chunkification costs

Stephen Weeks sweeks@research.nj.nec.com
Sun, 28 Feb 1999 21:16:36 -0500


To get a feel for how much chunkification is costing, I ran (most of
the benchmarks) in the old "one chunk" mode, where the entire program
is in one C chunk.  The numbers show that chunkification is still
costing (esp. in knuth-bendix, lexgen and logic).  Furthermore these
differences underestimate the costs of chunkification, since returns
always do a trampoline, even in the one chunk case.

		normal	one chunk
barnes-hut	13.8	13.0
count-graphs	12.8	11.9
fft		39.6	38.4
knuth-bendix	26.8	20.2
lexgen		48.8	41.5
life		57.6	55.6
logic		52.8	44.7
mandelbrot	16.3	15.8
matrix-multiply	13.3	13.0
nucleic		24.3	18.7
ratio-regions	25.4	22.3
simple		21.0	18.7
tsp		30.6	31.2
zern		44.9	45.3