src/hotspot/share/gc/shenandoah/shenandoahRootProcessor.cpp

rev 57866 : 8246100: Shenandoah: walk roots in more efficient order
Reviewed-by: zgu

*** 177,203 ****
    CodeBlobToOopClosure* codes_cl = ShenandoahConcurrentRoots::can_do_concurrent_class_unloading() ?
                                     static_cast<CodeBlobToOopClosure*>(&blobs_and_disarm_Cl) :
                                     static_cast<CodeBlobToOopClosure*>(&blobsCl);
    AlwaysTrueClosure always_true;
  
    _serial_roots.oops_do(oops, worker_id);
    _serial_weak_roots.weak_oops_do(oops, worker_id);
    if (_stw_roots_processing) {
      _vm_roots.oops_do<OopClosure>(oops, worker_id);
      _weak_roots.oops_do<OopClosure>(oops, worker_id);
    }
  
    if (_stw_class_unloading) {
      CLDToOopClosure clds(oops, ClassLoaderData::_claim_strong);
      _cld_roots.cld_do(&clds, worker_id);
      _code_roots.code_blobs_do(codes_cl, worker_id);
      _thread_roots.oops_do(oops, NULL, worker_id);
    } else {
      _thread_roots.oops_do(oops, codes_cl, worker_id);
    }
- 
-   _dedup_roots.oops_do(&always_true, oops, worker_id);
  }
  
  ShenandoahRootUpdater::ShenandoahRootUpdater(uint n_workers, ShenandoahPhaseTimings::Phase phase) :
    ShenandoahRootProcessor(phase),
    _serial_roots(phase),
--- 177,206 ----
    CodeBlobToOopClosure* codes_cl = ShenandoahConcurrentRoots::can_do_concurrent_class_unloading() ?
                                     static_cast<CodeBlobToOopClosure*>(&blobs_and_disarm_Cl) :
                                     static_cast<CodeBlobToOopClosure*>(&blobsCl);
    AlwaysTrueClosure always_true;
  
+   // Process serial-claiming roots first
    _serial_roots.oops_do(oops, worker_id);
    _serial_weak_roots.weak_oops_do(oops, worker_id);
+ 
+   // Process light-weight/limited parallel roots then
    if (_stw_roots_processing) {
      _vm_roots.oops_do<OopClosure>(oops, worker_id);
      _weak_roots.oops_do<OopClosure>(oops, worker_id);
    }
+   _dedup_roots.oops_do(&always_true, oops, worker_id);
  
+   // Process heavy-weight/fully parallel roots the last
    if (_stw_class_unloading) {
      CLDToOopClosure clds(oops, ClassLoaderData::_claim_strong);
      _cld_roots.cld_do(&clds, worker_id);
      _code_roots.code_blobs_do(codes_cl, worker_id);
      _thread_roots.oops_do(oops, NULL, worker_id);
    } else {
      _thread_roots.oops_do(oops, codes_cl, worker_id);
    }
  }
  
  ShenandoahRootUpdater::ShenandoahRootUpdater(uint n_workers, ShenandoahPhaseTimings::Phase phase) :
    ShenandoahRootProcessor(phase),
    _serial_roots(phase),
*** 230,249 ****
                                                static_cast<CodeBlobToOopClosure*>(&blobs_and_disarm_Cl) :
                                                static_cast<CodeBlobToOopClosure*>(&code_blob_cl);
    CLDToOopClosure adjust_cld_closure(oops, ClassLoaderData::_claim_strong);
    AlwaysTrueClosure always_true;
  
    _serial_roots.oops_do(oops, worker_id);
    _vm_roots.oops_do(oops, worker_id);
  
!   _thread_roots.oops_do(oops, NULL, worker_id);
    _cld_roots.cld_do(&adjust_cld_closure, worker_id);
    _code_roots.code_blobs_do(adjust_code_closure, worker_id);
! 
!   _serial_weak_roots.weak_oops_do(oops, worker_id);
!   _weak_roots.oops_do<OopClosure>(oops, worker_id);
!   _dedup_roots.oops_do(&always_true, oops, worker_id);
  }
  
  ShenandoahHeapIterationRootScanner::ShenandoahHeapIterationRootScanner() :
     ShenandoahRootProcessor(ShenandoahPhaseTimings::heap_iteration_roots),
     _serial_roots(ShenandoahPhaseTimings::heap_iteration_roots),
--- 233,255 ----
                                                static_cast<CodeBlobToOopClosure*>(&blobs_and_disarm_Cl) :
                                                static_cast<CodeBlobToOopClosure*>(&code_blob_cl);
    CLDToOopClosure adjust_cld_closure(oops, ClassLoaderData::_claim_strong);
    AlwaysTrueClosure always_true;
  
+   // Process serial-claiming roots first
    _serial_roots.oops_do(oops, worker_id);
+   _serial_weak_roots.weak_oops_do(oops, worker_id);
+ 
+   // Process light-weight/limited parallel roots then
    _vm_roots.oops_do(oops, worker_id);
+   _weak_roots.oops_do<OopClosure>(oops, worker_id);
+   _dedup_roots.oops_do(&always_true, oops, worker_id);
  
!   // Process heavy-weight/fully parallel roots the last
    _cld_roots.cld_do(&adjust_cld_closure, worker_id);
    _code_roots.code_blobs_do(adjust_code_closure, worker_id);
!   _thread_roots.oops_do(oops, NULL, worker_id);
  }
  
  ShenandoahHeapIterationRootScanner::ShenandoahHeapIterationRootScanner() :
     ShenandoahRootProcessor(ShenandoahPhaseTimings::heap_iteration_roots),
     _serial_roots(ShenandoahPhaseTimings::heap_iteration_roots),
*** 261,277 ****
     // Must use _claim_none to avoid interfering with concurrent CLDG iteration
     CLDToOopClosure clds(oops, ClassLoaderData::_claim_none);
     MarkingCodeBlobClosure code(oops, !CodeBlobToOopClosure::FixRelocations);
     ShenandoahParallelOopsDoThreadClosure tc_cl(oops, &code, NULL);
     AlwaysTrueClosure always_true;
     ResourceMark rm;
  
     _serial_roots.oops_do(oops, 0);
-    _vm_roots.oops_do(oops, 0);
-    _cld_roots.cld_do(&clds, 0);
-    _thread_roots.threads_do(&tc_cl, 0);
-    _code_roots.code_blobs_do(&code, 0);
- 
     _serial_weak_roots.weak_oops_do(oops, 0);
     _weak_roots.oops_do<OopClosure>(oops, 0);
     _dedup_roots.oops_do(&always_true, oops, 0);
   }
--- 267,288 ----
     // Must use _claim_none to avoid interfering with concurrent CLDG iteration
     CLDToOopClosure clds(oops, ClassLoaderData::_claim_none);
     MarkingCodeBlobClosure code(oops, !CodeBlobToOopClosure::FixRelocations);
     ShenandoahParallelOopsDoThreadClosure tc_cl(oops, &code, NULL);
     AlwaysTrueClosure always_true;
+ 
     ResourceMark rm;
  
+    // Process serial-claiming roots first
     _serial_roots.oops_do(oops, 0);
     _serial_weak_roots.weak_oops_do(oops, 0);
+ 
+    // Process light-weight/limited parallel roots then
+    _vm_roots.oops_do(oops, 0);
     _weak_roots.oops_do<OopClosure>(oops, 0);
     _dedup_roots.oops_do(&always_true, oops, 0);
+ 
+    // Process heavy-weight/fully parallel roots the last
+    _cld_roots.cld_do(&clds, 0);
+    _code_roots.code_blobs_do(&code, 0);
+    _thread_roots.threads_do(&tc_cl, 0);
   }