Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8213264

[Graal] 95% regression to C2 on AESGCMBench.decrypt AES/GCM/NoPadding

XMLWordPrintable

    • Icon: Enhancement Enhancement
    • Resolution: Fixed
    • Icon: P4 P4
    • 14
    • 12
    • hotspot
    • b25

       The open corpus of JMH micros is a maven project here: http://hg.openjdk.java.net/code-tools/jmh-jdk-microbenchmarks

      It contains various crypto micros. AESGCMBench.decrypt AES/GCM/NoPadding shows a case with an intrinsic for C2 where Graal does not yet have one and uses the jitted version, so one test system the score is:
      C2:
      Benchmark (algorithm) (dataSize) (keyLength) (provider) Mode Cnt Score Error Units
      AESGCMBench.decrypt AES/GCM/NoPadding 1024 128 thrpt 8 350500.214 ± 23761.865 ops/s
      Graal:
      Benchmark (algorithm) (dataSize) (keyLength) (provider) Mode Cnt Score Error Units
      AESGCMBench.decrypt AES/GCM/NoPadding 1024 128 thrpt 8 23644.262 ± 13400.185 ops/s

      You can see the difference by using JMH linux perfasm with hsdis, I will attach profiles later.

            jphalimi Jean-Philippe Halimi (Inactive)
            ecaspole Eric Caspole
            Votes:
            0 Vote for this issue
            Watchers:
            8 Start watching this issue

              Created:
              Updated:
              Resolved: