DeepSeek-V4.1-Flash
5b24dacee0
一、悬停高亮「几乎看不到、只是曲线换了颜色」（对比提交f845616的旧版观感） 根因：调色板第 1 色就是主题强调色（0xFF268BD2），而上一轮为提速把稠密信号的光晕改成 「按包络带原始几何 FillPath」——与曲线**逐像素同形**，于是曲线上的那层半透明强调色 完全看不出变化；旧版观感来自「lineWidth+4 的描边笔向两侧各外扩 2~3px」。 修法（PlotViewPaint.cpp::PaintSignalHaloFor）： 1) 逐列成对填四边形，上下左右都外扩 grow = (lineWidth + 4) * scale / 2 （竖向跃变处旧版 5px 笔涂出的正是那条整高色带）； 2) 不能把整条包络拼成一个多边形：满幅方波相邻列区间互相交叉，缠绕填充规则会在交叠处 相互抵消 ⇒ 填充变成一条条竖缝（实测只覆盖约一半带宽）； 3) alpha 从 0x55 调回 0x66（与旧版描边笔一致）。 代价仍只在「换信号/换视界/改样式」时付一次，此后每帧只 AlphaBlend。 三维对照（新增 haloshot 探针 +f845616工作树构建）：变化像素 旧 73726 / 上一轮 13529 / 本轮 74916，目视与旧版一致（跃变处的整高色带恢复）。 二、MATLAB .mat（v7.3 = HDF5）支持，含嵌套分组 新增只读 HDF5 子集（不引第三方库）：Utilities/Inc/MatH5.hpp + Utilities/Src/MatH5.cpp - 超级块 v0/v2；对象头 v1（含续块，主块长度以 headerSize 为准）；符号表（B-tree v1 + 本地堆）与 link 消息；数据空间 v1/v2；数据类型 v1；数据布局 v3（紧凑/连续/分块）； 分块数据集支持 deflate（zlib）；地址一律相对基地址解析。 新增 MATLAB 语义层：Utilities/Inc/MatLoader.hpp + Utilities/Src/MatLoader.cpp 1) Simulink SDI / SimulationData.Dataset（识别 #refs# + #sigstream#）： · #refs# 元素有 ElementType（signal/dataset）、Name、Values； · timeseries 的 Values/DataR2、Values/TimeR2 是 #sigstream#/<n> 的流编号； #data# 是 uint8 字节流，按 8 字节一组解释为 double（与 #length#/#sampleSize# 一致）； · 子数据集的 Values 是对象引用数组（子元素）⇒ 还原嵌套树； · 分组名 = 从根到信号的各级数据集名用 - 连接（如 A/A1/XYZ → 组 A-A1、信号 XYZ）， 信号名取元素 Name（空则退回 timeseries 的 Name，再退回 信号N）。 2) 普通 v7.3 变量：顶层数值矩阵按「列 = 信号、行 = 采样」读取（时间轴 = 采样序号）， 结构体按字段递归。 加载管线扩成多分组：新增 LoadSignalSourceGroups()（DataLoader）→ AsyncLoadState::groups → Application::LoadFile 逐个 AddGroup；其它格式仍只有一个分组。 文件对话框与 IsSupportedSignalFile 增加 .mat。 三、验证（全部离屏，不占用鼠标/键盘/屏幕） - 新增 haloshot（截图对照 f845616）、matcheck（15/15：4 组 8 信号、时间轴单调、采样数正确）、 matui（11/11：Application::LoadFile 端到端把 4 个分组加进数据层与信号列表）； - 回归 verify 49/49、panelanim 158/158、gridanim 93/93、hoverdrag 20/20、lanescroll 23/23、 datacheck 69/69、edgecheck 46/46、errcheck 19/19、halocheck 16/16、panflash 19/19、 panelstutter 8/8、stress 6/6； - make release / make debug 零警告；Makefile 与探针构建脚本补 -lz。 文档：Documents/agent_mem.md 追加本轮小节（含 4 条新规则：叠加层必须自带外扩、 缠绕填充不可用于自交包络带、HDF5 主消息块长度看 headerSize、 MATLAB 分块维度可能比数据空间多一维、按右侧补 1 对齐）。
954 lines
28 KiB
C++
954 lines
28 KiB
C++
// HDF5 只读子集：MATLAB 7.3 `.mat` 解析用（见 MatH5.hpp 的说明）。
|
||
#include "MatH5.hpp"
|
||
|
||
#include <zlib.h>
|
||
|
||
#include <algorithm>
|
||
#include <cstdio>
|
||
#include <cstring>
|
||
#include <filesystem>
|
||
|
||
namespace jdi
|
||
{
|
||
|
||
namespace
|
||
{
|
||
constexpr std::uint16_t kMessageDataspace = 0x0001;
|
||
constexpr std::uint16_t kMessageLink = 0x0006;
|
||
constexpr std::uint16_t kMessageDatatype = 0x0003;
|
||
constexpr std::uint16_t kMessageLayout = 0x0008;
|
||
constexpr std::uint16_t kMessageFilter = 0x000B;
|
||
constexpr std::uint16_t kMessageAttribute = 0x000C;
|
||
constexpr std::uint16_t kMessageContinuation = 0x0010;
|
||
constexpr std::uint16_t kMessageSymbolTable = 0x0011;
|
||
|
||
std::uint64_t ReadLE(const std::uint8_t* data, std::size_t size)
|
||
{
|
||
std::uint64_t value = 0;
|
||
for (std::size_t index = 0; index < size; ++index)
|
||
{
|
||
value |= static_cast<std::uint64_t>(data[index]) << (8 * index);
|
||
}
|
||
return value;
|
||
}
|
||
} // namespace
|
||
|
||
std::uint8_t MatHdf5::U8(std::uint64_t offset) const
|
||
{
|
||
return (offset < m_data.size()) ? m_data[static_cast<std::size_t>(offset)] : 0;
|
||
}
|
||
|
||
std::uint16_t MatHdf5::U16(std::uint64_t offset) const
|
||
{
|
||
return static_cast<std::uint16_t>(U8(offset)) | (static_cast<std::uint16_t>(U8(offset + 1)) << 8);
|
||
}
|
||
|
||
std::uint32_t MatHdf5::U32(std::uint64_t offset) const
|
||
{
|
||
return static_cast<std::uint32_t>(U16(offset)) | (static_cast<std::uint32_t>(U16(offset + 2)) << 16);
|
||
}
|
||
|
||
std::uint64_t MatHdf5::U64(std::uint64_t offset) const
|
||
{
|
||
return static_cast<std::uint64_t>(U32(offset)) | (static_cast<std::uint64_t>(U32(offset + 4)) << 32);
|
||
}
|
||
|
||
std::uint64_t MatHdf5::Addr(std::uint64_t offset) const
|
||
{
|
||
return (m_offsetSize == 8) ? U64(offset) : U32(offset);
|
||
}
|
||
|
||
std::uint64_t MatHdf5::Resolve(std::uint64_t address) const
|
||
{
|
||
if (address == 0 || address == 0xFFFFFFFFFFFFFFFFull)
|
||
{
|
||
return 0;
|
||
}
|
||
return address + m_base;
|
||
}
|
||
|
||
bool MatHdf5::Open(const std::wstring& path, std::wstring& error)
|
||
{
|
||
error.clear();
|
||
Close();
|
||
std::error_code code;
|
||
const std::uintmax_t size = std::filesystem::file_size(path, code);
|
||
if (code || size == 0)
|
||
{
|
||
error = L"无法读取文件大小。";
|
||
return false;
|
||
}
|
||
m_data.resize(static_cast<std::size_t>(size));
|
||
FILE* file = nullptr;
|
||
if (_wfopen_s(&file, path.c_str(), L"rb") != 0 || file == nullptr)
|
||
{
|
||
error = L"无法打开文件。";
|
||
return false;
|
||
}
|
||
const std::size_t read = std::fread(m_data.data(), 1, m_data.size(), file);
|
||
std::fclose(file);
|
||
if (read != m_data.size())
|
||
{
|
||
error = L"读取文件失败。";
|
||
return false;
|
||
}
|
||
|
||
// 超级块：签名可能出现在 0、512、1024…（MATLAB 用 512 字节 user block）。
|
||
std::size_t signature = m_data.size();
|
||
for (std::size_t index = 0; index + 8 <= m_data.size() && index <= 4096; index += 512)
|
||
{
|
||
if (m_data[index] == 0x89 && m_data[index + 1] == 'H' && m_data[index + 2] == 'D' &&
|
||
m_data[index + 3] == 'F')
|
||
{
|
||
signature = index;
|
||
break;
|
||
}
|
||
}
|
||
if (signature == m_data.size())
|
||
{
|
||
error = L"不是 HDF5（MATLAB 7.3）格式。";
|
||
return false;
|
||
}
|
||
const std::uint8_t version = U8(signature + 8);
|
||
if (version == 0 || version == 1)
|
||
{
|
||
m_offsetSize = U8(signature + 13);
|
||
m_lengthSize = U8(signature + 14);
|
||
m_base = U64(signature + 24);
|
||
const std::uint64_t rootEntry = signature + 24 + 32;
|
||
m_root = Resolve(Addr(rootEntry + 8));
|
||
}
|
||
else
|
||
{
|
||
m_offsetSize = U8(signature + 9);
|
||
m_lengthSize = U8(signature + 10);
|
||
m_base = U64(signature + 12);
|
||
m_root = Resolve(U64(signature + 36));
|
||
}
|
||
if (m_root == 0 || m_root >= m_data.size())
|
||
{
|
||
error = L"HDF5 超级块损坏。";
|
||
return false;
|
||
}
|
||
return true;
|
||
}
|
||
|
||
void MatHdf5::Close()
|
||
{
|
||
m_data.clear();
|
||
m_base = 0;
|
||
m_root = 0;
|
||
m_heaps.clear();
|
||
}
|
||
|
||
void MatHdf5::ReadMessages(std::uint64_t address, std::vector<Message>& messages) const
|
||
{
|
||
messages.clear();
|
||
if (address + 8 > m_data.size())
|
||
{
|
||
return;
|
||
}
|
||
if (U8(address) == 1)
|
||
{
|
||
const std::uint16_t count = U16(address + 2);
|
||
const std::uint32_t headerSize = U32(address + 8);
|
||
// 先收集消息块：主块（长度 = 对象头里的 headerSize）+ 若干续块（续块里没有对象头前缀）。
|
||
// ⚠️ **主块长度必须用 headerSize 限定**：消息数超过主块容量时，其余消息在续块里，
|
||
// 若在主块里继续按"消息数"往下读，就会读到块外的其它数据（字段全部解析失败）。
|
||
std::vector<std::pair<std::uint64_t, std::uint64_t>> blocks;
|
||
blocks.emplace_back(address + 16, headerSize);
|
||
std::size_t read = 0;
|
||
for (std::size_t blockIndex = 0; blockIndex < blocks.size(); ++blockIndex)
|
||
{
|
||
std::uint64_t offset = blocks[blockIndex].first;
|
||
const std::uint64_t end =
|
||
(blocks[blockIndex].second > 0) ? (offset + blocks[blockIndex].second) : 0;
|
||
while (read < count && offset + 8 <= m_data.size() && (end == 0 || offset + 8 <= end))
|
||
{
|
||
Message message;
|
||
message.type = U16(offset);
|
||
const std::uint16_t messageSize = U16(offset + 2);
|
||
offset += 8;
|
||
if (offset + messageSize > m_data.size())
|
||
{
|
||
break;
|
||
}
|
||
message.data.assign(m_data.begin() + static_cast<std::ptrdiff_t>(offset),
|
||
m_data.begin() + static_cast<std::ptrdiff_t>(offset + messageSize));
|
||
offset += (messageSize + 7u) & ~static_cast<std::uint16_t>(7u);
|
||
++read;
|
||
if (message.type == kMessageContinuation && message.data.size() >= 16)
|
||
{
|
||
const std::uint64_t blockAddress = Resolve(ReadLE(message.data.data(), 8));
|
||
const std::uint64_t blockSize = ReadLE(message.data.data() + 8, 8);
|
||
if (blockAddress != 0)
|
||
{
|
||
blocks.emplace_back(blockAddress, blockSize);
|
||
}
|
||
}
|
||
messages.push_back(std::move(message));
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
void MatHdf5::WalkBtree(std::uint64_t node, std::uint64_t heapData,
|
||
std::vector<std::pair<std::string, std::uint64_t>>& out) const
|
||
{
|
||
int guard = 0;
|
||
while (node != 0 && node + 24 < m_data.size() && guard++ < 100000)
|
||
{
|
||
if (std::memcmp(&m_data[static_cast<std::size_t>(node)], "TREE", 4) != 0)
|
||
{
|
||
return;
|
||
}
|
||
const std::uint8_t level = U8(node + 5);
|
||
const std::uint16_t entries = U16(node + 6);
|
||
std::uint64_t offset = node + 8 + 8 + 8;
|
||
std::vector<std::uint64_t> children;
|
||
for (std::uint16_t index = 0; index < static_cast<std::uint16_t>(entries + 1u); ++index)
|
||
{
|
||
offset += 8; // key（分组 B-tree 的 key 是堆偏移）
|
||
if (index < entries)
|
||
{
|
||
children.push_back(Resolve(U64(offset)));
|
||
offset += 8;
|
||
}
|
||
}
|
||
for (const std::uint64_t child : children)
|
||
{
|
||
if (level > 0)
|
||
{
|
||
WalkBtree(child, heapData, out);
|
||
}
|
||
else
|
||
{
|
||
WalkSymbolNode(child, heapData, out);
|
||
}
|
||
}
|
||
node = Resolve(U64(node + 8)); // 右兄弟
|
||
}
|
||
}
|
||
|
||
void MatHdf5::WalkSymbolNode(std::uint64_t node, std::uint64_t heapData,
|
||
std::vector<std::pair<std::string, std::uint64_t>>& out) const
|
||
{
|
||
if (node == 0 || node + 8 > m_data.size() ||
|
||
std::memcmp(&m_data[static_cast<std::size_t>(node)], "SNOD", 4) != 0 || heapData == 0)
|
||
{
|
||
return;
|
||
}
|
||
const std::uint16_t count = U16(node + 6);
|
||
std::uint64_t offset = node + 8;
|
||
for (std::uint16_t index = 0; index < count; ++index)
|
||
{
|
||
if (offset + 40 > m_data.size())
|
||
{
|
||
return;
|
||
}
|
||
const std::uint64_t nameOffset = U64(offset);
|
||
const std::uint64_t objectAddress = Resolve(U64(offset + 8));
|
||
offset += 40;
|
||
if (objectAddress == 0 || heapData + nameOffset >= m_data.size())
|
||
{
|
||
continue;
|
||
}
|
||
const char* name = reinterpret_cast<const char*>(m_data.data() + heapData + nameOffset);
|
||
out.emplace_back(std::string(name), objectAddress);
|
||
}
|
||
}
|
||
|
||
std::vector<std::pair<std::string, std::uint64_t>> MatHdf5::Members(std::uint64_t address) const
|
||
{
|
||
std::vector<std::pair<std::string, std::uint64_t>> out;
|
||
std::vector<Message> messages;
|
||
ReadMessages(address, messages);
|
||
for (const Message& message : messages)
|
||
{
|
||
if (message.type == kMessageSymbolTable && message.data.size() >= 16)
|
||
{
|
||
const std::uint64_t btree = Resolve(ReadLE(message.data.data(), 8));
|
||
const std::uint64_t heap = Resolve(ReadLE(message.data.data() + 8, 8));
|
||
std::uint64_t heapData = 0;
|
||
if (heap + 32 <= m_data.size() &&
|
||
std::memcmp(&m_data[static_cast<std::size_t>(heap)], "HEAP", 4) == 0)
|
||
{
|
||
heapData = Resolve(U64(heap + 24));
|
||
bool known = false;
|
||
for (const auto& entry : m_heaps)
|
||
{
|
||
if (entry.first == heap)
|
||
{
|
||
known = true;
|
||
break;
|
||
}
|
||
}
|
||
if (!known)
|
||
{
|
||
m_heaps.emplace_back(heap, heapData);
|
||
}
|
||
}
|
||
WalkBtree(btree, heapData, out);
|
||
}
|
||
else if (message.type == kMessageLink && message.data.size() >= 8)
|
||
{
|
||
std::size_t offset = 2;
|
||
const std::uint8_t flags = message.data[1];
|
||
if (flags & 0x08)
|
||
{
|
||
offset += 8;
|
||
}
|
||
if (flags & 0x04)
|
||
{
|
||
offset += 8;
|
||
}
|
||
if (flags & 0x10)
|
||
{
|
||
++offset;
|
||
}
|
||
if (offset >= message.data.size())
|
||
{
|
||
continue;
|
||
}
|
||
const std::uint8_t lengthSize = message.data[offset++];
|
||
std::uint64_t nameLength = 0;
|
||
for (std::uint8_t byte = 0; byte < lengthSize && offset < message.data.size(); ++byte)
|
||
{
|
||
nameLength |= static_cast<std::uint64_t>(message.data[offset + byte]) << (8 * byte);
|
||
}
|
||
offset += lengthSize;
|
||
if (offset + nameLength + 8 > message.data.size())
|
||
{
|
||
continue;
|
||
}
|
||
std::string name(reinterpret_cast<const char*>(message.data.data() + offset),
|
||
static_cast<std::size_t>(nameLength));
|
||
offset += static_cast<std::size_t>(nameLength);
|
||
const std::uint64_t target = Resolve(ReadLE(message.data.data() + offset, 8));
|
||
out.emplace_back(std::move(name), target);
|
||
}
|
||
}
|
||
return out;
|
||
}
|
||
|
||
std::uint64_t MatHdf5::Member(std::uint64_t address, const std::string& name) const
|
||
{
|
||
for (const auto& entry : Members(address))
|
||
{
|
||
if (entry.first == name)
|
||
{
|
||
return entry.second;
|
||
}
|
||
}
|
||
return 0;
|
||
}
|
||
|
||
bool MatHdf5::IsDataset(std::uint64_t address) const
|
||
{
|
||
std::vector<Message> messages;
|
||
ReadMessages(address, messages);
|
||
for (const Message& message : messages)
|
||
{
|
||
if (message.type == kMessageDataspace)
|
||
{
|
||
return true;
|
||
}
|
||
}
|
||
return false;
|
||
}
|
||
|
||
bool MatHdf5::ParseType(const std::uint8_t* data, std::size_t size, std::size_t offset, Type& out,
|
||
std::size_t& consumed) const
|
||
{
|
||
if (offset + 8 > size)
|
||
{
|
||
return false;
|
||
}
|
||
out.version = data[offset] >> 4;
|
||
out.cls = static_cast<TypeClass>(data[offset] & 0x0F);
|
||
const std::uint8_t bits0 = data[offset + 1];
|
||
out.size = ReadLE(data + offset + 4, 4);
|
||
std::size_t cursor = offset + 8;
|
||
if (out.cls == TypeClass::Integer)
|
||
{
|
||
out.bigEndian = (bits0 & 0x01) != 0;
|
||
out.isSigned = (bits0 & 0x08) != 0;
|
||
if (out.version >= 3)
|
||
{
|
||
cursor += 2; // bit offset
|
||
cursor += 2; // precision
|
||
}
|
||
else
|
||
{
|
||
cursor += 4;
|
||
}
|
||
}
|
||
else if (out.cls == TypeClass::Float)
|
||
{
|
||
out.bigEndian = (bits0 & 0x01) != 0;
|
||
cursor += 12;
|
||
}
|
||
else if (out.cls == TypeClass::String)
|
||
{
|
||
// 无需额外属性
|
||
}
|
||
else if (out.cls == TypeClass::Reference)
|
||
{
|
||
// 无需额外属性
|
||
}
|
||
else if (out.cls == TypeClass::VariableLength)
|
||
{
|
||
out.isVariableString = (bits0 & 0x0F) == 1;
|
||
std::size_t childConsumed = 0;
|
||
Type child;
|
||
if (ParseType(data, size, cursor, child, childConsumed))
|
||
{
|
||
cursor += childConsumed;
|
||
}
|
||
}
|
||
else if (out.cls == TypeClass::Compound)
|
||
{
|
||
const unsigned members = static_cast<unsigned>(bits0) | (static_cast<unsigned>(data[offset + 2]) << 8);
|
||
for (unsigned index = 0; index < members; ++index)
|
||
{
|
||
if (cursor >= size)
|
||
{
|
||
break;
|
||
}
|
||
const std::size_t nameStart = cursor;
|
||
while (cursor < size && data[cursor] != 0)
|
||
{
|
||
++cursor;
|
||
}
|
||
std::string name(reinterpret_cast<const char*>(data + nameStart), cursor - nameStart);
|
||
++cursor; // 结束符
|
||
cursor = (cursor + 7) & ~static_cast<std::size_t>(7);
|
||
if (out.version == 3)
|
||
{
|
||
// v3：先维度（1 字节 rank + 4 字节保留），再字节偏移（可变）
|
||
}
|
||
cursor += 4; // 字节偏移（v1/v2）
|
||
std::size_t childConsumed = 0;
|
||
Type child;
|
||
if (ParseType(data, size, cursor, child, childConsumed))
|
||
{
|
||
cursor += childConsumed;
|
||
out.members.emplace_back(std::move(name), std::move(child));
|
||
}
|
||
else
|
||
{
|
||
break;
|
||
}
|
||
}
|
||
}
|
||
else if (out.cls == TypeClass::Array)
|
||
{
|
||
if (cursor + 4 <= size)
|
||
{
|
||
const std::uint8_t rank = data[cursor];
|
||
cursor += 4;
|
||
cursor += static_cast<std::size_t>(rank) * 4; // 维度
|
||
cursor += static_cast<std::size_t>(rank) * 4; // 排列
|
||
std::size_t childConsumed = 0;
|
||
Type child;
|
||
if (ParseType(data, size, cursor, child, childConsumed))
|
||
{
|
||
cursor += childConsumed;
|
||
}
|
||
}
|
||
}
|
||
consumed = cursor - offset;
|
||
return true;
|
||
}
|
||
|
||
bool MatHdf5::DecodeChunked(const Dataset& dataset, std::size_t elementSize,
|
||
std::vector<std::uint8_t>& out) const
|
||
{
|
||
if (dataset.chunkAddress == 0 || dataset.chunkDims.empty() || elementSize == 0 ||
|
||
dataset.dims.empty())
|
||
{
|
||
return false;
|
||
}
|
||
// ⚠️ MATLAB 写出的分块维度可能与数据空间维度**维数不一致**（实测：1 维数据集的
|
||
// 分块维度是 `[N, 1]`）。把维度数组在**右侧补 1**（行优先的最快轴）对齐即可，
|
||
// 二者的元素总数与遍历顺序都不变。
|
||
std::vector<std::size_t> dims = dataset.dims;
|
||
std::vector<std::size_t> chunkDims = dataset.chunkDims;
|
||
while (dims.size() < chunkDims.size())
|
||
{
|
||
dims.push_back(1);
|
||
}
|
||
while (chunkDims.size() < dims.size())
|
||
{
|
||
chunkDims.push_back(1);
|
||
}
|
||
const std::size_t rank = dims.size();
|
||
const std::uint64_t total = [&]() -> std::uint64_t
|
||
{
|
||
std::uint64_t value = 1;
|
||
for (const std::size_t dim : dims)
|
||
{
|
||
value *= dim;
|
||
}
|
||
return value;
|
||
}();
|
||
out.assign(static_cast<std::size_t>(total) * elementSize, 0);
|
||
std::vector<std::uint8_t> chunk;
|
||
std::vector<std::uint8_t> inflated;
|
||
// 遍历分块 B-tree：key = 分块大小 + 过滤器掩码 + 各维偏移。
|
||
std::vector<std::uint64_t> stack{dataset.chunkAddress};
|
||
int guard = 0;
|
||
while (!stack.empty() && guard++ < 1000000)
|
||
{
|
||
const std::uint64_t node = stack.back();
|
||
stack.pop_back();
|
||
if (node == 0 || node + 24 > m_data.size() ||
|
||
std::memcmp(&m_data[static_cast<std::size_t>(node)], "TREE", 4) != 0)
|
||
{
|
||
continue;
|
||
}
|
||
const std::uint8_t level = U8(node + 5);
|
||
const std::uint16_t entries = U16(node + 6);
|
||
const std::size_t keySize = 8 + 8 * rank;
|
||
std::uint64_t offset = node + 8 + 8 + 8;
|
||
for (std::uint16_t index = 0; index < static_cast<std::uint16_t>(entries + 1u); ++index)
|
||
{
|
||
const std::uint32_t chunkSize = (offset + 4 <= m_data.size()) ? U32(offset) : 0;
|
||
std::vector<std::uint64_t> chunkOffset(rank, 0);
|
||
for (std::size_t axis = 0; axis < rank; ++axis)
|
||
{
|
||
chunkOffset[axis] = U64(offset + 8 + axis * 8);
|
||
}
|
||
offset += keySize;
|
||
if (index < entries)
|
||
{
|
||
const std::uint64_t child = Resolve(U64(offset));
|
||
offset += 8;
|
||
if (level > 0)
|
||
{
|
||
stack.push_back(child);
|
||
}
|
||
else if (child != 0 && chunkSize > 0)
|
||
{
|
||
chunk.assign(m_data.begin() + static_cast<std::ptrdiff_t>(child),
|
||
m_data.begin() + static_cast<std::ptrdiff_t>(child + chunkSize));
|
||
const std::uint8_t* source = chunk.data();
|
||
std::size_t sourceSize = chunk.size();
|
||
if (dataset.deflate)
|
||
{
|
||
// 分块未压缩时大小 = 分块元素数 × 元素大小：此时不需要解压。
|
||
std::size_t expected = elementSize;
|
||
for (const std::size_t dim : dataset.chunkDims)
|
||
{
|
||
expected *= dim;
|
||
}
|
||
if (chunk.size() != expected)
|
||
{
|
||
inflated.assign(expected, 0);
|
||
uLongf destLength = static_cast<uLongf>(expected);
|
||
if (uncompress(inflated.data(), &destLength, chunk.data(),
|
||
static_cast<uLong>(chunk.size())) != Z_OK)
|
||
{
|
||
continue;
|
||
}
|
||
source = inflated.data();
|
||
sourceSize = expected;
|
||
}
|
||
}
|
||
// 把该块拷进输出缓冲区（按行优先展开）。
|
||
const std::size_t rowBytes = chunkDims[rank - 1] * elementSize;
|
||
std::size_t rows = 1;
|
||
for (std::size_t axis = 0; axis + 1 < rank; ++axis)
|
||
{
|
||
rows *= chunkDims[axis];
|
||
}
|
||
for (std::size_t row = 0; row < rows; ++row)
|
||
{
|
||
std::vector<std::size_t> coordinate(rank - 1, 0);
|
||
std::size_t remainder = row;
|
||
for (std::size_t axis = rank - 1; axis-- > 0;)
|
||
{
|
||
coordinate[axis] = remainder % chunkDims[axis];
|
||
remainder /= chunkDims[axis];
|
||
}
|
||
std::uint64_t targetRow = 0;
|
||
bool inside = true;
|
||
std::uint64_t stride = 1;
|
||
for (std::size_t axis = rank - 1; axis-- > 0;)
|
||
{
|
||
const std::uint64_t index = chunkOffset[axis] + coordinate[axis];
|
||
if (index >= dims[axis])
|
||
{
|
||
inside = false;
|
||
break;
|
||
}
|
||
targetRow += index * stride;
|
||
stride *= dims[axis];
|
||
}
|
||
if (!inside)
|
||
{
|
||
continue;
|
||
}
|
||
const std::size_t sourceOffset = row * rowBytes;
|
||
const std::size_t targetOffset =
|
||
static_cast<std::size_t>(targetRow) * dims[rank - 1] * elementSize;
|
||
if (sourceOffset + rowBytes > sourceSize ||
|
||
targetOffset + rowBytes > out.size())
|
||
{
|
||
continue;
|
||
}
|
||
std::memcpy(out.data() + targetOffset, source + sourceOffset, rowBytes);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
}
|
||
return true;
|
||
}
|
||
|
||
bool MatHdf5::ReadDataset(std::uint64_t address, Dataset& out) const
|
||
{
|
||
out = Dataset();
|
||
std::vector<Message> messages;
|
||
ReadMessages(address, messages);
|
||
bool haveSpace = false;
|
||
bool haveType = false;
|
||
std::size_t compactSize = 0;
|
||
std::uint64_t dataAddress = 0;
|
||
std::uint64_t dataSize = 0;
|
||
for (const Message& message : messages)
|
||
{
|
||
if (message.type == kMessageDataspace && message.data.size() >= 8)
|
||
{
|
||
const std::uint8_t version = message.data[0];
|
||
const std::uint8_t rank = message.data[1];
|
||
std::size_t offset = (version == 1) ? 8 : 4;
|
||
out.dims.assign(rank, 0);
|
||
for (std::uint8_t axis = 0; axis < rank && offset + 8 <= message.data.size(); ++axis)
|
||
{
|
||
out.dims[axis] = static_cast<std::size_t>(ReadLE(&message.data[offset], 8));
|
||
offset += 8;
|
||
}
|
||
haveSpace = true;
|
||
}
|
||
else if (message.type == kMessageDatatype)
|
||
{
|
||
std::size_t consumed = 0;
|
||
haveType = ParseType(message.data.data(), message.data.size(), 0, out.type, consumed);
|
||
}
|
||
else if (message.type == kMessageLayout && message.data.size() >= 8)
|
||
{
|
||
const std::uint8_t version = message.data[0];
|
||
const std::uint8_t klass = message.data[1];
|
||
out.layout = static_cast<LayoutClass>(klass);
|
||
if (klass == 0)
|
||
{
|
||
// 紧凑：数据就放在消息里
|
||
if (version == 3)
|
||
{
|
||
compactSize = static_cast<std::size_t>(message.data[2]) |
|
||
(static_cast<std::size_t>(message.data[3]) << 8);
|
||
}
|
||
else
|
||
{
|
||
compactSize = static_cast<std::size_t>(ReadLE(&message.data[2], 4));
|
||
}
|
||
}
|
||
else if (klass == 1)
|
||
{
|
||
dataAddress = Resolve(ReadLE(&message.data[2], 8));
|
||
dataSize = ReadLE(&message.data[10], 8);
|
||
}
|
||
else if (klass == 2)
|
||
{
|
||
if (version == 3)
|
||
{
|
||
const std::uint8_t rank = message.data[2];
|
||
out.chunkAddress = Resolve(ReadLE(&message.data[3], 8));
|
||
out.chunkDims.assign(rank, 0);
|
||
for (std::uint8_t axis = 0; axis < rank; ++axis)
|
||
{
|
||
out.chunkDims[axis] = ReadLE(&message.data[11 + axis * 4], 4);
|
||
}
|
||
}
|
||
else
|
||
{
|
||
const std::uint8_t rank = message.data[2];
|
||
out.chunkAddress = Resolve(ReadLE(&message.data[3], 8));
|
||
out.chunkDims.assign(rank, 0);
|
||
for (std::uint8_t axis = 0; axis < rank; ++axis)
|
||
{
|
||
out.chunkDims[axis] = ReadLE(&message.data[11 + axis * 4], 4);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
else if (message.type == kMessageFilter && message.data.size() >= 2)
|
||
{
|
||
const int count = message.data[1];
|
||
std::size_t offset = 8;
|
||
for (int index = 0; index < count && offset + 8 <= message.data.size(); ++index)
|
||
{
|
||
const std::uint16_t id = static_cast<std::uint16_t>(message.data[offset]) |
|
||
(static_cast<std::uint16_t>(message.data[offset + 1]) << 8);
|
||
const std::uint16_t nameLength = static_cast<std::uint16_t>(message.data[offset + 2]) |
|
||
(static_cast<std::uint16_t>(message.data[offset + 3]) << 8);
|
||
const std::uint16_t values =
|
||
static_cast<std::uint16_t>(message.data[offset + 6]) |
|
||
(static_cast<std::uint16_t>(message.data[offset + 7]) << 8);
|
||
offset += 8;
|
||
offset += (nameLength + 7) & ~static_cast<std::size_t>(7);
|
||
offset += static_cast<std::size_t>(values) * 4;
|
||
if (id == 1)
|
||
{
|
||
out.deflate = true;
|
||
}
|
||
}
|
||
}
|
||
}
|
||
if (!haveSpace || !haveType || out.type.size == 0)
|
||
{
|
||
return false;
|
||
}
|
||
const std::size_t elementSize = out.type.size;
|
||
if (out.layout == LayoutClass::Compact)
|
||
{
|
||
// 紧凑数据在消息里，重新扫描一次取出字节。
|
||
for (const Message& message : messages)
|
||
{
|
||
if (message.type == kMessageLayout)
|
||
{
|
||
const std::size_t offset = (message.data[0] == 3) ? 4 : 8;
|
||
if (offset + compactSize <= message.data.size())
|
||
{
|
||
out.data.assign(message.data.begin() + static_cast<std::ptrdiff_t>(offset),
|
||
message.data.begin() +
|
||
static_cast<std::ptrdiff_t>(offset + compactSize));
|
||
}
|
||
}
|
||
}
|
||
}
|
||
else if (out.layout == LayoutClass::Contiguous)
|
||
{
|
||
if (dataAddress != 0 && dataAddress + dataSize <= m_data.size())
|
||
{
|
||
out.data.assign(m_data.begin() + static_cast<std::ptrdiff_t>(dataAddress),
|
||
m_data.begin() + static_cast<std::ptrdiff_t>(dataAddress + dataSize));
|
||
}
|
||
}
|
||
else if (out.layout == LayoutClass::Chunked)
|
||
{
|
||
if (!DecodeChunked(out, elementSize, out.data))
|
||
{
|
||
return false;
|
||
}
|
||
}
|
||
out.ok = true;
|
||
return true;
|
||
}
|
||
|
||
std::string MatHdf5::AttributeString(std::uint64_t address, const char* name) const
|
||
{
|
||
std::vector<Message> messages;
|
||
ReadMessages(address, messages);
|
||
for (const Message& message : messages)
|
||
{
|
||
if (message.type != kMessageAttribute || message.data.size() < 8)
|
||
{
|
||
continue;
|
||
}
|
||
const std::uint8_t version = message.data[0];
|
||
const std::size_t headerSize = (version == 1) ? 8 : (version == 2 ? 8 : 9);
|
||
const std::uint16_t nameSize = static_cast<std::uint16_t>(message.data[2]) |
|
||
(static_cast<std::uint16_t>(message.data[3]) << 8);
|
||
const std::uint16_t typeSize = static_cast<std::uint16_t>(message.data[4]) |
|
||
(static_cast<std::uint16_t>(message.data[5]) << 8);
|
||
const std::uint16_t spaceSize = static_cast<std::uint16_t>(message.data[6]) |
|
||
(static_cast<std::uint16_t>(message.data[7]) << 8);
|
||
std::size_t offset = headerSize;
|
||
if (offset + nameSize > message.data.size())
|
||
{
|
||
continue;
|
||
}
|
||
std::string attributeName(reinterpret_cast<const char*>(message.data.data() + offset),
|
||
nameSize);
|
||
while (!attributeName.empty() && attributeName.back() == '\0')
|
||
{
|
||
attributeName.pop_back();
|
||
}
|
||
offset += (nameSize + 7) & ~static_cast<std::size_t>(7);
|
||
const std::size_t typeOffset = offset;
|
||
offset += (typeSize + 7) & ~static_cast<std::size_t>(7);
|
||
const std::size_t spaceOffset = offset;
|
||
offset += (spaceSize + 7) & ~static_cast<std::size_t>(7);
|
||
if (attributeName != name)
|
||
{
|
||
continue;
|
||
}
|
||
if (offset > message.data.size())
|
||
{
|
||
return std::string();
|
||
}
|
||
Type type;
|
||
std::size_t consumed = 0;
|
||
if (!ParseType(message.data.data(), message.data.size(), typeOffset, type, consumed))
|
||
{
|
||
return std::string();
|
||
}
|
||
std::size_t count = 1;
|
||
if (spaceOffset + 4 <= message.data.size())
|
||
{
|
||
const std::uint8_t spaceVersion = message.data[spaceOffset];
|
||
const std::uint8_t rank = message.data[spaceOffset + 1];
|
||
std::size_t dimOffset = spaceOffset + ((spaceVersion == 1) ? 8 : 4);
|
||
count = 1;
|
||
for (std::uint8_t axis = 0; axis < rank && dimOffset + 8 <= message.data.size(); ++axis)
|
||
{
|
||
count *= static_cast<std::size_t>(ReadLE(&message.data[dimOffset], 8));
|
||
dimOffset += 8;
|
||
}
|
||
}
|
||
std::string text;
|
||
if (type.cls == TypeClass::String || type.cls == TypeClass::Integer)
|
||
{
|
||
