Diego Elias Damasceno Costa created LOG4J2-2478:
---------------------------------------------------
Summary: JMH Benchmarks in not consuming all computed variables
(risk of DCE)
Key: LOG4J2-2478
URL: https://issues.apache.org/jira/browse/LOG4J2-2478
Project: Log4j 2
Issue Type: Bug
Components: Tests
Environment: I am reporting the result of 5 full executions with
default parameters (iterations, forks, warmups, etc..)
Tests run on a Computational server with CPU: E5-1660-3.3GHZ (6 cores + HT), 64
GB RAM.
Reporter: Diego Elias Damasceno Costa
Attachments: logginglog4j2-jmhantipatterns-retu.csv
We are conducting a scientific study to investigate bad practices/anti-patterns
on creating micro-benchmarks using JMH, and we found an instance of harmfull
non-consumed computation in `AbstractStringLayoutStringEncodingBenchmark`.
As a good practice, every computation performed in inside the benchmark should
be consumed (see [JMH
Documentation|[http://hg.openjdk.java.net/code-tools/jmh/file/66fb723292d4/jmh-samples/src/main/java/org/openjdk/jmh/samples/JMHSample_08_DeadCode.java].]
This is partially done here with the self-made consume method, but on some
benchmarks (such as baseline, usAsciiGetBytes) the return is not consumed in
the benchmark, which opens the possibility of a Dead-Code Elimination to be
performed by JVM.
*There is a very simple solution:* just return the long primitive calculated by
_consume()_ at the end of each benchmark. This is done on other benchmarks of
the project as well.
In our tests, benchmarks were lightly affected (see attachment), with
Throughputs 5-35% lower after our fix patch.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
