mxnet-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Chris Olivier (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (MXNET-4) Performance: Refactor Random and ParallelRandom resources to use MKL for MKL builds
Date Tue, 16 Jan 2018 18:39:00 GMT

     [ https://issues.apache.org/jira/browse/MXNET-4?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

Chris Olivier updated MXNET-4:
------------------------------
    Description: 
Refactor Random and ParallelRandom resources to use MKL for MKL builds

Things such as RngUniform, etc.  Similarly to what is done for dropout operator.

It may need to allocate some temporary memory and generate random numbers in batches, then
serving them out from that batch. 

Also the Random classes could export a "fill buffer with randoms" function, which seems to
be a common use-case and fits the MKL API more closely.

Care must be taken regarding MKL's fixed output types for some of the API functions.

  was:
Refactor Random and ParallelRandom resources to use MKL for MKL builds

Things such as RngUniform, etc.  Similarly to what is done for dropout operator.


> Performance: Refactor Random and ParallelRandom resources to use MKL for MKL builds
> -----------------------------------------------------------------------------------
>
>                 Key: MXNET-4
>                 URL: https://issues.apache.org/jira/browse/MXNET-4
>             Project: Apache MXNet
>          Issue Type: Improvement
>            Reporter: Chris Olivier
>            Priority: Major
>
> Refactor Random and ParallelRandom resources to use MKL for MKL builds
> Things such as RngUniform, etc.  Similarly to what is done for dropout operator.
> It may need to allocate some temporary memory and generate random numbers in batches,
then serving them out from that batch. 
> Also the Random classes could export a "fill buffer with randoms" function, which seems
to be a common use-case and fits the MKL API more closely.
> Care must be taken regarding MKL's fixed output types for some of the API functions.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Mime
View raw message