Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add and use a list of methods that can skip null value store check #18464

Merged

Conversation

a7ehuo
Copy link
Contributor

@a7ehuo a7ehuo commented Nov 16, 2023

  • Add a list recognized methods in j9method that can skip null
    check on the value stored to array elements.
  • ILGen: If array flattening is disabled and
    skipNonNullableArrayNullStoreCheck is true, avoid generating
    array element store non-helper calls.
  • VP: Avoid generating the nonNullableArrayNullStoreCheck if
    skipNonNullableArrayNullStoreCheck is true when
    transforming array element store non-helper calls.

This change depends on

@a7ehuo a7ehuo added project:valhalla Used to track Project Valhalla related work depends:omr Pull request is dependent on a corresponding change in OMR labels Nov 16, 2023
@a7ehuo a7ehuo marked this pull request as ready for review November 16, 2023 14:57
@a7ehuo a7ehuo changed the title Add a list of methods that can skip null value store check WIP: Add a list of methods that can skip null value store check Nov 16, 2023
@a7ehuo a7ehuo requested a review from hzongaro November 16, 2023 14:57
@a7ehuo a7ehuo force-pushed the add-skip-null-value-store-check-pr branch from 79f3f82 to e4965d2 Compare November 20, 2023 18:49
@a7ehuo a7ehuo changed the title WIP: Add a list of methods that can skip null value store check Add a list of methods that can skip null value store check Nov 23, 2023
@a7ehuo
Copy link
Contributor Author

a7ehuo commented Nov 23, 2023

@hzongaro Removed WIP. eclipse/omr#7184 is in openj9-omr. This PR is ready for review. Thanks!

@hzongaro hzongaro self-assigned this Nov 28, 2023
Copy link
Member

@hzongaro hzongaro left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I still have to do a detailed review of the new list of recognized methods, but I wanted to provide some initial feedback.

Beside the minor suggestion for Value Propagation, it occurs to me that in storeArrayElement in Walker.cpp, it should be possible to avoid generating a reference to the storeFlattenableArrayElementNonHelper if non-nullable array element flattening is not enabled and IL is being generated for a safeToSkipNonNullableArrayNullStoreCheck method.

Finally, your change doesn't only add the list of methods. I would suggest that you mention in the commit and pull request description that it includes changes to Value Propagation (and maybe IL generation) to take advantage of the list of methods for which it is safe to skip the nonNullableArrayNullStoreCheck.

runtime/compiler/optimizer/J9ValuePropagation.cpp Outdated Show resolved Hide resolved
@a7ehuo a7ehuo force-pushed the add-skip-null-value-store-check-pr branch from e4965d2 to 84b8f38 Compare November 30, 2023 20:03
@a7ehuo a7ehuo requested a review from dsouzai as a code owner November 30, 2023 20:03
@a7ehuo a7ehuo changed the title Add a list of methods that can skip null value store check Add and use a list of methods that can skip null value store check Nov 30, 2023
@a7ehuo
Copy link
Contributor Author

a7ehuo commented Nov 30, 2023

... it should be possible to avoid generating a reference to the storeFlattenableArrayElementNonHelper ..

Great point!

All above comments are addressed in 84b8f38. Ready for another review. Thank you!

Copy link
Member

@hzongaro hzongaro left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for implementing this. I'm going to add a reference to this pull request in issue #17516, as its touches on some of the same problems with references to methods that no longer exist or methods that don't actually have array element stores that are already included in canSkipArrayStoreChecks.

runtime/compiler/il/J9MethodSymbol.cpp Outdated Show resolved Hide resolved
runtime/compiler/il/J9MethodSymbol.cpp Show resolved Hide resolved
runtime/compiler/il/J9MethodSymbol.cpp Outdated Show resolved Hide resolved
runtime/compiler/il/J9MethodSymbol.cpp Outdated Show resolved Hide resolved
runtime/compiler/il/J9MethodSymbol.cpp Show resolved Hide resolved
- Add a list recognized methods in j9method that can skip null
  check on the value stored to array elements.
- ILGen: If array flattening is disabled and
  `skipNonNullableArrayNullStoreCheck` is true, avoid generating
  array element store non-helper calls.
- VP: Avoid generating the `nonNullableArrayNullStoreCheck` if
  `skipNonNullableArrayNullStoreCheck` is true when
  transforming array element store non-helper calls.

Signed-off-by: Annabelle Huo <[email protected]>
@a7ehuo a7ehuo force-pushed the add-skip-null-value-store-check-pr branch from 84b8f38 to 066df6b Compare December 6, 2023 15:38
@a7ehuo
Copy link
Contributor Author

a7ehuo commented Dec 6, 2023

@hzongaro All comments are addressed. Ready for another review. Thanks!

@hzongaro
Copy link
Member

hzongaro commented Dec 6, 2023

Thinking about the discussion around ComparableTimSort a bit more, it occurs to me that avoiding a run-time check to see whether the component type of the array is null-restricted is only one potential benefit that can be derived from knowing what kinds of arrays a recognized method works with. The other potential benefit would to avoid unnecessary run-time checks of whether the elements of the arrays are flattened, and so avoid generating calls to storeFlattenableArrayElementNonHelper and loadFlattenableArrayElementNonHelper altogether.

As things stand, I think ComparableTimSort and TimSort contain methods that can avoid nonNullableArrayNullStoreCheck, but would still need storeFlattenableArrayElementNonHelper and loadFlattenableArrayElementNonHelper calls. The various toArray([Object;)[Object; methods in the collections classes might end up being in a similar situation, if the RI does something to eliminate the explicit stores of a null reference. I think that for all the other methods, both nonNullableArrayNullStoreCheck and {load|store}FlattenableArrayElementNonHelper calls could be avoided.

