gecko-dev/xpcom/glue/nsTArray-inl.h
Nathan Froyd 3d1192cadc Bug 1271751 - part 6 - make ShiftData actually move its elements rather than copying; r=erahm
This change eliminates the last use of the Copy* family of functions.
2016-07-08 16:59:05 -04:00

464 lines
16 KiB
C++

/* -*- 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
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
#ifndef nsTArray_h__
# error "Don't include this file directly"
#endif
template<class Alloc, class Copy>
nsTArray_base<Alloc, Copy>::nsTArray_base()
: mHdr(EmptyHdr())
{
MOZ_COUNT_CTOR(nsTArray_base);
}
template<class Alloc, class Copy>
nsTArray_base<Alloc, Copy>::~nsTArray_base()
{
if (mHdr != EmptyHdr() && !UsesAutoArrayBuffer()) {
Alloc::Free(mHdr);
}
MOZ_COUNT_DTOR(nsTArray_base);
}
template<class Alloc, class Copy>
const nsTArrayHeader*
nsTArray_base<Alloc, Copy>::GetAutoArrayBufferUnsafe(size_t aElemAlign) const
{
// Assuming |this| points to an nsAutoArray, we want to get a pointer to
// mAutoBuf. So just cast |this| to nsAutoArray* and read &mAutoBuf!
const void* autoBuf =
&reinterpret_cast<const AutoTArray<nsTArray<uint32_t>, 1>*>(this)->mAutoBuf;
// If we're on a 32-bit system and aElemAlign is 8, we need to adjust our
// pointer to take into account the extra alignment in the auto array.
static_assert(sizeof(void*) != 4 ||
(MOZ_ALIGNOF(mozilla::AlignedElem<8>) == 8 &&
sizeof(AutoTArray<mozilla::AlignedElem<8>, 1>) ==
sizeof(void*) + sizeof(nsTArrayHeader) +
4 + sizeof(mozilla::AlignedElem<8>)),
"auto array padding wasn't what we expected");
// We don't support alignments greater than 8 bytes.
MOZ_ASSERT(aElemAlign <= 4 || aElemAlign == 8,
"unsupported alignment.");
if (sizeof(void*) == 4 && aElemAlign == 8) {
autoBuf = reinterpret_cast<const char*>(autoBuf) + 4;
}
return reinterpret_cast<const Header*>(autoBuf);
}
template<class Alloc, class Copy>
bool
nsTArray_base<Alloc, Copy>::UsesAutoArrayBuffer() const
{
if (!mHdr->mIsAutoArray) {
return false;
}
// This is nuts. If we were sane, we'd pass aElemAlign as a parameter to
// this function. Unfortunately this function is called in nsTArray_base's
// destructor, at which point we don't know elem_type's alignment.
//
// We'll fall on our face and return true when we should say false if
//
// * we're not using our auto buffer,
// * aElemAlign == 4, and
// * mHdr == GetAutoArrayBuffer(8).
//
// This could happen if |*this| lives on the heap and malloc allocated our
// buffer on the heap adjacent to |*this|.
//
// However, we can show that this can't happen. If |this| is an auto array
// (as we ensured at the beginning of the method), GetAutoArrayBuffer(8)
// always points to memory owned by |*this|, because (as we assert below)
//
// * GetAutoArrayBuffer(8) is at most 4 bytes past GetAutoArrayBuffer(4), and
// * sizeof(nsTArrayHeader) > 4.
//
// Since AutoTArray always contains an nsTArrayHeader,
// GetAutoArrayBuffer(8) will always point inside the auto array object,
// even if it doesn't point at the beginning of the header.
//
// Note that this means that we can't store elements with alignment 16 in an
// nsTArray, because GetAutoArrayBuffer(16) could lie outside the memory
// owned by this AutoTArray. We statically assert that elem_type's
// alignment is 8 bytes or less in AutoTArray.
static_assert(sizeof(nsTArrayHeader) > 4,
"see comment above");
#ifdef DEBUG
ptrdiff_t diff = reinterpret_cast<const char*>(GetAutoArrayBuffer(8)) -
reinterpret_cast<const char*>(GetAutoArrayBuffer(4));
MOZ_ASSERT(diff >= 0 && diff <= 4,
"GetAutoArrayBuffer doesn't do what we expect.");
#endif
return mHdr == GetAutoArrayBuffer(4) || mHdr == GetAutoArrayBuffer(8);
}
// defined in nsTArray.cpp
bool IsTwiceTheRequiredBytesRepresentableAsUint32(size_t aCapacity,
size_t aElemSize);
template<class Alloc, class Copy>
template<typename ActualAlloc>
typename ActualAlloc::ResultTypeProxy
nsTArray_base<Alloc, Copy>::EnsureCapacity(size_type aCapacity,
size_type aElemSize)
{
// This should be the most common case so test this first
if (aCapacity <= mHdr->mCapacity) {
return ActualAlloc::SuccessResult();
}
// If the requested memory allocation exceeds size_type(-1)/2, then
// our doubling algorithm may not be able to allocate it.
