| IRC logs at
Set by AlexDaniel on 12 June 2018.
00:59 MasterDuke left 04:49 coverable6 left, bisectable6 left, unicodable6 left, quotable6 left, reportable6 left, evalable6 left, releasable6 left, nativecallable6 left, shareable6 left, sourceable6 left, squashable6 left, greppable6 left, linkable6 left, committable6 left, notable6 left, bloatable6 left, statisfiable6 left, benchable6 left, tellable6 left 04:50 greppable6 joined, unicodable6 joined, notable6 joined, statisfiable6 joined, benchable6 joined, linkable6 joined 04:51 evalable6 joined, tellable6 joined 04:52 bisectable6 joined, releasable6 joined, committable6 joined, bloatable6 joined, reportable6 joined, quotable6 joined, nativecallable6 joined, squashable6 joined, sourceable6 joined, shareable6 joined, coverable6 joined 05:11 Util left 07:15 MasterDuke joined 07:38 sena_kun joined 07:51 zakharyas joined 08:14 zakharyas left 08:16 zakharyas joined 08:39 Altai-man_ joined 08:42 sena_kun left
nwc10 good *, #moarvm 08:57
nine Good day nwc10! 09:33
MasterDuke this patch to add ops for allocation profiling may be getting away from me... 09:51
10:40 sena_kun joined 10:42 Altai-man_ left
MasterDuke anybody mind reviewing this for me it currently *sometimes* causes `MoarVM panic: Register types do not match between value and node` or a segfault when trying to profile something 10:43
unrelated, but it looks like we have 88 ops that have templates in src/jit/core_templates.expr, but no case in src/jit/graph.c 11:53
nine MasterDuke: since the panic is caused by the JIT, does MVM_SPESH_BLOCKING make it more predictable? 12:01
MasterDuke nine: it's more like profiling some things always causes it, but other things profile just fine 12:02
nine Then I'd just bisect that list to find out which addition causes the panic 12:03
12:39 Altai-man_ joined 12:42 sena_kun left
Geth_ MoarVM: MasterDuke17++ created pull request #1270:
Add more ops for allocation profiling
14:20 lucasb joined 14:41 sena_kun joined 14:42 Altai-man_ left 14:51 MasterDuke left 15:31 zakharyas1 joined 15:32 MasterDuke joined 15:33 zakharyas left 15:54 zakharyas1 left 15:55 zakharyas joined 16:39 Altai-man_ joined 16:42 sena_kun left
MasterDuke was a bit tedious. would there be a way to do this based on a new annotation for the oplist? might make it easier to remember when adding new ops 16:46
timotimo there may perhaps want to be something a little bit more intelligent about the whole instrumentation stuff 16:52
like, sometimes an op allocates more than just one object, we don't catch that at the moment
and the allocation logger also has a little heuristic to prevent double-counting and such, but it could also prevent counting of multiple objects made in the same op 16:54
18:18 zakharyas left 18:40 sena_kun joined 18:42 Altai-man_ left
lizmat and another Rakudo Weekly News hits the Net: 19:25
20:15 zakharyas joined 20:39 Altai-man_ joined 20:42 sena_kun left 20:48 MasterDuke left 20:54 Kaeipi left, Kaeipi joined 21:01 MasterDuke joined 21:12 zakharyas left 21:13 ggoebel joined 21:29 Altai-man_ left
ggoebel lizmat: thank you for your long time stewardship and publication of the weekly! 21:49
lizmat ggoebel: you're welcome!
22:00 lucasb left 23:12 ggoebel_ joined, ggoebel left 23:19 ggoebel_ left
Geth_ MoarVM/gc_measurement_debughelper: 46a3a5ebcb | (Timo Paulssen)++ | src/profiler/log.c
add MVM_profiler_measure_gc_entry_size to count GCs and PCNs