bsnes v0.038 released

[sweener2001]

what was ever stopping it from being distributed via torrent?

[byuu]

I left untranslated strings because of my lack of programming experience. If someone has the knowledge, feel free to modify my locale file.

No problem, I don't expect anyone to use those options anyway. Thank you and everyone else very much for the locales :D

I'm sure it's a pain given how fast I post new releases all the time, heh.

what was ever stopping it from being distributed via torrent?

I meant the executable is distributed like uTorrent. Eg on their download page, it's a direct link to an executable. You run it and that's the program. That's what I'm doing now, rather than putting the program inside a ZIP archive.

[diminish]

polish locale
Sorry for using polish file hosting service (to download press POBIERZ PLIK), but rapidshare says file can be downloaded only 10 times (free upload, I don't have any kind of premium account). Gratz and thanks for new version.

[DataPath]

If possible, use a single counter and a single "next necessary update count" variable to batch updates. A hierarchical event system may be even better.

When I had to simulate a circuit we were designing in college, I tried to do it with a hierarchal event system. I found it harder to debug, but that may have had more to do with my level of experience and limited time.

What you describe with a single counter and a single "next necessary update count" variable sounds a lot like a delta queue, which could actually be pretty cool.

If you kept a single central counter/clock and event queue, and then a couple accessor methods for adding events based on the clock speed of the source, it should be pretty readable. And most likely very compatible with byuu's current architecture.

[ZH/Franky]

Ugh, those games have horrible Huffman tables.

If you have the time, could you explain to me what a "huffman table" is?

[creaothceann]

Sorry for using polish file hosting service (to download press POBIERZ PLIK), but rapidshare says file can be downloaded only 10 times (free upload, I don't have any kind of premium account).

Rapidshare has a free account system - go to "Free Zone" | "Collector's Zone Login". Once logged in you can also manage the files you have uploaded.

@byuu:
Maybe bsnes is small enough (455 KB when compressed with WinRAR) to not invoke a download delay counter...

[Nach]

Code: Select all

#define forever for(;;)

...

forever
{
  do_stuff();
}

Don't limit yourself to the language you're using.

[h4tred]

If you have the time, could you explain to me what a "huffman table" is?

http://en.wikipedia.org/wiki/Huffman_coding

That article should explain the concept of Huffman encoding nicely, as well as Huffman trees.

[h4tred]

Yeah, we really don't have a sweet spot between portability, compatibility and speed right now. 9x uses an older opcode-based model, ZSNES uses x86 assembler and SNESGT is closed source.

I'd really like to team up with some other people (maybe AamirM? and work on something like that once I finish the cycle renderer. Would be fun working in a group via SVN or something.

That certainly ain't a bad idea. hiro, nall and ruby are certainly nice libraries, and it would be very nice to see something like SNESGT, but with it being open source for the Linux/MacOSX people. To be quite honest, if I decide (and if I am motivated), to get back into emu stuff, I might help you guys out .

PS: Sorry for not being my chirpy, and pleasant self lately . I didn't mean to turn into such a big asshole. Looking back, I feel absolutely terrible how I acted, and I guess I will take Aamir's advice and give peace a chance.

[byuu]

When I had to simulate a circuit we were designing in college, I tried to do it with a hierarchal event system.

What you describe with a single counter and a single "next necessary update count" variable sounds a lot like a delta queue, which could actually be pretty cool.

If you kept a single central counter/clock and event queue, and then a couple accessor methods for adding events based on the clock speed of the source, it should be pretty readable.

I have no idea what you're talking about, but delta queue sounds badass. Care to elaborate, or are you going to make me use a search engine? ;_;

If you have the time, could you explain to me what a "huffman table" is?

How the script in the game is encoded. See h4tred's link.

Maybe bsnes is small enough (455 KB when compressed with WinRAR) to not invoke a download delay counter...

Yuck. I'd rather host it locally than send people to an ad farm. I'll buy the bandwidth if I need it.

To be quite honest, if I decide (and if I am motivated), to get back into emu stuff, I might help you guys out

Sounds awesome to me. The more, the merrier. Plus you can't complain about if you worked on it :P

Man, I wish I wasn't so busy with all the bsnes stuff and the upcoming FEoEZ translation ... it'd be so much fun to start on this right now, heh.

