sparc.c changes

[Richard Henderson]

Sorry I dallied so long at looking at the changes you sent before.
I committed some of the work, but not all:

 - The Haifa scheduling bits for ultra are in.

 - The VIS bits are in.  At some point it would be nice to fix up the
   *_sp64 insns with the VIS alternatives as well.

 - The ldd deprication is in.

 - Most of the little cleanups are in as well.

What didn't go:

 - The I64_REG stuff.  I don't understand where Sun is saving the high
   bits across signals.  There is an ABI in the form of ucontext_t for
   modifying a register and continuing.  Supposing that the high bits
   are saved elsewhere, how do they know which register set to restore?
   Anyway, I'm nervous about this issue despite the fact that a simple
   test with SIGALRM did succeed on sol5.5.1.

 - V9 SImode/FP conversion.  I think I sent you a test case for this.

 - The return bits aren't in, partly because it was hard to distinguish
   them at a glance fro the I64_REG bits, and because I'm now tired. :-/
   They do look quite useful though -- if you resend them I'll commit them.

I sanity checked with a cross-build from my Alpha, so it's not even
gotten a foll bootstrap.  I'll start one here in a minute, but I'm
not going to wait up for it.


r~