[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17378494#comment-17378494
]
Antoine Pitrou commented on ARROW-10899:
Also worth taking a look at
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17360244#comment-17360244
]
Antoine Pitrou commented on ARROW-10899:
Ha, the search space for sorting algorithms looks a bit
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17360162#comment-17360162
]
Kirill Lykov commented on ARROW-10899:
--
Well, it looks like it is called "wolf sort"
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17360141#comment-17360141
]
Antoine Pitrou commented on ARROW-10899:
Good question about bandwidth!
A possible solution for
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17360115#comment-17360115
]
Kirill Lykov commented on ARROW-10899:
--
LSD sort is stable.
Do you want to try with Travis' LSD
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17359997#comment-17359997
]
Antoine Pitrou commented on ARROW-10899:
Thanks for the investigation! Were you able to check on
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17359990#comment-17359990
]
Kirill Lykov commented on ARROW-10899:
--
I have to drop this ticket but if someone will take it, I
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17341925#comment-17341925
]
Kirill Lykov commented on ARROW-10899:
--
I believe that the LSD implementation by Travis is stable
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17341893#comment-17341893
]
Antoine Pitrou commented on ARROW-10899:
Thanks a lot for running these benchmarks, by the way.
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17341891#comment-17341891
]
Antoine Pitrou commented on ARROW-10899:
We can't presume anything about the distribution of
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17341526#comment-17341526
]
Kirill Lykov commented on ARROW-10899:
--
Do we have some information about distribution of integers?
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17332655#comment-17332655
]
Kirill Lykov commented on ARROW-10899:
--
Actually, I don't see significant effect of prefetch in my
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17332650#comment-17332650
]
Antoine Pitrou commented on ARROW-10899:
You should probably bench against {{radix_sort6}}
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17332609#comment-17332609
]
Kirill Lykov commented on ARROW-10899:
--
I'm continuing experiments with MSD sort. Please ignore
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17324903#comment-17324903
]
Antoine Pitrou commented on ARROW-10899:
I'm not sure we need intense benchmarking. Admittedly,
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17324878#comment-17324878
]
Kirill Lykov commented on ARROW-10899:
--
Updates, I've contacted Travis and he wrote that he has a
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17319494#comment-17319494
]
Kirill Lykov commented on ARROW-10899:
--
Right, I agree that it is stable.
>From performance
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17319476#comment-17319476
]
Antoine Pitrou commented on ARROW-10899:
I don't know if Travis would be interested, as plumbing
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17319478#comment-17319478
]
Antoine Pitrou commented on ARROW-10899:
Also, AFAIU his radix sort implementation is simply a
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17318707#comment-17318707
]
Kirill Lykov commented on ARROW-10899:
--
Thanks for the reference to the blog, I read all of his
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17318133#comment-17318133
]
Antoine Pitrou commented on ARROW-10899:
Interesting read here:
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17314867#comment-17314867
]
Antoine Pitrou commented on ARROW-10899:
Thanks for the update [~klykov]. It will be useful if
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17314850#comment-17314850
]
Kirill Lykov commented on ARROW-10899:
--
I added a repository to put there some experiments i've
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17282484#comment-17282484
]
Kirill Lykov commented on ARROW-10899:
--
On a higher level I honestly thought that I will easily
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17282355#comment-17282355
]
Antoine Pitrou commented on ARROW-10899:
Even a radix sort is not necessarily stable. You have
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17282030#comment-17282030
]
Kirill Lykov commented on ARROW-10899:
--
Right, I will check also spinsort from boost
> [C++]
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17282026#comment-17282026
]
Antoine Pitrou commented on ARROW-10899:
Note that we need a stable sort, which spreadsort
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17282023#comment-17282023
]
Kirill Lykov commented on ARROW-10899:
--
Your intuition is correct, at least for int32_t key. I
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17281908#comment-17281908
]
Antoine Pitrou commented on ARROW-10899:
[~klykov] Feel free to experiment, but note that adding
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17281864#comment-17281864
]
Kirill Lykov commented on ARROW-10899:
--
1. From the paper "Fast Sort on CPUs and GPUs: A Case for
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17280272#comment-17280272
]
Kirill Lykov commented on ARROW-10899:
--
Thanks for the clarification, I will check it out next
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17280197#comment-17280197
]
Yibo Cai commented on ARROW-10899:
--
[~apitrou], please correct me if I missed somthing.
Code:
[
https://issues.apache.org/jira/browse/ARROW-10899?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=17279881#comment-17279881
]
Kirill Lykov commented on ARROW-10899:
--
Sounds interesting to me, I would like to have a look.
33 matches
Mail list logo
