[PATCH] sparc64-linux multilibs]

[Dan Nicolaescu]

Richard Henderson wrote:

  > On Tue, Dec 07, 1999 at 12:29:23PM +0100, Jakub Jelinek wrote:
  > 
  > > That would mean 6 different types of multilibs (64bit, 64bit
  > > no-app-regs, 64bit cmodel != medlow, 64bit no-appregs + cmodel !=
  > > medlow, 32bit, 32bit no-app-regs).  That seemed too much for me, so
  > > I have modified a little bit genmultilib to accept a new syntax and
  > > now I can have just 4 types of multilibs: (64bit, 64bit with
  > > no-app-regs or cmodel != medlow or both, 32bit, 32bit no-app-regs).
  >    
  > Is there any point to building no-app-regs versions of libgcc?
  >    
  > The C++ libraries I would expect most folks to use shared objects;
  > we'd do the same for fortran if we put the effort in.  And for   
  > shared objects we must of course use -mapp-regs.
   

I think at least _bb.o from libgcc.a should be built with
-mno-app-regs

Some of the __bb_* functions are not allowed to change the state of
the processor, if they use they use the app-regs then the
MACHINE_SAVE_STATE macro hould save those registers too.

What is the allowed use  of app-regs anywhay? At some point I was
looking for some info about this, but I didn't find anything...

--dan