Update yield logic for ARM processor

[lalitb]

Changes

Currently for ARM processor, we are using __yield intrinsic to yield processor while waiting in tight spin lock loop. This intrinsic is defined as part of ARM C Language Extension, and its implementation is not provided in GCC library. Being part of the language extension, the ARM Vendor may choose not to implement this intrinsic as part of its compiler toolchain. More details here: https://stackoverflow.com/questions/70069855/is-there-a-yield-intrinsic-on-arm. The otel-cpp build in that case fails with error:

/opentelemetry-cpp/api/include/opentelemetry/common/spin_lock_mutex.h: In member function ‘void opentelemetry::v1::common::SpinLockMutex::lock()’:
/opentelemetry-cpp/api/include/opentelemetry/common/spin_lock_mutex.h:105:9: error: ‘__yield’ was not declared in this scope
105 | __yield();

The change is to bypass using this intrinsic and directly use the assembly instruction. The changes are inspired from Microsoft mimalloc library and tested on a board using TI Cortex A9 processor.

For significant contributions please make sure you have completed the following items:

• CHANGELOG.md updated for non-trivial changes
• Unit tests have been added
• Changes in public API reviewed

[codecov]

Codecov Report

Merging #1325 (1be1b7b) into main (da2911c) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main    #1325   +/-   ##
=======================================
  Coverage   91.36%   91.36%           
=======================================
  Files         217      217           
  Lines        7842     7842           
=======================================
  Hits         7164     7164           
  Misses        678      678           

Impacted Files	Coverage Δ
api/include/opentelemetry/common/spin_lock_mutex.h	29.42% <ø> (ø)

[reyang]

Is there a way (e.g. macro or something) to determine if __yield is available?

[ThomsonTan]

Curious where this pattern comes from.

BTW, do we need add CI for arm?

[lalitb]

Is there a way (e.g. macro or something) to determine if __yield is available?

There is no standard approach to do this. ARM vendors may choose not to implement these extensions.

Curious where this pattern comes from.

This patten of yielding ? It was written by Josh. I would prefer to let these platform specific code go away as long term solution, and rely on yield provided by standard C++ library.

BTW, do we need add CI for arm?

Yes we need it. @esigo has done some initial work #1165, but seems builds are very slow, as there is no ARM virtual machines provided by github.