/**
 * @file Image.cpp
 * @brief vm 自带的 .stb 只读视图实现（执行器侧；compiler 各有实现，V2）
 * @author 
 * @date 2026-08-21
 */

#include "vm/Image.h"

#include <cstring>
#include <string>

namespace vm {

namespace {

    /// 映像魔数："STSC" 小端。
    const uint32_t kMagic = 0x43545353u;
    /// 映像格式版本（V2）。
    const uint32_t kVersion = 2;
    /// 映像头字节数（V2：128）。
    const size_t kHeaderSize = 128;
    /// 常量表一行字节数（value:8，无 tag）。
    const size_t kConstEntrySize = 8;
    /// 函数表一行字节数（nregs/code_offset/code_len 各 4）。
    const size_t kFuncRowSize = 12;
    /// 槽表一行字节数（addr:4 + 预留:4）。
    const size_t kSlotRowSize = 8;
    /// SHA-256 摘要长度（文件尾）。
    const size_t kSha256Size = 32;
    /// 型号标识长度（头内 @72，不足补 '\0'）。
    const size_t kModelIdSize = 32;

    // ---- 头字段偏移（与 Doc/isa/指令与映像.md §8.1 一致；compiler 侧另有同值一份）----
    constexpr size_t kOffMagic = 0;         ///< 魔数 "STSC"
    constexpr size_t kOffVersion = 4;       ///< 格式版本（V2 = 2）
    constexpr size_t kOffCycleLimit = 8;    ///< 每周期指令数上限
    constexpr size_t kOffDtMs = 12;         ///< 周期时长（毫秒）
    constexpr size_t kOffProjectHash = 16;  ///< 工程哈希（8 字节）
    constexpr size_t kOffEntryFnId = 24;    ///< 入口函数 fn_id
    constexpr size_t kOffNGlobals = 28;     ///< 全局槽数
    constexpr size_t kOffNI = 32;           ///< 输入（I）槽数（保留恒 0）
    constexpr size_t kOffNQ = 36;           ///< 输出（Q）槽数（保留恒 0）
    constexpr size_t kOffNM = 40;           ///< 中间（M）槽数（保留恒 0）
    constexpr size_t kOffNConsts = 44;      ///< 常量表条目数
    constexpr size_t kOffNFuncs = 48;       ///< 函数表行数
    constexpr size_t kOffConst = 52;        ///< 常量表段偏移
    constexpr size_t kOffFuncs = 56;        ///< 函数表段偏移
    constexpr size_t kOffCode = 60;         ///< 字节码段偏移
    constexpr size_t kOffSlots = 64;        ///< 槽表段偏移
    constexpr size_t kOffValues = 68;       ///< 值段偏移
    constexpr size_t kOffModelId = 72;      ///< 型号标识[32]
    constexpr size_t kOffNSlots = 104;      ///< 槽表条目数
    constexpr size_t kOffValuesSize = 108;  ///< 值段字节数
    constexpr size_t kOffMeta = 112;        ///< 元数据段偏移（0 = 空段）
    constexpr size_t kOffMaxStack = 116;    ///< 栈区字节数

    // ---- 段内行偏移 ----
    constexpr size_t kConstValueOff = 0;  ///< 常量表行内：value（8 字节）
    constexpr size_t kFuncNregsOff = 0;   ///< 函数表行内：nregs
    constexpr size_t kFuncCodeOff = 4;    ///< 函数表行内：code_offset（字节）
    constexpr size_t kFuncLenOff = 8;     ///< 函数表行内：code_len（字节）
    constexpr size_t kSlotAddrOff = 0;    ///< 槽表行内：addr

    uint32_t get_le32(const uint8_t* p) {
        return static_cast<uint32_t>(p[0])
            | (static_cast<uint32_t>(p[1]) << 8)
            | (static_cast<uint32_t>(p[2]) << 16)
            | (static_cast<uint32_t>(p[3]) << 24);
    }

    uint64_t get_le64(const uint8_t* p) {
        uint64_t v = 0;
        for (int i = 0; i < 8; ++i) {
            v |= static_cast<uint64_t>(p[i]) << (8 * i);
        }
        return v;
    }

    // ---- SHA-256（执行器侧实现，与 compiler 各一份）----

    const uint32_t kShaK[64] = {
        0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1,
        0x923f82a4, 0xab1c5ed5, 0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3,
        0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174, 0xe49b69c1, 0xefbe4786,
        0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
        0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147,
        0x06ca6351, 0x14292967, 0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13,
        0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85, 0xa2bfe8a1, 0xa81a664b,
        0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
        0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a,
        0x5b9cca4f, 0x682e6ff3, 0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208,
        0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2,
    };

    inline uint32_t rotr(uint32_t x, uint32_t n) { return (x >> n) | (x << (32 - n)); }

    struct Sha256 {
        uint32_t h[8] = {0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a,
                         0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19};  ///< 初始哈希值
        uint64_t total = 0;   ///< 已吸收字节数（final 时编码进长度域）
        uint8_t block[64];    ///< 当前块缓冲
        size_t block_len = 0; ///< 块缓冲已用字节数

        void update(const uint8_t* data, size_t len) {
            total += len;
            while (len > 0) {
                const size_t n = (block_len < 64) ? (64 - block_len) : 0;
                const size_t take = len < n ? len : n;
                if (take == 0) {
                    break;
                }
                for (size_t i = 0; i < take; ++i) {
                    block[block_len + i] = data[i];
                }
                block_len += take;
                data += take;
                len -= take;
                if (block_len == 64) {
                    process();
                    block_len = 0;
                }
            }
        }

        void process() {
            uint32_t w[64];
            for (int i = 0; i < 16; ++i) {
                w[i] = get_be32(block + i * 4);
            }
            for (int i = 16; i < 64; ++i) {
                const uint32_t s0 = rotr(w[i - 15], 7) ^ rotr(w[i - 15], 18) ^ (w[i - 15] >> 3);
                const uint32_t s1 = rotr(w[i - 2], 17) ^ rotr(w[i - 2], 19) ^ (w[i - 2] >> 10);
                w[i] = w[i - 16] + s0 + w[i - 7] + s1;
            }
            uint32_t a = h[0], b = h[1], c = h[2], d = h[3];
            uint32_t e = h[4], f = h[5], g = h[6], hh = h[7];
            for (int i = 0; i < 64; ++i) {
                const uint32_t s1 = rotr(e, 6) ^ rotr(e, 11) ^ rotr(e, 25);
                const uint32_t ch = (e & f) ^ (~e & g);
                const uint32_t t1 = hh + s1 + ch + kShaK[i] + w[i];
                const uint32_t s0 = rotr(a, 2) ^ rotr(a, 13) ^ rotr(a, 22);
                const uint32_t maj = (a & b) ^ (a & c) ^ (b & c);
                const uint32_t t2 = s0 + maj;
                hh = g;
                g = f;
                f = e;
                e = d + t1;
                d = c;
                c = b;
                b = a;
                a = t1 + t2;
            }
            h[0] += a; h[1] += b; h[2] += c; h[3] += d;
            h[4] += e; h[5] += f; h[6] += g; h[7] += hh;
        }

        void final(uint8_t out[32]) {
            const uint64_t bitlen = total * 8;
            const uint8_t pad = 0x80;
            update(&pad, 1);
            const uint8_t zeros[64] = {0};
            while (block_len != 56) {
                const size_t n = (block_len < 56) ? (56 - block_len) : (64 - block_len);
                update(zeros, n);
            }
            for (int i = 0; i < 8; ++i) {
                const uint8_t b2[1] = {static_cast<uint8_t>((bitlen >> (56 - 8 * i)) & 0xFF)};
                update(b2, 1);
            }
            for (int i = 0; i < 8; ++i) {
                out[i * 4 + 0] = static_cast<uint8_t>((h[i] >> 24) & 0xFF);
                out[i * 4 + 1] = static_cast<uint8_t>((h[i] >> 16) & 0xFF);
                out[i * 4 + 2] = static_cast<uint8_t>((h[i] >> 8) & 0xFF);
                out[i * 4 + 3] = static_cast<uint8_t>(h[i] & 0xFF);
            }
        }

        static uint32_t get_be32(const uint8_t* p) {
            return (static_cast<uint32_t>(p[0]) << 24) | (static_cast<uint32_t>(p[1]) << 16) |
                   (static_cast<uint32_t>(p[2]) << 8) | static_cast<uint32_t>(p[3]);
        }
    };

    void sha256(const uint8_t* data, size_t len, uint8_t out[32]) {
        Sha256 s;
        s.update(data, len);
        s.final(out);
    }

    void fill_model_id(const std::string& name, uint32_t version, char out[32]) {
        const std::string id = name + std::to_string(version);
        for (size_t i = 0; i < 32; ++i) {
            out[i] = i < id.size() ? id[i] : '\0';
        }
    }

}  // namespace

/**
 * @brief 从原始字节构造只读视图（不拷贝，调用方保证生命周期）。
 */
Image Image::from(const uint8_t* buf, size_t len) {
    Image v;
    v.buf_ = buf;
    v.len_ = len;
    if (buf == nullptr) {
        v.err_ = "null buffer";
        return v;
    }
    if (len < kHeaderSize) {
        v.err_ = "image too short";
        return v;
    }
    if (get_le32(buf + kOffMagic) != kMagic) {
        v.err_ = "bad magic";
        return v;
    }
    if (get_le32(buf + kOffVersion) != kVersion) {
        v.err_ = "bad version";
        return v;
    }
    ImageHeader& h = v.hdr_;
    h.cycle_limit = get_le32(buf + kOffCycleLimit);
    h.dt_ms = get_le32(buf + kOffDtMs);
    h.project_hash = get_le64(buf + kOffProjectHash);
    h.entry_fn_id = get_le32(buf + kOffEntryFnId);
    h.n_globals = get_le32(buf + kOffNGlobals);
    h.n_i = get_le32(buf + kOffNI);
    h.n_q = get_le32(buf + kOffNQ);
    h.n_m = get_le32(buf + kOffNM);
    h.n_consts = get_le32(buf + kOffNConsts);
    h.n_funcs = get_le32(buf + kOffNFuncs);
    h.offset_const = get_le32(buf + kOffConst);
    h.offset_funcs = get_le32(buf + kOffFuncs);
    h.offset_code = get_le32(buf + kOffCode);
    h.offset_slots = get_le32(buf + kOffSlots);
    h.offset_values = get_le32(buf + kOffValues);
    h.n_slots = get_le32(buf + kOffNSlots);
    h.values_size = get_le32(buf + kOffValuesSize);
    h.offset_meta = get_le32(buf + kOffMeta);
    h.max_stack = get_le32(buf + kOffMaxStack);

    // 段校验（V2：值段之后是 SHA-256 文件尾；元数据段 0 = 空段）
    if (len < static_cast<size_t>(h.offset_values) + kSha256Size) {
        v.err_ = "missing sha256 tail";
        return v;
    }
    const uint64_t offs[5] = {h.offset_const, h.offset_funcs, h.offset_code, h.offset_slots,
                              h.offset_values};
    for (int i = 0; i < 5; ++i) {
        if (offs[i] < kHeaderSize || offs[i] > len - kSha256Size) {
            v.err_ = "segment offset out of range";
            return v;
        }
        if (i > 0 && offs[i] < offs[i - 1]) {
            v.err_ = "segment offsets not monotonic";
            return v;
        }
    }
    if (offs[1] - offs[0] != static_cast<uint64_t>(h.n_consts) * kConstEntrySize) {
        v.err_ = "const table size mismatch";
        return v;
    }
    if (offs[2] - offs[1] != static_cast<uint64_t>(h.n_funcs) * kFuncRowSize) {
        v.err_ = "function table size mismatch";
        return v;
    }
    if (offs[4] - offs[3] != static_cast<uint64_t>(h.n_slots) * kSlotRowSize) {
        v.err_ = "slot table size mismatch";
        return v;
    }
    // 值段大小：meta 段为 0（空段）时值段到 SHA 尾；否则到 meta 起点
    if (h.offset_meta != 0) {
        if (static_cast<uint64_t>(h.offset_meta) - offs[4] != h.values_size) {
            v.err_ = "values segment size mismatch";
            return v;
        }
    } else if (static_cast<uint64_t>(len - kSha256Size) - offs[4] != h.values_size) {
        v.err_ = "values segment size mismatch";
        return v;
    }
    if ((h.offset_slots - h.offset_code) % 4 != 0) {
        v.err_ = "code segment not aligned";
        return v;
    }
    if (h.entry_fn_id >= h.n_funcs && h.n_funcs != 0) {
        v.err_ = "entry fn_id out of range";
        return v;
    }
    v.ok_ = true;
    v.err_.clear();
    return v;
}

