1 vs. 2 rounds of Cps simplification
Stephen Weeks
sweeks@research.nj.nec.com
Mon, 1 Mar 1999 13:13:25 -0500
Here are the running times and code sizes comparing running the Cps
simplifier 1 or 2 rounds. There was notable improvement in logic
(20%) and simple (33%). I don't think this justifies running the
simplifier twice in general, but it's probably worth looking into
those cases that did speed up to understand why and maybe change the
phase ordering or add an extra pass.
time(s) size(bytes)
1 2 1 2
---- ---- ------ ------
barnes-hut 13.8 13.8 40675 39043
count-graphs 12.8 12.7 40621 40793
fft 39.6 35.1 26393 23541
knuth-bendix 26.8 26.2 70334 69878
lexgen 48.8 48.2 187519 184327
life 57.6 54.6 37181 36005
logic 52.8 42.1 363741 415021
mandelbrot 16.3 16.3 12965 12965
matrix-multiply 13.3 13.1 13577 13577
mlyacc 24.6 23.9 840215 817287
nucleic 24.3 29.9 52807 74567
ratio-regions 25.4 24.3 40601 40113
simple 21.0 14.9 337544 274452
tsp 30.6 30.7 30591 30495
vliw 21.5 20.6 927819 889351
zern 44.9 45.1 21234 20634