DOSBox port questions...

[eXeler0]

@NovaCoder, (or MattHey)
In your opinion, would it have been possible to optimize it enough purely software-wise using asm to make it useful on a 060? Or is the only viable solution a faster CPU.. Does the prospect of 060+ performance of the Vampire revive its usefulness somehow?
Anyone tried it? (I know there are some beta testers of the Vampire 2b board around here ;-)

[matthey]

Originally Posted by eXeler0

@NovaCoder, (or MattHey)
In your opinion, would it have been possible to optimize it enough purely software-wise using asm to make it useful on a 060?

Yes, I believe so. A working dynrec should give 386 performance. Wolfenstein and Tomb Raider would probably go from a slide show on my 68060@75MHz to playable. There are many older games that don't need much processing power. I believe NovaCoder was wanting DOSBox to replace ScummVM on high end Amigas. I could optimize common C functions in assembler especially the functions emulating x86 instructions. Some of the DOSBox functions expect function arguments to be passed in registers (fastest way) but GCC doesn't have a global way of specifying what is necessary. I could fix this with enough time and it would be cleaner and faster than the modifications to the code which I made to get it to work. NovaCoder could probably have done some more optimizing at the high level too. A different version of GCC may generate better code. We were using 3.4.x for convenience. The code is in C++ and quite large so there aren't many other Amiga compilers which would work. We really needed an x86 DOS expert or one of the DOSBox programmers to help us get the dynrec going before it would have been worthwhile to optimize more.

Originally Posted by eXeler0

Or is the only viable solution a faster CPU.. Does the prospect of 060+ performance of the Vampire revive its usefulness somehow?
Anyone tried it? (I know there are some beta testers of the Vampire 2b board around here ;-)

Although the new Vampire's Apollo core in FPGA is not much faster than a fast 68060, it does have less bottlenecks. The caches are much bigger which should help with the large DOSBox executable. The Apollo core has the 64 bit integer MUL and DIV instructions in hardware making them faster than the 68060. DOSBox would be a good test of performance but the Apollo core needs to be mostly bug free to execute such a large program.

[Retro1234]

Is dosbax faster than PC-Task or PCX as there meant to have jit.

Hers a video of PC-Task ocs on Vampire.
[ Show youtube player ]

[eXeler0]

@BooBoo, Im guessing "yes". Could someone try to run a game a bit more complex than Solitaire and compare? ;-)

[matthey]

Originally Posted by Boo Boo

Is dosbax faster than PC-Task or PCX as there meant to have jit.

I believe PC-Task and PCX are considerably faster than DOSBox without dynrec. DOSBox is a big complex program with lots of options written in C++ for powerful computers.

Would a dosbox port compiled to run on power up boards be more performant and not too difficult to build? This would be a fun utility for me 😊

[AMike]

Originally Posted by esc

Would a dosbox port compiled to run on power up boards be more performant and not too difficult to build? This would be a fun utility for me 😊

That would be great - my CyberstormPPC is hungry. The sad truth is, that for the majority of the remained developers the PPC is not "real Amiga".

If it's possible to cross-compile in macosx or linux some ppc/wos binaries I can give it a shot. Could be fun But I'd rather not build a dev environment in UAE - is this possible? Googling has been a bit dubious so far.

[matthey]

Originally Posted by esc

Would a dosbox port compiled to run on power up boards be more performance and not too difficult to build? This would be a fun utility for me 😊

DOSBox performance should be better on high end and overclocked P5 PPC boards. I believe NovaCoder started with the PPC version of DOSBox not that it matters too much for a generic compile without dynrec.

http://os4depot.net/share/emulation/computer/dosbox.lha

Vbcc is an easy and modern way to compile for a WOS target but it doesn't support C++. Maybe Storm C v4 which uses GCC could be used?

@matthey - Yeah, was looking into Storm C v4. Unfortunately it's commercial software and I'm in Iraq right now, which means there's not much chance I can get a CD shipped to me. Especially if I want it quickly.