Keep the embedder off the Hexagon, and let the probe example ask for a runtime
On the tablet the engine compiled arcface for the NPU: the routing compared the form a rung wants with the form on offer, and for the embedder both are f32, so nothing said no. A rung now says which roles it serves at all, and the Hexagon does not serve the embedder (§7 — its vectors must compare across devices). Tested at the routing seam. dr-segment's onnx_probe example still named ort-tract, which is what stopped the workspace test build.
This commit is contained in:
@@ -95,12 +95,21 @@ impl Rung {
|
||||
|
||||
/// The model form this rung wants for a role.
|
||||
fn form(self, role: Role) -> Form {
|
||||
match (self, role) {
|
||||
(Rung::Hexagon, Role::Embedder) => Form::F32,
|
||||
(Rung::Hexagon, _) => Form::Int8,
|
||||
match self {
|
||||
Rung::Hexagon => Form::Int8,
|
||||
_ => Form::F32,
|
||||
}
|
||||
}
|
||||
|
||||
/// Whether this rung runs `role` at all. The Hexagon takes int8 graphs
|
||||
/// only, and the embedder is never int8 (§7) — it runs on the CPU
|
||||
/// beside a detector on the NPU, so its vectors compare across devices.
|
||||
fn serves(self, role: Role) -> bool {
|
||||
match self {
|
||||
Rung::Hexagon => role != Role::Embedder,
|
||||
_ => true,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// How long a session outlives its last use unless [`Config::decay`] says
|
||||
@@ -407,7 +416,7 @@ fn current_rung(s: &State) -> Rung {
|
||||
/// caller adds [`form_suffix`] to the `model_id` it records.
|
||||
pub fn resolve_model(role: Role, canonical: &Path) -> (PathBuf, Form) {
|
||||
let rung = current_rung(&state().lock().unwrap());
|
||||
if rung.form(role) == Form::Int8 {
|
||||
if rung.serves(role) && rung.form(role) == Form::Int8 {
|
||||
let sibling = int8_sibling(canonical);
|
||||
if sibling.is_file() {
|
||||
return (sibling, Form::Int8);
|
||||
@@ -451,7 +460,7 @@ pub fn open(role: Role, form: Form, bytes: &[u8]) -> Result<Model, Error> {
|
||||
/// the form on offer, or a missing engine says one lower (§6 step 4).
|
||||
fn effective_rung(s: &State, selected: Rung, role: Role, form: Form, bytes: &[u8]) -> Rung {
|
||||
let mut rung = selected;
|
||||
if rung.form(role) != form {
|
||||
if !rung.serves(role) || rung.form(role) != form {
|
||||
// The embedder on a Hexagon device, or an f32 detector where the int8
|
||||
// sibling was missing: neither can go to the NPU.
|
||||
rung = rung.fallback();
|
||||
@@ -541,6 +550,36 @@ mod tests {
|
||||
assert_eq!(loaded(), 1);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn the_hexagon_never_takes_the_embedder() {
|
||||
assert!(!Rung::Hexagon.serves(Role::Embedder));
|
||||
assert!(Rung::Hexagon.serves(Role::Detector));
|
||||
assert_eq!(Rung::Hexagon.form(Role::Detector), Form::Int8);
|
||||
// A detector offered in f32 on a Hexagon device lands on the CPU.
|
||||
let s = State {
|
||||
config: Config::default(),
|
||||
cache: Cache {
|
||||
rung: Some(Rung::Hexagon),
|
||||
..Cache::default()
|
||||
},
|
||||
probing: false,
|
||||
wanted: 0,
|
||||
};
|
||||
assert_eq!(
|
||||
effective_rung(&s, Rung::Hexagon, Role::Embedder, Form::F32, b""),
|
||||
Rung::Cpu
|
||||
);
|
||||
assert_eq!(
|
||||
effective_rung(&s, Rung::Hexagon, Role::Detector, Form::F32, b""),
|
||||
Rung::Cpu
|
||||
);
|
||||
// An int8 detector whose context is not compiled yet: also the CPU.
|
||||
assert_eq!(
|
||||
effective_rung(&s, Rung::Hexagon, Role::Detector, Form::Int8, b""),
|
||||
Rung::Cpu
|
||||
);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn the_status_line_reads_as_the_floor_before_init() {
|
||||
let s = status();
|
||||
|
||||
Reference in New Issue
Block a user