#include <metal_stdlib>
using namespace metal;

// Cube lines — Danil
// MSL port of cube_lines.vk.glsl. The original Shadertoy is ~780 lines and
// uses every GLSL trick (mat3 chains, arrays, fwidth-based AA, swap macros).
//
// Notes on compromises in this port:
//   - fcos() in the GLSL uses fwidth() with a fallback that depends on
//     iResolution.y. To avoid threading iResolution through ~10 helpers, this
//     port uses the simpler `fcos = cos` path (equivalent to defining AA_ALL
//     in the original). Edges may be slightly less smooth than the Vulkan/GL
//     versions; cosmetic only.
//   - Optional Shadertoy defines (ANIM_SHAPE, ANIM_COLOR, ROTATION_SPEED,
//     CAMERA_*, USE_COLOR, AA_*, ONLY_BOX, etc.) are left at their defaults
//     just like the .vk.glsl ships them.

struct ShadertoyUBO {
    float iTime;
    float2 iResolution;
};

struct PassthroughVOut {
    float4 pos [[position]];
    float2 uv;
};

constant float ROTATION_SPEED = 0.8999;
constant float tshift = 53.0;
constant float FDIST = 0.7;
constant float PI = 3.1415926;
constant float3 BOXDIMS = float3(0.75, 0.75, 1.25);
constant float IOR = 1.33;

static float3x3 rotx(float a) {
    float s = sin(a), c = cos(a);
    return float3x3(float3(1.0, 0.0, 0.0), float3(0.0, c, s), float3(0.0, -s, c));
}
static float3x3 roty(float a) {
    float s = sin(a), c = cos(a);
    return float3x3(float3(c, 0.0, s), float3(0.0, 1.0, 0.0), float3(-s, 0.0, c));
}
static float3x3 rotz(float a) {
    float s = sin(a), c = cos(a);
    return float3x3(float3(c, s, 0.0), float3(-s, c, 0.0), float3(0.0, 0.0, 1.0));
}

static float3 fcos(float3 x) { return cos(x); }

static float3 getColor(float3 p) {
    p = abs(p);
    p *= 1.25;
    p = 0.5 * p / dot(p, p);

    float t = 0.13 * length(p);
    float3 col = float3(0.3, 0.4, 0.5);
    col += 0.12 * fcos(6.28318 * t * 1.0 + float3(0.0, 0.8, 1.1));
    col += 0.11 * fcos(6.28318 * t * 3.1 + float3(0.3, 0.4, 0.1));
    col += 0.10 * fcos(6.28318 * t * 5.1 + float3(0.1, 0.7, 1.1));
    col += 0.10 * fcos(6.28318 * t * 17.1 + float3(0.2, 0.6, 0.7));
    col += 0.10 * fcos(6.28318 * t * 31.1 + float3(0.1, 0.6, 0.7));
    col += 0.10 * fcos(6.28318 * t * 65.1 + float3(0.0, 0.5, 0.8));
    col += 0.10 * fcos(6.28318 * t * 115.1 + float3(0.1, 0.4, 0.7));
    col += 0.10 * fcos(6.28318 * t * 265.1 + float3(1.1, 1.4, 2.7));
    col = clamp(col, 0.0, 1.0);
    return col;
}

static void calcColor(float3 ro, float3 rd, float3 nor, float d, float len, int idx,
                      bool si, float td,
                      thread float4& colx, thread float4& colsi) {
    float3 pos = ro + rd * d;
    float a = 1.0 - smoothstep(len - 0.15 * 0.5, len + 0.00001, length(pos));
    float3 col = getColor(pos);
    colx = float4(col, a);
    if (si) {
        pos = ro + rd * td;
        float ta = 1.0 - smoothstep(len - 0.15 * 0.5, len + 0.00001, length(pos));
        col = getColor(pos);
        colsi = float4(col, ta);
    }
}

