perf(kernel): linked cap tree, translate-based page lookup, next-fit frames/heap, targeted dequeue; host benches

crates/kernel/src/smp.rs

@@ -35,17 +35,17 @@ impl<P: Platform> Kernel<P> {
 
     /// Homes a new thread on the core with the fewest live threads.
     pub(crate) fn place(&mut self, tid: ThreadId) {
-        let cores = self.scheduler.num_cores().max(1);
-        let load = |core: CoreId| {
-            self.threads
-                .iter()
-                .filter(|(t, _)| *t != tid && self.home(*t) == core)
-                .count()
-        };
+        let cores = usize::from(self.scheduler.num_cores().max(1));
+        let mut load = [0usize; u8::MAX as usize + 1];
+        for (t, _) in self.threads.iter().filter(|(t, _)| *t != tid) {
+            if let Some(n) = load.get_mut(usize::from(self.home(t).raw())) {
+                *n += 1;
+            }
+        }
         let best = (0..cores)
-            .map(CoreId::from_raw)
-            .min_by_key(|c| load(*c))
-            .unwrap_or(CoreId::from_raw(0));
+            .min_by_key(|c| load.get(*c).copied().unwrap_or(usize::MAX))
+            .and_then(|c| u8::try_from(c).ok())
+            .map_or(CoreId::from_raw(0), CoreId::from_raw);
         self.set_home(tid, best);
     }