// RLX — versatile ML compiler + runtime.
// Copyright (C) 2026 Eugene Hauptmann, Nataliya Kosmyna.
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, version 3.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program. If not, see <https://www.gnu.org/licenses/>.
// Single-rounded fused multiply-add: out[i] = fma(a[i], b[i], c[i]).
// WGSL `fma` is a genuine fused op (one rounding) — required for the
// error-free transforms (TwoProduct) that compensated arithmetic needs.
struct Params {
n: u32,
a_off: u32,
b_off: u32,
c_off: u32,
out_off: u32,
_p0: u32, _p1: u32, _p2: u32,
};
@group(0) @binding(0) var<storage, read_write> arena: array<f32>;
@group(0) @binding(1) var<uniform> params: Params;
@compute @workgroup_size(64)
fn fma_main(@builtin(global_invocation_id) gid: vec3<u32>, @builtin(num_workgroups) ngs: vec3<u32>) {
let i = gid.x + gid.y * ngs.x * 64u;
if (i >= params.n) { return; }
let a = arena[params.a_off + i];
let b = arena[params.b_off + i];
let c = arena[params.c_off + i];
arena[params.out_off + i] = fma(a, b, c);
}