Dolphin, the GameCube and Wii emulator - Forums

Quote:Someone asked about running Dolphin on a PPC-based Mac because it used PPC, and then someone rejected the idea saying that it wasn't the same PPC, just the same micro-architecture series.

That's the exact opposite of the truth.

It's the same ISA (or rather a similar ISA) but a different microarchitecture (although the microarchitecture is heavily based on the G3 series microarchitectures).

It poses the same issue as above though. A lot of dolphin would need to be completely rewritten, debugged, and optimized. And the hardware used by PPC macs is so old that it would probably still be painfully slow.

What is the clock*ipc (my way of saying performance) of the PS3's main processor compared to broadway? Would it do well running a single thread? The Wii's CPU is garbage so maybe.

(10-25-2012, 07:08 AM)Axxer Wrote: [ -> ]What delroth just said is that not every instruction translated will be longer either. Some will be shorter. Some will be the same. Some will be longer. Just like a language.

*Woosh!* You hear that? That's the sound of something totally going over my head. Well, last time I try to read in a hurry like that (that's just an excuse, I'm probably really just clueless today).

I thought "That's actually a good analogy" was in reference to Axxer's original one, completely missed the fact that he was talking about this new one. My bad.

Quote:What is the clock*ipc (my way of saying performance) of the PS3's main processor compared to broadway? Would it do well running a single thread? The Wii's CPU is garbage so maybe.

We don't really know for sure. They're video game consoles with custom chips so it's not like there is a whole lot of benchmarking data available for them. It also depends a lot on the workload. IPC is completely different based on the architecture AND the workload. Historically PPC cpus have always done very well with floating point and vector arithmetic workloads, gekko and the PPE are no different in that regard.

There is a big clock rate difference between them
Gekko: 486 MHz
Broadway: 729 MHz
PPE: 3.2GHz
So about 7 times as high as gekko and 4 times as high as broadway.

As far as architecture goes:
Gekko/Broadway:
Based on the PowerPC 750CXe core
Cache: 32KB 8 way L1 Data cache, 32KB 8 way L1 instruction cache, and 256KB 2 way on-die L2 cache
Bus: 64 bit wide FSB, gekko runs it at 162MHz (1.3GB/s), broadway runs it at 243 MHz (1.94 GB/s)
Pipeline: out-of order superscalar, dual issue, six execution units (floating-point unit, branching unit, system regis
ter unit, load/store unit, two integer units), 32 bit integer units (one simple one complex), 64 bit floating point unit, supports 64 bit floating point vectors (two 32 bit floats in a 64 bit vector), supports static and dynamic branch prediction, and so on....
Lots more info. can be found online.

PPE:
Cache: The same except the L2 is 512KB and the L2 cache line size is 128 bytes (I think it's 64 bytes in the gekko/broadway)
Bus: Not even going to talk about this, since it has an IMC (integrated memory controller) so it's mostly irrelevant
Pipeline: in-order superscalar, dual issue, six execution units (integer, LD/ST, branch, floating point, two vector units), supports SMT with two threads (similar to Intels HT), 64 bit integer unit, 64 bit floating point unit, 128 bit vector units which do not support double precision floating point vectors, static branch prediction, once again a lot more information can be found online

I would expect them to have similar IPC based on the specs in most workloads but I don't really have any evidence to back that up. The PPE was specifically designed to trade away some IPC for a smaller and lower power core, thus allowing them to cram more SPEs onto the chip. A G4 and K8 should both kick its ass in IPC. I would expect it to a G3 or P6 in IPC. The IPC on vector arithmetic workloads though is definitely going to lean in favor of the PPE, big time.

There are synthetic benchmarks out there but they are pretty useless when trying to assess the IPC of normal applications. Synthetic tests are generally very lenient about IPC and generally favor higher clock rates, wider pipelines, and more cores. Yet even so the IPC in synthetic tests was pretty terrible, it got its ass kicked by pretty much every major desktop CPU from 2001/2002 and up.

For the record
Gekko ISA (broadway also has the same ISA): PowerPC 1.10
PPE ISA: Power 2.03

At one point in time PowerPC and Power where two separate but extremely similar ISAs developer by IBM. IBM eventually merged them and now it's just the Power ISA.

Also a few clarifications about earlier statements. A lot of these cpus use extremely similar ISAs but not totally identical. PowerPC can be used to refer to the PPC ISA or to a series of microarchitectures that implemented that ISA which were given that same name. Upon further inspection both the microarchitecture and ISA of gekko/broadway are either identical or nearly identical to the PowerPC G3 (PowerPC 750) used in macintosh computers.

I would think that, given the similarities if the ISAs and the (almost assuredly) more powerful CPU in the PS3, it would be theoretically possible to run the CPU portion of the Wii extremely well on the PS3. Idk about audio and graphics (what's going on with the DSP?) but those could be implemented and work well.

It is apparently possible to get a lot of games running full speed (I would assume, though all of this is an educated guess). Making the thing though... would be difficult. A lot of stuff would have to be rewritten, and I don't think developers are that interested in it. Porting to Xbox 360 would be easier since, as I understand it, DirectX on the 360 is pretty darn similar to its PC counterpart, and it runs a PPC too (and hey, it's triple core like Dolphin). I don't think it has been hacked yet though.

I'm highly skeptical. As we've seen with xbox emulators just because your ISA is the same or extremely similar doesn't mean emulation becomes lightweight. You're still going to need many times the amount of processing power that the original hardware had IF it's extremely well optimized.

I'm not an expert by any stretch of the imagination but from what I've seen the SPEs would be well suited to designing a fast software based video backend (which would be extremely accurate, potentially more accurate than what we have now) IF you could get someone to properly optimize it. But let's face it, the number of competent programmers experienced with SPE assembly or SPE C/C++ programming is pretty small. And debugging would be an absolute nightmare.

DSP emulation would be lightning fast with an HLE implementation. But I'm not so sure about LLE.

The xbox360 would indeed likely be a better choice. But the environment necessary is simply not there as you said.

I guess the PS3 development will increase anyway now the lvl0 key is given. Do you think it might be usefull when I crosspost this thread to some PS3 forums (homebrew, jailbreak, etc.) If there is enough interest from that side, are there some people here willing to help them out? Probably not into depth, but some directional and non low-level architecture based problems. Or maybe more

I wish I could help, but my knowledge is very very limited.


edit: here are the lvl0 keys eplained: http://psx-scene.com/forums/content/lv0-keys-leak-explanied-scene-developers-2669/

maybe this will be interesting:

Quote:Marcan, who is a valuable asset of the Wii and PS3 hacking scene has....

Are you still on that "Dolphin on PS3" topic? If so, it's not going to happen and the suggestion is about as realistic as the "Hey guys, what about Wii U emulation on Dolphin??" threads. I.e. I advise you to spend your time on something else

It's interesting doesn't mean it will happen. I really don't care whether or not it will happen or not since I don't have a PS3. It just is an interesting topic.

Quote:Do you think it might be usefull when I crosspost this thread to some PS3 forums (homebrew, jailbreak, etc.)

No. I'll just leave it at that.