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] peephole changing a move to itself with compare to a compare


Yes that worked, thanks,
I will be sending the new patches for {add,sub}disi, only problem I had was
I needed have zero_extendsidi2 be defined unconditional and have a split for
it when not TARGET_POWERPC64.


I have a question about your patch, should the constraint of the 2nd operand
of the first alternative should be `y' so that it is not bias for against using
other cr's than cr0?


Thanks,
Andrew Pinski

On Tuesday, May 13, 2003, at 16:34 US/Eastern, David Edelsohn wrote:

The poor code should not be fixed with a peephole, although GCC
should not generate a compare by moving to itself. If the common part of
the compiler is going to present the backend with this type of RTL, the
backend should choose better code from the beginning, like the appended
patch. Let me know if the patch works for you, it does in my test of your
example.


	Also, I agree with Richard that the {add,sub}disi peepholes should
be combiner patterns.  Please send revised patches for those two.

Thanks, David



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