Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8074383

Redundant "compare" in generated code for String.charAt bounds checks

    Details

    • Type: Enhancement
    • Status: Resolved
    • Priority: P4
    • Resolution: Fixed
    • Affects Version/s: None
    • Fix Version/s: None
    • Component/s: hotspot
    • Labels:
      None

      Description

      If you run a simple benchmark like this:

          private String s;
          private int idx;

          @Setup
          public void setup() {
              idx = 13;
              s = "Improve the scope of range check optimizations";
          }

          @Benchmark
          public void test_charAt() {
              trampoline_charAt();
          }

          @CompilerControl(CompilerControl.Mode.DONT_INLINE)
          public char trampoline_charAt() {
              return s.charAt(idx);
          }

      ...then you will see this generated code with the latest jdk9-dev:

       mov %eax,-0x14000(%rsp)
       push %rbp
       sub $0x20,%rsp
       mov 0xc(%rsi),%r11d ; get field $idx
       mov 0x10(%rsi),%r10d ; get field $s
       mov 0xc(%r12,%r10,8),%r9d ; get field $s.value
       test %r11d,%r11d
       jl 0x00007f6204b5cc19 ; range check 1, jump if ($idx < 0)
       mov 0xc(%r12,%r9,8),%ebp ; get $s.value.arraylength
       cmp %ebp,%r11d
       jge 0x00007f6204b5cc31 ; range check 2, jump if ($idx >= arraylength)
       cmp %ebp,%r11d ; <--- REDUNDANT COMPARE
       jae 0x00007f6204b5cc03 ; jump if overflow?
       lea (%r12,%r9,8),%r10 ; unpack $s.value array reference
       movzwl 0x10(%r10,%r11,2),%eax ; get $s.value[$idx]
       add $0x20,%rsp ; epilogue and return
       pop %rbp
       test %eax,0x15f1a3fe(%rip)
       retq

      As you can see, we do double compare on the same arguments, even though jumps are not affecting the flags, and there seem to be no incoming branches to the second cmp.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                shade Aleksey Shipilev
                Reporter:
                shade Aleksey Shipilev
              • Votes:
                0 Vote for this issue
                Watchers:
                2 Start watching this issue

                Dates

                • Created:
                  Updated:
                  Resolved: