8372251: Convert PartialArrayStepper/State to use Atomic<T>

Reviewed-by: iwalulya, tschatzl

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

@@ -47,7 +47,7 @@ PartialArrayState::PartialArrayState(oop src, oop dst,
 }
 
 void PartialArrayState::add_references(size_t count) {
-  size_t new_count = AtomicAccess::add(&_refcount, count, memory_order_relaxed);
+  size_t new_count = _refcount.add_then_fetch(count, memory_order_relaxed);
   assert(new_count >= count, "reference count overflow");
 }
 
@@ -92,7 +92,7 @@ PartialArrayState* PartialArrayStateAllocator::allocate(oop src, oop dst,
 }
 
 void PartialArrayStateAllocator::release(PartialArrayState* state) {
-  size_t refcount = AtomicAccess::sub(&state->_refcount, size_t(1), memory_order_release);
+  size_t refcount = state->_refcount.sub_then_fetch(1u, memory_order_release);
   if (refcount != 0) {
     assert(refcount + 1 != 0, "refcount underflow");
   } else {
@@ -116,25 +116,25 @@ PartialArrayStateManager::~PartialArrayStateManager() {
 }
 
 Arena* PartialArrayStateManager::register_allocator() {
-  uint idx = AtomicAccess::fetch_then_add(&_registered_allocators, 1u, memory_order_relaxed);
+  uint idx = _registered_allocators.fetch_then_add(1u, memory_order_relaxed);
   assert(idx < _max_allocators, "exceeded configured max number of allocators");
   return ::new (&_arenas[idx]) Arena(mtGC);
 }
 
 #ifdef ASSERT
 void PartialArrayStateManager::release_allocator() {
-  uint old = AtomicAccess::fetch_then_add(&_released_allocators, 1u, memory_order_relaxed);
-  assert(old < AtomicAccess::load(&_registered_allocators), "too many releases");
+  uint old = _released_allocators.fetch_then_add(1u, memory_order_relaxed);
+  assert(old < _registered_allocators.load_relaxed(), "too many releases");
 }
 #endif // ASSERT
 
 void PartialArrayStateManager::reset() {
-  uint count = AtomicAccess::load(&_registered_allocators);
-  assert(count == AtomicAccess::load(&_released_allocators),
+  uint count = _registered_allocators.load_relaxed();
+  assert(count == _released_allocators.load_relaxed(),
          "some allocators still active");
   for (uint i = 0; i < count; ++i) {
     _arenas[i].~Arena();
   }
-  AtomicAccess::store(&_registered_allocators, 0u);
-  DEBUG_ONLY(AtomicAccess::store(&_released_allocators, 0u);)
+  _registered_allocators.store_relaxed(0u);
+  DEBUG_ONLY(_released_allocators.store_relaxed(0u);)
 }

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

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2024, 2025, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -27,6 +27,7 @@
 
 #include "memory/allocation.hpp"
 #include "oops/oopsHierarchy.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/globalDefinitions.hpp"
 #include "utilities/macros.hpp"
 
@@ -60,8 +61,8 @@ class PartialArrayState {
   oop _source;
   oop _destination;
   size_t _length;
-  volatile size_t _index;
-  volatile size_t _refcount;
+  Atomic<size_t> _index;
+  Atomic<size_t> _refcount;
 
   friend class PartialArrayStateAllocator;
 
@@ -90,7 +91,7 @@ public:
 
   // A pointer to the start index for the next segment to process, for atomic
   // update.
-  volatile size_t* index_addr() { return &_index; }
+  Atomic<size_t>* index_addr() { return &_index; }
 };
 
 // This class provides memory management for PartialArrayStates.
@@ -178,8 +179,8 @@ class PartialArrayStateManager : public CHeapObj<mtGC> {
   // The number of allocators that have been registered/released.
   // Atomic to support concurrent registration, and concurrent release.
   // Phasing restriction forbids registration concurrent with release.
-  volatile uint _registered_allocators;
-  DEBUG_ONLY(volatile uint _released_allocators;)
+  Atomic<uint> _registered_allocators;
+  DEBUG_ONLY(Atomic<uint> _released_allocators;)
 