Thoughts? If you agree, then perhaps that could be handled as a follow-on pull request.

@a7ehuo
Copy link
Contributor Author

a7ehuo commented Dec 6, 2023

The other potential benefit would to avoid unnecessary run-time checks of whether the elements of the arrays are flattened, and so avoid generating calls to storeFlattenableArrayElementNonHelper and loadFlattenableArrayElementNonHelper altogether. ...
Thoughts? If you agree, then perhaps that could be handled as a follow-on pull request.

That's a great idea! If I understand it correctly, we could have another list of recognized methods that can skip generating the non helpers for flattenable arrays in ILGen, such as canSkipHelperCallForFlattenableArrays []. I agree it should be in a separate change/PR.

Copy link
Member

@hzongaro hzongaro left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the updates! Looks good.

May I ask you to open a follow-on issue as a reminder that we will have to deal with the various T[] xxx.toArray(T[]) methods?

@hzongaro
Copy link
Member

hzongaro commented Dec 7, 2023

Jenkins test sanity xlinuxval,xlinuxvalst jdknext

@a7ehuo
Copy link
Contributor Author

a7ehuo commented Dec 7, 2023

May I ask you to open a follow-on issue as a reminder that we will have to deal with the various T[] xxx.toArray(T[]) methods?

#18576 is created to track this work item

@hzongaro
Copy link
Member

hzongaro commented Dec 7, 2023

I forgot that I shouldn't be using xlinux for PR testing.

Jenkins test sanity alinuxval,alinuxvalst jdknext

@pshipton
Copy link
Member

pshipton commented Dec 7, 2023

I forgot that I shouldn't be using xlinux for PR testing.

Actually it might be working again soon. Plus a little bit of testing is fine, it's when there is too much that jenkins can restart or tests can time out. Adoptium have almost fixed the problem of downloading 100MB+ of data for every test run. There was one problem this morning with a PR open to fix it, still pending merge atm.

@pshipton
Copy link
Member

pshipton commented Dec 7, 2023

Your xlinux testing did run. Starting a new PR build doesn't cancel any old builds.

@hzongaro
Copy link
Member

hzongaro commented Dec 7, 2023

Your xlinux testing did run. Starting a new PR build doesn't cancel any old builds.

Right, but it had failed with some unexpected errors that were likely unrelated to this PR. When I went to rerun, I realized I ought not rerun on xlinux.

@hzongaro
Copy link
Member

hzongaro commented Dec 8, 2023

This shouldn't affect anything with value types support disabled, but just in case:

Jenkins test sanity all jdk17

@a7ehuo
Copy link
Contributor Author

a7ehuo commented Dec 11, 2023

openjdk17_j9_sanity.functional_x86-64_windows failed with the following error. It doesn't seem related to this PR.

12:41:56  -----------------------------------
12:41:56  cmdLineTester_AutoModuleClassloaderTest_0_PASSED
12:41:56  -----------------------------------
...
12:41:57  Attempting to destroy all caches in cacheDir C:\Users\jenkins.001\AppData\Local\javasharedresources\
12:41:57  
12:41:57  JVMSHRC241E Error: unable to delete shared class cache file
12:41:57  JVMSHRC336E Port layer error code = -100
12:41:57  JVMSHRC337E Platform error message: (32) The process cannot access the file because it is being used by another process.
12:41:57  JVMSHRC430I Failed to remove current generation of shared class cache "jenkins"
...
12:42:02  F:\Users\jenkins\workspace\Test_openjdk17_j9_sanity.functional_x86-64_windows_Personal_testList_1\aqa-tests\TKG/../TKG/settings.mk:351: \\TKG\\openj9Settings.mk: No such file or directory
12:42:02  make[6]: *** No rule to make target '\\TKG\\openj9Settings.mk'.  Stop.
...
12:42:03  Could not find test result, set build result to FAILURE.

@hzongaro
Copy link
Member

Now that Pull request #18584 has been merged, Valhalla standard testing should be more successful

Jenkins test sanity alinuxvalst jdknext

@hangshao0
Copy link
Contributor

The "access" shared library load error in Valhalla standard build should be fixed by: ibmruntimes/openj9-openjdk-jdk.valuetypes@89c45d9. I see @JasonFengJ9 pushed the fix in 2 hours ago.

@JasonFengJ9
Copy link
Member

The "access" shared library load error in Valhalla standard build should be fixed by: ibmruntimes/openj9-openjdk-jdk.valuetypes@89c45d9. I see @JasonFengJ9 pushed the fix in 2 hours ago.

Yeah, I am running a verification build https://openj9-jenkins.osuosl.org/job/Build_JDKnext_x86-64_windows_vt_standard_Personal/5/console which is waiting for a machine.

@hzongaro
Copy link
Member

Jenkins test sanity alinuxvalst jdknext

@hzongaro
Copy link
Member

Failure in aarch64 Linux vt_standard testing appears to be due to issue #17844. Merging.

@hzongaro hzongaro merged commit a3c7ae9 into eclipse-openj9:master Dec 13, 2023
17 of 21 checks passed
@a7ehuo a7ehuo deleted the add-skip-null-value-store-check-pr branch March 6, 2024 16:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
depends:omr Pull request is dependent on a corresponding change in OMR project:valhalla Used to track Project Valhalla related work
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants