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: Target FUNCTION_{PRO,EPI}LOGUE question


On Fri, 29 Jun 2001, Fergus Henderson wrote:

> Would doing the following steps be sufficient to solve the problem(s)?

Since we would want to get rid of the signed/unsigned/precision issues
that keep coming up (values getting the wrong one of sign-extended and
zero-extended), I don't think so - we really need a general integer
representation that keeps track of sign and value.  Conceivably the parts
of the exported interface that get used by the rest of the compiler could
be implemented using a fixed-width representation (if the rest of the
compiler only ever needs modulo arithmetic), but unless use of multiple
precision causes performance problems, I think defining clean interfaces
that operate on general integers, and making the rest of the compiler not
care about the internals of the representation, would be better.

-- 
Joseph S. Myers
jsm28@cam.ac.uk


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