Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: About compiler optimizations

Author: Dave Gomboc

Date: 19:23:36 12/19/02

Go up one level in this thread


Well, they bought out KAI C++, presumably that had a big effect.  KAI had a
really strong reputation for quality, and AFAIK Kuch (the 'K' in KAI :-) manages
their compiler division now.

Dave


On December 19, 2002 at 13:03:10, Eugene Nalimov wrote:

>I cannot be very objective here, as I am working on the directly competing
>product, but previous versions of Intel compiler were *really* buggy. Compiler
>choked on complex integer code regularly. I think it behaved better on FP code
>because one FP code resembles other FP code much more than one integer program
>resembles other integer program.
>
>Some time ago they started to pay attention not only for the speed of the code
>they emitted, but for compiler stability as well, so recent versions were much
>better in that respect.
>
>Not sure what the status is now, and don't want to speculate.
>
>Thanks,
>Eugene
>
>On December 19, 2002 at 12:29:09, Robert Hyatt wrote:
>
>>On December 19, 2002 at 12:20:55, Gian-Carlo Pascutto wrote:
>>
>>>On December 19, 2002 at 11:43:22, Robert Hyatt wrote:
>>>
>>>>Also he should try the Intel C/C++ compiler v 7.0 download from Intel's
>>>>web site.  It produces faster code for me than any version of gcc that is
>>>>available.
>>>
>>>It's also generally buggy for programs that are not in the SPEC
>>>benchmark set.
>>>
>>>--
>>>GCP
>>
>>
>>That is complete horse-do-do.  We are using this compiler on at least
>>a dozen production programs here.  And _none_ of those programs are in
>>SPEC.  And they have _all_ been validated with large test data sets to
>>confirm they produce identical results with gcc and even with some Sun
>>workstations.
>>
>>That is a vincent-like statement that is simply _false_.
>>
>>I'll be happy to give you the names of people here using that compiler on
>>my cluster and on machines outside my cluster, that will verify that they
>>have tested it carefully...
>>
>>It would make no sense for Intel to release something that doesn't work.



This page took 0 seconds to execute

Last modified: Thu, 15 Apr 21 08:11:13 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.