[jira] [Commented] (SYSTEMML-1561) Improve constant folding during compilation

["Mike Dusenberry (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/SYSTEMML-1561?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15995732#comment-15995732 ] 

Mike Dusenberry commented on SYSTEMML-1561:
-------------------------------------------

Just FYI, I'm making some progress on this.  Essentially, by rerunning static rewrites + IPA again immediately after the initial IPA pass as a kind of "second chance", we're able to apply this constant folding rewrite for this scenario.  This make sense because during the initial static rewrite pass, we can't apply constant folding to the {{Hout}}, {{Wout}}, etc. DAGs due to the leaf nodes being scalar transient reads.  After IPA with the new scalar replacement, these DAGs will become entirely operations on literal leaf nodes, and thus eligible for constant folding.  Then, after that second pass of static rewrites, we can benefit from IPA again by being able to now perform scalar replacement for functions/other DAGs that consume the {{Hout}}, {{Wout}}, etc. DAGs, which are now literals.  In terms of performance, I'm seeing the execution time cut in half (~500s faster) for SYSTEMML-1566.  I can open a PR soon.

> Improve constant folding during compilation
> -------------------------------------------
>
>                 Key: SYSTEMML-1561
>                 URL: https://issues.apache.org/jira/browse/SYSTEMML-1561
>             Project: SystemML
>          Issue Type: Improvement
>            Reporter: Mike Dusenberry
>             Fix For: SystemML 1.0
>
>         Attachments: scenario1_plan.txt, scenario1.py, scenario2_plan.txt, scenario2.py
>
>
> In our `nn` library, our convolution and pooling layers have to pass around the spatial dimensions (height and width) of the images that are stretched out into rows of the input/output matrices.  These output dimensions are computed within the forward functions of the above layers as small scalar equations.  From a mathematical standpoint, these sizes can be determined at compile time, and it is nice to have these size equations in DML (v.s. hiding them inside the engine within built-in functions).  However, we do not currently evaluate these expressions during compilation, and thus we are left with unknown sizes even during recompilation.  This naturally leads to max memory estimates and thus often leads to unnecessary distributed runtime ops rather than simple CP ones.
> I have two related scenarios for which this is a problem.  They both involve the {{Houtc1}} & {{Woutc1}} values that are returned from a `conv2d::forward(...)` function.  These represent the spatial dimensions of the volume with each of the rows of the output {{outc1}} of the function, and the third dimension is {{F1}}.  Thus, {{outc1}} has a number of columns equal to {{F1*Houtc1*Wouc1}}.
> In the first scenario ({{scenario1.py}}), a random matrix {{doutc1}} is created that should have the same dimensions as {{outc1}}.  For the columns, if I use {{cols=ncol(outc1)}} in this rand statement, the size will be propagated and CP ops will be compiled and run.  I I instead use {{cols=F1*Houtc1*Woutc1}}, the size will forever be unknown, even during recompilation, and thus Spark ops will be compiled and run.  I have included the recompile hops plan ({{scenario1_plan.txt}}).
> In the second scenario ({{scenario2.py}}), a {{max_pool2d::forward(...)}} function is inserted after the {{conv2d::forward(...)}} function that requires the {{Houtc1}} and {{Woutc1}} variables to be supplied as arguments.  Since those latter variables are not executed during compilation time, the max pooling sizes remain unknown, even during recompilation, and thus Spark ops will be compiled and run.  I have included the recompile hops plan ({{scenario2_plan.txt}}).
> We should either improve or fix our constant folding rewrites so that these scenarios are fixed, as they are necessary for performant deep learning applications.  Note too that this issue will be present in other non-deep learning scenarios as well.
> Mailing list thread: https://www.mail-archive.com/dev@systemml.incubator.apache.org/msg01657.html



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)