Re: [PATCH, middle-end] Improve num_sign_bit_copies when ANDing with a const

[Adam Nemet <anemet at caviumnetworks dot com>]

Mark Mitchell writes:
> (It's lame that there are HOST_WIDE_INT checks in this code; clearly,
> the host configuration has no impact on the validity of this
> optimization.  But, I know the whole function is like this, so what
> you're doing is in keeping with that.)

Thanks for the review.

I am not sure those checks really prevent any optimization but I might
be missing something.  For example if TImode constants cannot be
represented on the host machine then I don't think it is valid to have
something like:

  (and:TI (reg:TI ...) (const_int ...))

since a CONST_INT is represented as a HOST_WIDE_INT.

Adam