Comment #3 on issue 2513 by [email protected]: Patch: Build: add -fno-tree-vrp
to CXXFLAGS for gcc 4.7.0
http://code.google.com/p/lilypond/issues/detail?id=2513
Taking a look at the gcc bug data base, it would appear that the -ftree-vrp
optimization option has been around for a very long time now. So it is
likely that it is not solely responsible for the problem, but possibly in
interaction with other, possibly newer, optimization options. It is also
likely that this option has noticeable performance impact, given that it
has been included in GCC already a long time ago, and not being removed
since then like other optimization options that have proven not to be
overly effective.
A quick test might be to check the result of
g++ -O1 -ftree-vrp
If this is sufficient in itself for cranking out bad code, I agree that the
proposed fix is likely the best we can hope to do without foregoing even
more important optimizations. If the code delivered by that command is
good, then perhaps there is a different optimization option with possibly
less impact we can switch off.
Of course, this would also be interesting information to deliver in an
upstream bug report. And this should very very definitely be reported
upstream to the GCC bug data base.
