martin@mira.isdn.cs.tu-berlin.de

["David M. Ronis" <ronis at kirkwood dot chem dot mcgill dot ca>]

Hi 

   I too noticed a similar slowdown in a number crunching code, and
for the longest while, I thought it might have something to do with
glibc2.  I no longer believe that (I've posted some of this on the
gcc-lists), rather, there is a serious stack alignment bug that was
posted to the gcc-bugs list by Steve Johnson (one of the fftw authors).

Basically, things get messed up when you have multiple function calls,
e.g, in f(g(x),...)  g's stack won't be aligned.  Steve's last post
even has a simple test program and some ideas where the problem is
happening ing gcc, although none of the compiler folks have said
anything.


David