A Julia Example

[SchrodingerZhu]

https://discourse.julialang.org/t/julia-vs-microsofts-rust-inspired-verona-language-and-e-g-snmalloc-allocator/38610/3

Seems that snmalloc runs into some performance issues and wrong memory issues here.

[mjp41]

The crash looks like it is not freeing with snmalloc:

munmap_chunk

Is not part of snmalloc. So we are not intercepting something there.

The perf is hard unless we can repro. I wonder if it is the debug version they are running?

[SchrodingerZhu]

I am afraid so. I remember that mimalloc defaults the build type to release. maybe I can report something there.

[mjp41]

Perhaps, we should default the build, and then we would avoid this in the future.

[SchrodingerZhu]

BTW, I have checked that, indeed, Julia is still trying to free some memory using libc even with snmalloc preloaded. A possible solution to check the improvement with is to rebuild Julia with snmalloc (seems that libc is invoked by libjulia.so), but its build system is plain GNU Make and I do not feel like having a try now.

[mjp41]

A stack trace of that call is probably enough to work out what we aren't intercepting?

[mjp41]

It seems looking at the discource page it is calling cfree

[mjp41]

I've exposed cfree in #179 hopefully this will fix the failing example.

[SchrodingerZhu]

@mjp41 unfortunately the commit does not fix the issue.

munmap_chunk(): invalid pointer

signal (6): Aborted
in expression starting at REPL[1]:1
gsignal at /usr/bin/../lib/libc.so.6 (unknown line)
abort at /usr/bin/../lib/libc.so.6 (unknown line)
__libc_message at /usr/bin/../lib/libc.so.6 (unknown line)
malloc_printerr at /usr/bin/../lib/libc.so.6 (unknown line)
munmap_chunk at /usr/bin/../lib/libc.so.6 (unknown line)
close_unit_1 at /build/gcc/src/gcc/libgfortran/io/unit.c:743
close_units at /build/gcc/src/gcc/libgfortran/io/unit.c:798
_dl_fini at /lib64/ld-linux-x86-64.so.2 (unknown line)
__run_exit_handlers at /usr/bin/../lib/libc.so.6 (unknown line)
exit at /usr/bin/../lib/libc.so.6 (unknown line)
jl_exit at /usr/bin/../lib/libjulia.so.1 (unknown line)
unknown function (ip: 0x7fd3567f199b)
unknown function (ip: 0x7fd3567f19a5)
unknown function (ip: 0x7fd36b697855)
unknown function (ip: 0x7fd36b6974fa)
unknown function (ip: 0x7fd36b698d84)
unknown function (ip: 0x7fd36b699832)
unknown function (ip: 0x7fd36b6b7371)
unknown function (ip: 0x7fd36b6b7a08)
jl_toplevel_eval_in at /usr/bin/../lib/libjulia.so.1 (unknown line)
unknown function (ip: 0x7fd35658be04)
unknown function (ip: 0x7fd3567b1bf9)
unknown function (ip: 0x7fd3567b1ef4)
unknown function (ip: 0x7fd36b69d439)
unknown function (ip: (nil))
Allocations: 2504 (Pool: 2494; Big: 10); GC: 0

I do not think it is snmalloc's fault.

I have traced into unit.c file on the GCC mirror at GitHub, line 743 is simply a free.

[SchrodingerZhu]

I think the problem is not related to snmalloc anymore.