[aapcs64] Specify the __aarch64_sme_accessible SME support routine

[MacDue]

This is documenting an existing practice by LLVM's compiler-rt to depend (weakly) on __aarch64_sme_accessible. This is used to determine if SME is available on platforms where there is not a defined way of checking the CPU features.

The intent is to document this so libc implementations know what they need to provide.

[MacDue]

cc @sdesmalen-arm (note: I can't add reviewers in this repo).

[sdesmalen-arm]

I'm not sure that this is the right solution, because now we have two functions that define whether SME is available (__arm_sme_state and __aarch64_sme_accessible). __aarch64_sme_accessible is not explicitly specified, but it is also used by libgcc. A libc for a baremetal target could implement __aarch64_sme_accessible if they want LLVM's compiler-rt routines to handle SME code. Because the ABI doesn't really specify where these routines should live this is more of an implementation decision so that the routines can live in the compiler runtime. Perhaps the better solution would be to let LLVM's compiler-rt define __arm_sme_state as a weak symbol so that a libc for a bare-metal target can override it if they want to use a custom implementation that doesn't rely on __aarch64_sme_accessible.

[MacDue]

One thing to keep in mind is this does not just effect __arm_sme_state. Several ABI routines include a step that tests if SME is available. So I can see that it could make sense to separate the "is SME available" step (that compiler-rt can't implement for baremetal) to a common function.

Alternatively, I think it'd make more sense not to implement __arm_sme_state on baremetal at all (no weak symbol) and ensure all availability checks are in terms of __arm_sme_state.