home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!usc!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!unixhub!slacvm!dragon
- From: DRAGON@SLACVM.SLAC.STANFORD.EDU (415)
- Newsgroups: comp.unix.aix
- Subject: Re: xlf compiler preprocessors
- Message-ID: <92209.095733DRAGON@SLACVM.SLAC.STANFORD.EDU>
- Date: 27 Jul 92 17:57:33 GMT
- References: <4840@unixhub.SLAC.Stanford.EDU>
- Organization: Stanford Linear Accelerator Center
- Lines: 17
-
- You might take a look at the IBM "redbook" GG24-3611 _Performance Tuning
- for Numerically Intensive FORTRAN and C Programs_. The copy I have is old
- and does not even mention the special preprocessors, -Pk and -Pv. However,
- it does talk about the optimization (-O) and even mentions an undocumented
- feature (-qopt=3) which offers yet more optimization - at a cost of possibly
- giving the wrong answer! If this redbook has been updated for xlf v2.2,
- this would be the place for the explanation you need. Ask Jim Browne, our
- SE, if he can investigate. Also, you might ask him if there is an internal
- IBM "forum" on performance optimization. I have found the real-time forum
- a virtual treasure trove of useful (and undocumented) information.
-
- My guess would be to try the best preprocessor (-Pk ??) AND the -O
- optimization as suggested by info
- "For maximum optimization, the following command line options should be used:
- -Wp, -OPTIMIZE=5, -SCALAROPT=3, -ROUNDOFF=3"
-
- I would be interested to know what you finally come up with!
-