using Renderlet and renderlet_* shaders, fixed wrapping addition bug …

…in crabslab

Cargo.toml

@@ -16,13 +16,14 @@ resolver = "2"
 assert_approx_eq = "1.1.0"
 async-channel = "1.8"
 bytemuck = { version = "1.13.0", features = ["derive"] }
-crabslab = { version = "0.4.1", default-features = false }
+crabslab = { version = "0.4.5", path = "../crabslab/crates/crabslab", default-features = false }
 env_logger = "0.10.0"
 futures-lite = "1.13"
 gltf = { git = 'https://github.com/gltf-rs/gltf.git', features = ["KHR_lights_punctual", "KHR_materials_unlit", "KHR_materials_emissive_strength", "extras"] }
 image = "0.24"
 log = "0.4"
 naga = { version = "0.19", features = ["spv-in", "wgsl-out", "wgsl-in", "msl-out"] }
+plotters = "0.3"
 pretty_assertions = "1.4.0"
 proc-macro2 = { version = "1.0", features = ["span-locations"] }
 glam = { version = "0.24.2", default-features = false }

DEVLOG.md

@@ -1,5 +1,97 @@
 # devlog
 
+## Sat Apr 10, 2024 
+
+### Finishing the debugging session
+
+It WAS `crabslab`! Specifically it was `Slab::contains`, which is used to check that 
+a type with an `Id` _can be read_. 
+
+Previously the definition was: 
+
+```rust 
+fn contains<T: SlabItem>(&self, id: Id<T>) -> bool {
+    id.index() + T::SLAB_SIZE <= self.len()
+}
+```
+
+Which seems correct, and it functions correctly on the CPU. 
+But on the GPU (meaning `target_arch = "spirv"``) `usize` is a 32bit `u32`, 
+and so the `id.index() + T::SLAB_SIZE` will overflow if the id is `Id::NONE`, 
+because `Id::NONE = u32::MAX;`. 
+
+Indeed, the id is often `Id::NONE`, as that is the default!
+This was causing a silent panic in my shader, which then produced no output.
+
+Now the definition is this: 
+```rust
+fn contains<T: SlabItem>(&self, id: Id<T>) -> bool {
+    self.len() >= T::SLAB_SIZE && id.index() <= self.len() - T::SLAB_SIZE
+}
+```
+
+What a hard-to-diagnose bug! I really need trace statements on GPU.
+
+## Fri Apr 9, 2024 
+
+I have bugs after removing the SDF raymarching stuff. 
+
+Primarily I can't get any of my `stage_vertex`+`stage_fragment` tests passing. 
+Everything is blank. 
+
+### Debug with me!
+
+* it's not crabslab: I fixed some bugs in it and after testing through the `tutorial`
+  shaders I'm 80% certain it's not a (de)serialization problem.
+* NOTHING is being written to the depth texture...
+  - depth is cleared to 1.0 
+  - pipeline depth function is set to ALWAYS (always succeed) and still nothing is written
+  - face culling is off and still nothing is written
+  - running the vertex shader on CPU and printing out clip positions shows:
+    ```
+    clips: [
+        Vec4(
+            -1.0,
+            1.0,
+            0.25,
+            1.0,
+        ),
+        Vec4(
+            -1.0,
+            -1.0,
+            0.25,
+            1.0,
+        ),
+        Vec4(
+            1.0,
+            1.0,
+            0.25,
+            1.0,
+        ),
+    ]
+    ```
+    Which is a CCW triangle up in the top left of the clip space. 
+    So we should see SOMETHING in the depth texture at least, but we don't.
+    Why do we not? Is the render even happening on the GPU? Let's check logging
+    to see if we're issuing the calls..
+      - `stage_render` prints `drawing vertices 0..3 and instances 147..148`
+        so I'm certain we're actually rendering.
+
+At this point I'm a bit at a loss. The difference between the tutorial shaders (which are working) 
+and my stage shader is mainly that the stage shader first writes to an HDR surface and then 
+runs tonemapping and writes the result to the frame surface. I can't see any other course of action 
+than removing HDR and tonemapping to see if that works.
+
+I'm going to try rendering the `gltf_cmy_tri` slab with the `tutorial` shaders. 
+We'll see what happens.
+
+NOTHING! No rendering. No depth values. So this must have something to do with the data.
+
+### What to do about the wacky GLTF stage shaders
+
+I'm going to go back to a much simpler "renderlet" pbr shader program. The entire GLTF document can 
+still live on the GPU, but GLTF is too complicated a structure to use for the internal representation.
+
 ## Mon Feb 26, 2024 
 
 ### SDF stack lang on GPU considered harmful

README.md

@@ -1,7 +1,12 @@
 
 # renderling 🍖
 
-This aspires to be a modern "GPU-driven" renderer. It is alpha software. I'm still learning, but quickly!
+Renderling is an innovative, GPU-driven renderer designed for efficient scene rendering with a focus on leveraging 
+GPU capabilities for nearly all rendering operations. 
+Utilizing Rust for shader development, it ensures memory safety and cross-platform compatibility, including web platforms. 
+The project, currently in the alpha stage, aims for rapid loading of GLTF files, handling large scenes, and supporting numerous lights. 
+Development emphasizes performance, configurability, observability and the use of modern rendering techniques like forward+ rendering and 
+physically based shading.
 
 <img width="912" alt="ibl_environment_test" src="https://github.com/schell/renderling/assets/24942/297d6150-64b2-45b8-9760-12b27dc8cc3e">
 
@@ -155,10 +160,27 @@ RUSTFLAGS=--cfg=web_sys_unstable_apis trunk build crates/example-wasm/index.html
 
 ## 🫶 Sponsor this!
 
-This work will always be free and open source. If you use it (outright or for inspiration), please consider donating.
+This work will always be free and open source. 
+If you use it (outright or for inspiration), please consider donating.
 
 [💰 Sponsor 💝](https://github.com/sponsors/schell)
 
+### Related work & spin-off projects 
+
+Many projects were born from first solving a need within `renderling`. 
+Some of these solutions were then spun off into their own projects.
+
+- [`crabslab`](https://github.com/schell/crabslab)
+  A slab allocator for working across CPU/GPU boundaries.
+- [`loading-bytes`](crates/loading-bytes)
+  A cross-platform (including the web) way of loading files to bytes.
+- [`moongraph`](https://github.com/schell/moongraph)
+  A DAG and resource graph runner.
+- Contributions to [`naga`](https://github.com/gfx-rs/wgpu/issues/4489)
+  * Adding atomics support to the SPIR-V frontend (in progress)
+
+Sponsoring this project contributes to the ecosystem. 
+
 ## License
 Renderling is free and open source. All code in this repository is dual-licensed under either:
 

crates/img-diff/src/lib.rs

@@ -114,7 +114,10 @@ fn get_results(
 pub fn save(filename: &str, seen: impl Into<DynamicImage>) {
     let path = Path::new(TEST_OUTPUT_DIR).join(filename);
     std::fs::create_dir_all(path.parent().unwrap()).unwrap();
-    seen.into().save(path).unwrap();
+    let img: DynamicImage = seen.into();
+    let img_buffer = img.into_rgba8();
+    let img = DynamicImage::from(img_buffer);
+    img.save(path).unwrap();
 }
 
 pub fn assert_eq_cfg(

crates/renderling/Cargo.toml

@@ -15,7 +15,7 @@ readme = "../../README.md"
 crate-type = ["lib", "dylib"]
 
 [features]
-default = ["winit", "gltf", "sdf", "shaders"]
+default = ["winit", "sdf", "shaders", "tutorial"]
 shaders = [
   "array_test",
   "brdf_lut_convolution_fragment",
@@ -25,18 +25,25 @@ shaders = [
   "pbr_fragment",
   "prefilter_environment_cubemap_fragment",
   "prefilter_environment_cubemap_vertex",
+  "renderlet_fragment",
+  "renderlet_vertex",
   "skybox_cubemap_fragment",
   "skybox_cubemap_vertex",
   "skybox_equirectangular_fragment",
   "skybox_vertex",
-  "stage_fragment",
-  "stage_vertex",
   "test_i8_i16_extraction",
   "tonemapping_fragment",
   "tonemapping_vertex"
 ]
 gltf = ["dep:gltf"]
 sdf = []
+tutorial = [
+  "tutorial_passthru_fragment",
+  "tutorial_implicit_isosceles_vertex",
+  "tutorial_slabbed_vertices_no_instance",
+  "tutorial_slabbed_vertices",
+  "tutorial_slabbed_renderlet"
+]
 # shaders
 array_test = []
 brdf_lut_convolution_fragment = []
@@ -46,21 +53,27 @@ generate_mipmap_vertex = []
 pbr_fragment = [] 
 prefilter_environment_cubemap_fragment = []
 prefilter_environment_cubemap_vertex = []
-raymarch_fragment = []
-raymarch_rays_fragment = []
-raymarch_vertex = []
-sdf_shape_fragment = []
-sdf_shape_vertex = []
-sdf_prim_fragment_test = []
+#raymarch_fragment = []
+#raymarch_rays_fragment = []
+#raymarch_vertex = []
+#sdf_shape_fragment = []
+#sdf_shape_vertex = []
+#sdf_prim_fragment_test = []
+renderlet_fragment = []
+renderlet_vertex = []
 skybox_cubemap_fragment = []
 skybox_cubemap_vertex = []
 skybox_equirectangular_fragment = []
 skybox_vertex = []
-stage_fragment = []
-stage_vertex = []
 test_i8_i16_extraction = []
 tonemapping_fragment = []
 tonemapping_vertex = []
+tutorial_passthru_fragment = []
+tutorial_implicit_isosceles_vertex = []
+tutorial_slabbed_vertices_no_instance = []
+tutorial_slabbed_vertices = []
+tutorial_slabbed_renderlet = []
+
 wasm = ["wgpu/fragile-send-sync-non-atomic-wasm"]
 
 # dependencies for CPU and GPU code
@@ -103,6 +116,7 @@ env_logger = {workspace = true}
 icosahedron = "0.1"
 img-diff = { path = "../img-diff" }
 naga.workspace = true
+plotters.workspace = true
 pretty_assertions.workspace = true
 ttf-parser = "0.20.0"