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]

Re: Fix for gcc.c-torture/special/920520-1.c on powerpc



  In message <199809050325.NAA00528@geoffk.wattle.id.au>you write:
  > > Another possibility would be to see if tweaking PREFERRED_RELOAD_CLASS
  > > would avoid the problem to start with by having reload put the constant
  > > into memory for us.
  > 
  > But you don't want the constant in memory!
But I doubt the cases where it matters are all that important.

We're talking about getting an SFmode value into a GPR.  That shouldn't
be happening all that often.

If a simple solution like PREFERRED_RELOAD_CLASS can be used to make
things work, with (I suspect) an unmeasureable difference in performance
then it should be *considered*.

If you look at the way the PA did it in that one pattern, we didn't
force them to memory when loading in a GPR.  I still consider that
pattern a hack, but it is effective and should be considered as another
potential solution.

jeff


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