Uploaded image for project: 'JDK'
  1. JDK
  2. JDK-8190934

Regressions on Haswell Xeon due to JDK-8178811

    XMLWordPrintable

    Details

    • Subcomponent:
    • Resolved In Build:
      b37
    • CPU:
      x86_64

      Backports

        Description

        I tracked down various regressions in microbenchmarks on my Haswell-gen workstation (Intel(R) Xeon(R) CPU E5-2630 v3) to be due to the vzeroupper instructions emitted for performance reasons in JDK-8178811. It needs to be examined if the optimization in JDK-8178811 is profitable on pre-Skylake CPUs, and if so if we can improve on when and where the vzeroupper instructions are emitted.

        Example:

        Benchmark Mode Cnt Score Error Units
        Clazz.getClassLoader avgt 5 11.605 ± 0.350 ns/op # 10
        Clazz.getClassLoader avgt 5 7.802 ± 0.323 ns/op # 9 GA (similar score with -XX:UseAVX=0)

        package org.openjdk;

        import org.openjdk.jmh.annotations.Benchmark;
        import org.openjdk.jmh.annotations.BenchmarkMode;
        import org.openjdk.jmh.annotations.Mode;
        import org.openjdk.jmh.annotations.OutputTimeUnit;
        import org.openjdk.jmh.annotations.CompilerControl;

        import java.util.concurrent.TimeUnit;

        @BenchmarkMode(Mode.AverageTime)
        @OutputTimeUnit(TimeUnit.NANOSECONDS)
        public class Clazz {

            public static Class<?> c1 = Clazz.class;

            @Benchmark
            @CompilerControl(CompilerControl.Mode.DONT_INLINE)
            public ClassLoader getClassLoader() {
                return c1.getClassLoader();
            }
        }

        -prof perfasm on 9 or with -XX:UseAVX=0:

                               # {method} {0x00007f5804680aa0} &apos;getClassLoader&apos; &apos;()Ljava/lang/ClassLoader;&apos; in &apos;org/openjdk/Clazz&apos;
                               # [sp+0x30] (sp of caller)
                               0x00007f590176c7a0: mov 0x8(%rsi),%r10d
                               0x00007f590176c7a4: shl $0x3,%r10
                               0x00007f590176c7a8: cmp %r10,%rax
                               0x00007f590176c7ab: jne 0x00007f58f9b94000 ; {runtime_call ic_miss_stub}
                               0x00007f590176c7b1: xchg %ax,%ax
                               0x00007f590176c7b4: nopl 0x0(%rax,%rax,1)
                               0x00007f590176c7bc: xchg %ax,%ax
                             [Verified Entry Point]
          2.29% 2.33% 0x00007f590176c7c0: mov %eax,-0x14000(%rsp)
          4.86% 4.19% 0x00007f590176c7c7: push %rbp
          1.10% 1.20% 0x00007f590176c7c8: sub $0x20,%rsp ;*synchronization entry
                                                                             ; - org.openjdk.Clazz::getClassLoader@-1 (line 56)
          2.16% 2.24% 0x00007f590176c7cc: mov $0x5e95a20b8,%r10 ; {oop(a &apos;java/lang/Class&apos;{0x00000005e95a20b8} = &apos;org/openjdk/Clazz&apos;)}
          2.55% 2.29% 0x00007f590176c7d6: mov 0x70(%r10),%r11d ;*getstatic c1 {reexecute=0 rethrow=0 return_oop=0}
                                                                             ; - org.openjdk.Clazz::getClassLoader@0 (line 56)
          0.88% 1.01% 0x00007f590176c7da: mov 0x1c(%r12,%r11,8),%r11d ;*getfield classLoader {reexecute=0 rethrow=0 return_oop=0}
                                                                             ; - java.lang.Class::getClassLoader0@1 (line 811)
                                                                             ; - java.lang.Class::getClassLoader@1 (line 800)
                                                                             ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
                                                                             ; implicit exception: dispatches to 0x00007f590176c81d
          2.68% 2.77% 0x00007f590176c7df: test %r11d,%r11d
                          ╭ 0x00007f590176c7e2: je 0x00007f590176c806 ;*ifnonnull {reexecute=0 rethrow=0 return_oop=0}
                          │ ; - java.lang.Class::getClassLoader@6 (line 801)
                          │ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          1.51% 1.64% │ 0x00007f590176c7e4: mov $0x5e9a00d60,%r10 ; {oop(a &apos;java/lang/Class&apos;{0x00000005e9a00d60} = &apos;java/lang/System&apos;)}
          2.68% 2.85% │ 0x00007f590176c7ee: mov 0x7c(%r10),%ebp ;*getstatic security {reexecute=0 rethrow=0 return_oop=0}
                          │ ; - java.lang.System::getSecurityManager@0 (line 366)
                          │ ; - java.lang.Class::getClassLoader@11 (line 803)
                          │ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          0.76% 0.85% │ 0x00007f590176c7f2: test %ebp,%ebp
                          │╭ 0x00007f590176c7f4: jne 0x00007f590176c80a ;*ifnull {reexecute=0 rethrow=0 return_oop=0}
                          ││ ; - java.lang.Class::getClassLoader@16 (line 804)
                          ││ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          1.48% 1.59% ││ 0x00007f590176c7f6: lea (%r12,%r11,8),%rax ;*invokevirtual getClassLoader {reexecute=0 rethrow=0 return_oop=0}
                          ││ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          1.13% 0.97% ││↗ 0x00007f590176c7fa: add $0x20,%rsp
          2.37% 2.18% │││ 0x00007f590176c7fe: pop %rbp
          2.28% 2.50% │││ 0x00007f590176c7ff: test %eax,0x1bd9b7fb(%rip) # 0x00007f591d508000
                          │││ ; {poll_return}
          1.87% 2.13% │││ 0x00007f590176c805: retq
                          ↘││ 0x00007f590176c806: xor %eax,%eax
                           │╰ 0x00007f590176c808: jmp 0x00007f590176c7fa
                           ↘ 0x00007f590176c80a: mov $0xffffff5d,%esi
                               0x00007f590176c80f: mov %r11d,(%rsp)
                               0x00007f590176c813: callq 0x00007f58f9b95c00 ; ImmutableOopMap{rbp=NarrowOop [0]=NarrowOop }
                                                                             ;*ifnull {reexecute=1 rethrow=0 return_oop=0}
                                                                             ; - java.lang.Class::getClassLoader@16 (line 804)
                                                                             ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
                                                                             ; {runtime_call UncommonTrapBlob}


        10 ASM:
                              # {method} {0x00007f3f6d8aa640} &apos;getClassLoader&apos; &apos;()Ljava/lang/ClassLoader;&apos; in &apos;org/openjdk/Clazz&apos;
                               # [sp+0x30] (sp of caller)
                               0x00007f40658f14a0: mov 0x8(%rsi),%r10d
                               0x00007f40658f14a4: shl $0x3,%r10
                               0x00007f40658f14a8: cmp %r10,%rax
                               0x00007f40658f14ab: jne 0x00007f405dcfa400 ; {runtime_call ic_miss_stub}
                               0x00007f40658f14b1: xchg %ax,%ax
                               0x00007f40658f14b4: nopl 0x0(%rax,%rax,1)
                               0x00007f40658f14bc: xchg %ax,%ax
                             [Verified Entry Point]
          2.53% 3.10% 0x00007f40658f14c0: mov %eax,-0x14000(%rsp)
          4.92% 5.72% 0x00007f40658f14c7: push %rbp
          2.25% 3.08% 0x00007f40658f14c8: sub $0x20,%rsp ;*synchronization entry
                                                                             ; - org.openjdk.Clazz::getClassLoader@-1 (line 56)
          1.35% 2.12% 0x00007f40658f14cc: mov $0x5e95d34d0,%r10 ; {oop(a &apos;java/lang/Class&apos;{0x00000005e95d34d0} = &apos;org/openjdk/Clazz&apos;)}
          1.65% 2.14% 0x00007f40658f14d6: mov 0x70(%r10),%r11d ;*getstatic c1 {reexecute=0 rethrow=0 return_oop=0}
                                                                             ; - org.openjdk.Clazz::getClassLoader@0 (line 56)
          0.95% 1.27% 0x00007f40658f14da: mov 0x1c(%r12,%r11,8),%r11d ;*getfield classLoader {reexecute=0 rethrow=0 return_oop=0}
                                                                             ; - java.lang.Class::getClassLoader0@1 (line 814)
                                                                             ; - java.lang.Class::getClassLoader@1 (line 803)
                                                                             ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
                                                                             ; implicit exception: dispatches to 0x00007f40658f1522
          2.30% 3.11% 0x00007f40658f14df: test %r11d,%r11d
                          ╭ 0x00007f40658f14e2: je 0x00007f40658f1509 ;*ifnonnull {reexecute=0 rethrow=0 return_oop=0}
                          │ ; - java.lang.Class::getClassLoader@6 (line 804)
                          │ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          1.23% 1.35% │ 0x00007f40658f14e4: mov $0x5e9a00d60,%r10 ; {oop(a &apos;java/lang/Class&apos;{0x00000005e9a00d60} = &apos;java/lang/System&apos;)}
          1.29% 1.44% │ 0x00007f40658f14ee: mov 0x7c(%r10),%ebp ;*getstatic security {reexecute=0 rethrow=0 return_oop=0}
                          │ ; - java.lang.System::getSecurityManager@0 (line 370)
                          │ ; - java.lang.Class::getClassLoader@11 (line 806)
                          │ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          1.21% 1.14% │ 0x00007f40658f14f2: test %ebp,%ebp
                          │╭ 0x00007f40658f14f4: jne 0x00007f40658f150d ;*ifnull {reexecute=0 rethrow=0 return_oop=0}
                          ││ ; - java.lang.Class::getClassLoader@16 (line 807)
                          ││ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          0.94% 0.99% ││ 0x00007f40658f14f6: lea (%r12,%r11,8),%rax ;*invokevirtual getClassLoader {reexecute=0 rethrow=0 return_oop=0}
                          ││ ; - org.openjdk.Clazz::getClassLoader@3 (line 56)
          1.06% 1.71% ││↗ 0x00007f40658f14fa: vzeroupper
          4.46% 4.13% │││ 0x00007f40658f14fd: add $0x20,%rsp
          1.24% 0.74% │││ 0x00007f40658f1501: pop %rbp
          1.78% 1.66% │││ 0x00007f40658f1502: test %eax,0x1bde8af8(%rip) # 0x00007f40816da000
                          │││ ; {poll_return}
          1.01% 1.24% │││ 0x00007f40658f1508: retq
                          ↘││ 0x00007f40658f1509: xor %eax,%eax
                           │╰ 0x00007f40658f150b: jmp 0x00007f40658f14fa
                           ↘ 0x00007f40658f150d: mov $0xffffff55,%esi
                               0x00007f40658f1512: mov %r11d,(%rsp)
                               0x00007f40658f1516: xchg %ax,%ax
                               0x00007f40658f1518: vzeroupper
                               0x00007f40658f151b: callq 0x00007f405dcfc080 ; ImmutableOopMap{rbp=NarrowOop [0]=NarrowOop }
                                                                             ;*ifnull {reexecute=1 rethrow=0 return_oop=0}
                                                                             ; - java.lang.Class::getClassLoader@16 (line 807)

          Attachments

            Issue Links

              Activity

                People

                Assignee:
                vdeshpande Vivek Deshpande
                Reporter:
                redestad Claes Redestad
                Votes:
                0 Vote for this issue
                Watchers:
                8 Start watching this issue

                  Dates

                  Created:
                  Updated:
                  Resolved: