PSPSOne (Playstation Emulator for PSP) WIP Update
PacManFan posted this update concerning his Playstation Emulator for PSP Project.
<blockquote>Hey eveyone,
I want to let you all know the current status of PSPSOne. After I got my dyanmic recompiler(mostly) working, I realized that I just wasn't seein the speed increase I was hoping for. The actual CPU emulation was running 3-5X faster, but the screen updates were still very slow (2-4 fps).
Well, I took the plunge, and converted my project to use the latest version of the PSPSDK, and the HW accelerated graphic functions (ge functions). With this, there is good and bad news. The bad news is that I'll no longer be able to run my project under the PSPE emulator, the good news is now PSPOne has the ability to run much faster with ge functions,than with the previous SW-based rendering.
I found an open-source OpenGL graphics plugin for PCSX, and converted it to work with the PSP GL framework. The PSP GL functions take advantage of HW-based acceleration, instead of software based rendering that I was doing before.
I was amazed by the speed improvement immediatley. I am now getting 8-10fps, and that is running with the interpreted (not dynamically recompiled) version of the CPU core.
So, In short, PSPSOne is now running at 8-10fps using hardware accelerated functions through OpenGL.
Because the PSP OpenGL port is not 100% complete. My rendering has some issues. Most noticably, the textures are not in the correct format. I'm working today to manually change the code that specifies the texture pixel format.
I'm very excited by this huge jump in speed. This will be even faster once I work out some issues with my dynamic recompiler running on the PSP with OpenGL
-PMF</blockquote>
PSPSONE Advice (for PacMan Fan)
Hi,
What I meant was than PS1 is really easier to emulate.
To my understanding, it is just a "digital sample player" : or it plays the sounds from the CD,
or it plays audio sample directly to the DAC.
So it is FAR EASIER and less costly to emulate the sound of the PS1 than the Snes.
Now when it comes to the ps1 cpu... SURPRISE : A MIPS CPU TOO.
Which means that basically if you a program access the working ram, it just need a remapping of the code : no emulation to do.
Of course access to adress must "filtered" somehow by the emulator.
Basically a kind of hybrid between a cpu emulation and a JIT.(just in time compiler)
Theorytically, I would believe that the 33Mhz CPU of the PS1 could be emulated with only two or three time the CPU clock.
Which would basically mean as an example, that the ME chipset in the PSP could emulate the complete CPU of the PS1 to me (166 Mhz).
For the audio features of the ps1, except if you want to play mp3 instead of the cd track (well no cd drive on PSP right ? It should be pretty easy and not that costly to emulate.
Now for the graphic chipset of the ps1 -> VRAM structure may be causing some issue as the PSP and PS1 may have different format (=lot of work by cpu of the PSP to do conversion).
When it comes to rendering, the PS1 graphics system call could be translated to PSP graphic system call... I feel it would be rather simple. From my understanding the PS1 just has a display list that you transfer to the graphic chip which does the rendering for you.
In this case the emulator would just need to parse the renderlist and convert it to calls to the PSP render list... I bet with 5% or less CPU on the PSP all these calls could be translated to the PSP graphic chipset instruction format.
Until now the ps1 emu has been slow because they dont use ONE IMPORTANT FEATURE WHICH IS : THE CPU OF THE PS1 IS COMPATIBLE WITH THE PSP CPU.
In this case, they "emulate" completly the chip, which is very costly...
(33Mhz CPU cant be emulated on a 333Mhz cpu unlikely)
I heard also that the RENDERING was done in software, which is really a shame actually because translating the calls to PSP system call shouldnt be THAT difficult.
Of course, the reason for that is time and energy : using already existing C source of emu is easy, porting time is not that much... Writing a JIT and validate such technology would require many month man of work I believe.
Update : I read the site of PSPSOne, it seems that he implemented such technique but it didnt work as excepted (I tought register would be an issue but...)
I really believe that he should switch to hardware graphic acceleration. His first attempt gave him a 7x increase in cpu emulation performance. Even he can do may be x15 with a full JIT. I would have switched to the graphic emulation... Software is just TOO slow spending your time with for-loop to draw pixel.
