[jira] [Comment Edited] (ARROW-1403) [C++] Add variant of SerializeRecordBatch that accepts an writer-allocator callback

["Wes McKinney (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/ARROW-1403?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16139515#comment-16139515 ] 

Wes McKinney edited comment on ARROW-1403 at 8/24/17 3:42 AM:
--------------------------------------------------------------

I looked at this today. I think this might be overkill versus simply writing a version of {{SerializeRecordBatch}} specialized for GPU device memory


was (Author: wesmckinn):
I looked at this today. I think this might be overkill versus simply writing a version of {{SerializeRecordBatch}} specialize for GPU device memory

> [C++] Add variant of SerializeRecordBatch that accepts an writer-allocator callback
> -----------------------------------------------------------------------------------
>
>                 Key: ARROW-1403
>                 URL: https://issues.apache.org/jira/browse/ARROW-1403
>             Project: Apache Arrow
>          Issue Type: Improvement
>          Components: C++
>            Reporter: Wes McKinney
>
> When writing to other kinds of interfaces, like GPU, it would be useful to be able to pass a function or closure that can instantiate an instance of {{OutputStream}} (which might write to CPU memory, GPU, etc.) given the computed size of the record batch. Currently we allocate new CPU memory and write to that buffer, but this would eliminate an intermediate copy.
> So something like
> {code}
> typedef std::function<Status(const int64_t, std::unique_ptr<OutputStream>*)> StreamCreator;
> {code}



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)