Re: saving cycles?? New Tech?

diablovision@yahoo.com
14 Dec 2006 17:28:15 -0500

          From comp.compilers

Related articles
saving cycles?? New Tech? kumarssss@gmail.com (Kumar) (2006-12-11)
Re: saving cycles?? New Tech? diablovision@yahoo.com (2006-12-14)
| List of all articles for this month |

From: diablovision@yahoo.com
Newsgroups: comp.compilers
Date: 14 Dec 2006 17:28:15 -0500
Organization: Compilers Central
References: 06-12-048
Keywords: optimize
Posted-Date: 14 Dec 2006 17:28:15 EST

> Are there new optimization available other than ,,, Loop Unrolling,
> Loop peeling, scalar expansion, strip minig, loop unswitching.
>
> So that I can save the number of cycles by doing optimization.
> Can anybody tell me the technique that is going to large of cycles?


This is a bit like asking a group of English professors, "are there
any other authors besides Steinbeck, Dickens, Dostoevsky, and
Shakespeare that I should know about?"


Why yes, yes there are. The compiler field is huge--at this point,
there probably isn't a single person that knows more than an
appreciable fraction of what there is to know about
optimization. There are a handful of good books on compilers, and at
least 3 major academic conferences that are still full of optimization
and runtime tuning ideas. Look at the proceedings of PLDI, CC, LCTES,
CGO, etc for the past N years.


Before you rush out to study a bunch of irrelevant and esoteric
research, perhaps you could focus your problem a bit better. If there
is an application that you are interested in optimizing, or a class of
applications, perhaps you should try using a good profiler to
determine where the hot spots in the code are, and focusing
optimization efforts on those? Perhaps there are better worst-case
algorithms for the core of the problem, and focusing on compiler
optimizations is premature?


The open-endedness of your question makes it hard for anyone to pin
down just what it is you are after.


Post a followup to this message

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