Re: Compiler Verification/Optimization

johnrn@attglobal.net (john )
26 Apr 2000 02:46:21 -0400

          From comp.compilers

Related articles
Compiler Verification/Optimization jared@zucotto.com (Jared Dykstra) (2000-04-20)
Re: Compiler Verification/Optimization naveen_sharma@my-deja.com (2000-04-21)
Re: Compiler Verification/Optimization vmakarov@cygnus.com (Vladimir Makarov) (2000-04-25)
Re: Compiler Verification/Optimization johnrn@attglobal.net (2000-04-26)
| List of all articles for this month |

From: johnrn@attglobal.net (john )
Newsgroups: comp.compilers
Date: 26 Apr 2000 02:46:21 -0400
Organization: Global Network Services - Remote Access Mail & News Services
References: 00-04-143 00-04-155
Keywords: optimize, comment

My remarks concern Fortran compilers.


I have a few comments about your comments about optimization.
Optimizing compilers ALWAYS take more time to compile. But most
scientists who use Fortran don't care because they want their compiled
code to run very fast. Some productions systems run for 10 or more
hours continuosly and any gain in speed is very important.


Your concern about safe optimizations is misplaced. There are many
optimizations such as common expression elimination, strength
reduction, branch optimization, and moving non-loop dependent
calculations out of a loop have been proven to be quite safe.


John




On 21 Apr 2000 23:04:23 -0400, naveen_sharma@my-deja.com wrote:
> ...
>Optimization is another tricky business.There can be highly
>optimizing compilers but I think they impose a penality in terms
>of greater compilation times.Personally I feel compiler
>should do only safe optimizations because verifying that an
>optimization had no side effects is a painful exercise.
>
>Perhaps, I would be able to share more information if the precise
>context could be given.
[Actually, I've heard of optimizations that shortened compile time, e.g.
dead code elimination that removed code that would otherwise have to be
examined in later compiler phases, but it's true that optimizers tend to
be slow. On the other hand, it's also true that the vast majority of
compilations are for debugging, not production, in which case the speed
of the object code rarely matters. -John]


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.