render: improve fix for rendering shared 3d modules

lyra-game/src/render/render_job.rs

@@ -1,17 +1,21 @@
 use lyra_ecs::Entity;
 
+use super::transform_buffer_storage::TransformIndex;
+
 pub struct RenderJob {
     pub entity: Entity,
     pub shader_id: u64,
     pub mesh_uuid: uuid::Uuid,
+    pub transform_id: TransformIndex,
 }
 
 impl RenderJob {
-    pub fn new(entity: Entity, shader_id: u64, mesh_buffer_id: uuid::Uuid) -> Self {
+    pub fn new(entity: Entity, shader_id: u64, mesh_buffer_id: uuid::Uuid, transform_id: TransformIndex) -> Self {
         Self {
             entity,
             shader_id,
             mesh_uuid: mesh_buffer_id,
+            transform_id
         }
     }
 }

lyra-game/src/render/renderer.rs

@@ -29,7 +29,7 @@ use super::light_cull_compute::LightCullCompute;
 use super::material::Material;
 use super::render_buffer::BufferWrapper;
 use super::texture::RenderTexture;
-use super::transform_buffer_storage::TransformBuffers;
+use super::transform_buffer_storage::{TransformBuffers, TransformGroup};
 use super::vertex::Vertex;
 use super::{render_pipeline::FullRenderPipeline, render_buffer::BufferStorage, render_job::RenderJob};
 
@@ -374,12 +374,13 @@ impl BasicRenderer {
 
     /// Processes the mesh for the renderer, storing and creating buffers as needed. Returns true if a new mesh was processed.
     fn process_mesh(&mut self, entity: Entity, transform: Transform, mesh: &Mesh, mesh_uuid: Uuid) -> bool {
-        if self.transform_buffers.should_expand() {
+        let _ = transform;
+        /* if self.transform_buffers.should_expand() {
             self.transform_buffers.expand_buffers(&self.device);
         }
 
         self.transform_buffers.update_or_insert(&self.queue, &self.render_limits,
-            mesh_uuid, || ( transform.calculate_mat4(), glam::Mat3::from_quat(transform.rotation) ));
+            entity, || ( transform.calculate_mat4(), glam::Mat3::from_quat(transform.rotation) )); */
 
         #[allow(clippy::map_entry)]
         if !self.mesh_buffers.contains_key(&mesh_uuid) {
@@ -451,10 +452,14 @@ impl Renderer for BasicRenderer {
                         self.check_mesh_buffers(entity, &mesh_han);
                     }
                     
+                    let group = TransformGroup::EntityRes(entity, mesh_han.uuid());
+                    let transform_id = self.transform_buffers.update_or_push(&self.queue, &self.render_limits,
+                        group, || ( interop_pos.calculate_mat4(), glam::Mat3::from_quat(interop_pos.rotation) ));
+
                     let material = mesh.material.as_ref().unwrap()
                         .data_ref().unwrap();
                     let shader = material.shader_uuid.unwrap_or(0);
-                    let job = RenderJob::new(entity, shader, mesh_han.uuid());
+                    let job = RenderJob::new(entity, shader, mesh_han.uuid(), transform_id);
                     self.render_jobs.push_back(job);
                 }
             }
@@ -478,11 +483,16 @@ impl Renderer for BasicRenderer {
                                         && scene_epoch == last_epoch {
                                     self.check_mesh_buffers(entity, &mesh_han);
                                 }
+
+                                let scene_mesh_group = TransformGroup::Res(scene_han.uuid(), mesh_han.uuid());
+                                let group = TransformGroup::OwnedGroup(entity, scene_mesh_group.into());
+                                let transform_id = self.transform_buffers.update_or_push(&self.queue, &self.render_limits,
+                                    group, || ( mesh_interpo.calculate_mat4(), glam::Mat3::from_quat(mesh_interpo.rotation) ));
                                 
