Re: [PATCH] Complex MEM adjust_address instead of subreg (PR middle-end/21742)

[David Edelsohn <dje at watson dot ibm dot com>]

>>>>> Richard Henderson writes:

Richard> By performing the transformation and testing to see whether
Richard> the resulting address is valid.

Richard> Of course, this does mean that in the context of complex move
Richard> expansion we must have a fallback in case the new address isn't
Richard> valid.  Which will always be true on ia64.

Richard> Which means that the easiest solution is your first patch.
Richard> But we should update the comments to include all this fine
Richard> information.

	Just to be clear:

	You want this fixed in {read,write}_complex_part(), not loosening
the restriction for the MEM case in simplify_subreg(), right?

	You want the updated patch to use adjust_address_nv(), allowed to
fail, falling back to {extract,store}_bit_field(), right?

Thanks, David