Lesson 2 — Scalar parameters (SAXPY)

SAXPY (Single-precision A times X Plus Y) is the canonical BLAS-1 operation: y[i] = alpha * x[i] + y[i]. It is a staple benchmark because it is memory-bandwidth-bound — the GPU must stream every element through its caches at least twice.

Scalar uniforms

Scalars like alpha are not per-thread arrays; they are the same value for every thread. In Sarek you declare them as plain typed arguments (e.g., (alpha : float32)), and the transpiler automatically packs them into a uniform buffer that every thread can read.

For this lesson alpha = 2.0 is injected automatically — you can see it in Show generated WGSL as a field of the Params uniform struct.

Your task

Complete the kernel so that y.(i) is updated in-place: y[i] ← alpha * x[i] + y[i].

Kernel source fun (x : float32 vector) (y : float32 vector) (alpha : float32) -> let i = global_thread_id in y.(i) <- (* TODO *)

Click "Run on my GPU" to execute your kernel.

Hint

Float multiplication in OCaml is *. and addition is +.. Read alpha directly — it is a scalar argument available in every thread.