Another build problem under hpux 10.20 with current cvs source

[Richard Earnshaw]

dave@hiauly1.hia.nrc.ca said:
> I dropped to -O2 and I am passed the failure point.  This might be
> rename registers clobbering the return pointer again.  That one is
> hard to trace because the machine goes off to never-never-land.  We
> need to find a solution that you like for the return pointer bug. 

I've just fixed a similar bug on the ARM.  It was quite easy, just add a 
definition of HARD_REGNO_RENAME_OK(FROM,TO) that blocks the change when 
necessary (eg return register not stacked).  Only minor difficulty was 
that the macro isn't documented and I had to grub around in the source to 
find it (Grrrr!).

R.