diff --git a/runtime/platform/globals.h b/runtime/platform/globals.h
index 5de25a94722..66c3f273fc7 100644
--- a/runtime/platform/globals.h
+++ b/runtime/platform/globals.h
@@ -90,6 +90,7 @@
 #include <string.h>
 #include <sys/types.h>
 
+#include <bit>
 #include <cassert>  // For assert() in constant expressions.
 
 #if defined(_WIN32)
@@ -644,41 +645,7 @@ constexpr double MicrosecondsToMilliseconds(int64_t micros) {
 template <typename T>
 static inline void USE(T&&) {}
 
-// The type-based aliasing rule allows the compiler to assume that
-// pointers of different types (for some definition of different)
-// never alias each other. Thus the following code does not work:
-//
-// float f = foo();
-// int fbits = *(int*)(&f);
-//
-// The compiler 'knows' that the int pointer can't refer to f since
-// the types don't match, so the compiler may cache f in a register,
-// leaving random data in fbits.  Using C++ style casts makes no
-// difference, however a pointer to char data is assumed to alias any
-// other pointer. This is the 'memcpy exception'.
-//
-// The bit_cast function uses the memcpy exception to move the bits
-// from a variable of one type to a variable of another type. Of
-// course the end result is likely to be implementation dependent.
-// Most compilers (gcc-4.2 and MSVC 2005) will completely optimize
-// bit_cast away.
-//
-// There is an additional use for bit_cast. Recent gccs will warn when
-// they see casts that may result in breakage due to the type-based
-// aliasing rule. If you have checked that there is no breakage you
-// can use bit_cast to cast one pointer type to another. This confuses
-// gcc enough that it can no longer see that you have cast one pointer
-// type to another thus avoiding the warning.
-template <class D, class S>
-DART_FORCE_INLINE D bit_cast(const S& source) {
-  static_assert(sizeof(D) == sizeof(S),
-                "Source and destination must have the same size");
-
-  D destination;
-  // This use of memcpy is safe: source and destination cannot overlap.
-  memcpy(&destination, &source, sizeof(destination));
-  return destination;
-}
+using std::bit_cast;
 
 // Similar to bit_cast, but allows copying from types of unrelated
 // sizes. This method was introduced to enable the strict aliasing
diff --git a/runtime/platform/utils.h b/runtime/platform/utils.h
index bb6896ae8ab..1ea942d08d2 100644
--- a/runtime/platform/utils.h
+++ b/runtime/platform/utils.h
@@ -379,8 +379,7 @@ class Utils {
     ASSERT(0 <= rotate);
     ASSERT(rotate <= width);
     using Unsigned = typename std::make_unsigned<T>::type;
-    return (static_cast<Unsigned>(value) << rotate) |
-           (static_cast<T>(value) >> ((width - rotate) & (width - 1)));
+    return std::rotl(static_cast<Unsigned>(value), rotate);
   }
   template <typename T>
   static inline T RotateRight(T value, uint8_t rotate) {
@@ -388,8 +387,7 @@ class Utils {
     ASSERT(0 <= rotate);
     ASSERT(rotate <= width);
     using Unsigned = typename std::make_unsigned<T>::type;
-    return (static_cast<T>(value) >> rotate) |
-           (static_cast<Unsigned>(value) << ((width - rotate) & (width - 1)));
+    return std::rotr(static_cast<Unsigned>(value), rotate);
   }
 
   NO_SANITIZE_UNDEFINED("float-divide-by-zero")
diff --git a/runtime/vm/bss_relocs.cc b/runtime/vm/bss_relocs.cc
index ddd83b72fc6..55dcc097765 100644
--- a/runtime/vm/bss_relocs.cc
+++ b/runtime/vm/bss_relocs.cc
@@ -12,17 +12,16 @@ namespace dart {
 void BSS::InitializeBSSEntry(BSS::Relocation relocation,
                              uword new_value,
                              uword* bss_start) {
-  std::atomic<uword>* slot = reinterpret_cast<std::atomic<uword>*>(
-      &bss_start[BSS::RelocationIndex(relocation)]);
-  uword old_value = slot->load(std::memory_order_relaxed);
+  auto slot = std::atomic_ref(bss_start[BSS::RelocationIndex(relocation)]);
+  uword old_value = slot.load(std::memory_order_relaxed);
   // FullSnapshotReader::ReadProgramSnapshot, and thus BSS::Initialize, can
   // get called multiple times for the same isolate in different threads, though
   // the initialized value will be consistent and thus change only once. Avoid
   // calling compare_exchange_strong unless we actually need to change the
   // value, to avoid spurious read/write races by TSAN.
   if (old_value == new_value) return;
-  if (!slot->compare_exchange_strong(old_value, new_value,
-                                     std::memory_order_relaxed)) {
+  if (!slot.compare_exchange_strong(old_value, new_value,
+                                    std::memory_order_relaxed)) {
     RELEASE_ASSERT(old_value == new_value);
   }
 }
diff --git a/runtime/vm/field_table.cc b/runtime/vm/field_table.cc
index b15dc84bf4e..8dc5549ea98 100644
--- a/runtime/vm/field_table.cc
+++ b/runtime/vm/field_table.cc
@@ -117,7 +117,7 @@ void FieldTable::Grow(intptr_t new_capacity) {
   old_tables_->Add(old_table);
   // Ensure that new_table_ is populated before it is published
   // via store to table_.
-  reinterpret_cast<AcqRelAtomic<ObjectPtr*>*>(&table_)->store(new_table);
+  std::atomic_ref(table_).store(new_table, std::memory_order_release);
   if (isolate_group_ != nullptr) {
     isolate_group_->ForEachIsolate(
         [&](Isolate* isolate) {
diff --git a/runtime/vm/field_table.h b/runtime/vm/field_table.h
index 2414038711b..3f6e753ee01 100644
--- a/runtime/vm/field_table.h
+++ b/runtime/vm/field_table.h
@@ -63,8 +63,8 @@ class FieldTable {
     ASSERT(IsValidIndex(index));
     if (concurrent_use) {
       ObjectPtr* table =
-          reinterpret_cast<const AcqRelAtomic<ObjectPtr*>*>(&table_)->load();
-      return reinterpret_cast<AcqRelAtomic<ObjectPtr>*>(&table[index])->load();
+          std::atomic_ref(table_).load(std::memory_order::acquire);
+      return std::atomic_ref(table[index]).load(std::memory_order_acquire);
     } else {
       // There is no concurrent access expected for this field, so we avoid
       // using atomics. This will allow us to detect via TSAN if there are
@@ -106,7 +106,8 @@ class FieldTable {
   // element, last element contains -1.
   intptr_t free_head_;
 
-  ObjectPtr* table_;
+  // Mutable for atomic_ref in At.
+  mutable ObjectPtr* table_;
   // When table_ grows and have to reallocated, keep the old one here
   // so it will get freed when its are no longer in use.
   MallocGrowableArray<ObjectPtr*>* old_tables_;
diff --git a/runtime/vm/heap/scavenger.cc b/runtime/vm/heap/scavenger.cc
index e18315db5d5..4eb76776f29 100644
--- a/runtime/vm/heap/scavenger.cc
+++ b/runtime/vm/heap/scavenger.cc
@@ -117,14 +117,14 @@ static void objcpy(void* dst, const void* src, size_t size) {
 
 DART_FORCE_INLINE
 static uword ReadHeaderRelaxed(ObjectPtr obj) {
-  return reinterpret_cast<std::atomic<uword>*>(UntaggedObject::ToAddr(obj))
-      ->load(std::memory_order_relaxed);
+  return std::atomic_ref(*reinterpret_cast<uword*>(UntaggedObject::ToAddr(obj)))
+      .load(std::memory_order_relaxed);
 }
 
 DART_FORCE_INLINE
 static void WriteHeaderRelaxed(ObjectPtr obj, uword header) {
-  reinterpret_cast<std::atomic<uword>*>(UntaggedObject::ToAddr(obj))
-      ->store(header, std::memory_order_relaxed);
+  std::atomic_ref(*reinterpret_cast<uword*>(UntaggedObject::ToAddr(obj)))
+      .store(header, std::memory_order_relaxed);
 }
 
 class ScavengerVisitor : public ObjectPointerVisitor,
@@ -524,8 +524,9 @@ class ScavengerVisitor : public ObjectPointerVisitor,
   bool InstallForwardingPointer(uword addr,
                                 uword* old_header,
                                 uword new_header) {
-    return reinterpret_cast<std::atomic<uword>*>(addr)->compare_exchange_strong(
-        *old_header, new_header, std::memory_order_relaxed);
+    return std::atomic_ref(*reinterpret_cast<uword*>(addr))
+        .compare_exchange_strong(*old_header, new_header,
+                                 std::memory_order_relaxed);
   }
 
   DART_FORCE_INLINE
diff --git a/runtime/vm/isolate_test.cc b/runtime/vm/isolate_test.cc
index c4a8f3e7432..52ca0aac661 100644
--- a/runtime/vm/isolate_test.cc
+++ b/runtime/vm/isolate_test.cc
@@ -126,9 +126,9 @@ class InterruptChecker : public ThreadPool::Task {
       // Busy wait for interrupts.
       uword limit = 0;
       do {
-        limit = reinterpret_cast<RelaxedAtomic<uword>*>(
-                    thread_->stack_limit_address())
-                    ->load();
+        limit = std::atomic_ref(
+                    *reinterpret_cast<uword*>(thread_->stack_limit_address()))
+                    .load(std::memory_order_relaxed);
       } while (
           (limit == thread_->saved_stack_limit_) ||
           (((limit & Thread::kInterruptsMask) & Thread::kVMInterrupt) == 0));
diff --git a/runtime/vm/message_snapshot.cc b/runtime/vm/message_snapshot.cc
index 63cce5cc151..f0bb8b7d2e3 100644
--- a/runtime/vm/message_snapshot.cc
+++ b/runtime/vm/message_snapshot.cc
@@ -26,40 +26,19 @@
 
 namespace dart {
 
-static Dart_CObject cobj_sentinel = {Dart_CObject_kUnsupported, {false}};
-static Dart_CObject cobj_dynamic_type = {Dart_CObject_kUnsupported, {false}};
-static Dart_CObject cobj_void_type = {Dart_CObject_kUnsupported, {false}};
-static Dart_CObject cobj_empty_type_arguments = {Dart_CObject_kUnsupported,
-                                                 {false}};
-static Dart_CObject cobj_true = {Dart_CObject_kBool, {true}};
-static Dart_CObject cobj_false = {Dart_CObject_kBool, {false}};
-
-// Workaround for lack of designated initializers until we adopt c++20
-class PredefinedCObjects {
- public:
-  static PredefinedCObjects& getInstance() {
-    static PredefinedCObjects instance;
-    return instance;
-  }
-
-  static Dart_CObject* cobj_null() { return &getInstance().cobj_null_; }
-  static Dart_CObject* cobj_empty_array() {
-    return &getInstance().cobj_empty_array_;
-  }
-
- private:
-  PredefinedCObjects() {
-    cobj_null_.type = Dart_CObject_kNull;
-    cobj_null_.value.as_int64 = 0;
-    cobj_empty_array_.type = Dart_CObject_kArray;
-    cobj_empty_array_.value.as_array = {0, nullptr};
-  }
-
-  Dart_CObject cobj_null_;
-  Dart_CObject cobj_empty_array_;
-
-  DISALLOW_COPY_AND_ASSIGN(PredefinedCObjects);
-};
+static Dart_CObject cobj_sentinel = {.type = Dart_CObject_kUnsupported};
+static Dart_CObject cobj_dynamic_type = {.type = Dart_CObject_kUnsupported};
+static Dart_CObject cobj_void_type = {.type = Dart_CObject_kUnsupported};
+static Dart_CObject cobj_empty_type_arguments = {.type =
+                                                     Dart_CObject_kUnsupported};
+static Dart_CObject cobj_true = {.type = Dart_CObject_kBool,
+                                 .value = {.as_bool = true}};
+static Dart_CObject cobj_false = {.type = Dart_CObject_kBool,
+                                  .value = {.as_bool = false}};
+static Dart_CObject cobj_null = {.type = Dart_CObject_kNull,
+                                 .value = {.as_int64 = 0}};
+static Dart_CObject cobj_empty_array = {.type = Dart_CObject_kArray,
+                                        .value = {.as_array = {0, nullptr}}};
 
