Path: utzoo!mnetor!tmsoft!torsqnt!news-server.csri.toronto.edu!cs.utexas.edu!uunet!orca!javelin.es.com!bgeer
From: bgeer@javelin.es.com (Bob Geer)
Newsgroups: comp.dsp
Subject: Re: C cross compiler's for TMS320XX chips
Message-ID: <1991Feb13.140924.23433@javelin.es.com>
Date: 13 Feb 91 14:09:24 GMT
References: <1991Feb12.155138.5030@dgbt.doc.ca>
Reply-To: bgeer%javelin@dsd.es.com
Distribution: na
Organization: Evans & Sutherland Computer Corp., Salt Lake City, Utah
Lines: 20

don@dgbt.doc.ca (Donald McLachlan) writes:
>We have used TI's TMS320C25 C compiler, and are not impressed with the
>code it produces. Often hand assembled code is 8 time better, on average
>3. In CISC environments I always heard a good C compiler would give
>at most 50% overhead. What are reasonable numbers for RISC or DSP chips.

>If anyone can suggest a better C cross compiler for the TMS320XX family
>we would certainly like to hear about it.

Version 4.0 of TI's TMS320C30 C compiler has a much better optimizer
than vers. 3; it makes fuller use of registers & improves inner loop
code quite a bit.  Cursory measurements I've done show marked
improvement in our code but don't have publishable numbers yet.  I
don't know if TI has improved the _C25 compiler's optimizer as well,
but it would be worth a look-see.
-- 
<> Bob `Bear' Geer <>               bgeer@javelin.sim.es.com              <>
<>      Alta-holic <>   speaking only for myself, one of my many tricks   <>
<> Salt Lake City, <>    "We must strive to be more than we are, Lal."    <>
<>          Ootah  <>           -- Cmdr. Data, learning schmaltz          <>