C compilers for the 3B2

[Stephen J. Friedl]

Hi net.people,

     A number of recent postings have exhibited confusion about
the various C compilers available for the 3B2, so perhaps this
discussion will clear some things up.  We have been dealing with
these things for a couple of years and have been fairly successful
in tracking down information.  I have all four compilers mentioned
below immediately available to me, but some of these comments
are speculation; rebuttals are welcome.

     You can find out what kind of compiler you have by compiling
a small program with the "-V" flag.  I use:

        $ cat > main.c
        main() {}
        ^D
        $ cc -V main.c

     For each compiler we'll show the output from the above.
Also, our experience is with the 3B2 and 3B5 only, but some of
these may apply to the 3B15 as well.

C Issue 2
=========

     Also know as "pcc2", this has long been the standard C
compiler for the 3B2 and 3B5.  It is an old compiler with no MAU
(Math Accellerator Unit) support, so floating-point gets a very,
very slow (but very clever) interpreter.  Output from the
compiler is:

        $ cc -V main.c
        : command -System V Release 2.0.11.0 07/30/84
        WE32000 pcc2 compiler: @(#)local.c      SVR2
        : assembler - System V Release 2.0.11.0 07/30/84
        ld : System V Release 2.0.11.0 07/30/84 : Version 1.6: 2/27/84

C Issue 3
=========

     We know the least about this compiler so there is a little
more speculation here than for the rest.  This was the first
compiler to support a MAU.  Floating-point operations compile
down to function calls (_fpmul, _fpadd, etc.) that check for the
existance of the MAU: it is used if available, else the
interpreter is loaded.  This compiler is also known as "qcc".
Output from cc -V:

        $ cc -V main.c
        cc: command -C Issue 3.1 11/19/85
        WE32000 qcc compiler: SVR2.1.0
        as: assembler - C Issue 3.1 11/19/85
        ld : C Issue 3.1 11/19/85 : Version 1.8.1.3: 11/19/85

C-FP+
=====

     This is the highest performance floating point possible on a
3B2.  While the compiler can run on any machine, the resultant
code requires a MAU to execute.  Rather than generate the
floating-point function calls as above, direct math coprocessor
instructions are issued for absolute maximum speed.   This
compiler is also known as "rcc", and the compiler's default name
is "fpcc" rather than "cc".

        $ fpcc -V main.c
        cc: command -C-FP+ Release 1.0 12/2/85
        WE 32106 rcc compiler: C-FP+ Issue 0.1
        as: assembler - C Issue 3.1 11/19/85
        ld : C-FP+ Release 1.0 12/2/85 : Version 1.8: 1/3/85

C Issue 4.1
===========

     This is the latest (that I know of) compiler for the 3B2,
and it is similar to C Issue 3.  The current release is called "C
Programming Utilities Issue 4.1" and it goes ith "Advanced
Programming Utilities Issue 1.1".  It appears to handle
floating-point the same way as Issue 3 (with function calls) but
also supports a little more debugging info for sdb, shared
libraries, and probably more things that I don't know.  We
usually refer to this as "CPLU 4.1".

        $ cc -V main.c cc: command -C Compilation System,
        Issue 4.1 1/29/87 C Compilation System, Issue 4.1 1/29/87
        as: assembler - C Compilation System, Issue 4.1 1/29/87
        ld : C Compilation System, Issue 4.1 1/29/87

---------

     More random thoughts.  Even through floating-point is done
quite differently between Issue3/4, C-FP+ and pcc, older
interpreter-based object modules can be used with newer MAU-based
code.  This is helpful when linking with (say) Informix libraries
that have not been recompiled by the vendor.

     I have other thoughts on pcc2's floating-point interpreter
that I will post when I can put them together.  They do it in a
very, uh, um, er, "interesting" manner and you may be amused.
Try to get into C Issue 4 if you can because floating-point is
much faster even without a MAU, and C-FP+ is 237 times faster
than pcc2 on a 3B5 with a MAU.

     Another note: if you have System V Release 3, you need CPLU
4.1 to get SVR3-specific syscalls and library functions like
mkdir(2) and the new signal handler.  These capabilities exist in
the kernel, but the little interface functions to get there are
not included in the "older" libraries.  I don't think that
CPLU4.1 was out when Sys V Rel 3.0 was released, but with SVR3.1
it becaome available.  It is possible that AT&T will retrofit the
older compilers (possibly C-FP+) with the new libraries but you
never know.  A similar OS/compiler difference occurs with
#include header files but Unix provides a disk with these
important files so the problem is not so bad.

------

     I hope this helps, send mail this way for more questions.  I
can probably track down PECs (Price Element Codes) for these
things so you can order them from your Friendly Neighborhood AT&T
person.

-- 
Steve Friedl   V-Systems, Inc.   "Yes, I'm jeff at unh's brother"
friedl at vsi.com  {backbones}!vsi.com!friedl  attmail!vsi!friedl