[SAB] Move Atomics builtins to C++

src/builtins/builtins-sharedarraybuffer.cc

 // Copyright 2016 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
+#include "src/base/macros.h"
+#include "src/base/platform/mutex.h"
+#include "src/base/platform/time.h"
 #include "src/builtins/builtins-utils.h"
 #include "src/builtins/builtins.h"
 #include "src/code-factory.h"
 #include "src/code-stub-assembler.h"
+#include "src/conversions-inl.h"
+#include "src/factory.h"
+#include "src/futex-emulation.h"
+#include "src/globals.h"
 
 namespace v8 {
 namespace internal {
 
 // ES7 sharedmem 6.3.4.1 get SharedArrayBuffer.prototype.byteLength
 BUILTIN(SharedArrayBufferPrototypeGetByteLength) {
   HandleScope scope(isolate);
   CHECK_RECEIVER(JSArrayBuffer, array_buffer,
                  "get SharedArrayBuffer.prototype.byteLength");
   if (!array_buffer->is_shared()) {
@@ skipping 240 matching lines @@
   a.Bind(&u32);
   a.AtomicStore(MachineRepresentation::kWord32, backing_store,
                 a.WordShl(index_word, 2), value_word32);
   a.Return(value_integer);
 
   // This shouldn't happen, we've already validated the type.
   a.Bind(&other);
   a.Return(a.SmiConstant(0));
 }
 
+inline bool AtomicIsLockFree(uint32_t size) {
+  return size == 1 || size == 2 || size == 4;
+}
+
+// ES #sec-atomics.islockfree
+BUILTIN(AtomicsIsLockFree) {
+  HandleScope scope(isolate);
+  Handle<Object> size = args.atOrUndefined(isolate, 1);
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, size, Object::ToNumber(size));
+  return *isolate->factory()->ToBoolean(AtomicIsLockFree(size->Number()));
+}
+
+// ES #sec-validatesharedintegertypedarray
+MUST_USE_RESULT MaybeHandle<JSTypedArray> ValidateSharedIntegerTypedArray(
+    Isolate* isolate, Handle<Object> object, bool onlyInt32 = false) {

[binji, 2017/02/17 19:44:41]

nit: hacker_case for variables... or is the idea to keep the names matching the
ES spec?

[Dan Ehrenberg, 2017/02/20 21:34:21]

You're right, fixed.

+  if (object->IsJSTypedArray()) {
+    Handle<JSTypedArray> typedArray = Handle<JSTypedArray>::cast(object);
+    if (typedArray->GetBuffer()->is_shared()) {
+      if (onlyInt32) {
+        if (typedArray->type() == kExternalInt32Array) return typedArray;
+      } else {
+        if (typedArray->type() != kExternalFloat32Array &&
+            typedArray->type() != kExternalFloat64Array &&
+            typedArray->type() != kExternalUint8ClampedArray)
+          return typedArray;
+      }
+    }
+  }
+
+  THROW_NEW_ERROR(
+      isolate,
+      NewTypeError(onlyInt32 ? MessageTemplate::kNotInt32SharedTypedArray
+                             : MessageTemplate::kNotIntegerSharedTypedArray,
+                   object),
+      JSTypedArray);
+}
+
+// ES #sec-validateatomicaccess
+MUST_USE_RESULT Maybe<size_t> ValidateAtomicAccess(
+    Isolate* isolate, Handle<JSTypedArray> typedArray,
+    Handle<Object> requestIndex) {
+  // TOOD(v8:5961): Use ToIndex for indexes
+  ASSIGN_RETURN_ON_EXCEPTION_VALUE(
+      isolate, requestIndex, Object::ToNumber(requestIndex), Nothing<size_t>());
+  Handle<Object> offset;
+  ASSIGN_RETURN_ON_EXCEPTION_VALUE(isolate, offset,
+                                   Object::ToInteger(isolate, requestIndex),
+                                   Nothing<size_t>());
+  if (!requestIndex->SameValue(*offset)) {
+    isolate->Throw(*isolate->factory()->NewRangeError(
+        MessageTemplate::kInvalidAtomicAccessIndex));
+    return Nothing<size_t>();
+  }
+  size_t accessIndex;
+  uint32_t length = typedArray->length_value();
+  if (!TryNumberToSize(*requestIndex, &accessIndex) || accessIndex >= length) {
+    isolate->Throw(*isolate->factory()->NewRangeError(
+        MessageTemplate::kInvalidAtomicAccessIndex));
+    return Nothing<size_t>();
+  }
+  return Just<size_t>(accessIndex);
+}
+
+// ES #sec-atomics.wake
+// Atomics.wake( typedArray, index, count )
+BUILTIN(AtomicsWake) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);

[binji, 2017/02/17 19:44:41]

Can these just be at, since they're not optional? I don't really understand the
difference here. :)

[Dan Ehrenberg, 2017/02/20 21:34:21]

It looks like that would make a DCHECK fail if it's invoked with zero arguments,
on line 37 of src/builtins/builtins-utils.h.

+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> count = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array, true));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  uint32_t c;
+  if (count->IsUndefined(isolate)) {
+    c = kMaxUInt32;
+  } else {
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, count,
+                                       Object::ToInteger(isolate, count));
+    double countDouble = count->Number();
+    if (countDouble < 0)
+      countDouble = 0;
+    else if (countDouble > kMaxUInt32)
+      countDouble = kMaxUInt32;
+    c = static_cast<uint32_t>(countDouble);
+  }
+
+  Handle<JSArrayBuffer> array_buffer = sta->GetBuffer();
+  size_t addr = (i << 2) + NumberToSize(sta->byte_offset());
+
+  return FutexEmulation::Wake(isolate, array_buffer, addr, c);
+}
+
+// ES #sec-atomics.wait
+// Atomics.wait( typedArray, index, value, timeout )
+BUILTIN(AtomicsWait) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+  Handle<Object> timeout = args.atOrUndefined(isolate, 4);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array, true));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInt32(isolate, value));
+  int32_t valueInt32 = NumberToInt32(*value);
+
+  double timeoutNumber;
+  if (timeout->IsUndefined(isolate)) {
+    timeoutNumber = isolate->heap()->infinity_value()->Number();
+  } else {
+    ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, timeout,
+                                       Object::ToNumber(timeout));
+    timeoutNumber = timeout->Number();
+    if (std::isnan(timeoutNumber))
+      timeoutNumber = isolate->heap()->infinity_value()->Number();
+    else if (timeoutNumber < 0)
+      timeoutNumber = 0;
+  }
+
+  if (!isolate->allow_atomics_wait()) {
+    THROW_NEW_ERROR_RETURN_FAILURE(
+        isolate, NewTypeError(MessageTemplate::kAtomicsWaitNotAllowed));
+  }
+
+  Handle<JSArrayBuffer> array_buffer = sta->GetBuffer();
+  size_t addr = (i << 2) + NumberToSize(sta->byte_offset());
+
+  return FutexEmulation::Wait(isolate, array_buffer, addr, valueInt32,
+                              timeoutNumber);
+}
+
+namespace {
+
+#if V8_CC_GNU
+
+template <typename T>
+inline T CompareExchangeSeqCst(T* p, T oldval, T newval) {
+  (void)__atomic_compare_exchange_n(p, &oldval, newval, 0, __ATOMIC_SEQ_CST,
+                                    __ATOMIC_SEQ_CST);
+  return oldval;
+}
+
+template <typename T>
+inline T AddSeqCst(T* p, T value) {
+  return __atomic_fetch_add(p, value, __ATOMIC_SEQ_CST);
+}
+
+template <typename T>
+inline T SubSeqCst(T* p, T value) {
+  return __atomic_fetch_sub(p, value, __ATOMIC_SEQ_CST);
+}
+
+template <typename T>
+inline T AndSeqCst(T* p, T value) {
+  return __atomic_fetch_and(p, value, __ATOMIC_SEQ_CST);
+}
+
+template <typename T>
+inline T OrSeqCst(T* p, T value) {
+  return __atomic_fetch_or(p, value, __ATOMIC_SEQ_CST);
+}
+
+template <typename T>
+inline T XorSeqCst(T* p, T value) {
+  return __atomic_fetch_xor(p, value, __ATOMIC_SEQ_CST);
+}
+
+template <typename T>
+inline T ExchangeSeqCst(T* p, T value) {
+  return __atomic_exchange_n(p, value, __ATOMIC_SEQ_CST);
+}
+
+#elif V8_CC_MSVC
+
+#define InterlockedCompareExchange32 _InterlockedCompareExchange
+#define InterlockedExchange32 _InterlockedExchange
+#define InterlockedExchangeAdd32 _InterlockedExchangeAdd
+#define InterlockedAnd32 _InterlockedAnd
+#define InterlockedOr32 _InterlockedOr
+#define InterlockedXor32 _InterlockedXor
+#define InterlockedExchangeAdd16 _InterlockedExchangeAdd16
+#define InterlockedCompareExchange8 _InterlockedCompareExchange8
+#define InterlockedExchangeAdd8 _InterlockedExchangeAdd8
+
+#define ATOMIC_OPS(type, suffix, vctype)                                    \
+  inline type AddSeqCst(type* p, type value) {                              \
+    return InterlockedExchangeAdd##suffix(reinterpret_cast<vctype*>(p),     \
+                                          bit_cast<vctype>(value));         \
+  }                                                                         \
+  inline type SubSeqCst(type* p, type value) {                              \
+    return InterlockedExchangeAdd##suffix(reinterpret_cast<vctype*>(p),     \
+                                          -bit_cast<vctype>(value));        \
+  }                                                                         \
+  inline type AndSeqCst(type* p, type value) {                              \
+    return InterlockedAnd##suffix(reinterpret_cast<vctype*>(p),             \
+                                  bit_cast<vctype>(value));                 \
+  }                                                                         \
+  inline type OrSeqCst(type* p, type value) {                               \
+    return InterlockedOr##suffix(reinterpret_cast<vctype*>(p),              \
+                                 bit_cast<vctype>(value));                  \
+  }                                                                         \
+  inline type XorSeqCst(type* p, type value) {                              \
+    return InterlockedXor##suffix(reinterpret_cast<vctype*>(p),             \
+                                  bit_cast<vctype>(value));                 \
+  }                                                                         \
+  inline type ExchangeSeqCst(type* p, type value) {                         \
+    return InterlockedExchange##suffix(reinterpret_cast<vctype*>(p),        \
+                                       bit_cast<vctype>(value));            \
+  }                                                                         \
+                                                                            \
+  inline type CompareExchangeSeqCst(type* p, type oldval, type newval) {    \
+    return InterlockedCompareExchange##suffix(reinterpret_cast<vctype*>(p), \
+                                              bit_cast<vctype>(newval),     \
+                                              bit_cast<vctype>(oldval));    \
+  }
+
+ATOMIC_OPS(int8_t, 8, char)
+ATOMIC_OPS(uint8_t, 8, char)
+ATOMIC_OPS(int16_t, 16, short)  /* NOLINT(runtime/int) */
+ATOMIC_OPS(uint16_t, 16, short) /* NOLINT(runtime/int) */
+ATOMIC_OPS(int32_t, 32, long)   /* NOLINT(runtime/int) */
+ATOMIC_OPS(uint32_t, 32, long)  /* NOLINT(runtime/int) */
+
+#undef ATOMIC_OPS_INTEGER
+#undef ATOMIC_OPS
+
+#undef InterlockedCompareExchange32
+#undef InterlockedExchange32
+#undef InterlockedExchangeAdd32
+#undef InterlockedAnd32
+#undef InterlockedOr32
+#undef InterlockedXor32
+#undef InterlockedExchangeAdd16
+#undef InterlockedCompareExchange8
+#undef InterlockedExchangeAdd8
+
+#else
+
+#error Unsupported platform!
+
+#endif
+
+template <typename T>
+T FromObject(Handle<Object> number);
+
+template <>
+inline uint8_t FromObject<uint8_t>(Handle<Object> number) {
+  return NumberToUint32(*number);
+}
+
+template <>
+inline int8_t FromObject<int8_t>(Handle<Object> number) {
+  return NumberToInt32(*number);
+}
+
+template <>
+inline uint16_t FromObject<uint16_t>(Handle<Object> number) {
+  return NumberToUint32(*number);
+}
+
+template <>
+inline int16_t FromObject<int16_t>(Handle<Object> number) {
+  return NumberToInt32(*number);
+}
+
+template <>
+inline uint32_t FromObject<uint32_t>(Handle<Object> number) {
+  return NumberToUint32(*number);
+}
+
+template <>
+inline int32_t FromObject<int32_t>(Handle<Object> number) {
+  return NumberToInt32(*number);
+}
+
+inline Object* ToObject(Isolate* isolate, int8_t t) { return Smi::FromInt(t); }
+
+inline Object* ToObject(Isolate* isolate, uint8_t t) { return Smi::FromInt(t); }
+
+inline Object* ToObject(Isolate* isolate, int16_t t) { return Smi::FromInt(t); }
+
+inline Object* ToObject(Isolate* isolate, uint16_t t) {
+  return Smi::FromInt(t);
+}
+
+inline Object* ToObject(Isolate* isolate, int32_t t) {
+  return *isolate->factory()->NewNumber(t);
+}
+
+inline Object* ToObject(Isolate* isolate, uint32_t t) {
+  return *isolate->factory()->NewNumber(t);
+}
+
+template <typename T>
+inline Object* DoCompareExchange(Isolate* isolate, void* buffer, size_t index,
+                                 Handle<Object> oldobj, Handle<Object> newobj) {
+  T oldval = FromObject<T>(oldobj);
+  T newval = FromObject<T>(newobj);
+  T result =
+      CompareExchangeSeqCst(static_cast<T*>(buffer) + index, oldval, newval);
+  return ToObject(isolate, result);
+}
+
+template <typename T>
+inline Object* DoAdd(Isolate* isolate, void* buffer, size_t index,
+                     Handle<Object> obj) {
+  T value = FromObject<T>(obj);
+  T result = AddSeqCst(static_cast<T*>(buffer) + index, value);
+  return ToObject(isolate, result);
+}
+
+template <typename T>
+inline Object* DoSub(Isolate* isolate, void* buffer, size_t index,
+                     Handle<Object> obj) {
+  T value = FromObject<T>(obj);
+  T result = SubSeqCst(static_cast<T*>(buffer) + index, value);
+  return ToObject(isolate, result);
+}
+
+template <typename T>
+inline Object* DoAnd(Isolate* isolate, void* buffer, size_t index,
+                     Handle<Object> obj) {
+  T value = FromObject<T>(obj);
+  T result = AndSeqCst(static_cast<T*>(buffer) + index, value);
+  return ToObject(isolate, result);
+}
+
+template <typename T>
+inline Object* DoOr(Isolate* isolate, void* buffer, size_t index,
+                    Handle<Object> obj) {
+  T value = FromObject<T>(obj);
+  T result = OrSeqCst(static_cast<T*>(buffer) + index, value);
+  return ToObject(isolate, result);
+}
+
+template <typename T>
+inline Object* DoXor(Isolate* isolate, void* buffer, size_t index,
+                     Handle<Object> obj) {
+  T value = FromObject<T>(obj);
+  T result = XorSeqCst(static_cast<T*>(buffer) + index, value);
+  return ToObject(isolate, result);
+}
+
+template <typename T>
+inline Object* DoExchange(Isolate* isolate, void* buffer, size_t index,
+                          Handle<Object> obj) {
+  T value = FromObject<T>(obj);
+  T result = ExchangeSeqCst(static_cast<T*>(buffer) + index, value);
+  return ToObject(isolate, result);
+}
+
+// Uint8Clamped functions

