This is the mail archive of the
mailing list for the GCC project.
Re: PR middle-end/8408
> Oh, I see. We'd have to have two forms of nonimmediate_operand,
> one of which didn't accept rvalues like addressof.
Same story for register_operand and pmode_register_operand, because both
accept ADDRESSOF according to recog.c .
> Something to put on the to-do list might be to create an
> lvalue_operand predicate which does not accept addressof, use
> that instead of nonimmediate_operand where necessary.
Then we'll need lvalue_register_operand and lvalue_pmode_register_operand too.