On Mon, 28 Sep 2026 10:25:55 GMT, Aleksey Shipilev <[email protected]> wrote:
> Maybe that _also_ means we need to obsess about the PCR>1 overheads more, so
> that more workloads get on the right side of that balance. Although I do not
> see clearly how to do it.
This indeed looks like a path length issue more than anything.
Using this one as the canary test:
$ build/linux-x86_64-server-release/images/jdk/bin/java -jar
renaissance-jmh-0.16.1.jar -f 1 -prof perfasm --jvmArgsAppend
"-XX:TieredStopAtLevel=3 -XX:+UnlockExperimentalVMOptions
-XX:ProfileCaptureRatio=4" -wi 0 -i 1 JmhScrabble
# Old shape
2.03% 0x0000715bec40b3dc: movabs $0x715ba001b230,%rdi ;
{metadata(method data for {method} {0x00000000655a3ac0}
0.07% 0x0000715bec40b3e6: mov 0x70(%rdi),%ebx
2.01% 0x0000715bec40b3e9: lea 0x2(%rbx),%rbx
0.08% 0x0000715bec40b3ed: mov %ebx,0x70(%rdi)
0.43% 0x0000715bec40b3f0: and $0x1ffffe,%ebx
0.00% 0x0000715bec40b3f6: je 0x0000715bec40b4f1
0x0000715bec40b3fc: movabs $0x715ba001b230,%rdi ;
{metadata(method data for {method} {0x00000000655a3ac0}
0x0000715bec40b406: addq $0x1,0xa8(%rdi)
# New shape
0x000079c2e040f23b: movabs $0x79c29401b630,%rdi ;
{metadata(method data for {method} {0x00000000175a3ac0}
0.00% 0x000079c2e040f245: mov 0x70(%rdi),%ebx
1.97% 0x000079c2e040f248: or %rbx,%rbx
0.11% 0x000079c2e040f24b: je 0x000079c2e040f419
0.00% 0x000079c2e040f251: cmp $0x40000000,%r14d
0.00% 0x000079c2e040f258: jb 0x000079c2e040f419
0x000079c2e040f25e: xor %ebx,%ebx
0.01% 0x000079c2e040f260: crc32 %ebx,%r14d
0.00% 0x000079c2e040f266: movabs $0x79c29401b630,%rdi ;
{metadata(method data for {method} {0x00000000175a3ac0}
0x000079c2e040f270: mov 0xa8(%rdi),%rbx
2.29% 0x000079c2e040f277: or %rbx,%rbx
0.10% ╭ 0x000079c2e040f27a: je 0x000079c2e040f289
0.01% │ 0x000079c2e040f280: cmp $0x40000000,%r14d
0.00% │╭ 0x000079c2e040f287: jae 0x000079c2e040f291
↘│ 0x000079c2e040f289: addq $0x4,0xa8(%rdi)
0.47% ↘ 0x000079c2e040f291: xor %ebx,%ebx
0.01% 0x000079c2e040f293: crc32 %ebx,%r14d
-------------
PR Comment: https://git.openjdk.org/jdk/pull/28541#issuecomment-5868602726
