Implement tree hash caching (#584)

* Implement basic tree hash caching

* Use spaces to indent top-level Cargo.toml

* Optimize BLS tree hash by hashing bytes directly

* Implement tree hash caching for validator registry

* Persist BeaconState tree hash cache to disk

* Address Paul's review comments

eth2/utils/tree_hash/benches/benches.rs

@@ -1,8 +1,6 @@
-#[macro_use]
-extern crate lazy_static;
-
 use criterion::Criterion;
 use criterion::{black_box, criterion_group, criterion_main, Benchmark};
+use lazy_static::lazy_static;
 use types::test_utils::{generate_deterministic_keypairs, TestingBeaconStateBuilder};
 use types::{BeaconState, EthSpec, Keypair, MainnetEthSpec, MinimalEthSpec};
 
@@ -27,25 +25,61 @@ fn build_state<T: EthSpec>(validator_count: usize) -> BeaconState<T> {
     state
 }
 
+// Note: `state.canonical_root()` uses whatever `tree_hash` that the `types` crate
+// uses, which is not necessarily this crate. If you want to ensure that types is
+// using this local version of `tree_hash`, ensure you add a workspace-level
+// [dependency
+// patch](https://doc.rust-lang.org/cargo/reference/manifest.html#the-patch-section).
 fn bench_suite<T: EthSpec>(c: &mut Criterion, spec_desc: &str, validator_count: usize) {
-    let state = build_state::<T>(validator_count);
+    let state1 = build_state::<T>(validator_count);
+    let state2 = state1.clone();
+    let mut state3 = state1.clone();
+    state3.build_tree_hash_cache().unwrap();
 
     c.bench(
-        &format!("{}/{}_validators", spec_desc, validator_count),
+        &format!("{}/{}_validators/no_cache", spec_desc, validator_count),
         Benchmark::new("genesis_state", move |b| {
             b.iter_batched_ref(
-                || state.clone(),
-                // Note: `state.canonical_root()` uses whatever `tree_hash` that the `types` crate
-                // uses, which is not necessarily this crate. If you want to ensure that types is
-                // using this local version of `tree_hash`, ensure you add a workspace-level
-                // [dependency
-                // patch](https://doc.rust-lang.org/cargo/reference/manifest.html#the-patch-section).
+                || state1.clone(),
                 |state| black_box(state.canonical_root()),
                 criterion::BatchSize::SmallInput,
             )
         })
         .sample_size(10),
     );
+
+    c.bench(
+        &format!("{}/{}_validators/empty_cache", spec_desc, validator_count),
+        Benchmark::new("genesis_state", move |b| {
+            b.iter_batched_ref(
+                || state2.clone(),
+                |state| {
+                    assert!(!state.tree_hash_cache.is_initialized());
+                    black_box(state.update_tree_hash_cache().unwrap())
+                },
+                criterion::BatchSize::SmallInput,
+            )
+        })
+        .sample_size(10),
+    );
+
+    c.bench(
+        &format!(
+            "{}/{}_validators/up_to_date_cache",
+            spec_desc, validator_count
+        ),
+        Benchmark::new("genesis_state", move |b| {
+            b.iter_batched_ref(
+                || state3.clone(),
+                |state| {
+                    assert!(state.tree_hash_cache.is_initialized());
+                    black_box(state.update_tree_hash_cache().unwrap())
+                },
+                criterion::BatchSize::SmallInput,
+            )
+        })
+        .sample_size(10),
+    );
 }
 
 fn all_benches(c: &mut Criterion) {