[M3commit] CVS Update: cm3

[Jay Krell]

CVSROOT:	/usr/cvs
Changes by:	jkrell at birch.	10/06/02 20:31:34

Modified files:
	cm3/m3-sys/m3cc/gcc/gcc/config/: sol2.h 
	cm3/m3-sys/m3cc/gcc/gcc/config/sparc/: sparc.h 
	cm3/m3-sys/m3cc/gcc/gcc/m3cg/: m3gty43.h m3gty45.h parse.c 

Log message:
	significantly remove use of volatile
	Allowing the optimizer to actually do anything
	
	volatile remains on Solaris/sparc32
	probably just need to experiment extensively
	
	volatile remains on all floating point loads,
	quite unfortunate for the scientific computing crowed..
	
	volatile is added to more locals and temporaries
	within functions that call setjmp/vfork
	(setjmp is very common!?)
	
	volatile store after insert_mn, small lame hack,
	because otherwise we have a read before write
	
	a few optimizations are turned off,
	though I did solve the "unit at a time" problem, with TREE_USED
	or so I thought, but no...I had confused "ui" and "vbtkit" packages
	
	prototype marking fault_proc as noreturn
	but then I get a warning that it does return,
	even if I removed the return
	
	Provide and use m3_build_pointer_type that presently is pointless,
	but maybe will mean something in future
	There is a bit to turn off alias analysis.
	pre/fre still crashed. I didn't try vrp, it is
	the most painful to test when it goes wrong, because
	it successfully compiles everything.
	
	Mostly tested on AMD64_DARWIN.
	Needs broader coverage.
	Need to consider the warnings it causes.
	Need to get fault_proc marked noreturn first.
	Could use better for Solaris/sparc32, but machine
	is so slow experimentation is painful.