That certainly ain't a bad idea. hiro, nall and ruby are certainly nice libraries

I'm a big fan of ruby and nall, but hiro's extremely limited. Qt would've been so much better, if only it were licensed under the LGPL.

PS: Sorry for not being my chirpy, and pleasant self lately

I'm used to it </rimshot> :P
Nah, you had me riled up and acting the same way, sadly. Lesson learned.

Seriously, I completely respect your opinion to disagree. Just please don't bring it up on these two specific boards. It forces me to respond, and that just creates all kinds of unnecessary drama that I want to avoid.

[Jipcy]

Yuck. I'd rather host it locally than send people to an ad farm. I'll buy the bandwidth if I need it.

http://aws.amazon.com/s3/
http://jungledisk.com/

I've been using Jungle Disk (which uses Amazon S3) for backup and shared drive, and I like it pretty well. Reliable file serving, only pay for what you use, and no limit.

[creaothceann]

I didn't notice any advertising on rapidshare, just counters. Meh.

[Verdauga Greeneyes]

I have no idea what you're talking about, but delta queue sounds badass. Care to elaborate, or are you going to make me use a search engine? ;_;

I didn't either, but it was actually very easy to find: http://compilers.cs.ucla.edu/avrora/api ... Queue.html

Completely off-topic, but maybe you guys can help me with this. I'm trying to do something like the following:
(x-a)³ = (x³ - 3ax² + 3a²x - a³)
However, the moment I do this the compiler completely explodes.

[byuu]

I didn't either, but it was actually very easy to find:

Oh wow, yeah. That's actually exactly what my CPU core needs.

I currently schedule IRQ and NMI hold delays, but I will also need to schedule ALU events, joypad polling steps, etc in the future.

If I'm able to compact the IRQ and NMI to a single test, then it'll actually be faster, despite testing even more things.

If it weren't such an amazing PITA to calculate how many cycles in the future were needed to trigger a given IRQ, and for all the crazy edge cases and race conditions, that would be a great way to implement range testing, too. I've tried doing that in the past, sans the delta queue part.

Way cool, thanks.

Completely off-topic, but maybe you guys can help me with this. I'm trying to do something like the following:
(x-a)³ = (x³ - 3ax² + 3a²x - 3a³)

o.O

Never seen an algebraic transformation like that before -- almost looks like the crap the FOIL method spits out.

... you want help implementing the math algorithm ... or?

[Verdauga Greeneyes]

(x-a)³ = (x³ - 3ax² + 3a²x - a³)

o.O

Never seen an algebraic transformation like that before -- almost looks like the crap the FOIL method spits out.

Really? (x-a)³ = (x-a)*(x² - 2ax + a²) = (x³ - 3ax² + 3a²x - a³)
I need it so I can split the a from the x. Anyway I've implemented the maths, but I'm still working on my problem.

Good grief. Okay, whatever the problem is, it's caused by Direct3D9's CreateDevice. It returns successfully so I don't know what's going on, but in a greatly reduced testcase I was able to observe that (x-a)³ and (x³ - 3ax² + 3a²x - 3a³) are equivalent (as they should be) when I leave the call out.

Here is the call in question:

Code: Select all

memset(&presentation, 0, sizeof(presentation));
presentation.SwapEffect           = D3DSWAPEFFECT_FLIP;
presentation.hDeviceWindow        = (HWND)param;
presentation.Windowed             = true;
presentation.Flags                = D3DPRESENTFLAG_VIDEO;
presentation.PresentationInterval = D3DPRESENT_INTERVAL_IMMEDIATE;
d3d->CreateDevice(D3DADAPTER_DEFAULT, D3DDEVTYPE_HAL, (HWND)param, D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED, &presentation, &device);

I don't think I'm doing anything strange there - param is the client rect hwnd and rendering seems to work fine.

I've fully reinstalled MinGW and gotten the latest DirectX SDK - no luck.

[Jonas Quinn]

(x-a)³ = (x³ - 3ax² + 3a²x - 3a³)

o.O

Never seen an algebraic transformation like that before -- almost looks like the crap the FOIL method spits out.

