https://github.com/nemanjai commented:
Do we really want to support only `__builtin_cpu_is` on AIX? It doesn't seem
like this would achieve the desired goal. Most users will use these builtins to
test for some capability on the target machine. It almost never really matters
to a user whether the CPU is a Power10. They are much more likely to care about
whether the system supports MMA so they can insert calls to MMA functions;
prefixed instructions so they can add them in inline asm, etc.
It is not clear to me what goal is achieved by just providing the processor
identification and not its capabilities.
I think it would be better if (in consultation with the AIX team), we determine
what it means when the kernel reports that the CPU is for example `Power10` and
then we emulate `__builtin_cpu_supports` as well based on that.
For example, a call to `__builtin_cpu_supports("mma")` ends up emitting a check
for whether the CPU is `Power10` or above.
https://github.com/llvm/llvm-project/pull/80069
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