[binji, 2017/02/17 19:44:41]

Oops, the Uint8Clamped versions should have been removed. Might as well do it
now!

[Dan Ehrenberg, 2017/02/20 21:34:21]

Would you mind if I removed this in a follow-on patch? This is such a big
refactoring; it'd be easier if I can leave the tests as is so that I can have
the confidence that I didn't mess up both the code and the tests.

+
+uint8_t ClampToUint8(int32_t value) {
+  if (value < 0) return 0;
+  if (value > 255) return 255;
+  return value;
+}
+
+inline Object* DoCompareExchangeUint8Clamped(Isolate* isolate, void* buffer,
+                                             size_t index,
+                                             Handle<Object> oldobj,
+                                             Handle<Object> newobj) {
+  typedef int32_t convert_type;
+  uint8_t oldval = ClampToUint8(FromObject<convert_type>(oldobj));
+  uint8_t newval = ClampToUint8(FromObject<convert_type>(newobj));
+  uint8_t result = CompareExchangeSeqCst(static_cast<uint8_t*>(buffer) + index,
+                                         oldval, newval);
+  return ToObject(isolate, result);
+}
+
+#define DO_UINT8_CLAMPED_OP(name, op)                                        \
+  inline Object* Do##name##Uint8Clamped(Isolate* isolate, void* buffer,      \
+                                        size_t index, Handle<Object> obj) {  \
+    typedef int32_t convert_type;                                            \
+    uint8_t* p = static_cast<uint8_t*>(buffer) + index;                      \
+    convert_type operand = FromObject<convert_type>(obj);                    \
+    uint8_t expected;                                                        \
+    uint8_t result;                                                          \
+    do {                                                                     \
+      expected = *p;                                                         \
+      result = ClampToUint8(static_cast<convert_type>(expected) op operand); \
+    } while (CompareExchangeSeqCst(p, expected, result) != expected);        \
+    return ToObject(isolate, expected);                                      \
+  }
+
+DO_UINT8_CLAMPED_OP(Add, +)
+DO_UINT8_CLAMPED_OP(Sub, -)
+DO_UINT8_CLAMPED_OP(And, &)
+DO_UINT8_CLAMPED_OP(Or, |)
+DO_UINT8_CLAMPED_OP(Xor, ^)
+
+#undef DO_UINT8_CLAMPED_OP
+
+inline Object* DoExchangeUint8Clamped(Isolate* isolate, void* buffer,
+                                      size_t index, Handle<Object> obj) {
+  typedef int32_t convert_type;
+  uint8_t* p = static_cast<uint8_t*>(buffer) + index;
+  uint8_t result = ClampToUint8(FromObject<convert_type>(obj));
+  uint8_t expected;
+  do {
+    expected = *p;
+  } while (CompareExchangeSeqCst(p, expected, result) != expected);
+  return ToObject(isolate, expected);
+}
+
+}  // anonymous namespace
+
+// Duplicated from objects.h
+// V has parameters (Type, type, TYPE, C type, element_size)
+#define INTEGER_TYPED_ARRAYS(V)          \
+  V(Uint8, uint8, UINT8, uint8_t, 1)     \
+  V(Int8, int8, INT8, int8_t, 1)         \
+  V(Uint16, uint16, UINT16, uint16_t, 2) \
+  V(Int16, int16, INT16, int16_t, 2)     \
+  V(Uint32, uint32, UINT32, uint32_t, 4) \
+  V(Int32, int32, INT32, int32_t, 4)
+
+// ES #sec-atomics.wait
+// Atomics.compareExchange( typedArray, index, expectedValue, replacementValue )
+BUILTIN(AtomicsCompareExchange) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> expectedValue = args.atOrUndefined(isolate, 3);
+  Handle<Object> replacementValue = args.atOrUndefined(isolate, 4);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, expectedValue,
+                                     Object::ToInteger(isolate, expectedValue));
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, replacementValue, Object::ToInteger(isolate, replacementValue));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size)            \
+  case kExternal##Type##Array:                                         \
+    return DoCompareExchange<ctype>(isolate, source, i, expectedValue, \
