VOGONS


PowerPC Dynamic Recompiler (patch)

Topic actions

Reply 100 of 137, by samo79

User metadata
Rank Newbie
Rank
Newbie

Mmm why do you want to force me to say exactly how much more speed I would like to achieve?
I don't know exactly how much space for optimization it will possible; that's why i asked jmarsh... but even if we could gain only 1 or 2 fps more it will be pretty good 😀
For sure i will not expect the impossible

Reply 101 of 137, by DosFreak

User metadata
Rank l33t++
Rank
l33t++

because "I want more speed" is too vague? More speed for what? If someone worked on it with no goal let's say they put 100hrs into it for 2fps....for what?

How To Ask Questions The Smart Way
Make your games work offline

Reply 102 of 137, by Dominus

User metadata
Rank DOSBox Moderator
Rank
DOSBox Moderator

It's nice to have more speed for PowerPCs but only theoretically. So it really depends on why you are using this old machine for gaming and why would 1-2 fps actually help? anything below 5fps is almost a matter of a certain system service running at the time or not and not really a good measurement 😀

Windows 3.1x guide for DOSBox
60 seconds guide to DOSBox
DOSBox SVN snapshot for macOS (10.4-11.x ppc/intel 32/64bit) notarized for gatekeeper

Reply 104 of 137, by samo79

User metadata
Rank Newbie
Rank
Newbie

@Jmarsh

Thanks for replying 😀
My testing hardware is a Sam440ep Flex 800 Mhz (Radeon 9250) running AmigaOS 4.1

More info about this board: https://en.m.wikipedia.org/wiki/Sam440ep

As a test I'm using the latest version ported to AmigaOS4 by kas1e (with your Dynamic core enabled)
As far as performance goes, according to PCPBENCH the performance's difference between core "normal" and core "dynamic" on my sytem are :

core=normal --- 0.7 (sometimes even 0.8
core=dynamic --- give me a stable 2.7 as a result

There is a big difference in term of performance using Dynamic instead of normal of course, and this is visible not only in that specific benchmark test but also in real games .. but unfortunately not quite enough to play many of that old titles on a slow hardware like mine 🙁

These are my current settings:

fullscreen = false
vsync = false
fullresolution = original
windowresolution = original
output = surface

Reply 105 of 137, by Dominus

User metadata
Rank DOSBox Moderator
Rank
DOSBox Moderator

So, the question remains, do you actually want to play on this slow machine (which no matter what won't be much faster) or just be able to play? Huge difference and IMO you are way better off with any newer system if you really WANT to play.
I *do* get it to be able to play. I'm still releasing snapshots of DOSBox SVN for PPC macOS but I definitely do not want to play on my old Mac Mini anymore.

Windows 3.1x guide for DOSBox
60 seconds guide to DOSBox
DOSBox SVN snapshot for macOS (10.4-11.x ppc/intel 32/64bit) notarized for gatekeeper

Reply 106 of 137, by samo79

User metadata
Rank Newbie
Rank
Newbie

Just playing some more complex game with an usable framerate, for what i can see even a slight improvement could allow the use of slightly more complex games (for example Mortal Kombat or Super Street Fighter 2 just to say a couple)

Reply 109 of 137, by jmarsh

User metadata
Rank Oldbie
Rank
Oldbie

I mean that setting does not even exist in regular DOSBox. Whatever build you're using has other patches/modifications. Since your speed is lower than expected it's probably due to those.
On a 729MHz CPU I get 4.2fps in PCPBench (the regular 640x400 test), up to 4.5 if I turn frameskipping on.

Reply 110 of 137, by samo79

User metadata
Rank Newbie
Rank
Newbie

Yes, indeed as i said that settings i've used in config was "experimental" ..

