| --- Log opened Fri Jul 23 00:00:17 2010 |
| 00:01 | | cpux [Moo@Nightstar-20a84089.dyn.optonline.net] has joined #code |
| 00:10 | | Tarinaky [Tarinaky@Nightstar-f349ca6d.plus.com] has joined #code |
| 00:17 | | You're now known as TheWatcher[T-2] |
| 00:21 | | You're now known as TheWatcher[zZzZ] |
| 00:35 | < celticminstrel> | Which qualifiers are we talking about now? |
| 00:40 | <@Vornicus> | "inline", among others. |
| 00:41 | <@Vornicus> | "register", that pile. |
| 00:51 | <@McMartin> | "static", on the other hand, is mandatory~ |
| 00:54 | < Tarinaky> | What're we talking about? |
| 00:56 | <@ToxicFrog> | C+ |
| 00:56 | <@ToxicFrog> | C++, rather |
| 00:56 | <@ToxicFrog> | Don't do it too much or you'll go blind~ |
| 00:57 | < Tarinaky> | Too late. |
| 00:57 | < Tarinaky> | Most the uses of static you could -probably- fake with namespaces and globals :/ |
| 00:58 | < Tarinaky> | Although I suspect that's not what was being discussed. |
| 00:58 | <@McMartin> | "mandatory" means "the compiler isn't allowed to ignore it". |
| 00:59 | <@McMartin> | "inline" and "register" are hints and the compiler can override the programmer at will while remaining compliant. |
| 00:59 | < Tarinaky> | Ah. That. |
| 01:00 | < Tarinaky> | Isn't there also the requirement for an inline function to still be allocated memory as a function even if it's no longer actually called? |
| 01:03 | | Attilla [Obsolete@Nightstar-303ce1dd.threembb.co.uk] has quit [[NS] Quit: ] |
| 01:07 | | AnnoDomini [annodomini@Nightstar-ba4df6b6.adsl.tpnet.pl] has quit [[NS] Quit: leaving] |
| 01:08 | <@Vornicus> | Dunno, but with block-scoping I suspect you can actually allocate additional stack memory partway through any function without calling more functions. |
| 01:08 | <@Vornicus> | ...actually, uh |
| 01:09 | <@Vornicus> | an inline function may just end up telling its parent how much memory it needs in the compile step, if all the memory is grabbed up at call time. |
| 01:12 | < Tarinaky> | Vornicus: I was of the understanding that inline functions had to be allocated because of function pointers? |
| 01:13 | < Tarinaky> | I seem to recall under specific situations (I think if they're inside a class maybe?) they can't be optimised away. |
| 01:14 | < celticminstrel> | The register keyword is being repurposed in C++1x, as I recall. |
| 01:14 | < celticminstrel> | But I don't remember as what. |
| 01:15 | < Tarinaky> | celticminstrel: I think it's being made more abstract to refer to any speed optimisation. |
| 01:17 | < celticminstrel> | It might be just changed to mean "local variable", a function previously taken by "auto". |
| 01:39 | <@Vornicus> | Hokay. Given four points of known distance to each other (in 3 space) I can, given distances from those four points to any other, locate that point in 3 space. The question is, how. |
| 01:50 | <@Vornicus> | (three points gives you two possibilities.) |
| 01:57 | < celticminstrel> | This reminds me of Stargate, for some reason... |
| 02:11 | | Tarinaky [Tarinaky@Nightstar-f349ca6d.plus.com] has quit [Client closed the connection] |
| 02:11 | | cpux [Moo@Nightstar-20a84089.dyn.optonline.net] has quit [Ping timeout: 121 seconds] |
| 02:13 | | cpux [Moo@Nightstar-20a84089.dyn.optonline.net] has joined #code |
| 02:18 | <@Vornicus> | HOkay. Given distances between three points I can place all three on a plane; the fourth, then, gets placed via intersection of three spheres, and I just pick the positive z value. |
| 02:47 | | cpux [Moo@Nightstar-20a84089.dyn.optonline.net] has quit [Ping timeout: 121 seconds] |
