Transformer architecture implemented in TypeScript with WebGPU.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
nn/shader/tmul.wgsl

34 lines
854 B

alias number = f32;
struct PackedMatrixBatch {
size: vec4<u32>,
data: array<vec4<number>>,
}
struct MatrixBatch {
size: vec4<u32>,
data: array<number>,
}
@group(0) @binding(0)
var<storage, read> l: PackedMatrixBatch;
@group(0) @binding(1)
var<storage, read> tr: PackedMatrixBatch;
@group(0) @binding(2)
var<storage, read_write> result: MatrixBatch;
@compute @workgroup_size(1)
fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
result.size = l.size;
result.size.x = tr.size.y;
var s = 0f;
let vc = l.size.x / 4;
let lo = (global_id.z * l.size.y + global_id.y) * vc;
let ro = (global_id.z * tr.size.y + global_id.x) * vc;
for (var i = 0u; i < vc; i++) {
s += dot(l.data[lo + i], tr.data[ro + i]);
}
result.data[global_id.z * result.size.y * result.size.x + global_id.y * result.size.x + global_id.x] = s;
}