Vector API - Matrix Benchmark Contribution

Paul Sandoz paul.sandoz at oracle.com
Thu Aug 17 20:05:04 UTC 2023


Here it is:

  https://bugs.openjdk.org/browse/JDK-8314544

Paul.

On Aug 17, 2023, at 11:37 AM, Martin Stypinski <mstypinski at gmail.com> wrote:

OK, so can you open an issue and send me the number? I can redo the PR next week.

Best
Martin.

Am Do., 17. Aug. 2023 um 18:57 Uhr schrieb Paul Sandoz <paul.sandoz at oracle.com<mailto:paul.sandoz at oracle.com>>:
Thanks, part of step 3 is I need create an issue and associate it with the PR. (There are some failed pre-integration checks that block integration and this is one of them)

Paul.

On Aug 17, 2023, at 3:12 AM, Martin Stypinski <mstypinski at gmail.com<mailto:mstypinski at gmail.com>> wrote:

ok, fixed it as discussed

1. is done - a long time ago.
2. PR is done, testing went through
3. Go ahead

Any reasons why it failed?

Am Mi., 16. Aug. 2023 um 23:01 Uhr schrieb Paul Sandoz <paul.sandoz at oracle.com<mailto:paul.sandoz at oracle.com>>:
That looks good. I recommend:

- Removing the system property for VECTOR_ACCESS_OOB_CHECK
- The benchmark methods should return the result (JMH will send it down a blackhole)
- You may also want to remove the allocation of the result matrix (pass it in as a pre-allocated argument and return it), since what we are interested here is how this all optimizes for the loop structure.

The process is:

1. Sign the OCA https://oca.opensource.oracle.com/
2. Submit a PR to https://github.com/openjdk/jdk/<https://urldefense.com/v3/__https://github.com/openjdk/jdk/__;!!ACWV5N9M2RV99hQ!I32aHuZI7cW86Y3E969ZLyt4xH5nc2970bNrLwvfU1kc_rvxq1863brm4DAQrlmUr8XbLIuCX__gW28g2LC-$>, placing the class in test/micro/org/openjdk/bench/jdk/incubator/vector/
(Building the JDK and testing should be straightforward, for testing see https://github.com/openjdk/jdk/blob/master/doc/testing.md<https://urldefense.com/v3/__https://github.com/openjdk/jdk/blob/master/doc/testing.md__;!!ACWV5N9M2RV99hQ!I32aHuZI7cW86Y3E969ZLyt4xH5nc2970bNrLwvfU1kc_rvxq1863brm4DAQrlmUr8XbLIuCX__gW0i3Z52N$>)
3. I can review/sponsor

Paul.

On Aug 11, 2023, at 12:24 AM, Martin Stypinski <mstypinski at gmail.com<mailto:mstypinski at gmail.com>> wrote:

Hi all,

I started this thing a long time ago and wanted to actually contribute my matrix benchmark as suggested by Paul. I made some changes: https://gist.github.com/Styp/618cf9d139151727f46396235aff09a3<https://urldefense.com/v3/__https://gist.github.com/Styp/618cf9d139151727f46396235aff09a3__;!!ACWV5N9M2RV99hQ!I32aHuZI7cW86Y3E969ZLyt4xH5nc2970bNrLwvfU1kc_rvxq1863brm4DAQrlmUr8XbLIuCX__gW7I6h23g$>

Paul; What do you think? And how does the process work?

Best
Martin



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/panama-dev/attachments/20230817/30024ff6/attachment-0001.htm>


More information about the panama-dev mailing list