Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Aklyon]

Set up Boss Runs for station battles, just to give them a bit more oomph.

Oh, and this might make it:

That thing is looks like it'll either be a really big, easy target to hit, or a very dangerous target to hit.

[xenoargh]

A bit of both, with maybe some surprises thrown in. 

Haven't coded it up yet, but I have plans, muahaha.  Will probably code the sucker as soon as I get done writing descriptions and all that; I'm doing some profiling this afternoon on a few things that I want to make just a wee bit faster.

[xenoargh]

Having way too much fun with profilers. 

I will make the PD AI fast-yet-good, somehow, lol. 

It's the most expensive single piece of AI code, because so many of the little pew-pew guns are PD, natch. 

I've made it roughly 3X as fast as it was, roughly on-par with the regular, non-PD, which is pretty good.

The biggest issue with the PD AI is that when there are hundreds of missiles on the battlefield, things get pretty painful; SS wasn't ever really designed for Macross Missile Massacres on epic scales.  This comes up mainly with the Glaug, where the Flyter remains a lag-inducing thing. 

Sure, I can just change their weapon, but why not work on fundamental issues?

I think I'm going to try something out; there's probably a way to iterate through the ships once every few frames and build an array for that ship that's public; then the individual weapon AIs can check that list, instead of maintaining private lists.  If that's doable, it'd speed up weapon AI performance enormously; one ship's list can get used for all of the weapons on that ship, reducing the number of distance checks by a huge amount.  For ships with just one weapon, there wouldn't be any speedup, of course... but for stuff with more than 1, it'd be a huge savings, since it's necessary to do a huge amount of distance checks per update.

Buuut... interestingly enough, the most expensive single calculation in the whole shebang is WeaponAPI distanceFromArc();  I've been thinking that I should write a rough approximation that uses two triangles instead, as I think it'll result in considerable speedups with no really obvious loss in fidelity.

Anyhow, got through a chunk of the writeups; will try to get the rest done and kick out a release tomorrow if at all possible

[xenoargh]

Huh.  I got the cost of getting a rough approximation of a ship's position inside of an arc down to lower than the per-frame costs of grabbing the location (i.e., it fails the distance check, which is inherently faster, more often than it needs to check the arc, but the costs of getting the approximate solution are lower overall).

Still kind of lag-city when there is enough stuff going on with major fighter battles, but it's progress.

Other fun fact:  sound processing appears to be a major bottleneck; the last snapshot shows it as about half the load. 

That kind of sucks, because there is no way to optimize sound at all, mod-side; I can't set up priority queues and SS doesn't have a typical priority system for sound (or if it does, it's internal).   I'd hate to have to give up machineguns, but perhaps I should use the technique I developed elsewhere to fake 'em...

[xenoargh]

You ever have one of those days where you're like, "Yeah, I'm good"?

This is definitely one of mine.  I figure LazyWizard might find this intriguing, heh:

The PD AI, once the largest eater-of-CPU in the entire sim, is almost 4X times faster than the standard Autofire AI (which means when that is converted since it started out being roughly 2X faster... huge fights are going to be quite playable even on fairly old hardware).  That snapshot is from a profile taken during a huge fighter / fighter massacre with lots of missiles flying about.  I guess the ultimate test is what happens in a Glaug / Glaug fight, but I'm very happy about this.

This is quite a lot faster than Vanilla's autofire AI at this point, and I am frankly amazed at how much more speed was actually available.  I'm glad I tried some crazy ideas out.

Tech explanations of this stuff, if anybody wants some, will follow when I get around to doing further testing / bulletproofing / spanking of any remaining bugeroos.

Anyhow, I think I'm going to go ahead and finish this mini-project before release; I'm fairly confident that it's crash-free and I'm utterly convinced it's the right approach.  On the good side, I got most of the text stuff done, so tomorrow night looks like it's time, short of anything really weird coming up.

[Debido]

The StandardAutoFireAI really took me by surprise, 12 seconds is a fair chunk of change when it comes to CPU processing time. Glad to hear you were able to make a more efficient Auto Fire AI for yourself to get some more frames out of it.

It seems that there are places in the game engine there are some places some optimisations can be made by Alex, particularly the sound system. It would be interesting to know if Alex could improve that sound system code for very large battles.

EDIT: Actually I was thinking about the sound system, that occurs in another thread anyway so the entire load should be able to be balanced on one whole other core in a multi-core CPU system - so in theory shouldn't put a burden on the framerate...in theory.

[Richyread]

You ever have one of those days where you're like, "Yeah, I'm good"?

This is definitely one of mine.  I figure LazyWizard might find this intriguing, heh:

I really wish I understood half of what you've written (wanna be codejunkie...but RL keeps getting in the way )

But one thing I can say is I definitely appreciate a good bar chart now and again! 

Seriously though I've been following your mod intently and I'm very much looking forward to playing this updated version.

Keep up the good work.

Richy,

[xenoargh]

EDIT: Actually I was thinking about the sound system, that occurs in another thread anyway so the entire load should be able to be balanced on one whole other core in a multi-core CPU system - so in theory shouldn't put a burden on the framerate...in theory.

There is communications load between threads and all that, so not necessarily.  That said, I think it was an anomalous result from that profiling run, but IDK right now and either way, I can't really do anything about it.

Anyhow, I got the last big problem solved on paper, and speed is way, way, way up for the core issues that were a concern, but I want it to go even faster; Glaug on Glaug is still too slow for me, I want Flyter swarms spitting thousands of missiles at each other running at at least 15FPS, if that's technologically possible on this hardware.  That's probably asking for the moon, but we'll see. 

Need to hack the paper solution out and see where I'm at, speed-wise, but for anything less than severe missile madness, speeds are hugely improved over the last build.  And there are quite a lot of low-hanging fruit left; this has mainly been about optimization of the turret AIs, which are the biggest drains, but ship AIs and System AIs and a few other things can also contribute fairly significant gains.

[xenoargh]

Success

Glaug huge fleet vs. Punkjunker huge fleet, 120 FP in action on both sides:

The profiler results tell it all:


Now I just need to hack the changes into the AIs that haven't been optimized yet (I've just done PD, Standard and Strike here); there are other areas where the cost-savings can be realized as well.

[Debido]

*whistles* wowzers, great work Xeno. These new PD AIs will probably be useful in Eternal Nox eventually with all the weapons it has.

[xenoargh]

I think it's finally solid.  There's a fair amount of savings still left to find in AI-land, though; what I learned / built has a bunch of applications.  It's kind of exciting for me because the total cost changes should be appreciable. 

Too tired to finish that stuff tonight, though.  Gah, I'm starting to sound like Chris Roberts here: no release this evening folks, must sleeeeep.

[Protoman]

I've got to say... one of the best mods out there . There is one thing that bugs me though, and that is the burn drive. Now I don't have any problem with the ability but the fact that the AI just doesn't seem to do well with it. Ships with it tend to just circle the target (Haven't tested vs without doing much but a few shots here and there. That doesn't really work against ships with armor regeneration at all. Just do a Mastiff class vs a Archer class in the simulater and you can replicate this bland battle experience. Maybe I am doing something wrong or I am not playing the mod right? Please tell me! 

Also when in space I can't see enemy ships, just thrusters of them and their rank. BUT I am able to see my own ships.

[dmaiski]

burn drive vs armor regen :
aparently armor regen is getting a rework
(@xeno if you have not done this already)
burn drive ai probably needs a fix to prevent it from being on unless it is chasing a target ship

only seeing thrusters : thats intentional
(@xeno you may want to add description of this into front page so people dont ask so much )

@xeno
btw do you feel that AI fleets arent agressive enough in campaign map? (ive rarely gotten attacked or chased by them... ever)

also >.< release, i has exams to study for, but im bored, i need your mod to procrastinate effectively!!!

[xenoargh]

Burn Drive AI has been fixed for over a month, same with the Armor Regen (very much nerfed, but still useful, but now Neutronium is competitive).

Agreed about the AI fleets, but that's going to have to wait.

Also when in space I can't see enemy ships, just thrusters of them and their rank. BUT I am able to see my own ships.

That's on purpose, both to save memory and to make it so that players can't know fleet compositions in advance.

Anyhow... I have some minor optimization stuff I still want to get done while the ideas are fresh; lots of cost savings are possible still in various areas, due to the core changes I made for the weapon AIs.  Writeups still aren't done, but I may just release and not worry about that until later this week; I think that it's more important to finish the code stuff and let people play it for a while.

[xenoargh]

Got an hour; going to try and finish the text entries.  Everything else is go for launch.