#if 0 /*
#/  ================================================================
#/
#/    graphics.c
#/
#/  ----------------------------------------------------------------
#/
#/    (C) 2025 Mitya Selivanov <guattari.tech>
#/
#/  ================================================================
#/
#/    Self-testing shell script
#/
#/  ================================================================
SRC=${0##*./}
BIN=${SRC%.*}
gcc                                 \
  -Wall -Wextra -Werror -pedantic   \
  -Wno-missing-braces               \
  -Wno-old-style-declaration        \
  -Wno-overlength-strings           \
  -O3 -D NDEBUG                     \
  -D GRAPHICS_TEST_SUITE            \
  -lX11 -lm -lasound                \
  -o $BIN $SRC &&                   \
  ./$BIN $@
STATUS=$?
rm -f $BIN
exit $? # */
#endif
//  ================================================================

#ifndef GRAPHICS_HEADER_GUARD_
#define GRAPHICS_HEADER_GUARD_

#ifdef EVERY_TEST_SUITE
#define GRAPHICS_TEST_SUITE
#endif

#ifdef GRAPHICS_HEADER
#define REDUCED_SYSTEM_LAYER_HEADER
#endif

#include "reduced_system_layer.c"

//  ================================================================
//
//    Options
//
//  ================================================================

#ifndef EPSILON
#define EPSILON (1e-9)
#endif

#ifndef ANTIALIASING_SCALE
#define ANTIALIASING_SCALE 3
#endif

//  ================================================================
//
//    Vector math
//
//  ================================================================

vec3_f32 vec3_from_vec4_f32(vec4_f32 v);
vec4_f32 vec4_from_vec3_f32(vec3_f32 v, f32 w);
vec3_f32 vec3_f32_lerp     (vec3_f32 a, vec3_f32 b, f32 t);
vec4_f32 vec4_f32_lerp     (vec4_f32 a, vec4_f32 b, f32 t);

//  ================================================================

typedef struct {
  i64       width;
  i64       height;
  i64       line_size;
  vec4_f32 *pixels;
} Pixel_Buffer;

typedef struct {
  Pixel_Buffer buffer;
  vec2         position;
  vec2         scale;
  b8           quick     : 1; // If set, anti-aliasing is skipped.
  b8           alpha     : 1; // If set, the resulting color is: dst * (1 - alpha) + src * alpha.
  b8           xor_color : 1; // If set, the source color is XORed with the destination color.
  vec4_f32     color;
} Brush;

vec3_f32 lab_from_rgb(vec3_f32 rgb);
vec3_f32 rgb_from_lab(vec3_f32 lab);
vec3_f32 lab_from_lch(vec3_f32 lch);
vec3_f32 lch_from_lab(vec3_f32 lab);
vec3_f32 rgb_from_lch(vec3_f32 lch);
vec3_f32 lch_from_rgb(vec3_f32 rgb);

//  NOTE:
//  General rule to work with gamma is removing it,
//  then doing computations like blending, interpolation, etc,
//  then adding it back.

vec3_f32 rgb_gamma_add    (vec3_f32 rgb);
vec3_f32 rgb_gamma_remove (vec3_f32 rgb);
vec4_f32 rgba_gamma_add   (vec4_f32 rgb);
vec4_f32 rgba_gamma_remove(vec4_f32 rgb);

vec3_f32 rgb_gamma_lerp (vec3_f32 a, vec3_f32 b, f32 t);
vec4_f32 rgba_gamma_lerp(vec4_f32 a, vec4_f32 b, f32 t);
vec3_f32 rgb_mix        (vec3_f32 a, vec3_f32 b, f32 t);
vec4_f32 rgba_mix       (vec4_f32 a, vec4_f32 b, f32 t);
vec3_f32 lch_mix        (vec3_f32 a, vec3_f32 b, f32 t);
vec3_f32 lch_lerp       (vec3_f32 a, vec3_f32 b, f32 t);

u32      rgb_u32_from_f32(vec3_f32 color);
vec3_f32 rgb_f32_from_u32(u32 color);
u32      rgba_u32_from_f32(vec4_f32 color);
vec4_f32 rgba_f32_from_u32(u32 color);

#define RGB(...)  ((Brush) { .color = { __VA_ARGS__, 1.f } })
#define RGBA(...) ((Brush) { .color = { __VA_ARGS__ } })
#define LCH(...)  ((Brush) { .color = vec4_from_vec3_f32(rgb_from_lch((vec3_f32) { __VA_ARGS__ }), 1.f) } })
#define LCHA(...) ((Brush) { .color = vec4_from_vec3_f32(rgb_from_lch(vec3_from_vec4_f32((vec4_f32) { __VA_ARGS__ })), ((vec4_f32) { __VA_ARGS__ }).w) } })

b8 rectangle_contains(f64 x0, f64 y0, f64 width, f64 height,                          f64 px, f64 py);
b8 triangle_contains (f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2,                 f64 px, f64 py);
b8 quad_contains     (f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2, f64 x3, f64 y3, f64 px, f64 py);
b8 ellipse_contains  (f64 x0, f64 y0, f64 width, f64 height,                          f64 px, f64 py);
b8 line_contains     (f64 x0, f64 y0, f64 x1, f64 y1, f64 width,                      f64 px, f64 py);

Pixel_Buffer frame_pixels (void);
Pixel_Buffer sketch_pixels(i64 width, i64 height);
Pixel_Buffer subimage     (Pixel_Buffer image, i64 x, i64 y, i64 width, i64 height);

void put_pixel       (Brush brush, i64 x, i64 y);
void draw_pixels     (Brush brush, f64 x, f64 y, f64 width, f64 height, Pixel_Buffer src);
void fill_rectangle  (Brush brush, f64 x, f64 y, f64 width, f64 height);
void fill_triangle   (Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2);
void fill_quad       (Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2, f64 x3, f64 y3);
void fill_ellipse    (Brush brush, f64 x, f64 y, f64 width, f64 height);
void fill_line       (Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 width);
void draw_text_area  (Brush brush, f64 x, f64 y, f64 width, f64 height, f64 max_scale_x, f64 max_scale_y, i64 num_chars, c32 *text);
void draw_text_cursor(Brush brush, f64 x, f64 y, f64 width, f64 height, f64 max_scale_x, f64 max_scale_y, i64 cursor, i64 selection, i64 num_chars, c32 *text);

#endif // GRAPHICS_HEADER_GUARD_

//  ================================================================

#ifndef GRAPHICS_HEADER
#ifndef GRAPHICS_IMPL_GUARD_
#define GRAPHICS_IMPL_GUARD_

static u64 _bitfont[] = {
  0xbc0000000000, 0xc00300000, 0x5fd5040093f24fc9, 0xa00a2c2a1a280105, 0xc000415e6f, 0x400000020be0000, 0x1c38a8400000007d, 0x40002043e1020215, 0x408102000000010, 0x9800000000020002, 0xf913e00000033, 0x53200000207c8800, 0x3654880000099, 0x54b800000f840e00, 0xe953c000001a, 0x953e000000674080, 0x1e54b800000f, 0x490000000000240, 0x88a08000000, 0x20a220050a142850, 0x6520800000, 0x912f801eab260be, 0x800034952bf0001f, 0xc850bf0000921427, 0xf00010a54afc0003, 0xd29427800002142b, 0x840007e1023f0000, 0x7d09100000217e, 0x3f000188a08fc000, 0xc30c0cfc00000810, 0x27803f101013f00f, 0xc244bf0000f214, 0x4bf0002f21427800, 0xc254a480006c24, 0x407c00102fc08100, 0xf208080f0000fa0, 0x531007d81c607c0, 0xc208288c031141, 0x83fc00046954b10, 0x180e03000000, 0x41040000000ff04, 0x8102040810000404, 0x2a54600000000101, 0x309123e0000e, 0xc912180000a22447, 0x8000062a54700007, 0xe52a4300000029f0, 0xa0000602043e0001, 0x1d48000002074, 0x1f000003610f8000, 0x13e04f800000010, 0x470000780813e00f, 0x184893e0000e224, 0x23e0001f12243000, 0x82a54100000008, 0x40780000009f0200, 0xe208080e0001f20, 0xa22007981860780, 0x82082888022282, 0x16c200004ca95320, 0x7f000004, 0x408200000086d04, 0x8204,
};

enum {
  CHAR_NUM_BITS_X_ = 6,
  CHAR_NUM_BITS_Y_ = 7,
  CHAR_NUM_BITS_   = CHAR_NUM_BITS_X_ * CHAR_NUM_BITS_Y_,
};

static f64 min3_(f64 a, f64 b, f64 c) {
  if (a < b && a < c)
    return a;
  if (b < c)
    return b;
  return c;
}

static f64 max3_(f64 a, f64 b, f64 c) {
  if (a > b && a > c)
    return a;
  if (b > c)
    return b;
  return c;
}

static f64 min4_(f64 a, f64 b, f64 c, f64 d) {
  if (a < b && a < c && a < d)
    return a;
  if (b < c && b < d)
    return b;
  if (c < d)
    return c;
  return d;
}

static f64 max4_(f64 a, f64 b, f64 c, f64 d) {
  if (a > b && a > c && a > d)
    return a;
  if (b > c && b > d)
    return b;
  if (c > d)
    return c;
  return d;
}

static b8 same_sign_(f64 a, f64 b) {
  if (a >=  EPSILON && b <= -EPSILON) return 0;
  if (a <= -EPSILON && b >=  EPSILON) return 0;
  return 1;
}

static f64 gamma_(f64 x) {
  if (x >= 0.0031308)
    return 1.055 * pow(x, 1.0 / 2.4) - 0.055;
  return 12.92 * x;
}

static f64 gamma_re_(f64 x) {
  if (x >= 0.04045)
    return pow((x + 0.055) / 1.055, 2.4);
  return x / 12.92;
}

static void put_pixel_color_(Brush brush, i64 x, i64 y) {
  brush.buffer.pixels[y * brush.buffer.line_size + x] = brush.color;
}

static void put_pixel_alpha_(Brush brush, i64 x, i64 y) {
  i64 n = y * brush.buffer.line_size + x;

  vec4_f32 dst = brush.buffer.pixels[n];
  f64      a   = brush.color.w;

  brush.buffer.pixels[n] = (vec4_f32) {
    .x = (f32) (dst.x * (1. - a) + brush.color.x * a),
    .y = (f32) (dst.y * (1. - a) + brush.color.y * a),
    .z = (f32) (dst.z * (1. - a) + brush.color.z * a),
    .w = 1.,
  };
}

static void put_pixel_xor_(Brush brush, i64 x, i64 y) {
  i64 n = y * brush.buffer.line_size + x;

  vec3_f32 c = rgb_f32_from_u32(
      rgb_u32_from_f32(vec3_from_vec4_f32(brush.buffer.pixels[n]))
    ^ rgb_u32_from_f32(vec3_from_vec4_f32(brush.color))
  );

  brush.color = vec4_from_vec3_f32(c, brush.color.w);

  if (brush.alpha)
    put_pixel_alpha_(brush, x, y);
  else
    brush.buffer.pixels[n] = brush.color;
}

static void put_pixel_(Brush brush, i64 x, i64 y) {
  if (brush.xor_color)
    put_pixel_xor_(brush, x, y);
  else if (brush.alpha)
    put_pixel_alpha_(brush, x, y);
  else
    put_pixel_color_(brush, x, y);
}

f64 transform_x_(Brush brush, f64 x) {
  return (brush.position.x + x) * brush.scale.x;
}

f64 transform_y_(Brush brush, f64 y) {
  return (brush.position.y + y) * brush.scale.y;
}

void transform_box_(Brush brush, f64 x, f64 y, f64 width, f64 height, f64 *x0, f64 *y0, f64 *x1, f64 *y1) {
  if (brush.scale.x < 0) {
    *x1 = transform_x_(brush, x);
    *x0 = *x1 + width * brush.scale.x;
  } else {
    *x0 = transform_x_(brush, x);
    *x1 = *x0 + width  * brush.scale.x;
  }

  if (brush.scale.y < 0) {
    *y1 = transform_y_(brush, y);
    *y0 = *y1 + height * brush.scale.y;
  } else {
    *y0 = transform_y_(brush, y);
    *y1 = *y0 + height * brush.scale.y;
  }
}

Brush antialiasing_brush_(Brush brush) {
  return (Brush) {
    .buffer    = sketch_pixels(brush.buffer.width * ANTIALIASING_SCALE, brush.buffer.height * ANTIALIASING_SCALE),
    .position  = brush.position,
    .scale     = {
      .x = brush.scale.x * ANTIALIASING_SCALE,
      .y = brush.scale.y * ANTIALIASING_SCALE,
    },
    .quick     = 1,
    .alpha     = 0,
    .xor_color = 0,
    .color     = brush.alpha
             ? brush.color
             : vec4_from_vec3_f32(vec3_from_vec4_f32(brush.color), 1.f),
  };
}

Brush copy_brush_(Brush brush) {
  return (Brush) {
    .buffer    = brush.buffer,
    .position  = brush.position,
    .scale     = brush.scale,
    .alpha     = 1,
    .xor_color = brush.xor_color,
  };
}

void draw_pixels_(Brush brush, f64 x, f64 y, f64 width, f64 height, Pixel_Buffer src) {
  //  FIXME PERF

  f64 x0, y0, x1, y1;
  transform_box_(brush, x, y, width, height, &x0, &y0, &x1, &y1);

  f64 w = x1 - x0;
  f64 h = y1 - y0;

  if (w < EPSILON || h < EPSILON)
    return;

  i64 i0 = (i64) floor(x0);
  i64 i1 = (i64) floor(x1);
  i64 j0 = (i64) floor(y0);
  i64 j1 = (i64) floor(y1);

  if (i0 <  0)                   i0 = 0;
  if (i1 >= brush.buffer.width)  i1 = brush.buffer.width - 1;
  if (j0 <  0)                   j0 = 0;
  if (j1 >= brush.buffer.height) j1 = brush.buffer.height - 1;

  f64 w_inv = 1. / w;
  f64 h_inv = 1. / h;

  if (!brush.quick)
    for (i64 j = j0; j <= j1; ++j) {
      i64 src_j0 = (i64) floor(((j     - y0) * src.height) * h_inv + .5);
      i64 src_j1 = (i64) floor(((j + 1 - y0) * src.height) * h_inv + .5);
      if (src_j0 == src_j1) src_j1 = src_j0 + 1;
      if (src_j1 <= 0 || src_j0 >= src.height) continue;
      if (src_j0 < 0)          src_j0 = 0;
      if (src_j1 > src.height) src_j1 = src.height;
      for (i64 i = i0; i <= i1; ++i) {
        i64 src_i0 = (i64) floor(((i     - x0) * src.width) * w_inv + .5);
        i64 src_i1 = (i64) floor(((i + 1 - x0) * src.width) * w_inv + .5);
        if (src_i0 == src_i1) src_i1 = src_i0 + 1;
        if (src_i1 <= 0 || src_i0 >= src.width) continue;
        if (src_i0 < 0)         src_i0 = 0;
        if (src_i1 > src.width) src_i1 = src.width;
        brush.color = (vec4_f32) {0};
        i64 n = 0;
        for (i64 jj = src_j0; jj < src_j1; ++jj)
          for (i64 ii = src_i0; ii < src_i1; ++ii) {
            brush.color.x += gamma_re_(src.pixels[jj * src.line_size + ii].x);
            brush.color.y += gamma_re_(src.pixels[jj * src.line_size + ii].y);
            brush.color.z += gamma_re_(src.pixels[jj * src.line_size + ii].z);
            brush.color.w += gamma_re_(src.pixels[jj * src.line_size + ii].w);
            ++n;
          }
        if (n == 0)
          continue;
        f32 n_inv = 1.f / n;
        brush.color.x = gamma_(brush.color.x * n_inv);
        brush.color.y = gamma_(brush.color.y * n_inv);
        brush.color.z = gamma_(brush.color.z * n_inv);
        brush.color.w = gamma_(brush.color.w * n_inv);
        put_pixel_(brush, i, j);
      }
    }
  else
    for (i64 j = j0; j <= j1; ++j) {
      i64 src_j = (i64) floor(((j - y0) * src.height) * h_inv + .5);
      if (src_j < 0 || src_j >= src.height) continue;
      for (i64 i = i0; i <= i1; ++i) {
        i64 src_i = (i64) floor(((i - x0) * src.width) * w_inv + .5);
        if (src_i < 0 || src_i >= src.width) continue;
        brush.color = src.pixels[src_j * src.line_size + src_i];
        put_pixel_(brush, i, j);
      }
    }
}

void draw_pixels_from_(Brush brush, f64 x, f64 y, f64 width, f64 height, Pixel_Buffer src, i64 src_x, i64 src_y, i64 src_width, i64 src_height) {
  //  NOTE: In the future we may need to implement
  //        pixel buffers with bounding boxes.

  f64 bx = 0;
  f64 by = 0;
  f64 bw = 0;
  f64 bh = 0;

  if (src_width > src.width - src_x) {
    bw += src_width - src.width + src_x;
    src_width = src.width - src_x;
  }

  if (src_height > src.height - src_y) {
    bh += src_height - src.height + src_y;
    src_height = src.height - src_y;
  }

  if (src_x < 0) {
    bx -= src_x;
    bw -= src_x;
    src_x = 0;
  }

  if (src_y < 0) {
    by -= src_y;
    bh -= src_y;
    src_y = 0;
  }

  if (src_x + src_width > src.width || src_y + src_height > src.height || src_width <= 0 || src_height <= 0)
    return;

  f64 w_inv = 1. / src_width;
  f64 h_inv = 1. / src_height;

  bx *= width  * w_inv;
  by *= height * h_inv;
  bw *= width  * w_inv;
  bh *= height * h_inv;

  draw_pixels_(brush, x + bx, y + by, width - bw, height - bh, subimage(src, src_x, src_y, src_width, src_height));
}

void fill_rectangle_(Brush brush, f64 x, f64 y, f64 width, f64 height) {
  if (width < EPSILON || height < EPSILON)
    return;

  f64 x0, y0, x1, y1;
  transform_box_(brush, x, y, width, height, &x0, &y0, &x1, &y1);

  if (!brush.quick) {
    i64 i0 = (i64) ceil (x0);
    i64 i1 = (i64) floor(x1);
    i64 j0 = (i64) ceil (y0);
    i64 j1 = (i64) floor(y1);

    i64 i00 = i0;
    i64 j00 = j0;
    i64 i10 = i1;
    i64 j10 = j1;

    if (i0 < 0)                   i0 = 0;
    if (i1 > brush.buffer.width)  i1 = brush.buffer.width;
    if (j0 < 0)                   j0 = 0;
    if (j1 > brush.buffer.height) j1 = brush.buffer.height;

    f64 kx0 = i00 - x0;
    f64 kx1 = x1  - i10;
    f64 ky0 = j00 - y0;
    f64 ky1 = y1  - j10;

    if (brush.xor_color)
      for (i64 j = j0; j < j1; ++j)
        for (i64 i = i0; i < i1; ++i)
          put_pixel_xor_(brush, i, j);
    else if (brush.alpha)
      for (i64 j = j0; j < j1; ++j)
        for (i64 i = i0; i < i1; ++i)
          put_pixel_alpha_(brush, i, j);
    else
      for (i64 j = j0; j < j1; ++j)
        for (i64 i = i0; i < i1; ++i)
          put_pixel_color_(brush, i, j);

    if (i00 > i10) {
      kx0 *= kx1;
      kx1 = 0.;
    }

    if (j00 > j10) {
      ky0 *= ky1;
      ky1 = 0.;
    }

    f64 alpha   = brush.alpha ? brush.color.w : 1.;
    brush.alpha = 1;

    if (i00 - 1 >= 0 && i00 - 1 < brush.buffer.width) {
      brush.color.w = gamma_(gamma_re_(alpha) * kx0);
      for (i64 j = j0; j < j1; ++j)
        put_pixel_(brush, i00 - 1, j);
    }

    if (i10 >= 0 && i10 < brush.buffer.width) {
      brush.color.w = gamma_(gamma_re_(alpha) * kx1);
      for (i64 j = j0; j < j1; ++j)
        put_pixel_(brush, i10, j);
    }

    if (j00 - 1 >= 0 && j00 - 1 < brush.buffer.height) {
      brush.color.w = gamma_(gamma_re_(alpha) * ky0);
      for (i64 i = i0; i < i1; ++i)
        put_pixel_(brush, i, j00 - 1);
    }

    if (j10 >= 0 && j10 < brush.buffer.height) {
      brush.color.w = gamma_(gamma_re_(alpha) * ky1);
      for (i64 i = i0; i < i1; ++i)
        put_pixel_(brush, i, j10);
    }

    if (   i00 - 1 >= 0 && i00 - 1 < brush.buffer.width
        && j00 - 1 >= 0 && j00 - 1 < brush.buffer.height) {
      brush.color.w = gamma_(gamma_re_(alpha) * kx0 * ky0);
      put_pixel_(brush, i00 - 1, j00 - 1);
    }

    if (   i10     >= 0 && i10     < brush.buffer.width
        && j00 - 1 >= 0 && j00 - 1 < brush.buffer.height) {
      brush.color.w = gamma_(gamma_re_(alpha) * kx1 * ky0);
      put_pixel_(brush, i10, j00 - 1);
    }

    if (   i00 - 1 >= 0 && i00 - 1 < brush.buffer.width
        && j10     >= 0 && j10     < brush.buffer.height) {
      brush.color.w = gamma_(gamma_re_(alpha) * kx0 * ky1);
      put_pixel_(brush, i00 - 1, j10);
    }

    if (   i10 >= 0 && i10 < brush.buffer.width
        && j10 >= 0 && j10 < brush.buffer.height) {
      brush.color.w = gamma_(gamma_re_(alpha) * kx1 * ky1);
      put_pixel_(brush, i10, j10);
    }
  } else {
    i64 i0 = (i64) floor(x0);
    i64 i1 = (i64) ceil (x1);
    i64 j0 = (i64) floor(y0);
    i64 j1 = (i64) ceil (y1);

    if (i0 < 0)                   i0 = 0;
    if (i1 > brush.buffer.width)  i1 = brush.buffer.width;
    if (j0 < 0)                   j0 = 0;
    if (j1 > brush.buffer.height) j1 = brush.buffer.height;

    if (brush.xor_color)
      for (i64 j = j0; j < j1; ++j)
        for (i64 i = i0; i < i1; ++i)
          put_pixel_xor_(brush, i, j);
    else if (brush.alpha)
      for (i64 j = j0; j < j1; ++j)
        for (i64 i = i0; i < i1; ++i)
          put_pixel_alpha_(brush, i, j);
    else
      for (i64 j = j0; j < j1; ++j)
        for (i64 i = i0; i < i1; ++i)
          put_pixel_color_(brush, i, j);
  }
}

void fill_triangle_(Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2) {
  //  FIXME PERF:
  //  Implement better algorithm.

  x0 = transform_x_(brush, x0);
  y0 = transform_y_(brush, y0);

  x1 = transform_x_(brush, x1);
  y1 = transform_y_(brush, y1);

  x2 = transform_x_(brush, x2);
  y2 = transform_y_(brush, y2);

  i64 i0 = (i64) floor(min3_(x0, x1, x2));
  i64 j0 = (i64) floor(min3_(y0, y1, y2));
  i64 i1 = (i64) ceil (max3_(x0, x1, x2));
  i64 j1 = (i64) ceil (max3_(y0, y1, y2));

  if (j0 < 0)                    j0 = 0;
  if (j1 >= brush.buffer.height) j1 = brush.buffer.height - 1;

  for (i64 j = j0; j <= j1; ++j) {
    i64 left  = i1;
    i64 right = i0;

    for (i64 i = i0; i <= i1; ++i)
      if (triangle_contains(x0, y0, x1, y1, x2, y2, (f64) i, (f64) j)) {
        left = i;
        break;
      }

    for (i64 i = i1; i >= i0; --i)
      if (triangle_contains(x0, y0, x1, y1, x2, y2, (f64) i, (f64) j)) {
        right = i;
        break;
      }

    if (left  <  0)                  left  = 0;
    if (right >= brush.buffer.width) right = brush.buffer.width - 1;

