Path: utzoo!utgpu!news-server.csri.toronto.edu!rutgers!apple!decwrl!sgi!calcite!vjs
From: vjs@calcite.UUCP (Vernon Schryver)
Newsgroups: comp.sys.intel
Subject: Re: Intel-8086/80186-Assembler for System-V available ?
Message-ID: <96@calcite.UUCP>
Date: 13 Oct 90 08:23:05 GMT
References: <:C96DR1@xds13.ferranti.com> <15919@bfmny0.BFM.COM> <1990Oct11.084750.1183@orfeo.radig.de>
Organization: Rhyolite Software, Mountain View, CA
Lines: 25


Maybe PLM*86 would run faster as a native UNIX program.  Who cares?  It
would generate the same code.

PLM86 was mind boggling bad even at peep-hole optimizing in MDS-311 days.
(MDS-311 was the ISIS 8085 cross development package you could get for your
very own low speed 8086 in 1978.)  The "code" generated by PLM386 is not
significantly different, at least in the versions I've used.  PLM386 still
wastes registers and cycles with abandon--as if the *86 had any to spare.
It still cannot strength reduce multiplies of constant powers of two.  It
usually ignores and so recomputes condition codes generated by preceding
statements.  And so on and on.

Besides, the beasts that are the slowest are ASM386 and BLD386, at least
under VP/ix and DOSMerge.  It takes longer to link the stuff I do than it
does to compile or assemble it, despite having stripped symbols in
preceding BND386 passes.  The assembler seems slower, in lines/hour, than
the compiler.

Anyone with freedom of choice (i.e. no dusty albatrosses) would choose
almost any C compiler.



Vernon Schryver,     vjs@calcite.uucp