This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Entry for the LITERATURE
- To: egcs at cygnus dot com
- Subject: Entry for the LITERATURE
- From: Toon Moene <toon at moene dot indiv dot nluug dot nl>
- Date: Fri, 7 Nov 97 17:09:14 +0100
- Organization: Moene Computational Physics, Maartensdijk, The Netherlands
I just came across this jewel:
*** QUOTE ***
From: metcalf@mail.cern.ch (Mike METCALF)
Organization: CERN - European Organization for Nuclear Research
Subject: XLF - all is revealed
Date: Thu, 6 Nov 1997 14:12:21 GMT
Newsgroups: comp.lang.fortran
There are often discussions/speculations here about how given
compilers optimize source code, and in particular about how they
deal with optimal use of cache memory. I have just noticed a paper
describing exactly how this is done in IBM's xlf compiler, using the
-qhot option, as of version 4. The reference is: IBM J. Res.
Develop., 41, 3, May 1997, 233-264 - Automatic selection of
high-order transformations in the IBM XL Fortran compiler, V.
Sarkar.
Highly recommended.
Mike Metcalf
*** UNQUOTE ***
Mike Metcalf is a long time (i.e. since mid-60's) CERN scientist
who has been on various Fortran Standardisation bodies.
I hope this paper saves me a lot of explaining my `after the
release' wish-list for Fortran specific optimisations :-)
David (Edelsohn), is this paper online somewhere ?
Cheers,
Toon.