Paolo Bonzini wrote:
As for copies, I think it would be a bad decision to stick only to
original (after the code selection) alternative and generate copies to
satisfy this alternative. For example, if pseudo got memory instead of
hard-register required by the alternative, it would be bad to generate a
copy (ld/st in this case) if memory is accepted by the insn.
Yes, spilling is a special case (see also PR19398). If something is
spilled, "prereloads" could be rolled back for affected instructions and
recomputed (possibly forcing no change on the constraints of other
registers).
Ick. This reminds me of the old old LRS code -- while it worked and was
reasonably effective, it was bloody hard to understand and follow (one
of major reasons the code never was contributed). I don't see what
this kind of approach buys us when compared to having IRA deal with
spill code directly.
That is in brief how I see it and there are a lot of reload details
missed (like virtual register eliminations or addressing displacement
constraints etc).
I suppose those would stay in reload?
Ideally they'd all move into IRA. But I don't mind putting them off
until after we have a solid handle on how to deal with the most common
cases.
Jeff
