any real chip on 20nm mass production, don't expect any 16nm chip in less than 3 years. Besides FINFET technology is very complex, it might take more times than expected.


Take 2 runner, one a lot ahead of the other. The first guy hit a wall so it need to climb it, it the mean time the other guy is still runing fast. With some math you can prove that the latter will come back and even will be well ahead of the former king. Sadly it is the real world, the guy will hit the wall too and it will need to climb it too. A15 was designed with a higher frequency in mind that A9, that the main reason why A15 got a bigger pipeline. As of today, A15 leaks so much that it can't even reach A9 frequency. Intel did the same with P4, bigger pipeline => bigger frequency => ouch too much power => lower frequency => smaller pipeline.

One difficulty to optimize on ARM cpu is the difference between all ARM cpu. To reduce power consumption, some SoC vendor reduce the float performance or the SIMD engine performance, they are useless on reallife but we need them for VU emulation. Or maybe we could try to use GPU instead. But GPU side it is worst, they got totally different engine with probably a lot of differents driver bug. Even if you can get a PCSX2 port on ARM, it will actually run only a single SoC vendor. 70% of the user will complain Wink
Reply
#14
I never -not for one second- have I implied that we'll have 16nm by 2015, but we *will* have A57s. A15 was taped out in October 2011 and went in full production in November 2012 (Chromebooks, Nexus 10) and in a few weeks we're going to have the first quads in mass release. I do not have a reason to believe that (with) A57 will be different, I'll be very surprised if by the first half of 2015 A57's won't be amongst us with 16nm being released in the following year (2016). We'll be here we'd both see how that will go... same with the fabled wall (and you're right I may be wrong 'bout this, but -again- we'd see).

As for optimizations, you're right, the plethora of ARM offerings, even though it is an advantage to the pace that industry is moving also creates fragmentation. Obviously you'd have to develop for only one vendor over another at first, no different how you did with Intel on the desktop arena. I don't see how that can be a problem though, you managed well to the desktop, I don't see why people should be complaining on ARM, they ought to know for why's that. But even if you capture just 30% of the market it's gonna be so massive, so much more massive than desktops, that it would be enough to put in the shadow your current market reach...

BTW you probably know but the Dolphin guys already started their port job... Not that it is much relevant as it is a completely different project with different challenges, but it *does* show where the market is moving to ... *for better or worse*.
Reply
#15
Ok sorry. What I mean is an ARM test chip on 16nm vs a full SoC on 22nm have nothings in common. Even if the later seem to include the former. That said you probably right that we will see A57 sooner. My guess is that phone maker would compensate the lake of frequency by an increase of RAM => which imply a 64bits instruction sets.

ARM are totally different than X86. For example Tegra2 didn't include any SIMD instruction (2 years old I know, probably better know). Some floats instructions are emulated by the CPU so they are 10x slower. Mobile have heavy constrain on power so you need to do some big trade-off. AMD and Intel are close, besides the latter own 80% of the market. Let's wait a couple years to see where the market settle down. I agree with you market is moving toward pad/smartphone but gamers will probably stay longer on desktop Wink Nobody know which ISA will win, maybe both will stay alive (worst case for dev)! If android apps were C, ARM would have win by now but JAVA change a lot the equation.
Reply
#16
No, android is c+java. And I am going to port the pcsx2 to android.
Reply
#17
Interesting link on new tech limitation

http://www.semiwiki.com/forum/content/22..._234194577
Reply
#18
I think the crux of the matter here is that the Linux port isn't well polished at the moment and has major issues graphics acceleration wise (in that it's currently non-existent in a working form). So porting that wouldn't help.

Tie to that the fact that Arm is very much a moving target right now, sure, it's coming along in leaps and bounds but it's changing shape each time hence why Gregory stated everything would need to be trashed and you'd have to start over each time it went through another incarnation.

(05-15-2013, 12:59 AM)ning Wrote: No, android is c+java. And I am going to port the pcsx2 to android.

[Image: GoodLuckWithThat.jpg]
Reply




Users browsing this thread: 1 Guest(s)