// Additionally, if it exceeds uint32_t(-1) then we couldn't fit in the
// Header::mCapacity member. Just bail out in cases like that. We don't want
// to be allocating 2 GB+ arrays anyway.
if (!IsTwiceTheRequiredBytesRepresentableAsUint32(aCapacity, aElemSize)) {
ActualAlloc::SizeTooBig((size_t)aCapacity * aElemSize);
return ActualAlloc::FailureResult();
}
size_t reqSize = sizeof(Header) + aCapacity * aElemSize;
if (mHdr == EmptyHdr()) {
// Malloc() new data
Header* header = static_cast<Header*>(ActualAlloc::Malloc(reqSize));
if (!header) {
return ActualAlloc::FailureResult();
}
header->mLength = 0;
header->mCapacity = aCapacity;
header->mIsAutoArray = 0;
mHdr = header;
return ActualAlloc::SuccessResult();
}
// We increase our capacity so that the allocated buffer grows exponentially,
// which gives us amortized O(1) appending. Below the threshold, we use
// powers-of-two. Above the threshold, we grow by at least 1.125, rounding up
// to the nearest MiB.
const size_t slowGrowthThreshold = 8 * 1024 * 1024;
size_t bytesToAlloc;
if (reqSize >= slowGrowthThreshold) {
size_t currSize = sizeof(Header) + Capacity() * aElemSize;
size_t minNewSize = currSize + (currSize >> 3); // multiply by 1.125
bytesToAlloc = reqSize > minNewSize ? reqSize : minNewSize;
// Round up to the next multiple of MiB.
const size_t MiB = 1 << 20;
bytesToAlloc = MiB * ((bytesToAlloc + MiB - 1) / MiB);
} else {
// Round up to the next power of two.
bytesToAlloc = mozilla::RoundUpPow2(reqSize);
}
Header* header;
if (UsesAutoArrayBuffer() || !Copy::allowRealloc) {
// Malloc() and copy
header = static_cast<Header*>(ActualAlloc::Malloc(bytesToAlloc));
if (!header) {
return ActualAlloc::FailureResult();
}
Copy::MoveNonOverlappingRegionWithHeader(header, mHdr, Length(), aElemSize);
if (!UsesAutoArrayBuffer()) {
ActualAlloc::Free(mHdr);
}
} else {
// Realloc() existing data
header = static_cast<Header*>(ActualAlloc::Realloc(mHdr, bytesToAlloc));
if (!header) {
return ActualAlloc::FailureResult();
}
}
// How many elements can we fit in bytesToAlloc?
size_t newCapacity = (bytesToAlloc - sizeof(Header)) / aElemSize;
MOZ_ASSERT(newCapacity >= aCapacity, "Didn't enlarge the array enough!");
header->mCapacity = newCapacity;
mHdr = header;
return ActualAlloc::SuccessResult();
}
// We don't need use Alloc template parameter specified here because failure to
// shrink the capacity will leave the array unchanged.
template<class Alloc, class Copy>
void
nsTArray_base<Alloc, Copy>::ShrinkCapacity(size_type aElemSize,
size_t aElemAlign)
{
if (mHdr == EmptyHdr() || UsesAutoArrayBuffer()) {
return;
}
if (mHdr->mLength >= mHdr->mCapacity) { // should never be greater than...
return;
}
size_type length = Length();
if (IsAutoArray() && GetAutoArrayBuffer(aElemAlign)->mCapacity >= length) {
Header* header = GetAutoArrayBuffer(aElemAlign);
// Move the data, but don't copy the header to avoid overwriting mCapacity.
header->mLength = length;
Copy::MoveNonOverlappingRegion(header + 1, mHdr + 1, length, aElemSize);
nsTArrayFallibleAllocator::Free(mHdr);
mHdr = header;
return;
}
if (length == 0) {
MOZ_ASSERT(!IsAutoArray(), "autoarray should have fit 0 elements");
nsTArrayFallibleAllocator::Free(mHdr);
mHdr = EmptyHdr();
return;
}
size_type size = sizeof(Header) + length * aElemSize;
void* ptr = nsTArrayFallibleAllocator::Realloc(mHdr, size);
if (!ptr) {
return;
}
mHdr = static_cast<Header*>(ptr);
mHdr->mCapacity = length;
}
template<class Alloc, class Copy>
template<typename ActualAlloc>
void
nsTArray_base<Alloc, Copy>::ShiftData(index_type aStart,
size_type aOldLen, size_type aNewLen,
size_type aElemSize, size_t aElemAlign)
{
if (aOldLen == aNewLen) {
return;
}
// Determine how many elements need to be shifted
size_type num = mHdr->mLength - (aStart + aOldLen);
// Compute the resulting length of the array
mHdr->mLength += aNewLen - aOldLen;
if (mHdr->mLength == 0) {
ShrinkCapacity(aElemSize, aElemAlign);
} else {
// Maybe nothing needs to be shifted
if (num == 0) {
return;
}
// Perform shift (change units to bytes first)
aStart *= aElemSize;
aNewLen *= aElemSize;
aOldLen *= aElemSize;
char* baseAddr = reinterpret_cast<char*>(mHdr + 1) + aStart;
Copy::MoveOverlappingRegion(baseAddr + aNewLen, baseAddr + aOldLen, num, aElemSize);
}
}
template<class Alloc, class Copy>
template<typename ActualAlloc>
bool
nsTArray_base<Alloc, Copy>::InsertSlotsAt(index_type aIndex, size_type aCount,
size_type aElemSize,
size_t aElemAlign)
{
MOZ_ASSERT(aIndex <= Length(), "Bogus insertion index");
size_type newLen = Length() + aCount;
EnsureCapacity<ActualAlloc>(newLen, aElemSize);
// Check for out of memory conditions
if (Capacity() < newLen) {
return false;
}
// Move the existing elements as needed. Note that this will
// change our mLength, so no need to call IncrementLength.
ShiftData<ActualAlloc>(aIndex, 0, aCount, aElemSize, aElemAlign);
return true;
}
// nsTArray_base::IsAutoArrayRestorer is an RAII class which takes
// |nsTArray_base &array| in its constructor. When it's destructed, it ensures
// that
//
// * array.mIsAutoArray has the same value as it did when we started, and
// * if array has an auto buffer and mHdr would otherwise point to sEmptyHdr,
// array.mHdr points to array's auto buffer.
template<class Alloc, class Copy>
nsTArray_base<Alloc, Copy>::IsAutoArrayRestorer::IsAutoArrayRestorer(
nsTArray_base<Alloc, Copy>& aArray,
size_t aElemAlign)
: mArray(aArray)
, mElemAlign(aElemAlign)
, mIsAuto(aArray.IsAutoArray())
{
}
template<class Alloc, class Copy>
nsTArray_base<Alloc, Copy>::IsAutoArrayRestorer::~IsAutoArrayRestorer()
{
// Careful: We don't want to set mIsAutoArray = 1 on sEmptyHdr.
if (mIsAuto && mArray.mHdr == mArray.EmptyHdr()) {
// Call GetAutoArrayBufferUnsafe() because GetAutoArrayBuffer() asserts
// that mHdr->mIsAutoArray is true, which surely isn't the case here.
mArray.mHdr = mArray.GetAutoArrayBufferUnsafe(mElemAlign);
mArray.mHdr->mLength = 0;
} else if (mArray.mHdr != mArray.EmptyHdr()) {
mArray.mHdr->mIsAutoArray = mIsAuto;
}
}
template<class Alloc, class Copy>
template<typename ActualAlloc, class Allocator>
typename ActualAlloc::ResultTypeProxy
nsTArray_base<Alloc, Copy>::SwapArrayElements(nsTArray_base<Allocator,
Copy>& aOther,
size_type aElemSize,
size_t aElemAlign)
{
// EnsureNotUsingAutoArrayBuffer will set mHdr = sEmptyHdr even if we have an
// auto buffer. We need to point mHdr back to our auto buffer before we
// return, otherwise we'll forget that we have an auto buffer at all!
// IsAutoArrayRestorer takes care of this for us.
IsAutoArrayRestorer ourAutoRestorer(*this, aElemAlign);
typename nsTArray_base<Allocator, Copy>::IsAutoArrayRestorer
otherAutoRestorer(aOther, aElemAlign);
// If neither array uses an auto buffer which is big enough to store the
// other array's elements, then ensure that both arrays use malloc'ed storage
// and swap their mHdr pointers.
if ((!UsesAutoArrayBuffer() || Capacity() < aOther.Length()) &&
(!aOther.UsesAutoArrayBuffer() || aOther.Capacity() < Length())) {
if (!EnsureNotUsingAutoArrayBuffer<ActualAlloc>(aElemSize) ||
!aOther.template EnsureNotUsingAutoArrayBuffer<ActualAlloc>(aElemSize)) {
return ActualAlloc::FailureResult();
}
Header* temp = mHdr;
mHdr = aOther.mHdr;
aOther.mHdr = temp;
return ActualAlloc::SuccessResult();
}
// Swap the two arrays by copying, since at least one is using an auto
// buffer which is large enough to hold all of the aOther's elements. We'll
// copy the shorter array into temporary storage.
//
// (We could do better than this in some circumstances. Suppose we're
// swapping arrays X and Y. X has space for 2 elements in its auto buffer,
// but currently has length 4, so it's using malloc'ed storage. Y has length
// 2. When we swap X and Y, we don't need to use a temporary buffer; we can
// write Y straight into X's auto buffer, write X's malloc'ed buffer on top
// of Y, and then switch X to using its auto buffer.)
if (!ActualAlloc::Successful(EnsureCapacity<ActualAlloc>(aOther.Length(), aElemSize)) ||
!Allocator::Successful(aOther.template EnsureCapacity<Allocator>(Length(), aElemSize))) {
return ActualAlloc::FailureResult();
}
// The EnsureCapacity calls above shouldn't have caused *both* arrays to
// switch from their auto buffers to malloc'ed space.
MOZ_ASSERT(UsesAutoArrayBuffer() || aOther.UsesAutoArrayBuffer(),
"One of the arrays should be using its auto buffer.");
size_type smallerLength = XPCOM_MIN(Length(), aOther.Length());
size_type largerLength = XPCOM_MAX(Length(), aOther.Length());
void* smallerElements;
void* largerElements;
if (Length() <= aOther.Length()) {
smallerElements = Hdr() + 1;
largerElements = aOther.Hdr() + 1;
} else {
smallerElements = aOther.Hdr() + 1;
largerElements = Hdr() + 1;
}
// Allocate temporary storage for the smaller of the two arrays. We want to
// allocate this space on the stack, if it's not too large. Sounds like a
// job for AutoTArray! (One of the two arrays we're swapping is using an
// auto buffer, so we're likely not allocating a lot of space here. But one
// could, in theory, allocate a huge AutoTArray on the heap.)
AutoTArray<nsTArray_Impl<uint8_t, ActualAlloc>, 64> temp;
if (!ActualAlloc::Successful(temp.template EnsureCapacity<ActualAlloc>(smallerLength,
aElemSize))) {
return ActualAlloc::FailureResult();
}
Copy::MoveNonOverlappingRegion(temp.Elements(), smallerElements, smallerLength, aElemSize);
Copy::MoveNonOverlappingRegion(smallerElements, largerElements, largerLength, aElemSize);
Copy::MoveNonOverlappingRegion(largerElements, temp.Elements(), smallerLength, aElemSize);
// Swap the arrays' lengths.
MOZ_ASSERT((aOther.Length() == 0 || mHdr != EmptyHdr()) &&
(Length() == 0 || aOther.mHdr != EmptyHdr()),
"Don't set sEmptyHdr's length.");
size_type tempLength = Length();
// Avoid writing to EmptyHdr, since it can trigger false
// positives with TSan.
if (mHdr != EmptyHdr()) {
mHdr->mLength = aOther.Length();
}
if (aOther.mHdr != EmptyHdr()) {
aOther.mHdr->mLength = tempLength;
}
return ActualAlloc::SuccessResult();
}
template<class Alloc, class Copy>
template<typename ActualAlloc>
bool
nsTArray_base<Alloc, Copy>::EnsureNotUsingAutoArrayBuffer(size_type aElemSize)
{
if (UsesAutoArrayBuffer()) {
// If you call this on a 0-length array, we'll set that array's mHdr to
// sEmptyHdr, in flagrant violation of the AutoTArray invariants. It's
// up to you to set it back! (If you don't, the AutoTArray will forget
// that it has an auto buffer.)
if (Length() == 0) {
mHdr = EmptyHdr();
return true;
}
size_type size = sizeof(Header) + Length() * aElemSize;
Header* header = static_cast<Header*>(ActualAlloc::Malloc(size));
if (!header) {
return false;
}
Copy::MoveNonOverlappingRegionWithHeader(header, mHdr, Length(), aElemSize);
header->mCapacity = Length();
mHdr = header;
}
return true;
}