[ https://issues.apache.org/jira/browse/SYSTEMML-2050?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Matthias Boehm updated SYSTEMML-2050: ------------------------------------- Description: As pointed out by [~dusenberrymw] - prints currently show up in reverse order after the merge of sequences of statement blocks. For instance, the following example {code} print("test1"); print("test"+2); print("test"+3); if(1==1){} print("test4"); print("test"+5); {code} currently produces the following output {code} test4 test5 test1 test2 test3 {code} The new instruction scheduling introduced with SYSTEMML-2039 ensures (as a byproduct) that prints within one DAG are scheduled in their original order. However, after the merge of statement blocks, prints without data dependencies can be reordered. This task aims to simply sort all DAG output instructions accordingly to their line numbers. Note that his only affects the new instruction scheduling as used for spark, hybrid_spark, and singlenode. was: As pointed out by [~dusenberrymw] - prints currently show up in reverse order after the merge of sequences of statement blocks. For instance, the following example {code} print("test1"); print("test"+2); print("test"+3); if(1==1){} print("test4"); print("test"+5); {code} currently produces the following output {code} test4 test5 test1 test2 test3 {code} The new instruction scheduling introduced with SYSTEMML-2039 ensures (as a byproduct) that prints within one DAG are scheduled in their original order. However, after the merge of statement blocks, prints without data dependencies can be reordered. This task aims to simply sort all DAG output instructions accordingly to their line numbers. > Prints show in reverse order of merge of statement blocks > --------------------------------------------------------- > > Key: SYSTEMML-2050 > URL: https://issues.apache.org/jira/browse/SYSTEMML-2050 > Project: SystemML > Issue Type: Bug > Reporter: Matthias Boehm > > As pointed out by [~dusenberrymw] - prints currently show up in reverse order > after the merge of sequences of statement blocks. For instance, the following > example > {code} > print("test1"); > print("test"+2); > print("test"+3); > if(1==1){} > print("test4"); > print("test"+5); > {code} > currently produces the following output > {code} > test4 > test5 > test1 > test2 > test3 > {code} > The new instruction scheduling introduced with SYSTEMML-2039 ensures (as a > byproduct) that prints within one DAG are scheduled in their original order. > However, after the merge of statement blocks, prints without data > dependencies can be reordered. > This task aims to simply sort all DAG output instructions accordingly to > their line numbers. Note that his only affects the new instruction scheduling > as used for spark, hybrid_spark, and singlenode. -- This message was sent by Atlassian JIRA (v6.4.14#64029)