GCC Compile-Time Regressions

[Karel Gardas]

> Currently, there is some controversy as to whether 3.4/TOT is actually
> slower than 3.3.  My measurements
> consistently confirm this thesis, although some folks apparently found
> the reverse to be true.  A suggestion
> has been made (on this list, I believe) that 3.4 may be slower on small
> files but faster on big ones.  So,
> is my test case considered small or large? The structparser.cpp file
> itself is indeed very small, but it
> pulls in a lot of Qt and STL gunk via #include and balloons up to 1 Mb,
> which I would no longer consider
> small.  So, if someone (other than me) could do some systematic
> 3.3-vs-3.4 measurements and post the results
> here, that would be great.
>

Interesting results. I've found exactly the oposite for MICO sources:
http://gcc.gnu.org/ml/gcc/2003-02/msg00800.html

Anyway, for hello world like C++ example is gcc3.4 really a bit slower
than gcc3.3:

gcc3.3: 1.76s
gcc3.4: 1.85s

What surprised me is that you've found gcc3.3 faster than gcc3.4 while
compiling Qt app. My quess: Qt is not using _much_ template code, while
MICO is using templates quite lot. If this is true, then gcc3.4 will be
faster on templatized code, while gcc3.3 on simple class/struct code...

I agree that this is just speculation, maybe some machine generated test
file (big) will be good here...

Cheers,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com