text.assign(reinterpret_cast<const char*>(message.data.data() + offset),
|
||
std::min(message.data.size() - offset, std::max<std::size_t>(count, 1)));
|
||
while (!text.empty() && text.back() == '\0')
|
||
{
|
||
text.pop_back();
|
||
}
|
||
}
|
||
return text;
|
||
}
|
||
return std::string();
|
||
}
|
||
|
||
bool MatHdf5::ReadText(std::uint64_t address, std::wstring& out) const
|
||
{
|
||
out.clear();
|
||
Dataset dataset;
|
||
if (!ReadDataset(address, dataset))
|
||
{
|
||
return false;
|
||
}
|
||
if (dataset.type.cls == TypeClass::String)
|
||
{
|
||
out.assign(dataset.data.begin(), dataset.data.end());
|
||
while (!out.empty() && out.back() == L'\0')
|
||
{
|
||
out.pop_back();
|
||
}
|
||
return true;
|
||
}
|
||
if (dataset.type.size == 2)
|
||
{
|
||
for (std::size_t index = 0; index + 1 < dataset.data.size(); index += 2)
|
||
{
|
||
const wchar_t character = static_cast<wchar_t>(
|
||
static_cast<std::uint16_t>(dataset.data[index]) |
|
||
(static_cast<std::uint16_t>(dataset.data[index + 1]) << 8));
|
||
if (character != 0)
|
||
{
|
||
out.push_back(character);
|
||
}
|
||
}
|
||
return true;
|
||
}
|
||
if (dataset.type.size == 1)
|
||
{
|
||
for (const std::uint8_t byte : dataset.data)
|
||
{
|
||
if (byte != 0)
|
||
{
|
||
out.push_back(static_cast<wchar_t>(byte));
|
||
}
|
||
}
|
||
return true;
|
||
}
|
||
return false;
|
||
}
|
||
|
||
bool MatHdf5::ReadNumbers(std::uint64_t address, std::vector<double>& out) const
|
||
{
|
||
out.clear();
|
||
Dataset dataset;
|
||
if (!ReadDataset(address, dataset) || !dataset.ok)
|
||
{
|
||
return false;
|
||
}
|
||
const std::size_t elementSize = dataset.type.size;
|
||
if (elementSize == 0 || dataset.data.size() % elementSize != 0)
|
||
{
|
||
return false;
|
||
}
|
||
const std::size_t count = dataset.data.size() / elementSize;
|
||
out.resize(count);
|
||
const bool integer = (dataset.type.cls == TypeClass::Integer);
|
||
const bool floating = (dataset.type.cls == TypeClass::Float);
|
||
if (!integer && !floating)
|
||
{
|
||
return false;
|
||
}
|
||
for (std::size_t index = 0; index < count; ++index)
|
||
{
|
||
const std::uint8_t* source = dataset.data.data() + index * elementSize;
|
||
if (floating && elementSize == 8)
|
||
{
|
||
double value = 0.0;
|
||
std::memcpy(&value, source, sizeof(value));
|
||
out[index] = value;
|
||
}
|
||
else if (floating && elementSize == 4)
|
||
{
|
||
float value = 0.0f;
|
||
std::memcpy(&value, source, sizeof(value));
|
||
out[index] = static_cast<double>(value);
|
||
}
|
||
else if (integer && dataset.type.isSigned)
|
||
{
|
||
std::int64_t value = 0;
|
||
for (std::size_t byte = 0; byte < elementSize; ++byte)
|
||
{
|
||
value |= static_cast<std::int64_t>(source[byte]) << (8 * byte);
|
||
}
|
||
// 符号扩展
|
||
const int bits = static_cast<int>(elementSize) * 8;
|
||
if (bits < 64 && (value & (1ll << (bits - 1))) != 0)
|
||
{
|
||
value |= ~((1ll << bits) - 1);
|
||
}
|
||
out[index] = static_cast<double>(value);
|
||
}
|
||
else if (integer)
|
||
{
|
||
std::uint64_t value = 0;
|
||
for (std::size_t byte = 0; byte < elementSize; ++byte)
|
||
{
|
||
value |= static_cast<std::uint64_t>(source[byte]) << (8 * byte);
|
||
}
|
||
out[index] = static_cast<double>(value);
|
||
}
|
||
}
|
||
return true;
|
||
}
|
||
|
||
bool MatHdf5::ReadReferences(std::uint64_t address, std::vector<std::uint64_t>& out) const
|
||
{
|
||
out.clear();
|
||
Dataset dataset; if (!ReadDataset(address, dataset) || !dataset.ok ||
|
||
dataset.type.cls != TypeClass::Reference)
|
||
{
|
||
return false;
|
||
}
|
||
const std::size_t count = dataset.data.size() / 8;
|
||
out.reserve(count);
|
||
for (std::size_t index = 0; index < count; ++index)
|
||
{
|
||
const std::uint64_t raw = ReadLE(dataset.data.data() + index * 8, 8);
|
||
out.push_back(Resolve(raw));
|
||
}
|
||
return true;
|
||
}
|
||
|
||
} // namespace jdi
|