
This pattern (read-only base, writable subclass) was picked to maintain a *Buffer option that doesn't copy the source bits when parsing. ByteBuffer and Buffer both copy. I'm open to discussion on what the type relationship would be, though :) Tests have been added to ensure the symmetric nature of read/write operations. BUG= R=bcornell@google.com, pthatcher@webrtc.org Review URL: https://webrtc-codereview.appspot.com/45259005 Cr-Commit-Position: refs/heads/master@{#9107}
282 lines
8.6 KiB
C++
282 lines
8.6 KiB
C++
/*
|
|
* Copyright 2015 The WebRTC 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 in the root of the source
|
|
* tree. An additional intellectual property rights grant can be found
|
|
* in the file PATENTS. All contributing project authors may
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
*/
|
|
|
|
#include "webrtc/base/bitbuffer.h"
|
|
|
|
#include <algorithm>
|
|
#include <limits>
|
|
|
|
#include "webrtc/base/checks.h"
|
|
|
|
namespace {
|
|
|
|
// Returns the lowest (right-most) |bit_count| bits in |byte|.
|
|
uint8 LowestBits(uint8 byte, size_t bit_count) {
|
|
DCHECK_LE(bit_count, 8u);
|
|
return byte & ((1 << bit_count) - 1);
|
|
}
|
|
|
|
// Returns the highest (left-most) |bit_count| bits in |byte|, shifted to the
|
|
// lowest bits (to the right).
|
|
uint8 HighestBits(uint8 byte, size_t bit_count) {
|
|
DCHECK_LE(bit_count, 8u);
|
|
uint8 shift = 8 - static_cast<uint8>(bit_count);
|
|
uint8 mask = 0xFF << shift;
|
|
return (byte & mask) >> shift;
|
|
}
|
|
|
|
// Returns the highest byte of |val| in a uint8.
|
|
uint8 HighestByte(uint64 val) {
|
|
return static_cast<uint8>(val >> 56);
|
|
}
|
|
|
|
// Returns the result of writing partial data from |source|, of
|
|
// |source_bit_count| size in the highest bits, to |target| at
|
|
// |target_bit_offset| from the highest bit.
|
|
uint8 WritePartialByte(uint8 source, size_t source_bit_count,
|
|
uint8 target, size_t target_bit_offset) {
|
|
DCHECK(target_bit_offset < 8);
|
|
DCHECK(source_bit_count < 9);
|
|
DCHECK(source_bit_count <= (8 - target_bit_offset));
|
|
// Generate a mask for just the bits we're going to overwrite, so:
|
|
uint8 mask =
|
|
// The number of bits we want, in the most significant bits...
|
|
static_cast<uint8>(0xFF << (8 - source_bit_count))
|
|
// ...shifted over to the target offset from the most signficant bit.
|
|
>> target_bit_offset;
|
|
|
|
// We want the target, with the bits we'll overwrite masked off, or'ed with
|
|
// the bits from the source we want.
|
|
return (target & ~mask) | (source >> target_bit_offset);
|
|
}
|
|
|
|
// Counts the number of bits used in the binary representation of val.
|
|
size_t CountBits(uint64 val) {
|
|
size_t bit_count = 0;
|
|
while (val != 0) {
|
|
bit_count++;
|
|
val >>= 1;
|
|
}
|
|
return bit_count;
|
|
}
|
|
|
|
} // namespace
|
|
|
|
namespace rtc {
|
|
|
|
BitBuffer::BitBuffer(const uint8* bytes, size_t byte_count)
|
|
: bytes_(bytes), byte_count_(byte_count), byte_offset_(), bit_offset_() {
|
|
DCHECK(static_cast<uint64>(byte_count_) <=
|
|
std::numeric_limits<uint32>::max());
|
|
}
|
|
|
|
uint64 BitBuffer::RemainingBitCount() const {
|
|
return (static_cast<uint64>(byte_count_) - byte_offset_) * 8 - bit_offset_;
|
|
}
|
|
|
|
bool BitBuffer::ReadUInt8(uint8* val) {
|
|
uint32 bit_val;
|
|
if (!ReadBits(&bit_val, sizeof(uint8) * 8)) {
|
|
return false;
|
|
}
|
|
DCHECK(bit_val <= std::numeric_limits<uint8>::max());
|
|
*val = static_cast<uint8>(bit_val);
|
|
return true;
|
|
}
|
|
|
|
bool BitBuffer::ReadUInt16(uint16* val) {
|
|
uint32 bit_val;
|
|
if (!ReadBits(&bit_val, sizeof(uint16) * 8)) {
|
|
return false;
|
|
}
|
|
DCHECK(bit_val <= std::numeric_limits<uint16>::max());
|
|
*val = static_cast<uint16>(bit_val);
|
|
return true;
|
|
}
|
|
|
|
bool BitBuffer::ReadUInt32(uint32* val) {
|
|
return ReadBits(val, sizeof(uint32) * 8);
|
|
}
|
|
|
|
bool BitBuffer::PeekBits(uint32* val, size_t bit_count) {
|
|
if (!val || bit_count > RemainingBitCount() || bit_count > 32) {
|
|
return false;
|
|
}
|
|
const uint8* bytes = bytes_ + byte_offset_;
|
|
size_t remaining_bits_in_current_byte = 8 - bit_offset_;
|
|
uint32 bits = LowestBits(*bytes++, remaining_bits_in_current_byte);
|
|
// If we're reading fewer bits than what's left in the current byte, just
|
|
// return the portion of this byte that we need.
|
|
if (bit_count < remaining_bits_in_current_byte) {
|
|
*val = HighestBits(bits, bit_offset_ + bit_count);
|
|
return true;
|
|
}
|
|
// Otherwise, subtract what we've read from the bit count and read as many
|
|
// full bytes as we can into bits.
|
|
bit_count -= remaining_bits_in_current_byte;
|
|
while (bit_count >= 8) {
|
|
bits = (bits << 8) | *bytes++;
|
|
bit_count -= 8;
|
|
}
|
|
// Whatever we have left is smaller than a byte, so grab just the bits we need
|
|
// and shift them into the lowest bits.
|
|
if (bit_count > 0) {
|
|
bits <<= bit_count;
|
|
bits |= HighestBits(*bytes, bit_count);
|
|
}
|
|
*val = bits;
|
|
return true;
|
|
}
|
|
|
|
bool BitBuffer::ReadBits(uint32* val, size_t bit_count) {
|
|
return PeekBits(val, bit_count) && ConsumeBits(bit_count);
|
|
}
|
|
|
|
bool BitBuffer::ConsumeBytes(size_t byte_count) {
|
|
return ConsumeBits(byte_count * 8);
|
|
}
|
|
|
|
bool BitBuffer::ConsumeBits(size_t bit_count) {
|
|
if (bit_count > RemainingBitCount()) {
|
|
return false;
|
|
}
|
|
|
|
byte_offset_ += (bit_offset_ + bit_count) / 8;
|
|
bit_offset_ = (bit_offset_ + bit_count) % 8;
|
|
return true;
|
|
}
|
|
|
|
bool BitBuffer::ReadExponentialGolomb(uint32* val) {
|
|
if (!val) {
|
|
return false;
|
|
}
|
|
// Store off the current byte/bit offset, in case we want to restore them due
|
|
// to a failed parse.
|
|
size_t original_byte_offset = byte_offset_;
|
|
size_t original_bit_offset = bit_offset_;
|
|
|
|
// Count the number of leading 0 bits by peeking/consuming them one at a time.
|
|
size_t zero_bit_count = 0;
|
|
uint32 peeked_bit;
|
|
while (PeekBits(&peeked_bit, 1) && peeked_bit == 0) {
|
|
zero_bit_count++;
|
|
ConsumeBits(1);
|
|
}
|
|
|
|
// We should either be at the end of the stream, or the next bit should be 1.
|
|
DCHECK(!PeekBits(&peeked_bit, 1) || peeked_bit == 1);
|
|
|
|
// The bit count of the value is the number of zeros + 1. Make sure that many
|
|
// bits fits in a uint32 and that we have enough bits left for it, and then
|
|
// read the value.
|
|
size_t value_bit_count = zero_bit_count + 1;
|
|
if (value_bit_count > 32 || !ReadBits(val, value_bit_count)) {
|
|
CHECK(Seek(original_byte_offset, original_bit_offset));
|
|
return false;
|
|
}
|
|
*val -= 1;
|
|
return true;
|
|
}
|
|
|
|
void BitBuffer::GetCurrentOffset(
|
|
size_t* out_byte_offset, size_t* out_bit_offset) {
|
|
CHECK(out_byte_offset != NULL);
|
|
CHECK(out_bit_offset != NULL);
|
|
*out_byte_offset = byte_offset_;
|
|
*out_bit_offset = bit_offset_;
|
|
}
|
|
|
|
bool BitBuffer::Seek(size_t byte_offset, size_t bit_offset) {
|
|
if (byte_offset > byte_count_ || bit_offset > 7 ||
|
|
(byte_offset == byte_count_ && bit_offset > 0)) {
|
|
return false;
|
|
}
|
|
byte_offset_ = byte_offset;
|
|
bit_offset_ = bit_offset;
|
|
return true;
|
|
}
|
|
|
|
BitBufferWriter::BitBufferWriter(uint8* bytes, size_t byte_count)
|
|
: BitBuffer(bytes, byte_count), writable_bytes_(bytes) {
|
|
}
|
|
|
|
bool BitBufferWriter::WriteUInt8(uint8 val) {
|
|
return WriteBits(val, sizeof(uint8) * 8);
|
|
}
|
|
|
|
bool BitBufferWriter::WriteUInt16(uint16 val) {
|
|
return WriteBits(val, sizeof(uint16) * 8);
|
|
}
|
|
|
|
bool BitBufferWriter::WriteUInt32(uint32 val) {
|
|
return WriteBits(val, sizeof(uint32) * 8);
|
|
}
|
|
|
|
bool BitBufferWriter::WriteBits(uint64 val, size_t bit_count) {
|
|
if (bit_count > RemainingBitCount()) {
|
|
return false;
|
|
}
|
|
size_t total_bits = bit_count;
|
|
|
|
// For simplicity, push the bits we want to read from val to the highest bits.
|
|
val <<= (sizeof(uint64) * 8 - bit_count);
|
|
|
|
uint8* bytes = writable_bytes_ + byte_offset_;
|
|
|
|
// The first byte is relatively special; the bit offset to write to may put us
|
|
// in the middle of the byte, and the total bit count to write may require we
|
|
// save the bits at the end of the byte.
|
|
size_t remaining_bits_in_current_byte = 8 - bit_offset_;
|
|
size_t bits_in_first_byte =
|
|
std::min(bit_count, remaining_bits_in_current_byte);
|
|
*bytes = WritePartialByte(
|
|
HighestByte(val), bits_in_first_byte, *bytes, bit_offset_);
|
|
if (bit_count <= remaining_bits_in_current_byte) {
|
|
// Nothing left to write, so quit early.
|
|
return ConsumeBits(total_bits);
|
|
}
|
|
|
|
// Subtract what we've written from the bit count, shift it off the value, and
|
|
// write the remaining full bytes.
|
|
val <<= bits_in_first_byte;
|
|
bytes++;
|
|
bit_count -= bits_in_first_byte;
|
|
while (bit_count >= 8) {
|
|
*bytes++ = HighestByte(val);
|
|
val <<= 8;
|
|
bit_count -= 8;
|
|
}
|
|
|
|
// Last byte may also be partial, so write the remaining bits from the top of
|
|
// val.
|
|
if (bit_count > 0) {
|
|
*bytes = WritePartialByte(HighestByte(val), bit_count, *bytes, 0);
|
|
}
|
|
|
|
// All done! Consume the bits we've written.
|
|
return ConsumeBits(total_bits);
|
|
}
|
|
|
|
bool BitBufferWriter::WriteExponentialGolomb(uint32 val) {
|
|
// We don't support reading UINT32_MAX, because it doesn't fit in a uint32
|
|
// when encoded, so don't support writing it either.
|
|
if (val == std::numeric_limits<uint32>::max()) {
|
|
return false;
|
|
}
|
|
uint64 val_to_encode = static_cast<uint64>(val) + 1;
|
|
|
|
// We need to write CountBits(val+1) 0s and then val+1. Since val (as a
|
|
// uint64) has leading zeros, we can just write the total golomb encoded size
|
|
// worth of bits, knowing the value will appear last.
|
|
return WriteBits(val_to_encode, CountBits(val_to_encode) * 2 - 1);
|
|
}
|
|
|
|
} // namespace rtc
|