|
github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm Set by AlexDaniel on 12 June 2018. |
|||
|
01:04
leont left
02:19
MasterDuke left
03:53
Geth left
03:54
Geth joined
05:05
Kaiepi left,
Kaiepi joined
07:39
patrickb joined
09:24
sena_kun left
09:26
sena_kun joined
12:03
leont joined
12:13
SmokeMachine joined,
SmokeMachine left
12:15
SmokeMachine joined
13:05
patrickb left
13:21
patrickb joined
14:46
patrickb left,
vrurg_ left
14:47
vrurg joined
15:03
zakharyas joined
16:17
codesect` joined
16:18
codesect` left
16:19
codesections left,
codesections joined
17:10
leont_ joined
17:11
vrurg_ joined
17:12
domidumont joined
17:13
sivoais_ left
17:18
vrurg left,
leont left
17:21
vrurg joined,
leont joined
17:22
sivoais joined,
vrurg left,
leont left
17:45
domidumont left
19:36
sivoais left,
vrurg_ left,
Util left,
jjatria left,
moritz left
19:40
sivoais joined,
vrurg_ joined,
timotimo joined,
Util joined,
jjatria joined,
moritz joined
20:03
zakharyas left
20:13
MasterDuke joined
|
|||
| MasterDuke | my branch calls allocate_frame 470k more times than master | 21:01 | |
| i'm not sure i trust perf's annotation of MVM_frame_invoke. at least to me, it looks like it's saying there's time spent in the `#if MVM_SPESH_CHECK_PRESELECTION` block, but MVM_SPESH_CHECK_PRESELECTION is *not* set | 21:10 | ||
| jnthn | Curious. In the profile output is the inline ratio lower to match that? | 21:25 | |
| MasterDuke | not sure. btw, should i check master with that last decont logging change applied on top? or leave it as is (where it has the 688k deopts, but is still faster) | 21:27 | |
| jnthn | I think let's get that change in sooner | 21:29 | |
| MasterDuke | master (stock): In total, 12746473 call frames were entered and exited by the profiled code. Inlining eliminated the need to create 9760662 call frames (that's 43.37%). | 21:30 | |
| remove branch: In total, 11604551 call frames were entered and exited by the profiled code. Inlining eliminated the need to create 10902584 call frames (that's 48.44%). | |||
|
21:30
linkable6 left
|
|||
| jnthn | Wat, that should be less allocate frame calls then | 21:31 | |
|
21:31
linkable6 joined
|
|||
| MasterDuke | master (plus decont logging change): In total, 11604381 call frames were entered and exited by the profiled code. Inlining eliminated the need to create 10902754 call frames (that's 48.44%). | 21:32 | |
| my earlier counts were by just sticking an fprintf in allocate_frame, maybe that's not quite the exact same counting method? | 21:33 | ||
| japhb | MasterDuke: Wouldn't the I/O call (the fprintf) affect timings enough to change thread scheduling? | 22:35 | |
| MasterDuke | well, the only threads should be the main one and spesh, and i had MVM_SPESH_BLOCKING set. so possibly, but i tried to minimize it | 22:36 | |
| japhb | Hmmm, fair. | 22:37 | |
| jnthn | Darn it. New Ryzen showed up today. Was just going to stick an OS on it. Turn it out. Very unpleasant scraping sound from one of the fans. :/ | 22:55 | |
| *Turn it on, even | |||
| MasterDuke | an important one? or one you can temporarily leave off? | 22:58 | |
| jnthn | I don't know yet :/ | ||
| Too tired to figure it out right now | 22:59 | ||
|
23:54
Geth left
|
|||