// SPDX-License-Identifier: AGPL-3.0-or-later
// VectorAdd: C = A * alpha + B (f64 canonical)
// CUDA equivalent: cublas::axpy
// Use cases: Gradient updates, residual connections
@group(0) @binding(0) var<storage, read> a: array<f64>;
@group(0) @binding(1) var<storage, read> b: array<f64>;
@group(0) @binding(2) var<storage, read_write> output: array<f64>;
struct Params {
size: u32,
alpha: f64,
}
@group(0) @binding(3) var<uniform> params: Params;
@compute @workgroup_size(256)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
let idx = global_id.x;
if (idx >= params.size) {
return;
}
// C = A * alpha + B
output[idx] = a[idx] * params.alpha + b[idx];
}