🚀 go-pugleaf

In article <753@h-three.UUCP>, tony@h-three.UUCP (t e bennett) writes:
> 
> I'm looking for info on i960 versions of gcc et.al.

Pardon me if this sounds like a late-night TV ad - I'm just cramming as
much info into as little typing as possible.

[stop here if you object to a mini-ad]

As of Friday, 12/15, Intel is shipping copies of a product with the
part # GNU960T, which can be ordered from your local Intel field sales
office for $250.  For this you get a QIC-11 tape with gcc960 (1.36), gas960,
gld960, gnu binary tools (gnm, gar, etc), gdb960 (based on version 2.8 -
sorry, no more up-to-date version available), and a C library that is
not copylefted, and can be included in normal programs.  Also included
is a retargetable ROM monitor (NINDY), to which gdb960 can talk for
remote debugging.  Several other support programs are provided.  You
also get a 200-page manual (GNU documents, manual pages, release notes,
and other documentation), and a copy of Harbison & Steele's C book,
which documents the library, as well as the C language.

The tape includes binaries for Sun-3, Vax, and System V/386.  Most of
the tools also work on Sun 386i.

Of course, full source code for everything is included.

If you don't want to pay anything, I will be putting some tar files on
willing FTP sites - I had some volunteers some time ago, and I have to go
look them up and ensure that they are still willing.  Of course, with
these you don't get binaries (too big, not a policy), and no printed
manual or Harbison & Steele (a $40 value! :-)).

If people desire it, I can post the machine description to the compiler
and patches to it to this newsgroup, but the changes to the assembler,
linker, and other tools are too major to post, and apply to old versions
that may not be easily accessible.

> I have gcc 1.34 for
> the i960 as well as gcc 1.36 (which does not include the 960 port.  why?).

We were not synch'd with RMS in time for 1.36.  He had his hands full with
other stuff, and we had to make some changes to the base compiler to
accomodate our calling sequence (compatible with Intel's standard
compiler).

We're starting to work with RMS to get the 960 code included in the
1.37 release, but we still have some unresolved issues regarding changes
to the base - we can't accept an incompatible calling sequence with our
other compilers.  Hopefully we'll work these out prior to the next
release.

> Have also seen reference to gcc 1.35 for the i960.
> 
> Is the current 960 work considered too preliminary to include as part of
> the standard distribution?

The current release has been tested very well - we have passed some of the
Plum-Hall tests, and compiled numerous internal test cases.   We will continue
to improve the code, but it's not "preliminary" any more.  The current bugs
are:
	no 'long double' support (80-bit extended precision FP)
	no support for floating-point emulation
	bitfields which attempt to span word boundaries break (base problem)
	no 'long long' support (64-bit integers)
	no instruction scheduling

We hope to correct these in another release in Q1.


For other questions, e-mail to:

	bug-gnu960@mipon2.intel.com

or FAX to

	GNU/960 - (503) 696-4930

sorry, but we can't accept phone calls.

S. McGeady
Intel Corporation