currying

[Stephen Weeks]

Here is a comparison of the dynamic counts for the syntactic Sxml
unknown calls.  Both of these are with polyvariance.  The difference
is that in the left hand column, curried function applications were
counted as known, while in the right hand column, they were counted as
known.  As we can see, counting the curried applications as known
eliminates most of the spurious unknown calls.  I conjecture that many
of the remaining ones that are caught by the flow analysis are in fact
loop invariants.

		curry unknown	curry known
barnes-hut	2		2
count-graphs	17,891,342	742
fft		45,440,290	0
knuth-bendix	16,654,135	16,502
lexgen		120,300,788	85,294,868
life		194,437,790	33
logic		168,178,500	17,824,510
mandelbrot	0		0
matrix-multiply	0		0
mlyacc		2,707,688	600,602
nucleic		62,783,000	12,139,400
ratio-regions	0		0
simple		5,285,980	4,832,999
tsp		2		2
vliw		7,372,074	4,670,889
zern		1,952,000	0