00:00:00 formal upper bounds 00:00:07 not just for the sorting algorithm 00:00:12 but for the whole app 00:00:25 They do. It's 2^{word size} :P 00:00:34 hah. hah. 00:01:17 GregorR, seriously, pre-OSX had a good thing then, you could know for sure it would keep within limits 00:01:37 each app would have a minimum and maximum size set in the info box for the program file 00:01:48 the OS would give it something between those when the app was started 00:02:05 to change you would have to quit the app, open the info box and change the values, start the app again 00:02:16 OOM ahead of time 00:02:17 wow 00:02:25 that indeed does sound pretty stupid! 00:03:02 oklopol, talking about Mac OS 6-9 (and probably older ones, but never used them. and yes it is technically System 6, System 7, Mac OS 8, Mac OS 9...) 00:03:21 oklopol, it meant you never ran into swap trash 00:03:31 of course you had to reboot to change virtual memory size 00:03:36 which is even sillier 00:03:42 doesn't windows also force it? 00:03:45 for the page file 00:03:56 while linux can change any time you want 00:04:13 actually not sure about page file size in Windows 00:04:15 oklopol, tell me ^ 00:04:24 i don't know it 00:04:24 iirc you use it? 00:04:29 page file size that is 00:04:42 i also don't know whether you need to reboot to change the size of virtual memory 00:04:47 why would i :| 00:04:49 well pretty sure you need(ed) to restart in windows to change the size 00:04:58 why would i change it? 00:05:12 and Mac OS virtual memory == Windows pagefile, == *nix swap 00:05:40 if they chose it badly, i'll just buy a new computer. 00:05:45 ... 00:05:48 -_- 00:06:04 don't be stupid 00:06:43 if i need to look beneath the abstraction, the abstraction is bad. 00:07:35 i try to keep the abstractions i make and the abstractions i use separate 00:09:39 * AnMaster wonders if he should optionally support using hugepagetlb thing for allocation in cfunge 00:10:05 translation lookaside buffer? 00:10:10 if tlb cache currently has issues with overflowing it could help speed a bit 00:10:18 oklopol, yeah it allows you to get larger pages 00:10:28 than the standard 4096 byte ones 00:10:32 i know what it is 00:10:35 # ls /sys/kernel/mm/hugepages 00:10:35 hugepages-2048kB 00:10:38 oh 00:10:42 no that's not what a tlb is 00:10:48 oklopol, indeed not 00:11:13 /usr/src/linux/Documentation/vm/hugetlbpage.txt 00:11:17 oklopol, ^ 00:11:20 point is 00:11:26 a TLB is used to find a page 00:11:47 too many pages and the cache can't fit all pages used by app 00:12:00 oh hugepagetlb does that 00:12:09 right, misunderstood you 00:12:13 thus you could save some tiny amount of time by using fewer, but larger, pages 00:12:30 oklopol, I will use oprofile tomorrow to check tlb misses 00:12:30 yeah makes sense 00:12:33 iirc it can check for it 00:12:40 to see if it is worthwhile at all 00:13:17 ehird. Look above. Insane micro optimising ^ 00:13:18 ;P 00:13:54 anyway i don't really care for that kinda optimization 00:13:56 hmm 00:14:14 oklopol, well really what would be useful I guess is changing stack direction 00:14:19 currently it grows up 00:14:26 copying strings to it means reversing 00:14:39 that is what the current bottleneck is 00:14:56 however making it grow down introduces other issues 00:15:08 like growing it not being a simple realloc() 00:15:08 but not because it's micro-optimization, caching just doesn't fit my mental calculation model. maybe because the usual computation models don't have anything ilke it. 00:15:35 s/calculation/computation/ 00:16:05 oklopol, that is the difference between "theoretical model" and "model useful in real life" 00:16:43 oklopol, still I'd say making string copying not have to swap string direction... 