[MLton] cvs commit: rewrote x86.Block.compress to run in linear time

Stephen Weeks MLton@mlton.org
Sat, 3 Jul 2004 12:54:09 -0700


I added -copy-prop-cutoff <n> and compiled HOL with

	-copy-prop-cutoff 1000

Here are the summary stats, comparing with the old runs where we
compiled with -native-copy-prop {true,false}

			no cutoff	1000		0
			----------	----------	----------
compile time (s)  	4993		4309		4192
code size (bytes)	25,179,782	25,181,702	25,293,974
run time (s)		272		272		278

So, a cutoff of 1000 saves most of the time without a noticeable loss
in optimization.

I'm less motivated to fix the pass now, but hopefully still motivated
enough.