Optimize rollback hot paths

Reviewed-on: #52
Co-authored-by: Álex Román Núñez <eirexe@eirteam.moe>

.gitignore

@@ -10,3 +10,7 @@ imported_assets
 *_old.glb
 /replay.json
 *.replay.tmp
+
+# Python debugger formatter tests
+__pycache__/
+*.pyc

README.md

@@ -2,6 +2,15 @@
 
 Fancy heavily advanced platform fighting game
 
+## Debugging fixed-point values with lldb
+
+If using CodeLLDB:
+```json
+"preRunCommands": [
+    "command script import ${workspaceFolder}/tools/lldb_fixed.py"
+]
+```
+
 ## Direct P2P Debugging
 
 ```text
@@ -16,18 +25,9 @@ cargo run -- --debug-match --players 2 --local-port 7001 \
 
 ## Production Build
 
-Production builds use the `production` Cargo profile and do not dynamically
-link Bevy or Rust. Development builds retain Bevy dynamic linking for faster
-iteration. Cargo stores the production profile under `target/production`,
-separate from development artifacts under `target/debug`.
-
 ```sh
 cargo run --locked \
   --bin process-assets --no-default-features --features debug_features
 cargo build --locked \
   --profile production --no-default-features --features debug_features
-```
+```
-
-The asset preprocessing command must be run from the repository root. It
-populates `imported_assets/Default`, which is then consumed by the production
-game without running the asset processor at startup.

src/fighter/baked_animation/mod.rs

@@ -4,8 +4,8 @@ mod asset_baking;
 mod runtime;
 mod visual_binding;
 
-pub use asset_baking::{BakedAnimationAssetProcessingPlugin, BakedAnimationPlugin};
 pub use crate::math::fixed_transform::FixedMat4;
+pub use asset_baking::{BakedAnimationAssetProcessingPlugin, BakedAnimationPlugin};
 pub use runtime::{
     BakedAnimationClip, BakedFighterAnimations, FighterBoneIndexMap, FighterBoneMatrices,
     update_fighter_bone_matrices,
@@ -164,6 +164,20 @@ mod tests {
         );
     }
 
+    #[test]
+    fn repeated_pose_updates_reuse_scratch_and_output_buffers() {
+        let mut matrices = FighterBoneMatrices::default();
+        let names = ["root".into(), "hand".into()];
+        let parents = [None, Some(0)];
+        let pose = [FixedMat4::IDENTITY, FixedMat4::IDENTITY];
+
+        matrices.set_pose(&names, &parents, &pose, FGi32::ONE);
+        let capacities = matrices.buffer_capacities();
+        matrices.set_pose(&names, &parents, &pose, FGi32::ONE);
+
+        assert_eq!(capacities, matrices.buffer_capacities());
+    }
+
     #[test]
     fn model_scale_affects_gameplay_bones_without_changing_visual_locals() {
         let mut root = FixedMat4::IDENTITY;

src/fighter/baked_animation/runtime.rs

@@ -3,7 +3,7 @@ use std::collections::HashMap;
 use bevy::prelude::*;
 use serde::{Deserialize, Serialize};
 
-use super::{BAKED_ANIMATION_FPS};
+use super::BAKED_ANIMATION_FPS;
 
 use crate::{
     fighter::{
@@ -12,8 +12,8 @@ use crate::{
         manifest::FighterManifest,
         visual::FighterAnimations,
     },
+    math::fixed_transform::FixedMat4,
     math::int::FGi32,
-    math::fixed_transform::FixedMat4
 };
 
 #[derive(Clone, Debug, Serialize, Deserialize)]
@@ -70,12 +70,24 @@ pub struct FighterBoneMatrices {
     previous_frame_matrices: Vec<FixedMat4>,
     local_matrices: Vec<FixedMat4>,
     previous_local_matrices: Vec<FixedMat4>,
+    model_matrices: Vec<Option<FixedMat4>>,
 }
 
 #[derive(Component, Clone, Debug, Default)]
 pub struct FighterBoneIndexMap(pub HashMap<String, usize>);
 
 impl FighterBoneMatrices {
+    #[cfg(test)]
+    pub(super) fn buffer_capacities(&self) -> [usize; 5] {
+        [
+            self.matrices.capacity(),
+            self.local_matrices.capacity(),
+            self.previous_frame_matrices.capacity(),
+            self.previous_local_matrices.capacity(),
+            self.model_matrices.capacity(),
+        ]
+    }
+
     pub fn get(&self, bone: &str) -> Option<(FixedMat4, FixedMat4)> {
         let index = self.bone_names.iter().position(|name| name == bone)?;
         Some((
@@ -142,13 +154,19 @@ impl FighterBoneMatrices {
             assert_eq!(self.bone_names, bone_names, "baked bone names changed");
             assert_eq!(self.parents, parents, "baked bone parents changed");
         }
+        self.matrices.reserve(local.len());
+        self.previous_frame_matrices.reserve(local.len());
+        self.local_matrices.reserve(local.len());
+        self.previous_local_matrices.reserve(local.len());
+        self.model_matrices.reserve(local.len());
         std::mem::swap(&mut self.matrices, &mut self.previous_frame_matrices);
         std::mem::swap(&mut self.local_matrices, &mut self.previous_local_matrices);
         self.local_matrices.clear();
         self.local_matrices.extend_from_slice(local);
-        let mut model = vec![None; local.len()];
+        self.model_matrices.resize(local.len(), None);
+        self.model_matrices.fill(None);
         for index in 0..local.len() {
-            resolve_model_transform(index, parents, local, &mut model)
+            resolve_model_transform(index, parents, local, &mut self.model_matrices)
                 .expect("baked bone hierarchy should be valid");
         }
         let model_scale = FixedMat4 {
@@ -159,10 +177,12 @@ impl FighterBoneMatrices {
                 [FGi32::ZERO, FGi32::ZERO, FGi32::ZERO, FGi32::ONE],
             ],
         };
-        self.matrices = model
+        self.matrices.clear();
-            .into_iter()
+        self.matrices.extend(
-            .map(|matrix| model_scale.mul(matrix.expect("baked bone hierarchy should be valid")))
+            self.model_matrices.iter().map(|matrix| {
-            .collect();
+                model_scale.mul(matrix.expect("baked bone hierarchy should be valid"))
+            }),
+        );
     }
 }
 

src/fighter/collision.rs

@@ -50,31 +50,29 @@ pub fn air_collide_with_stage(
         translation.0 + ecb.get_bottom_point(),
     );
 
-    let mut intersect_results = vec![];
+    stage_collision
-
+        .stage_polys
-    for (poly_idx, poly) in stage_collision.stage_polys.iter().enumerate() {
-        if let Some(intersect_result) = poly.intersect_ray(ray_segment) {
-            intersect_results.push((
-                AirCollisionWithStageResult {
-                    line_id: {
-                        StageLineID {
-                            polygon: poly_idx,
-                            segment: intersect_result.segment_idx,
-                        }
-                    },
-                    hit_position: intersect_result.position,
-                    hit_normal: intersect_result.normal,
-                },
-                intersect_result
-                    .position
-                    .distance_squared(ray_segment.point2()),
-            ));
-        }
-    }
-    intersect_results.sort_by(|(_, a), (_, b)| a.cmp(b));
-
-    intersect_results
         .iter()
-        .last()
+        .enumerate()
-        .map(|(val, _)| val.to_owned().clone())
+        .filter_map(|(poly_idx, poly)| {
+            poly.intersect_ray(ray_segment).map(|intersect_result| {
+                (
+                    AirCollisionWithStageResult {
+                        line_id: {
+                            StageLineID {
+                                polygon: poly_idx,
+                                segment: intersect_result.segment_idx,
+                            }
+                        },
+                        hit_position: intersect_result.position,
+                        hit_normal: intersect_result.normal,
+                    },
+                    intersect_result
+                        .position
+                        .distance_squared(ray_segment.point2()),
+                )
+            })
+        })
+        .max_by(|(_, a), (_, b)| a.cmp(b))
+        .map(|(result, _)| result)
 }

src/replay.rs

@@ -152,6 +152,20 @@ fn validate(replay: &Replay) -> Result<(), String> {
     {
         return Err("replay frames must be strictly ordered".into());
     }
+    if replay
+        .frames
+        .windows(2)
+        .any(|frames| frames[0].frame.checked_add(1) != Some(frames[1].frame))
+    {
+        return Err("replay frames must be contiguous".into());
+    }
+    if replay
+        .checksums
+        .windows(2)
+        .any(|checksums| checksums[0].frame >= checksums[1].frame)
+    {
+        return Err("replay checksums must be strictly ordered".into());
+    }
     if replay
         .frames
         .iter()
@@ -173,6 +187,21 @@ impl Replay {
         handles.sort_unstable();
         handles != (0..handles.len()).collect::<Vec<_>>()
     }
+
+    fn frame(&self, frame: i32) -> Option<&ReplayFrame> {
+        let first_frame = self.frames.first()?.frame;
+        let index = frame.checked_sub(first_frame)? as usize;
+        self.frames
+            .get(index)
+            .filter(|recorded| recorded.frame == frame)
+    }
+
+    fn checksum(&self, frame: i32) -> Option<u128> {
+        self.checksums
+            .binary_search_by_key(&frame, |checksum| checksum.frame)
+            .ok()
+            .map(|index| self.checksums[index].checksum)
+    }
 }
 
 fn start_replay(mut commands: Commands, args: Res<Args>, loaded: Option<Res<LoadedReplay>>) {
@@ -210,11 +239,7 @@ fn replay_inputs(
                 .insert(frame.0, inputs.iter().map(|(input, _)| *input).collect());
         }
         ReplayMode::Playback { replay, mismatch } if !*mismatch => {
-            let Some(recorded) = replay
+            let Some(recorded) = replay.frame(frame.0) else {
-                .frames
-                .iter()
-                .find(|recorded| recorded.frame == frame.0)
-            else {
                 error!("Replay has no input for frame {}", frame.0);
                 *mismatch = true;
                 return;
@@ -235,21 +260,15 @@ fn record_checksum(
     let Some(mut runtime) = runtime else { return };
     if matches!(&runtime.mode, ReplayMode::Record { .. }) {
         runtime.checksums.insert(frame.0, checksum.0);
-    } else if let ReplayMode::Playback { replay, mismatch } = &mut runtime.mode {
+    } else if let ReplayMode::Playback { replay, mismatch } = &mut runtime.mode
-        if let Some(expected) = replay
+        && let Some(expected) = replay.checksum(frame.0)
-            .checksums
+        && expected != checksum.0
-            .iter()
+    {
-            .find(|value| value.frame == frame.0)
+        error!(
-            .map(|value| value.checksum)
+            "Replay desync at frame {}: expected {:X}, got {:X}",
-        {
+            frame.0, expected, checksum.0
-            if expected != checksum.0 {
+        );
-                error!(
+        *mismatch = true;
-                    "Replay desync at frame {}: expected {:X}, got {:X}",
-                    frame.0, expected, checksum.0
-                );
-                *mismatch = true;
-            }
-        }
     }
 }
 
@@ -407,4 +426,41 @@ mod tests {
             "unsupported replay version 1"
         );
     }
+
+    #[test]
+    fn playback_lookups_use_frame_offsets_and_support_sparse_checksums() {
+        let mut replay = sample();
+        replay.frames[0].frame = 10;
+        replay.frames.push(ReplayFrame {
+            frame: 11,
+            inputs: replay.frames[0].inputs.clone(),
+        });
+        replay.checksums.push(ReplayChecksum {
+            frame: 11,
+            checksum: 84,
+        });
+        replay.checksums[0].frame = 10;
+        validate(&replay).unwrap();
+
+        assert_eq!(replay.frame(10).unwrap().frame, 10);
+        assert_eq!(replay.frame(11).unwrap().frame, 11);
+        assert!(replay.frame(9).is_none());
+        assert_eq!(replay.checksum(10), Some(42));
+        assert_eq!(replay.checksum(11), Some(84));
+        assert_eq!(replay.checksum(12), None);
+    }
+
+    #[test]
+    fn noncontiguous_input_frames_are_rejected() {
+        let mut replay = sample();
+        replay.frames.push(ReplayFrame {
+            frame: 3,
+            inputs: replay.frames[0].inputs.clone(),
+        });
+
+        assert_eq!(
+            validate(&replay).unwrap_err(),
+            "replay frames must be contiguous"
+        );
+    }
 }

src/ui/fighter_info.rs

@@ -1,13 +1,17 @@
 use bevy::prelude::*;
 
