shaders/elementwise_mul.wgsl

// elementwise_mul.wgsl — a[i] = a[i] * b[i], in-place.
// Used in Mamba mixer for gate: gated = hidden_y * silu(gate).

struct Params { n: u32, }
@group(0) @binding(0) var<storage, read_write> a_buf: array<f32>;
@group(0) @binding(1) var<storage, read>       b_buf: array<f32>;
@group(1) @binding(0) var<uniform> params: Params;

@compute @workgroup_size(64)
fn main(@builtin(global_invocation_id) gid: vec3<u32>) {
    let i = gid.x;
    if (i >= params.n) { return; }
    a_buf[i] = a_buf[i] * b_buf[i];
}