target register load optimizations

[Joern Rennecke]

It turned out that there was a problem with running flow betweeen
reload and epilogue generation.  mark_regs_live_at_end uses
HAVE_epilogue && reload_completed to check if an rtl epilogue
has already been generated.  I've introduced a new variable
epilogue_comleted for this and patched rtl.h, flow.c, toplev.c,
function.c, config/ia64/ia64.c, config/rs6000/rs6000.c,
config/sh/sh.c and config/sparc/sparc.c accordingly.

It would be nice if we had an enumeration variable that replaces
these variables as well as rtx_equal_function_value_matters, and
have a setting function that also sets default values for
cse_not_expected, no_new_pseudos etc., but that would really be a
separete project. 

-- 
--------------------------
SuperH (UK) Ltd.
2410 Aztec West / Almondsbury / BRISTOL / BS32 4QX
T:+44 1454 465658