VOGONS


First post, by pupnik

User metadata
Rank Newbie
Rank
Newbie

Title says it all: I'm trying to get solid playability for 286-class games on the Nokia n810 internet tablet. CPU is 400mhz TI OMAP2420.

I generally can get around 1600-1800 cycles without sound blaster emulation, but only 700-1000 cycles with soundblaster (1) before music starts breaking-up (regardless of buffer sizes). Hz of all sound is 8000.

Initial oprofile results show that core cpu and memory writes and reads are big performance eaters. For e.g. top 3 in the Wing Commander INtro are:
144808 31.9373 dosbox CPU_Core_Simple_Run()
43815 9.6634 dosbox mem_writew(unsigned int, unsigned short)
39148 8.6341 dosbox OPL2:: YM3812UpdateOne(int, short*, int

I've been perusing n0p's tweaks and he apparently did not include the verysimple cpu source on his page (BOOOO). If anyone could help me get my hands on this i'd be very grateful.

Reply 1 of 7, by wd

User metadata
Rank DOSBox Author
Rank
DOSBox Author

Suppose that "verysimple" core is pretty much what the simple core in the
later sources is, that is a stripped-down normal core where more direct
access to memory is used to fetch code (not using the more complex
paging-supporting ones).

Reply 2 of 7, by pupnik

User metadata
Rank Newbie
Rank
Newbie

According to n0p's page it's a 286-only core.

The performance i'm getting is just sad. Betrayal at krondor with sound blaster 1 emulation needs to drop down to 500 cycles. Without music it can run ~ 1500 cycles.

Prince of persia with sb1 can run 900 cycles.

I'm using moe's fmopl version. I dropped all the internal soundblaster frequencies and dosbox.conf frequencies to 11025, which has helped a bit. The dream would be to get the sound blaster / adlib running entirely on the nokia DSP.

I released a new version for the Nokia N800 and N810 today - installs straight from webpage link without dependency issues, or the need to touch any config files.

http://pupnik.de/dosbox.html

Reply 3 of 7, by leileilol

User metadata
Rank l33t++
Rank
l33t++
pupnik wrote:

The performance i'm getting is just sad. Betrayal at krondor with sound blaster 1 emulation needs to drop down to 500 cycles. Without music it can run ~ 1500 cycles.

Considering this is anything but a 286 class game this is not surprising given the full 256 color VGA and 3d graphics processing needed for first person travel view. This is also slow on a 386 (a 486 80MHz works best for this game, at the least)

This is not "just sad". You are intentionally overdoing it.

apsosig.png
long live PCem

Reply 5 of 7, by Freddo

User metadata
Rank Oldbie
Rank
Oldbie
leileilol wrote:

Considering this is anything but a 286 class game this is not surprising given the full 256 color VGA and 3d graphics processing needed for first person travel view. This is also slow on a 386 (a 486 80MHz works best for this game, at the least)

This is not "just sad". You are intentionally overdoing it.

I think it's pretty cool 😁 I would have been more interested if it had worked on the Nokia S60 smartphones instead of their internet tablets, though, even if it had been slightly slower (fastest Nokia S60 phone currently runs at 369 MHz).

I would love to be able to play UFO: Enemy Unknown and Terror of the Deep on my phone during train rides or whatever. Could play without sound and since it's turnbased it doesn't matter if it's a bit slow.

DosFreak wrote:

Cool, didn't know about that one 😀 I love these "old games - new engines" stuff 😁

Made me wonder about this, but it still looks utterly dead. Too bad 😢

Reply 6 of 7, by pupnik

User metadata
Rank Newbie
Rank
Newbie

Well you're right about BaK of course. I shouldn't have mentioned it.

I've just exhausted my options for speedups, with the exception of the final rendering backend, which could use a framebuffer and hardware scaling.

My two hopes for the future are that someone will manage to build the fmopl FM-synthesis to run on the TI cx3110x DSP, and that eventually we might get a dynamic recompiler for ARM on dosbox.

Reply 7 of 7, by swaaye

User metadata
Rank l33t++
Rank
l33t++

It's too bad that the Arm processors in these mobile devices have sorta stopped scaling. Years ago Intel projected a 1 GHz Xscale CPU, but years passed and we're slower then we actually were for a while because devices are smaller and have tighter power demands (phones vs. PDAs). Ignoring clockspeed, these processors are also very simple little things compared to desktop/notebook CPUs.

I'm sure a dynamic core would help, but that is a huge undertaking. I really wonder how much it would help though. DOS games are obviously x86-based and a x86 dyncore probably has big advantages over the same for another ISA.