16:11 <irker238> systemtap: tg systemtap.git:master * release-4.3-60-gf70d74ef6 / stapbpf/stapbpf.cxx: print formatting fixes for stapbpf rlim_t

19:59 <agentzh> fche: howdy! do you have any clues about this stap optimization problem? https://sourceware.org/bugzilla/show_bug.cgi?id=26392

20:00 <agentzh> since there are so many different optimizers in stap, not sure which one would be responsible for this.

20:07 <agentzh> any hints would be highly appreciated :)

20:13 <fche> hi agentzh

20:16 <fche> there is a pass for memoizing pure embedded-c functions

20:17 <fche> stable_functionall_visitor ?

20:17 <agentzh> oh, so the right fix is to disable this memoization?

20:17 <fche> well

20:17 <fche> maybe we need to make it more clever

20:18 <fche> this would be kind of sort of like that

20:19 <fche> but it's weird, it looks as thought the l->__tmp0 value is just not stored anywhere?

20:20 <agentzh> nope. it's discarded.

20:20 <agentzh> the only useful thing is l->l___stable___global_ubacktrace__overload_0_value.

20:20 <agentzh> that's where the value is actually used.

20:20 <fche> right, that works

20:20 <fche> weird to keep copying it so much, but anyway that's a separate question

20:21 <agentzh> aye

20:21 <fche> I think this memoization scheme presumes that it's run only on lightweight functions

20:22 <fche> maybe the *backtrace ones should not get the /* stable */ marker even though they are

20:22 <fche> that way this optimization will not kick in for them

20:22 <agentzh> okay, so a patch is welcome for this change?

20:22 <agentzh> glad we don't have to touch the C++ codebase.

20:24 <fche> well worth a test

20:24 <fche> if that works, sure commit at will

20:25 <fche> it was really intended for things like pid() tid() etc.

20:25 <fche> where an unnecessary call is still very lightweight

20:26 <agentzh> got it.

20:26 <agentzh> thanks

20:26 <fche> nice find, good luck :)

20:26 <agentzh> it was caught by my colleague, Junlong Li and i won't take his credit for this one ;)

20:26 <fche> sssh, he never needs to find out :)

20:27 <agentzh> heh

20:27 <fche> btw azhang is making some floating-point progress

20:28 <fche> will see over the next little while how best to make progress

20:31 <agentzh> fche: sorry, i never got the time to clear up our patch for floating-point numbers.

20:31 <agentzh> it's quite messy and big.

20:35 <agentzh> fche: patch done and tested: https://gist.github.com/agentzh/daba38cfdb1863c54bf7c6a95c512721

20:35 <agentzh> it's indeed the cause.

20:58 <irker238> systemtap: yichun systemtap.git:master * release-4.3-61-gd4d890cb2 / tapset/linux/context-unwind.stp tapset/linux/ucontext-unwind.stp: PR26392: removed the "stable" flag from unwunding tapset functions

20:58 <agentzh> committed.

21:19 <fche> thanks dude

21:19 <fche> nice doing business with you :)

22:03 <agentzh> my pleasure :)