[Jprogramming] A surprise (to me)

['Michael Day' via Programming]

Yes, I get similar timings in J602, J807 and J901 under Windows 10.

I happen to have found that rl performs better (for this example) on the 
list of indices

than on the parent string,  inadvertently, as I was looking for the 
maximum of q!

In JQt 9.01:

Callinjg the long q "ql",

    iql =: i.~ q

    timex'rl iql'
2.40118
    timex'rl ql'
7.50985

while this clunky work in progress gives:

    timex'(+/\ inv @: I.@: ~: >:@(>./),~}.) iql'
1.24205

(It's almost right: !

    10 {.+/\ inv ( I.@: ~: >:@(>./),~}.) iql
0 1 1 1 1 1 3 1 1 1

    10 {.rl iql
1 1 1 1 1 1 3 1 1 1
)

Space isn't so good:

    ts'(+/\ inv @: I.@: ~: >:@(>./),~}.) iql'  NB. time and space
1.32368 1.67773e9
    ts'rl iql'
2.42318 6.49995e8

So - why is the elegant rl "worse" for timing than my horrible cobble?

Mike



On 15/01/2020 08:18, bill lam wrote:
> I have no idea, but I got similar results on linux
>
>          time    space
> j907 : 5.05662 6.49995e8
> j807 : 1.24932 6.49991e8
> j602 : 0.514155 6.03982e8
>
> A bigger surprise is j602 being the fastest, almost 10 times
> faster than j901.
>
> Tue, 14 Jan 2020, Joey K Tuttle написал(а):
>> I was fiddling with patterns in the first 10e9 digits of Pi this evening and was surprised by a timing.
>>
>> Rather than what I was fooling with, I put it into a smaller easily reproduced example, to wit -
>>
>> MBpro-3:JKTststs jkt$ ja
>>     JVERSION
>> Engine: j807/j64/darwin
>> Release-e: commercial/2019-11-04T12:57:58
>> Library: 8.07.26
>> Platform: Darwin 64
>> Installer: J807 install
>> InstallPath: /applications/j64-807
>> Contact: www.jsoftware.com
>>     20?.@$ 2
>> 0 1 0 1 1 0 0 1 1 1 0 0 0 0 1 0 0 0 1 0
>>     rl
>> [: #;.1~ 1 , 2 ~:/\ ]
>>     rl 20?.@$ 2
>> 1 1 1 2 2 3 4 1 3 1 1
>>     q=. (50e6 ?.@$ 10){'0123456789'
>>     timex 'z=. rl q'
>> 1.720735 649991296
>>     hist
>> ~. ,: #/.~
>>     hist z
>>         1      3       2     4    5  7   6 8 9
>> 40500537 405764 4048327 40574 4095 35 410 4 1
>>     NB. i.e. one run of 9 digits, 4 of length 8 etc.
>>     timex 'hist z'
>> 0.295335 536872960
>>     exit 0
>>
>> MBpro-3:JKTststs jkt$ jb
>>     JVERSION
>> Engine: j901/j64avx2/darwin
>> Release-c: commercial/2020-01-11T12:24:26
>> Library: 9.01.18
>> Platform: Darwin 64
>> Installer: J901 install
>> InstallPath: /applications/j901
>> Contact: www.jsoftware.com
>>     q=. (50e6 ?.@$ 10){'0123456789'
>>     timex 'z=. rl q'
>> 6.56145 649995392
>>     hist z
>>         1      3       2     4    5  7   6 8 9
>> 40500537 405764 4048327 40574 4095 35 410 4 1
>>     6.56145 % 1.720735
>> 3.813167048
>>     exit 0
>> MBpro-3:JKTststs jkt$
>>
>> What surprised me was that the run length function rl took more space and about 4 times  as long in J9.01 as in J8.07 - most things I try are the other way around.
>>
>> Have others had similar experiences?
>>   
>> Does anyone see a reason my run length verb should bog down?
>>
>> - joey
>>
>>
>> ----------------------------------------------------------------------
>> For information about J forums see http://www.jsoftware.com/forums.htm