feat: renderer: PBR, IBL, Passes by sp0lsh · Pull Request #38 · Pillware/Pill

sp0lsh · 2026-05-28T16:30:11Z

60k pills @ 42 FPS, macbook M1 1920x180. Single threaded yet, wasm is single threaded too.

https://github.com/KhronosGroup/glTF-Sample-Models/tree/d7a3cc8e51d7c573771ae77a57f16b0662a905c6/2.0/MetalRoughSpheres

https://github.com/KhronosGroup/glTF-Sample-Models/tree/d7a3cc8e51d7c573771ae77a57f16b0662a905c6/2.0/DamagedHelmet

sequenceDiagram
    autonumber
    participant ECS as ECS · SoA storage
    participant GP as Gameplay systems
    participant RS as RenderingSystem (stateless)
    participant RN as renderer.render()
    participant PB as PassPBROpaque
    participant GPU as GPU

    note over GP,ECS: 2. Game — mutate world
    GP->>ECS: pill_particle / hero / drift write Transform[]

    note over RS,RN: 3. PostGame — stateless plumbing
    RS->>ECS: borrow active scene
    RS->>RN: render(active_camera, scene, delta_time, resources)
    RN->>RN: get_current_texture → create_view → create_command_encoder
    RN->>RN: WorldQuery::new(active_camera, scene, delta_time, resources)

    note over RN,GPU: Scene Passes — each Pass queries the world, owns its queue
    RN->>PB: draw(encoder, frame, view, &world)
    PB->>ECS: a. QUERY  world.query::<T>() → zip transform[]+pbr[], skip None
    PB->>PB: b. GROUP  find-or-create by material+mesh (cut PSO/bind switches)
    PB->>PB: c. STAGE  concat instances → buffer, assign base_offset
    PB->>GPU: d. UPLOAD  write per_draw storage (x1)
    PB->>GPU: e. PASS   begin_render_pass: Clear color=BLACK, depth=1.0
    PB->>GPU: f. ENCODE 6 groups → 6x draw_indexed (instanced, 60,001 total)
    PB->>GPU: g. BG     background sub-draw (fullscreen tri @ far plane)
    GPU->>GPU: vertex shader per instance: model = T·R·S → clip = viewProj·model

    cube                  497 KB raw   207 KB gzip
    Three.js r167 webgpu  779 KB raw   210 KB gzip
    Three.js r167 module  668 KB raw   165 KB gzip
    → cube beats both on raw, on par on gzip

                            cube
    total (final)         497.3 KB
    total (pre-opt)        1.83 MB

    --- final binary sections (post wasm-opt, no symbol names) ---
      code                379.3 KB
      data                100.1 KB
      wasm-meta            17.3 KB

    --- pre-opt symbol attribution (only place with crate info) ---
      engine total         117.5 KB
        pill_engine         87.0 KB
        pill_core           22.9 KB
        pill_web             7.6 KB
        pill_renderer          0 B    (merged into pill_engine this branch)
      game code              1.4 KB   (monitor only)
      [rust-std]           194.9 KB
      [rodata]             100.9 KB   (was ~130 KB before ASCII-case fix; −30 KB Unicode tables)
      winit                 75.0 KB
      wgpu                  41.7 KB
      hashbrown             41.9 KB
      web_sys               26.2 KB
      wasm_bindgen          10.4 KB

      stripped by wasm-opt:
        [debug:names]      491.5 KB
        [wasm-bindgen]     437.8 KB

    --- full deployment (wasm + JS glue) ---
      cube     571.8 KB   (wasm 497.3 + js 74.6)

# Conflicts: # engine/Cargo.toml # engine/pill_core/Cargo.toml # engine/pill_core/src/utils.rs # engine/pill_engine/Cargo.toml # engine/pill_engine/res/shaders/default_lit_fragment.wgsl # engine/pill_engine/res/shaders/default_unlit_fragment.wgsl # engine/pill_engine/src/ecs/systems/rendering_system.rs # engine/pill_engine/src/engine.rs # engine/pill_engine/src/graphics/dummy_renderer.rs # engine/pill_engine/src/graphics/render_queue.rs # engine/pill_engine/src/graphics/renderer.rs # engine/pill_engine/src/lib.rs # engine/pill_engine/src/renderer/drawers/mesh_drawer.rs # engine/pill_engine/src/renderer/resources/engine_parameters.rs # engine/pill_engine/src/renderer/resources/renderer_mesh.rs # engine/pill_engine/src/renderer/resources/renderer_shader.rs # engine/pill_engine/src/renderer/resources/renderer_texture.rs # engine/pill_engine/src/resources/material.rs # engine/pill_engine/src/resources/shader.rs # engine/pill_engine/src/resources/texture.rs # engine/pill_renderer/Cargo.toml # engine/pill_renderer/src/renderer.rs # engine/pill_renderer/src/resources/renderer_material.rs # engine/pill_renderer/src/resources/renderer_resource_storage.rs

…kspace path - div_ceil/repeat_n: replace manual bit-alignment arithmetic with .div_ceil() and std::iter::repeat_n in pass_pbr_static.rs and state.rs - needless_range_loop: rewrite mat4_mul inner loop as (0..4).map(...).sum() in glb_to_cooked_mesh.rs - egui native-only: move egui + egui-wgpu from unconditional deps to [target.'cfg(not(target_arch = "wasm32"))'.dependencies]; gate egui_client.rs, EguiClient re-export, RenderStateComponent.egui_client field, and init_default_passes signature/call sites with #[cfg(not(target_arch = "wasm32"))] to restore WASM binary size below 499 KB budget - codesign: gate codesign_adhoc function and call sites with #[cfg(target_os = "macos")] so Linux CI does not try to invoke the tool - workspace path: restore NO_PATH placeholder in engine/Cargo.toml so the launcher can substitute the correct game path on any machine - pill_tunel: add egui = "0.32.1" direct dependency (game.rs uses egui:: types); fix workspace field to NO_PATH

JDuchniewicz · 2026-06-07T16:04:31Z

+    }
+}
+
+fn orbit_camera_system(engine: &mut Engine) -> Result<()> {


we really ought to have a common place for such systems/components

Cool, these are just emerging. A code that get's duplicated 3 times is moved to a proper common place.

JDuchniewicz · 2026-06-07T16:09:11Z


+#[cfg(target_os = "macos")]
+fn codesign_adhoc(path: &PathBuf) -> Result<()> {
+    let status = Command::new("codesign")


what does this do?

On macos executables would faild duoe to .dylib (.dll on mac) being not signed, even if it's development version #think_different

JDuchniewicz · 2026-06-07T16:10:20Z


-use pill_game::WebGame;
+#[cfg(target_arch = "wasm32")]
+#[global_allocator]


what was the allocator situation for Wasm before this allocator was added?

Allocator for wasm was changed to single threaded allocator with less binary size.

JDuchniewicz · 2026-06-07T16:11:30Z


 [dependencies]
 pill_core = {path = "../pill_core"}
+glam = { version = "0.30.8", features = ["serde", "bytemuck"] }


why are we pulling it in here? Isn't everything typedeffed already in core?

Good question

JDuchniewicz · 2026-06-07T16:14:00Z


+/// Parse an RTEX binary header.
+/// Returns `(pixel_bytes, width, height, mip_count, version)`.
+fn decode_rtex_header(bytes: &[u8]) -> (&[u8], u32, u32, u32, u32) {


defo not place for this here, move to utils?

utils in what domain?

I though for now just a file in the crate - if we start getting a lot of common functions/code across the whole project then maybe a new pill_utils crate?

JDuchniewicz · 2026-06-07T16:15:10Z

+
+    /// Provides equirect background texture bytes; used by PassBackground on first-frame init.
+    /// Must be called before the first rendered frame (i.e. from `start()`).
+    pub fn set_background_texture(&mut self, bytes: Vec<u8>) -> Result<()> {


this also has utilish taste to it.

also, I think the less we require to be called by the user the better - it should all happen automagically or at least minimize it to be called inside start. I think such a requirement is too stark. The game should run if the user does not provide these textures. Also, what if the user wants to change it mid-run?

Good catch. Engine is expected to have no knowledge about domain specific assets. Remove set_background_texture() and keep it as component state

JDuchniewicz · 2026-06-07T16:20:41Z

+pub type RenderQueueKeyType = u64;
+
+// 64-bit render sort key layout (MSB → LSB):
+// bit: 63        59 58       51 50       43 42       35 34       27 26       19 18       11 10        0


not sure if I don't prefer the old more verbose ASCII diagram

JDuchniewicz · 2026-06-07T16:23:25Z

    ) -> Result<()> {
        let shader = engine.get_resource::<Shader>(&self.shader_handle)?;

-        // Get texture to be set


not sure about removing these comments

JDuchniewicz · 2026-06-07T16:32:03Z

+    #[cfg(not(target_arch = "wasm32"))]
+    fn new(window: Arc<winit::window::Window>, config: EngineConfig) -> Result<Self> {
+        info!(LogContext::Rendering => "Initializing {}", "Renderer".module_object_style());
+        let state = pollster::block_on(State::new(window, config))?;


will this run the renderer on another thread? I am trying to grasp the point behind having it inited async

JDuchniewicz · 2026-06-07T16:32:45Z

+        vertex_wgsl: &str,
+        fragment_wgsl: &str,
+        texture_slots: &HashMap<String, crate::resources::ShaderTextureSlot>,
+        parameter_slots: &[(String, crate::resources::ShaderParameterSlot)],


Why the other is a hashmap and this is an array? Also, maybe a custom typedef so we avoid long names?

