On Fri, Jan 04, 2002 at 08:54:14AM -0500, Rafal Boni wrote:

 > As a heads-up to everyone, profiling is still broken with the new tool-
 > chain (unless Jason's fixed it in the last few days and I haven't caught
 > up to that email yet 8-), but the good news is I have a fix just need to
 > find the time to get it in the tree.

Actually, I committed a fix for the profiling problem.  I removed the
.noreorder/.reorder from the FUNCTION_PROFILER() macro in gcc (and adjusted
the location of the "jal _mcount" accordingly).  Profiled programs work
fine (I tested this specifically).

I figured you were just too busy with other stuff, and knowing that there
as only that one remaining issue, I just thought "aw, hell, I just do it" :-)

It's amazing how long this problem has been discussed on the master GCC
mailing lists (and even though an acceptable patch for it was posted by
Ethan in 2000, the problem is still un-fixed in gcc-current!, because of
some mis-communication after a slight change to the patch was made).

 > The *really* good news is that my last x86->mipseb (sgimips) cross-built 
 > with the new toolchain worked *really* well and fixed a bunch of cross-
 > build issues I was having using the old toolchain (like broken awk, some
 > libc brokenness that caused the ftp progress bar to display nonsense, ...).

Yes, I cross-built a kernel and full userland for my P-5064 to test this.

Folks with slower MIPS systems (like old DECstations, etc.) will really
like being able to cross-build easily from a faster machine, I suspect :-)

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>