Re: [tree-ssa, LNO] Analysis of scalar evolutions and data dependences

[law at redhat dot com]

In message <1071617117.28689.150.camel@frodo.toronto.redhat.com>, Diego Novillo
 writes:
 >On Fri, 2003-12-12 at 06:11, Pop Sébastian wrote:
 >> Second hunk that is a good candidate for going into tree-ssa.
 >> 
 >> In some cases it is simpler to expect that the whole dump file
 >> matches.  For this you save a copy of the expected output in the same
 >> directory as the testcase.  
 >> 
 >> I find this simpler to use and maintain testcases than writing long
 >> regexps with lots of backslashes.
 >> 
 >Hmm, but this is very brittle wrt changes in the formatting of the
 >dump.  The regexp match has to be done once and is more localized.  This
 >approach has the potential of forcing people to change a whole file at
 >every minor change in the dump.
 >
 >I'm not sure I like the idea.  Thoughts?
It's also brittle in that it's tied to what version # variables get, what
address temporaries get and the like.  Those are the primary reasons I
didn't go that route when starting to build out the tree-ssa optimization
tests.

After working with the tree-ssa optimization tests for a while some
of their limitations are clearer.  The most glaring one is that we 
can't easily do things like code motion tests.

jeff