Bootstrap time regressions

[Greg Schafer <gschafer at zip dot com dot au>]

Hi

First, GCC continues to advance in leaps and bounds and I'm very grateful to
all involved for making such a great product. The bugfixing rate of late is
amazing.

Second, I have some concerns about mainline bootstrap speed. Please see this
BZ for the background:

  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19223

The bottom line is:

 1) STAGE1_CHECKING slows down C only bootstraps by up to 19.5% (on my setup
    at least)

 2) C only bootstrap times with GCC-4 have almost exactly doubled since
    3.4.3 (again, on my setup at least)

While 2) is not good, there is obviously loads of new code in the guts of
GCC and probably no easy fixes.

However, for 1), STAGE1_CHECKING is set by default in current GCC-4 with no
easy way to switch it off. The main reason for this unconditional setting of
STAGE1_CHECKING is paraphrased as "to catch more bugs". I can't argue with
that, but I do worry about it when it's at the expense of major bootstrap
slow downs in a release compiler.

I guess I'm asking if anyone else thinks that STAGE1_CHECKING should be
switched off in the released 4.0 compiler in the face of hard numbers?

Thanks
Greg

PS - the time taken when compiling other C code with GCC-4 is only
fractionally slower than 3.4.3 in my tests which is very heartening
considering all the changes. It's the slow bootstrap times that worry me.