 enum class MessagePhase {
   kBeforeTypes = 0,
@@ -2374,7 +2353,7 @@ class ArrayMessageSerializationCluster : public MessageSerializationCluster {
     for (intptr_t i = 0; i < count; i++) {
       Dart_CObject* array = reinterpret_cast<Dart_CObject*>(objects_[i]);
       intptr_t length = array->value.as_array.length;
-      s->WriteRef(PredefinedCObjects::cobj_null());  // TypeArguments
+      s->WriteRef(&cobj_null);  // TypeArguments
       for (intptr_t j = 0; j < length; j++) {
         s->WriteRef(array->value.as_array.values[j]);
       }
@@ -2819,7 +2798,7 @@ bool ApiMessageSerializer::Trace(Dart_CObject* object) {
   intptr_t cid;
   switch (object->type) {
     case Dart_CObject_kNull:
-      ForwardRef(object, PredefinedCObjects::cobj_null());
+      ForwardRef(object, &cobj_null);
       return true;
     case Dart_CObject_kBool:
       ForwardRef(object, object->value.as_bool ? &cobj_true : &cobj_false);
@@ -3234,9 +3213,9 @@ void MessageDeserializer::AddBaseObjects() {
 }
 
 void ApiMessageSerializer::AddBaseObjects() {
-  AddBaseObject(PredefinedCObjects::cobj_null());
+  AddBaseObject(&cobj_null);
   AddBaseObject(&cobj_sentinel);
-  AddBaseObject(PredefinedCObjects::cobj_empty_array());
+  AddBaseObject(&cobj_empty_array);
   AddBaseObject(&cobj_dynamic_type);
   AddBaseObject(&cobj_void_type);
   AddBaseObject(&cobj_empty_type_arguments);
@@ -3245,9 +3224,9 @@ void ApiMessageSerializer::AddBaseObjects() {
 }
 
 void ApiMessageDeserializer::AddBaseObjects() {
-  AddBaseObject(PredefinedCObjects::cobj_null());
+  AddBaseObject(&cobj_null);
   AddBaseObject(&cobj_sentinel);
-  AddBaseObject(PredefinedCObjects::cobj_empty_array());
+  AddBaseObject(&cobj_empty_array);
   AddBaseObject(&cobj_dynamic_type);
   AddBaseObject(&cobj_void_type);
   AddBaseObject(&cobj_empty_type_arguments);
diff --git a/runtime/vm/object.h b/runtime/vm/object.h
index 45c7a07b449..a3f9d0ccc7f 100644
--- a/runtime/vm/object.h
+++ b/runtime/vm/object.h
@@ -896,9 +896,8 @@ class Object {
 
   template <typename FieldType, std::memory_order order>
   FieldType LoadNonPointer(const FieldType* addr) const {
-    return reinterpret_cast<std::atomic<FieldType>*>(
-               const_cast<FieldType*>(addr))
-        ->load(order);
+    return std::atomic_ref<FieldType>(*const_cast<FieldType*>(addr))
+        .load(order);
   }
 
   // Needs two template arguments to allow assigning enums to fixed-size ints.
@@ -913,8 +912,8 @@ class Object {
   void StoreNonPointer(const FieldType* addr, ValueType value) const {
     // Can't use Contains, as it uses tags_, which is set through this method.
     ASSERT(reinterpret_cast<uword>(addr) >= UntaggedObject::ToAddr(ptr()));
-    reinterpret_cast<std::atomic<FieldType>*>(const_cast<FieldType*>(addr))
-        ->store(value, order);
+    std::atomic_ref<FieldType>(*const_cast<FieldType*>(addr))
+        .store(value, order);
   }
 
   // Provides non-const access to non-pointer fields within the object. Such
diff --git a/runtime/vm/raw_object.h b/runtime/vm/raw_object.h
index 65a32d24da6..bd3cc12c285 100644
--- a/runtime/vm/raw_object.h
+++ b/runtime/vm/raw_object.h
@@ -602,8 +602,7 @@ class UntaggedObject {
   }
   template <typename type, std::memory_order order>
   type LoadPointer(type const* addr) const {
-    return reinterpret_cast<std::atomic<type>*>(const_cast<type*>(addr))
-        ->load(order);
+    return std::atomic_ref<type>(*const_cast<type*>(addr)).load(order);
   }
   template <typename type, typename compressed_type>
   type LoadCompressedPointer(compressed_type const* addr) const {
@@ -612,9 +611,9 @@ class UntaggedObject {
   }
   template <typename type, typename compressed_type, std::memory_order order>
   type LoadCompressedPointer(compressed_type const* addr) const {
-    compressed_type v = reinterpret_cast<std::atomic<compressed_type>*>(
-                            const_cast<compressed_type*>(addr))
-                            ->load(order);
+    compressed_type v =
+        std::atomic_ref<compressed_type>(*const_cast<compressed_type*>(addr))
+            .load(order);
     return static_cast<type>(v.Decompress(heap_base()));
   }
   template <typename type, typename compressed_type>
@@ -638,8 +637,7 @@ class UntaggedObject {
   }
   template <typename type, std::memory_order order>
   void StorePointer(type const* addr, type value) {
-    reinterpret_cast<std::atomic<type>*>(const_cast<type*>(addr))
-        ->store(value, order);
+    std::atomic_ref<type>(*const_cast<type*>(addr)).store(value, order);
     if (value.IsHeapObject()) {
       CheckHeapPointerStore(value, Thread::Current());
     }
@@ -654,9 +652,8 @@ class UntaggedObject {
   }
   template <typename type, typename compressed_type, std::memory_order order>
   void StoreCompressedPointer(compressed_type const* addr, type value) {
-    reinterpret_cast<std::atomic<compressed_type>*>(
-        const_cast<compressed_type*>(addr))
-        ->store(static_cast<compressed_type>(value), order);
+    std::atomic_ref<compressed_type>(*const_cast<compressed_type*>(addr))
+        .store(static_cast<compressed_type>(value), order);
     if (value.IsHeapObject()) {
       CheckHeapPointerStore(value, Thread::Current());
     }
@@ -698,8 +695,7 @@ class UntaggedObject {
   }
   template <typename type, std::memory_order order, typename value_type = type>
   void StoreArrayPointer(type const* addr, value_type value) {
-    reinterpret_cast<std::atomic<type>*>(const_cast<type*>(addr))
-        ->store(type(value), order);
+    std::atomic_ref<type>(*const_cast<type*>(addr)).store(type(value), order);
     if (value->IsHeapObject()) {
       CheckArrayPointerStore(addr, value, Thread::Current());
     }
@@ -722,9 +718,8 @@ class UntaggedObject {
   }
   template <typename type, typename compressed_type, std::memory_order order>
   void StoreCompressedArrayPointer(compressed_type const* addr, type value) {
-    reinterpret_cast<std::atomic<compressed_type>*>(
-        const_cast<compressed_type*>(addr))
-        ->store(static_cast<compressed_type>(value), order);
+    std::atomic_ref<compressed_type>(*const_cast<compressed_type*>(addr))
+        .store(static_cast<compressed_type>(value), order);
     if (value->IsHeapObject()) {
       CheckArrayPointerStore(addr, value, Thread::Current());
     }
@@ -734,9 +729,8 @@ class UntaggedObject {
   void StoreCompressedArrayPointer(compressed_type const* addr,
                                    type value,
                                    Thread* thread) {
-    reinterpret_cast<std::atomic<compressed_type>*>(
-        const_cast<compressed_type*>(addr))
-        ->store(static_cast<compressed_type>(value), order);
+    std::atomic_ref<compressed_type>(*const_cast<compressed_type*>(addr))
+        .store(static_cast<compressed_type>(value), order);
     if (value->IsHeapObject()) {
       CheckArrayPointerStore(addr, value, thread);
     }
@@ -757,9 +751,8 @@ class UntaggedObject {
             std::memory_order order = std::memory_order_relaxed>
   type ExchangeCompressedPointer(compressed_type const* addr, type value) {
     compressed_type previous_value =
-        reinterpret_cast<std::atomic<compressed_type>*>(
-            const_cast<compressed_type*>(addr))
-            ->exchange(static_cast<compressed_type>(value), order);
+        std::atomic_ref<compressed_type>(*const_cast<compressed_type*>(addr))
+            .exchange(static_cast<compressed_type>(value), order);
     if (value.IsHeapObject()) {
       CheckHeapPointerStore(value, Thread::Current());
     }
@@ -771,8 +764,7 @@ class UntaggedObject {
   }
   template <std::memory_order order>
   SmiPtr LoadSmi(SmiPtr const* addr) const {
-    return reinterpret_cast<std::atomic<SmiPtr>*>(const_cast<SmiPtr*>(addr))
-        ->load(order);
+    return std::atomic_ref<SmiPtr>(*const_cast<SmiPtr*>(addr)).load(order);
   }
   SmiPtr LoadCompressedSmi(CompressedSmiPtr const* addr) const {
     return static_cast<SmiPtr>(
@@ -780,10 +772,10 @@ class UntaggedObject {
   }
   template <std::memory_order order>
   SmiPtr LoadCompressedSmi(CompressedSmiPtr const* addr) const {
-    return static_cast<SmiPtr>(reinterpret_cast<std::atomic<CompressedSmiPtr>*>(
-                                   const_cast<CompressedSmiPtr*>(addr))
-                                   ->load(order)
-                                   .DecompressSmi());
+    return static_cast<SmiPtr>(
+        std::atomic_ref<CompressedSmiPtr>(*const_cast<CompressedSmiPtr*>(addr))
+            .load(order)
+            .DecompressSmi());
   }
 
   // Use for storing into an explicitly Smi-typed field of an object
@@ -794,17 +786,15 @@ class UntaggedObject {
   }
   template <typename type, std::memory_order order>
   void StoreSmi(type const* addr, type value) {
-    reinterpret_cast<std::atomic<type>*>(const_cast<type*>(addr))
-        ->store(value, order);
+    std::atomic_ref<type>(*const_cast<type*>(addr)).store(value, order);
   }
   void StoreCompressedSmi(CompressedSmiPtr const* addr, SmiPtr value) {
     *const_cast<CompressedSmiPtr*>(addr) = value;
   }
   template <std::memory_order order>
   void StoreCompressedSmi(CompressedSmiPtr const* addr, SmiPtr value) {
-    reinterpret_cast<std::atomic<CompressedSmiPtr>*>(
-        const_cast<CompressedSmiPtr*>(addr))
-        ->store(static_cast<CompressedSmiPtr>(value), order);
+    std::atomic_ref<CompressedSmiPtr>(*const_cast<CompressedSmiPtr*>(addr))
+        .store(static_cast<CompressedSmiPtr>(value), order);
   }
 
  private:
diff --git a/runtime/vm/simulator_arm64.cc b/runtime/vm/simulator_arm64.cc
index 4218f5c2929..afadc2fb59b 100644
--- a/runtime/vm/simulator_arm64.cc
+++ b/runtime/vm/simulator_arm64.cc
@@ -2560,15 +2560,15 @@ void Simulator::DecodeAtomicMemory(Instr* instr) {
 
   if (size == 3) {
     uint64_t in = get_register(rs, R31IsZR);
-    auto addr =
-        reinterpret_cast<std::atomic<uint64_t>*>(get_register(rn, R31IsSP));
+    auto addr = std::atomic_ref(
+        *reinterpret_cast<uint64_t*>(get_register(rn, R31IsSP)));
     uint64_t out;
     switch (opc) {
       case 1:
-        out = addr->fetch_and(~in, order);
+        out = addr.fetch_and(~in, order);
         break;
       case 3:
-        out = addr->fetch_or(in, order);
+        out = addr.fetch_or(in, order);
         break;
       default:
         UNIMPLEMENTED();
@@ -2577,15 +2577,15 @@ void Simulator::DecodeAtomicMemory(Instr* instr) {
   } else if (size == 2) {
     ASSERT(size == 2);
     uint32_t in = get_wregister(rs, R31IsZR);
-    auto addr =
-        reinterpret_cast<std::atomic<uint32_t>*>(get_register(rn, R31IsSP));
+    auto addr = std::atomic_ref(
+        *reinterpret_cast<uint32_t*>(get_register(rn, R31IsSP)));
     uint32_t out;
     switch (opc) {
       case 1:
-        out = addr->fetch_and(~in, order);
+        out = addr.fetch_and(~in, order);
         break;
       case 3:
-        out = addr->fetch_or(in, order);
+        out = addr.fetch_or(in, order);
         break;
       default:
         UNIMPLEMENTED();
diff --git a/runtime/vm/simulator_memory.h b/runtime/vm/simulator_memory.h
index bfb7694b7fe..e814a5422a8 100644
--- a/runtime/vm/simulator_memory.h
+++ b/runtime/vm/simulator_memory.h
@@ -26,20 +26,12 @@ class DirectSimulatorMemory {
 
   template <typename T>
   T Load(uword addr, std::memory_order order) {
-    // TODO(42074): Once we switch to C++20 we should change this to use use
-    // `std::atomic_ref<T>` which supports performing atomic operations on
-    // non-atomic data.
-    static_assert(sizeof(std::atomic<T>) == sizeof(T));
-    return reinterpret_cast<std::atomic<T>*>(addr)->load(order);
+    return std::atomic_ref(*reinterpret_cast<T*>(addr)).load(order);
   }
 
   template <typename T>
   void Store(uword addr, T value, std::memory_order order) {
-    // TODO(42074): Once we switch to C++20 we should change this to use use
-    // `std::atomic_ref<T>` which supports performing atomic operations on
-    // non-atomic data.
-    static_assert(sizeof(std::atomic<T>) == sizeof(T));
-    reinterpret_cast<std::atomic<T>*>(addr)->store(value, order);
+    std::atomic_ref(*reinterpret_cast<T*>(addr)).store(value, order);
   }
 
   template <typename T>
@@ -47,12 +39,8 @@ class DirectSimulatorMemory {
                        T& old_value,
                        T value,
                        std::memory_order order) {
-    // TODO(42074): Once we switch to C++20 we should change this to use use
-    // `std::atomic_ref<T>` which supports performing atomic operations on
-    // non-atomic data.
-    static_assert(sizeof(std::atomic<T>) == sizeof(T));
-    return reinterpret_cast<std::atomic<T>*>(addr)->compare_exchange_weak(
-        old_value, value, order);
+    return std::atomic_ref(*reinterpret_cast<T*>(addr))
+        .compare_exchange_weak(old_value, value, order);
   }
 
   void FlushAddress(uword addr) {}
@@ -112,21 +100,13 @@ class BufferedSimulatorMemory {
   template <typename T>
   T Load(uword addr, std::memory_order order) {
     FlushAddress(addr);
-    // TODO(42074): Once we switch to C++20 we should change this to use use
-    // `std::atomic_ref<T>` which supports performing atomic operations on
-    // non-atomic data.
-    static_assert(sizeof(std::atomic<T>) == sizeof(T));
-    return reinterpret_cast<std::atomic<T>*>(addr)->load(order);
+    return std::atomic_ref(*reinterpret_cast<T*>(addr)).load(order);
   }
 
   template <typename T>
   void Store(uword addr, T value, std::memory_order order) {
     FlushAddress(addr);
-    // TODO(42074): Once we switch to C++20 we should change this to use use
-    // `std::atomic_ref<T>` which supports performing atomic operations on
-    // non-atomic data.
-    static_assert(sizeof(std::atomic<T>) == sizeof(T));
-    reinterpret_cast<std::atomic<T>*>(addr)->store(value, order);
+    std::atomic_ref(*reinterpret_cast<T*>(addr)).store(value, order);
   }
 
   template <typename T>
@@ -135,12 +115,8 @@ class BufferedSimulatorMemory {
                        T value,
                        std::memory_order order) {
     FlushAddress(addr);
-    // TODO(42074): Once we switch to C++20 we should change this to use use
-    // `std::atomic_ref<T>` which supports performing atomic operations on
-    // non-atomic data.
-    static_assert(sizeof(std::atomic<T>) == sizeof(T));
-    return reinterpret_cast<std::atomic<T>*>(addr)->compare_exchange_weak(
-        old_value, value, order);
+    return std::atomic_ref<T>(*reinterpret_cast<T*>(addr))
+        .compare_exchange_weak(old_value, value, order);
   }
 
   void FlushAddress(uword addr) {
diff --git a/runtime/vm/simulator_riscv.cc b/runtime/vm/simulator_riscv.cc
index 7ccf4b7e69f..0e0789eccd6 100644
--- a/runtime/vm/simulator_riscv.cc
+++ b/runtime/vm/simulator_riscv.cc
@@ -2398,9 +2398,9 @@ void Simulator::InterpretAMOSWAP(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
   type desired = get_xreg(instr.rs2());
-  type result = atomic->exchange(desired, instr.memory_order());
+  type result = atomic.exchange(desired, instr.memory_order());
   set_xreg(instr.rd(), sign_extend(result));
 }
 
