// 75% SMPTE color bars + moving square, Rec.709, 10-bit Y'CbCr, packed as v210 (4:2:2).
// One work-item per 16-byte block = 6 pixels (6 Y + 3 Cb + 3 Cr).

struct Params {
    width: u32,
    height: u32,
    frame: u32, // animation tick (0,1,2,... per generated frame), NOT the raw grain index
    _pad0: u32,
};

@group(0) @binding(0) var<storage, read_write> out: array<u32>;
@group(0) @binding(1) var<uniform> params: Params;

fn bar_index(px: u32) -> u32 {
    return min((px * 7u) / params.width, 6u);
}

@compute @workgroup_size(64)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
    let block = gid.x;
    let total = (params.width * params.height) / 6u;
    if (block >= total) {
        return;
    }

    let x = (block * 6u) % params.width;
    let py = (block * 6u) / params.width;

    // Bar order: 75% white, yellow, cyan, green, magenta, red, blue
    let y_tab  = array<u32, 7>(721u, 674u, 581u, 534u, 251u, 204u, 111u);
    let cb_tab = array<u32, 7>(512u, 176u, 589u, 253u, 771u, 435u, 848u);
    let cr_tab = array<u32, 7>(512u, 543u, 176u, 207u, 817u, 848u, 481u);

    // Moving square: horizontal oscillation around screen center.
    // frame is a small tick; converting the huge raw grain index here
    // would destroy f32 precision and freeze the motion.
    let centerX = f32(params.width) / 2.0;
    let centerY = f32(params.height) / 2.0;
    let fps = 25.0;
    let t = f32(params.frame) / fps;
    const squareSize = 150u;
    let half = f32(squareSize) / 2.0;
    let offsetPixels = sin(t * 0.5) * (centerX - half);

    let sq_x_min = centerX - half + offsetPixels;
    let sq_x_max = centerX + half + offsetPixels;
    let sq_y_min = centerY - half;
    let sq_y_max = centerY + half;
    let py_f = f32(py);

    var y: array<u32, 6>;
    var cb: array<u32, 6>;
    var cr: array<u32, 6>;
    for (var i = 0u; i < 6u; i++) {
        let px = f32(x + i);
        let b = bar_index(x + i);
        if (px >= sq_x_min && px < sq_x_max && py_f >= sq_y_min && py_f < sq_y_max) {
            // inverted bar: luma mirrored across studio range (64..940),
            // chroma mirrored across neutral 512
            y[i]  = 1004u - y_tab[b];
            cb[i] = 1024u - cb_tab[b];
            cr[i] = 1024u - cr_tab[b];
        } else {
            y[i] = y_tab[b];
            cb[i] = cb_tab[b];
            cr[i] = cr_tab[b];
        }
    }

    // v210 word layout, chroma co-sited with luma samples 0/2/4:
    // w0 = Cb0|Y0<<10|Cr0<<20; w1 = Y1|Cb2<<10|Y2<<20;
    // w2 = Cr2|Y3<<10|Cb4<<20; w3 = Y4|Cr4<<10|Y5<<20
    out[block * 4u + 0u] = (cb[0] & 0x3FFu) | ((y[0] & 0x3FFu) << 10u) | ((cr[0] & 0x3FFu) << 20u);
    out[block * 4u + 1u] = (y[1] & 0x3FFu) | ((cb[2] & 0x3FFu) << 10u) | ((y[2] & 0x3FFu) << 20u);
    out[block * 4u + 2u] = (cr[2] & 0x3FFu) | ((y[3] & 0x3FFu) << 10u) | ((cb[4] & 0x3FFu) << 20u);
    out[block * 4u + 3u] = (y[4] & 0x3FFu) | ((cr[4] & 0x3FFu) << 10u) | ((y[5] & 0x3FFu) << 20u);
}