--- Log opened Thu Jul 18 00:00:13 2013 | ||
-!- munaafghumran [~mg0950@cpc10-aztw24-2-0-cust177.aztw.cable.virginmedia.com] has quit [Ping timeout: 264 seconds] | 00:30 | |
-!- jeremybennett_ [~jeremy@92.40.159.240.threembb.co.uk] has joined #mageec | 07:30 | |
-!- mode/#mageec [+o jeremybennett_] by ChanServ | 07:30 | |
-!- jeremybennett_ [~jeremy@92.40.159.240.threembb.co.uk] has quit [Ping timeout: 268 seconds] | 08:44 | |
-!- munaafghumran [~mg0950@eduroam-wireless-pat4.nomadic.bris.ac.uk] has joined #mageec | 09:04 | |
-!- jeremybennett_ [~jeremy@host86-166-112-128.range86-166.btcentralplus.com] has joined #mageec | 09:47 | |
-!- mode/#mageec [+o jeremybennett_] by ChanServ | 09:47 | |
-!- jeremybennett_ [~jeremy@host86-166-112-128.range86-166.btcentralplus.com] has quit [Read error: Connection reset by peer] | 09:49 | |
-!- jeremybennett_ [~jeremy@host86-166-112-128.range86-166.btcentralplus.com] has joined #mageec | 09:54 | |
-!- mode/#mageec [+o jeremybennett_] by ChanServ | 09:54 | |
-!- jeremybennett_ [~jeremy@host86-166-112-128.range86-166.btcentralplus.com] has quit [Read error: Connection reset by peer] | 09:55 | |
-!- jeremybennett_ [~jeremy@host86-166-112-128.range86-166.btcentralplus.com] has joined #mageec | 09:58 | |
-!- mode/#mageec [+o jeremybennett_] by ChanServ | 09:58 | |
munaafghumran | Am I in charge of creating the training set for the ML (or a small set thats big enough to test out a few different methods). On that topic, the training set could help to have associated power consumption for each set of flags that programs are run on, so there's a measure of what helped reduce power. | 10:01 |
---|---|---|
munaafghumran | I wanted to try to recreate the approach MILEPOST used, just on a small scale to get the method right even if for now I just use it for execution time | 10:02 |
jamespallister | If you want some initial data, I can collect some energy data for you | 10:31 |
@jeremybennett_ | munaafghumran: You should draw on jamespallister expertise. | 10:32 |
@jeremybennett_ | We are going to need training sets and evaluation sets, the latter could include bigger programs. | 10:32 |
@jeremybennett_ | jamespallister started with 10 programs, but he had another 10 under consideration. | 10:33 |
@jeremybennett_ | We do need to explicitly package these up. | 10:33 |
@jeremybennett_ | We can discuss this when we meet on Monday | 10:33 |
@jeremybennett_ | jamespallister: How's the jetlag? | 10:34 |
jamespallister | jeremybennett_: redbull helps, but I'm mostly back on UK time, if a bit tired | 10:34 |
@jeremybennett_ | Nothing liek | 10:34 |
@jeremybennett_ | Nothing like a 6am start to shock you back into local time :( | 10:35 |
jamespallister | Are you up at the TSB? | 10:35 |
@jeremybennett_ | Yup - sitting next to your supervisor | 10:35 |
jamespallister | Ah well in that case I am working very hard | 10:35 |
@jeremybennett_ | I've already briefed her on your hard work in California :) | 10:36 |
munaafghumran | jamespallister: I think the energy data would be good, though I thought AWhetter was getting this information from the boards simon gave him from benchmark programs? | 10:55 |
jamespallister | I don't think we have the boards up and running yet? | 10:56 |
AWhetter | nope | 10:56 |
AWhetter | nearly but not yet | 10:56 |
jamespallister | If you need something to experiment with immediately, I've should be able to generate some data | 10:56 |
