running times - coalescing vs not coalescing

Stephen Weeks sweeks@research.nj.nec.com
Tue, 9 Mar 1999 09:47:45 -0500


Here are the running times for the benchmarks with and without
coalescing.  I also give the difference in the number of trampolines
(without - with) and the number saved per second.  

In the cases wher coalescing caused a slowdown, a plausible
explanation is decreased performance of gcc.

		coales	don't	tramps saved	tramps/sec
		------	------	--------------	------------
barnes-hut	11.97	13.01	16,405,990	15,774,990
count-graphs	10.0	12.05	38,447,002	18,754,635
fft		42.87	41.7	302
knuth-bendix	15.28	24.02	107,324,502	12,279,691
lexgen		31.76	43.16	183,751,176	16,118,524
life		54.15	54.16	8,973,844	897,384,400
logic		49.26	51.14	98,447,550	52,365,718
mandelbrot	15.98	14.31	0
matrix-multiply	12.05	11.65	40
mlyacc		20.58	20.72	29,803,014	212,878,671
nucleic		17.07	21.14	51,750,600	12,715,135
ratio-regions	18.98	23.99	66,487,172	13,270,892
simple		18.4	19.3	44,465,075	49,405,638
tsp		25.42	27.14	23,645,916	13,747,625
vliw		19.11	19.25	12,748,634	91,061,671
zern		44.33	43.37	6	
smlc		101.39	106.16	228,308,072	47,863,327