Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8237401

[Graal] 50% regression to C2 on crypto.small.CipherBench.GCM.decrypt

XMLWordPrintable

    • Icon: Enhancement Enhancement
    • Resolution: Duplicate
    • Icon: P4 P4
    • tbd
    • 13
    • hotspot

      This is similar to JDK-8213396, the main difference seems to be the non optimized clearing of the buffer in doFinal, see here:

      http://hg.openjdk.java.net/jdk/jdk14/file/cf3ce4ee50bf/src/java.base/share/classes/com/sun/crypto/provider/CipherCore.java

      at line 959.

      All the crypto intrinsics look good otherwise -

       19.11% runtime stub StubRoutines::counterMode_AESCrypt (1102 bytes)
        8.68% runtime stub StubRoutines::ghash_processBlocks (774 bytes)

            Unassigned Unassigned
            ecaspole Eric Caspole
            Votes:
            0 Vote for this issue
            Watchers:
            2 Start watching this issue

              Created:
              Updated:
              Resolved: