Re: Testsuite failure ?

[David Ronis <ronis at montroll dot chem dot mcgill dot ca>]

OK, I'm convinced, I just did a complete bootstrap configure/build
with CFLAGS, CXXFLAGS, BOOT_CFLAGS set to:

-O3 -march=i686 -funroll-loops -ffast-math -fomit-frame-pointer
 -fforce-mem -fforce-addr -malign-jumps=2 -malign-functions=2
 -malign-loops=2

This the same as what was giving me problems, except for the
-malign-double flag.  I now get the same testsuite results as what I
get with the default flags. (The results are posted at:

     http://gcc.gnu.org/ml/gcc-testresults/2000-08/msg00229.html

). 

So the upshot of all this is that -malign-double is bad while the
other flags are OK.

Given that -malign-double is so "evil" (Jeff Law), why have it as an
option at all?  Clearly it must be worthwhile and safe in some cases?
For example, how about a fortran code that doesn't have structures?
I'm sure I've compiled 3rd-party libraries (usually fortran) that had
it enabled in the default makefile.  Perhaps the documentation should
be expanded to discuss this in more detail.

As long as we're at it, are the other -malign-flags (or
-mpreferred-stack-boundary) necessary if one sets -march=i686?  If
they are, what are the optimal values to set them two, the docs say
that 4 is the default, but suggest that 3 would be better.

In any event, thanks to everyone to setting me straight on
malign-double--I've been having these problems as for quite some
time.

David