Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-4755487

(-O) dividing and multiplying powers of two versus shift left and right

    Details

    • Subcomponent:
    • CPU:
      x86
    • OS:
      linux

      Description



      Name: gm110360 Date: 09/30/2002


      FULL PRODUCT VERSION :
      java version "1.4.0_01"
      Java(TM) 2 Runtime Environment, Standard Edition (build 1.4.0_01-b03)
      Java HotSpot(TM) Client VM (build 1.4.0_01-b03, mixed mode)


      FULL OPERATING SYSTEM VERSION :
      Linux Slackware 8.1

      ADDITIONAL OPERATING SYSTEMS :
      All (I think)

      A DESCRIPTION OF THE PROBLEM :
      When dividing or multiplying by an inline constant (or an
      final constant in another class), and this constant is a
      power of two, a trivial (though very helpful) optimalisation
      could be applied:
      Check if the constant is an power of two, for example with
      (x & (x-1)) == 0 . If so, compile it to shift left(multiply)
      or shift right (divide) with the two-log of that number.
      Of course, this optimalisation can be done in the VM, but I
      think that's not the case right now. (judging from the great
      speed increase in my applet when I abandoned / for >>)


      STEPS TO FOLLOW TO REPRODUCE THE PROBLEM :
      1.Compile the source code fragment
      2."javap -c" it
      3.Check the decompiled code

      EXPECTED VERSUS ACTUAL BEHAVIOR :
      actual:
      /128 gets compiled to idiv 128
      *128 gets compiled to imul 128
      "perfect":
      /128 gets compiled to ishr 7 (ishr = with sign extension)
      *128 gets compiled to ishl 7


      REPRODUCIBILITY :
      This bug can be reproduced always.

      ---------- BEGIN SOURCE ----------
      /* generated javac output in comments */
      class bla {

      /*
      Method void main(java.lang.String[])
      */
      public static void main(String args[]) {
      /*
      0 bipush 100
          2 istore_1
      */
      int i = 100;
      /*
      3 iload_1
          4 sipush 128
          7 idiv
          8 istore_1
      */
      i = i / 128;
      /*
      9 iload_1
         10 sipush 128
         13 imul
         14 istore_1
      */
      i = i * 128;
      }

      }

      ---------- END SOURCE ----------
      (Review ID: 165087)
      ======================================================================

        Attachments

          Activity

            People

            • Assignee:
              darcy Joe Darcy
              Reporter:
              gmanwanisunw Girish Manwani (Inactive)
            • Votes:
              0 Vote for this issue
              Watchers:
              0 Start watching this issue

              Dates

              • Created:
                Updated:
                Resolved:
                Imported:
                Indexed: