Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8049103

investigate whether inlining still needs to be inhibited in objectMonitor.cpp and synchronizer.cpp

    Details

    • Type: Bug
    • Status: Resolved
    • Priority: P4
    • Resolution: Not an Issue
    • Affects Version/s: 9
    • Fix Version/s: 11
    • Component/s: hotspot
    • Labels:
      None
    • Subcomponent:
    • CPU:
      generic
    • OS:
      linux

      Description

      The following issue came up during the code review cycle for the following:

      JDK-8047104 cleanup misc issues prior to Contended Locking
                      reorder and cache line bucket

      Both objectMonitor.cpp and synchronizer.cpp contain:

      #if defined(__GNUC__) && !defined(IA64) && !defined(PPC64)
        // Need to inhibit inlining for older versions of GCC to avoid build-time failures
        #define NOINLINE __attribute__((noinline))
      #else
        #define NOINLINE
      #endif

      and it would be good to determine if this is still needed.


      >> src/share/vm/runtime/objectMonitor.cpp
      >>
      >> Wondering if "Need to inhibit inlining for older versions of GCC to avoid build-time failures" is actually still relevant? Something for a later set of changes.
      >
      > Good question. I'll chase down the history on that one and try
      > to narrow down the GCC era/version where this used to happen.
      > Will probably result in a new follow up bug to propose removal
      > of the inhibit inlining stuff.

      Had to chase this back into the TeamWare history:

      $ sgv src/share/vm/runtime/synchronizer.cpp | grep noinline
      1.119
      1.69.1.6 #define ATTR __attribute__((noinline))
      4716 lines
      No id keywords (cm7)

      $ sp -r1.69.1.6 src/share/vm/runtime/synchronizer.cpp
      src/share/vm/runtime/SCCS/s.synchronizer.cpp:

      D 1.69.1.6 05/08/16 17:22:05 dice 150 149 00267/00163/03862
      MRs:
      COMMENTS:

      $ sccs sccsdiff -r1.69.1.{5,6} src/share/vm/runtime/synchronizer.cpp | head
      11a12,18
      > #if defined(__GNUC__)
      > // Need to inhibit inlining for older versions of GCC to avoid build-time failures
      > #define ATTR __attribute__((noinline))
      > #else
      > #define ATTR
      > #endif
      >

      so the noinline stuff was committed (sccs delget) back on
      August 16, 2005 in a delta without any bug ID. Sigh.

      The first release that contains that delta is JDK1.6.0 so
      here's the Linux compilers in use back then:

      $ strings linux-i586/jre/lib/i386/client/libjvm.so | grep 'with gcc'
      Java HotSpot(TM) Client VM (1.6.0-b105) for linux-x86, built on Nov 29 2006 01:24:38 by "java_re" with gcc 3.2.1-7a (J2SE release)

      $ strings linux-amd64/jre/lib/amd64/server/libjvm.so | grep 'with gcc'
      Java HotSpot(TM) 64-Bit Server VM (1.6.0-b105) for linux-amd64, built on Nov 29 2006 01:44:03 by "uucp" with gcc 3.2.2 (SuSE Linux)


      In contrast, for JDK8 we used the following on Linux:

      $ strings linux-i586/jre/lib/i386/client/libjvm.so | grep 'with gcc'
      Java HotSpot(TM) Client VM (25.0-b70) for linux-x86 JRE (1.8.0-b132), built on Mar 4 2014 03:27:19 by "java_re" with gcc 4.3.0 20080428 (Red Hat 4.3.0-8)

      $ strings linux-x64/jre/lib/amd64/server/libjvm.so | grep 'with gcc'
      Java HotSpot(TM) 64-Bit Server VM (25.0-b70) for linux-amd64 JRE (1.8.0-b132), built on Mar 4 2014 03:07:25 by "java_re" with gcc 4.3.0 20080428 (Red Hat 4.3.0-8)


      And in JDK9, we're using the following on Linux:

      $ strings linux-i586/jre/lib/i386/client/libjvm.so | grep 'with gcc'
      Java HotSpot(TM) Client VM (1.9.0-ea-b19) for linux-x86 JRE (1.9.0-ea-b19), built on Jun 19 2014 00:08:12 by "java_re" with gcc 4.8.2

      $ strings linux-x64/jre/lib/amd64/server/libjvm.so | grep 'with gcc'
      Java HotSpot(TM) 64-Bit Server VM (1.9.0-ea-b19) for linux-amd64 JRE (1.9.0-ea-b19), built on Jun 18 2014 23:57:36 by "java_re" with gcc 4.8.2

      So I'm thinking that I need to roll this historical info into a
      new bug for someone to investigate whether to remove the noinline
      stuff. Agreed?

      Dan

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                coleenp Coleen Phillimore
                Reporter:
                dcubed Daniel Daugherty
              • Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: