Re: Newbie SSA question

Anthony PIRON <apiron@ulb.ac.be>
1 Feb 2001 17:38:56 -0500

          From comp.compilers

Related articles
Newbie SSA question Kevin.Tucker@Microchip.com (2001-01-26)
Re: Newbie SSA question crmorgan@venus.star.net (Bob Morgan) (2001-02-01)
Re: Newbie SSA question apiron@ulb.ac.be (Anthony PIRON) (2001-02-01)
Re: Newbie SSA question Martin.Ward@durham.ac.uk (2001-02-01)
Re: Newbie SSA question pmk@visi.com (2001-02-12)
| List of all articles for this month |
From: Anthony PIRON <apiron@ulb.ac.be>
Newsgroups: comp.compilers
Date: 1 Feb 2001 17:38:56 -0500
Organization: VUB/ULB Computing Centre
References: 01-01-147
Keywords: analysis
Posted-Date: 01 Feb 2001 17:38:56 EST

Hi,


> If the latter, then converting out of SSA form would mean
> simply "dropping the subscripts", as it were.


No, it's not simply dropping the subscript because optimisation passes
can modify the well constructed initial structure of SSA. See for more
information: Briggs, P; Cooper K Harvey, T; Simpson, T 1999. Practical
Improvements to the Construction and destruction of Static Single
Assignment. (I found this online but I cannot remember the url, probably
the Cooper Keith or Briggs Preston homepage)


The SSA is used to simplify and to speed up the optimisation passes. The
SSA give explicitly the USE-DEF relation. More the USE-DEF relation is
factored thanks to the phi nodes. This factorisation drops the complexity
of many optimisations from quadratic complexity to linear complexity.
  The variables renaming is also very useful for instructions scheduling,
for example :
        a <- 1
                <- a
        a <- 2
Become
        a1 <-1
                    <- a1
          a2 <- 2
You will see there is no more write write conflict and the 2 a
assignations can be parallelized.


SSA have got many more useful properties. You can find more references on
my web page http://www.ulb.ac.be/di/ssd/apiron/researchsection.html. My
MS thesis (sorry in french) was about sparse evaluation. There is the
bibliography online.




--
Anthony PIRON - Researcher
Universite Libre de Bruxelles, Department of Computer Science
CP 212 - Boulevard du Triomphe, B-1050 Bruxelles, Belgium
Tel. +32 2 650 50 55, Fax +32 2 650 56 09, GSM +32 478 530 022
Email: apiron@ulb.ac.be Web: http://www.ulb.ac.be/di/ssd/apiron


Post a followup to this message

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