                                 let material = mesh.material.as_ref().unwrap()
                                     .data_ref().unwrap();
                                 let shader = material.shader_uuid.unwrap_or(0);
-                                let job = RenderJob::new(entity, shader, mesh_han.uuid());
+                                let job = RenderJob::new(entity, shader, mesh_han.uuid(), transform_id);
                                 self.render_jobs.push_back(job);
                             }
                         }
@@ -576,9 +586,8 @@ impl Renderer for BasicRenderer {
                     }
 
                     // Get the bindgroup for job's transform and bind to it using an offset.
-                    let transform_indices = *self.transform_buffers.transform_indices(job.mesh_uuid).unwrap();
+                    let bindgroup = self.transform_buffers.bind_group(job.transform_id);
-                    let bindgroup = self.transform_buffers.bind_group(transform_indices).unwrap();
+                    let offset = self.transform_buffers.buffer_offset(job.transform_id);
-                    let offset = TransformBuffers::index_offset(&self.render_limits, transform_indices) as u32;
                     render_pass.set_bind_group(1, bindgroup, &[ offset, offset, ]);
 
                     render_pass.set_bind_group(2, &self.camera_buffer.bindgroup(), &[]);

lyra-game/src/render/transform_buffer_storage.rs

@@ -1,44 +1,167 @@
-use std::{collections::{VecDeque, HashMap}, num::NonZeroU64};
+use std::{collections::{HashMap, VecDeque}, hash::{BuildHasher, DefaultHasher, Hash, Hasher, RandomState}, num::NonZeroU64};
 
+use lyra_ecs::Entity;
 use uuid::Uuid;
 use wgpu::Limits;
 
 use std::mem;
 
+/// A group id created from a [`TransformGroup`].
+/// 
+/// This is mainly created so that [`TransformGroup::OwnedGroup`] can use another group inside of it.
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub struct TransformGroupId(u64);
+
+impl From<TransformGroup> for TransformGroupId {
+    fn from(value: TransformGroup) -> Self {
+        let mut hasher = DefaultHasher::new();
+        value.hash(&mut hasher);
+        let hash = hasher.finish();
+        
+        TransformGroupId(hash)
+    }
+}
+
+/// Used as a key into the [`TransformBuffers`].
+/// 
+/// This enum is used as a key to identify a transform for a RenderJob. The renderer uses this
+/// to differentiate a transform between two entities that share a resource handle to the same
+/// scene:
+/// ```nobuild
+/// // The group of the mesh in the scene.
+/// let scene_mesh_group = TransformGroup::Res(scene_handle.uuid(), mesh_handle.uuid());
+/// // The group of the owned entity that has mesh in a scene.
+/// let finished_group = TransformGroup::OwnedGroup(entity, scene_mesh_group.into());
+/// ```
+/// 
+/// A simpler example of the use of a transform group is when processing lone mesh handles
+/// owned by entities:
+/// ```nobuild
+/// let group = TransformGroup::EntityRes(entity, mesh_handle.uuid());
+/// ```
+/// 
+/// These were made to fix [#6](https://git.seanomik.net/SeanOMik/lyra-engine/issues/6).
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
+pub enum TransformGroup {
+    /// Just an entity.
+    Entity(Entity),
+    /// An entity that owns another group.
+    OwnedGroup(Entity, TransformGroupId),
+    /// A resource uuid grouped with an owning Entity.
+    EntityRes(Entity, Uuid),
+    /// A resource uuid grouped with another resource uuid.
+    Res(Uuid, Uuid),
+}
+
+/// The index of a specific Transform inside of the buffers.
 #[derive(Default, Copy, Clone, PartialEq, Eq, Debug)]
-pub(crate) struct TransformBufferIndices {
+pub struct TransformIndex {
-    pub buffer_index: usize,
+    /// The index of the entry in the buffer chain.
-    pub transform_index: usize,
+    entry_index: usize,
+    /// The index of the transform in the entry.
+    transform_index: usize,
 }
 
 /// A struct representing a single transform buffer. There can be multiple of these
-pub(crate) struct TransformBufferEntry {
+struct BufferEntry {
     pub len: usize,
-    pub transform_buf: wgpu::Buffer,
-    pub normal_mat_buf: wgpu::Buffer,
     pub bindgroup: wgpu::BindGroup,
+    pub transform_buffer: wgpu::Buffer,
+    pub normal_buffer: wgpu::Buffer,
+}
+
+/// A HashMap that caches values for reuse.
+/// 
+/// The map detects dead values by tracking which entries were not updated since the last time
+/// [`CachedValMap::update`] was ran. When dead values are collected, they can be reused on an
+/// [`insert`](CachedValMap::insert) into the map.
+struct CachedValMap<K, V, S = RandomState> {
+    latest: HashMap<K, V, S>,
+    old: HashMap<K, V, S>,
+    dead: VecDeque<V>,
+}
+
+impl<K, V, S: Default> Default for CachedValMap<K, V, S> {
+    fn default() -> Self {
+        Self {
+            latest: Default::default(),
+            old: Default::default(),
+            dead: Default::default()
+        }
+    }
+}
+
+#[allow(dead_code)]
+impl<K: Hash + Eq + PartialEq + Clone, V: Clone, S: BuildHasher> CachedValMap<K, V, S> {
+    /// Insert a key, possibly reusing a value in the map.
+    /// 
+    /// Returns the reused value, if one was reused. If its `None`, then the value was retrieved
+    /// from running `val_fn`.
+    pub fn insert<F>(&mut self, key: K, mut val_fn: F) -> Option<V>
+    where
+        F: FnMut() -> V
+    {
+        if self.latest.contains_key(&key) {
+            self.latest.insert(key, val_fn());
+            None
+        } else {
+            let val = self.dead.pop_front()
+                .unwrap_or_else(val_fn);
+            self.latest.insert(key, val.clone());
+            Some(val)
+        }
+    }
+
+    /// Returns a reference to the value corresponding to the key.
+    pub fn get(&mut self, key: K) -> Option<&V> {
+        if let Some(v) = self.old.remove(&key) {
+            self.latest.insert(key.clone(), v);
+        }
+
+        self.latest.get(&key)
+    }
+
+    /// Keep a key alive without updating its value.
+    pub fn keep_alive(&mut self, key: K) {
+        if let Some(v) = self.old.remove(&key) {
+            self.latest.insert(key, v);
+        }
+    }
+
+    /// Returns `true` if the map contains a value for the specified key.
+    pub fn contains(&self, key: K) -> bool {
+        self.old.contains_key(&key) || self.latest.contains_key(&key)
+    }
+
+    /// Collects the now dead values for reuse.
+    /// 
+    /// This detects dead values by tracking which entries were not updated since the last time
+    /// update was ran.
+    pub fn update(&mut self) {
+        // drain the dead values into the dead queue
+        self.dead.extend(self.old.drain().map(|(_, v)| v));
+
+        // now drain the latest entries into the old entries
+        self.old.extend(self.latest.drain());
+    }
 }
 
 /// A helper struct for managing the Transform buffers for meshes.
 /// 
-/// This struct manages a "chain" of uniform buffers that store Transform for meshes. When
+/// This struct manages a "chain" of uniform buffers that store Transform for [`TransformGroup`]s.
-/// first created it only has a single "chain-link" with a buffer that is the maximum length
+/// When first created it only has a single "chain-link" with a buffer that is the maximum length
 /// the GPU supports. When the first buffer fills up, a new one should be created which will also
 /// be the maximum length the GPU supports. When the new buffer fills up, a new one will be
 /// created once again, and so on.
 /// 
-/// `Uuid`'s are used to represent entries (usually Meshes) in the buffer. The Uuid's can be used
+/// [`TransformGroup`]s are used to represent entries in the buffer. They are used to insert,
-/// to insert, update, and retrieve the transforms.
+/// update, and retrieve the transforms.
-pub(crate) struct TransformBuffers {
+pub struct TransformBuffers {
     pub bindgroup_layout: wgpu::BindGroupLayout,
-    pub just_updated: HashMap<Uuid, TransformBufferIndices>,
+    groups: CachedValMap<TransformGroupId, TransformIndex>,
-    pub not_updated: HashMap<Uuid, TransformBufferIndices>,
+    entries: Vec<BufferEntry>,
-    pub dead_indices: VecDeque<TransformBufferIndices>,
+    limits: wgpu::Limits,
-    pub next_indices: TransformBufferIndices,
+    max_transform_count: usize,
-    /// (transform count, buffer, bindgroup)
-    pub buffer_bindgroups: Vec<TransformBufferEntry>,
-    /// The max amount of transforms in a buffer
-    pub max_transform_count: usize,
 }
 
 impl TransformBuffers {
@@ -72,13 +195,11 @@ impl TransformBuffers {
         });
 
         let mut s = Self {
-            max_transform_count: limits.max_uniform_buffer_binding_size as usize / (mem::size_of::<glam::Mat4>() * 2),
-            buffer_bindgroups: Vec::new(),
             bindgroup_layout,
-            just_updated: HashMap::new(),
+            groups: Default::default(),
-            not_updated: HashMap::new(),
+            entries: Default::default(),
-            dead_indices: VecDeque::new(),
+            max_transform_count: (limits.max_uniform_buffer_binding_size / 2) as usize / (mem::size_of::<glam::Mat4>()),
-            next_indices: TransformBufferIndices::default(),
+            limits,
         };
 
         // create the first uniform buffer
@@ -87,106 +208,84 @@ impl TransformBuffers {
         s
     }
 
-    /// Update an transform in the buffer.
+    /// Update an existing transform in the buffers.
     /// 
     /// # Panics
-    /// Panics if the entity isn't stored, you can check if it is before with [`TransformBuffers:contains`].
+    /// Panics if the `entity_group` is not already inside of the buffers.
-    pub fn update_transform(&mut self, queue: &wgpu::Queue, limits: &Limits, uuid: Uuid, transform: glam::Mat4, normal_matrix: glam::Mat3) -> TransformBufferIndices {
+    pub fn update_transform(&mut self, queue: &wgpu::Queue, limits: &Limits, entity_group: TransformGroup, transform: glam::Mat4, normal_matrix: glam::Mat3) -> TransformIndex {
-        let indices = self.not_updated.remove(&uuid)
+        let index = *self.groups.get(entity_group.into())
-            .or_else(|| self.just_updated.remove(&uuid))
+            .expect("Use 'push_transform' for new entities");
-            .expect("Use 'insert_entity' for new entities");
+        let entry = self.entries.get_mut(index.entry_index).unwrap();
-        self.just_updated.insert(uuid, indices);
 
         let normal_matrix = glam::Mat4::from_mat3(normal_matrix);
 
-        let buffer = self.buffer_bindgroups.get(indices.buffer_index).unwrap();
+        // write the transform and normal to the end of the transform
-        let offset = Self::index_offset(limits, indices);
+        let offset = Self::get_buffer_offset(limits, index) as _;
-        queue.write_buffer(&buffer.transform_buf, offset, bytemuck::bytes_of(&transform));
+        queue.write_buffer(&entry.transform_buffer, offset, bytemuck::bytes_of(&transform));
-        queue.write_buffer(&buffer.normal_mat_buf, offset, bytemuck::bytes_of(&normal_matrix));
+        queue.write_buffer(&entry.normal_buffer, offset, bytemuck::bytes_of(&normal_matrix));
-        indices
+
+        index
     }
 
