Re: Solaris 2.7 and 64-bit compiling

[David Miller <davem at twiddle dot net>]

   Date: Sat, 10 Apr 1999 23:58:24 -0700
   From: Richard Henderson <rth@cygnus.com>

   The only big problem at the moment is the subreg thing right?

And also fixing up all of the quad fp routines to pass args in the way
v9 wants, remember all of the "passing pointers to libcall" stuff we
were talking about last time I was looking into all of this?

Also, the "passing/returning small struct/union in registers"
mechanism was removed from the final V9/ELF specs I believe, and our
code needs to be updated to remove this stuff.

Then there will be a slew of bugs to deal with.

   It would be possible to disable this problem for the time 
   being by disallowing MODE_INT values in FP regs.  This will
   of course result in code not nearly as nice as possible,
   but it would produce a working compiler.

I don't believe this would be nearly enough to get a working
compiler.   I think even with this hack it would take a few weeks of
full time debugging just to get a bootstrap using sunpro to build the
first stage.  This is how long it took me to stabilize the sparc64
target that far last time I had to do it.

   Shall I put together such a patch?

Feel free, but as stated, I anticipate that a lot more work will be
needed to get sparc64 to bootstrap, much less be reliable.

I would not feel comfortable at all claiming that sparc64 is a
supported target in the next egcs release.

Later,
David S. Miller
davem@redhat.com