Really? (x-a)³ = (x-a)*(x² - 2ax + a²) = (x³ - 3ax² + 3a²x - 3a³)
I need it so I can split the a from the x. Anyway I've implemented the maths, but I'm still working on my problem.

Good grief. Okay, whatever the problem is, it's caused by Direct3D9's CreateDevice. It returns successfully so I don't know what's going on, but in a greatly reduced testcase I was able to observe that (x-a)³ and (x³ - 3ax² + 3a²x - 3a³) are equivalent (as they should be) when I leave the call out.

Here is the call in question:

Code: Select all

d3d->CreateDevice(D3DADAPTER_DEFAULT, D3DDEVTYPE_HAL, (HWND)param, D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED, &presentation, &device);

I don't think I'm doing anything strange there - param is the client rect hwnd and rendering seems to work fine.

I've fully reinstalled MinGW and gotten the latest DirectX SDK - no luck.

It might help to replace the call with that:

Code: Select all

d3d->CreateDevice(D3DADAPTER_DEFAULT, D3DDEVTYPE_HAL, (HWND)param, D3DCREATE_FPU_PRESERVE | D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED, &presentation, &device);

[Verdauga Greeneyes]

It might help to replace the call with that:

Code: Select all

d3d->CreateDevice(D3DADAPTER_DEFAULT, D3DDEVTYPE_HAL, (HWND)param, D3DCREATE_FPU_PRESERVE | D3DCREATE_SOFTWARE_VERTEXPROCESSING | D3DCREATE_MULTITHREADED, &presentation, &device);

Wow, you're absolutely right. Now that's something I never would have found out on my own. For the record, without that flag it also fails on MSVC. Thanks a million!

[henke37]

I have a wild guess why the performance hit exists. Cpu instruction caches. It might possibly have been short enough to fit in one transfer or something like that. But yeah, I am just guessing here.

[franpa]

Would the memory controller being integrated into the CPU chip affect it at all? the new i7 Intel CPU's and most of AMD's do that.

[Verdauga Greeneyes]

By the way, for those of you who might be interested in the details of my problem, Direct3D was forcing my calculations into single precision and I was doing a summation that really needed double. I could have mitigated the effects of this with Kahan Summation, had I known of its existence at the time - but of course finding an actual solution is even better.

Edit: actually I just tried it, and even using Kahan Summation it completely fails. So yeah, I don't know what's up with this flag or the behaviour without it, but I guess I'm using it.

[grinvader]

Jonas has a gift for that kind of stuff


he is a great voodoo expert

[FirebrandX]

byuu, if you'd like an extra mirror to host the bsnes file, I can host it on my site. I pay for my site so there are no ads, and I have plenty of bandwidth per month.

Anyway, just want to say the new version works great on my Q6600 system. I did some testing and the speed hit was about 8%, though the lowest FPS was still well above 90, so I'm well in the clear of the 60 minimum I like to have.

[byuu]

byuu, if you'd like an extra mirror to host the bsnes file, I can host it on my site. I pay for my site so there are no ads, and I have plenty of bandwidth per month.

I should be okay, my site only draws ~10% of cinnamonpirate's traffic. The main limitation of my site is space, not bandwidth.

tukuyomi has a great archive that goes back pretty far. I used to have older ones all the way back to v002, but like an idiot I formatted the drive they were on by mistake.

I did some testing and the speed hit was about 8%, though the lowest FPS was still well above 90, so I'm well in the clear of the 60 minimum I like to have.

Darn, only 90. At least v036 or so's S-DSP state machine approach boosted it 10% before this hit; so we're back to the same performance as then. And hopefully the delta queue method will pan out for some extra speed.

[deltaphc]

I'm a big fan of ruby and nall, but hiro's extremely limited. Qt would've been so much better, if only it were licensed under the LGPL.

One thing that a lot of people overlook about Qt's open-source license is that it has an exception:
http://doc.trolltech.com/4.4/license-gp ... tions.html

So, if I'm reading it correctly, Qt-based open-source software can use a non-GPL license just as long as it's one of the licenses listed there. So you probably can't use a custom license, but at the very least it doesn't have to be GPL.

[funkyass]

should it not be

(x-a)³ = (x³ - 3ax² + 3a²x - a³)