[Bug target/69299] [6 Regression] -mavx performance degradation with r232088

["jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69299

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Maybe we really need to have two types of memory
constraints, ones which can be worst case always satisfied by reloading
their address into an address register and another ones which can be worst
case always satisfied by loading the memory into a temporary register (for
loads) or storing it from a temporary register.

Or consider the constraint as CT_MEMORY only if the operand satisfies the
constraint predicate and as CT_FIXED_FORM (or whatever is the default)
otherwise?  Only normal CT_MEMORY, for Bm as long as it satisfies
memory_operand, the exact address form doesn't really matters, but what matters
is something inherent in the memory (and ISA flags).