| 03:40 | | PinkFreud [WhyNot@Pinkfreud.is.really.fuckin.lame.nightstar.net] has quit [Ping timeout: 121 seconds] |
| 03:43 | | PinkFreud [WhyNot@NetworkAdministrator.Nightstar.Net] has joined #code |
| 03:50 | | * Vornicus hands someone convenient the "NO SHORTCUTS" stick. |
| 04:02 | | Rhamphoryncus [rhamph@Nightstar-bbc709c4.abhsia.telus.net] has joined #code |
| 04:08 | | Stalker [Z@3A600C.A966FF.5BF32D.8E7ABA] has quit [Ping timeout: 121 seconds] |
| 04:23 | | cpux [Moo@Nightstar-20a84089.dyn.optonline.net] has joined #code |
| 04:29 | | Thaqui [Thaqui@27B34E.D54D49.F53FA1.6A113C] has joined #code |
| 05:24 | <@Vornicus> | I have hatred for linalg! |
| 05:40 | | gnolam [lenin@Nightstar-38637aa0.priv.bahnhof.se] has quit [[NS] Quit: Z?] |
| 06:11 | | celticminstrel [celticminstre@Nightstar-f8b608eb.cable.rogers.com] has quit [[NS] Quit: And lo! The computer falls into a deep sleep, to awake again some other day!] |
| 06:17 | | cpux is now known as shade_of_cpux |
| 08:16 | | AnnoDomini [annodomini@Nightstar-ce54b4b1.adsl.tpnet.pl] has joined #code |
| 08:16 | | mode/#code [+o AnnoDomini] by Reiver |
| 09:08 | | You're now known as TheWatcher |
| 10:04 | | Vornicus is now known as Vornicus-Latens |
| 10:46 | | aoanla [AndChat@37647E.0002A6.6F1B60.CA0BCB] has joined #code |
| 10:46 | | aoanla [AndChat@37647E.0002A6.6F1B60.CA0BCB] has quit [[NS] Quit: ] |
| 11:47 | | * TheWatcher args at emacs, wonders why its syntax highlighting has suddenly gone and messed up |
| 11:49 | | Tarinaky [Tarinaky@Nightstar-f349ca6d.plus.com] has joined #code |
| 12:32 | | Rhamphoryncus [rhamph@Nightstar-bbc709c4.abhsia.telus.net] has quit [Client exited] |
| 12:40 | | shade_of_cpux [Moo@Nightstar-20a84089.dyn.optonline.net] has quit [[NS] Quit: ChatZilla 0.9.86 [Firefox 3.6.7/20100713130626]] |
| 13:33 | < Tarinaky> | Right. -now- I can start worrying about making 650 operations~ (Re: Something someone said... I think it was either VOrn or TF) |
| 13:35 | <@ToxicFrog> | It was Vorn, and you were worried about the performance hit of using the Mersenne Twister despite not having done any sort of analysis or experimentation. |
| 13:35 | <@ToxicFrog> | Performance is nothing until you can prove you don't have it. |
| 13:36 | <@TheWatcher> | Make it work, then make it work better, then make it work faster~ |
| 13:45 | < Tarinaky> | I wasn't worried. |
| 13:45 | < Tarinaky> | I'm currently at the stage where it works 'better'. |
| 13:46 | < Tarinaky> | But it's noticably slow. |
| 13:46 | < Tarinaky> | So I wanted to know. How do you analyse a program to work out what bits to optimise? |
| 13:46 | < Tarinaky> | Aside from just guessing from the documentation. |
| 13:49 | <@ToxicFrog> | Your two main weapons are algorithmic analysis and profiling. |
| 13:50 | < Tarinaky> | My current 'guess' is I need to cut down the number of SDL blits I perform. |
| 13:50 | <@ToxicFrog> | A profiler is a program you run your program inside that, afterwards, tells you how the time broke down - which functions did it call, how long did it spend in them, etc. |
| 13:50 | <@ToxicFrog> | This lets you identify where your bottlenecks are. |
| 13:51 | <@ToxicFrog> | Once you know that, you can figure out how to optimize them - maybe you're using an O(n^2) algorithm where you could be using O(nlogn). Maybe you're making lots of small IO operations when you could be making a few big ones. Maybe this function here was just written without an eye to efficiency and needs to be tightened up because it turns out to be called a lot more than expected. Maybe your design is inherently inefficient and needs |