    if (brush.xor_color)
      for (i64 i = left; i <= right; ++i)
        put_pixel_xor_(brush, i, j);
    else if (brush.alpha)
      for (i64 i = left; i <= right; ++i)
        put_pixel_alpha_(brush, i, j);
    else
      for (i64 i = left; i <= right; ++i)
        put_pixel_color_(brush, i, j);
  }
}

void fill_quad_(Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2, f64 x3, f64 y3) {
  fill_triangle_(brush, x0, y0, x1, y1, x2, y2);
  fill_triangle_(brush, x0, y0, x2, y2, x3, y3);
}

void fill_ellipse_(Brush brush, f64 x, f64 y, f64 width, f64 height) {
  f64 x0, y0, x1, y1;
  transform_box_(brush, x, y, width, height, &x0, &y0, &x1, &y1);

  i64 i0 = (i64) floor(x0);
  i64 i1 = (i64) ceil (x1);
  i64 j0 = (i64) floor(y0);
  i64 j1 = (i64) ceil (y1);

  if (j0 < 0)                    j0 = 0;
  if (j1 >= brush.buffer.height) j1 = brush.buffer.height - 1;

  for (i64 j = j0; j <= j1; ++j) {
    i64 left  = i1;
    i64 right = i0;

    for (i64 i = i0; i <= i1; ++i)
      if (ellipse_contains(x0, y0, x1 - x0, y1 - y0, (f64) i, (f64) j)) {
        left  = i;
        right = (i64) ceil(x0 + x1 - left); // symmetry
        break;
      }

    for (i64 i = right; i >= i0; --i)
      if (ellipse_contains(x0, y0, x1 - x0, y1 - y0, (f64) i, (f64) j)) {
        right = i;
        break;
      }

    if (left  <  0)                  left  = 0;
    if (right >= brush.buffer.width) right = brush.buffer.width - 1;

    if (brush.xor_color)
      for (i64 i = left; i <= right; ++i)
        put_pixel_xor_(brush, i, j);
    else if (brush.alpha)
      for (i64 i = left; i <= right; ++i)
        put_pixel_alpha_(brush, i, j);
    else
      for (i64 i = left; i <= right; ++i)
        put_pixel_color_(brush, i, j);
  }
}

static i64 char_column_offset_(c32 c, i64 column_index) {
  if (column_index < 0 || column_index >= CHAR_NUM_BITS_X_)
    return -1;
  return (c - 32) * CHAR_NUM_BITS_ + column_index * CHAR_NUM_BITS_Y_;
}

static b8 char_bit_(i64 column_offset, i64 row_index) {
  if (column_offset < 0 || row_index < 0 || row_index >= CHAR_NUM_BITS_Y_)
    return 0;

  i64 bit_index   = column_offset + row_index;
  i64 qword_index = bit_index / 64;
  if (qword_index < 0 || qword_index >= (i64) (sizeof _bitfont / sizeof *_bitfont))
    return 0;
  u64 mask        = 1ull << (bit_index % 64);
  return !!(_bitfont[qword_index] & mask);
}

static u64 char_column_convolved_(c32 c, i64 column_index) {
  if (column_index < 0 || column_index >= CHAR_NUM_BITS_X_)
    return 0;

  u64 column = 0;
  i64 offset = char_column_offset_(c, column_index);

  for (i64 y = 0; y < CHAR_NUM_BITS_Y_; ++y)
    if (char_bit_(offset, y))
      column |= 3ull << y;

  return column;
}

static b8 char_column_empty_(c32 c, i64 column_index) {
  if (column_index < 0 || column_index >= CHAR_NUM_BITS_X_)
    return 1;

  i64 offset = char_column_offset_(c, column_index);

  for (i64 y = 0; y < CHAR_NUM_BITS_Y_; ++y)
    if (char_bit_(offset, y))
      return 0;

  return 1;
}

static i64 char_width_(c32 c) {
  if (c < 32)
    return 0;
  if (c == ' ' || c > 127)
    return 4;

  i64 width = 0;

  for (; width < CHAR_NUM_BITS_X_; ++width)
    if (char_column_empty_(c, width) && char_column_empty_(c, width + 1))
      break;

  return width;
}

static i64 char_spacing_(i64 num_chars, c32 *text, i64 index) {
  if (text == NULL)
    return 0;

  if (index < 0 || index + 1 >= num_chars)
    return 0;

  u64 a = char_column_convolved_(text[index], char_width_(text[index]) - 1);
  u64 b = char_column_convolved_(text[index + 1], 0);

  if (!!(a & b))
    return 1;

  return 0;
}

static i64 text_cursor_(i64 num_chars, c32 *text) {
  if (text == NULL)
    return 0;

  i64 cursor = 0;

  for (i64 i = 0; i < num_chars; ++i) {
    if (text[i] <= ' ') {
      if (text[i] == '\n')
        cursor = 0;
      else if (text[i] == '\b' && i > 0)
        cursor -= char_width_(text[i - 1]) + char_spacing_(num_chars, text, i - 1);
      else if (text[i] == '\r')
        cursor = 0;
      else
        cursor += char_width_(' ') + char_spacing_(num_chars, text, i);
      continue;
    }
    cursor += char_width_(text[i]) + char_spacing_(num_chars, text, i);
  }

  return cursor;
}

static i64 enum_text_columns_(i64 num_chars, c32 *text) {
  if (text == NULL)
    return 0;

  i64 cols = 0;
  i64 n    = 0;

  for (i64 i = 0; i < num_chars; ++i) {
    if (text[i] <= ' ') {
      if (text[i] == '\n') {
        if (cols < n)
          cols = n;
        n = 0;
      } else if (text[i] == '\b' && i > 0) {
        if (cols < n)
          cols = n;
        n -= char_width_(text[i - 1]) + char_spacing_(num_chars, text, i - 1);
      } else if (text[i] == '\r') {
        if (cols < n)
          cols = n;
        n = 0;
      } else
        n += char_width_(' ') + char_spacing_(num_chars, text, i);
      continue;
    }
    n += char_width_(text[i]) + char_spacing_(num_chars, text, i);
  }

  if (cols < n)
    cols = n;

  return cols;
}

static i64 enum_text_rows_(i64 num_chars, c32 *text) {
  if (text == NULL)
    return 0;

  i64 rows = 0;

  for (i64 i = 0; i <= num_chars; ++i)
    if (i == num_chars || text[i] == '\n') {
      if (rows > 0)
        ++rows;
      rows += CHAR_NUM_BITS_Y_;
    }

  return rows;
}

static void draw_text_(Brush brush, f64 x0, f64 y0, f64 scale_x, f64 scale_y, i64 num_chars, c32 *text) {
  x0 = transform_x_(brush, x0);
  y0 = transform_y_(brush, y0);

  if (brush.scale.x < 0) scale_x *= -brush.scale.x;
  else                   scale_x *=  brush.scale.x;
  if (brush.scale.y < 0) scale_y *= -brush.scale.y;
  else                   scale_y *=  brush.scale.y;

  f64 kx = scale_x;
  f64 h  = scale_y * CHAR_NUM_BITS_Y_;

  if (h < EPSILON)
    return;

  f64 x = x0;
  f64 y = y0;

  for (i64 n = 0; n < num_chars; ++n) {
    if (text[n] <= ' ') {
      if (text[n] == '\n') {
        x  = x0;
        y += scale_y * (CHAR_NUM_BITS_Y_ + 1);
      }
      else if (text[n] == '\b' && n > 0)
        x -= kx * (char_width_(text[n - 1]) + char_spacing_(num_chars, text, n - 1));
      else if (text[n] == '\r')
        x  = x0;
      else
        x += kx * (char_width_(' ') + char_spacing_(num_chars, text, n));
      continue;
    }

    i64 num_cols = char_width_(text[n]);
    f64 w        = num_cols * kx;

    if (w >= EPSILON) {
      i64 i0 = (i64) floor(x);
      i64 i1 = (i64) ceil (x + w);
      i64 j0 = (i64) floor(y);
      i64 j1 = (i64) ceil (y + h);

      if (i0 <  0)                   i0 = 0;
      if (i1 >= brush.buffer.width)  i1 = brush.buffer.width - 1;
      if (j0 <  0)                   j0 = 0;
      if (j1 >= brush.buffer.height) j1 = brush.buffer.height - 1;

