// engine/dxt —— DDS(DXT1/3/5) + 未压缩 BGRA8 → RGBA8 软解。M1 T3。
#include "dxt.h"
#include <cstdio>
#include <cstring>

namespace mtgodot {
namespace {

inline uint32_t rd_le32(const uint8_t* p) {
    return uint32_t(p[0]) | (uint32_t(p[1]) << 8) | (uint32_t(p[2]) << 16) | (uint32_t(p[3]) << 24);
}

// 5:6:5 → r,g,b (0..255)
inline void unpack565(uint16_t c, int& r, int& g, int& b) {
    r = ((c >> 11) & 0x1f); r = (r << 3) | (r >> 2);
    g = ((c >> 5) & 0x3f);  g = (g << 2) | (g >> 4);
    b = (c & 0x1f);         b = (b << 3) | (b >> 2);
}

// 解一个 DXT1 颜色块（8B）→ 16 像素 RGB（不写 alpha；dxt1_alpha=true 时按 1-bit alpha 写）
void decode_color_block(const uint8_t* blk, uint8_t out[16][4], bool dxt1_alpha) {
    uint16_t c0 = uint16_t(blk[0] | (blk[1] << 8));
    uint16_t c1 = uint16_t(blk[2] | (blk[3] << 8));
    int r[4], g[4], b[4], a[4] = {255, 255, 255, 255};
    unpack565(c0, r[0], g[0], b[0]);
    unpack565(c1, r[1], g[1], b[1]);
    if (c0 > c1 || !dxt1_alpha) {
        r[2] = (2 * r[0] + r[1]) / 3; g[2] = (2 * g[0] + g[1]) / 3; b[2] = (2 * b[0] + b[1]) / 3;
        r[3] = (r[0] + 2 * r[1]) / 3; g[3] = (g[0] + 2 * g[1]) / 3; b[3] = (b[0] + 2 * b[1]) / 3;
    } else {
        r[2] = (r[0] + r[1]) / 2; g[2] = (g[0] + g[1]) / 2; b[2] = (b[0] + b[1]) / 2;
        r[3] = g[3] = b[3] = 0; a[3] = 0;   // 透明
    }
    uint32_t bits = rd_le32(blk + 4);
    for (int i = 0; i < 16; ++i) {
        int idx = (bits >> (i * 2)) & 3;
        out[i][0] = uint8_t(r[idx]); out[i][1] = uint8_t(g[idx]);
        out[i][2] = uint8_t(b[idx]); out[i][3] = uint8_t(a[idx]);
    }
}

// DXT3：alpha 块（8B）= 16 个 4-bit alpha，直接展开
void decode_dxt3_alpha(const uint8_t* blk, uint8_t out[16][4]) {
    for (int i = 0; i < 8; ++i) {
        int a0 = blk[i] & 0x0f, a1 = (blk[i] >> 4) & 0x0f;
        out[i * 2 + 0][3] = uint8_t(a0 * 17);   // 0..15 → 0..255
        out[i * 2 + 1][3] = uint8_t(a1 * 17);
    }
}

// DXT5：alpha 块（8B）= 2 端点 + 16×3-bit 索引
void decode_dxt5_alpha(const uint8_t* blk, uint8_t out[16][4]) {
    int a0 = blk[0], a1 = blk[1];
    int a[8];
    a[0] = a0; a[1] = a1;
    if (a0 > a1) {
        for (int i = 1; i < 7; ++i) a[i + 1] = ((7 - i) * a0 + i * a1) / 7;
    } else {
        for (int i = 1; i < 5; ++i) a[i + 1] = ((5 - i) * a0 + i * a1) / 5;
        a[6] = 0; a[7] = 255;
    }
    uint64_t bits = 0;
    for (int i = 0; i < 6; ++i) bits |= uint64_t(blk[2 + i]) << (8 * i);
    for (int i = 0; i < 16; ++i) {
        int idx = int((bits >> (i * 3)) & 7);
        out[i][3] = uint8_t(a[idx]);
    }
}

enum Fmt { F_NONE, F_DXT1, F_DXT3, F_DXT5, F_BGRA8 };

Image decode(const uint8_t* d, size_t len) {
    Image img;
    if (len < 128 || std::memcmp(d, "DDS ", 4) != 0) return img;
    uint32_t hsize = rd_le32(d + 4);
    if (hsize != 124) return img;
    uint32_t h = rd_le32(d + 12);
    uint32_t w = rd_le32(d + 16);
    uint32_t pf_flags = rd_le32(d + 80);
    const uint8_t* fourcc = d + 84;
    uint32_t rgb_bitcount = rd_le32(d + 88);

    Fmt fmt = F_NONE;
    if (pf_flags & 0x4) {   // DDPF_FOURCC
        if (!std::memcmp(fourcc, "DXT1", 4)) fmt = F_DXT1;
        else if (!std::memcmp(fourcc, "DXT3", 4)) fmt = F_DXT3;
        else if (!std::memcmp(fourcc, "DXT5", 4)) fmt = F_DXT5;
    } else if ((pf_flags & 0x40) && rgb_bitcount == 32) {   // DDPF_RGB, 32bpp
        fmt = F_BGRA8;   // Metin2 的非压缩 DDS 一般是 B8G8R8A8
    }
    if (fmt == F_NONE || w == 0 || h == 0 || w > 8192 || h > 8192) return img;

    const uint8_t* src = d + 128;
    size_t avail = len - 128;
    img.w = uint16_t(w);
    img.h = uint16_t(h);
    img.rgba.assign(size_t(w) * h * 4, 0);

    if (fmt == F_BGRA8) {
        if (avail < size_t(w) * h * 4) { img = Image{}; return img; }
        for (size_t i = 0; i < size_t(w) * h; ++i) {
            img.rgba[i * 4 + 0] = src[i * 4 + 2];
            img.rgba[i * 4 + 1] = src[i * 4 + 1];
            img.rgba[i * 4 + 2] = src[i * 4 + 0];
            img.rgba[i * 4 + 3] = src[i * 4 + 3];
        }
        img.format = "BGRA8";
        return img;
    }

    const int block_bytes = (fmt == F_DXT1) ? 8 : 16;
    const size_t bx = (w + 3) / 4, by = (h + 3) / 4;
    if (avail < bx * by * block_bytes) { img = Image{}; return img; }

    for (size_t byi = 0; byi < by; ++byi) {
        for (size_t bxi = 0; bxi < bx; ++bxi) {
            const uint8_t* blk = src + (byi * bx + bxi) * block_bytes;
            uint8_t px[16][4];
            if (fmt == F_DXT1) {
                decode_color_block(blk, px, /*dxt1_alpha=*/true);
            } else {
                decode_color_block(blk + 8, px, /*dxt1_alpha=*/false);
                if (fmt == F_DXT3) decode_dxt3_alpha(blk, px);
                else               decode_dxt5_alpha(blk, px);
            }
            for (int py = 0; py < 4; ++py) {
                size_t y = byi * 4 + py;
                if (y >= h) break;
                for (int pxx = 0; pxx < 4; ++pxx) {
                    size_t x = bxi * 4 + pxx;
                    if (x >= w) break;
                    uint8_t* o = &img.rgba[(y * w + x) * 4];
                    const uint8_t* s = px[py * 4 + pxx];
                    o[0] = s[0]; o[1] = s[1]; o[2] = s[2]; o[3] = s[3];
                }
            }
        }
    }
    img.format = (fmt == F_DXT1) ? "DXT1" : (fmt == F_DXT3) ? "DXT3" : "DXT5";
    return img;
}

} // namespace

Image load_dds(const uint8_t* bytes, size_t len) { return decode(bytes, len); }

Image load_dds_path(const char* path) {
    FILE* f = std::fopen(path, "rb");
    if (!f) return {};
    std::fseek(f, 0, SEEK_END);
    long n = std::ftell(f);
    std::fseek(f, 0, SEEK_SET);
    std::vector<uint8_t> buf(n > 0 ? size_t(n) : 0);
    size_t rd = buf.empty() ? 0 : std::fread(buf.data(), 1, buf.size(), f);
    std::fclose(f);
    if (rd != buf.size()) return {};
    return decode(buf.data(), buf.size());
}

} // namespace mtgodot