Add mcfgthread

[lhmouse]

The mcfgthread library https://github.com/lhmouse/mcfgthread implements fast mutex and condition variables on Windows. To test mcfgthread, it's necessary to set up an environment in an MSYS2 UCRT64 shell:

  # pacman -S mingw-w64-ucrt-x86_64-{cmake,pkgconf,ninja,boost,mcfgthread}
  # cmake .
  # ninja
  # ./MutexShootout.exe
  Usage: D:\lh_mouse\GitHub\MutexShootout\MutexShootout.exe threadCount nolocks|std|boost|cs0|cs4k|srw|mcf|mcf0i

mcf allows the locking and unlocking functions to be inlined, while mcf0i calls external functions.

Results:

  $ ./MutexShootout.exe 16 srw
  srw_mutex:        29.3014
  $ ./MutexShootout.exe 16 mcf
  mcf_mutex:        14.2265
  $ ./MutexShootout.exe 16 mcf0i
  mcf_mutex (noinline): 13.4116

[lhmouse]

Implementation details: https://github.com/lhmouse/mcfgthread/blob/ed201825f6d9298def77bc7c29aa1a8f86d634bb/MUTEX.md

[lhmouse]

Here are test results on Windows 11 on ARM 24H2 (Dev channel) on Huawei MateBook E Go 2023, with an 8-core Snapdragon 8cx Gen 3, using Clang 18 as the compiler:

1 thread

• CLANGARM64 ~/GitHub/MutexShootout
• $ for s in nolocks std boost cs0 cs4k srw mcf mcf0i; do ./MutexShootout.exe 1 $s; done
  No locking: 35.8718
  std::mutex: 37.1544
  boost::mutex: 37.821
  cs_mutex_nospin: 36.7659
  cs_mutex_4K: 36.7416
  srw_mutex: 36.4696
  mcf_mutex: 36.675
  mcf_mutex (noinline): 36.6938

4 threads

• CLANGARM64 ~/GitHub/MutexShootout
• $ for s in std boost cs0 cs4k srw mcf mcf0i; do ./MutexShootout.exe 4 $s; done
  std::mutex: 43.9401
  boost::mutex: 59.4707
  cs_mutex_nospin: 68.5111
  cs_mutex_4K: 49.7991
  srw_mutex: 44.9692
  mcf_mutex: 51.0564
  mcf_mutex (noinline): 51.0674

8 threads

• CLANGARM64 ~/GitHub/MutexShootout
• $ for s in std boost cs0 cs4k srw mcf mcf0i; do ./MutexShootout.exe 8 $s; done
  std::mutex: 53.0173
  boost::mutex: 62.0042
  cs_mutex_nospin: 64.7601
  cs_mutex_4K: 63.4763
  srw_mutex: 54.9078
  mcf_mutex: 57.6587
  mcf_mutex (noinline): 56.6434