uniform float4x4 ViewProj;
uniform texture2d image;
uniform texture2d colorLUT;
uniform float multiplier;
uniform float sdrWhiteLevel;
uniform int outputTransfer; /* 0: sRGB, 1: scRGB, 2: PQ decoded to scRGB */
uniform bool matrixEnabled;
uniform bool lutEnabled;
uniform float4 colorMatrix0;
uniform float4 colorMatrix1;
uniform float4 colorMatrix2;
uniform float colorScalar;

sampler_state def_sampler
{
  Filter   = Linear;
  AddressU = Clamp;
  AddressV = Clamp;
};

sampler_state lut_sampler
{
  Filter   = Point;
  AddressU = Clamp;
  AddressV = Clamp;
};

struct VertData
{
  float4 pos : POSITION;
  float2 uv  : TEXCOORD0;
};

VertData VSDefault(VertData vert_in)
{
  VertData vert_out;
  vert_out.pos = mul(float4(vert_in.pos.xyz, 1.0), ViewProj);
  vert_out.uv  = vert_in.uv;
  return vert_out;
}

float srgbToLinearChannel(float value)
{
  return value <= 0.04045 ? value / 12.92 :
    pow((value + 0.055) / 1.055, 2.4);
}

float3 srgbToLinear(float3 color)
{
  return float3(
    srgbToLinearChannel(color.r),
    srgbToLinearChannel(color.g),
    srgbToLinearChannel(color.b));
}

float3 pow3(float3 value, float exponent)
{
  return pow(value, float3(exponent, exponent, exponent));
}

float3 linearToSrgb(float3 value)
{
  float3 low = value * 12.92;
  float3 high = 1.055 * pow3(max(value, 0.0), 1.0 / 2.4) - 0.055;
  return lerp(low, high, step(0.0031308, value));
}

float3 bt709ToXYZ(float3 rgb)
{
  return float3(
    dot(rgb, float3(0.4123908, 0.3575843, 0.1804808)),
    dot(rgb, float3(0.2126390, 0.7151687, 0.0721923)),
    dot(rgb, float3(0.0193308, 0.1191948, 0.9505322)));
}

float3 xyzToBT709(float3 xyz)
{
  return float3(
    dot(xyz, float3( 3.2409699, -1.5373832, -0.4986108)),
    dot(xyz, float3(-0.9692436,  1.8759675,  0.0415551)),
    dot(xyz, float3( 0.0556301, -0.2039770,  1.0569715)));
}

float3 xyzToBT2020(float3 xyz)
{
  return float3(
    dot(xyz, float3( 1.7166512, -0.3556708, -0.2533663)),
    dot(xyz, float3(-0.6666844,  1.6164812,  0.0157685)),
    dot(xyz, float3( 0.0176399, -0.0427706,  0.9421031)));
}

float3 bt709ToBT2020(float3 rgb)
{
  return xyzToBT2020(bt709ToXYZ(rgb));
}

float3 bt2020ToBT709(float3 rgb)
{
  return float3(
    dot(rgb, float3( 1.6604910, -0.5876411, -0.0728499)),
    dot(rgb, float3(-0.1245505,  1.1328999, -0.0083494)),
    dot(rgb, float3(-0.0181508, -0.1005789,  1.1187297)));
}

float3 linearToPQ(float3 value)
{
  const float m1 = 0.1593017578125;
  const float m2 = 78.84375;
  const float c1 = 0.8359375;
  const float c2 = 18.8515625;
  const float c3 = 18.6875;
  float3 p = pow3(max(value, 0.0), m1);
  return pow3((c1 + c2 * p) / (1.0 + c3 * p), m2);
}

float3 pqToLinear(float3 value)
{
  const float m1 = 0.1593017578125;
  const float m2 = 78.84375;
  const float c1 = 0.8359375;
  const float c2 = 18.8515625;
  const float c3 = 18.6875;
  float3 p = pow3(max(value, 0.0), 1.0 / m2);
  return pow3(max(p - c1, 0.0) / max(c2 - c3 * p, 0.000001),
      1.0 / m1);
}

float3 applyLUT(float3 value)
{
  float3 pos = clamp(value, 0.0, 1.0) * 4095.0;
  float3 lo = floor(pos);
  float3 hi = min(lo + 1.0, 4095.0);
  float3 f = frac(pos);
  float2 rlo = float2((lo.r + 0.5) / 4096.0, 0.5);
  float2 rhi = float2((hi.r + 0.5) / 4096.0, 0.5);
  float2 glo = float2((lo.g + 0.5) / 4096.0, 0.5);
  float2 ghi = float2((hi.g + 0.5) / 4096.0, 0.5);
  float2 blo = float2((lo.b + 0.5) / 4096.0, 0.5);
  float2 bhi = float2((hi.b + 0.5) / 4096.0, 0.5);
  return float3(
    lerp(colorLUT.Sample(lut_sampler, rlo).r,
         colorLUT.Sample(lut_sampler, rhi).r, f.r),
    lerp(colorLUT.Sample(lut_sampler, glo).g,
         colorLUT.Sample(lut_sampler, ghi).g, f.g),
    lerp(colorLUT.Sample(lut_sampler, blo).b,
         colorLUT.Sample(lut_sampler, bhi).b, f.b));
}

float4 PSCursor(VertData vert_in) : TARGET
{
  float4 pixel = image.Sample(def_sampler, vert_in.uv);

  if (pixel.a > 0.0)
  {
    /* Cursor RGB is premultiplied in nonlinear sRGB. Recover the straight
     * color before decoding it, then premultiply again in linear light. */
    float3 srgb = clamp(pixel.rgb / pixel.a, 0.0, 1.0);
    float3 value = srgbToLinear(srgb);
    if (matrixEnabled)
    {
      float3 xyz = bt709ToXYZ(value);
      xyz = float3(
        dot(float4(xyz, 1.0), colorMatrix0),
        dot(float4(xyz, 1.0), colorMatrix1),
        dot(float4(xyz, 1.0), colorMatrix2)) * colorScalar;
      value = outputTransfer == 2 ? xyzToBT2020(xyz) : xyzToBT709(xyz);
    }
    else if (outputTransfer == 2)
      value = bt709ToBT2020(value);

    if (outputTransfer == 2)
    {
      value = linearToPQ(value * (sdrWhiteLevel / 10000.0));
      if (lutEnabled)
        value = applyLUT(value);
      value = bt2020ToBT709(pqToLinear(value)) * 125.0;
    }
    else if (outputTransfer == 1)
    {
      value *= multiplier;
      if (lutEnabled)
        value = applyLUT(value);
    }
    else
    {
      value = linearToSrgb(value);
      if (lutEnabled)
        value = applyLUT(value);
    }

    pixel.rgb = value * pixel.a;
  }
  else
    pixel.rgb = float3(0.0, 0.0, 0.0);

  return pixel;
}

technique Draw
{
  pass
  {
    vertex_shader = VSDefault(vert_in);
    pixel_shader  = PSCursor(vert_in);
  }
}