Current Backend Bugs

[Bill.Chiles@CHILES.SLISP.CS.CMU.EDU]

This error aborted two files' compilations:

   Error in function C::FAILED-TO-PACK-LOAD-TN-ERROR.
   Unable to pack a Load-TN in SC RT::SIGNED-REG for the third argument to
   the RT::8BIT-SYSTEM-SET VOP.
   Perhaps all SC elements already in use by VOP?

A quick look made it look like further lifetime hacking would not solve the
problem, and Rob seemed to think he may have to relax the constraint that
arg and result load-tns must be distinct registers ... or something like
that.


Declarations of result types didn't fix these two:
1.
   Error in function C::NO-LOAD-SCS-ALLOWED-BY-PRIMITIVE-TYPE-ERROR.
   #<TN t1[CS3]> is not valid as the first result to VOP:
     RT::MOVE-FROM-WORD/FIXNUM,
   since the TN's primitive type T doesn't allow any of the SCs
   allowed by the operand restriction:
     (RT::ANY-REG)
   Current cost info inconsistent with that in effect at compile time.  Recompile.
   Compilation order may be incorrect.

2.
   Error in function C::NO-LOAD-SCS-ALLOWED-BY-PRIMITIVE-TYPE-ERROR.
   #<TN t1[CS3]> is not valid as the first result to VOP:
     RT::MOVE-FROM-BASE-CHARACTER,
   since the TN's primitive type T doesn't allow any of the SCs
   allowed by the operand restriction:
     (RT::ANY-REG)

Probably Rob has to determine why these are losing.

Other than these two problems, it looks like the float stuff is the only
thing before the RT port.

William, this probably means you should whip up a ldb for the RT.  Since
Rob will probably blow away the float shit this week, I could start working
on building a core at the end of this week or the beginning of next week.

Bill