This is the port we actually using:

http://os4depot.net/?function=showfile&file=e … uter/dosbox.lha

Anf that are the default settings we have in our port for Amiga

fullscreen = false
vsync = false
fullresolution = original
windowresolution = original
output = texture
texture_renderer = compositing
capture_mouse = seamless middlerelease
sensitivity = 100
waitonerror = true
priority = higher,normal
mapperfile = mapper-0.74-3.map

In my cass, on my little machine in general i only changing the output=texture
-- to use "surface" instead ... this will help a bit on Sam440

About the CPU.. yes we have similar mhz on our PowerPC boards but probably that CPU cannot be exactly compared .. Sam440 uses a SoC processor

https://en.m.wikipedia.org/wiki/Sam440ep

Reply 111 of 137, by DosFreak

User metadata
Rank l33t++
Rank
l33t++

You'll need to test if the port your are using has regression, since you are also using dosbox-staging patches then you'll need to talk to them about those if those are and issue and you will also need to compare to the the official version of dosbox.
We'd prefer it if you keep dosbox-staging discussions to the dosbox-staging site since one of the maintainers of that projector is banned here.

Not that this is related to your issue but does the port you use still use a "0.74-3" keyboard file?
I guess this port is supposedly still 0.74-3 even with all of the changes? So not using the latest SVN code?

Last edited by DosFreak on 2020-10-30, 18:32. Edited 9 times in total.

How To Ask Questions The Smart Way
Make your games work offline

Reply 112 of 137, by kas1e

User metadata
Rank Newbie
Rank
Newbie

@jmarsh
My amigaos build is original dosbox + all your ppc patches + sdl2 patch (so sdl2 swith surely may have impact) + cdrom based part from dosbox-staging as it support audio from cd better than in original.

But, to add on that, all lowend custom ppc amiga hardware are crap if we compare it with even old ppc macs or wii, as it done by entusiast in low quanity, and mostly for labor of love than business. As such all the things related to cache's speeds, speed of memory tranfer most of time suck.

Add on top that it controlled by AmigaOS4: not the OS writen by genius, but just additional components and improvements at top of old, BCPL language based, AmigaOS3 from begining of 199x era. There even still some 68k code left which is internally emulated.

So sdl2 patch, custom slow ppc hw, amigaos4 : nothing surprised. And i do not know what samo expect at all with his hardware: stuff which works slow for him, will continue works slow. Be in hope that anyone will spend hours because "i want 1 fps" to have the same slow speed i see no sense.

My PPC machine from othet side are 2ghz ppc and comparable at least by speed with latest ppc macs, and dosbox works good with your jit.

Sure if there possibility to speed things up for cheap, that will be cool, but loosing a lot of time for +0.1 fps make no sense.

But for real, what is need it in terms of ppc support is 3dfx . That surely more important than improve benchmarks on 0.1 fps.

Reply 113 of 137, by krcroft

User metadata
Rank Oldbie
Rank
Oldbie

samo79,

The Pi3B (non-plus edition) at 1.35 GHz allows DOSBox to handle roughly 30,000 cycles, which produces a score of 17.6 FPS in PC Player Bench at 320x200) If we proportionally scale that score down to your 800 MHz system (800 / 1350), we get 10.4 FPS.

But you're only getting ~4 FPS, so what explains the > two-fold reduction? My linear calculation assumes your PPC 800 Mhz performs the same number of instructions-per-cycle as the Pi3, but given your PPC is likely many generations older, my guess is its reduced IPC efficiency likely fully explains this performance gap.

If your PPC's IPC turns out to be less than 1/3rd the Pi3, then the PPC dynrec implementation is even be more efficient than DOSBox's ARM dynrec 😀 (but I haven't dug up those numbers).

Reply 114 of 137, by jmarsh

User metadata
Rank Oldbie
Rank
Oldbie
krcroft wrote on 2020-10-31, 08:12:

But you're only getting ~4 FPS, so what explains the > two-fold reduction?

The default mode that PCPBench uses is 640x400, four times the size of 320x200. By that comparison the 800MHz PowerPC is kicking the 1.3GHz Pi's butt for IPC.

Reply 116 of 137, by Stretch

User metadata
Rank Member
Rank
Member

For Nintendo Wii, is this patch intended for the source code from https://github.com/dborth/dosbox-wii or from another repo? I get many errors of "(different line endings)" even though I converted the patch to Windows line endings with unix2dos.

Win 11 - Intel i7-1360p - 32 GB - Intel Iris Xe - Sound BlasterX G5