02:14.40 | *** join/#debian.gr stavrosg (~blah@athedsl-96971.home.otenet.gr) |
02:14.40 | *** join/#debian.gr stavrosg (~blah@unaffiliated/stavrosg) |
07:58.19 | *** join/#debian.gr simosx (~simosx@ubuntu/member/simosx) |
08:16.02 | *** join/#debian.gr stavrosg_ (~blah@unaffiliated/stavrosg) |
08:25.55 | *** join/#debian.gr mikeX (~mike@adsl-72.109.242.139.tellas.gr) |
08:34.43 | *** join/#debian.gr mikeX (~mike@adsl-72.109.242.139.tellas.gr) |
09:02.30 | *** join/#debian.gr LjL (~ljl@ubuntu/member/ljl) |
10:17.18 | *** join/#debian.gr kalyplato (~kalyplato@unaffiliated/kalyplato) |
10:23.17 | *** join/#debian.gr michail (~michalis@host2.digea.ondsl.gr) |
10:23.17 | *** join/#debian.gr michail (~michalis@unaffiliated/michail) |
11:39.23 | *** join/#debian.gr YianPan71 (~YianPan@athedsl-111872.home.otenet.gr) |
12:23.21 | *** join/#debian.gr cmpahar (~cmpahar@athedsl-217455.home.otenet.gr) |
15:31.20 | *** join/#debian.gr gnugr (~gnugr@athedsl-214981.home.otenet.gr) |
15:31.20 | *** mode/#debian.gr [+o gnugr] by ChanServ |
15:31.56 | *** mode/#debian.gr [-o gnugr] by gnugr |
16:39.22 | *** join/#debian.gr markos_ (~markos@71.22.9.118) |
16:49.32 | johnp | Gcc vs Ekopath4 http://www.pathscale.com/ekopath4-open-source-announcement |
16:57.59 | markos_ | vai 8a exei plaka |
17:08.15 | *** join/#debian.gr mikeX (~mike@adsl-72.109.242.139.tellas.gr) |
17:16.04 | *** join/#debian.gr mikeX (~mike@adsl-72.109.242.139.tellas.gr) |
17:33.09 | *** join/#debian.gr markos_ (~markos@71.22.9.118) |
18:52.34 | johnp | to na ginei to linux 4 fores poio grigoro den exei katholou plaka |
19:05.38 | *** join/#debian.gr simosx (~simosx@ubuntu/member/simosx) |
19:10.39 | markos_ | johnp, giati? |
19:49.43 | johnp | markos_: apo tests pou exoun ginei einai poio grigoro to pathscale |
19:50.12 | johnp | tha eprepe na to paroun oloi polh sovara |
19:50.35 | markos_ | vai |
19:50.43 | johnp | den mporo na skefto kati allo efoson exei ginei kai ato open |
19:50.48 | markos_ | alla dev to blepw va allazouv oloi apo tn mia mera stnv alln |
19:51.03 | johnp | pantos tha eprepe na allaksoun |
19:51.16 | markos_ | sigoura 8a uparxouv polla failures |
19:51.42 | markos_ | opote 8a perasei kairos mexri va givei evtelws drop-in replacement |
19:51.57 | markos_ | kai pros to parov trexie movo se x86 arxitektovikes |
19:52.03 | johnp | mporei pali na doun tou gcc kapoia tips kai na diothothoun |
19:52.09 | johnp | e nai |
19:52.36 | markos_ | kai dev eivai gia OLA ta programmata |
19:52.46 | markos_ | dev 8a kavei ola ta programmata dnladn 4 fores pio grngora |
19:53.05 | markos_ | apla xrnsimopoiei e3upva to autovectorization |
19:53.12 | markos_ | kai ta SSE* engines |
19:53.28 | johnp | (sthn mana sou to eipes ;) ) |
19:53.38 | markos_ | ? |
19:53.56 | johnp | ti me perases ? guru tou gcc ? |
19:54.24 | markos_ | evvow oti dev kavouv kava magiko |
19:55.13 | markos_ | apla xrnsimopoiouv auto-vectorization, dnladn automatn xrnsn twv SSE3 engines, gia *orismevous* algori8mous kai petuxaivouv 2-4x speedup |
19:55.18 | markos_ | px |
19:55.22 | markos_ | av exeis eva aplo loop |
19:55.57 | markos_ | int a[16]; for (i=0; i < 16; i++) { a[i] = 10*i; } |
19:55.57 | johnp | ok, peripou katalaba |
19:56.21 | johnp | kseroun kalitera ta spec tou CPU |
19:56.35 | markos_ | auto se assembly 8a ntav eva aplo loop pou 8a etrexe 16 fores, kai ka8e fora to avtistoixo stoixeio tou pivaka a[i] 8a ntav 10*i |
19:56.43 | markos_ | alla 8a etrexe 16 fores |
19:56.53 | markos_ | av xrnsimopoieis kapoio SIMD engine, px SSE |
19:56.56 | markos_ | auto 8a givotav |
19:57.16 | johnp | gia pes... |
19:58.31 | markos_ | int a[16]; for (i=0; i < 16; i+=4) { vector_add(a, values); } |
19:58.42 | markos_ | twra to vector_add eivai pseudokwdikas |
19:58.56 | markos_ | diale3a la8os paradeigma giati to 10*i eivai ligo tricky |
19:59.07 | markos_ | alla av ntav px a[i] = 10; 8a ntav pio eukolo |
19:59.11 | johnp | ok, sto point |
19:59.17 | markos_ | koita to loop |
19:59.22 | markos_ | stnv prwtn periptwsn eivai i++ |
19:59.28 | markos_ | evw meta eivai i+=4 |
19:59.47 | markos_ | stn mia periptwsn au3avetai kata 1, opote exei 16 bnmata |
19:59.56 | markos_ | evw edw au3avetai kata 4, opote exei 16/4 = 4 bnmata |
20:00.02 | markos_ | hence 4 fores pio grngoro |
20:00.12 | markos_ | autn n basikn idea |
20:00.37 | johnp | o kodikas den diaferei apo CPU ? pos ginete ayto me ton compiler? |
20:00.48 | markos_ | eivai diaforetikos kwdikas |
20:00.57 | markos_ | paragei evtoles gia SSE |
20:01.03 | johnp | mporeis na kaneis ena exe kai na to trexeis se opoio arch theleis? |
20:01.09 | johnp | aaaaa |
20:01.31 | johnp | ok |
20:01.34 | markos_ | fusika oi algori8moi eivai polu polu pio duskolo va metatrapouv se vector kwdika stnv pragmatikotnta auto pou diale3a eivai eva aplousteumevo paradeigma |
20:01.46 | markos_ | opote otav blepeis eva programma va leei SSE-optimized klp |
20:01.49 | johnp | exoun polh dromo mprosta |
20:02.16 | markos_ | stnv pragmatikotnta kapoios egrapse optimized kwdika gia eva algori8mo kai kavei kapoia sugkekrimevn douleia pio grngora epeidn eivai se SSE |
20:02.18 | markos_ | anyway |
20:02.20 | markos_ | brb |
20:02.43 | johnp | :) ok |
20:02.45 | johnp | tnks |
22:04.42 | *** join/#debian.gr YianPan71 (~YianPan@ppp-94-69-241-51.home.otenet.gr) |