JDK-8303827 : C2 SuperWord: allow more fine grained alignment for +AlignVector
  • Type: Enhancement
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 21
  • Priority: P4
  • Status: Closed
  • Resolution: Duplicate
  • Submitted: 2023-03-08
  • Updated: 2024-01-11
  • Resolved: 2024-01-11
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.
Other
tbdResolved
Related Reports
Blocks :  
JDK-8298935 - fix independence bug in create_pack logic in SuperWord::find_adjacent_refs
Duplicate :  
JDK-8310190 - C2 SuperWord: AlignVector is broken, generates misaligned packs
Relates :  
JDK-8305055 - IR check fails on some aarch64 platforms
Relates :  
JDK-8310190 - C2 SuperWord: AlignVector is broken, generates misaligned packs
Description
Platforms with alignment requirement do not need to have alignment modulo vector_width, that is too restrictive. It prevents vectorization of some simple conversion tests. Platforms often only require 4-byte or 8-byte alignment.

This is more relevant after JDK-8298935 (where we disable "resurrecting" rejected memops). This prevents certain conversion-tests to fail under +AlignVector. For example some tests in TestVectorizeTypeConversion.java.
Comments
Next to bugfix, I also improved the alignment constraints
11-01-2024
JDK-8310190 made alignment more fine-grained: We should look into what tests can now have AlignVector = false removed from the IR test rules. And maybe we see some cases that are still not possible.
10-01-2024
We should re-enable the IR checks that were disabled for "+AlignVector" in JDK-8305055.
30-03-2023
Hi [~epeter], I only have ThunderX2 machines, but I checked that "+AlignVector" involves other old aarch64 cpus and also some x86 machines. I don't have enough confidence unless we can do a thorough test on all these different machines. Thanks.
14-03-2023
[~fgao] Would you be interested in doing this, or know anybody else who is interested?
09-03-2023