🚀 go-pugleaf

I've been trying to get gcc to generate far functions call (jalr
ra,$reg) instead of (jal symbol_name) while optimizing.  With no
optimization it generates the register based jump but turning on
optimization causes that to be reduced to the 28-bit jump
(understandable since it saves 2 instructions).

I have need to do calls from the virtual memory space to the fixed
physical memory space of the MIPS processor.  This requires a full 32
bit jump.  I've tried playing with the no-function-cse flags and
#define's.  Additionally I've tried modifying calls.c directly to do a
force_reg() but to no avail.  I suspect the optimizer is coming along
and deciding that it can do better.

Can anyone point me in the direction of where in the optimizer to look
for this (or, if I'm totally in left field, where else I should be
looking).

						Thanks in advance,
-- 
          __  /    __  /    /   /    ___/	     Doug Gerecht
         /   /    /   /    /   /    /
        /   /    /   /    /   /    / - /	     dougg@kentek.com
      _____/   _____/   _____/   _____/