06:19 <discocaml> <sim642> Doesn't the profiler also give you stack traces?

06:20 <discocaml> <sim642> I think somehow dumping the lambda IR from the compiler shows the numbers. Not sure how convenient that is to do in a real project though

10:16 <neiluj> yikes, got a memleak to debug

10:58 <discocaml> <contextfreebeer> not by default but you can enable them it seems, I'm using `perf`

12:50 <discocaml> <kakadu18> Folks, is it normal that `Float.(max (-.nan) nan)` is `-nan`?

12:53 <neiluj> Hey! how to interpret this flamegraph? http://lutim.altay.fr/6NX1hPcw/ucwxOSTt.png I don't understand where all these heap allocations are coming from

12:54 <discocaml> <kakadu18> I'm trying to read https://iremi.univ-reunion.fr/IMG/pdf/ieee-754-2008.pdf#page=40 and d3i says that negative Nans should be below positive ones.

12:55 <neiluj> the code for with_clean_stack6 is https://bpa.st/LT5Q

12:56 <neiluj> the C code gerepilemany6 does not allocate to the heap

13:11 <discocaml> <contextfreebeer> what is your cpu?

13:13 <discocaml> <contextfreebeer> on x86 at least iirc there is no such thing as signed NaN, there is a distinction between quiet and signaling instead, and in that case an ordering on the NaNs might as well just be based on the bit pattern with msb set = higher, right?

13:14 <discocaml> <contextfreebeer> this is not something covered by the ieee754 standard AFAICT, so there is no correct behavior

13:23 <discocaml> <contextfreebeer> my mistake, it does seem to be specified based on what you sent

13:24 <discocaml> <contextfreebeer> however, `signaling orders below quiet for +NaN, reverse for −NaN`, not sure what this means exactly, if you just read the first part, it implies that quiet > signaling which means `-nan` would be higher, but I don't understand the `for +NaN, reverse for −NaN`

13:32 <neiluj> okay got it, the garbage collector is not called enough

13:35 <neiluj> that's annoying thoug

13:36 <discocaml> <Kakadu> I'm on amd64 4.14 where nans are signalling. So the distinction between quiet and signalling seems not to be relevant there....

13:36 <discocaml> <contextfreebeer> what do you mean, *all* NaNs are signalling?

13:36 <discocaml> <contextfreebeer> is it an AMD cpu, not intel?

13:37 <discocaml> <contextfreebeer> I am reading the intel x86_64 documentation and it looks like *any* operation performed on both signaling and quiet nans returns the quiet nan

13:38 <discocaml> <contextfreebeer> which is consistent with the results you showed

13:39 <discocaml> <contextfreebeer> well, it doesn't return the quiet nan operand exactly unless it is an x87 operation, but the signalling nan is converted to a quiet one for sse floating point operations

13:39 <discocaml> <kakadu18> Why it returns -nan? I expected just a nan.

13:40 <discocaml> <kakadu18> I meant that Float.nan is a signalling one on 4.14

13:40 <discocaml> <contextfreebeer> yep, makes sense

13:41 <discocaml> <contextfreebeer> and -nan is quiet

13:43 <discocaml> <contextfreebeer> so, https://godbolt.org/z/5bYnf6169 it looks like the comparison instruction here is `comisd` and then the flags are checked to see which is the maximum

13:44 <discocaml> <contextfreebeer> the documentation says that sse instructions, if both operands are NaNs, will convert the first operand to quiet nan if it is signalling, otherwise return it, in any case, the result is a quiet nan

13:44 <discocaml> <contextfreebeer> so the max will always be `-nan`

14:43 <discocaml> <kakadu18> I think, that I extended my understanded, thanks!

15:58 <neiluj> what could be the reasons the memory consumption keeps increasing (excluding memory leak?)

16:18 <discocaml> <contextfreebeer> bad space complexity?

16:35 <neiluj> thanks!

16:57 <neiluj> 'cause I got a bit loop that computes a sequence from the previous terms, so the memory consumption should be constant

16:58 <neiluj> however, the used memory keeps growing

16:58 <neiluj> with the code I posted earlier: https://bpa.st/LT5Q

21:48 <discocaml> <holmdunc> Not sure how relevant it is, but OCaml 5(.1.0) is now directly installable from Homebrew. It was previously stuck on 4.14.0 for a long time