Path: utzoo!mnetor!uunet!husc6!mit-eddie!killer!chasm
From: chasm@killer.UUCP (Charles Marslett)
Newsgroups: comp.arch
Subject: Re: RISC is a nasty no-no!
Message-ID: <3723@killer.UUCP>
Date: 17 Mar 88 14:10:44 GMT
References: <179@wsccs.UUCP: <696@nuchat.UUCP> <284@scdpyr.UUCP> <641@bnr-rsc.UUCP>
Organization: The Unix(R) Connection, Dallas, Texas
Lines: 32
Summary: \"register\" is not . . .

In article <641@bnr-rsc.UUCP>, crick@bnr-rsc.UUCP (Bill Crick) writes:
> The comment aboutgood programmers putting the array pointers in 
> registers is probably moot. Don't most modern(?) compilers ignore
> the register directive and do their own allocation?


If my experience is any indication, register usage in modern micro C compilers
is drifting toward the the golden ideal of 1970-era Fortran compilers (IBM's
Fortran H for example) -- Microsoft C 5.1 seems to completly ignore register
declarations except for generating some warnings if they are used incorrectly
and earlier beta copies of 5.1 seemed to have real problems with having more
than 4 or 5 register declarations (couldn't juggle its own register allocation
and the programer's too!).  At least one of the 68000 compilers I used a
couple of years ago also had the characteristic that if it needed a register
(for its optimization) it would take it from your set of register variables
and might (sometimes) let you use the register elsewhere in the code.  This
is a much harder task for the 68000 since there are more candidates for
register variables and optimization sequences.

I guess my point is, if the demand for good code generation from compilers
wins out over fast debugging cycles for compiler company resources, we
may reach the stage where the author of the program need not worry about
the architecture of the underlying machine when writing code because the
compiler writer is optimizing generic-CPU code well enough that he can
ignore the application writer's optimizations.  I hope we get there before
everyone gets sidetracked to more "interesting" or "marketable" features
in compilers -- whatever they might be in 1992.

                                         ========================
Charles Marslett                         |Said the iconoclast:  |
chasm@killer.UUCP                        | This is better?      |
                                          ========================