This is the mail archive of the gcc-patches@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: PR 30194: temporary workaround


Daniel Berlin wrote on 12/14/06 15:42:

I assume you are thinking of two things here:

Many more than 2, actually.

1. Prioritze putting call clobbered vars into partitions ahead of putting non-call clobbered vars into partitions 2. Put call
clobbered vars into call clobbered partitions and non-call clobbered
vars into non-call clobbered partitions, to avoid adding more call
clobbering than necessary.


We already do #1.  #2 is possible.  Another is see what variables are
actually referenced on their own and probably *where*.  Ad nauseaum.
Any static heuristic will be easily beat by one testcase or another.

Long term, I will experiment to get closer to dynamic partitioning.  But
for now, there are many tweaks that can be done on the static partitioner.


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