@@ -2411,9 +2411,9 @@ void Simulator::InterpretAMOADD(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
   type arg = get_xreg(instr.rs2());
-  type result = atomic->fetch_add(arg, instr.memory_order());
+  type result = atomic.fetch_add(arg, instr.memory_order());
   set_xreg(instr.rd(), sign_extend(result));
 }
 
@@ -2424,9 +2424,9 @@ void Simulator::InterpretAMOXOR(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
   type arg = get_xreg(instr.rs2());
-  type result = atomic->fetch_xor(arg, instr.memory_order());
+  type result = atomic.fetch_xor(arg, instr.memory_order());
   set_xreg(instr.rd(), sign_extend(result));
 }
 
@@ -2437,9 +2437,9 @@ void Simulator::InterpretAMOAND(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
   type arg = get_xreg(instr.rs2());
-  type result = atomic->fetch_and(arg, instr.memory_order());
+  type result = atomic.fetch_and(arg, instr.memory_order());
   set_xreg(instr.rd(), sign_extend(result));
 }
 
@@ -2450,9 +2450,9 @@ void Simulator::InterpretAMOOR(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
   type arg = get_xreg(instr.rs2());
-  type result = atomic->fetch_or(arg, instr.memory_order());
+  type result = atomic.fetch_or(arg, instr.memory_order());
   set_xreg(instr.rd(), sign_extend(result));
 }
 
@@ -2463,14 +2463,15 @@ void Simulator::InterpretAMOMIN(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
-  type expected = atomic->load(std::memory_order_relaxed);
+  // TODO(c++26): fetch_max
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
+  type expected = atomic.load(std::memory_order_relaxed);
   type compare = get_xreg(instr.rs2());
   type desired;
   do {
     desired = expected < compare ? expected : compare;
   } while (
-      !atomic->compare_exchange_weak(expected, desired, instr.memory_order()));
+      !atomic.compare_exchange_weak(expected, desired, instr.memory_order()));
   set_xreg(instr.rd(), sign_extend(expected));
 }
 
@@ -2481,14 +2482,15 @@ void Simulator::InterpretAMOMAX(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
-  type expected = atomic->load(std::memory_order_relaxed);
+  // TODO(c++26): fetch_max
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
+  type expected = atomic.load(std::memory_order_relaxed);
   type compare = get_xreg(instr.rs2());
   type desired;
   do {
     desired = expected > compare ? expected : compare;
   } while (
-      !atomic->compare_exchange_weak(expected, desired, instr.memory_order()));
+      !atomic.compare_exchange_weak(expected, desired, instr.memory_order()));
   set_xreg(instr.rd(), sign_extend(expected));
 }
 
@@ -2499,8 +2501,8 @@ void Simulator::InterpretLOADORDERED(Instr instr) {
     Fault("Misaligned atomic memory operation");
   }
   memory_.FlushAddress(addr);
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
-  type value = atomic->load(instr.memory_order());
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
+  type value = atomic.load(instr.memory_order());
   set_xreg(instr.rd(), sign_extend(value));
 }
 
@@ -2512,8 +2514,8 @@ void Simulator::InterpretSTOREORDERED(Instr instr) {
   }
   memory_.FlushAddress(addr);
   type value = get_xreg(instr.rs2());
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
-  atomic->store(value, instr.memory_order());
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
+  atomic.store(value, instr.memory_order());
 }
 
 template <typename type>
@@ -2524,8 +2526,8 @@ void Simulator::InterpretAMOCAS(Instr instr) {
   }
   type expected = get_xreg(instr.rd());
   type desired = get_xreg(instr.rs2());
-  std::atomic<type>* atomic = reinterpret_cast<std::atomic<type>*>(addr);
-  atomic->compare_exchange_weak(expected, desired, instr.memory_order());
+  std::atomic_ref<type> atomic(*reinterpret_cast<type*>(addr));
+  atomic.compare_exchange_weak(expected, desired, instr.memory_order());
   set_xreg(instr.rd(), expected);
 }