JDuchniewicz · 2026-06-07T16:37:07Z

+
+        let frame = self.state.surface.get_current_texture();
+        let frame = match frame {
+            std::result::Result::Ok(frame) => frame,


I am not sure we need to have qualification everywhere, especially on std types or pill_* ones. Defo good to have it on wgpu etc, but then again we create long chains of types and names and its hard to read. If we use it often, just use it at the top.

Yeah, we need to decide on one project-wide approach.
But I'm also not sure which to pick

I think the rules should be simple:
If used often - import, if it's an one-off and is not a::super::long::qualification::type::name then we can do it with the qualification. Otherwise, let's just import at the top and save the future code readers additional strain. LSPs are good enough at providing the full type info if necessary.

JDuchniewicz · 2026-06-07T16:39:48Z

+        Ok(buffer)
+    }
+
+    fn create_pipeline_v2(&mut self, desc: PipelineV2Desc) -> Result<PipelineV2> {


v2? why then keep v1?

JDuchniewicz · 2026-06-07T16:41:03Z

+    fn create_texture_from_pixels(
+        &mut self,
+        name: &str,
+        mip_pixels: &[&[u8]],


quite weird type, why it's not contiguous or a separate specialized type?

JDuchniewicz · 2026-06-07T16:46:37Z

                module: &vertex_shader,
                entry_point: Some("vs_main"),
-                buffers: vertex_layouts, // Specifies structure of vertices that will be passed to the vertex shader
+                buffers: vertex_layouts,


a lot of comments removed - sure we want that?

JDuchniewicz · 2026-06-07T16:47:10Z

                wgpu::VertexAttribute {
-                    // Vertex texture coordinates
-                    // slangc maps TEXCOORD0 → @location(4), not 1
                    offset: mem::size_of::<[f32; 3]>() as wgpu::BufferAddress,


comments removed

JDuchniewicz · 2026-06-07T16:48:07Z

+    pub(crate) fn write_parameters_to_buffer(
+        queue: &wgpu::Queue,
+        buffer: &wgpu::Buffer,
+        parameter_slots: &[(String, ShaderParameterSlot)],


why this array of pairs?

JDuchniewicz · 2026-06-07T17:46:35Z

+// --- Pass API types ---
+
+/// Read-only view of the world handed to each pass. A pass calls `query::<T>()` for whatever
+/// component types it needs and builds its OWN draw list (opaque, translucent, shadow, …).


I think this is in line with what @MattSzymonski is implementing in the ECS refactor. Might be good to commonalize the query-approach to using just a slice of mut engine.

JDuchniewicz · 2026-06-07T17:47:52Z

+}
+
+#[derive(Clone, Debug)]
+pub struct PipelineV2Desc<'a> {


I still don't get the V2 - will it be merged without a rename?

JDuchniewicz · 2026-06-07T17:49:53Z

-        textures: &[(String, MaterialTexture)],
-    ) -> Result<()>;
+    /// Returns the OS window the renderer is bound to.
+    fn get_window(&self) -> std::sync::Arc<winit::window::Window>;


Don't we have a coding convention for using qualified type names? If not, it's good to set it as I don't know when we use -import a type and when to use a full qualified type name.

JDuchniewicz · 2026-06-07T17:51:29Z

 pub struct RenderQueueItem {
    pub key: RenderQueueKey,
    pub entity_index: u32,
+    /// Raw transform copied (no trig) during the cache-warm sequential ECS scan;


whats trig in this context? Trigonometry?

Yes, building LocalMatrix requires computing sin/cos from angle values

JDuchniewicz · 2026-06-07T18:09:36Z

+                entry_func: "fs_main",
+            },
+            vertex_buffers: &[
+                <crate::renderer::resources::RendererMesh as crate::renderer::resources::Vertex>::data_layout_descriptor(),


yeah, a bit too long

JDuchniewicz · 2026-06-07T18:11:20Z

+        &mut self,
+        encoder: &mut wgpu::CommandEncoder,
+        renderer: &mut dyn PillRenderer,
+        _frame: &wgpu::SurfaceTexture,


why do we need this param?

Some Passes may need surface to get it's format or size

JDuchniewicz · 2026-06-07T18:12:21Z

+        // Read active camera and transform.
+        let active_camera_index = world.active_camera.data().index as usize;
+        let active_camera_component = camera_components
+            .data


why not handle the error gracefully?

JDuchniewicz · 2026-06-07T18:18:06Z

+        let active_camera_component = camera_components
+            .data
+            .get(active_camera_index)
+            .unwrap()


safer failing?

JDuchniewicz · 2026-06-07T18:30:17Z

 // --- Transform Component ---

-// NOTE: Setting position/rotation/scale directly is not possible since we need to update matrices after each change
+// 36 bytes hot (pos+rot+scale only); model_matrix/normal_matrix were dead weight — GPU


Wait, so model_matrix was used only for rendering?

There is no Scene hierarchy and transform tree, isn't there?

In this case ALU on GPU is faster than Memory Bandwidth of pushing whole matrices through CPU->GPU.

JDuchniewicz · 2026-06-07T18:44:25Z

 {
    //entry: HashMapVacantEntry<'a, TypeId, Box<(dyn Any + Send + Sync)>>,
-    entry: HashMapVacantEntry<'a, TypeId, Box<dyn Any + Send>>,
+    entry: HashMapVacantEntry<'a, TypeId, Box<dyn Any>>,


they used to be Send + Sync now they are no longer either? Why?

Did we ever have threadness implemented, because that's the reason why we add these traits. If we want to have multithreaded iteration over the typemaps then we would probably need to either slice them up in non-overlapping partitions (which is cleaner) or have these traits added.

JDuchniewicz · 2026-06-07T18:44:56Z

@@ -0,0 +1,9 @@
+// WHY: zero-size marker types so Handle<RendererMeshTag> and Handle<RendererTextureTag> are different types — passing the wrong one is a compile error, not a runtime bug.


JDuchniewicz · 2026-06-07T18:46:35Z

+    }
+}
+
+pub(crate) fn lerp3(start: [f32; 3], end: [f32; 3], factor: f32) -> [f32; 3] {


can't some of these pill_assets helper functions be shared? We are starting to have a lot of math/util funcitons flying around here

Agree, math in core looks like good place.

As per previous reply, pill_assets are based on examples using it and as code emerges the functions are moved into proper modules.

JDuchniewicz · 2026-06-07T18:52:16Z

LGTM, minor remarks. Would really appreciate some documentation at this point, especially since I am no rendering expert and it's not that easy to review it without a deeper understanding. Read through Aaltonen's presentations and at least I get the gist of the rendering optimization thanks to that 😁

sp0lsh added 2 commits May 28, 2026 18:29

sp0lsh marked this pull request as draft May 28, 2026 16:30

sp0lsh added 7 commits May 28, 2026 19:04

feat: extract shaders to files

724f949

review

d51c262

review

95d4ada

review

cdcb249

review

285f712

review

6e64c83

example pbr helmet

040bc05

sp0lsh changed the title ~~feat: renderer~~ feat: renderer: PBR, Passes May 29, 2026

example: pbr balls

fc02e60

sp0lsh changed the title ~~feat: renderer: PBR, Passes~~ feat: renderer: PBR, IBL, Passes May 30, 2026

sp0lsh added 13 commits May 30, 2026 18:12

example: pbr helmet

e26b2e8

fix ci

252a327

fix ci

24c56e9

fix: do not embedd hdr by default

d38a328

fix wasm size

279fb43

fix: fog and ibl tweaks

c10032a

feat: bg sub pass for better perf

c4f86fd

feat: hdr for helmet

b4e06fb

perf: pill_tunel: 3 -> 12 FPS

124ded8

perf: pill_tunel: 12 -> 30 FPS

98e9e15

feat: rendering system cleanup

f0bca71

feat: renderer cleanup

dba3e0f

sp0lsh force-pushed the feat/renderer branch from 30c15f1 to dba3e0f Compare May 31, 2026 15:34

sp0lsh added 3 commits May 31, 2026 18:49

feat: rust safety review

3ecedfa

add pbr_balls, pbr_helmet to wasm ci

9ef8851

fix ci

256c3a4

JDuchniewicz reviewed Jun 7, 2026

View reviewed changes

JDuchniewicz mentioned this pull request Jun 8, 2026

hot-reload: Add UI to display status. Don't crash when build fails #42

Merged

		@@ -0,0 +1,9 @@
		// WHY: zero-size marker types so Handle<RendererMeshTag> and Handle<RendererTextureTag> are different types — passing the wrong one is a compile error, not a runtime bug.

Uh oh!

Conversation

sp0lsh commented May 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sp0lsh Jun 7, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

sp0lsh commented May 28, 2026 •

edited

Loading

sp0lsh Jun 7, 2026 •

edited

Loading