github.com/moarvm/moarvm | IRC logs at colabti.org/irclogger/irclogger_logs/moarvm
Set by AlexDaniel on 12 June 2018.
brrt \o 15:29
nwc10 o/ 15:30
nine Could it be that it may actually take longer to trigger a full collection on a 32 bit system? 16:12
brrt how so? 16:17
(I've startd the change to have the guard ops use the deopt index... this is mostly going well but I have a few failing tests) 16:18
jnthn nine: Yes 16:25
nine: Pointers are smaller, so objects are smaller, and grep for MVM_GC_GEN2_THRESHOLD_MINIMUM for the rest of the story :) 16:26
brrt d'oh, never thought of that 16:27
nine Ooooh....that makes so much sense :) 16:34
With that explanation and the one test run where a different GC test failed, I'm now confident in just raising the iteration limit of my tests :) 16:35
dogbert17 nine: looking for more GC-related errors? 16:37
nine dogbert17: it's about an Inline::Perl5 issue 16:38
dogbert17 aha 16:39
Geth MoarVM/expr-jit-devirtualize: 142f51068b | (Bart Wiegmans)++ | 10 files
[Spesh] Make deopt use index rather than target address

We'd use the index to search for the deopt address during compilation, which makes the deopt index 'special'. And then we'd use the deopt address to search for the deopt index, in order to find the objects that we'd need to rematerialize.
... (9 more lines)		 16:56
MoarVM/expr-jit-devirtualize: 147f269792 | (Bart Wiegmans)++ | tools/jit-bisect.pl
[Tools] Spesh bisect should learn about PEA
dogbert17 libuv 1.32 is now out - github.com/libuv/libuv/blob/v1.x/ChangeLog 17:07
timotimo when running a script with heap snapshots activated, seen takes up almost 10% and saw just over 5% 17:58
to be fair, zstd also takes like 9% in total 17:59
i've been meaning to look into how multiple threads could work on compressing different attribute streams at the same time to get the heap snapshot speed up a little 18:00
i see 12.61% of perf's hits inside __strcmp_sse2_unaligned 18:03
i suppose i'll have to run without jit and with callgrind to get accurate info on this
timotimo there was a get_type_index-related cache that wasn't being used at all yet, I put that in and its hit rate is at 100% after the initial miss that sets up the cache slot 19:08
but now there's a crash, too
reducing the debug value makes it go away 19:09
which is always welcome
MasterDuke how/where is a cache existing but not being used?
timotimo existing as in "there's an array put into a struct" 19:10
not being used as in "never read from, never written to"
it's just for the heap dump profiler
MasterDuke huh
timotimo ==31911== Invalid read of size 8
==31911== Address 0x901bc8e0 is not stack'd, malloc'd or (recently) free'd
bleh
at least i got the crash in rr now 19:19
oh, that's a different crash 19:20
well ain't that fun
timotimo i'm seeing a string be put into the strings list of a CU, allocated in the gen2, then later on being overwritten by an object coming from the nursery 20:05
i should have looked at GC_DEBUG earlier 20:10
MoarVM panic: Collectable 0x5d9fd00 in a gen2 freelist accessed 20:18
yes, exactly
lizmat and another Perl 6 Weekly hits the Net: p6weekly.wordpress.com/2019/09/09/...repidated/ 21:34

Powered by Cro and the Raku® Programming Language.

Please report any issues / comments / feature requests as an issue on App::Raku::Log.

Thank you!