static bool iBilinearPatch(float3 ro, float3 rd, float4 ps, float4 ph, float sz,
                           thread float& t, thread float3& norm,
                           thread bool& si, thread float& tsi, thread float3& normsi,
                           thread float& fade, thread float& fadesi) {
    float3 va = float3(0.0, 0.0, ph.x + ph.w - ph.y - ph.z);
    float3 vb = float3(0.0, ps.w - ps.y, ph.z - ph.x);
    float3 vc = float3(ps.z - ps.x, 0.0, ph.y - ph.x);
    float3 vd = float3(ps.xy, ph.x);
    t = -1.0;
    tsi = -1.0;
    si = false;
    fade = 1.0;
    fadesi = 1.0;
    norm = float3(0.0, 1.0, 0.0);
    normsi = float3(0.0, 1.0, 0.0);

    float tmp = 1.0 / (vb.y * vc.x);
    float a = 0.0, b = 0.0, c = 0.0;
    float d = va.z * tmp;
    float e = 0.0, f = 0.0;
    float g = (vc.z * vb.y - vd.y * va.z) * tmp;
    float h = (vb.z * vc.x - va.z * vd.x) * tmp;
    float i = -1.0;
    float j = (vd.x * vd.y * va.z + vd.z * vb.y * vc.x) * tmp - (vd.y * vb.z * vc.x + vd.x * vc.z * vb.y) * tmp;

    float p = dot(float3(a, b, c), rd.xzy * rd.xzy) + dot(float3(d, e, f), rd.xzy * rd.zyx);
    float q = dot(float3(2.0, 2.0, 2.0) * ro.xzy * rd.xyz, float3(a, b, c)) + dot(ro.xzz * rd.zxy, float3(d, d, e)) +
              dot(ro.yyx * rd.zxy, float3(e, f, f)) + dot(float3(g, h, i), rd.xzy);
    float r = dot(float3(a, b, c), ro.xzy * ro.xzy) + dot(float3(d, e, f), ro.xzy * ro.zyx) + dot(float3(g, h, i), ro.xzy) + j;

    if (abs(p) < 0.000001) {
        float tt = -r / q;
        if (tt <= 0.0) return false;
        t = tt;
        float3 pos = ro + t * rd;
        if (length(pos) > sz) return false;
        float3 grad = float3(2.0) * pos.xzy * float3(a, b, c) + pos.zxz * float3(d, d, e) + pos.yyx * float3(f, e, f) + float3(g, h, i);
        norm = -normalize(grad);
        return true;
    } else {
        float sq = q * q - 4.0 * p * r;
        if (sq < 0.0) return false;
        float s = sqrt(sq);
        float t0 = (-q + s) / (2.0 * p);
        float t1 = (-q - s) / (2.0 * p);
        float tt1 = min(t0 < 0.0 ? t1 : t0, t1 < 0.0 ? t0 : t1);
        float tt2 = max(t0 > 0.0 ? t1 : t0, t1 > 0.0 ? t0 : t1);
        float tt0 = tt1;
        if (tt0 <= 0.0) return false;
        float3 pos = ro + tt0 * rd;
        bool ru = step(sz, length(pos)) > 0.5;
        if (ru) {
            tt0 = tt2;
            pos = ro + tt0 * rd;
        }
        if (tt0 <= 0.0) return false;
        bool ru2 = step(sz, length(pos)) > 0.5;
        if (ru2) return false;

        if ((tt2 > 0.0) && (!ru) && !(step(sz, length(ro + tt2 * rd)) > 0.5)) {
            si = true;
            fadesi = s;
            tsi = tt2;
            float3 tpos = ro + tsi * rd;
            float3 tgrad = float3(2.0) * tpos.xzy * float3(a, b, c) + tpos.zxz * float3(d, d, e) +
                           tpos.yyx * float3(f, e, f) + float3(g, h, i);
            normsi = -normalize(tgrad);
        }

        fade = s;
        t = tt0;
        float3 grad = float3(2.0) * pos.xzy * float3(a, b, c) + pos.zxz * float3(d, d, e) + pos.yyx * float3(f, e, f) + float3(g, h, i);
        norm = -normalize(grad);
        return true;
    }
}

static float dot2(float3 v) { return dot(v, v); }

static float segShadow(float3 ro, float3 rd, float3 pa, float sh) {
    float dm = dot(rd.yz, rd.yz);
    float k1 = (ro.x - pa.x) * dm;
    float k2 = (ro.x + pa.x) * dm;
    float2 k5 = (ro.yz + pa.yz) * dm;
    float k3 = dot(ro.yz + pa.yz, rd.yz);
    float2 k4 = (pa.yz + pa.yz) * rd.yz;
    float2 k6 = (pa.yz + pa.yz) * dm;

    for (int i = 0; i < 4; i++) {
        float2 s = float2(i & 1, i >> 1);
        float t = dot(s, k4) - k3;
        if (t > 0.0) {
            sh = min(sh, dot2(float3(clamp(-rd.x * t, k1, k2), k5 - k6 * s) + rd * t) / (t * t));
        }
    }
    return sh;
}

static float boxSoftShadow(float3 ro, float3 rd, float3 rad, float sk) {
    rd += 0.0001 * (1.0 - abs(sign(rd)));
    float3 rdd = rd;
    float3 roo = ro;

    float3 m = 1.0 / rdd;
    float3 n = m * roo;
    float3 k = abs(m) * rad;

    float3 t1 = -n - k;
    float3 t2 = -n + k;

    float tN = max(max(t1.x, t1.y), t1.z);
    float tF = min(min(t2.x, t2.y), t2.z);

    if (tN < tF && tF > 0.0) return 0.0;

    float sh = 1.0;
    sh = segShadow(roo.xyz, rdd.xyz, rad.xyz, sh);
    sh = segShadow(roo.yzx, rdd.yzx, rad.yzx, sh);
    sh = segShadow(roo.zxy, rdd.zxy, rad.zxy, sh);
    sh = clamp(sk * sqrt(sh), 0.0, 1.0);
    return sh * sh * (3.0 - 2.0 * sh);
}

static float boxRay(float3 ro, float3 rd, float3 r, thread float3& nn, bool entering) {
    rd += 0.0001 * (1.0 - abs(sign(rd)));
    float3 dr = 1.0 / rd;
    float3 n = ro * dr;
    float3 k = r * abs(dr);

    float3 pin = -k - n;
    float3 pout = k - n;
    float tin = max(pin.x, max(pin.y, pin.z));
    float tout = min(pout.x, min(pout.y, pout.z));
    if (tin > tout) return -1.0;
    if (entering) {
        nn = -sign(rd) * step(pin.zxy, pin.xyz) * step(pin.yzx, pin.xyz);
    } else {
        nn = sign(rd) * step(pout.xyz, pout.zxy) * step(pout.xyz, pout.yzx);
    }
    return entering ? tin : tout;
}

static float3 bgcol(float3 rd) {
    return mix(float3(0.01), float3(0.336, 0.458, 0.668), 1.0 - pow(abs(rd.z + 0.25), 1.3));
}

static float3 background(float3 ro, float3 rd, float3 l_dir, thread float& alpha) {
    float t = (-BOXDIMS.z - ro.z) / rd.z;
    alpha = 0.0;
    float3 bgc = bgcol(rd);
    if (t < 0.0) return bgc;
    float2 uv = ro.xy + t * rd.xy;
    float shad = boxSoftShadow((ro + t * rd), normalize(l_dir + float3(0.0, 0.0, 1.0)) * rotz(PI * 0.65), BOXDIMS, 1.5);
    float aofac = smoothstep(-0.95, 0.75, length(abs(uv) - min(abs(uv), float2(0.45))));
    aofac = min(aofac, smoothstep(-0.65, 1.0, shad));
    float lght = max(dot(normalize(ro + t * rd + float3(0.0, -0.0, -5.0)), normalize(l_dir - float3(0.0, 0.0, 1.0)) * rotz(PI * 0.65)), 0.0);
    float3 col = mix(float3(0.4), float3(0.71, 0.772, 0.895), lght * lght * aofac + 0.05) * aofac;
    alpha = 1.0 - smoothstep(7.0, 10.0, length(uv));
    return mix(col * length(col) * 0.8, bgc, smoothstep(7.0, 10.0, length(uv)));
}

static float4 insides(float3 ro, float3 rd, float3 nor_c, float3 l_dir, thread float& tout) {
    tout = -1.0;

    float pi = 3.1415926;

    if (abs(nor_c.x) > 0.5) {
        rd = rd.xzy * nor_c.x;
        ro = ro.xzy * nor_c.x;
    } else if (abs(nor_c.z) > 0.5) {
        l_dir = l_dir * roty(pi);
        rd = rd.yxz * nor_c.z;
        ro = ro.yxz * nor_c.z;
    } else if (abs(nor_c.y) > 0.5) {
        l_dir = l_dir * rotz(-pi * 0.5);
        rd = rd * nor_c.y;
        ro = ro * nor_c.y;
    }

    const float curvature = 0.5;
    float bil_size = 1.0;
    float4 ps = float4(-bil_size, -bil_size, bil_size, bil_size) * curvature;
    float4 ph = float4(-bil_size, bil_size, bil_size, -bil_size) * curvature;

    float4 colx[3] = { float4(0.0), float4(0.0), float4(0.0) };
    float3 dx[3] = { float3(-1.0), float3(-1.0), float3(-1.0) };
    float4 colxsi[3] = { float4(0.0), float4(0.0), float4(0.0) };
    int order[3] = { 0, 1, 2 };

    for (int i = 0; i < 3; i++) {
        if (abs(nor_c.x) > 0.5) {
            ro = ro * rotz(-pi * (1.0 / 3.0));
            rd = rd * rotz(-pi * (1.0 / 3.0));
        } else if (abs(nor_c.z) > 0.5) {
            ro = ro * rotz(pi * (1.0 / 3.0));
            rd = rd * rotz(pi * (1.0 / 3.0));
        } else if (abs(nor_c.y) > 0.5) {
            ro = ro * rotx(pi * (1.0 / 3.0));
            rd = rd * rotx(pi * (1.0 / 3.0));
        }
        float3 normnew;
        float tnew;
        bool si;
        float tsi;
        float3 normsi;
        float fade;
        float fadesi;

        if (iBilinearPatch(ro, rd, ps, ph, bil_size, tnew, normnew, si, tsi, normsi, fade, fadesi)) {
            if (tnew > 0.0) {
                float4 tcol, tcolsi;
                calcColor(ro, rd, normnew, tnew, bil_size, i, si, tsi, tcol, tcolsi);
                if (tcol.a > 0.0) {
                    dx[i] = float3(tnew, float(si), tsi);

                    float dif = clamp(dot(normnew, l_dir), 0.0, 1.0);
                    float amb = clamp(0.5 + 0.5 * dot(normnew, l_dir), 0.0, 1.0);

