Tail recursion problem (ptr_mode != Pmode)

[Jeffrey Naset <jnaset at rsn dot hp dot com>]

Hi,

  I've found a problem that affects the ia64-hp-hpux11.23 target at the 
very least.

  gcc (snapshot 2003 0310) is generating 'gengtype' as part of its 
bootstrap, and gengtype has incorrect code.  In the function 
'output_mangled_typename', tail recursion optimization has occurred, 
with the flaw that 32-bit pointers aren't getting swizzled prior to use.

  In this case, output_mangled_typename takes two args (both pointers), 
and they are swizzled at the very top of the function (probably in the 
prologue).  Later, when a recursive call is made, the args are loaded 
(unswizzled, per the calling conventions) and a branch is made, to a 
point just after where the pointers are swizzled.  Then the program 
dies trying to reference memory location in a register that doesn't 
have a valid 64-bit pointer value.  (It is a valid 32-bit pointer)   
[Pmode for the architecture is DImode, ptr_mode in this case is SImode]

  If anyone has any ideas on how to fix this, please let me know!

Thanks,
Jeffrey Naset
jeffrey (dot) naset (at) hp (dot) com