Just for the record, concerning considering -O6 to be 'safe'
I've been compiling stuff with pgcc on a cyrix 6x86, for (the same)
cyrix6x86, using
march=cyrix6x86 and various -O[ ,2,3,6,9] including the kernel (2.1 series)
and I have seen a clear, dependable, predictable, repeatable pattern
that -O2 gives me stable programs and -O3 does not. therefore, at least for
this compiler/cpu combo,
-O6 is out of the question. (although technically, possibly the exact
proposed settings so far would be ok, as I have not tried -march=pentium -O6
, or rather, I beleive I actually have done that before I found out
about -march=cyrix6x86, but I do not have any organized memory from those
early runs, so no report to offer. )
I'm at work, can't tell you the exact pcgg and glibc numbers, they are both
from the pgcc site as of about a month ago. and I installed the contents of
the rpms by hand using midnight commander, and painstakingly cleaned
out/changed the references and files to the gcc2.8 and glibc that shipped
with RH5.1, with the exception of leaving the
even setting -march=cyrix6x86 -O3 or greater in the toplevel makefile in
hostcflags, and then forcing a re-compile of the lxdialog for menuconfig,
yeilds a very buggy dialog that basically can't make it through the whole
process of picking all my options fresh, unless I'm quick, and don't do much
back-tracking. backing it off from -O3 to -O2 clears it up. I shudder to
think about the poor birth-defect kernels I tried to run where the kernel
itself was compiled like that. -mpentium and -O3 at least were fine useing
gcc2.8, as I recall.
I have another box I can compile 2.0 series kernels on using pgcc
and -march=amdk6 + other options if testing and reports are wanted, but as
you know, with 2.0 needing patches just to use pgcc, that may not be an
informative test case unless all I have to report is that everything is
fine. I don't get to boot that box to linux enough to have seen any kind of
pattern at present, but the cyrix is clear as a bell, -O2 guuud, -O3 baaad
Perhaps since it *is* -march=cyrix6x86, the better place to report this
would be some gcc list or group, but,
I can't follow a mail list for every file on my system and so I'm not on any
such gcc list.
-- Brian~
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/