effects of chunk coalescing

Stephen Weeks sweeks@research.nj.nec.com
Mon, 8 Mar 1999 17:44:39 -0500


Here are the trampoline counts for the benchmarks with and without
random chunk coalescing.

These numbers are probably a bit misleading, because the small size of
the benchmarks makes it more likely for coalescing to work well.

		trampolines	returns		inter-chunk returns
		----------	------------	-------------------
barnes-hut	 1,482,484	21,175,411	  741,241
(no coalesce)	17,888,474	21,175,411	8,236,031

count-graphs	     1,102	20,948,776	       650
(no coalesce)	38,448,104	20,948,776	13,553,376

fft		20,974,184	10,487,242	10,487,091
(no coalesce)	20,974,486	10,487,242	10,487,242

knuth-bendix	 11,116,052	20,883,251	 5,865,100
(no coalesce)	118,440,554	20,883,251	57,741,301

lexgen		 16,958,142	111,061,245	10,826,810
(no coalesce)	200,709,318	111,061,245	99,916,628

life			2	5,789,414		 0
(no coalesce)	8,973,846	5,789,414	 4,619,132

logic		116,370,472	136,706,110	19,473,700
(no coalesce)	214,818,022	136,706,110	78,772,610

mandelbrot	2		0		0
(no coalesce)	2		0		0

matrix-multiply	 2		20		 0
(no coalesce)	42		20		20

mlyacc		19,839,376	50,907,041	 9,902,515
(no coalesce)	49,642,390	50,907,041	24,147,102

nucleic			 2	25,871,000		 0
(no coalesce)	51,750,602	25,871,000	25,658,800

ratio-regions	 5,183,525	35,835,339	 3,118,291
(no coalesce)	71,670,697	35,835,339	35,835,339

simple		 61,011,165	81,662,139	30,604,657
(no coalesce)	105,476,240	81,662,139	52,401,889

tsp			62	12,802,173		60
(no coalesce)	23,645,978	12,802,173	11,826,813

vliw		50,817,713	34,952,200	26,002,984
(no coalesce)	63,566,347	34,952,200	31,558,885

zern		2		3		0
(no coalesce)	8		3		3