[incubator-mxnet] branch v1.2.0 updated: Scala inference memory leak fix #11204 (#11216)

[an...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

anirudh2290 pushed a commit to branch v1.2.0
in repository https://gitbox.apache.org/repos/asf/incubator-mxnet.git


The following commit(s) were added to refs/heads/v1.2.0 by this push:
     new 864c750  Scala inference memory leak fix #11204 (#11216)
864c750 is described below

commit 864c750fb5c689731adac4e79c582c40a9c88b23
Author: Andrew Ayres <an...@gmail.com>
AuthorDate: Thu Jun 14 10:50:43 2018 -0700

    Scala inference memory leak fix #11204 (#11216)
    
    * Fixes Scala memory leak (#10436)
    
    * Replaced the copy and disposed of sliced ndArray to resolve memory leak
    
    * Wrapped disposes in a finally to ensure they are called.
---
 .../main/scala/org/apache/mxnet/FeedForward.scala  | 25 ++++++++++++++++------
 1 file changed, 18 insertions(+), 7 deletions(-)

diff --git a/scala-package/core/src/main/scala/org/apache/mxnet/FeedForward.scala b/scala-package/core/src/main/scala/org/apache/mxnet/FeedForward.scala
index 7289df1..87c9bc7 100644
--- a/scala-package/core/src/main/scala/org/apache/mxnet/FeedForward.scala
+++ b/scala-package/core/src/main/scala/org/apache/mxnet/FeedForward.scala
@@ -224,13 +224,24 @@ class FeedForward private(
     var i = 0
     while (data.hasNext && i != numBatch) {
       val batch = data.next()
-      i += 1
-      ExecutorManager.loadData(batch, dataArrays)
-      predExec.forward(isTrain = false)
-      val padded = batch.pad
-      val realSize = batchSize - padded
-      for ((list, nd) <- outputs zip predExec.outputs) {
-        list += nd.slice(0, realSize).copy()
+      try {
+        i += 1
+        ExecutorManager.loadData(batch, dataArrays)
+        predExec.forward(isTrain = false)
+        val padded = batch.pad
+        val realSize = batchSize - padded
+        for ((list, nd) <- outputs zip predExec.outputs) {
+          // The slice is being written to a value so that dispose can be called after the copy.
+          // The one liner nd.slice().copy() leads to leaking the memory of the slice.
+          val ndSliced = nd.slice(0, realSize)
+          try {
+            list += ndSliced.copy()
+          } finally {
+            ndSliced.dispose()
+          }
+        }
+      } finally {
+        batch.dispose()
       }
     }
     // TODO(Yizhi): we can use Symbol.concat to do the same thing. Can it be more efficient?

-- 
To stop receiving notification emails like this one, please contact
anirudh2290@apache.org.