> So the moral of the story seems to be that with the new benchmarks `both' is
> a tie with `call'. I.e., 7 got faster and 7 got slower, but none by very
> much. I'm still confused about the slow-down. I suppose it could be `luck',
> where some heuristic just got fooled, but I would have thought that more
> contification was always better.
Here are the numbers using the native backend. They show that a lot of the
weirdness was caused by using C. Now, both is always as good as call with the
exception of tensor. And both is always as good as either with the exception of
knuth-bendix and tensor.
none call cont both
barnes-hut 6.9 4.8 6.9 4.8
checksum 10.8 4.2 4.2 4.2
count-graphs 9.8 7.4 7.9 7.0
fft 23.4 17.5 17.7 17.5
fib 4.1 4.3 4.3 4.3
knuth-bendix 8.9 8.7 8.3 8.7
lexgen 20.6 13.9 15.2 13.7
life 31.8 26.8 28.2 26.7
logic 37.4 37.2 37.3 36.8
mandelbrot 25.6 7.3 7.3 7.3
matrix-multiply 9.4 6.4 6.1 6.1
merge 45.0 43.3 42.7 42.6
mlyacc 23.8 13.3 21.6 13.2
mpuz 33.4 18.1 28.6 18.1
nucleic 13.5 11.6 13.1 11.6
ratio-regions 14.1 9.6 9.6 9.6
simple 8.5 7.4 8.7 7.4
smith-normal-form 1.2 1.2 1.2 1.1
tak 10.5 11.0 11.0 11.0
tensor 5.5 5.2 5.0 5.3
tsp 19.4 12.8 13.0 12.8
vliw 14.8 8.0 11.2 7.9
wc-input1 3.9 3.1 3.1 3.1
wc-scanStream 26.2 12.3 12.7 12.3
zern 63.8 33.4 33.4 33.3
(text/plain)