Source code
Revision control
Copy as Markdown
Other Tools
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
* vim: set ts=8 sts=2 et sw=2 tw=80:
* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
#ifndef vm_TypedArrayObject_inl_h
#define vm_TypedArrayObject_inl_h
/* Utilities and common inline code for TypedArray */
#include "vm/TypedArrayObject.h"
#include "mozilla/Assertions.h"
#include "mozilla/FloatingPoint.h"
#include <algorithm>
#include <type_traits>
#include "jsnum.h"
#include "gc/Zone.h"
#include "jit/AtomicOperations.h"
#include "js/Conversions.h"
#include "js/ScalarType.h" // js::Scalar::Type
#include "js/Value.h"
#include "util/DifferentialTesting.h"
#include "util/Memory.h"
#include "vm/ArrayObject.h"
#include "vm/BigIntType.h"
#include "vm/NativeObject.h"
#include "vm/Uint8Clamped.h"
#include "gc/ObjectKind-inl.h"
#include "vm/NativeObject-inl.h"
#include "vm/ObjectOperations-inl.h"
namespace js {
template <typename To, typename From>
inline To ConvertNumber(From src);
template <>
inline int8_t ConvertNumber<int8_t, float>(float src) {
return JS::ToInt8(src);
}
template <>
inline uint8_t ConvertNumber<uint8_t, float>(float src) {
return JS::ToUint8(src);
}
template <>
inline uint8_clamped ConvertNumber<uint8_clamped, float>(float src) {
return uint8_clamped(src);
}
template <>
inline int16_t ConvertNumber<int16_t, float>(float src) {
return JS::ToInt16(src);
}
template <>
inline uint16_t ConvertNumber<uint16_t, float>(float src) {
return JS::ToUint16(src);
}
template <>
inline int32_t ConvertNumber<int32_t, float>(float src) {
return JS::ToInt32(src);
}
template <>
inline uint32_t ConvertNumber<uint32_t, float>(float src) {
return JS::ToUint32(src);
}
template <>
inline int64_t ConvertNumber<int64_t, float>(float src) {
return JS::ToInt64(src);
}
template <>
inline uint64_t ConvertNumber<uint64_t, float>(float src) {
return JS::ToUint64(src);
}
template <>
inline int8_t ConvertNumber<int8_t, double>(double src) {
return JS::ToInt8(src);
}
template <>
inline uint8_t ConvertNumber<uint8_t, double>(double src) {
return JS::ToUint8(src);
}
template <>
inline uint8_clamped ConvertNumber<uint8_clamped, double>(double src) {
return uint8_clamped(src);
}
template <>
inline int16_t ConvertNumber<int16_t, double>(double src) {
return JS::ToInt16(src);
}
template <>
inline uint16_t ConvertNumber<uint16_t, double>(double src) {
return JS::ToUint16(src);
}
template <>
inline int32_t ConvertNumber<int32_t, double>(double src) {
return JS::ToInt32(src);
}
template <>
inline uint32_t ConvertNumber<uint32_t, double>(double src) {
return JS::ToUint32(src);
}
template <>
inline int64_t ConvertNumber<int64_t, double>(double src) {
return JS::ToInt64(src);
}
template <>
inline uint64_t ConvertNumber<uint64_t, double>(double src) {
return JS::ToUint64(src);
}
template <typename To, typename From>
inline To ConvertNumber(From src) {
static_assert(
!std::is_floating_point_v<From> ||
(std::is_floating_point_v<From> && std::is_floating_point_v<To>),
"conversion from floating point to int should have been handled by "
"specializations above");
return To(src);
}
template <typename NativeType>
struct TypeIDOfType;
template <>
struct TypeIDOfType<int8_t> {
static const Scalar::Type id = Scalar::Int8;
static const JSProtoKey protoKey = JSProto_Int8Array;
};
template <>
struct TypeIDOfType<uint8_t> {
static const Scalar::Type id = Scalar::Uint8;
static const JSProtoKey protoKey = JSProto_Uint8Array;
};
template <>
struct TypeIDOfType<int16_t> {
static const Scalar::Type id = Scalar::Int16;
static const JSProtoKey protoKey = JSProto_Int16Array;
};
template <>
struct TypeIDOfType<uint16_t> {
static const Scalar::Type id = Scalar::Uint16;
static const JSProtoKey protoKey = JSProto_Uint16Array;
};
template <>
struct TypeIDOfType<int32_t> {
static const Scalar::Type id = Scalar::Int32;
static const JSProtoKey protoKey = JSProto_Int32Array;
};
template <>
struct TypeIDOfType<uint32_t> {
