This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH]: Reduce memory usage and compile time necessary for pruning
On Mon, 2006-03-13 at 10:16 -0500, Diego Novillo wrote:
> On 03/12/06 16:29, Daniel Berlin wrote:
>
> > Any idea what to do here?
> >
> Not really. Introducing another temporary might be tempting, but we're
> already temporary-happy, and these are fairly bulky objects.
>
> With any luck, any workaround you implement will be temporary. The
> whole tag-is-used-alone thing will not be needed with the new scheme.
> How about a bit in the virtual operand itself to indicate whether it's
> an LHS or an RHS V_MAY_DEF?
This is my inclination, but it would have to be two bits, because it
could be because of both.
In fact, in this case, it *is*, because the SMT is def'd by both the
call, *and* the store on the LHS.
We just don't add two may defs because we don't add duplicate may-defs.
I'm actually going to see how bad things are as a temporary hack if we
1. try to determine what tags the lhs may def in this special case of
having a call store to something that may-def things.
or
2. Try to take the conservative route of marking may-defs from calls
when the lhs may generate may-defs (the problem with this is that it may
cause random changes to call statements to need used_alone
recalculation, so it may not work).
If it becomes ugly at all, i will add some bits :)