/*
 * gfx_raw_16.c — 16-colour (mode 0x82) raw primitives.
 *
 * Pixel format (verified empirically 2026-05-31): each byte at
 * 0xC000+xb holds two horizontal pixels —
 *       high nibble (bits 7-4) = pixel at x = 2*xb     (LEFT,  even)
 *       low  nibble (bits 3-0) = pixel at x = 2*xb + 1 (RIGHT, odd)
 *
 * Row addressing matches mode 0x81: Port_Y (0x89) = y, 320 bytes/row,
 * 320 × 2 = 640 pixels.
 *
 * Accelerator: byte-wise — one Fill burst paints up to 256 bytes =
 * 512 pixels.  For a solid colour the byte value is the nibble in
 * both halves: b = c | (c<<4).
 *
 * RMW pattern for unaligned edges & vertical lines:
 *   read byte at (HL); mask out target nibble; OR in new nibble; write.
 * ~10 instructions per pixel — slow but unavoidable since a byte
 * spans two horizontal pixels.
 *
 * "Raw" = W3-naive: caller wraps a sequence with one
 * _gfx_w3_video_begin / _gfx_w3_video_end pair.
 */

#include <gfx.h>
#include <stdint.h>

extern uint16_t _gfx_addr_base;

/* ---- Scratch shared across asm helpers ---------------------------- */
static uint8_t  g16_y;
static uint8_t  g16_byte;     /* combined byte: nibble | (nibble<<4) */
static uint8_t  g16_nibble;   /* color in correct half (low or high) */
static uint8_t  g16_mask;     /* mask preserving the OTHER half */
static uint8_t  g16_len;      /* accel block size (0 = 256) */
static uint16_t g16_addr;

/* ---- Accel horizontal Fill burst ---------------------------------- *
 *
 * Caller has W3 mapped and DI active.  Only the block-size byte uses
 * SMC.  Colour is preloaded into C and shipped via `ld a, c` (0x79).
 * Inserting another `ld a, #n` between LD C,C and the firing LD (HL),A
 * breaks the burst — the accel FSM re-reads the immediate as a fresh
 * block size.
 */
static void g16_hfill_chunk(void) __naked
{
    __asm
        ld    a, (_g16_len)
        ld    (_g16_h_len_imm), a

        ld    a, (_g16_byte)
        ld    c, a

        ld    a, (_g16_y)
        out   (#0x89), a

        ld    hl, (_g16_addr)

        ld    d, d              ; 0x52 — set block size
        ld    a, #0             ; 0x3E nn — length (patched)
    _g16_h_len_imm = . - 1
        ld    c, c              ; 0x49 — horizontal Fill
        ld    a, c              ; 0x79 — A = colour byte
        ld    (hl), a           ; fires accel
        ld    b, b              ; 0x40 — disable
        ret
    __endasm;
}

/* ---- RMW one nibble at (g16_addr, g16_y) ------------------------- */
static void g16_rmw_pixel(void) __naked
{
    __asm
        ld    a, (_g16_y)
        out   (#0x89), a

        ld    hl, (_g16_addr)
        ld    a, (_g16_mask)
        ld    b, a              ; B = preserve mask
        ld    a, (_g16_nibble)
        ld    c, a              ; C = new nibble (in correct half)
        ld    a, (hl)
        and   a, b              ; clear target nibble
        or    a, c              ; OR in new
        ld    (hl), a
        ret
    __endasm;
}

/* ---- Raw primitives (W3-naive, composable) ----------------------- */

void _gfx_putpixel16_raw(int x, int y, uint8_t color)
{
    if ((unsigned)x >= GFX_WIDTH_16 || (unsigned)y >= GFX_HEIGHT_16) return;
    g16_y    = (uint8_t)y;
    g16_addr = (uint16_t)(_gfx_addr_base + ((unsigned)x >> 1));
    if (x & 1) {
        g16_nibble = (uint8_t)(color & 0x0F);
        g16_mask   = 0xF0;
    } else {
        g16_nibble = (uint8_t)((color & 0x0F) << 4);
        g16_mask   = 0x0F;
    }
    g16_rmw_pixel();
}

void _gfx_hline16_raw(int x, int y, int len, uint8_t color)
{
    if ((unsigned)y >= GFX_HEIGHT_16) return;
    if (x < 0)                  { len += x; x = 0; }
    if (x >= GFX_WIDTH_16)      return;
    if (x + len > GFX_WIDTH_16) len = GFX_WIDTH_16 - x;
    if (len <= 0)               return;

    g16_y = (uint8_t)y;
    uint8_t cnib = color & 0x0F;
    g16_byte = (uint8_t)(cnib | (cnib << 4));

    /* Leading unaligned pixel: x odd → RIGHT half of left-most byte. */
    if (x & 1) {
        g16_addr   = (uint16_t)(_gfx_addr_base + ((unsigned)x >> 1));
        g16_nibble = cnib;
        g16_mask   = 0xF0;
        g16_rmw_pixel();
        x++;
        len--;
        if (len <= 0) return;
    }

    /* Even x; emit (len/2) full bytes via accel hfill. */
    int full = len >> 1;
    g16_addr = (uint16_t)(_gfx_addr_base + ((unsigned)x >> 1));
    while (full > 0) {
        int chunk = full > 256 ? 256 : full;
        g16_len = (chunk == 256) ? 0 : (uint8_t)chunk;
        g16_hfill_chunk();
        full -= chunk;
        g16_addr += chunk;
    }

    /* Trailing odd-length pixel: LEFT half of the next byte. */
    if (len & 1) {
        g16_nibble = (uint8_t)(cnib << 4);
        g16_mask   = 0x0F;
        g16_rmw_pixel();
    }
}

void _gfx_vline16_raw(int x, int y, int len, uint8_t color)
{
    if ((unsigned)x >= GFX_WIDTH_16) return;
    if (y < 0)                   { len += y; y = 0; }
    if (y >= GFX_HEIGHT_16)      return;
    if (y + len > GFX_HEIGHT_16) len = GFX_HEIGHT_16 - y;
    if (len <= 0)                return;

    g16_addr = (uint16_t)(_gfx_addr_base + ((unsigned)x >> 1));
    if (x & 1) {
        g16_nibble = (uint8_t)(color & 0x0F);
        g16_mask   = 0xF0;
    } else {
        g16_nibble = (uint8_t)((color & 0x0F) << 4);
        g16_mask   = 0x0F;
    }
    for (int i = 0; i < len; i++) {
        g16_y = (uint8_t)(y + i);
        g16_rmw_pixel();
    }
}

/* Clear the whole 640×256 area with `color`.  Row-major hfill —
 * 256 rows × 2 bursts each (256+64 bytes). */
void _gfx_clear16_raw(uint8_t color)
{
    g16_byte = (uint8_t)((color & 0x0F) | ((color & 0x0F) << 4));
    for (int y = 0; y < GFX_HEIGHT_16; y++) {
        g16_y    = (uint8_t)y;
        g16_addr = _gfx_addr_base;
        g16_len  = 0;
        g16_hfill_chunk();
        g16_addr = (uint16_t)(_gfx_addr_base + 0x100);
        g16_len  = 64;
        g16_hfill_chunk();
    }
}