Hi all !
Here are the same measurements with Bill Lams proposed way to measure
this. As we can all see it is lousy because the overhead in the
measurement loop is much higher than that in @.
n=.500000
v=.?~n
$v
500000
10{.v
414253 208033 396027 190208 112248 196018 51503 465441 277514 114236
ts'-AtTacit-AtTacit-AtTacit-AtTacit-@- v'
0.258554 1.32199e8
ts'-AtTacit-AtTacit-AtTacit-AtTacit-AtopTacit- v'
0.260208 1.322e8
ts'-At-At-At-At-Atop- v'
2.98929 1.32204e8
ts'-@:-@:-@:-@:-@- v'
0.154225 1.32197e8
ts'-@- v'
0.0907723 1.32195e8
v=.1
n*{.n ts'-@:-@:-@:-@:- v'
0.455432
Cheers,
Erling
On 2017-10-02 17:24, Raul Miller wrote:
Believe me, I understood that.
But this particular performance measure is rather like measuring the
performance of tires sliding sideways (as opposed to rolling) in a
parking lot while carrying several hundred pounds of meatloaf.
In other words, it's not something I would feel comfortable optimizing
for, though I can sort of almost see some sort of vague connection to
real applications if I do not think about it too much.
Thanks,
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
