arm64: Expand the use of Armv8.1-A atomics

[andrew]

Summary

When targeting Armv8.1 we can assume FEAT_LSE is available and can use
the atomic instructions this provides without needing to check for
support first.

Sponsored by: Arm Ltd

[imp]

Do we care about Arm 8.0? Or will __ARM_FEATURE_ATOMICS not be defined there?

Otherwise I like it.

[emaste]

https://developer.arm.com/documentation/101028/0010/Feature-test-macros

Large System Extensions

__ARM_FEATURE_ATOMICS is defined if the Large System Extensions introduced in the Armv8.1-A architecture are supported on this target. Note: It is strongly recommended that standardized C11/C++11 atomics are used to implement atomic operations in user code.

[mmel]

While it probably does not directly apply to this case, we should be very careful with similar constructions.
Clang is slightly schizophrenic and declare minimal common list of extension if -march is passed to the compiler, but maximal list of extensions (including optional ones) if -mcpu is passed to the compiler.
See https://github.com/llvm/llvm-project/issues/90365.
To my eyes this is a bug, nothing else.

So with the current behavior of Clang, we should consider ACLE as a definition for compiler supported extension instructions, not the target CPU.

[andrew]

In D46159#1052031, @imp wrote:

Do we care about Arm 8.0? Or will __ARM_FEATURE_ATOMICS not be defined there?

__ARM_FEATURE_ATOMICS is only defined when the atomic operations are available