Commands benchmarking (#2334)

# Objective

- Currently the `Commands` and `CommandQueue` have no performance testing. 
- As `Commands` are quite expensive due to the `Box<dyn Command>` allocated for each command, there should be perf tests for implementations that attempt to improve the performance.  

## Solution

- Add some benchmarking for `Commands` and `CommandQueue`.

benches/Cargo.toml

@@ -16,6 +16,11 @@ name = "system_stage"
 path = "benches/bevy_ecs/stages.rs"
 harness = false
 
+[[bench]]
+name = "commands"
+path = "benches/bevy_ecs/commands.rs"
+harness = false
+
 [[bench]]
 name = "iter"
 path = "benches/bevy_tasks/iter.rs"

benches/benches/bevy_ecs/commands.rs

@@ -0,0 +1,177 @@
+use bevy::ecs::{
+    system::{Command, CommandQueue, Commands},
+    world::World,
+};
+use criterion::{black_box, criterion_group, criterion_main, Criterion};
+
+criterion_group!(
+    benches,
+    empty_commands,
+    spawn_commands,
+    fake_commands,
+    zero_sized_commands,
+    medium_sized_commands,
+    large_sized_commands
+);
+criterion_main!(benches);
+
+struct A;
+struct B;
+struct C;
+
+fn empty_commands(criterion: &mut Criterion) {
+    let mut group = criterion.benchmark_group("empty_commands");
+    group.warm_up_time(std::time::Duration::from_millis(500));
+    group.measurement_time(std::time::Duration::from_secs(4));
+
+    group.bench_function("0_entities", |bencher| {
+        let mut world = World::default();
+        let mut command_queue = CommandQueue::default();
+
+        bencher.iter(|| {
+            command_queue.apply(&mut world);
+        });
+    });
+
+    group.finish();
+}
+
+fn spawn_commands(criterion: &mut Criterion) {
+    let mut group = criterion.benchmark_group("spawn_commands");
+    group.warm_up_time(std::time::Duration::from_millis(500));
+    group.measurement_time(std::time::Duration::from_secs(4));
+
+    for entity_count in (1..5).map(|i| i * 2 * 1000) {
+        group.bench_function(format!("{}_entities", entity_count), |bencher| {
+            let mut world = World::default();
+            let mut command_queue = CommandQueue::default();
+
+            bencher.iter(|| {
+                let mut commands = Commands::new(&mut command_queue, &world);
+                for i in 0..entity_count {
+                    let mut entity = commands.spawn();
+
+                    if black_box(i % 2 == 0) {
+                        entity.insert(A);
+                    }
+
+                    if black_box(i % 3 == 0) {
+                        entity.insert(B);
+                    }
+
+                    if black_box(i % 4 == 0) {
+                        entity.insert(C);
+                    }
+
+                    if black_box(i % 5 == 0) {
+                        entity.despawn();
+                    }
+                }
+                drop(commands);
+                command_queue.apply(&mut world);
+            });
+        });
+    }
+
+    group.finish();
+}
+
+struct FakeCommandA;
+struct FakeCommandB(u64);
+
+impl Command for FakeCommandA {
+    fn write(self: Box<Self>, world: &mut World) {
+        black_box(self);
+        black_box(world);
+    }
+}
+
+impl Command for FakeCommandB {
+    fn write(self: Box<Self>, world: &mut World) {
+        black_box(self);
+        black_box(world);
+    }
+}
+
+fn fake_commands(criterion: &mut Criterion) {
+    let mut group = criterion.benchmark_group("fake_commands");
+    group.warm_up_time(std::time::Duration::from_millis(500));
+    group.measurement_time(std::time::Duration::from_secs(4));
+
+    for command_count in (1..5).map(|i| i * 2 * 1000) {
+        group.bench_function(format!("{}_commands", command_count), |bencher| {
+            let mut world = World::default();
+            let mut command_queue = CommandQueue::default();
+
+            bencher.iter(|| {
+                let mut commands = Commands::new(&mut command_queue, &world);
+                for i in 0..command_count {
+                    if black_box(i % 2 == 0)
+                        commands.add(FakeCommandA);
+                    } else {
+                        commands.add(FakeCommandB(0));
+                    }
+                }
+                drop(commands);
+                command_queue.apply(&mut world);
+            });
+        });
+    }
+
+    group.finish();
+}
+
+#[derive(Default)]
+struct SizedCommand<T: Default + Send + Sync + 'static>(T);
+
+impl<T: Default + Send + Sync + 'static> Command for SizedCommand<T> {
+    fn write(self: Box<Self>, world: &mut World) {
+        black_box(self);
+        black_box(world);
+    }
+}
+
+struct LargeStruct([u64; 64]);
+
+impl Default for LargeStruct {
+    fn default() -> Self {
+        Self([0; 64])
+    }
+}
+
+fn sized_commands_impl<T: Default + Command>(criterion: &mut Criterion) {
+    let mut group =
+        criterion.benchmark_group(format!("sized_commands_{}_bytes", std::mem::size_of::<T>()));
+    group.warm_up_time(std::time::Duration::from_millis(500));
+    group.measurement_time(std::time::Duration::from_secs(4));
+
+    for command_count in (1..5).map(|i| i * 2 * 1000) {
+        group.bench_function(format!("{}_commands", command_count), |bencher| {
+            let mut world = World::default();
+            let mut command_queue = CommandQueue::default();
+
+            bencher.iter(|| {
+                let mut commands = Commands::new(&mut command_queue, &world);
+                for _ in 0..command_count {
+                    commands.add(T::default());
+                }
+                drop(commands);
+                command_queue.apply(&mut world);
+            });
+        });
+    }
+
+    group.finish();
+}
+
+fn zero_sized_commands(criterion: &mut Criterion) {
+    sized_commands_impl::<SizedCommand<()>>(criterion);
+}
+
+fn medium_sized_commands(criterion: &mut Criterion) {
+    sized_commands_impl::<SizedCommand<(u32, u32, u32)>>(criterion);
+}
+
+fn large_sized_commands(criterion: &mut Criterion) {
+    sized_commands_impl::<SizedCommand<LargeStruct>>(criterion);
+}