Anonymous user
Talk:Self numbers: Difference between revisions
→Improvement to the clever sieving of purefox: using modulus base*base+1
(Speedup for Go, link to tio.run) |
(→Improvement to the clever sieving of purefox: using modulus base*base+1) |
||
| (7 intermediate revisions by 3 users not shown) | |||
Line 6:
So the upper limit is 10,000 +(max digitcount)*9 -1 here 10,107<BR>
For example:<BR>
After marking the first 10000 (n = 0--9999) not selfnumbers the last marked is 9999+4*9 =
Now counting all selfnumbers 0..9999 than copy 10000- (max digitcount)*9 to Position 0..- (max digitcount)*9 and clear the rest to the upper limit.
This can all be done lightning fast in Level I cache. [[user Horst.h|Horst.h]] 18:20, 7 October 2020 (UTC)
:New observation using Base*Base+1 = 101. Line 0 is modified 1,3,5 are deleted<BR>
:Many same lines, changes in constant distances only a rotation of line 1 with some.
:There shall be a build system for this pattern.
:mostly 10 out of 101 are selfnumbers, every 9*101 -2 , every 99*101 -3 , every 999*101 -4
:
<pre>
annotation number of self numbers up to
101 13
1010 103
10100 991
101000 9880
1010000 98761
10100000 987562
101000000 9875563
1010000000 98755565
10100000000 987555566
n Div 101
0 0 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
101 1 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
202 2 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
303 3 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
404 4 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
505 5 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
606 6 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
707 7 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
808 8 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
909 9 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
-- rotate <<<<<<<<<
-- v deleted
1010 10 00000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000000000010
-- v inserted
1111 11 10000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000000000010
--
1818 18 10000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000000000010
-- v
1919 19 10000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000000000000
2020 20 00100000000001000000000010000000000100000000001000000000010000000000100000000001000000000010100000000
--
2828 28 00100000000001000000000010000000000100000000001000000000010000000000100000000001000000000010100000000
-- vvv
2929 29 00100000000001000000000010000000000100000000001000000000010000000000100000000001000000000000001000000
3030 30 00001000000000010000000000100000000001000000000010000000000100000000001000000000010100000000001000000
-- vvv
3939 39 00001000000000010000000000100000000001000000000010000000000100000000001000000000000001000000000010000
4040 40 00000010000000000100000000001000000000010000000000100000000001000000000010100000000001000000000010000
--
4848 48 00000010000000000100000000001000000000010000000000100000000001000000000010100000000001000000000010000
vvv
4949 49 00000010000000000100000000001000000000010000000000100000000001000000000000001000000000010000000000100
5050 50 00000000100000000001000000000010000000000100000000001000000000010100000000001000000000010000000000100
-- vvv
5959 59 00000000100000000001000000000010000000000100000000001000000000000001000000000010000000000100000000001
6060 60 00000000001000000000010000000000100000000001000000000010100000000001000000000010000000000100000000001
-- vvv
6969 69 00000000001000000000010000000000100000000001000000000000001000000000010000000000100000000001000000000
7070 70 01000000000010000000000100000000001000000000010100000000001000000000010000000000100000000001000000000
-- vvv
7979 79 01000000000010000000000100000000001000000000000001000000000010000000000100000000001000000000010000000
8080 80 00010000000000100000000001000000000010100000000001000000000010000000000100000000001000000000010000000
-- vvv
8989 89 00010000000000100000000001000000000000001000000000010000000000100000000001000000000010000000000100000
9090 90 00000100000000001000000000010100000000001000000000010000000000100000000001000000000010000000000100000
-- V VVV
9999 99 00000100000000000000000000000000010000000000100000000001000000000010000000000100000000001000000000010
----
99889 989 01000000000010000000000100000000001000000000010100000000001000000000010000000000100000000001000000000
-- v v vvv
99990 990 01000000000010000000000000000000000000000000000000000100000000001000000000010000000000100000000001000
First line found:
..8810 matches found... last one
100992930 999930 00000001010000000000100000000001000000000010000000000100000000001000000000010000000000100000000001000
</pre>
[[user:Horsth|Horsth]] [[User:Horsth|Horsth]] ([[User talk:Horsth|talk]]) 07:01, 9 October 2020 (UTC)
== Go, tweaked ==
Line 14 ⟶ 88:
Just saw Horst's idea, haven't incorporated it yet.--[[User:Enter your username|Enter your username]] ([[User talk:Enter your username|talk]]) 19:47, 7 October 2020 (UTC)
:Actually, I wondered after I'd done the 'sieve based' version whether using partial sums for each loop would quicken it up but, when it came in so much faster than the 'low memory' version, I didn't bother investigating further.
:Anyway I've now tested it and it does indeed improve performance by around 25% so I've incorporated it into the main page. Have done it slighly different to you (using an array of partial sums) to keep it more 'go fmt' friendly.
:Incidentally, the timings are for my core i7 which is why they're much faster than tio.run's.
:Might do a separate version incorporating Horst.h's ideas as I suspect it might come in a bit slower for the basic task of 100 million numbers while allowing us to stretch that to 1 billion. But we'll see. --[[User:PureFox|PureFox]] ([[User talk:PureFox|talk]]) 09:06, 8 October 2020 (UTC)
::I was wrong. Horst.h's approach is not only quicker for 100 million numbers but I can't even extend the second version to do up to 1 billion numbers without running into memory issues. So I've added a third version. A little slower than Pascal but not too bad :) --[[User:PureFox|PureFox]] ([[User talk:PureFox|talk]]) 12:53, 8 October 2020 (UTC)
| |||