-    /// Insert a new transform into the buffer, returns where in the buffer it was stored.
+    /// Push a new transform into the buffers.
-    pub fn insert_transform(&mut self, queue: &wgpu::Queue, limits: &Limits, uuid: Uuid, transform: glam::Mat4, normal_matrix: glam::Mat3) -> TransformBufferIndices {
+    pub fn push_transform(&mut self, queue: &wgpu::Queue, limits: &Limits, entity_group: TransformGroup, transform: glam::Mat4, normal_matrix: glam::Mat3) -> TransformIndex {
-        let indices = match self.dead_indices.pop_front() {
+        self.groups.insert(entity_group.into(), || {
-            Some(indices) => indices,
+            // this closure is only called when there are no values that can be reused,
-            None => {
+            // so we get a brand new index at the end of the last entry in the chain.
-                let indices = &mut self.next_indices;
+            let last = self.entries.last_mut().unwrap();
-                let this_idx = *indices;
+            
-                let entry = self.buffer_bindgroups.get_mut(indices.buffer_index).unwrap();
+            // ensure the gpu buffer is not overflown
+            debug_assert!(last.len < self.max_transform_count,
+                "Transform buffer is filled and 'next_indices' was not incremented! \
+                    Was a new buffer created?");
+            
+            let tidx = last.len;
+            last.len += 1;
 
-                if entry.len >= self.max_transform_count {
+            TransformIndex {
-                    panic!("Transform buffer is filled and 'next_indices' was not incremented! Was a new buffer created?");
+                entry_index: self.entries.len() - 1,
-                }
+                transform_index: tidx
-
-                entry.len += 1;
-                indices.transform_index += 1;
-                this_idx
             }
-        };
+        });
 
-        self.just_updated.insert(uuid, indices);
+        self.update_transform(queue, limits, entity_group, transform, normal_matrix)
-        self.update_transform(queue, limits, uuid, transform, normal_matrix)
-    }
-
-    /// Update or insert a transform
-    pub fn update_or_insert<TFn>(&mut self, queue: &wgpu::Queue, limits: &Limits, uuid: Uuid, transform_fn: TFn) -> TransformBufferIndices
-        where TFn: Fn() -> (glam::Mat4, glam::Mat3)
-    {   
-        let (tran, norm) = transform_fn();
-        if self.contains(uuid) {
-            self.update_transform(queue, limits, uuid, tran, norm)
-        } else {
-            self.insert_transform(queue, limits, uuid, tran, norm)
-        }
-    }
-
-    /// Returns true if the transform related to the `uuid` is stored (does not mean its up-to-date).
-    pub fn contains(&self, uuid: Uuid) -> bool {
-        self.not_updated.contains_key(&uuid) || self.just_updated.contains_key(&uuid)
     }
 
     /// Collect the dead transforms and prepare self to check next time.
     pub fn tick(&mut self) {
-        // take the dead entities, these were ones that were not updated this tick
+        self.groups.update();
-        let dead: VecDeque<TransformBufferIndices> = self.not_updated.values().copied().collect();
-        self.dead_indices = dead;
-
-        self.not_updated = self.just_updated.clone();
-        self.just_updated.clear();
     }
 
-    /// Returns the offset for the transform index in the buffer
+    /// Returns a boolean indicating if the buffer contains this group.
-    pub fn index_offset(limits: &Limits, indices: TransformBufferIndices) -> u64 {
+    pub fn contains(&self, group: TransformGroup) -> bool {
-        indices.transform_index as u64 * limits.min_uniform_buffer_offset_alignment as u64
+        self.groups.contains(group.into())
     }
 
-    /// Returns whether or not the transform buffers should be expanded
+    /// Update an existing transform group or if its not existing yet, pushes it to the buffer.
-    pub fn should_expand(&self) -> bool {
+    /// 
-        if let Some(entry) = self.buffer_bindgroups.last() {
+    /// Returns: the index that the transform is at in the buffers.
-            entry.len >= self.max_transform_count
+    pub fn update_or_push<F>(&mut self, queue: &wgpu::Queue, limits: &Limits, group: TransformGroup, transform_fn: F) -> TransformIndex
+        where F: Fn() -> (glam::Mat4, glam::Mat3)
+    {   
+        let (transform, normal_matrix) = transform_fn();
+        if self.contains(group) {
+            self.update_transform(queue, limits, group, transform, normal_matrix)
         } else {
-            true
+            self.push_transform(queue, limits, group, transform, normal_matrix)
         }
     }
 
-    /// Returns the bind group for the index
-    pub fn bind_group(&self, index: TransformBufferIndices) -> Option<&wgpu::BindGroup> {
-        self.buffer_bindgroups.get(index.buffer_index)
-            .map(|entry| &entry.bindgroup)
-    }
-
     /// Expand the Transform buffers by adding another uniform buffer binding.
     /// 
     /// This object has a chain of uniform buffers, when the buffers are expanded, a new
     /// "chain-link" is created.
     pub fn expand_buffers(&mut self, device: &wgpu::Device) {
         let limits = device.limits();
-        let max_buffer_sizes = (limits.max_uniform_buffer_binding_size as u64) / 2;
+        let max_buffer_sizes = self.max_transform_count as u64 * limits.min_uniform_buffer_offset_alignment as u64;
 
         let transform_buffer = device.create_buffer(
             &wgpu::BufferDescriptor {
-                label: Some(&format!("B_Transform_{}", self.buffer_bindgroups.len())),
+                label: Some(&format!("B_Transform_{}", self.entries.len())),
                 usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
                 size: max_buffer_sizes,
                 mapped_at_creation: false,
@@ -195,7 +294,7 @@ impl TransformBuffers {
 
         let normal_mat_buffer = device.create_buffer(
             &wgpu::BufferDescriptor {
-                label: Some(&format!("B_NormalMatrix_{}", self.buffer_bindgroups.len())),
+                label: Some(&format!("B_NormalMatrix_{}", self.entries.len())),
                 usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
                 size: max_buffer_sizes,
                 mapped_at_creation: false,
@@ -234,17 +333,43 @@ impl TransformBuffers {
             label: Some("BG_Transforms"),
         });
 
-        let entry = TransformBufferEntry {
+        let entry = BufferEntry {
             bindgroup: transform_bind_group,
-            transform_buf: transform_buffer,
+            transform_buffer,
-            normal_mat_buf: normal_mat_buffer,
+            normal_buffer: normal_mat_buffer,
             len: 0,
         };
-        self.buffer_bindgroups.push(entry);
+        self.entries.push(entry);
     }
 
-    /// Returns the indices of the Transform 
+    /// Returns the bind group for the transform index.
-    pub fn transform_indices(&self, uuid: Uuid) -> Option<&TransformBufferIndices> {
+    pub fn bind_group(&self, transform_id: TransformIndex) -> &wgpu::BindGroup {
-        self.just_updated.get(&uuid).or_else(|| self.not_updated.get(&uuid))
+        let entry = self.entries.get(transform_id.entry_index).unwrap();
+        &entry.bindgroup
     }
+
+    /// Get the buffer offset for a transform using wgpu limits.
+    /// 
+    /// If its possible to borrow immutably, use [`TransformBuffers::buffer_offset`].
+    fn get_buffer_offset(limits: &wgpu::Limits, transform_index: TransformIndex) -> u32 {
+        transform_index.transform_index as u32 * limits.min_uniform_buffer_offset_alignment as u32
+    }
+
+    /// Returns the offset of the transform inside the bind group buffer.
+    /// 
+    /// ```nobuild
+    /// let bindgroup = transform_buffers.bind_group(job.transform_id);
+    /// let offset = transform_buffers.buffer_offset(job.transform_id);
+    /// render_pass.set_bind_group(1, bindgroup, &[ offset, offset, ]);
+    /// ```
+    pub fn buffer_offset(&self, transform_index: TransformIndex) -> u32 {
+        Self::get_buffer_offset(&self.limits, transform_index)
+    }
+}
+
+#[repr(C)]
+#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
+struct TransformNormalMatPair {
+    transform: glam::Mat4,
+    normal_mat: glam::Mat4,
 }