|
02:16
colomon joined
03:44
cognominal joined
04:32
jnap joined
04:43
krunen joined
07:11
FROGGS joined
08:20
sorear joined
08:21
woolfy joined
08:57
odc joined
10:01
FROGGS joined
11:58
krunen joined
12:00
FROGGS joined
14:26
jnap joined
15:21
frankbutt joined,
frankbutt left
15:51
krunen joined
16:50
FROGGS joined
17:24
FROGGS joined
18:13
tgt joined
18:16
benabik joined
18:41
jnap joined
|
|||
| diakopter | nwc10: patch looks good to me | 19:30 | |
| FROGGS | nwc++ | 19:32 | |
| nwc10++ | |||
| timotimo | so who has measured the speedup? | 19:46 | |
| nwc10 | me, just once and once, on a multi user machine, for the setting | 19:48 | |
| specifically, not the spectests | 19:49 | ||
| and not startup time | |||
| timotimo | right. | 19:50 | |
| so nobody did it properly | |||
| let me do it, then :) | |||
| timotimo builds all the backends for nqp | 19:53 | ||
|
19:55
tgt joined
|
|||
| timotimo | first timing with your patch: Stage parse : 44.563 Stage mast : 20.349 | 20:02 | |
| nwc10 | I'm assuming that the setting compilation has an awful lot of gen2. Atypically high | 20:03 | |
| so will be the best-case win | |||
| timotimo | maybe. | ||
| i'll run spectests afterwards | |||
| Stage parse : 44.670 Stage mast : 20.307 | 20:04 | ||
| one more, then i'll undo the patch and run those again | 20:05 | ||
| actually, i should probably directly run a spectest | |||
| since that won't need any extra recompilation | |||
| Stage parse : 44.742 Stage mast : 20.439 | 20:06 | ||
| unfortunately the spectests still take a while :P | 20:08 | ||
| ah, i already reached S32 | 20:09 | ||
| there's kind of a double hump there :P | |||
| TEST_JOBS=4 make m-spectest 771,24s user 30,92s system 373% cpu 3:34,72 total | 20:10 | ||
| unfortunately i didn't get the maxresidentk for some reason :o | |||
| nwc10 | I doubt that that will change | ||
| timotimo | Files=777, Tests=29654, 205 wallclock secs ( 4.97 usr 0.78 sys + 762.92 cusr 28.87 csys = 797.54 CPU) | ||
| ISTR that was quite a bit more the last time i tried | 20:11 | ||
| doing a second run nw. | |||
| and then i'll do measurements without your patch | |||
| Files=777, Tests=29654, 206 wallclock secs ( 4.94 usr 0.76 sys + 763.32 cusr 29.49 csys = 798.51 CPU) | 20:14 | ||
| TEST_JOBS=4 make m-spectest 770,69s user 31,34s system 377% cpu 3:32,50 total | 20:15 | ||
| so that's not very unstable | |||
| nwc10 | is sometimes slower for me, but I don't know what else is running on the machine | 20:16 | |
| seems nothing now I look again | |||
| timotimo | Stage parse : 44.971 Stage mast : 20.451 | ||
| nwc10 | OK, I got better than that :-( | 20:17 | |
| but it *is* in the right direction. | |||
| timotimo | that's slower than both my previous measurements for stage parse, but no change for stage mast | ||
| i suppose things in stage mast will die very young? | |||
| nwc10 | I don't know enough to be sure | ||
| timotimo | Stage parse : 44.775 Stage mast : 20.669 | 20:18 | |
| nwc10 | I'm also building with -Og, so not maxing out gcc's optmiser, so I might get better speedups than real optimsed runs | ||
| timotimo | i'm using -O1 | ||
| Stage parse : 45.013 Stage mast : 20.571 | 20:19 | ||
| so it's a little bit better, but not much | |||
| nwc10 | OK. :-/ | 20:20 | |
| but it's in the right direction. | |||
| timotimo | Files=777, Tests=29654, 205 wallclock secs ( 4.97 usr 0.77 sys + 761.93 cusr 29.87 csys = 797.54 CPU) | 20:25 | |
| TEST_JOBS=4 make m-spectest 771,09s user 31,94s system 358% cpu 3:44,21 total | |||
| i see much less cpu percentage | |||
| but more time all in all? | |||
| i'll have to re-run that at least once. | |||
| er, wait | 20:28 | ||
| i looked the wrong way around of course | |||
| your change gave a noticable win, actually | |||
| better cpu utilisation and 12 seconds less run time | 20:29 | ||
| second run coming up | |||
| nwc10 | let's hope that the second run agrees | ||
| timotimo | Files=777, Tests=29654, 205 wallclock secs ( 4.94 usr 0.72 sys + 761.93 cusr 29.62 csys = 797.21 CPU) | ||
| TEST_JOBS=4 make m-spectest 769,11s user 31,57s system 377% cpu 3:32,16 total | |||
| so the wallclock seconds seem to be the same | 20:30 | ||
| nwc10: still, good catch! | 20:36 | ||
| 0.19user 0.03system 0:00.23elapsed 98%CPU (0avgtext+0avgdata 97668maxresident)k ← without patch | 20:37 | ||
| (perl6-m -e 'say 1') | |||
| 0.20user 0.02system 0:00.22elapsed 98%CPU (0avgtext+0avgdata 97892maxresident)k | |||
| with your patch | |||
| unfortunately it's very noisy at that level, so i'd say there's no change there | 20:38 | ||
| diakopter | heh | ||
|
20:41
FROGGS joined
|
|||
| timotimo | well, the startup time being way too small is a *good* problem to have :D | 20:41 | |
| dalek | arVM/small_big_ints: 482343c | (Timo Paulssen)++ | src/math/smallbigintsupport. (2 files): use MVMObject in smallbigintsupport. |
21:27 | |
| arVM/small_big_ints: cc17337 | (Timo Paulssen)++ | src/ (4 files): saner (but still not very sane) bigint coercion in ops |
|||
| arVM/small_big_ints: 449ad14 | (Timo Paulssen)++ | src/math/bigintops.c: this may be wrong, but it gets us further i think i'll have to get the data back into the bigint object at some point? |
21:36 | ||
| timotimo | i think i need another rethink of this :p | ||
| hoelzro | hehe, small big ints... | 21:48 | |
| timotimo | you're not the first ... :) | ||
|
23:23
tgt joined
23:40
lizmat joined
23:52
benabik joined
|
|||