00:16:46 would help a lot 00:16:46 meh, no asymptotical difference 00:16:53 really 00:16:59 about 4-8 times faster 00:17:03 depending on CPU 00:17:10 because you can copy in word size 00:17:16 also less cache misses 00:17:22 and 00:17:27 you could SIMD it 00:17:38 *yawn*, let the compiler worry about that 00:17:39 to get 128-bit copying on x86/amd64 00:17:51 oklopol, I will, but it can't do it atm 00:17:54 oh simd! 00:17:56 because stack grows up 00:17:59 now that's something i do like. 00:18:03 and strings need to be *REVERSED* 00:18:08 when pushed on stack 00:18:13 which means byte by byte copy 00:18:22 which sucks 00:18:28 *WHY* do they need to be reversed? 00:18:47 oklopol, because they are 0"gnirts" in cfunge and my stack implementation grows upwards 00:18:56 which is what I want to change 00:19:00 so it grows downwards 00:19:11 so I can just memcpy() or such 00:19:58 oklopol, that is why 00:20:03 any questions? 00:20:07 no not really 00:20:34 wait 00:20:36 that quote 00:20:42 was ehird... in #esoteric 00:20:46 2008-03-21/FreeNode-#esoteric.log.bz2:okt 31 23:20:13 ah yes -- quoting alice in wonderland. The post-ironic hippy way to be cool on the internet. 00:20:51 AnMaster: What is the big question about the meaning of life, the universe and everything? 00:21:01 oerjan, I don't know :P 00:21:13 well it was worth a shot 00:21:34 i can't smell people 00:21:54 nov 06 02:39:21 "We're all mad here. I am mad. You're mad." "How do you know that I am mad?" asked Alice. "You must be," said the Cat, "or you wouldn't have come here." 00:22:00 oklopol: maybe there aren't any nearby. you were a lizard, right? 00:22:15 okt 31 23:19:45 "We're all mad here. I am mad. You're mad." "How do you know that I am mad?" asked Alice. "You must be," said the Cat, "or you wouldn't have come here." 00:22:16 that too 00:22:27 oerjan, you were obsessed with that line? 00:22:34 okt 31 23:20:13 ah yes -- quoting alice in wonderland. The post-ironic hippy way to be cool on the internet. 00:22:39 oerjan: i mean i don't have the whole sense. of smelling people. 00:22:54 AnMaster: it fits #esoteric so well 00:23:02 quoting Terry Pratchett -> no, that's the post-hippy ironic way 00:23:17 post-hippy post-ironic way -> killing yourself 00:23:40 holy shit 00:23:42 2:23 00:23:43 :DD 00:23:49 oklopol, true 00:23:53 i have 00:23:57 too much stuff to do 00:24:08 should probably 00:24:09 liek 00:24:11 go. 00:24:23 I have stuff to do tomorrow morning. For values of morning in the range like 11:00 00:24:27 to 12:00 00:24:33 so yeah I better go to sleep 00:24:44 oklopol: your haiku misses the metric a bit 00:24:44 night 00:25:02 oerjan: i know 00:25:05 ^ fixed 00:26:01 okay 00:26:07 algebra time now, i think 00:26:08 -> 00:26:21 err 00:26:42 oerjan, what is the name for stuff like hexameter and pentameter and such 00:26:48 "versmått" in Swedish 00:26:51 but in English? 00:28:07 btw: Iambic pentameter > haiku 00:28:07 "meter" i think 00:28:41 http://en.wikipedia.org/wiki/Meter_(poetry) 00:28:55 Blank verse :) 00:33:43 night 00:35:44 indeed it is 01:57:50 Blank verse is better than rhyming, 01:57:55 For there's no need to worry of timing, 01:57:59 The problem you see, 01:58:02 With blank verse, to me, 01:58:17 Is that it's just an excuse for lazy people to write prose and call it poetry. 01:58:37 *sniff* so beautiful 02:04:28 -!- MizardX has quit (Read error: 104 (Connection reset by peer)). 02:06:27 -!- oklopol has quit ("( www.nnscript.com :: NoNameScript 4.2 :: www.regroup-esports.com )"). 02:17:44 A poem is 02:17:51 Just like prose, but 02:17:57 Separated by random 02:17:59 Newlines 02:21:15 Or sometimes // Double slashes // Just for conciseness. 02:22:07 That // Too. 03:05:06 -!- psygnisfive has joined. 05:19:07 -!- oerjan has quit ("leaving"). 05:30:46 -!- MizardX has joined. 07:51:48 -!- oklopol has joined. 07:59:59 -!- clog has quit (ended). 08:00:00 -!- clog has joined. 08:10:13 -!- oklopol has quit (Read error: 54 (Connection reset by peer)). 08:10:29 -!- oklopol has joined. 08:36:08 -!- GreaseMonkey has quit ("Client Excited"). 09:12:17 -!- MigoMipo has joined. 09:27:46 -!- ais523 has joined. 10:10:00 -!- jix has joined. 10:32:17 http://news.bbc.co.uk/1/hi/uk_politics/7908498.stm haha 10:41:06 hi 10:41:11 hi 10:41:21 ais523, quite interesting speed test: dd if=/dev/zero of=/dev/null& pid=$! ; sleep 30; kill -USR1 $pid; sleep 1; kill $pid 10:41:37 on FreeBSD replace USR1 with INFO 10:41:52 why are you sending sigusr1 to dd? 10:42:00 ais523, because it dumps stats 10:42:02 like speed 10:42:15 20219142656 bytes (20 GB) copied, 30.0494 s, 673 MB/s too 10:42:21 also, you don't want that to run too much, your /dev will end up running out of zeros 10:42:24 err remove too 10:42:32 ais523, hah hah 10:45:47 ais523: Also, your bit bucket will fill up. 10:46:04 pikhq: most of the parts of this laptop are starting to break, anyway 10:46:08 I think the bit bucket is leaky 10:46:20 and the more zeros in there, the more pressure on the leaks and the faster they get out 10:47:22 no you are both wrong 10:47:38 2.6.23 and later recycles from /dev/null 10:47:58 all zeros are sent for using in /dev/zero 10:48:01 -!- ais523 has quit. 10:48:11 -!- ais523 has joined. 10:48:20 ais523, did you miss what I said? 10:48:27 yes 10:48:32 and the more zeros in there, the more pressure on the leaks and the faster they get out 10:48:32 no you are both wrong 10:48:32 2.6.23 and later recycles from /dev/null 10:48:32 all zeros are sent for using in /dev/zero 10:48:32 * ais523 has quit () 10:48:51 I was wondering when someone would think of recycling 10:49:19 ais523, other bits are used for other stuff, usually /dev/random and /dev/urandom 10:49:42 wouldn't that leave /dev/random spewing out more ones than zeros? 10:50:05 ais523, random gets 0 from the zero buffer when it needs it 10:50:08 duh 10:50:31 * AnMaster wonders what /dev/cpu_dma_latency is 10:50:45 that's the sort of thing I'd expect in /proc not /dev 10:50:56 /dev/network_latency exists too 10:51:04 # cat /dev/network_latency 10:51:04 cat: /dev/network_latency: Invalid argument 10:51:13 network_throughput as well 10:51:29 heh 10:51:32 ais523, and indeed 10:51:58 ais523: Obviously, it can toggle 0s to 1s. 10:52:26 pikhq, and the other way, it does that only if it runs out of the buffers 10:52:38 to avoid the overhead when possible 10:52:44 Well, of course. 11:07:37 -!- BeholdMyGlory has joined. 11:11:45 -!- ais523 has quit ("lunch"). 11:30:41 So can you do echo "really really much" > /dev/network_throughput when you want a faster connection? 11:39:26 And /dev/network_{latency,throughput} seems to be intended for processes who want to register/monitor QoS requirements. 11:41:48 That's a funny interface; you open /dev/network_throughput and write in what you want to get, then keep a file handle open as long as you want it to be active. I would have expected some ugly ioctl nastiness. 11:49:42 -!- KingOfKarlsruhe has joined. 11:51:49 fizzie, heh 11:51:53 yeah funny interface 11:56:32 strange, seems like time(NULL) takes a lot of the system time. Ah I guess the HRTI test calling y so often causes that... 12:04:47 -!- oerjan has joined. 12:05:06 oh right... that is callgrind's time().. 12:10:27 for (ssize_t i = len; i >= 0; i--) 12:10:27 stack->entries[top - (size_t)i] = str[i]; 12:10:38 gcc generates very suboptimal code for that, even at -O3 12:10:39 hm 12:10:54 fizzie, in jitfunge, which way does the stack grow? 12:11:07 or stacks 12:11:35 hrrm I just got an idea for fast asm on x86_64 for that stack push... 12:12:21 oh yes... abusing BSWAP I'm sure should be possible... 12:12:23 Up, I think. To higher addresses. I'm not completely sure how I did the stack-stackery, I think I just keep the topmost stack in the specially handled memory area, and copy things on stack-stack operations. 12:13:28 fizzie, well up is bad for string pushing on stack 12:13:31 I found 12:13:39 basically that is the main bottle neck in cfunge atm 12:13:56 since pushing strings mean reversing every byte to push a 0"gnirts" 12:14:09 mostly related to all those y in the HRTI test that mycology does 12:14:24 and huge env 12:14:34 Yes, well, my STRN implementation might not be very optimal anyway. 12:14:38 $ env | wc -c 12:14:38 5685 12:15:44 fizzie, well STRN isn't a bottle neck in mycology, and my STRN is "not too bad", and actually since it is defined to use strings it isn't well defined how values outside the range of char works in STRN 12:16:15 in cfunge it depends on which instruction, since some uses unsigned char*, and other ones just copies directly to/from funge space 12:16:47 should probably implement some sort of fungeCell* pushing thing... would be faster 12:17:17 (less checks for available stack space, since unlike you I don't mess with catching SEGV and poking registers to avoid checking for stack size) 12:19:09 There's no special reason why I couldn't have a downward-growing stack, except that theoretically it might be a bit more likely to find free pages after the stack than before it. But that's just a guess. And anyway I do that half-assed attempt to stick the stack out there in the wilderness if possible. 12:19:34 yeah.. down growing stacks means grow isn't a simple realloc() any more 12:19:47 for me 12:21:18 Actually, heh, I don't think current jitfunge even grows the stack at all. Haven't touched that code in a while, but all I'm seeing here are the underflow checks. 12:23:49 -!- ais523 has joined. 12:45:20 -!- MigoMipo has quit ("QuitIRCServerException: MigoMipo disconnected from IRC Server"). 13:05:02 -!- KingOfKarlsruhe has quit (Remote closed the connection). 13:11:43 ais523, hi 13:11:49 hi 13:14:15 ais523, is there any way to tell gcc that how a specific memory block is a aligned, I mean I have a pointer to an array of ints and want to tell gcc it is properly aligned for SSE access, since gcc decides to not try to vectorize the loop due to: "note: Unknown alignment for access: *array_25" 13:14:21 but I know it is properly aligned 13:14:27 -!- jix has quit (Read error: 113 (No route to host)). 13:14:42 there's an __attribute__ for it, I'm pretty sure 13:14:44 check the docs 13:15:47 ais523, that is for aligning variables yes, but there doesn't seem to be one to tell it how a malloc()ed block is aligned (and gcc can't possible trace this back to the malloc since the pointer has been stored in a struct for quite a while before and passed around) 13:16:21 couldn't you put the attribute in the right place in a pointer definition to constrain what it's pointing to, rather than what it is? 13:16:42 tried that. it seemed to align the pointer itself 13:23:19 There's that "aligned (x)" attribute for types; the example has "typedef int more_aligned_int __attribute__ ((aligned (8)));". Maybe you could (ab)use a similar typedef -- like do "typedef unsigned char sse_aligned_char __attribute__ ((aligned (42)));" and then keep a sse_aligned_char* type pointer. 13:23:31 (There's 42 because I don't really know how many bytes you want there.) 13:23:47 I don't think aligning to 42-byte boundaries would be particularly useful... 13:24:47 hm 13:25:12 I need to align 128-bits probably 13:25:23 lets try that with 64 bits 13:26:01 (8) 13:26:07 -!- impomatic has joined. 13:27:03 Also: 13:27:09 As another example, 13:27:15 char *__attribute__((aligned(8))) *f; 13:27:22 specifies the type "pointer to 8-byte-aligned pointer to `char'". 13:27:26 ais523: but but - clearly that must be the answer 13:27:33 actually that doesn't help. What I need is to tell it *base* is aligned 13:27:48 error: alignment of array elements is greater than element size 13:27:53 if I try fizzie's suggestion 13:28:17 Yes, well, you can try something like "char __attribute__((aligned(16))) *foo" too. 13:28:27 But I'm no GCCer. 13:28:43 eh 13:28:58 hm 13:29:19 Oh, maybe I should've read the rest of that paragraph. 