This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Problem with your new SRA code


Thanks for fixing this so quickly!

    Initial instantiation for bin_op_types<D1060>
    Scalarization disabled for bin_op_types<D1060>

Clearly in this case, it should indeed be disabled.

Hoever, does this disable scalarization any time a non-constant index
is found?  If so, I think that's overly pessimistic.

Suppose I have an array of four ints and I reference each one with
a constant index 20 times.  And then *once* I read the array with a
non-constant index.  I'd expect to see it scalarized in that case, but
the four elements copied back before the non-constant read.

I can't tell if that's what this code does or not.

It's what happens for a record, right?  An assignment to or from the
whole variable doesn't disable scalarization.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]