Steel Bank Common Lisp sbcl

Rearrange interpreted TAGBODY parser.

Fixes problem reported by Tito Latini on sbcl-devel.

install.sh: don't create site-systems.

There's no asdf-install anymore.

Speed up GC on ppc and arm Linux.

PPC-Linux had -O3 disabled some twelve years ago due to some bug,
which disabled inlining of GC helper functions, which slowed done
garbage collection by a factor of 25. ARM config files were copied
from PPC config files and inherited the slow down.

Update NEWS for the last several commits worth of MIPS changes

* No further explanation needed.

Enable specific internal error codes for %TYPE-CHECK-ERROR on MIPS

* Now that MIPS is on the newer EMIT-ERROR-BREAK interface, we
can also tie it into MAKE-TYPE-CHECK-FORM and %TYPE-CHECK-ERROR/C.

Convert MIPS to new-style EMIT-ERROR-BREAK interface

* This is basically f33fdd489e9012e5064d35ca7edc7d4bc3c4a0c2
(1.0.28.68), but for MIPS rather than PPC. Just ALPHA and HPPA
remaining on the old style, apparently.

Mark exhaust.impure.lisp / (:EXHAUST :BINDING-STACK) as broken on MIPS

* This probably should be made to work at some point, but for
now let's just make it obvious that it's not expected to work.

Fix alignment issues for 64-bit callback arguments on MIPS

* Another KLUDGE to a broken mechanism: MIPS requires that
64-bit parameters be 64-bit aligned, but the callback logic only
provides 32-bit alignment on 32-bit targets. Compute the amount
of padding required (which will be zero most of the time), apply
it to the offset in the SAP+ forms, and return it so that
subsequent parameters will have their offsets correct.

* Seriously, why can't we have something like the :ARG-TN
methods for callback args?

Allow backends to supply additional displacement for callback arguments

* As the comment says, at least one backend (MIPS, that I'm
aware of) requires the ability to insert additional padding into
the callback argument vector in order to accomodate arguments with
alignment requirements beyond the single word already supplied.

* Allowing an optional second return value as the additional
displacement required for the alignment padding is the least
disruptive mechanism that I could find, though I'm badly
unconvinced about the entire callback implementation at this point.

Add two new callback.impure.lisp test cases

* Testing for alignment of double-machine-word integer values,
at least in theory.