IJ12156: INCORRECT FLOATING POINT RESULT ON X86 MACHINES SUPPORTING AVX

APAR status

• Closed as program error.

Error description

• Error Message: The user will report an incorrect floating point
  math result when running on an x86 CPU which supports the AVX
  instruction set extensions. The bad result will be generated by
  a three operand AVX floating point instruction (usually an add
  or subtract).
  .
  Stack Trace: N/A
  .
  The problem can only occur on x86 hardware and only when running
  Java 8 SR5 FP18 through FP26 (8.0.5.18 - 8.0.5.26) when run on
  x86 machines which support the AVX instruction set extensions.
  

Local fix

• The problem is more likely to occur in cold compiles since some
  of the opportunities for the bug to occur are eliminated by
  higher optimization strategies. Use of -Xjit:dontDowngradeToCold
  may help reduce the frequency of errors but will not eliminate
  it. There is no workaround guaranteed to fully prevent the error
  

Problem summary

• In some circumstances the JIT code generator may incorrectly
  reload a value from memory rather than reusing a previously
  loaded result which can lead to different data being read and an
  incorrect result.
  

Problem conclusion

• The code generator was fixed to ensure that loads were not
  repeated incorrectly - the generated code sequence will now
  reuse the previously loaded value correctly.
  .
  This APAR will be fixed in the following Java Releases:
   8 SR5 FP27 (8.0.5.27)
  .
  Contact your IBM Product's Service Team for these Service
  Refreshes and Fix Packs.
  For those running stand-alone, information about the available
  Service Refreshes and Fix Packs can be found at:
   https://www.ibm.com/developerworks/java/jdk/
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

• Fixed component name

  JIT

• Fixed component ID

  620700124

Applicable component levels