munaafghumran | oh right okay, if you can give me a few programs energy data and the associated files so I can generate the feature vectors from milepost then I can start trying some stuff out | 10:57 |
jamespallister | Which flags, and how many data points do you need? | 11:01 |
munaafghumran | standard ones with -O3 for reference, and just some rnadomly switched ones for now (just for some data as we are goign to use bluecrystal to do it with 1000 random flags I presumed?) | 11:05 |
jamespallister | well we can't use bluecrystal to generate the energy data. I guess we can use it if the training takes a significant amount of time | 11:06 |
jamespallister | there are ~82 flags enabled for O3, in GCC 4.7 | 11:07 |
munaafghumran | okay, right well then for now could I get some data with just -O3, and then if you can run it with 70 iterations on flags randomly set on and off and collect that data for 4 or 5 programs if that's not too much hassle? | 11:11 |
jamespallister | by just -O3, do you mean the 9 flags enabled on top of O0? | 11:14 |
jamespallister | i.e. O1 enables 37 flags over O0, O2 enables 37 more (36+37 total flags) | 11:15 |
jamespallister | O3 enables 9 further flags (37+36+9 = 82) | 11:15 |
munaafghumran | oh right I meant just so it's run with -O3, like gcc -O3 hello.c so if that enables 82 flags in total, then with those 82 | 11:16 |
munaafghumran | those 82 turned on* | 11:17 |
munaafghumran | if you can save the random set of on/off flags (or I can provide some if that's easier) along with the energy output for each run with those flags, that's what I am looking for | 11:18 |
-!- munaafghunran_ [~androirc@eduroam-wireless-pat6.nomadic.bris.ac.uk] has joined #mageec | 11:56 | |
-!- munaafghunran_ is now known as munaafghumran_ | 11:57 | |
-!- munaafghumran [~mg0950@eduroam-wireless-pat4.nomadic.bris.ac.uk] has quit [Ping timeout: 248 seconds] | 12:10 | |
-!- munaafghumran_ [~androirc@eduroam-wireless-pat6.nomadic.bris.ac.uk] has quit [Ping timeout: 268 seconds] | 12:11 | |
-!- munaafghumran [~mg0950@eduroam-wireless-pat4.nomadic.bris.ac.uk] has joined #mageec | 13:28 | |
jamespallister | munaafghumran: just generating the file now | 13:33 |
munaafghumran | cheers! | 13:34 |
jamespallister | takes about 2-3 hours per benchmark | 13:34 |
jamespallister | how were you planning on extracting the features from the benchmarks? | 13:34 |
jamespallister | using milepost? | 13:34 |
munaafghumran | yeah | 13:34 |
jamespallister | ah | 13:35 |
munaafghumran | that's fine I can get them all once it's all done | 13:35 |
munaafghumran | is that a problem using milepost? | 13:35 |
jamespallister | I'm just thinking that the features may not be correct - these benchmarks were compiled with GCC 4.7 | 13:35 |
munaafghumran | if it's the benchmark programs you have given ashley, I have the feature files from him already - the low-powerbenchmarks on git? | 13:36 |
jamespallister | yes | 13:36 |
munaafghumran | hmm why would it not be correct sorry? | 13:36 |
jamespallister | I'm guessing you were planning on running the benchmarks through milepost with the same compiler options, and extracting them during the compilation process | 13:36 |
jamespallister | unless there is a method that can extract the features from a compiled binary | 13:37 |
munaafghumran | milepost can just extract the features given just the .c file so I was going to just run that extract_features option | 13:37 |