-use crate::{fighter::{attack::FighterDamage, manifest::{FighterManifest}}, match_loading::MatchPlayer, player::Player};
+use crate::{
+    fighter::{attack::FighterDamage, manifest::FighterManifest}, match_loading::{MatchPlayer}, match_runtime::MatchState, player::Player, ui::stock_icon::BattleUIStockIcon,
+};
 
 #[derive(Component, FromTemplate)]
 pub struct FighterInfoUI {
     percentage_text: Entity,
+    stock_icon_container: Entity,
     _name_text: Entity,
     player_handle: usize,
     visible_damage: FighterDamage,
+    visible_stocks: u8,
 }
 
 impl FighterInfoUI {
@@ -16,6 +20,7 @@ impl FighterInfoUI {
         bsn! {
             FighterInfoUI {
                 percentage_text: #PercentageText,
+                stock_icon_container: #StockIconContainer,
                 _name_text: #NameText,
                 player_handle: {match_player.handle}
             }
@@ -25,6 +30,10 @@ impl FighterInfoUI {
                 justify_content: JustifyContent::Center,
             }
             Children [
+                (
+                    #StockIconContainer
+                    Node
+                )
                 (
                     #PercentageText
                     Text("0.00%")
@@ -41,17 +50,54 @@ impl FighterInfoUI {
     }
 }
 
-pub fn update_fighter_info_damage(fighters: Query<(&Player, &FighterDamage)>, ui_elements: Query<&mut FighterInfoUI>, mut commands: Commands) {
+pub fn update_fighter_info_damage(
+    fighters: Query<(&Player, &FighterDamage)>,
+    ui_elements: Query<&mut FighterInfoUI>,
+    mut commands: Commands,
+) {
     for mut ui_element in ui_elements {
-        let Some((_, damage)) = fighters.iter().find(|(player, _)| {
+        let Some((_, damage)) = fighters
-            player.handle == ui_element.player_handle
+            .iter()
-        }) else {
+            .find(|(player, _)| player.handle == ui_element.player_handle)
+        else {
             continue;
         };
 
         if *damage != ui_element.visible_damage {
-            commands.entity(ui_element.percentage_text).insert(Text(format!("{}%", damage)));
+            commands
+                .entity(ui_element.percentage_text)
+                .insert(Text(format!("{}%", damage)));
             ui_element.visible_damage = damage.clone();
         }
     }
-}
+}
+
+pub fn update_fighter_info_stocks(
+    match_state: Res<MatchState>,
+    ui_elements: Query<&mut FighterInfoUI>,
+    mut commands: Commands,
+) {
+
+    for mut ui_element in ui_elements {
+        let handle = ui_element.player_handle;
+        let Some(match_player) = match_state.players.get(handle) else {
+            continue;
+        };
+
+        let Some(stocks) = match_player.stocks_remaining else {
+            continue;
+        };
+
+        if stocks != ui_element.visible_stocks {
+            commands
+                .entity(ui_element.stock_icon_container)
+                .despawn_children()
+                .with_children(|spawner| {
+                    for _ in 0..stocks {
+                        spawner.spawn_empty().apply_scene(BattleUIStockIcon::scene());
+                    }
+                });
+            ui_element.visible_stocks = stocks;
+        }
+    }
+}

src/ui/mod.rs

@@ -1,10 +1,15 @@
 use bevy::prelude::*;
 
-use crate::AppState;
+use crate::{
+    AppState,
+    match_loading::{MatchConfig, MatchRules},
+    match_runtime::MatchState,
+};
 
 pub mod battle_ui;
 pub mod errors;
 pub mod fighter_info;
+pub mod stock_icon;
 
 #[derive(Default)]
 pub struct InGameUIPlugin;
@@ -13,6 +18,20 @@ impl Plugin for InGameUIPlugin {
     fn build(&self, app: &mut App) {
         app.add_systems(OnEnter(AppState::InMatch), battle_ui::spawn_battle_ui)
             .add_systems(OnExit(AppState::InMatch), battle_ui::cleanup_battle_ui)
-            .add_systems(PostUpdate, fighter_info::update_fighter_info_damage);
+            .add_systems(PostUpdate, fighter_info::update_fighter_info_damage)
+            .add_systems(
+                PostUpdate,
+                fighter_info::update_fighter_info_stocks.run_if(
+                    resource_exists::<MatchState>
+                        .and_then(resource_exists::<MatchConfig>)
+                        .and_then(|match_config: Res<MatchConfig>| {
+                            if let MatchRules::Stock { .. } = match_config.rules {
+                                true
+                            } else {
+                                false
+                            }
+                        }),
+                ),
+            );
     }
 }

src/ui/stock_icon.rs

@@ -0,0 +1,17 @@
+use bevy::prelude::*;
+
+#[derive(SceneComponent, Clone, FromTemplate)]
+pub struct BattleUIStockIcon;
+
+impl BattleUIStockIcon {
+    pub fn scene() -> impl Scene {
+        bsn! {
+            Node {
+                width: px(32),
+                height: px(32),
+                border_radius: BorderRadius::MAX,
+            }
+            BackgroundColor({bevy::color::palettes::css::RED})
+        }
+    }
+}

tools/lldb_fixed.py

@@ -0,0 +1,123 @@
+"""LLDB summaries for Rust's `fixed` crate.
+
+Load this file from LLDB with:
+
+    command script import /path/to/shinespark/tools/lldb_fixed.py
+
+The implementation does not use floating-point arithmetic, so displayed values
+are exact even for FixedI128/FixedU128 values.
+"""
+
+import re
+
+try:
+    import lldb
+except ImportError:
+    # Keep the pure parsing and formatting helpers testable without LLDB.
+    lldb = None
+
+
+_FIXED_TYPE_RE = re.compile(r"(?:^|::)Fixed([IU])(8|16|32|64|128)<(.+)>$")
+_DIRECT_TYPENUM_RE = re.compile(r"(?:^|::)U(\d+)$")
+_TYPENUM_BIT_RE = re.compile(r"(?:^|::)B([01])(?=\s*[,>])")
+
+
+def parse_fixed_type(type_name):
+    """Return (signed, storage_bits, fractional_bits), or None."""
+    if not type_name:
+        return None
+
+    match = _FIXED_TYPE_RE.search(type_name.replace(" ", ""))
+    if match is None:
+        return None
+
+    signed = match.group(1) == "I"
+    storage_bits = int(match.group(2))
+    fractional_type = match.group(3)
+
+    direct = _DIRECT_TYPENUM_RE.search(fractional_type)
+    if direct is not None:
+        fractional_bits = int(direct.group(1))
+    else:
+        # typenum represents UInt<N, B> as a binary tree. Its textual Rust
+        # type lists the bits from most significant to least significant.
+        digits = _TYPENUM_BIT_RE.findall(fractional_type)
+        if not digits:
+            return None
+        fractional_bits = int("".join(digits), 2)
+
+    if fractional_bits > storage_bits:
+        return None
+    return signed, storage_bits, fractional_bits
+
+
+def normalize_bits(raw_bits, signed, storage_bits):
+    """Interpret a possibly unsigned LLDB integer using the fixed type's width."""
+    mask = (1 << storage_bits) - 1
+    raw_bits &= mask
+    if signed and raw_bits >= 1 << (storage_bits - 1):
+        raw_bits -= 1 << storage_bits
+    return raw_bits
+
+
+def format_fixed(raw_bits, fractional_bits):
+    """Format a scaled integer exactly as a decimal string."""
+    if fractional_bits == 0:
+        return str(raw_bits)
+
+    negative = raw_bits < 0
+    magnitude = -raw_bits if negative else raw_bits
+    whole = magnitude >> fractional_bits
+    remainder = magnitude & ((1 << fractional_bits) - 1)
+
+    if remainder == 0:
+        result = str(whole)
+    else:
+        # remainder / 2^n == (remainder * 5^n) / 10^n, which gives a finite,
+        # exact decimal representation without converting through a float.
+        decimal = str(remainder * 5**fractional_bits).zfill(fractional_bits)
+        result = f"{whole}.{decimal.rstrip('0')}"
+
+    return f"-{result}" if negative else result
+
+
+def _parse_lldb_integer(value):
+    text = value.GetValue()
+    if text is not None:
+        try:
+            return int(text, 0)
+        except ValueError:
+            pass
+
+    # GetValueAsUnsigned is also useful on LLDB versions that do not render
+    # primitive values through GetValue. It is sufficient for all <= 64-bit
+    # fixed types; current LLDB versions render i128/u128 through GetValue.
+    return value.GetValueAsUnsigned()
+
+
+def fixed_summary(value, _internal_dict):
+    layout = parse_fixed_type(value.GetTypeName())
+    if layout is None:
+        return None
+
+    signed, storage_bits, fractional_bits = layout
+    bits = value.GetChildMemberWithName("bits")
+    if not bits.IsValid():
+        return None
+
+    raw_bits = _parse_lldb_integer(bits)
+    raw_bits = normalize_bits(raw_bits, signed, storage_bits)
+    return format_fixed(raw_bits, fractional_bits)
+
+
+def __lldb_init_module(debugger, _internal_dict):
+    # LLDB's type-name regex sees the concrete fixed type rather than project
+    # aliases such as FGi32 and FGWide.
+    debugger.HandleCommand(
+        "type summary add --python-function lldb_fixed.fixed_summary "
+        "--category Shinespark --regex "
+        "'^fixed::Fixed(I|U)(8|16|32|64|128)<.+>$'"
+    )
+    # A dedicated category makes the formatter work even when CodeLLDB has not
+    # enabled rustc's standard `Rust` formatter category yet.
+    debugger.HandleCommand("type category enable Shinespark")

