>I'm confused.  You're not seeing same problem if you're getting
>assembly output.  I die with internal compiler errors like the following:

ah. I'm using a `stage1' egcs: egcs compiled by gcc 2.7.2.2+myc1.
that works fine on the code-frag you gave me.  I havent tried a stage2
egcs (egcs compiled with egcs) yet.

>My binutils 2.8.1 cross environment has no problems synthesizing the
>instruction sequence.  The sequence looks something like

So does mine, after all.  False alarm there. Please ignore.

I'm using -O2. does that make a difference? if not: I'd guess the egcs
1.1 you're using now is buggy due to a bug in the egcs 1.0.2 you used
to compile the cross-compiler with.  my egcs 1.0.2 (same source
version as yours) works fine.

Do you have a gcc2.7.2.2 for the compile host?  Compiling your egcs
cross-compiler (i.e., targeted at netbsd/mipsel) with 2.7.2.2 should
work better on this specific test-case. at least, mine does.

It would be interesting to keep both egcs cc1 binaries around and
compare the output of cc1 -da from the two, and see where the RTL is
getting hosed.  (recompiling egcs 1.1. again with both compilers and
and comparing would also be informative).

>As I said earlier, since you're getting assembly code, it looks like
>there's some difference between the cross compilation and native
>environments.  Darn.

I'd investigate the results made by different bootstrap compilers
further first...