736 Add mimalloc as opt-in replacement memory allocator

[lifflander]

Fixes #736

CMake options to enable:

-Dvt_mimalloc_enabled:BOOL=1
-Dvt_mimalloc_static:BOOL=ON
-DMI_INTERPOSE:BOOL=ON
-DMI_OVERRIDE:BOOL=ON

Environment options to add stats at the end of run

export MIMALLOC_SHOW_STATS=1

Verbose mode:

export MIMALLOC_VERBOSE=1

[PhilMiller]

Does your import modify the upstream mimalloc code at all, or is it 100% stock?

[PhilMiller]

Ugh, this actually implements its own full allocator. That's way more intrusive than just being able to get instrumentation. It's not clear without digging into the docs - does it have a mode that just tracks (de)allocations, and can provide statistics, without overriding use of whatever would otherwise be called?

[codecov]

Codecov Report

Merging #737 into develop will not change coverage by %.
The diff coverage is n/a.

@@           Coverage Diff            @@
##           develop     #737   +/-   ##
========================================
  Coverage    79.13%   79.13%           
========================================
  Files          335      335           
  Lines        10206    10206           
========================================
  Hits          8077     8077           
  Misses        2129     2129           

[lifflander]

@PhilMiller Yes, it implements it's own allocator. I couldn't find a instrumenting one that is correctly delegates to the original. I tried many variants.

I just pushed the changes to mimalloc, which adds the incremental statistic.

[PhilMiller]

Matching style and such, this should probably be named mi_get_allocated_size()

[lifflander]

Good point

[pnstickne]

Would recommend adding a 'fetch' script which also prunes all non-necessary files, such as docs, tests, examples in the TPL to reduce the additions to that which is required for inclusion in VT's repo.

Here is the script for google test, eg. https://github.com/DARMA-tasking/vt/blob/develop/tests/extern/fetch-googletest.sh (so might be lib/fetch-mimalloc.sh if pairing, although maybe the fetch scripts should be elsewhere?)

[pnstickne]

It might also be useful to collect the changes to mimalloc in this work into a patch (or set of revisions) also included somewhere in VT for when/if mimalloc is ever updated. This could be part of the 'fetch' script to apply.

The resulting patch (applied after the changes) would be a no-diff, or result in the same being applied to the input. Can be added as last commit. Primary purpose is for future work assistance.