|
This channel is intended for people just starting with the Raku Programming Language (raku.org). Logs are available at irclogs.raku.org/raku-beginner/live.html Set by lizmat on 8 June 2022. |
|||
|
00:05
teatwo joined
00:09
teatime left
04:29
rantanplan joined
04:30
snonux left
05:30
CIAvash joined
07:40
lizmat left
07:42
lizmat joined
07:57
dakkar joined
08:00
dakkar left
08:01
dakkar joined
08:03
dakkar left
08:04
dakkar joined
|
|||
| rcmlz | :-) | 08:51 | |
|
09:50
CIAvash left
|
|||
| if you replace the for ^($prime**$power) -> $val { in line 18 with (^($prime**$power)).race(batch => 1, degree => Kernel.cpu-cores).map( -> $val { ... perhaps you can get faster ... | 10:16 | ||
| (I assume batch size 1 is appropriate here ...) | 10:17 | ||
| and put in line 41 the colosing ) for the map() | 10:19 | ||
| nemokosch | could for not consume the race sequence as well? | 10:42 | |
| I don't actually know but it would be reasonable and practical | 10:43 | ||
|
11:30
raiph joined
11:40
tea3po joined
11:43
teatwo left
11:57
raiph left
11:59
raiph joined
12:20
raiph left
|
|||
| rcmlz | hyper for ^100 -> $a { works as well - and is even simpler - thanks for the hint. | 12:49 | |
| nemokosch | no race? | 12:51 | |
| you know the difference between hyper and race, right? | |||
| lizmat | and yet another Rakudo Weekly News hits the Net: rakudoweekly.blog/2023/10/16/2023-42-2000/ | 13:07 | |
| rcmlz | race as well | ||
| yes | |||
| nhail | What does batch size mean? I don’t know much about race | 14:57 | |
| Well I threw in the multithreading and it did speed it up quite a bit | 15:47 | ||
| I then decided to run it looking for solutions mod 17^6 | 15:49 | ||
| ~24 million | |||
| update: it took 30 mins | 16:39 | ||
|
16:52
dakkar left
|
|||
| rcmlz | batch: how many items from the input list get processed at once, when a single item takes long to be processed, batch should be small, when a single item is quickly calculated, batch can be largeer | 17:02 | |
| nhail | So actually batch should be kinda big here | 17:13 | |
| It's just a polynomial and a % | |||
| rcmlz | it controls the trade-off between sending data to a thread | ||
| nhail | Okay, makes sense | ||
| I'm working on a new version that uses some more number theoretic machinery instead of a raw brute force search | 17:14 | ||
| rcmlz | I suggest to test the impact of various batch sizes. Have a look at rosetta code, I put an example for parallel quick and merge sort there. | 17:15 | |
| I mean for benchmarking the impact of batch size. | 17:16 | ||
| nhail | For 17^5: batch 10 took 49s. batch 100 took 39s. batch 500 took 36s. | 17:20 | |
| This is on an intel i5-3340M, on my laptop | |||
| rcmlz | and 2**15? | 17:21 | |
| it all depends on the total number of items, of course, and also on the number of cores. | 17:22 | ||
| I was under the assumption that number items / cores should be optimal, but my experiments did not support that assumption. | 17:23 | ||
| nhail | As the prime power? All sizes finished in under a second | 17:24 | |
| With 2**15 as the batch size for 17**5, it was 38 seconds | 17:26 | ||
|
23:40
teatwo joined
23:43
tea3po left
|
|||