   // These are all for sole use of the befriended allocator class.
   Arena* register_allocator();

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

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020, 2024, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2020, 2025, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -26,6 +26,7 @@
 #define SHARE_GC_SHARED_PARTIALARRAYTASKSTEPPER_HPP
 
 #include "oops/arrayOop.hpp"
+#include "runtime/atomic.hpp"
 #include "utilities/globalDefinitions.hpp"
 
 class PartialArrayState;
@@ -73,7 +74,7 @@ private:
   uint _task_fanout;
 
   // For unit tests.
-  inline Step next_impl(size_t length, volatile size_t* index_addr) const;
+  inline Step next_impl(size_t length, Atomic<size_t>* index_addr) const;
 };
 
 #endif // SHARE_GC_SHARED_PARTIALARRAYTASKSTEPPER_HPP

src/hotspot/share/gc/shared/partialArrayTaskStepper.inline.hpp

@@ -46,15 +46,13 @@ PartialArrayTaskStepper::start(size_t length) const {
 }
 
 PartialArrayTaskStepper::Step
-PartialArrayTaskStepper::next_impl(size_t length, volatile size_t* index_addr) const {
+PartialArrayTaskStepper::next_impl(size_t length, Atomic<size_t>* index_addr) const {
   // The start of the next task is in the state's index.
   // Atomically increment by the chunk size to claim the associated chunk.
   // Because we limit the number of enqueued tasks to being no more than the
   // number of remaining chunks to process, we can use an atomic add for the
   // claim, rather than a CAS loop.
-  size_t start = AtomicAccess::fetch_then_add(index_addr,
-                                              _chunk_size,
-                                              memory_order_relaxed);
+  size_t start = index_addr->fetch_then_add(_chunk_size, memory_order_relaxed);
 
   assert(start < length, "invariant: start %zu, length %zu", start, length);
   assert(((length - start) % _chunk_size) == 0,

test/hotspot/gtest/gc/shared/test_partialArrayTaskStepper.cpp

@@ -24,6 +24,7 @@
 
 #include "gc/shared/partialArrayTaskStepper.inline.hpp"
 #include "memory/allStatic.hpp"
+#include "runtime/atomic.hpp"
 #include "unittest.hpp"
 
 using Step = PartialArrayTaskStepper::Step;
@@ -33,7 +34,7 @@ class PartialArrayTaskStepper::TestSupport : AllStatic {
 public:
   static Step next(const Stepper* stepper,
                    size_t length,
-                   size_t* to_length_addr) {
+                   Atomic<size_t>* to_length_addr) {
     return stepper->next_impl(length, to_length_addr);
   }
 };
@@ -42,9 +43,9 @@ using StepperSupport = PartialArrayTaskStepper::TestSupport;
 
 static uint simulate(const Stepper* stepper,
                      size_t length,
-                     size_t* to_length_addr) {
+                     Atomic<size_t>* to_length_addr) {
   Step init = stepper->start(length);
-  *to_length_addr = init._index;
+  to_length_addr->store_relaxed(init._index);
   uint queue_count = init._ncreate;
   uint task = 0;
   for ( ; queue_count > 0; ++task) {
@@ -57,9 +58,9 @@ static uint simulate(const Stepper* stepper,
 
 static void run_test(size_t length, size_t chunk_size, uint n_workers) {
   const PartialArrayTaskStepper stepper(n_workers, chunk_size);
-  size_t to_length;
+  Atomic<size_t> to_length;
   uint tasks = simulate(&stepper, length, &to_length);
-  ASSERT_EQ(length, to_length);
+  ASSERT_EQ(length, to_length.load_relaxed());
   ASSERT_EQ(tasks, length / chunk_size);
 }