JDK-8299179 : ArrayFill with store on backedge needs to reduce length by 1
  • Type: Bug
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 16,17,19,20,21
  • Priority: P3
  • Status: In Progress
  • Resolution: Unresolved
  • Submitted: 2022-12-21
  • Updated: 2023-01-10
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
JDK 21
21Unresolved
Related Reports
Relates :  
JDK-8299808 - ArrayFill should be preferred over unrolling
Relates :  
JDK-8275047 - Optimize existing fill stubs for AVX-512 target
Relates :  
JDK-8247307 - C2: Loop array fill stub routines are not called
Description
Please change the generic bug title to something more descriptive once the root cause is known.

The attached Java Fuzzer test results in a wrong execution with C2 compared to a run with the interpreter/C1 after JDK-8247307 on aarch64 and x64:
- aarch64: fails after JDK-8247307 since -XX:+OptimizeFill is set by default
- x64: only fails with -XX:UseAVX=3 (and before JDK-8275047, -XX:+OptimizeFill must be set as it was false by default):

To reproduce:
$ java -Xint Test.java > interpreter.log
$ java -XX:-TieredCompilation -Xcomp -XX:CompileOnly=Test (-XX:UseAVX=3) Test.java > c2_test.log
$ java -XX:-TieredCompilation -Xcomp -XX:CompileOnly=Reduced (-XX:UseAVX=3) Reduced.java > c2_reduced.log

$ diff interpreter.log c2_test.log:
7c7
< vMeth1_check_sum: 585775794890188912
---
> vMeth1_check_sum: 585775794890188648
16c16
< vMeth1_check_sum: 1171551947176573060
---
> vMeth1_check_sum: 1171551947176572796
25c25
< vMeth1_check_sum: 1757328099462957208
---
> vMeth1_check_sum: 1757328099462956944
34c34
< vMeth1_check_sum: 2343104251749341356
---
> vMeth1_check_sum: 2343104251749341092
43c43
< vMeth1_check_sum: 2928880404035725504
---
> vMeth1_check_sum: 2928880404035725240
52c52
< vMeth1_check_sum: 3514656556322109652
---
> vMeth1_check_sum: 3514656556322109388
61c61
< vMeth1_check_sum: 4100432708608493800
---
> vMeth1_check_sum: 4100432708608493536
70c70
< vMeth1_check_sum: 4686208860894877948
---
> vMeth1_check_sum: 4686208860894877684
79c79
< vMeth1_check_sum: 5271985013181262096
---
> vMeth1_check_sum: 5271985013181261832
88c88
< vMeth1_check_sum: 5857761165467646244
---
> vMeth1_check_sum: 5857761165467645980

$ diff interpreter.log c2_reduced.log:
134c134
< iArrFld[63]: 0
---
> iArrFld[63]: 4
204c204
< iArrFld[63]: 0
---
> iArrFld[63]: 4
274c274
< iArrFld[63]: 0
---
> iArrFld[63]: 4
344c344
< iArrFld[63]: 0
---
> iArrFld[63]: 4
414c414
< iArrFld[63]: 0
---
> iArrFld[63]: 4
484c484
< iArrFld[63]: 0
---
> iArrFld[63]: 4
554c554
< iArrFld[63]: 0
---
> iArrFld[63]: 4
624c624
< iArrFld[63]: 0
---
> iArrFld[63]: 4
694c694
< iArrFld[63]: 0
---
> iArrFld[63]: 4
Comments
A pull request was submitted for review. URL: https://git.openjdk.org/jdk/pull/11904 Date: 2023-01-09 14:48:00 +0000
10-01-2023
JDK-8247307 has probably just revealed an existing issue as it just enables more patterns to use array fill stub.
21-12-2022
ILW = Wrong execution of C2 compiled array stub, single Java Fuzzer test, use -XX:-OptimizeFill = HLM = P3
21-12-2022