/**
 * @brief 从 vector 构造只读视图。
 */
Image Image::from(const std::vector<uint8_t>& buf) {
    return from(buf.data(), buf.size());
}

/**
 * @brief 读函数表一行。
 */
FuncRow Image::func_row(size_t i) const {
    FuncRow r;
    if (ok_ && i < hdr_.n_funcs) {
        const uint8_t* p = buf_ + hdr_.offset_funcs + i * kFuncRowSize;
        r.nregs = get_le32(p + kFuncNregsOff);
        r.code_offset = get_le32(p + kFuncCodeOff);
        r.code_len = get_le32(p + kFuncLenOff);
    }
    return r;
}

/**
 * @brief 读常量表一项（8B 原始值）。
 */
ConstEntry Image::const_entry(size_t i) const {
    ConstEntry e;
    if (ok_ && i < hdr_.n_consts) {
        const uint8_t* p = buf_ + hdr_.offset_const + i * kConstEntrySize;
        e.value = get_le64(p + kConstValueOff);
    }
    return e;
}

/**
 * @brief 读槽表一项的地址（值段内偏移）。
 */
uint32_t Image::slot_addr(size_t i) const {
    if (ok_ && i < hdr_.n_slots) {
        const uint8_t* p = buf_ + hdr_.offset_slots + i * kSlotRowSize;
        return get_le32(p + kSlotAddrOff);
    }
    return 0;
}

/**
 * @brief 槽表段起点。
 */
const uint8_t* Image::slots_bytes() const {
    return ok_ ? buf_ + hdr_.offset_slots : nullptr;
}

/**
 * @brief 字节码段起点。
 */
const uint8_t* Image::code_bytes() const {
    return ok_ ? buf_ + hdr_.offset_code : nullptr;
}

/**
 * @brief 值段起点。
 */
const uint8_t* Image::data_bytes() const {
    return ok_ ? buf_ + hdr_.offset_values : nullptr;
}

/**
 * @brief 值段字节数。
 */
size_t Image::data_len() const {
    return ok_ ? (len_ - kSha256Size) - hdr_.offset_values : 0;
}

/**
 * @brief 型号标识字符串。
 */
std::string Image::model_id() const {
    if (!ok_) {
        return "";
    }
    std::string s(reinterpret_cast<const char*>(buf_ + kOffModelId), kModelIdSize);
    const size_t z = s.find('\0');
    if (z != std::string::npos) {
        s.resize(z);
    }
    return s;
}

/**
 * @brief 型号标识是否匹配。
 */
bool Image::model_matches(const std::string& name, uint32_t version) const {
    char want[kModelIdSize];
    fill_model_id(name, version, want);
    return std::memcmp(buf_ + kOffModelId, want, kModelIdSize) == 0;
}

/**
 * @brief 文件尾 SHA-256 完整性校验。
 */
bool Image::sha_ok() const {
    if (!ok_) {
        return false;
    }
    const size_t content_len = len_ - kSha256Size;
    uint8_t digest[32];
    sha256(buf_, content_len, digest);
    return std::memcmp(buf_ + content_len, digest, 32) == 0;
}

}  // namespace vm