                    {
                        float3 shad = float3(0.32, 0.43, 0.54) * amb + float3(1.0, 0.9, 0.7) * dif;
                        const float3 tcr = float3(1.0, 0.21, 0.11);
                        float ta = clamp(length(tcol.rgb), 0.0, 1.0);
                        tcol = clamp(tcol * tcol * 2.0, 0.0, 1.0);
                        float4 tvalx = float4(tcol.rgb * shad * 1.4 + 3.0 * (tcr * tcol.rgb) * clamp(1.0 - (amb + dif), 0.0, 1.0), min(tcol.a, ta));
                        tvalx.rgb = clamp(2.0 * tvalx.rgb * tvalx.rgb, 0.0, 1.0);
                        tvalx *= min(fade * 5.0, 1.0);
                        colx[i] = tvalx;
                    }
                    if (si) {
                        dif = clamp(dot(normsi, l_dir), 0.0, 1.0);
                        amb = clamp(0.5 + 0.5 * dot(normsi, l_dir), 0.0, 1.0);
                        float3 shad = float3(0.32, 0.43, 0.54) * amb + float3(1.0, 0.9, 0.7) * dif;
                        const float3 tcr = float3(1.0, 0.21, 0.11);
                        float ta = clamp(length(tcolsi.rgb), 0.0, 1.0);
                        tcolsi = clamp(tcolsi * tcolsi * 2.0, 0.0, 1.0);
                        float4 tvalx = float4(tcolsi.rgb * shad + 3.0 * (tcr * tcolsi.rgb) * clamp(1.0 - (amb + dif), 0.0, 1.0), min(tcolsi.a, ta));
                        tvalx.rgb = clamp(2.0 * tvalx.rgb * tvalx.rgb, 0.0, 1.0);
                        tvalx.rgb *= min(fadesi * 5.0, 1.0);
                        colxsi[i] = tvalx;
                    }
                }
            }
        }
    }

    // sort by dx[*].x descending (3 passes of bubble sort like the GLSL)
    if (dx[0].x < dx[1].x) { float3 tv = dx[0]; dx[0] = dx[1]; dx[1] = tv; int ti = order[0]; order[0] = order[1]; order[1] = ti; }
    if (dx[1].x < dx[2].x) { float3 tv = dx[1]; dx[1] = dx[2]; dx[2] = tv; int ti = order[1]; order[1] = order[2]; order[2] = ti; }
    if (dx[0].x < dx[1].x) { float3 tv = dx[0]; dx[0] = dx[1]; dx[1] = tv; int ti = order[0]; order[0] = order[1]; order[1] = ti; }

    tout = max(max(dx[0].x, dx[1].x), dx[2].x);

    float a = 1.0;
    if (dx[0].y < 0.5) {
        a = colx[order[0]].a;
    }

    bool rul[3] = {
        ((dx[0].y > 0.5) && (dx[1].x <= 0.0)),
        ((dx[1].y > 0.5) && (dx[0].x > dx[1].z)),
        ((dx[2].y > 0.5) && (dx[1].x > dx[2].z))
    };
    for (int k = 0; k < 3; k++) {
        if (rul[k]) {
            float4 tcolxsi = colxsi[order[k]];
            float4 tcolx = colx[order[k]];
            float4 tvalx = mix(tcolxsi, tcolx, tcolx.a);
            float4 tvalx2 = mix(float4(0.0), tvalx, max(tcolx.a, tcolxsi.a));
            colx[order[k]] = tvalx2;
        }
    }

    float a1 = (dx[1].y < 0.5) ? colx[order[1]].a : ((dx[1].z > dx[0].x) ? colx[order[1]].a : 1.0);
    float a2 = (dx[2].y < 0.5) ? colx[order[2]].a : ((dx[2].z > dx[1].x) ? colx[order[2]].a : 1.0);
    float3 col = mix(mix(colx[order[0]].rgb, colx[order[1]].rgb, a1), colx[order[2]].rgb, a2);
    a = max(max(a, a1), a2);
    return float4(col, a);
}

fragment float4 cube_lines_fs(PassthroughVOut in [[stage_in]],
                               constant ShadertoyUBO& U [[buffer(0)]]) {
    float2 fragCoord = in.uv * U.iResolution;
    float iTime = U.iTime;

    float3 l_dir = normalize(float3(0.0, 1.0, 0.0));
    l_dir = l_dir * rotz(0.5);
    float mouseY = PI * 0.49 - smoothstep(0.0, 8.5, fmod((iTime + tshift) * 0.33, 25.0))
                            * (1.0 - smoothstep(14.0, 24.0, fmod((iTime + tshift) * 0.33, 25.0))) * 0.55 * PI;
    float mouseX = -2.0 * PI - 0.25 * (iTime * ROTATION_SPEED + tshift);

    float3 eye = 4.0 * float3(cos(mouseX) * cos(mouseY), sin(mouseX) * cos(mouseY), sin(mouseY));
    float3 w = normalize(-eye);
    float3 up = float3(0.0, 0.0, 1.0);
    float3 u = normalize(cross(w, up));
    float3 v = cross(u, w);

    float4 tot = float4(0.0);
    float2 uv = (fragCoord - 0.5 * U.iResolution) / U.iResolution.x;
    float3 rd = normalize(w * FDIST + uv.x * u + uv.y * v);

    float3 ni;
    float t = boxRay(eye, rd, BOXDIMS, ni, true);
    float3 ro = eye + t * rd;
    float2 coords = ro.xy * ni.z / BOXDIMS.xy + ro.yz * ni.x / BOXDIMS.yz + ro.zx * ni.y / BOXDIMS.zx;
    float fadeborders = (1.0 - smoothstep(0.915, 1.05, abs(coords.x))) * (1.0 - smoothstep(0.915, 1.05, abs(coords.y)));

    if (t > 0.0) {
        float3 col = float3(0.0);
        float R0 = (IOR - 1.0) / (IOR + 1.0);
        R0 *= R0;

        float2 theta = float2(0.0);
        float3 n = float3(cos(theta.x) * sin(theta.y), sin(theta.x) * sin(theta.y), cos(theta.y));

        float3 nr = n.zxy * ni.x + n.yzx * ni.y + n.xyz * ni.z;
        float3 rdr = reflect(rd, nr);
        float talpha;
        float3 reflcol = background(ro, rdr, l_dir, talpha);

        float3 rd2 = refract(rd, nr, 1.0 / IOR);

        float accum = 1.0;
        float3 no2 = ni;
        float3 ro_refr = ro;

        float4 colo[2] = { float4(0.0), float4(0.0) };

        for (int j = 0; j < 2; j++) {
            float tb;
            float2 coords2 = ro_refr.xy * no2.z + ro_refr.yz * no2.x + ro_refr.zx * no2.y;
            float3 eye2 = float3(coords2, -1.0);
            float3 rd2trans = rd2.yzx * no2.x + rd2.zxy * no2.y + rd2.xyz * no2.z;

            rd2trans.z = -rd2trans.z;
            float4 internalcol = insides(eye2, rd2trans, no2, l_dir, tb);
            if (tb > 0.0) {
                internalcol.rgb *= accum;
                colo[j] = internalcol;
            }

            if ((tb <= 0.0) || (internalcol.a < 1.0)) {
                float tout = boxRay(ro_refr, rd2, BOXDIMS, no2, false);
                no2 = n.zyx * no2.x + n.xzy * no2.y + n.yxz * no2.z;
                float3 rout = ro_refr + tout * rd2;
                float3 rdout = refract(rd2, -no2, IOR);
                float fresnel2 = R0 + (1.0 - R0) * pow(1.0 - dot(rdout, no2), 1.3);
                rd2 = reflect(rd2, -no2);

                ro_refr = rout;
                ro_refr.z = max(ro_refr.z, -0.999);

                accum *= fresnel2;
            }
        }
        float fresnel = R0 + (1.0 - R0) * pow(1.0 - dot(-rd, nr), 5.0);
        col = mix(mix(colo[1].rgb * colo[1].a, colo[0].rgb, colo[0].a) * fadeborders, reflcol, pow(fresnel, 1.5));
        col = clamp(col, 0.0, 1.0);

        float cineshader_alpha = clamp(0.15 * dot(eye, ro), 0.0, 1.0);
        tot += float4(col, cineshader_alpha);
    } else {
        float alpha;
        tot += float4(background(eye, rd, l_dir, alpha), 0.15);
    }

    float4 outColor = tot;
    outColor.rgb = clamp(outColor.rgb, 0.0, 1.0);
    return outColor;
}