Path: utzoo!attcan!uunet!cs.utexas.edu!tut.cis.ohio-state.edu!rutgers!cbmvax!martin
From: martin@cbmvax.commodore.com (Martin Hunt)
Newsgroups: comp.sys.amiga
Subject: Re: Lattice C questions.
Message-ID: <13334@cbmvax.commodore.com>
Date: 20 Jul 90 16:05:06 GMT
References: <1083@orange9.qtp.ufl.edu> <13270@cbmvax.commodore.com> <3448@ssc-vax.UUCP>
Reply-To: martin@cbmvax (Martin Hunt)
Distribution: na
Organization: Commodore, West Chester, PA
Lines: 27

In article <3448@ssc-vax.UUCP> coy@ssc-vax.UUCP (Stephen B Coy) writes:
>In article <13270@cbmvax.commodore.com>, martin@cbmvax.commodore.com (Martin Hunt) writes:
>> >* 5 Packard-Bell 386SX w/out /387	 127.39		 0.69
>> >*   16 MHz floating point emulation
>> Amiga 500                                146              .79
>> 
>> It is rather interesting that an Amiga 500 runs this benchmark faster
>> than a 386SX.  
>
>Since the benchmark is heavily floating point oriented I'd think
>that the results would depend on the floating point library used.
>Was the Amiga benchmark compiled for IEEE or FFP?  Microsoft C also
>offers an alternate floating point library which gives better speed
>at the cost of accuaracy.  Without this information the benchmarks
>are just noise.
>
Most benchmarks are just noise.  I should have put a :^) at the end of
my post.  This benchmark is very dependent on what compilers
and math libraries are used.

I used double precision IEEE libraries, of course.  FFP is too inaccurate
for scientific work.  In 2.0 we have a single precision IEEE library
that is as fast as FFP, but more accurate.

-- 
Martin Hunt                     martin@cbmvax.commodore.com
Commodore-Amiga Engineering     {uunet|pyramid|rutgers}!cbmvax!martin