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: Patch [rs6000]: support FP modes in CTR/LR


On Feb 20, 2004, at 4:36 PM, Geoff Keating wrote:
Dale Johannesen <dalej@apple.com> writes:
Note that the CTR-to-LR case is not handled by any of the patterns, and
can't be without a temporary somewhere. What I'd like to do is
remove support
for keeping FP objects in CTR/LR (it looks like changing
HARD_REGNO_MODE_OK
is sufficient); while it's possible to construct cases where this is
useful, most of the
time the code is going to be really bad. Does that seem like a
reasonable idea?

You will also need to change CANNOT_CHANGE_MODE_CLASS, I think.


It doesn't seem like an unreasonable idea, but if it's really so bad
then the register allocator shouldn't be doing it in the first place;
could you try to work out what's happening?

I agree. I think the problem is that REGISTER_MOVE_COST is too
generous for this case, but I'd rather disable it than try to improve things there.



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