13:29:44 "Note again that this does not work with most attributes; for example, the usage of `aligned' and `noreturn' attributes given above is not yet supported." immediately after that char *__.. example. 13:33:35 still doesn't help 13:33:51 fizzie, aligned would cause it to insert padding between each member in array 13:34:11 As far as the "attribute in typedef" is considered, certainly with a "typedef unsigned char auchar __attribute__ ((aligned (16)));" I can't do "auchar foo[10];" without that error, but a "char foo[10]; auchar* bar = (auchar*)foo;" works just fine. 13:34:15 not tell it that the first member is aligned on even 16 bytes 13:34:17 which it is 13:34:58 Not that I have any clue whether gcc would understand that access via that "bar" pointer there would be 16-byte-aligned. Or what auchar[1] would mean. Maybe there would be padding. 13:35:12 Er, I mean bar[1] there. 13:35:19 fizzie, it can do it for static variables 13:35:23 err 13:35:25 yeah 13:35:27 and stack ones 13:37:56 Yes, well, my interpretation of "typedef foo afoo __attribute__ ((aligned (x))); ... afoo* x = ...;" would be "afoo is a foo-type thing that is aligned at x-byte boundaries, so x is therefore a pointer to a foo-element aligned thusly", but I really can't know how GCC reads it. 13:38:11 icc seems to understand this with no hints btw... 13:41:03 GCC does have that vector-size attribute, too, for more explicit vectorizablitiesies. 13:42:27 The stuff at http://gcc.gnu.org/onlinedocs/gcc/Vector-Extensions.html I mean. But I haven't really found any concrete examples about "here's how to tell the alignment of a memory block for auto-vectorization purposes". 13:54:40 o 13:54:57 oko 13:55:33 okleosynthesis 13:59:46 The basis of all matter. 14:00:26 (Musts away.) 14:00:27 ...no *your* matter 14:00:32 except the matter that doesn't matter 14:00:57 I was this close to adding "at least all interesting matter". 14:01:01 Well, no matter. 14:01:04 -!- oklopol has quit (Read error: 104 (Connection reset by peer)). 14:01:19 -!- oklopol has joined. 14:02:18 16:00… fizzie: Well, no matter. ||| 16:00… * Disconnected <<< for second there i perceived this as if fizzie had quit just after that message, and somehow it would've looked like i was disconnecting from my perspective. 14:02:26 it was weird 14:05:08 Are talk pages on the esolang wiki covered by the same license as the articles pages? 14:07:15 I think so 14:07:22 there's an agree-to-PD message below the edit box 14:07:26 and it's the only way to change them 14:07:46 -!- MizardX has quit (Read error: 104 (Connection reset by peer)). 14:09:57 Thanks :-) 14:10:12 I wanted to use something from the Underload talk page 14:11:36 which one? 14:11:44 if I wrote it, it's definitely PD, with my compliments 14:14:32 Thanks, but it's Keymaker's quine :-) 14:14:59 what did you want it for, by the way? 14:15:06 i think that's in the article too? 14:15:17 no, my quine's the one in the article 14:15:20 I thought I'd include it here with an explanation of how it works http://corewar.co.uk/assembly/underload.htm 14:19:00 GCC does have that vector-size attribute, too, for more explicit vectorizablitiesies. <-- doesn't do the same thing in this case 14:42:32 -!- dbc has quit (Read error: 110 (Connection timed out)). 14:50:23 -!- dbc has joined. 15:03:35 Just starting my implementation of Forth in Redcode. Still open to suggestions 15:07:26 -!- ehird has left (?). 15:07:28 -!- ehird has joined. 15:08:23 i was gone a day 15:08:25 did you miss me 15:08:31 good 15:08:37 * impomatic did :-) 15:08:51 see, clearly I'm the most important member of this channel. 15:09:15 ehird: I was gone a day too, so no 15:09:19 I didn't know you weren't there to miss 15:09:49 -!- upyr[emacs] has joined. 15:10:03 hi upyr[emacs] 15:10:05 you new? 15:10:14 wait, ehird was gone a day? *duck* 15:10:28 :< 15:12:49 ehird: yes :/ 15:13:03 why the :/? nervous about the goat sacrifice? 15:13:04 it'll be ok. 15:13:20 also, has nobody noticed the logs link is wrong? 