Re: Code Optimization questions

sanjay@equalizer.cray.com (+Sanjay Krishnamurthy)
Sat, 14 Nov 1992 01:43:32 GMT

          From comp.compilers

Related articles
Code optimization questions cse264ce@cs.ucsd.edu (1992-11-04)
Re: Code optimization questions preston@miranda.cs.rice.edu (1992-11-11)
Re: Code optimization questions Jonathan.Bowen@prg.oxford.ac.uk (1992-11-13)
Re: Code Optimization questions sanjay@equalizer.cray.com (1992-11-14)
Re: Code optimization questions jeremy@sw.oz.au (1992-11-16)
Re: Code optimization questions glew@pdx007.intel.com (1992-11-18)
| List of all articles for this month |

Newsgroups: comp.compilers
From: sanjay@equalizer.cray.com (+Sanjay Krishnamurthy)
Organization: Compilers Central
Date: Sat, 14 Nov 1992 01:43:32 GMT
References: 92-11-015
Keywords: optimize, performance

Steve Boswell <whatis@ucsd.edu> asks:


>[1] Is there any info on optimizers that take advantage of profiler
> information?


You need to look at all the papers (~zillion) on the IMPACT compiler at
the University of Illinois. This compiler bases its optimizations on a
"Superblock" which is basically a special trace. I use "The Superblock: An
Effective Technique for VLIW and Superscalar compilation," Hwu et al, to
appear in Journal of Supercomputing, 1993, as my primary reference. You
should be able to get a bunch of tech. reports from hwu@crhc.uiuc.edu
describing the use of profiling info to do everything from procedure
inlining to scheduling.


>[2] Is there any info on optimizers that take advantage of runtime
> assertions? This can help with anti dependencies,......


There's a paper titled "Run time disambiguation" written by Alex Nicolau
which appeared in the IEEE Transactions on Computers sometime in the past
4 years (very helpful, eh??) You should look at this to find out how
static dependence analysis can be aided with runtime assertions.


>[3] If statement A has a def-use chain to statement B, does it follow
> that statement B has a use-def chain to statement A? Is this an
> "if and only if" relation?


You mean "Is this a symmetric relation?" Sure appears to me.


-Sanjay M. Krishnamurthy
Cray Research Superservers
--


Post a followup to this message

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