8257145: Performance regression with -XX:-ResizePLAB after JDK-8079555

Co-authored-by: Junjun Lin <linjunjun@huawei.com>
Reviewed-by: tschatzl, sjohanss

src/hotspot/share/gc/g1/g1EvacStats.cpp

@@ -27,6 +27,7 @@
 #include "gc/shared/gcId.hpp"
 #include "logging/log.hpp"
 #include "memory/allocation.inline.hpp"
+#include "runtime/globals.hpp"
 
 void G1EvacStats::log_plab_allocation() {
   PLABStats::log_plab_allocation();
@@ -88,8 +89,8 @@ size_t G1EvacStats::compute_desired_plab_sz() {
   return cur_plab_sz;
 }
 
-G1EvacStats::G1EvacStats(const char* description, size_t desired_plab_sz_, unsigned wt) :
-  PLABStats(description, desired_plab_sz_, wt),
+G1EvacStats::G1EvacStats(const char* description, size_t default_per_thread_plab_size, unsigned wt) :
+  PLABStats(description, default_per_thread_plab_size, default_per_thread_plab_size * ParallelGCThreads, wt),
   _region_end_waste(0),
   _regions_filled(0),
   _direct_allocated(0),

src/hotspot/share/gc/g1/g1EvacStats.hpp

@@ -56,7 +56,7 @@ class G1EvacStats : public PLABStats {
   virtual size_t compute_desired_plab_sz();
 
  public:
-  G1EvacStats(const char* description, size_t desired_plab_sz_, unsigned wt);
+  G1EvacStats(const char* description, size_t default_per_thread_plab_size, unsigned wt);
 
   ~G1EvacStats();
 

src/hotspot/share/gc/shared/plab.cpp

@@ -135,6 +135,9 @@ void PLABStats::log_sizing(size_t calculated_words, size_t net_desired_words) {
 
 // Calculates plab size for current number of gc worker threads.
 size_t PLABStats::desired_plab_sz(uint no_of_gc_workers) {
+  if (!ResizePLAB) {
+      return _default_plab_sz;
+  }
   return align_object_size(clamp(_desired_net_plab_sz / no_of_gc_workers, min_size(), max_size()));
 }
 

src/hotspot/share/gc/shared/plab.hpp

@@ -151,6 +151,7 @@ class PLABStats : public CHeapObj<mtGC> {
   size_t _wasted;             // of which wasted (internal fragmentation)
   size_t _undo_wasted;        // of which wasted on undo (is not used for calculation of PLAB size)
   size_t _unused;             // Unused in last buffer
+  size_t _default_plab_sz;
   size_t _desired_net_plab_sz;// Output of filter (below), suitably trimmed and quantized
   AdaptiveWeightedAverage
          _filter;             // Integrator with decay
@@ -169,13 +170,14 @@ class PLABStats : public CHeapObj<mtGC> {
   virtual size_t compute_desired_plab_sz();
 
  public:
-  PLABStats(const char* description, size_t desired_net_plab_sz_, unsigned wt) :
+  PLABStats(const char* description, size_t default_per_thread_plab_size, size_t desired_net_plab_sz, unsigned wt) :
     _description(description),
     _allocated(0),
     _wasted(0),
     _undo_wasted(0),
     _unused(0),
-    _desired_net_plab_sz(desired_net_plab_sz_),
+    _default_plab_sz(default_per_thread_plab_size),
+    _desired_net_plab_sz(desired_net_plab_sz),
     _filter(wt)
   { }
 

test/hotspot/jtreg/gc/g1/plab/TestPLABPromotion.java

@@ -72,7 +72,7 @@ public class TestPLABPromotion {
     private static final int PLAB_SIZE_HIGH = 65536;
     private static final int OBJECT_SIZE_SMALL = 10;
     private static final int OBJECT_SIZE_MEDIUM = 100;
-    private static final int OBJECT_SIZE_HIGH = 1000;
+    private static final int OBJECT_SIZE_HIGH = 3500;
     private static final int GC_NUM_SMALL = 1;
     private static final int GC_NUM_MEDIUM = 3;
     private static final int GC_NUM_HIGH = 7;