      f64 w_inv = 1. / w;
      f64 h_inv = 1. / h;

      for (i64 i = i0; i <= i1; ++i) {
        i64 column = (i64) floor(((i - x) * num_cols) * w_inv + .5);
        i64 offset = char_column_offset_(text[n], column);

        if (brush.xor_color)
          for (i64 j = j0; j <= j1; ++j) {
            i64 row = (i64) floor(((j - y) * CHAR_NUM_BITS_Y_) * h_inv + .5);
            if (char_bit_(offset, row))
               put_pixel_xor_(brush, i, j);
          }
        else if (brush.alpha)
          for (i64 j = j0; j <= j1; ++j) {
            i64 row = (i64) floor(((j - y) * CHAR_NUM_BITS_Y_) * h_inv + .5);
            if (char_bit_(offset, row))
              put_pixel_alpha_(brush, i, j);
          }
        else
          for (i64 j = j0; j <= j1; ++j) {
            i64 row = (i64) floor(((j - y) * CHAR_NUM_BITS_Y_) * h_inv + .5);
            if (char_bit_(offset, row))
               put_pixel_color_(brush, i, j);
          }
      }
    }

    x += kx * (num_cols + char_spacing_(num_chars, text, n));
  }
}

static Brush brush_defaults_(Brush b) {
  if (g_platform.frame_width > 0) {
    if (b.scale.x == 0.) b.scale.x = ((f64) g_platform.frame_width)  / g_platform.real_width;
    if (b.scale.y == 0.) b.scale.y = ((f64) g_platform.frame_height) / g_platform.real_height;
  }

  if (b.buffer.pixels == NULL)
    b.buffer = frame_pixels();

  if (b.buffer.line_size == 0)
    b.buffer.line_size = b.buffer.width;

  return b;
}

//  ================================================================

vec3_f32 vec3_from_vec4_f32(vec4_f32 v) {
  return (vec3_f32) {
    .x = v.x,
    .y = v.y,
    .z = v.z,
  };
}

vec4_f32 vec4_from_vec3_f32(vec3_f32 v, f32 w) {
  return (vec4_f32) {
    .x = v.x,
    .y = v.y,
    .z = v.z,
    .w = w,
  };
}

vec3_f32 vec3_f32_lerp(vec3_f32 a, vec3_f32 b, f32 t) {
  return (vec3_f32) {
    .x = a.x + (b.x - a.x) * t,
    .y = a.y + (b.y - a.y) * t,
    .z = a.z + (b.z - a.z) * t,
  };
}

vec4_f32 vec4_f32_lerp(vec4_f32 a, vec4_f32 b, f32 t) {
  return (vec4_f32) {
    .x = a.x + (b.x - a.x) * t,
    .y = a.y + (b.y - a.y) * t,
    .z = a.z + (b.z - a.z) * t,
    .w = a.w + (b.w - a.w) * t,
  };
}

//  ================================================================

vec3_f32 rgb_gamma_add(vec3_f32 rgb) {
  return (vec3_f32) {
    .x = (f32) gamma_(rgb.x),
    .y = (f32) gamma_(rgb.y),
    .z = (f32) gamma_(rgb.z),
  };
}

vec3_f32 rgb_gamma_remove(vec3_f32 rgb) {
  return (vec3_f32) {
    .x = (f32) gamma_re_(rgb.x),
    .y = (f32) gamma_re_(rgb.y),
    .z = (f32) gamma_re_(rgb.z),
  };
}

vec4_f32 rgba_gamma_add(vec4_f32 rgba) {
  return (vec4_f32) {
    .x = (f32) gamma_(rgba.x),
    .y = (f32) gamma_(rgba.y),
    .z = (f32) gamma_(rgba.z),
    .w = (f32) gamma_(rgba.w),
  };
}

vec4_f32 rgba_gamma_remove(vec4_f32 rgba) {
  return (vec4_f32) {
    .x = (f32) gamma_re_(rgba.x),
    .y = (f32) gamma_re_(rgba.y),
    .z = (f32) gamma_re_(rgba.z),
    .w = (f32) gamma_re_(rgba.w),
  };
}

vec3_f32 lab_from_rgb(vec3_f32 rgb) {
  f64 l = 0.4122214708 * rgb.x + 0.5363325363 * rgb.y + 0.0514459929 * rgb.z;
  f64 m = 0.2119034982 * rgb.x + 0.6806995451 * rgb.y + 0.1073969566 * rgb.z;
  f64 s = 0.0883024619 * rgb.x + 0.2817188376 * rgb.y + 0.6299787005 * rgb.z;

  f64 l_ = cbrt(l);
  f64 m_ = cbrt(m);
  f64 s_ = cbrt(s);

  return (vec3_f32) {
    .x = (f32) (0.2104542553 * l_ + 0.7936177850 * m_ - 0.0040720468 * s_),
    .y = (f32) (1.9779984951 * l_ - 2.4285922050 * m_ + 0.4505937099 * s_),
    .z = (f32) (0.0259040371 * l_ + 0.7827717662 * m_ - 0.8086757660 * s_),
  };
}

vec3_f32 rgb_from_lab(vec3_f32 lab) {
  f64 l_ = lab.x + 0.3963377774 * lab.y + 0.2158037573 * lab.z;
  f64 m_ = lab.x - 0.1055613458 * lab.y - 0.0638541728 * lab.z;
  f64 s_ = lab.x - 0.0894841775 * lab.y - 1.2914855480 * lab.z;

  f64 l = l_ * l_ * l_;
  f64 m = m_ * m_ * m_;
  f64 s = s_ * s_ * s_;

  return (vec3_f32) {
    .x = (f32) (+4.0767416621 * l - 3.3077115913 * m + 0.2309699292 * s),
    .y = (f32) (-1.2684380046 * l + 2.6097574011 * m - 0.3413193965 * s),
    .z = (f32) (-0.0041960863 * l - 0.7034186147 * m + 1.7076147010 * s),
  };
}

vec3_f32 lch_from_lab(vec3_f32 lab) {
  f64 a = lab.y;
  f64 b = lab.z;

  return (vec3_f32) {
    .x = lab.x,
    .y = (f32) sqrt (a * a + b * b),
    .z = (f32) atan2(b, a),
  };
}

vec3_f32 lab_from_lch(vec3_f32 lch) {
  return (vec3_f32) {
    .x = lch.x,
    .y = (f32) (lch.y * cos(lch.z)),
    .z = (f32) (lch.y * sin(lch.z)),
  };
}

vec3_f32 rgb_from_lch(vec3_f32 lch) {
  return rgb_from_lab(lab_from_lch(lch));
}

vec3_f32 lch_from_rgb(vec3_f32 rgb) {
  return lch_from_lab(lab_from_rgb(rgb));
}

vec3_f32 rgb_gamma_lerp (vec3_f32 a, vec3_f32 b, f32 t) {
  return rgb_gamma_add(vec3_f32_lerp(
    rgb_gamma_remove(a),
    rgb_gamma_remove(b),
    t
  ));
}

vec4_f32 rgba_gamma_lerp(vec4_f32 a, vec4_f32 b, f32 t) {
  return rgba_gamma_add(vec4_f32_lerp(
    rgba_gamma_remove(a),
    rgba_gamma_remove(b),
    t
  ));
}

vec3_f32 rgb_mix(vec3_f32 a, vec3_f32 b, f32 t) {
  vec3_f32 a_lch = lch_from_rgb(a);
  vec3_f32 b_lch = lch_from_rgb(b);

  return rgb_from_lch((vec3_f32) {
    a_lch.x + (b_lch.x - a_lch.x) * t,
    a_lch.y + (b_lch.y - a_lch.y) * t,
    lch_from_rgb(rgb_gamma_lerp(a, b, t)).z,
  });
}

vec4_f32 rgba_mix(vec4_f32 a, vec4_f32 b, f32 t) {
  vec3_f32 a_lch = lch_from_rgb(vec3_from_vec4_f32(a));
  vec3_f32 b_lch = lch_from_rgb(vec3_from_vec4_f32(b));
  vec4_f32 ab    = rgba_gamma_lerp(a, b, t);

  return vec4_from_vec3_f32(rgb_from_lch((vec3_f32) {
    a_lch.x + (b_lch.x - a_lch.x) * t,
    a_lch.y + (b_lch.y - a_lch.y) * t,
    lch_from_rgb(vec3_from_vec4_f32(ab)).z,
  }), ab.w);
}

vec3_f32 lch_mix(vec3_f32 a, vec3_f32 b, f32 t) {
  return (vec3_f32) {
    a.x + (b.x - a.x) * t,
    a.y + (b.y - a.y) * t,
    lch_from_rgb(rgb_gamma_lerp(rgb_from_lch(a), rgb_from_lch(b), t)).z,
  };
}

vec3_f32 lch_lerp(vec3_f32 a, vec3_f32 b, f32 t) {
  f32 delta_hue = b.z - a.z;
  if (delta_hue >  M_PI) delta_hue -= M_PI * 2;
  if (delta_hue < -M_PI) delta_hue += M_PI * 2;

  return (vec3_f32) {
    .x = a.x + (b.x - a.x) * t,
    .y = a.y + (b.y - a.y) * t,
    .z = a.z +  delta_hue  * t,
  };
}

u32 rgb_u32_from_f32(vec3_f32 color) {
  i32 ir = (i32) floor(color.x * 255. + .5);
  i32 ig = (i32) floor(color.y * 255. + .5);
  i32 ib = (i32) floor(color.z * 255. + .5);

  u32 r = ir < 0 ? 0u : ir > 255 ? 255u : (u32) ir;
  u32 g = ig < 0 ? 0u : ig > 255 ? 255u : (u32) ig;
  u32 b = ib < 0 ? 0u : ib > 255 ? 255u : (u32) ib;

  return (r << 16) | (g << 8) | b;
}

vec3_f32 rgb_f32_from_u32(u32 color) {
  return (vec3_f32) {
    .x = ((color & 0xff0000) >> 16) / 255.f,
    .y = ((color & 0x00ff00) >>  8) / 255.f,
    .z =  (color & 0x0000ff)        / 255.f,
  };
}

u32 rgba_u32_from_f32(vec4_f32 color) {
  i32 ir = (i32) floor(color.x * 255. + .5);
  i32 ig = (i32) floor(color.y * 255. + .5);
  i32 ib = (i32) floor(color.z * 255. + .5);
  i32 ia = (i32) floor(color.w * 255. + .5);

  u32 r = ir < 0 ? 0u : ir > 255 ? 255u : (u32) ir;
  u32 g = ig < 0 ? 0u : ig > 255 ? 255u : (u32) ig;
  u32 b = ib < 0 ? 0u : ib > 255 ? 255u : (u32) ib;
  u32 a = ia < 0 ? 0u : ia > 255 ? 255u : (u32) ia;

  return (a << 24) | (r << 16) | (g << 8) | b;
}

vec4_f32 rgba_f32_from_u32(u32 color) {
  return (vec4_f32) {
    .x = ((color & 0x00ff0000) >> 16) / 255.f,
    .y = ((color & 0x0000ff00) >>  8) / 255.f,
    .z =  (color & 0x000000ff)        / 255.f,
    .w = ((color & 0xff000000) >> 24) / 255.f,
  };
}

b8 rectangle_contains(f64 x0, f64 y0, f64 width, f64 height, f64 px, f64 py) {
  return px >= x0 && px < x0 + width && py >= y0 && py < y0 + height;
}

b8 triangle_contains(f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2, f64 px, f64 py) {
  //  Z-components of cross-products
  //

  f64 x10 = x1 - x0;
  f64 x21 = x2 - x1;
  f64 x02 = x0 - x2;

  f64 y10 = y1 - y0;
  f64 y21 = y2 - y1;
  f64 y02 = y0 - y2;

  f64 z0  = x02 * y10 - x10 * y02;
  f64 z1  = x10 * y21 - x21 * y10;
  f64 z2  = x21 * y02 - x02 * y21;

  f64 pz0 = x02 * (py - y0) - (px - x0) * y02;
  f64 pz1 = x10 * (py - y1) - (px - x1) * y10;
  f64 pz2 = x21 * (py - y2) - (px - x2) * y21;

  return same_sign_(z0, pz0)
      && same_sign_(z1, pz1)
      && same_sign_(z2, pz2);
}

b8 quad_contains(f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2, f64 x3, f64 y3, f64 px, f64 py) {
  return triangle_contains(x0, y0, x1, y1, x2, y2, px, py)
      || triangle_contains(x0, y0, x2, y2, x3, y3, px, py);
}

b8 ellipse_contains(f64 x0, f64 y0, f64 width, f64 height, f64 px, f64 py) {
  f64 dw = width  / 2;
  f64 dh = height / 2;

  if (dw < EPSILON || dh < EPSILON)
    return 0;

  f64 cx = x0 + dw;
  f64 cy = y0 + dh;
  f64 kx = 1. / dw;
  f64 ky = 1. / dh;

  f64 dx = (px - cx) * kx;
  f64 dy = (py - cy) * ky;

  return dx * dx + dy * dy - 1.0 < EPSILON;
}

b8 line_contains(f64 x0, f64 y0, f64 x1, f64 y1, f64 width, f64 px, f64 py) {
  f64 dx = x1 - x0;
  f64 dy = y1 - y0;

  //  Tangent
  //
  f64 tx = -dy;
  f64 ty = dx;
  f64 tl = sqrt(tx * tx + ty * ty);
  if (tl >= EPSILON) {
    tx /= tl;
    ty /= tl;
  }
  tx *= width * .5;
  ty *= width * .5;

  return
       triangle_contains(x0 - tx, y0 - ty, x0 + tx, y0 + ty, x1 + tx, y1 + ty, px, py)
    || triangle_contains(x0 - tx, y0 - ty, x1 + tx, y1 + ty, x1 - tx, y1 - ty, px, py);
}

Pixel_Buffer frame_pixels(void) {
  return (Pixel_Buffer) {
    .width     = g_platform.frame_width,
    .height    = g_platform.frame_height,
    .line_size = g_platform.frame_width,
    .pixels    = g_platform.pixels,
  };
}

Pixel_Buffer sketch_pixels(i64 width, i64 height) {
  if (width <= 0 || height <= 0) {
    width  = g_platform.frame_width;
    height = g_platform.frame_height;
  }

  if (width * height > MAX_NUM_PIXELS) {
    width  = 0;
    height = 0;
  }

  return (Pixel_Buffer) {
    .width     = width,
    .height    = height,
    .line_size = width,
    .pixels    = g_platform.sketch,
  };
}

Pixel_Buffer subimage(Pixel_Buffer image, i64 x, i64 y, i64 width, i64 height) {
  if (x < 0 || y < 0 || width <= 0 || height <= 0 || x + width > image.width || y + height > image.height || image.pixels == NULL)
    return (Pixel_Buffer) {
      .width     = 0,
      .height    = 0,
      .line_size = 0,

      // Set pixels pointer to prevent default initialization.
      .pixels = g_platform.sketch,
    };

  return (Pixel_Buffer) {
    .width     = width,
    .height    = height,
    .line_size = image.line_size,
    .pixels    = image.pixels + (y * image.line_size + x),
  };
}

void put_pixel(Brush brush, i64 x, i64 y) {
  brush = brush_defaults_(brush);

  if (x < 0 || x >= brush.buffer.width || y < 0 || y >= brush.buffer.height)
    return;

  put_pixel_(brush, x, y);
}

void draw_pixels(Brush brush, f64 x, f64 y, f64 width, f64 height, Pixel_Buffer src) {
  if (width < EPSILON || height < EPSILON || src.width <= 0 || src.height <= 0 || src.line_size <= 0 || src.pixels == NULL)
    return;

  draw_pixels_(brush_defaults_(brush), x, y, width, height, src);
}

void fill_rectangle(Brush brush, f64 x, f64 y, f64 width, f64 height) {
  fill_rectangle_(brush_defaults_(brush), x, y, width, height);
}

void fill_triangle(Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2) {
  brush = brush_defaults_(brush);

  if (!brush.quick) {
    Brush aa = antialiasing_brush_(brush);

    f64 x      = min3_(x0, x1, x2);
    f64 y      = min3_(y0, y1, y2);
    f64 width  = max3_(x0, x1, x2) - x;
    f64 height = max3_(y0, y1, y2) - y;

    Brush clear = aa;
    clear.color = vec4_from_vec3_f32(vec3_from_vec4_f32(brush.color), 0.f);
    fill_rectangle_(clear, 0, 0, width, height);

    fill_triangle_(aa, x0 - x, y0 - y, x1 - x, y1 - y, x2 - x, y2 - y);

    f64 x0, y0, x1, y1;
    transform_box_(aa, 0, 0, width, height, &x0, &y0, &x1, &y1);

    i64 i0 = (i64) floor(x0);
    i64 i1 = (i64) ceil (x1);
    i64 j0 = (i64) floor(y0);
    i64 j1 = (i64) ceil (y1);

    draw_pixels_from_(
      copy_brush_(brush),
      x, y, width, height,
      aa.buffer, i0, j0, i1 - i0, j1 - j0
    );
  } else
    fill_triangle_(brush, x0, y0, x1, y1, x2, y2);
}

void fill_quad(Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 x2, f64 y2, f64 x3, f64 y3) {
  brush = brush_defaults_(brush);

  if (!brush.quick) {
    Brush aa = antialiasing_brush_(brush);

    f64 x      = min4_(x0, x1, x2, x3);
    f64 y      = min4_(y0, y1, y2, y3);
    f64 width  = max4_(x0, x1, x2, x3) - x;
    f64 height = max4_(y0, y1, y2, y3) - y;

    Brush clear = aa;
    clear.color = vec4_from_vec3_f32(vec3_from_vec4_f32(brush.color), 0.f);
    fill_rectangle_(clear, 0, 0, width, height);

    fill_quad_(aa, x0 - x, y0 - y, x1 - x, y1 - y, x2 - x, y2 - y, x3 - x, y3 - y);

    f64 x0, y0, x1, y1;
    transform_box_(aa, 0, 0, width, height, &x0, &y0, &x1, &y1);

    i64 i0 = (i64) floor(x0);
    i64 i1 = (i64) ceil (x1);
    i64 j0 = (i64) floor(y0);
    i64 j1 = (i64) ceil (y1);

    draw_pixels_from_(
      copy_brush_(brush),
      x, y, width, height,
      aa.buffer, i0, j0, i1 - i0, j1 - j0
    );
  } else
    fill_quad_(brush, x0, y0, x1, y1, x2, y2, x3, y3);
}

void fill_ellipse(Brush brush, f64 x, f64 y, f64 width, f64 height) {
  //  FIXME PERF:
  //  Implement better algorithm.

  if (width < EPSILON || height < EPSILON)
    return;

  brush = brush_defaults_(brush);

  if (!brush.quick) {
    Brush aa = antialiasing_brush_(brush);

    Brush clear = aa;
    clear.color = vec4_from_vec3_f32(vec3_from_vec4_f32(brush.color), 0.f);
    fill_rectangle_(clear, 0, 0, width, height);

    fill_ellipse_(aa, 0, 0, width, height);

    f64 x0, y0, x1, y1;
    transform_box_(aa, 0, 0, width, height, &x0, &y0, &x1, &y1);

    i64 i0 = (i64) floor(x0);
    i64 i1 = (i64) ceil (x1);
    i64 j0 = (i64) floor(y0);
    i64 j1 = (i64) ceil (y1);

    draw_pixels_from_(
      copy_brush_(brush),
      x, y, width, height,
      aa.buffer, i0, j0, i1 - i0, j1 - j0
    );
  } else
    fill_ellipse_(brush, x, y, width, height);
}

void fill_line(Brush brush, f64 x0, f64 y0, f64 x1, f64 y1, f64 width) {
  if (width < EPSILON)
    return;

  f64 dx = x1 - x0;
  f64 dy = y1 - y0;

  //  Tangent
  //
  f64 tx = -dy;
  f64 ty =  dx;
  f64 tl = sqrt(tx * tx + ty * ty);

  if (tl < EPSILON)
    return;

  f64 k = .5 / tl;
  tx *= width * k;
  ty *= width * k;

  fill_quad(brush, x0 - tx, y0 - ty, x0 + tx, y0 + ty, x1 + tx, y1 + ty, x1 - tx, y1 - ty);
}

void draw_text_area(Brush brush, f64 x, f64 y, f64 width, f64 height, f64 max_scale_x, f64 max_scale_y, i64 num_chars, c32 *text) {
  if (text == NULL || num_chars <= 0 || max_scale_x < EPSILON || max_scale_y < EPSILON)
    return;

  i64 num_columns = enum_text_columns_(num_chars, text);
  i64 num_rows    = enum_text_rows_(num_chars, text);

  f64 scale_x = width  / num_columns;
  f64 scale_y = height / num_rows;

  f64 kx = scale_x / max_scale_x;
  f64 ky = scale_y / max_scale_y;

  f64 k = kx < ky ? kx : ky;

  kx = k * max_scale_x;
  ky = k * max_scale_y;

  brush = brush_defaults_(brush);

  if (!brush.quick) {
    Brush aa = antialiasing_brush_(brush);

    Brush clear = aa;
    clear.color = vec4_from_vec3_f32(vec3_from_vec4_f32(brush.color), 0.f);
    fill_rectangle_(clear, 0, 0, width, height);

    draw_text_(aa, 0, 0, kx, ky, num_chars, text);

    f64 x0, y0, x1, y1;
    transform_box_(aa, 0, 0, width, height, &x0, &y0, &x1, &y1);

    i64 i0 = (i64) floor(x0);
    i64 i1 = (i64) ceil (x1);
    i64 j0 = (i64) floor(y0);
    i64 j1 = (i64) ceil (y1);

    draw_pixels_from_(
      copy_brush_(brush),
      x, y, width, height,
      aa.buffer, i0, j0, i1 - i0, j1 - j0
    );
  } else
    draw_text_(brush, x, y, kx, ky, num_chars, text);
}

void draw_text_cursor(Brush brush, f64 x, f64 y, f64 width, f64 height, f64 max_scale_x, f64 max_scale_y, i64 cursor, i64 selection, i64 num_chars, c32 *text) {
  if (max_scale_x < EPSILON || max_scale_y < EPSILON)
    return;

  i64 num_columns = enum_text_columns_(num_chars, text);
  i64 num_rows    = enum_text_rows_(num_chars, text);
  i64 cursor_x    = text_cursor_(cursor, text);
  i64 cursor_y    = enum_text_rows_(cursor, text);

  f64 scale_x = width  / num_columns;
  f64 scale_y = height / num_rows;

  f64 kx = scale_x / max_scale_x;
  f64 ky = scale_y / max_scale_y;

  f64 k = kx < ky ? kx : ky;

  kx = k * max_scale_x;
  ky = k * max_scale_y;

  brush = brush_defaults_(brush);

  if (selection != 0) {
    i64 selection_x, selection_y;

    if (selection > 0) {
      selection_x = text_cursor_(cursor + selection, text);
      selection_y = enum_text_rows_(cursor + selection, text);
    } else {
      selection_x = cursor_x;
      selection_y = cursor_y;
      cursor_x    = text_cursor_(cursor + selection, text);
      cursor_y    = enum_text_rows_(cursor + selection, text);
    }

    if (cursor_y == selection_y)
      fill_rectangle_(
        brush,
        x + kx * cursor_x,
        y + ky * cursor_y - ky * (CHAR_NUM_BITS_Y_ + 1),
        kx * (selection_x - cursor_x),
        ky * (CHAR_NUM_BITS_Y_ + 1)
      );
    else {
      fill_rectangle_(
        brush,
        x + kx * cursor_x,
        y + ky * cursor_y - ky * (CHAR_NUM_BITS_Y_ + 1),
        kx * (num_columns - cursor_x),
        ky * (CHAR_NUM_BITS_Y_ + 1)
      );
      for (i64 j = cursor_y + CHAR_NUM_BITS_Y_ + 1; j < selection_y; j += CHAR_NUM_BITS_Y_ + 1)
        fill_rectangle_(
          brush,
          x,
          y + ky * j - ky * (CHAR_NUM_BITS_Y_ + 1),
          kx * num_columns,
          ky * (CHAR_NUM_BITS_Y_ + 1)
        );
      fill_rectangle_(
        brush,
        x,
        y + ky * selection_y - ky * (CHAR_NUM_BITS_Y_ + 1),
        kx * selection_x,
        ky * (CHAR_NUM_BITS_Y_ + 1)
      );
    }
  } else
    fill_rectangle_(
      brush,
      x + kx * cursor_x,
      y + ky * cursor_y - ky * CHAR_NUM_BITS_Y_,
      kx * .5,
      ky * (CHAR_NUM_BITS_Y_ - 1)
    );
}

//  ================================================================
//
//    Test suite
//
//  ================================================================

#ifdef GRAPHICS_TEST_SUITE

#define TEST_FILE graphics
#include "test.c"

TEST("colors") {
  REQUIRE_EQ(rgb_u32_from_f32((vec3_f32) { 1.f, 0.f, 0.f, }), 0xff0000);
  REQUIRE_EQ(rgb_u32_from_f32((vec3_f32) { 0.f, 1.f, 0.f, }), 0x00ff00);
  REQUIRE_EQ(rgb_u32_from_f32((vec3_f32) { 0.f, 0.f, 1.f, }), 0x0000ff);

  REQUIRE_EQ(rgb_u32_from_f32((vec3_f32) { .5f, 0.f, 0.f, }), 0x800000);
  REQUIRE_EQ(rgb_u32_from_f32((vec3_f32) { 0.f, .5f, 0.f, }), 0x008000);
  REQUIRE_EQ(rgb_u32_from_f32((vec3_f32) { 0.f, 0.f, .5f, }), 0x000080);

  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 1.f, 0.f, 0.f, 0.f, }), 0x00ff0000);
  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 0.f, 1.f, 0.f, 0.f, }), 0x0000ff00);
  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 0.f, 0.f, 1.f, 0.f, }), 0x000000ff);
  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 0.f, 0.f, 0.f, 1.f, }), 0xff000000);

  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { .5f, 0.f, 0.f, 0.f, }), 0x00800000);
  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 0.f, .5f, 0.f, 0.f, }), 0x00008000);
  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 0.f, 0.f, .5f, 0.f, }), 0x00000080);
  REQUIRE_EQ(rgba_u32_from_f32((vec4_f32) { 0.f, 0.f, 0.f, .5f, }), 0x80000000);

  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 0.f, 0.f, }).x * 100,   0);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 1.f, 1.f, 1.f, }).x * 100, 100);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 1.f, 0.f, 0.f, }).x * 100,  62);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 1.f, 0.f, }).x * 100,  86);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 0.f, 1.f, }).x * 100,  45);

  REQUIRE_EQ(lch_from_rgb((vec3_f32) { .5f, .5f, .5f, }).y * 100,  0);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 1.f, 0.f, 0.f, }).y * 100, 25);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 1.f, 0.f, }).y * 100, 29);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 0.f, 1.f, }).y * 100, 31);

  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 1.f, 0.f, 0.f, }).z * 100,   51);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 1.f, 0.f, }).z * 100,  248);
  REQUIRE_EQ(lch_from_rgb((vec3_f32) { 0.f, 0.f, 1.f, }).z * 100, -167);

  vec3_f32 white = rgb_from_lch(lch_from_rgb((vec3_f32) { 1.f, 1.f, 1.f, }));
  REQUIRE_EQ(white.x * 100, 100);
  REQUIRE_EQ(white.y * 100, 100);
  REQUIRE_EQ((white.z + 3e-7) * 100, 100);

  vec3_f32 black = rgb_from_lch(lch_from_rgb((vec3_f32) { 0.f, 0.f, 0.f, }));
  REQUIRE_EQ(black.x * 100, 0);
  REQUIRE_EQ(black.y * 100, 0);
  REQUIRE_EQ(black.z * 100, 0);

  vec3_f32 grey = rgb_from_lch(lch_from_rgb((vec3_f32) { .5f, .5f, .5f, }));
  REQUIRE_EQ(grey.x * 100, 50);
  REQUIRE_EQ((grey.y + 3e-8) * 100, 50);
  REQUIRE_EQ((grey.z + 3e-7) * 100, 50);

  vec3_f32 red = rgb_from_lch(lch_from_rgb((vec3_f32) { 1.f, 0.f, 0.f, }));
  REQUIRE_EQ(red.x * 100, 100);
  REQUIRE_EQ(red.y * 100,   0);
  REQUIRE_EQ(red.z * 100,   0);

  vec3_f32 green = rgb_from_lch(lch_from_rgb((vec3_f32) { 0.f, 1.f, 0.f, }));
  REQUIRE_EQ(green.x * 100,   0);
  REQUIRE_EQ(green.y * 100, 100);
  REQUIRE_EQ(green.z * 100,   0);

  vec3_f32 blue = rgb_from_lch(lch_from_rgb((vec3_f32) { 0.f, 0.f, 1.f, }));
  REQUIRE_EQ(blue.x * 100, 0);
  REQUIRE_EQ(blue.y * 100, 0);
  REQUIRE_EQ((blue.z + 6e-8) * 100, 100);

  vec3_f32 yellow = rgb_from_lch(lch_from_rgb((vec3_f32) { 1.f, 1.f, 0.f, }));
  REQUIRE_EQ(yellow.x * 100, 100);
  REQUIRE_EQ((yellow.y + 6e-8) * 100, 100);
  REQUIRE_EQ(yellow.z * 100,   0);

  vec3_f32 cyan = rgb_from_lch(lch_from_rgb((vec3_f32) { 0.f, 1.f, 1.f, }));
  REQUIRE_EQ(cyan.x * 100,   0);
  REQUIRE_EQ((cyan.y + 2e-7) * 100, 100);
  REQUIRE_EQ((cyan.z + 2e-7) * 100, 100);

  vec3_f32 purple = rgb_from_lch(lch_from_rgb((vec3_f32) { 1.f, 0.f, 1.f, }));
  REQUIRE_EQ(purple.x * 100, 100);
  REQUIRE_EQ(purple.y * 100, 0);
  REQUIRE_EQ((purple.z + 2e-7) * 100, 100);
}

Brush brush = {
  .buffer = {
    .width     = 1280,
    .height    = 720,
    .line_size = 1280,
    .pixels    = g_platform.pixels,
  },
  .position = { 0.f, 0.f, },
  .scale    = { 1.f, 1.f, },
  .color    = { 1.f, 1.f, 1.f, 1.f },
};

BENCHMARK("fill rectangle") {
  BENCHMARK_BEGIN;
  {
    fill_rectangle(brush, 100, 100, 300, 200);
  }
  BENCHMARK_END;
}

BENCHMARK("fill triangle") {
  BENCHMARK_BEGIN;
  {
    fill_triangle(brush, 100, 100, 300, 100, 200, 250);
  }
  BENCHMARK_END;
}

BENCHMARK("fill quad") {
  BENCHMARK_BEGIN;
  {
    fill_quad(brush, 100, 100, 300, 100, 300, 200, 100, 200);
  }
  BENCHMARK_END;
}

BENCHMARK("fill ellipse") {
  BENCHMARK_BEGIN;
  {
    fill_ellipse(brush, 80, 80, 340, 240);
  }
  BENCHMARK_END;
}

BENCHMARK("fill line") {
  BENCHMARK_BEGIN;
  {
    fill_line(brush, 100, 100, 300, 200, 40);
  }
  BENCHMARK_END;
}

BENCHMARK("draw text area") {
  BENCHMARK_BEGIN;
  {
    c32 text[] = { 'H', 'e', 'l', 'l', 'o', ',', ' ', 'S', 'a', 'i', 'l', 'o', 'r', '!', };
    draw_text_area(brush, 100, 100, 300, 200, 100, 200, sizeof text / sizeof *text, text);
  }
  BENCHMARK_END;
}

void update_and_render_frame(void) {}

i32 main(i32 argc, c8 **argv) {
  return run_tests_and_benchmarks(argc, argv);
}

#undef TEST_FILE
#endif // GRAPHICS_TEST_SUITE

//  ================================================================

#endif // GRAPHICS_IMPL_GUARD_
#endif // GRAPHICS_HEADER