#include <string.h>

#include "ppu.h"

//#define NESE_DEBUG "PPU"
#include "log.h"

#include "serdes.h"


void nes_ppu_init(nes_PPU* ppu, nes_PPU_Memory* mem) {
    uint8_t* pal = mem->palette;
    pal[0] = pal[4] = pal[8] = pal[12] =
        pal[16] = pal[20] = pal[24] =
        pal[28] = 0xFFU;

    ppu->hit_line = -1;
}


static const uint8_t mirror_schemes[][4] = {
    [nes_Mirror_Horizontal] = {0, 0, 1, 1},
    [nes_Mirror_Vertical]   = {0, 1, 0, 1},
    [nes_Mirror_Only_0]     = {0, 0, 0, 0},
    [nes_Mirror_Only_1]     = {1, 1, 1, 1},
    [nes_Mirror_Four]       = {0, 1, 2, 3},
};


void nes_ppu_set_mirroring(nes_PPU_Memory* mem,
                           nes_Nametable_Mirroring mirror) {
    for (int i = 0; i < 4; ++i) {
        mem->bank[nes_PPU_Nametable_Bank_Index + i] =
                vram_page(mem, (int)mirror_schemes[mirror][i]);
    }
}

void nes_ppu_find_hit_line(nes_PPU* ppu, nes_PPU_Memory* mem) {
    int sprite_bank = !!(mem->ctrl & ppu_Control_Sprite_Bank);
    oam_sprite sprite = *(oam_sprite*)mem->oam;
    int sprite_height = (mem->ctrl & ppu_Control_Sprite_Size) ?
                        16 : 8;
    int stride = 1;
    int y_off = 0;
    if (sprite.attr & oam_Attr_Flip_Y) {
        stride = -1;
        y_off = sprite_height - 1;
        sprite_bank = (sprite.index & 1);
        sprite.index &= 0xFEU;
    }

    sprite_bank = (sprite_bank << 2) | (sprite.index >> 6);

    const int data_off = ((sprite.index & 0x3FU) << 4) +
                         ((y_off & 8) << 1) + (y_off & 7);
    uint8_t* data = mem->bank[sprite_bank] + data_off;

    ppu->hit_line = -1;

    if (sprite.y + 1 <= nes_ppu_render_h && sprite.y > 0) {
        for (int line = 0; line < sprite_height; ++line) {
            if (data[0] | data[8]) {
                ppu->hit_line = sprite.y + 1 + line + nes_ppu_visible_line;
                break;
            }
            data += stride;
        }
    }
}

static const uint8_t sprite_rev[256] = {
    0x00, 0x40, 0x80, 0xC0, 0x10, 0x50, 0x90, 0xD0,
    0x20, 0x60, 0xA0, 0xE0, 0x30, 0x70, 0xB0, 0xF0,
    0x04, 0x44, 0x84, 0xC4, 0x14, 0x54, 0x94, 0xD4,
    0x24, 0x64, 0xA4, 0xE4, 0x34, 0x74, 0xB4, 0xF4,
    0x08, 0x48, 0x88, 0xC8, 0x18, 0x58, 0x98, 0xD8,
    0x28, 0x68, 0xA8, 0xE8, 0x38, 0x78, 0xB8, 0xF8,
    0x0C, 0x4C, 0x8C, 0xCC, 0x1C, 0x5C, 0x9C, 0xDC,
    0x2C, 0x6C, 0xAC, 0xEC, 0x3C, 0x7C, 0xBC, 0xFC,
    0x01, 0x41, 0x81, 0xC1, 0x11, 0x51, 0x91, 0xD1,
    0x21, 0x61, 0xA1, 0xE1, 0x31, 0x71, 0xB1, 0xF1,
    0x05, 0x45, 0x85, 0xC5, 0x15, 0x55, 0x95, 0xD5,
    0x25, 0x65, 0xA5, 0xE5, 0x35, 0x75, 0xB5, 0xF5,
    0x09, 0x49, 0x89, 0xC9, 0x19, 0x59, 0x99, 0xD9,
    0x29, 0x69, 0xA9, 0xE9, 0x39, 0x79, 0xB9, 0xF9,
    0x0D, 0x4D, 0x8D, 0xCD, 0x1D, 0x5D, 0x9D, 0xDD,
    0x2D, 0x6D, 0xAD, 0xED, 0x3D, 0x7D, 0xBD, 0xFD,
    0x02, 0x42, 0x82, 0xC2, 0x12, 0x52, 0x92, 0xD2,
    0x22, 0x62, 0xA2, 0xE2, 0x32, 0x72, 0xB2, 0xF2,
    0x06, 0x46, 0x86, 0xC6, 0x16, 0x56, 0x96, 0xD6,
    0x26, 0x66, 0xA6, 0xE6, 0x36, 0x76, 0xB6, 0xF6,
    0x0A, 0x4A, 0x8A, 0xCA, 0x1A, 0x5A, 0x9A, 0xDA,
    0x2A, 0x6A, 0xAA, 0xEA, 0x3A, 0x7A, 0xBA, 0xFA,
    0x0E, 0x4E, 0x8E, 0xCE, 0x1E, 0x5E, 0x9E, 0xDE,
    0x2E, 0x6E, 0xAE, 0xEE, 0x3E, 0x7E, 0xBE, 0xFE,
    0x03, 0x43, 0x83, 0xC3, 0x13, 0x53, 0x93, 0xD3,
    0x23, 0x63, 0xA3, 0xE3, 0x33, 0x73, 0xB3, 0xF3,
    0x07, 0x47, 0x87, 0xC7, 0x17, 0x57, 0x97, 0xD7,
    0x27, 0x67, 0xA7, 0xE7, 0x37, 0x77, 0xB7, 0xF7,
    0x0B, 0x4B, 0x8B, 0xCB, 0x1B, 0x5B, 0x9B, 0xDB,
    0x2B, 0x6B, 0xAB, 0xEB, 0x3B, 0x7B, 0xBB, 0xFB,
    0x0F, 0x4F, 0x8F, 0xCF, 0x1F, 0x5F, 0x9F, 0xDF,
    0x2F, 0x6F, 0xAF, 0xEF, 0x3F, 0x7F, 0xBF, 0xFF,
};

void nes_ppu_render_line(nes_PPU* ppu, nes_PPU_Memory* mem) {
    uint8_t* scanline_ptr = ppu->screen_data +
                   (    nes_ppu_render_w *
                        (ppu->scanline - nes_ppu_visible_line));
    uint8_t* ptr = scanline_ptr;

    if (!(mem->mask & ppu_Mask_Back)) {
        memset(ptr, 0xFFU, nes_ppu_render_w);

    } else {
        int back_bank = nes_PPU_Nametable_Bank_Index +
                   ((mem->addr >> 10) & 3);

        int y_coarse = (mem->addr >> 5) & 0x1F;
        int y_fine = mem->addr >> 12;
        int x_coarse = mem->addr & 0x1FU;

        const int bank_off = !!(mem->ctrl & ppu_Control_Back_Bank) << 2;

        const uint8_t* nametable = mem->bank[back_bank] + (y_coarse * 32) + x_coarse;
        const uint8_t* attrs = mem->bank[back_bank] + 0x3C0U + ((y_coarse >> 2) << 3);

        // Partial left column
        ptr += 8 - mem->x;

        // Omit if left column is masked
        if (mem->mask & ppu_Mask_Left_Back) {
            const uint8_t* pal = mem->palette +
                    (((attrs[x_coarse >> 2] >>
                        (   (x_coarse & 2) +
                            ((y_coarse & 2) << 1))) & 3) << 2);
            const int ch = *nametable;
            const int bank = (ch >> 6) + bank_off;
            const int addr_off = ((ch & 0x3F) << 4) + y_fine;
            const uint8_t* data = mem->bank[bank] + addr_off;
            const uint8_t pl0 = data[0];
            const uint8_t pl1 = data[8];
            const int pat0 = (pl0 & 0x55) | ((pl1 << 1) & 0xAA);
            const int pat1 = ((pl0 >> 1) & 0x55) | (pl1 & 0xAA);
            switch (mem->x) {
                case 0: ptr[-8] = pal[(pat1 >> 6) & 3];
                case 1: ptr[-7] = pal[(pat0 >> 6) & 3];
                case 2: ptr[-6] = pal[(pat1 >> 4) & 3];
                case 3: ptr[-5] = pal[(pat0 >> 4) & 3];
                case 4: ptr[-4] = pal[(pat1 >> 2) & 3];
                case 5: ptr[-3] = pal[(pat0 >> 2) & 3];
                case 6: ptr[-2] = pal[(pat1 >> 0) & 3];
                case 7: ptr[-1] = pal[(pat0 >> 0) & 3];
            }
        }

        // TODO: Mapper ppu_bus

        ++x_coarse;
        ++nametable;

        void __attribute__((always_inline)) inline render_bg(void) {
            const uint8_t* pal = mem->palette +
                (((attrs[x_coarse >> 2] >>
                    (   (x_coarse & 2) +
                        ((y_coarse & 2) << 1))) & 3) << 2);
            const int ch = *nametable;
            const int bank = (ch >> 6) + bank_off;
            const int addr_off = ((ch & 0x3F) << 4) + y_fine;
            const uint8_t* data = mem->bank[bank] + addr_off;
            const uint8_t pl0 = data[0];
            const uint8_t pl1 = data[8];
            const int pat0 = (pl0 & 0x55) | ((pl1 << 1) & 0xAA);
            const int pat1 = ((pl0 >> 1) & 0x55) | (pl1 & 0xAA);

            ptr[0] = pal[(pat1 >> 6) & 3];
            ptr[1] = pal[(pat0 >> 6) & 3];
            ptr[2] = pal[(pat1 >> 4) & 3];
            ptr[3] = pal[(pat0 >> 4) & 3];
            ptr[4] = pal[(pat1 >> 2) & 3];
            ptr[5] = pal[(pat0 >> 2) & 3];
            ptr[6] = pal[(pat1 >> 0) & 3];
            ptr[7] = pal[(pat0 >> 0) & 3];
            ptr += 8;
        }

        // Left Nametable
        for (; x_coarse < 32; ++x_coarse) {
            render_bg();
            // TODO: Mapper ppu_bus
            ++nametable;
        }

        // Right Nametable
        back_bank ^= 0b01;
        nametable = mem->bank[back_bank] + (y_coarse * 32);
        attrs = mem->bank[back_bank] + 0x3C0U + ((y_coarse >> 2) << 3);

        for (x_coarse = 0; x_coarse < (mem->addr & 0x1FU); ++x_coarse) {
            render_bg();
            // TODO: Mapper ppu_bus
            ++nametable;
        }

        // Partial right column
        {
            const uint8_t* pal = mem->palette +
                    (((attrs[x_coarse >> 2] >>
                        (   (x_coarse & 2) +
                            ((y_coarse & 2) << 1))) & 3) << 2);
            const int ch = *nametable;
            const int bank = (ch >> 6) + bank_off;
            const int addr_off = ((ch & 0x3F) << 4) + y_fine;
            const uint8_t* data = mem->bank[bank] + addr_off;
            const uint8_t pl0 = data[0];
            const uint8_t pl1 = data[8];
            const int pat0 = (pl0 & 0x55) | ((pl1 << 1) & 0xAA);
            const int pat1 = ((pl0 >> 1) & 0x55) | (pl1 & 0xAA);
            switch (mem->x) {
                case 8: ptr[7] = pal[(pat0 >> 0) & 3];
                case 7: ptr[6] = pal[(pat1 >> 0) & 3];
                case 6: ptr[5] = pal[(pat0 >> 2) & 3];
                case 5: ptr[4] = pal[(pat1 >> 2) & 3];
                case 4: ptr[3] = pal[(pat0 >> 4) & 3];
                case 3: ptr[2] = pal[(pat1 >> 4) & 3];
                case 2: ptr[1] = pal[(pat0 >> 6) & 3];
                case 1: ptr[0] = pal[(pat1 >> 6) & 3];
            }
        }
    }

    // TODO: Mapper ppu_bus

    if (!(mem->mask & ppu_Mask_Left_Back)) {
        memset(scanline_ptr, 0xFFU, 8);
    }

    // TODO: Mapper VROM switch

    if (mem->mask & ppu_Mask_Sprite) {
        const int sprite_height = (mem->ctrl & ppu_Control_Sprite_Size) ?
                                  16 : 8;
        const int scanline = ppu->scanline - nes_ppu_visible_line;

        int sprite_bank = !!(mem->ctrl & ppu_Control_Sprite_Bank) << 2;

        uint8_t foreground[nes_ppu_render_w];
        memset(foreground, 0xFFU, nes_ppu_render_w);

        int n_sprites = 0;
        const oam_sprite* sprites = (oam_sprite*)mem->oam;
        const oam_sprite* sprite = sprites +
                                   NES_PPU_SPRITE_COUNT - 1;
        for ( ; sprite >= sprites && n_sprites < 8; --sprite) {
            int y = sprite->y + 1;
            if (y > scanline || y + sprite_height <= scanline) {
                continue;
            }

            int y_off = scanline - y;
            if (sprite->attr & oam_Attr_Flip_Y) {
                y_off = sprite_height - y_off - 1;
            }

            int bank = sprite_bank;
            int ch = sprite->index;
            if (mem->ctrl & ppu_Control_Sprite_Size) {
                bank = (ch & 1) << 2;
                ch &= 0xFEU;
            }

            bank += (ch >> 6);
            const int addr_off = ((ch & 0x3fU) << 4) +
                                 ((y_off & 8) << 1) +
                                 (y_off & 7);
            const uint8_t* data = mem->bank[bank] + addr_off;
            const uint8_t pl0 = data[0];
            const uint8_t pl1 = data[8];
            int pat0 = (pl0 & 0x55) | ((pl1 << 1) & 0xAA);
            int pat1 = ((pl0 >> 1) & 0x55) | (pl1 & 0xAA);

            const uint8_t* pal = &mem->palette[0x10 + ((sprite->attr & oam_Attr_Pal_Mask) << 2)];
            const uint8_t* back =
                    (sprite->attr & oam_Attr_Background) ?
                    (scanline_ptr + sprite->x) : NULL;
            uint8_t* dst = foreground + sprite->x;

            const int over_x = ((int)sprite->x + 8) - nes_ppu_render_w;

            if (sprite->attr & oam_Attr_Flip_X) {
                uint8_t tmp = pat0;
                pat0 = sprite_rev[pat1];
                pat1 = sprite_rev[tmp];
            }

#define do_sprite(idx, pi, shift)                           \
                if (back && back[idx] != 0xFFU) {           \
                    dst[idx] = back[idx];                   \
                } else {                                    \
                    int pal_idx = (pat##pi >> shift) & 3;   \
                    if (pal_idx) dst[idx] = pal[pal_idx];   \
                }

            switch (over_x) {
            default:
                do_sprite(7, 0, 0);
            case 1:
                do_sprite(6, 1, 0);
            case 2:
                do_sprite(5, 0, 2);
            case 3:
                do_sprite(4, 1, 2);
            case 4:
                do_sprite(3, 0, 4);
            case 5:
                do_sprite(2, 1, 4);
            case 6:
                do_sprite(1, 0, 6);
            case 7:
                do_sprite(0, 1, 6);
            }

            ++n_sprites;
        }

        if (n_sprites >= 8) {
            mem->status |= ppu_Status_Overflow;
        } else {
            mem->status &= ~ppu_Status_Overflow;
        }

        // Mask column 0?
        if (!(mem->mask & ppu_Mask_Left_Sprite)) {
            memset(foreground, 0xFFU, 8);
        }

        // Overlay the sprites
        const uint8_t* fg = foreground;
        uint8_t* dst = scanline_ptr;
        for (int i = 0; i < nes_ppu_render_w; ++i) {
            if (*fg != 0xFFU) *dst = *fg;
            ++fg;
            ++dst;
        }
    }

    // Increment internal registers
    if (mem->mask & (ppu_Mask_Sprite | ppu_Mask_Back)) {
        uint16_t mask = 0b10000011111;
        mem->addr = (mem->addr & ~mask) | (mem->t & mask);

        int y_scroll = (mem->addr >> 12) |
                       ((mem->addr >> 2) & 0xF8U);
        if (nes_ppu_render_h - 1 == y_scroll) {
            y_scroll = 0;
            mem->addr ^= 0x800;
        } else if (0xFFU == y_scroll) {
            y_scroll = 0;
        } else {
            ++y_scroll;
        }
        mem->addr = (mem->addr & ~0b111001111100000) |
                    ((y_scroll & 7) << 12) |
                    ((y_scroll & 0xF8U) << 2);
    }
}

/*
static int nes_ppu_chr_ram_size(const void* _ppu, const void*) {
    const nes_PPU_Memory* ppu = (const nes_PPU_Memory*)_ppu;
    return (ppu->chr_ram ?
            ppu->n_chr_banks * NES_CHR_ROM_PAGE_SIZE :
            0);
}

static int nes_ppu_read_chr_ram(void* dst, const void* src,
                                int avail, const void*) {
    int size = 0;
    nes_PPU_Memory* ppu = (nes_PPU_Memory*)dst;
    if (ppu->chr_ram) {
        size = ppu->n_chr_banks * NES_CHR_ROM_PAGE_SIZE;
        if (size > avail) size = avail;
        memcpy(ppu->chr_ram, src, size);
    }
    return size;
}

static int nes_ppu_write_chr_ram(void* dst, const void* src,
                                 int avail, const void*) {
    int size = 0;
    const nes_PPU_Memory* ppu = (const nes_PPU_Memory*)src;
    if (ppu->chr_ram) {
        size = ppu->n_chr_banks * NES_CHR_ROM_PAGE_SIZE;
        if (size > avail) size = avail;
        memcpy(dst, ppu->chr_ram, size);
    }
    return size;
}

static const Serdes_IO ppu_serdes_io = {
    .read = nes_ppu_read_chr_ram,
    .write = nes_ppu_write_chr_ram,
    .in_size = nes_ppu_chr_ram_size,
    .out_size = nes_ppu_chr_ram_size,
};
*/

static void* ppu_chr_ram_ptr(const void* _ppu) {
    return ((nes_PPU_Memory*)_ppu)->chr_ram;
}

static size_t ppu_chr_ram_size(const void* _ppu) {
    return ((nes_PPU_Memory*)_ppu)->n_chr_banks * NES_CHR_ROM_PAGE_SIZE;
}

static Serdes_Ptr_Ref ppu_chr_ram_ref = {
    .ptr = ppu_chr_ram_ptr,
    .size = ppu_chr_ram_size,
};

const Serdes_Item nes_ppu_memory_serdes[] = {
    {"CRAM", 0, &serdes_mem_ptr, &ppu_chr_ram_ref},
    {"VRAM", offsetof(nes_PPU_Memory, ctrl), &serdes_mem, (void*)(sizeof(nes_PPU_Memory) - offsetof(nes_PPU_Memory, ctrl))},
    {0}
};