I am following up on the performance issue where the culprit seems to be the too low OptoLoopAlignment. To perform better experiments, I suggest allowing larger alignments.
Note that we cannot make OptoLoopAlignment larger than CodeEntryAlignment, because nmethod copy would break it, see assert in `MacroAssembler::align`. SeeJDK-8273459 for latest discussion about it. So CodeEntryAlignment needs to be configurable as well.
No default value is changed, this only unblocks experiments.
Note that we cannot make OptoLoopAlignment larger than CodeEntryAlignment, because nmethod copy would break it, see assert in `MacroAssembler::align`. See
No default value is changed, this only unblocks experiments.
- relates to
-
JDK-8281829 runtime/CommandLine/OptionsValidation/TestOptionsWithRanges.java fails after JDK-8281467
-
- Resolved
-
-
JDK-8281936 compiler/arguments/TestCodeEntryAlignment.java fails on AVX512 machines
-
- Closed
-
-
JDK-8281586 x86: Default to larger OptoLoopAlignment
-
- Open
-
-
JDK-8282355 compiler/arguments/TestCodeEntryAlignment.java failed "guarantee(sect->end() <= tend) failed: sanity"
-
- Closed
-
-
JDK-8273459 Update code segment alignment to 64 bytes
-
- Resolved
-
(2 links to)