8336640: Shenandoah: Parallel worker use in parallel_heap_region_iterate

Reviewed-by: shade
Backport-of: e74edbaea9f09169f597a470f647f3b7d10cc71b

Author: William Kemper

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

@@ -1479,19 +1479,20 @@ class ShenandoahParallelHeapRegionTask : public WorkerTask {
 private:
   ShenandoahHeap* const _heap;
   ShenandoahHeapRegionClosure* const _blk;
+  size_t const _stride;
 
   shenandoah_padding(0);
   volatile size_t _index;
   shenandoah_padding(1);
 
 public:
-  ShenandoahParallelHeapRegionTask(ShenandoahHeapRegionClosure* blk) :
+  ShenandoahParallelHeapRegionTask(ShenandoahHeapRegionClosure* blk, size_t stride) :
           WorkerTask("Shenandoah Parallel Region Operation"),
-          _heap(ShenandoahHeap::heap()), _blk(blk), _index(0) {}
+          _heap(ShenandoahHeap::heap()), _blk(blk), _stride(stride), _index(0) {}
 
   void work(uint worker_id) {
     ShenandoahParallelWorkerSession worker_session(worker_id);
-    size_t stride = ShenandoahParallelRegionStride;
+    size_t stride = _stride;
 
     size_t max = _heap->num_regions();
     while (Atomic::load(&_index) < max) {
@@ -1510,8 +1511,20 @@ class ShenandoahParallelHeapRegionTask : public WorkerTask {
 
 void ShenandoahHeap::parallel_heap_region_iterate(ShenandoahHeapRegionClosure* blk) const {
   assert(blk->is_thread_safe(), "Only thread-safe closures here");
-  if (num_regions() > ShenandoahParallelRegionStride) {
-    ShenandoahParallelHeapRegionTask task(blk);
+  const uint active_workers = workers()->active_workers();
+  const size_t n_regions = num_regions();
+  size_t stride = ShenandoahParallelRegionStride;
+  if (stride == 0 && active_workers > 1) {
+    // Automatically derive the stride to balance the work between threads
+    // evenly. Do not try to split work if below the reasonable threshold.
+    constexpr size_t threshold = 4096;
+    stride = n_regions <= threshold ?
+            threshold :
+            (n_regions + active_workers - 1) / active_workers;
+  }
+
+  if (n_regions > stride && active_workers > 1) {
+    ShenandoahParallelHeapRegionTask task(blk, stride);
     workers()->run_task(&task);
   } else {
     heap_region_iterate(blk);

src/hotspot/share/gc/shenandoah/shenandoah_globals.hpp

@@ -314,9 +314,10 @@
           "checking for cancellation, yielding, etc. Larger values improve "\
           "marking performance at expense of responsiveness.")              \
                                                                             \
-  product(uintx, ShenandoahParallelRegionStride, 1024, EXPERIMENTAL,        \
+  product(uintx, ShenandoahParallelRegionStride, 0, EXPERIMENTAL,           \
           "How many regions to process at once during parallel region "     \
-          "iteration. Affects heaps with lots of regions.")                 \
+          "iteration. Affects heaps with lots of regions. "                 \
+          "Set to 0 to let Shenandoah to decide the best value.")           \
                                                                             \
   product(size_t, ShenandoahSATBBufferSize, 1 * K, EXPERIMENTAL,            \
           "Number of entries in an SATB log buffer.")                       \