This is the mail archive of the gcc@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]

Re: Graph coloring for register allocation?


>>>>> "Zack Weinberg" writes:

Zack> (c4x/c4x.c)  When they say "can screw up", they mean "can make reload
Zack> abort."

	Isn't the question whether the new allocator can replace reload,
or at least most of it?  If the new allocator can replace the part of
reload which crashes in this type of case, there is not problem.

	MPY-ADD allocation seems more like carefully obeying the register
classes as opposed to a constant that cannot be loaded directly or a value
ending up in completely the wrong register which requires secondary and
tertiary reloads to fix up.

David

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