| 13:51 | <@ToxicFrog> | to be rethought. |
| 13:51 | <@ToxicFrog> | But it starts with profiling. |
| 13:52 | < Tarinaky> | Right. What'd I use to profile a python script? |
| 13:52 | <@TheWatcher> | http://docs.python.org/library/profile.html ~ |
| 13:58 | < Tarinaky> | 1882460 function calls (1882218 primitive calls) in 21.552 CPU seconds \o/ |
| 14:03 | | celticminstrel [celticminstre@Nightstar-f8b608eb.cable.rogers.com] has joined #code |
| 14:04 | < Tarinaky> | I'm not really... sure how I'm meant to use these statistics :/ |
| 14:08 | < Tarinaky> | I -think- it's telling me that I'm spending a lot of time in python2.6/random.py |
| 14:09 | < Tarinaky> | It certainly has a lot of calls. |
| 14:09 | | gnolam [lenin@Nightstar-38637aa0.priv.bahnhof.se] has joined #code |
| 14:12 | <@TheWatcher> | Yes, but that's probably what you'd expect - how much time per call, compared to other functions you are calling, does it take? |
| 14:13 | < Tarinaky> | More than it takes to perform an SDL blit by the looks of this. |
| 14:15 | < Tarinaky> | http://tarinaky.pastebin.com/Hf3H5NB6 |
| 14:15 | < Tarinaky> | Oh wait. No. |
| 14:15 | < Tarinaky> | I can't read. |
| 14:15 | < Tarinaky> | Yeah. SDL blit's up the list higher. |
| 14:17 | < Tarinaky> | So from this it's telling me i need to cut down on the number of sdl blits I perform, yes? |
| 14:18 | <@ToxicFrog> | Looks like it. |
| 14:18 | <@ToxicFrog> | 200k blits for ~250 frames seems a bit excessive. |
| 14:21 | < Tarinaky> | Three layers of about 30 tiles. Every frame's made from 'scratch' without reusing anything from the previous frame. So yeah... pretty excessive. |
| 14:57 | | Stalker [Z@3A600C.A966FF.5BF32D.8E7ABA] has joined #code |
| 15:07 | < Tarinaky> | Also, what is a 'CPU second'? |
| 15:25 | <@ToxicFrog> | A second of CPU time. |
| 15:25 | <@ToxicFrog> | That is to say, a second spent actually executing your code on the chip, which may be longer than a second realtime (if on a multitasking system) or shorter (if it's a threaded program on a multicore system) |
| 15:28 | <@ToxicFrog> | Also - three layers of ~30 tiles each? That's around 100 blits/frame. You're pushing 800. |
| 15:30 | <@ToxicFrog> | I suspect a lot of this is starfield generation; your program spent 80% of its execution time in or under starfield/get_frame |
| 15:37 | < Tarinaky> | Yes. |
| 15:38 | < Tarinaky> | Well. My plan is to make the buffers bigger than the screen by about a row and a column in all directions. |
| 15:38 | < Tarinaky> | Then actually scroll it (blitting in the new rows/columns when the buffer reaches the top. |
| 15:40 | < Tarinaky> | That way it's only performing 3 blits per frame for -most- frames. |
| 15:40 | < Tarinaky> | And fewer than it is now every few frames. |
| 15:49 | < gnolam> | http://www.lysator.liu.se/~gnolam/temp/intensimeter_28_WIP_5.png |
| 15:49 | < gnolam> | http://www.lysator.liu.se/~gnolam/temp/intensimeter_28_WIP_6.png |
| 15:50 | | Stalker [Z@3A600C.A966FF.5BF32D.8E7ABA] has quit [Ping timeout: 121 seconds] |
| 15:51 | < Tarinaky> | gnolam: Wassat? |
| 15:54 | < gnolam> | Intensimeter 28/T. Slightly modified AN/UDR-13 pocket radiac. |
| 15:54 | < gnolam> | This little feller: http://www.canberra.com/products/1130.asp |
| 16:00 | <@ToxicFrog> | Tarinaky: are you regenerating the starfield slices every frame right now? |
| 16:05 | < Tarinaky> | I'm regenerating everything every frame right now. |