munaafghumran | but like I said if the benchmarks are the one you have given ashley, I have the feature vectors from them as ashleys already run milepost on all of them | 13:39 |
jamespallister | ah ok yes that should be good enough | 13:39 |
jamespallister | it may be slightly inaccurate due to parsing difference between different compiler versions | 13:39 |
jamespallister | but should be good enough for a first attempt | 13:40 |
munaafghumran | I think that'll be fine as yeah exactly it's just a first attempt for now, just need rough numbers and correct data to work with :) | 13:40 |
jamespallister | ok cool, how many data points do you need per benchmark? | 13:40 |
munaafghumran | what do you mean by data points exactly | 13:41 |
jamespallister | iterations - each data point is one set of flags with the corresponding energy value | 13:41 |
jamespallister | im currently doing 1000 tests per benchmark, each test has a different set of flags | 13:41 |
jamespallister | but if less are required, I can get you the data faster | 13:42 |
munaafghumran | oh right, yeah, I think for now, 500 might be enough to get something going | 13:42 |
-!- munaafghumran [~mg0950@eduroam-wireless-pat4.nomadic.bris.ac.uk] has quit [Ping timeout: 240 seconds] | 15:23 | |
-!- munaafghumran [~mg0950@eduroam-wireless-pat4.nomadic.bris.ac.uk] has joined #mageec | 15:24 | |
-!- munaafghumran [~mg0950@eduroam-wireless-pat4.nomadic.bris.ac.uk] has quit [Ping timeout: 276 seconds] | 15:58 | |
-!- jeremybennett_ [~jeremy@host86-166-112-128.range86-166.btcentralplus.com] has quit [Ping timeout: 240 seconds] | 16:21 | |
-!- jamespallister [~jamespall@it000477.users.bris.ac.uk] has quit [Ping timeout: 240 seconds] | 16:26 | |
-!- munaafghumran_ [~android@cpc10-aztw24-2-0-cust177.aztw.cable.virginmedia.com] has joined #mageec | 17:14 | |
-!- munaafghumran [~mg0950@cpc10-aztw24-2-0-cust177.aztw.cable.virginmedia.com] has joined #mageec | 17:28 | |
-!- jeremybennett_ [~jeremy@188.28.157.43.threembb.co.uk] has joined #mageec | 17:49 | |
-!- mode/#mageec [+o jeremybennett_] by ChanServ | 17:49 | |
-!- munaafghumran [~mg0950@cpc10-aztw24-2-0-cust177.aztw.cable.virginmedia.com] has quit [] | 17:50 | |
-!- jeremybennett_ [~jeremy@188.28.157.43.threembb.co.uk] has quit [Ping timeout: 264 seconds] | 18:01 | |
-!- jeremybennett_ [~jeremy@188.28.157.43.threembb.co.uk] has joined #mageec | 18:06 | |
-!- mode/#mageec [+o jeremybennett_] by ChanServ | 18:06 | |
-!- munaafghumran_ [~android@cpc10-aztw24-2-0-cust177.aztw.cable.virginmedia.com] has quit [Quit: AndroIRC - Android IRC Client ( http://www.androirc.com )] | 18:45 | |
-!- munaafghumran_ [~android@178.99.114.249] has joined #mageec | 18:46 | |
-!- jeremybennett_ [~jeremy@188.28.157.43.threembb.co.uk] has quit [Ping timeout: 240 seconds] | 18:46 | |
-!- munaafghumran_ [~android@178.99.114.249] has quit [Read error: Connection reset by peer] | 18:49 | |
-!- munaafghumran_ [~android@178.99.114.249] has joined #mageec | 18:49 | |
-!- munaafghumran_ [~android@178.99.114.249] has quit [Read error: Connection reset by peer] | 22:38 | |
-!- munaafghumran_ [~android@31.123.169.141] has joined #mageec | 23:12 | |
-!- munaafghumran_ [~android@31.123.169.141] has quit [Read error: Connection reset by peer] | 23:13 | |
-!- munaafghumran_ [~android@31.123.169.141] has joined #mageec | 23:13 | |
--- Log closed Fri Jul 19 00:00:19 2013 |
Generated by irclog2html.py 2.12.1 by Marius Gedminas - find it at mg.pov.lt!