RFC: Implement Intel ia64 instrinsics

Jan Beulich JBeulich@novell.com
Tue Aug 10 10:31:00 GMT 2004


a couple of questions/comments:

- Why define a processor specific popcnt builtin when there already is
a generic one?
- As for some of the __asm-s (e.g. for fwb, srlz.[di], mf, sync.i,
invala, fclrf.sf0) - can't they become builtins the same as most
everything else? It seems to me that the difference to those implemented
as builtins is that none of these here take arguments, which I don't
think prevents them from being builtins.
- Some of the __asm constructs use ::: with no intervening blanks.
Isn't this prone to cause problems for C++, where :: is supposed to be a
single token (3.3.3 accepts all of :::, : ::, and :: :, but 3.4.1 as
well as mainline choke on : ::, even though the intention appears to be
to support any combination)?


More information about the Gcc-patches mailing list