| 16:10 | <@ToxicFrog> | Aah. |
| 16:11 | <@ToxicFrog> | Yeah, one quick and easy optimization you can do here is generate a starfield slice only the first time it's seen, and save it. |
| 16:12 | | Thaqui [Thaqui@27B34E.D54D49.F53FA1.6A113C] has quit [Connection closed] |
| 16:13 | < Tarinaky> | Well. I was thinking holding onto a larger buffer and only generating a row at a time would probably be faster and not involve growing memory. |
| 16:13 | < Tarinaky> | Since you're only going to be generating the same starfield slices if you're going back and forth. |
| 16:14 | < Tarinaky> | Which is a bit of an arbitrary criteria. |
| 16:24 | <@ToxicFrog> | "holding on to a larger buffer"? |
| 16:29 | < Tarinaky> | <Tarinaky> Well. My plan is to make the buffers bigger than the screen by about a row and a column in all directions. |
| 16:29 | < Tarinaky> | <Tarinaky> Then actually scroll it (blitting in the new rows/columns when the buffer reaches the top. |
| 16:29 | < Tarinaky> | Emphasis on the holding onto. |
| 16:30 | < Tarinaky> | That way it doesn't touch the starfield module at all except when it needs to :/ |
| 16:35 | < Tarinaky> | Anyway. Might as well do it now. |
| 16:42 | <@ToxicFrog> | Rows and columns? I thought this only scrolled on one axis? |
| 16:42 | < Tarinaky> | Oh no. I have it 2D atm :/ |
| 16:42 | < Tarinaky> | The maths is the same and it doesn't really make it any more complicated to do 2D. |
| 17:05 | | Attilla [Obsolete@Nightstar-ad930fff.threembb.co.uk] has joined #code |
| 17:05 | | mode/#code [+o Attilla] by Reiver |
| 17:33 | | aoanla [AndChat@37647E.0002A6.6F1B60.CA0BCB] has joined #code |
| 17:37 | | jeroid [jerith@687AAB.5E3E50.627C4E.9E55C7] has joined #code |
| 17:50 | | Attilla [Obsolete@Nightstar-ad930fff.threembb.co.uk] has quit [[NS] Quit: ] |
| 17:59 | | Syloqs-AFH [Syloq@NetworkAdministrator.Nightstar.Net] has quit [Ping timeout: 121 seconds] |
| 18:32 | | aoanla [AndChat@37647E.0002A6.6F1B60.CA0BCB] has quit [[NS] Quit: ] |
| 19:01 | | Vornicus-Latens is now known as VOrnicus |
| 19:01 | | VOrnicus is now known as Vornicus |
| 19:03 | | McMartin is now known as McMartin[Ohio] |
| 19:09 | | You're now known as TheWatcher[afk] |
| 19:13 | | jeroid [jerith@687AAB.5E3E50.627C4E.9E55C7] has quit [[NS] Quit: Bye] |
| 19:32 | | aoanla [AndChat@Nightstar-2c9ea6d0.range81-129.btcentralplus.com] has joined #code |
| 19:38 | | Syloqs_AFH [Syloq@NetworkAdministrator.Nightstar.Net] has joined #code |
| 19:39 | | Syloqs_AFH is now known as Syloqs-AFH |
| 20:11 | | TarinakyKai [Tarinaky@Nightstar-f349ca6d.plus.com] has joined #code |
| 20:15 | | Tarinaky [Tarinaky@Nightstar-f349ca6d.plus.com] has quit [Ping timeout: 121 seconds] |
| 21:46 | | You're now known as TheWatcher |
| 21:54 | | Zed_ [Zed@Nightstar-e4835f03.or.comcast.net] has quit [Connection reset by peer] |
| 21:54 | | Zed_ [Zed@Nightstar-e4835f03.or.comcast.net] has joined #code |
| 22:15 | | aoanla [AndChat@Nightstar-2c9ea6d0.range81-129.btcentralplus.com] has quit [[NS] Quit: ] |
| 23:23 | < TarinakyKai> | Anyone about? |
| 23:26 | | * simon_ abouts |
| 23:36 | < TarinakyKai> | Currently musing about a problem I had several months ago with C++. |
| 23:37 | < TarinakyKai> | Specifically, how to make pointers as painless as possible. |
| 23:40 | | TarinakyKai [Tarinaky@Nightstar-f349ca6d.plus.com] has quit [Connection closed] |
| 23:56 | | AnnoDomini [annodomini@Nightstar-ce54b4b1.adsl.tpnet.pl] has quit [[NS] Quit: SLEEP CYCLE, REBOOT.] |
| --- Log closed Sat Jul 24 00:00:18 2010 |