+                                    replacementValue);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoCompareExchangeUint8Clamped(isolate, source, i, expectedValue,
+                                           replacementValue);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
+// ES #sec-atomics.add
+// Atomics.add( typedArray, index, value )
+BUILTIN(AtomicsAdd) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInteger(isolate, value));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size) \
+  case kExternal##Type##Array:                              \
+    return DoAdd<ctype>(isolate, source, i, value);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoAddUint8Clamped(isolate, source, i, value);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
+// ES #sec-atomics.sub
+// Atomics.sub( typedArray, index, value )
+BUILTIN(AtomicsSub) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInteger(isolate, value));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size) \
+  case kExternal##Type##Array:                              \
+    return DoSub<ctype>(isolate, source, i, value);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoSubUint8Clamped(isolate, source, i, value);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
+// ES #sec-atomics.and
+// Atomics.and( typedArray, index, value )
+BUILTIN(AtomicsAnd) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInteger(isolate, value));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size) \
+  case kExternal##Type##Array:                              \
+    return DoAnd<ctype>(isolate, source, i, value);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoAndUint8Clamped(isolate, source, i, value);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
+// ES #sec-atomics.or
+// Atomics.or( typedArray, index, value )
+BUILTIN(AtomicsOr) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInteger(isolate, value));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size) \
+  case kExternal##Type##Array:                              \
+    return DoOr<ctype>(isolate, source, i, value);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoOrUint8Clamped(isolate, source, i, value);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
+// ES #sec-atomics.xor
+// Atomics.xor( typedArray, index, value )
+BUILTIN(AtomicsXor) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInteger(isolate, value));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size) \
+  case kExternal##Type##Array:                              \
+    return DoXor<ctype>(isolate, source, i, value);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoXorUint8Clamped(isolate, source, i, value);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
+// ES #sec-atomics.exchange
+// Atomics.exchange( typedArray, index, value )
+BUILTIN(AtomicsExchange) {
+  HandleScope scope(isolate);
+  Handle<Object> array = args.atOrUndefined(isolate, 1);
+  Handle<Object> index = args.atOrUndefined(isolate, 2);
+  Handle<Object> value = args.atOrUndefined(isolate, 3);
+
+  Handle<JSTypedArray> sta;
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(
+      isolate, sta, ValidateSharedIntegerTypedArray(isolate, array));
+
+  Maybe<size_t> maybeIndex = ValidateAtomicAccess(isolate, sta, index);
+  if (maybeIndex.IsNothing()) return isolate->heap()->exception();
+  size_t i = maybeIndex.FromJust();
+
+  ASSIGN_RETURN_FAILURE_ON_EXCEPTION(isolate, value,
+                                     Object::ToInteger(isolate, value));
+
+  uint8_t* source = static_cast<uint8_t*>(sta->GetBuffer()->backing_store()) +
+                    NumberToSize(sta->byte_offset());
+
+  switch (sta->type()) {
+#define TYPED_ARRAY_CASE(Type, typeName, TYPE, ctype, size) \
+  case kExternal##Type##Array:                              \
+    return DoExchange<ctype>(isolate, source, i, value);
+
+    INTEGER_TYPED_ARRAYS(TYPED_ARRAY_CASE)
+#undef TYPED_ARRAY_CASE
+
+    case kExternalUint8ClampedArray:
+      return DoExchangeUint8Clamped(isolate, source, i, value);
+
+    default:
+      break;
+  }
+
+  UNREACHABLE();
+  return isolate->heap()->undefined_value();
+}
+
 }  // namespace internal
 }  // namespace v8