[ 
https://issues.apache.org/jira/browse/SYSTEMML-2092?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Matthias Boehm updated SYSTEMML-2092:
-------------------------------------
    Description: 
So far, parfor loops only support left and right indexing operations over 
result variables. The related result merge then simply merges the partial 
worker results with or without compare matrix. However, there are many use 
cases that would benefit from additional result aggregation functions.

 

This task aims to introduce a new {{R += A}} accumulator operation to address 
the above challenge. The semantics are equivalent to {{R = R + A}}, but the 
parfor dependency analysis will recognize this as a special aggregation 
function instead of throwing an error due to output dependencies.

  was:
So far, parfor loops only support left and right indexing operations over 
result variables. The related result merge then simply merges the partial 
worker results with or without compare matrix. However, there are many use 
cases that would benefit from additional result aggregation functions.

 

This task aims to introduce a new \{{R += A}} accumulator operation to address 
the above challenge. The semantics are equivalent to R = R + A, but the parfor 
dependency analysis will recognize this as a special aggregation function 
instead of throwing an error due to output dependencies.


> Parfor result aggregation
> -------------------------
>
>                 Key: SYSTEMML-2092
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-2092
>             Project: SystemML
>          Issue Type: Task
>            Reporter: Matthias Boehm
>            Priority: Major
>
> So far, parfor loops only support left and right indexing operations over 
> result variables. The related result merge then simply merges the partial 
> worker results with or without compare matrix. However, there are many use 
> cases that would benefit from additional result aggregation functions.
>  
> This task aims to introduce a new {{R += A}} accumulator operation to address 
> the above challenge. The semantics are equivalent to {{R = R + A}}, but the 
> parfor dependency analysis will recognize this as a special aggregation 
> function instead of throwing an error due to output dependencies.



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

Reply via email to