From: | Christopher Glaeser <cdg@nullstone.com> |
Newsgroups: | comp.compilers |
Date: | 17 May 1997 22:55:49 -0400 |
Organization: | Nullstone Corporation |
References: | 97-05-183 97-05-195 |
Keywords: | C, benchmarks, errors |
David Chase wrote:
> then you can try to get the largest and most ugly of
> the codes your "compiler" generates incorporated into an
> "industry-standard benchmark".
This will ensure your program will run very fast, but will not ensure
correct answers, and may introduce defects. As an example, in the
race for the best possible performance, at least five major suppliers
of chips, computer systems, and compilers generated incorrect code the
SPEC EQNTOTT benchmark. See
http://www.nullstone.com/eqntott/eqntott.htm for additional
information.
It should be noted that all of these defects were introduced into
these compilers as a direct result of trying to improve EQNTOTT
performance. In addition, some compilers still exhibit these defects.
In other words, if you are actually trying to use the EQNTOTT program
(in contrast to just timing it) you may want to revalidate the program
each time you use a different compiler.
Best regards,
Christopher Glaeser cdg@nullstone.com
Nullstone Corporation http://www.nullstone.com
--
Return to the
comp.compilers page.
Search the
comp.compilers archives again.