static const Scalar::Type id = Scalar::Uint32;
static const JSProtoKey protoKey = JSProto_Uint32Array;
};
template <>
struct TypeIDOfType<int64_t> {
static const Scalar::Type id = Scalar::BigInt64;
static const JSProtoKey protoKey = JSProto_BigInt64Array;
};
template <>
struct TypeIDOfType<uint64_t> {
static const Scalar::Type id = Scalar::BigUint64;
static const JSProtoKey protoKey = JSProto_BigUint64Array;
};
template <>
struct TypeIDOfType<float> {
static const Scalar::Type id = Scalar::Float32;
static const JSProtoKey protoKey = JSProto_Float32Array;
};
template <>
struct TypeIDOfType<double> {
static const Scalar::Type id = Scalar::Float64;
static const JSProtoKey protoKey = JSProto_Float64Array;
};
template <>
struct TypeIDOfType<uint8_clamped> {
static const Scalar::Type id = Scalar::Uint8Clamped;
static const JSProtoKey protoKey = JSProto_Uint8ClampedArray;
};
class SharedOps {
public:
template <typename T>
static T load(SharedMem<T*> addr) {
return js::jit::AtomicOperations::loadSafeWhenRacy(addr);
}
template <typename T>
static void store(SharedMem<T*> addr, T value) {
js::jit::AtomicOperations::storeSafeWhenRacy(addr, value);
}
template <typename T>
static void memcpy(SharedMem<T*> dest, SharedMem<T*> src, size_t size) {
js::jit::AtomicOperations::memcpySafeWhenRacy(dest, src, size);
}
template <typename T>
static void memmove(SharedMem<T*> dest, SharedMem<T*> src, size_t size) {
js::jit::AtomicOperations::memmoveSafeWhenRacy(dest, src, size);
}
template <typename T>
static void podCopy(SharedMem<T*> dest, SharedMem<T*> src, size_t nelem) {
js::jit::AtomicOperations::podCopySafeWhenRacy(dest, src, nelem);
}
template <typename T>
static void podMove(SharedMem<T*> dest, SharedMem<T*> src, size_t nelem) {
js::jit::AtomicOperations::podMoveSafeWhenRacy(dest, src, nelem);
}
static SharedMem<void*> extract(TypedArrayObject* obj) {
return obj->dataPointerEither();
}
};
class UnsharedOps {
public:
template <typename T>
static T load(SharedMem<T*> addr) {
return *addr.unwrapUnshared();
}
template <typename T>
static void store(SharedMem<T*> addr, T value) {
*addr.unwrapUnshared() = value;
}
template <typename T>
static void memcpy(SharedMem<T*> dest, SharedMem<T*> src, size_t size) {
::memcpy(dest.unwrapUnshared(), src.unwrapUnshared(), size);
}
template <typename T>
static void memmove(SharedMem<T*> dest, SharedMem<T*> src, size_t size) {
::memmove(dest.unwrapUnshared(), src.unwrapUnshared(), size);
}
template <typename T>
static void podCopy(SharedMem<T*> dest, SharedMem<T*> src, size_t nelem) {
// std::copy_n better matches the argument values/types of this
// function, but as noted below it allows the input/output ranges to
// overlap. std::copy does not, so use it so the compiler has extra
// ability to optimize.
const auto* first = src.unwrapUnshared();
const auto* last = first + nelem;
auto* result = dest.unwrapUnshared();
std::copy(first, last, result);
}
template <typename T>
static void podMove(SharedMem<T*> dest, SharedMem<T*> src, size_t n) {
// std::copy_n copies from |src| to |dest| starting from |src|, so
// input/output ranges *may* permissibly overlap, as this function
// allows.
const auto* start = src.unwrapUnshared();
auto* result = dest.unwrapUnshared();
std::copy_n(start, n, result);
}
static SharedMem<void*> extract(TypedArrayObject* obj) {
return SharedMem<void*>::unshared(obj->dataPointerUnshared());
}
};
template <typename T, typename Ops>
class ElementSpecific {
public:
/*
* Copy |source|'s elements into |target|, starting at |target[offset]|.
* Act as if the assignments occurred from a fresh copy of |source|, in
* case the two memory ranges overlap.
*/
static bool setFromTypedArray(Handle<TypedArrayObject*> target,
size_t targetLength,
Handle<TypedArrayObject*> source,
size_t sourceLength, size_t offset) {
// WARNING: |source| may be an unwrapped typed array from a different
// compartment. Proceed with caution!
MOZ_ASSERT(TypeIDOfType<T>::id == target->type(),
"calling wrong setFromTypedArray specialization");
MOZ_ASSERT(!target->hasDetachedBuffer(), "target isn't detached");
MOZ_ASSERT(!source->hasDetachedBuffer(), "source isn't detached");
MOZ_ASSERT(*target->length() >= targetLength, "target isn't shrunk");
MOZ_ASSERT(*source->length() >= sourceLength, "source isn't shrunk");
MOZ_ASSERT(offset <= targetLength);
MOZ_ASSERT(sourceLength <= targetLength - offset);
if (TypedArrayObject::sameBuffer(target, source)) {
return setFromOverlappingTypedArray(target, targetLength, source,
sourceLength, offset);
}
SharedMem<T*> dest =
target->dataPointerEither().template cast<T*>() + offset;
size_t count = sourceLength;
if (source->type() == target->type()) {
Ops::podCopy(dest, source->dataPointerEither().template cast<T*>(),
count);
return true;
}
SharedMem<void*> data = Ops::extract(source);
switch (source->type()) {
case Scalar::Int8: {
SharedMem<int8_t*> src = data.cast<int8_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Uint8:
case Scalar::Uint8Clamped: {
SharedMem<uint8_t*> src = data.cast<uint8_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Int16: {
SharedMem<int16_t*> src = data.cast<int16_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Uint16: {
SharedMem<uint16_t*> src = data.cast<uint16_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Int32: {
SharedMem<int32_t*> src = data.cast<int32_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Uint32: {
SharedMem<uint32_t*> src = data.cast<uint32_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::BigInt64: {
SharedMem<int64_t*> src = data.cast<int64_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::BigUint64: {
SharedMem<uint64_t*> src = data.cast<uint64_t*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Float32: {
SharedMem<float*> src = data.cast<float*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
case Scalar::Float64: {
SharedMem<double*> src = data.cast<double*>();
for (size_t i = 0; i < count; ++i) {
Ops::store(dest++, ConvertNumber<T>(Ops::load(src++)));
}
break;
}
default:
MOZ_CRASH("setFromTypedArray with a typed array with bogus type");
}
return true;
}
/*
* Copy |source[0]| to |source[len]| (exclusive) elements into the typed
* array |target|, starting at index |offset|. |source| must not be a
* typed array.
*/
static bool setFromNonTypedArray(JSContext* cx,
Handle<TypedArrayObject*> target,
HandleObject source, size_t len,
size_t offset = 0) {
MOZ_ASSERT(target->type() == TypeIDOfType<T>::id,
"target type and NativeType must match");
MOZ_ASSERT(!source->is<TypedArrayObject>(),
"use setFromTypedArray instead of this method");
MOZ_ASSERT_IF(target->hasDetachedBuffer(), target->length().isNothing());
size_t i = 0;
if (source->is<NativeObject>()) {
size_t targetLength = target->length().valueOr(0);
if (offset <= targetLength && len <= targetLength - offset) {
// Attempt fast-path infallible conversion of dense elements up to
// the first potentially side-effectful lookup or conversion.
size_t bound = std::min<size_t>(
source->as<NativeObject>().getDenseInitializedLength(), len);
SharedMem<T*> dest =
target->dataPointerEither().template cast<T*>() + offset;
MOZ_ASSERT(!canConvertInfallibly(MagicValue(JS_ELEMENTS_HOLE)),
"the following loop must abort on holes");
const Value* srcValues = source->as<NativeObject>().getDenseElements();
for (; i < bound; i++) {
if (!canConvertInfallibly(srcValues[i])) {
break;
}
Ops::store(dest + i, infallibleValueToNative(srcValues[i]));
}
if (i == len) {
return true;
}
}
}
// Convert and copy any remaining elements generically.
RootedValue v(cx);
for (; i < len; i++) {
if constexpr (sizeof(i) == sizeof(uint32_t)) {
if (!GetElement(cx, source, source, uint32_t(i), &v)) {
return false;
}
} else {
if (!GetElementLargeIndex(cx, source, source, i, &v)) {
return false;
}
}
T n;
if (!valueToNative(cx, v, &n)) {
return false;
}
// Ignore out-of-bounds writes, but still execute getElement/valueToNative
// because of observable side-effects.
if (offset + i >= target->length().valueOr(0)) {
continue;
}
MOZ_ASSERT(!target->hasDetachedBuffer());
// Compute every iteration in case getElement/valueToNative
// detaches the underlying array buffer or GC moves the data.
SharedMem<T*> dest =
target->dataPointerEither().template cast<T*>() + offset + i;
Ops::store(dest, n);
}
return true;
}
/*
* Copy |source| into the typed array |target|.
*/
static bool initFromIterablePackedArray(
JSContext* cx, Handle<FixedLengthTypedArrayObject*> target,
Handle<ArrayObject*> source) {
MOZ_ASSERT(target->type() == TypeIDOfType<T>::id,
"target type and NativeType must match");
MOZ_ASSERT(!target->hasDetachedBuffer(), "target isn't detached");
MOZ_ASSERT(IsPackedArray(source), "source array must be packed");
MOZ_ASSERT(source->getDenseInitializedLength() <= target->length());
size_t len = source->getDenseInitializedLength();
size_t i = 0;
// Attempt fast-path infallible conversion of dense elements up to the
// first potentially side-effectful conversion.
SharedMem<T*> dest = target->dataPointerEither().template cast<T*>();
const Value* srcValues = source->getDenseElements();
for (; i < len; i++) {
if (!canConvertInfallibly(srcValues[i])) {
break;
}
Ops::store(dest + i, infallibleValueToNative(srcValues[i]));
}
if (i == len) {
return true;
}
// Convert any remaining elements by first collecting them into a
// temporary list, and then copying them into the typed array.
RootedValueVector values(cx);
if (!values.append(srcValues + i, len - i)) {
return false;
}
RootedValue v(cx);
for (size_t j = 0; j < values.length(); i++, j++) {
v = values[j];
T n;
if (!valueToNative(cx, v, &n)) {
return false;
}
// |target| is a newly allocated typed array and not yet visible to
// content script, so valueToNative can't detach the underlying
// buffer.
MOZ_ASSERT(i < target->length());
// Compute every iteration in case GC moves the data.
SharedMem<T*> newDest = target->dataPointerEither().template cast<T*>();
Ops::store(newDest + i, n);
}
return true;
}
private:
static bool setFromOverlappingTypedArray(Handle<TypedArrayObject*> target,
size_t targetLength,
Handle<TypedArrayObject*> source,
size_t sourceLength, size_t offset) {
// WARNING: |source| may be an unwrapped typed array from a different
// compartment. Proceed with caution!
MOZ_ASSERT(TypeIDOfType<T>::id == target->type(),
"calling wrong setFromTypedArray specialization");
MOZ_ASSERT(!target->hasDetachedBuffer(), "target isn't detached");
MOZ_ASSERT(!source->hasDetachedBuffer(), "source isn't detached");
MOZ_ASSERT(*target->length() >= targetLength, "target isn't shrunk");
MOZ_ASSERT(*source->length() >= sourceLength, "source isn't shrunk");
MOZ_ASSERT(TypedArrayObject::sameBuffer(target, source),
"the provided arrays don't actually overlap, so it's "
"undesirable to use this method");
MOZ_ASSERT(offset <= targetLength);
MOZ_ASSERT(sourceLength <= targetLength - offset);
SharedMem<T*> dest =
target->dataPointerEither().template cast<T*>() + offset;
size_t len = sourceLength;
if (source->type() == target->type()) {
SharedMem<T*> src = source->dataPointerEither().template cast<T*>();
Ops::podMove(dest, src, len);
return true;
}
// Copy |source| in case it overlaps the target elements being set.
size_t sourceByteLen = len * source->bytesPerElement();
void* data = target->zone()->template pod_malloc<uint8_t>(sourceByteLen);
if (!data) {
return false;
}
Ops::memcpy(SharedMem<void*>::unshared(data), source->dataPointerEither(),
sourceByteLen);
switch (source->type()) {
case Scalar::Int8: {
int8_t* src = static_cast<int8_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Uint8:
case Scalar::Uint8Clamped: {
uint8_t* src = static_cast<uint8_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Int16: {
int16_t* src = static_cast<int16_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Uint16: {
uint16_t* src = static_cast<uint16_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Int32: {
int32_t* src = static_cast<int32_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Uint32: {
uint32_t* src = static_cast<uint32_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::BigInt64: {
int64_t* src = static_cast<int64_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::BigUint64: {
uint64_t* src = static_cast<uint64_t*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Float32: {
float* src = static_cast<float*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
case Scalar::Float64: {
double* src = static_cast<double*>(data);
for (size_t i = 0; i < len; ++i) {
Ops::store(dest++, ConvertNumber<T>(*src++));
}
break;
}
default:
MOZ_CRASH(
"setFromOverlappingTypedArray with a typed array with bogus type");
}
js_free(data);
return true;
}
static bool canConvertInfallibly(const Value& v) {
if (TypeIDOfType<T>::id == Scalar::BigInt64 ||
TypeIDOfType<T>::id == Scalar::BigUint64) {
// Numbers, Null, Undefined, and Symbols throw a TypeError. Strings may
// OOM and Objects may have side-effects.
return v.isBigInt() || v.isBoolean();
}
// BigInts and Symbols throw a TypeError. Strings may OOM and Objects may
// have side-effects.
return v.isNumber() || v.isBoolean() || v.isNull() || v.isUndefined();
}
static T infallibleValueToNative(const Value& v) {
if (TypeIDOfType<T>::id == Scalar::BigInt64) {
if (v.isBigInt()) {
return T(BigInt::toInt64(v.toBigInt()));
}
return T(v.toBoolean());
}
if (TypeIDOfType<T>::id == Scalar::BigUint64) {
if (v.isBigInt()) {
return T(BigInt::toUint64(v.toBigInt()));
}
return T(v.toBoolean());
}
if (v.isInt32()) {
return T(v.toInt32());
}
if (v.isDouble()) {
return doubleToNative(v.toDouble());
}
if (v.isBoolean()) {
return T(v.toBoolean());
}
if (v.isNull()) {
return T(0);
}
MOZ_ASSERT(v.isUndefined());
return TypeIsFloatingPoint<T>() ? T(JS::GenericNaN()) : T(0);
}
static bool valueToNative(JSContext* cx, HandleValue v, T* result) {
MOZ_ASSERT(!v.isMagic());
if (MOZ_LIKELY(canConvertInfallibly(v))) {
*result = infallibleValueToNative(v);
return true;
}
if (std::is_same_v<T, int64_t>) {
JS_TRY_VAR_OR_RETURN_FALSE(cx, *result, ToBigInt64(cx, v));
return true;
}
if (std::is_same_v<T, uint64_t>) {
JS_TRY_VAR_OR_RETURN_FALSE(cx, *result, ToBigUint64(cx, v));
return true;
}
double d;
MOZ_ASSERT(v.isString() || v.isObject() || v.isSymbol() || v.isBigInt());
if (!(v.isString() ? StringToNumber(cx, v.toString(), &d)
: ToNumber(cx, v, &d))) {
return false;
}
*result = doubleToNative(d);
return true;
}
static T doubleToNative(double d) {
if (TypeIsFloatingPoint<T>()) {
// The JS spec doesn't distinguish among different NaN values, and
// it deliberately doesn't specify the bit pattern written to a
// typed array when NaN is written into it. This bit-pattern
// inconsistency could confuse differential testing, so always
// canonicalize NaN values in differential testing.
if (js::SupportDifferentialTesting()) {
d = JS::CanonicalizeNaN(d);
}
return T(d);
}
if (MOZ_UNLIKELY(std::isnan(d))) {
return T(0);
}
if (TypeIDOfType<T>::id == Scalar::Uint8Clamped) {
return T(d);
}
if (TypeIsUnsigned<T>()) {
return T(JS::ToUint32(d));
}
return T(JS::ToInt32(d));
}
};
inline gc::AllocKind js::FixedLengthTypedArrayObject::allocKindForTenure()
const {
// Fixed length typed arrays in the nursery may have a lazily allocated
// buffer. Make sure there is room for the array's fixed data when moving the
// array.
if (hasBuffer()) {
return NativeObject::allocKindForTenure();
}
gc::AllocKind allocKind;
if (hasInlineElements()) {
allocKind = AllocKindForLazyBuffer(byteLength());
} else {
allocKind = gc::GetGCObjectKind(getClass());
}
return gc::ForegroundToBackgroundAllocKind(allocKind);
}
/* static */ gc::AllocKind
js::FixedLengthTypedArrayObject::AllocKindForLazyBuffer(size_t nbytes) {
MOZ_ASSERT(nbytes <= INLINE_BUFFER_LIMIT);
if (nbytes == 0) {
nbytes += sizeof(uint8_t);
}
size_t dataSlots = AlignBytes(nbytes, sizeof(Value)) / sizeof(Value);
MOZ_ASSERT(nbytes <= dataSlots * sizeof(Value));
return gc::GetGCObjectKind(FIXED_DATA_START + dataSlots);
}
} // namespace js
#endif // vm_TypedArrayObject_inl_h