tools/test_lldb_fixed.py

@@ -0,0 +1,58 @@
+import unittest
+
+import lldb_fixed
+
+
+class ParseFixedTypeTests(unittest.TestCase):
+    def test_direct_typenum_alias(self):
+        self.assertEqual(
+            lldb_fixed.parse_fixed_type("fixed::FixedI32<typenum::uint::U16>"),
+            (True, 32, 16),
+        )
+
+    def test_expanded_typenum(self):
+        type_name = (
+            "fixed::FixedI32<"
+            "typenum::uint::UInt<"
+            "typenum::uint::UInt<"
+            "typenum::uint::UInt<"
+            "typenum::uint::UInt<"
+            "typenum::uint::UInt<typenum::uint::UTerm, typenum::bit::B1>, "
+            "typenum::bit::B0>, typenum::bit::B0>, typenum::bit::B0>, "
+            "typenum::bit::B0>>"
+        )
+        self.assertEqual(lldb_fixed.parse_fixed_type(type_name), (True, 32, 16))
+
+    def test_unsigned_type(self):
+        self.assertEqual(
+            lldb_fixed.parse_fixed_type("fixed::FixedU64<typenum::uint::U24>"),
+            (False, 64, 24),
+        )
+
+    def test_unrelated_type(self):
+        self.assertIsNone(lldb_fixed.parse_fixed_type("core::primitive::i32"))
+
+
+class FormatFixedTests(unittest.TestCase):
+    def test_q16_16(self):
+        self.assertEqual(lldb_fixed.format_fixed(3 << 15, 16), "1.5")
+        self.assertEqual(lldb_fixed.format_fixed(-(3 << 15), 16), "-1.5")
+        self.assertEqual(lldb_fixed.format_fixed(1, 16), "0.0000152587890625")
+
+    def test_integer_value_has_no_decimal_suffix(self):
+        self.assertEqual(lldb_fixed.format_fixed(12 << 16, 16), "12")
+
+    def test_q96_32(self):
+        raw = (1 << 100) + (1 << 31)
+        self.assertEqual(
+            lldb_fixed.format_fixed(raw, 32),
+            "295147905179352825856.5",
+        )
+
+    def test_twos_complement_normalization(self):
+        self.assertEqual(lldb_fixed.normalize_bits(0xFFFF8000, True, 32), -32768)
+        self.assertEqual(lldb_fixed.normalize_bits(0xFFFF8000, False, 32), 0xFFFF8000)
+
+
+if __name__ == "__main__":
+    unittest.main()