15:13:37 :/ decouse i don't sleep 2 days 15:13:51 that could be an issue 15:13:52 fucking wiki locking up just as i'm saving :< 15:13:56 clog: are you sure you didn't mean oklopol? 15:19:56 http://blogs.msdn.com/michkap/archive/2005/09/17/469941.aspx 15:24:10 15:59:55 IMO all apps should have upper limits on how much they will allocate 15:24:10 16:00:00 formal upper bounds 15:24:10 16:00:07 not just for the sorting algorithm 15:24:11 16:00:12 but for the whole app 15:24:20 Alternative solution: don't write memory hogging apps. 15:24:28 Crazy, I know. 15:24:49 AnMaster's method looks like a way to make sure you don't write a memory hogging app 15:25:14 err, you can easily write a memory hogging app by doing that 15:25:18 it gains you approximately nothing. 15:25:36 it lets you know exactly how memory hogging it is 15:25:57 no, it doesn't, and to boot, even if it did it's a tedious, stupid way of finding that out 15:30:34 ubiquite (ubiquiting, ubiquited), v. To make a property more ubiquitous. 15:30:46 (also, ubiquties) 15:31:06 maybe it should be ubiquit 15:31:14 , ubiquiting, ubiquited, ubiquites 15:36:23 Yay, Forth interpreter is working! 15:36:35 well done 15:36:41 Thanks :-) 15:36:50 can it multiply? 15:37:15 Yes, it supports about 20 words 15:37:38 impomatic: does "SEE" work? 15:38:12 Hmmm... I don't know what SEE is, so probably not! 15:38:48 impomatic: prints to screen the forth source code of a defined word, or the disassembled machine code of a primitive 15:39:09 also, does IF, ELSE, THEN-style branch control work? 15:41:22 -!- MizardX has joined. 15:41:37 SEE doesn't work and probably won't ever. IF ELSE THEN not yet. BEGIN UNTIL is the only flow control so far. 15:41:37 It only 45 Redcode instructions! 15:41:58 how is BEGIN UNTIL implemented? 15:42:12 It's just, you can implement the language Forth conventionally, but it isn't really a forth unless you do it the "proper way" 15:42:17 threaded machine code that is 15:42:39 -!- jix has joined. 15:43:55 the following are implemented: u. space * = 1+ - + swap ?dup dup depth abs negate r> >r r@ begin drop until 15:44:06 It's threaded 15:45:13 ehird: redcode's all about threaded machine code 15:45:17 that, and crazy side effects 15:45:23 err, no. 15:45:32 i mean you have to compile forth to threaded machine code 15:45:42 also redcode isn't _threaded_ machine code 15:45:47 ehird: yes it is 15:45:47 that makes no sense, you can only compile to that 15:45:53 have you not seen the SPL instruction? 15:46:00 I don't recall it 15:46:15 ehird: well, you're missing the point if you miss the instruction that does machine-code threading 15:46:17 -!- MizardX- has joined. 15:46:34 SPL is one of the most important opcodes in redcode, it massively changes the way the language works 15:46:41 oh wait 15:47:03 "threading" means something different in forth, nothing to do with concurrency 15:47:10 -!- MizardX has quit (Connection reset by peer). 15:47:16 iirc 15:47:27 no duh 15:47:34 -!- MizardX- has changed nick to MizardX. 15:47:45 ais523: so no, you don't know what threaded means 15:47:54 oh, ok, one word with two meanings? 15:47:59 threading in Forth means the program is compiled to a long series of subroutine addresses 15:48:08 I'm tempted to say forth's meaning came first. 15:48:16 impomatic: exactly 15:48:26 2 2 + -> {&LITERAL,2,&LITERAL,2,&PLUS} 15:48:33 (where LITERAL looks one ahead in the ip) 15:49:47 Yes, that's exactly how it compiles :-) 15:50:02 i was explaining for ais523 15:51:51 what? ais523 didn't know what threading in forth was? ^_^ 15:51:57 no 15:52:03 oh ok 15:52:23 that's pretty interesting, it's sort-of like coercing an ordinary processor into being a bytecode interpreter 15:52:26 ais523, ever seen bashforth? string-threaded iirc 15:52:47 redcode doesn't have subroutines, so you'd need to implement a stack yourself to do it like that 15:52:58 mhm 15:53:22