#version 450
// license:BSD-3-Clause
// copyright-holders:Ryan Holtz,ImJezze
// Adapted from mame_ntsc_decode.slang for SDL3 GPU / Vulkan SPIRV

layout(location=0) in vec2 v_uv;
layout(location=0) out vec4 FragColor;

layout(set=2, binding=0) uniform sampler2D Source;

layout(set=3, binding=0) uniform NTSCParams {
    float source_width;
    float source_height;
    float a_value;
    float b_value;
    float cc_value;
    float scan_time;
    float notch_width;
    float y_freq;
    float i_freq;
    float q_freq;
    float _pad0;
    float _pad1;
} u;

const float PI  = 3.1415927;
const float PI2 = PI * 2.0;

const vec3 RDot = vec3(1.0,  0.956,  0.621);
const vec3 GDot = vec3(1.0, -0.272, -0.647);
const vec3 BDot = vec3(1.0, -1.106,  1.703);

const vec4 NotchOffset = vec4(0.0, 1.0, 2.0, 3.0);
const int SampleCount     = 64;
const int HalfSampleCount = 32;

void main() {
    vec2 source_dims = vec2(u.source_width, u.source_height);
    vec4 BaseTexel   = texture(Source, v_uv);

    float CCValue        = u.cc_value;
    float ScanTime       = u.scan_time;
    float NotchHalfWidth = u.notch_width / 2.0;
    float YFreqResponse  = u.y_freq;
    float IFreqResponse  = u.i_freq;
    float QFreqResponse  = u.q_freq;
    float AValue         = u.a_value;
    float BValue         = u.b_value;

    float TimePerSample = ScanTime / (source_dims.x * 4.0);

    float Fc_y1 = (CCValue - NotchHalfWidth) * TimePerSample;
    float Fc_y2 = (CCValue + NotchHalfWidth) * TimePerSample;
    float Fc_y3 = YFreqResponse * TimePerSample;
    float Fc_i  = IFreqResponse * TimePerSample;
    float Fc_q  = QFreqResponse * TimePerSample;

    float Fc_i_2    = Fc_i  * 2.0;
    float Fc_q_2    = Fc_q  * 2.0;
    float Fc_y1_2   = Fc_y1 * 2.0;
    float Fc_y2_2   = Fc_y2 * 2.0;
    float Fc_y3_2   = Fc_y3 * 2.0;
    float Fc_i_pi2  = Fc_i  * PI2;
    float Fc_q_pi2  = Fc_q  * PI2;
    float Fc_y1_pi2 = Fc_y1 * PI2;
    float Fc_y2_pi2 = Fc_y2 * PI2;
    float Fc_y3_pi2 = Fc_y3 * PI2;
    float PI2Length = PI2 / float(SampleCount);

    float W    = PI2 * CCValue * ScanTime;
    float WoPI = W / PI;

    float HOffset = BValue / WoPI;
    float VScale  = AValue * source_dims.y / WoPI;

    vec4 YAccum = vec4(0.0);
    vec4 IAccum = vec4(0.0);
    vec4 QAccum = vec4(0.0);

    vec4 Cy        = vec4(v_uv.y);
    vec4 VPosition = Cy;

    for (float i = 0.0; i < float(SampleCount); i += 4.0) {
        float n  = i - float(HalfSampleCount);
        vec4  n4 = n + NotchOffset;

        vec4 Cx        = vec4(v_uv.x) + (n4 * 0.25) / source_dims.x;
        vec4 HPosition = Cx;

        vec4 C  = texture(Source, vec2(Cx.r, Cy.r));
        vec4 T  = HPosition + vec4(HOffset) + VPosition * vec4(VScale);
        vec4 WT = W * T;

        vec4 SincKernel = 0.54 + 0.46 * cos(PI2Length * n4);

        vec4 SincYIn1 = Fc_y1_pi2 * n4;
        vec4 SincYIn2 = Fc_y2_pi2 * n4;
        vec4 SincYIn3 = Fc_y3_pi2 * n4;
        vec4 SincIIn  = Fc_i_pi2  * n4;
        vec4 SincQIn  = Fc_q_pi2  * n4;

        vec4 SincY1, SincY2, SincY3;
        SincY1.x = (SincYIn1.x != 0.0) ? sin(SincYIn1.x) / SincYIn1.x : 1.0;
        SincY1.y = (SincYIn1.y != 0.0) ? sin(SincYIn1.y) / SincYIn1.y : 1.0;
        SincY1.z = (SincYIn1.z != 0.0) ? sin(SincYIn1.z) / SincYIn1.z : 1.0;
        SincY1.w = (SincYIn1.w != 0.0) ? sin(SincYIn1.w) / SincYIn1.w : 1.0;
        SincY2.x = (SincYIn2.x != 0.0) ? sin(SincYIn2.x) / SincYIn2.x : 1.0;
        SincY2.y = (SincYIn2.y != 0.0) ? sin(SincYIn2.y) / SincYIn2.y : 1.0;
        SincY2.z = (SincYIn2.z != 0.0) ? sin(SincYIn2.z) / SincYIn2.z : 1.0;
        SincY2.w = (SincYIn2.w != 0.0) ? sin(SincYIn2.w) / SincYIn2.w : 1.0;
        SincY3.x = (SincYIn3.x != 0.0) ? sin(SincYIn3.x) / SincYIn3.x : 1.0;
        SincY3.y = (SincYIn3.y != 0.0) ? sin(SincYIn3.y) / SincYIn3.y : 1.0;
        SincY3.z = (SincYIn3.z != 0.0) ? sin(SincYIn3.z) / SincYIn3.z : 1.0;
        SincY3.w = (SincYIn3.w != 0.0) ? sin(SincYIn3.w) / SincYIn3.w : 1.0;

        vec4 IdealY = Fc_y1_2 * SincY1 - Fc_y2_2 * SincY2 + Fc_y3_2 * SincY3;

        vec4 IdealI, IdealQ;
        IdealI.x = Fc_i_2 * ((SincIIn.x != 0.0) ? sin(SincIIn.x) / SincIIn.x : 1.0);
        IdealI.y = Fc_i_2 * ((SincIIn.y != 0.0) ? sin(SincIIn.y) / SincIIn.y : 1.0);
        IdealI.z = Fc_i_2 * ((SincIIn.z != 0.0) ? sin(SincIIn.z) / SincIIn.z : 1.0);
        IdealI.w = Fc_i_2 * ((SincIIn.w != 0.0) ? sin(SincIIn.w) / SincIIn.w : 1.0);
        IdealQ.x = Fc_q_2 * ((SincQIn.x != 0.0) ? sin(SincQIn.x) / SincQIn.x : 1.0);
        IdealQ.y = Fc_q_2 * ((SincQIn.y != 0.0) ? sin(SincQIn.y) / SincQIn.y : 1.0);
        IdealQ.z = Fc_q_2 * ((SincQIn.z != 0.0) ? sin(SincQIn.z) / SincQIn.z : 1.0);
        IdealQ.w = Fc_q_2 * ((SincQIn.w != 0.0) ? sin(SincQIn.w) / SincQIn.w : 1.0);

        vec4 FilterY = SincKernel * IdealY;
        vec4 FilterI = SincKernel * IdealI;
        vec4 FilterQ = SincKernel * IdealQ;

        YAccum += C * FilterY;
        IAccum += C * cos(WT) * FilterI;
        QAccum += C * sin(WT) * FilterQ;
    }

    vec3 YIQ = vec3(
        (YAccum.r + YAccum.g + YAccum.b + YAccum.a),
        (IAccum.r + IAccum.g + IAccum.b + IAccum.a) * 2.0,
        (QAccum.r + QAccum.g + QAccum.b + QAccum.a) * 2.0);

    vec3 RGB = vec3(
        dot(YIQ, RDot),
        dot(YIQ, GDot),
        dot(YIQ, BDot));

    FragColor = vec4(RGB, BaseTexel.a);
}