I think this version of the patch addresses all the issues Richard 
raised with the last version, unless I've gotten confused again and let 
something slip through the cracks.  I experimented with tweaking all of 
MOVE_RATIO/CLEAR_RATIO/SET_RATIO and came up with values that produce 
good -Os results in CSiBE as well as making sense in terms of the base 
MIPS_CALL_RATIO versus how many instructions would be required for each 
bytewise move/clear/set.  Overall I am getting just over 0.5% 
improvement with -Os from this patch now.

Mark, could you take another look at this as well, since I included 
Richard's suggested change to remove the optimize_size check from the 
target-independent code in builtins.c?

-Sandra