src/share/vm/gc/cms/parNewGeneration.cpp

rev 8461 : 8086056: ParNew: auto-tune ParGCCardsPerStrideChunk
Summary: Automatically calculate ParGCCardsPerStrideChunk based on the old gen capacity.
Reviewed-by:
rev 8441 : 8081007: Remove redundant active worker variables and calls in ParNewGeneration::collect
Reviewed-by: pliden, kbarrett, ehelin
rev 8409 : 8080879: Remove FlexibleWorkGang::set_for_termination
Reviewed-by: brutisso, kbarrett, pliden
rev 8404 : 8080876: Replace unnecessary MAX2(ParallelGCThreads, 1) calls with ParallelGCThreads
Reviewed-by: kbarrett, mgerdin
rev 8399 : 8080113: Remove CollectedHeap::set_par_threads()
Reviewed-by: jmasa, kbarrett
rev 8397 : 8080111: Remove SubTaskDone::_n_threads
Reviewed-by: jmasa, kbarrett
rev 8396 : 8080110: Remove usage of CollectedHeap::n_par_threads() from root processing
Reviewed-by: jmasa, kbarrett
rev 8362 : 8079792: GC directory structure cleanup
Reviewed-by: brutisso, stefank, david

*** 618,628 ****
  ParNewGeneration::
  ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level)
    : DefNewGeneration(rs, initial_byte_size, level, "PCopy"),
    _overflow_list(NULL),
    _is_alive_closure(this),
!   _plab_stats(YoungPLABSize, PLABWeight)
  {
    NOT_PRODUCT(_overflow_counter = ParGCWorkQueueOverflowInterval;)
    NOT_PRODUCT(_num_par_pushes = 0;)
    _task_queues = new ObjToScanQueueSet(ParallelGCThreads);
    guarantee(_task_queues != NULL, "task_queues allocation failure.");
--- 618,629 ----
  ParNewGeneration::
  ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level)
    : DefNewGeneration(rs, initial_byte_size, level, "PCopy"),
    _overflow_list(NULL),
    _is_alive_closure(this),
!   _plab_stats(YoungPLABSize, PLABWeight),
!   _should_adjust_cards_per_stride(false)
  {
    NOT_PRODUCT(_overflow_counter = ParGCWorkQueueOverflowInterval;)
    NOT_PRODUCT(_num_par_pushes = 0;)
    _task_queues = new ObjToScanQueueSet(ParallelGCThreads);
    guarantee(_task_queues != NULL, "task_queues allocation failure.");
*** 656,665 ****
--- 657,670 ----
      const char* cname =
           PerfDataManager::counter_name(_gen_counters->name_space(), "threads");
      PerfDataManager::create_constant(SUN_GC, cname, PerfData::U_None,
                                       ParallelGCThreads, CHECK);
    }
+ 
+   // we'll automatically calculate ParGCCardsPerStrideChunk if it has
+   // not been explicitly set on the command line
+   _should_adjust_cards_per_stride = FLAG_IS_DEFAULT(ParGCCardsPerStrideChunk);
  }
  #ifdef _MSC_VER
  #pragma warning( pop )
  #endif
  
*** 878,887 ****
--- 883,928 ----
    }
    // Reset the PromotionFailureALot counters.
    NOT_PRODUCT(gch->reset_promotion_should_fail();)
  }
  
+ void ParNewGeneration::adjust_cards_per_stride(const Generation& old_gen) {
+   if (!_should_adjust_cards_per_stride) return;
+ 
+   const size_t MinOldGenCapacity =      G;
+   const size_t MaxOldGenCapacity = 16 * G;
+   assert(MinOldGenCapacity <= MaxOldGenCapacity, "sanity");
+   const size_t OldGenCapacityDiff = MaxOldGenCapacity - MinOldGenCapacity;
+ 
+   const intx MinStrideChunkSize =   256;
+   const intx MaxStrideChunkSize = 8 * K;
+   assert(MinStrideChunkSize <= MaxStrideChunkSize, "sanity");
+   const intx StrideChunkSizeDiff = MaxStrideChunkSize - MinStrideChunkSize;
+ 
+   const size_t capacity = old_gen.capacity();
+   // adj_capacity = capacity bounded by {Min,Max}OldGenCapacity
+   size_t adj_capacity = capacity;
+   adj_capacity = MAX2(adj_capacity, MinOldGenCapacity);
+   adj_capacity = MIN2(adj_capacity, MaxOldGenCapacity);
+ 
+   const size_t capacity_offset = adj_capacity - MinOldGenCapacity;
+   assert(capacity_offset && capacity_offset <= OldGenCapacityDiff,
+          "sanity");
+   const double r = (double) capacity_offset / (double) OldGenCapacityDiff;
+   const intx stride_chunk_offset = (intx) (r * (double) StrideChunkSizeDiff);
+   const int stride_chunk_size_log =
+             log2_intptr((intptr_t) (MinStrideChunkSize + stride_chunk_offset));
+   const intx stride_chunk_size = (intx) 1 << stride_chunk_size_log;
+ 
+   assert(MinStrideChunkSize <= stride_chunk_size &&
+          stride_chunk_size <= MaxStrideChunkSize &&
+          is_power_of_2((intptr_t) stride_chunk_size),
+          err_msg("stride_chunk_size: "INTX_FORMAT, stride_chunk_size));
+ 
+   FLAG_SET_ERGO(intx, ParGCCardsPerStrideChunk, stride_chunk_size);
+ }
+ 
  void ParNewGeneration::collect(bool   full,
                                 bool   clear_all_soft_refs,
                                 size_t size,
                                 bool   is_tlab) {
    assert(full || size > 0, "otherwise we don't want to collect");
*** 923,932 ****
--- 964,974 ----
    // Capture heap used before collection (for printing).
    size_t gch_prev_used = gch->used();
  
    age_table()->clear();
    to()->clear(SpaceDecorator::Mangle);
+   adjust_cards_per_stride(*_old_gen);
  
    gch->save_marks();
  
    // Set the correct parallelism (number of queues) in the reference processor
    ref_processor()->set_active_mt_degree(active_workers);