Document std::hardware_ destructive/constructive _interference_size

[gonzalobg]

For aarch64, LLVM and GCC set the values of std::hardware_destructive_interference_size and std::hardware_constructive_interference_size to 256B and 64B, respectively. These values are often used to align data-structures, and changing them would break the platform ABI on Linux at least.

It may make sense to document these values as part of the ABI properly somewhere.

[smithp35]

Thanks for reporting.

At the moment I don't think those values are expected to be ABI stable. There was a long thread in https://discourse.llvm.org/t/rfc-c-17-hardware-constructive-destructive-interference-size/48674/ if the information in the thread is still correct then the value returned can vary depending on target CPU.

In particular https://discourse.llvm.org/t/rfc-c-17-hardware-constructive-destructive-interference-size/48674/43?page=2

It is understood that those values are not ABI stable and not even stable across versions of the library. A diagnostic would be great but I will not block implementing this feature on such a diagnostic (I guess I am hoping that having the feature in main will provide a sufficient motivation for a diagnostic to be added).

Maybe worth adding a comment to cppabi64 "C++ ABI" that these values are implementation defined, may vary between targets, do not depend on them being ABI stable etc.

I think we'd want to know that the C++ library implementations had committed to a stable value before documenting it.

[gonzalobg]

Maybe worth adding a comment to cppabi64 "C++ ABI" that these values are implementation defined, may vary between targets, do not depend on them being ABI stable etc.

I think this would be more than sufficient.
Apologies for not finding that llvm discourse thread.