wip/gltf on the slab build out (#69)

* Support for indexed vertices, ie native vertices can be indexed

* debugging gltf on-the-slab textures

* debugging gltf on-the-slab textures, still

* gltf on-the-slab can use textures

* porting over more gltf tests

* setting stage images resets the atlas size on the slab

* ported over more tests to the slab

* ported over atlas_uv_mapping test

* ported uv_wrapping test

* ported negative_uv_wrapping

* WIP porting scene_cube_directional (test of directional lighting)

* WIP CPU-side sampling for testing

* ported scene_cube_directional test of directional lighting

* fixed accessors, they now take their own offset into account

* fix accessors - view stuff shouldn't be in GltfAccessor, it should be in GltfBufferView

* WIP GLTF is the only internal representation

* WIP more work on GLTF is the only internal representation

* finished porting tests to gltf-on-the-slab

* criterion benchmark stub

* added criterion benchmark and ported scene_parent_sanity test

* ported pbr test minus bloom

* pbr metallic spheres test doesn't use bloom

* chore/remove bytemuck from renderling shader (#70)

* parted out slab into slabcrab

* parted out slab into crabslab

* parted out slab into crabslab

* Update README.md

* readme

* WIP part out crabslab, remove non-slab shaders, remove bloom (to be reimplemented later)

* WIP

* img-diff takes into account pixel threshold and image threshold

* fixed problems with the skybox after slabbing

* fixed tests to use explicit linear transfer where appropriate

Cargo.toml

@@ -6,19 +6,22 @@ members = [
     "crates/renderling",
     "crates/renderling-shader",
     "crates/renderling-gpui",
-    "crates/renderling-derive",
+    "crates/crabslab",
+    "crates/crabslab-derive",
 ]
 
 exclude = ["./shaders"]
 
 resolver = "2"
 
 [workspace.dependencies]
+async-channel = "1.8"
 bytemuck = { version = "1.13.0", features = ["derive"] }
 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"
-glam = "0.24.2"
+glam = { version = "0.24.2", default-features = false }
+snafu = "0.7"
 winit = { version = "0.27" }
 wgpu = { version = "0.17" }

DEVLOG.md

@@ -1,5 +1,118 @@
 # devlog
 
+## Sat Dec 23, 2023
+
+I've ported over a majority of the tests to the GLTF-on-the-slab implementation.
+I'm currently working on the big PBR test and having trouble with the skybox, which
+is rendering all black...
+
+Debugging rabbit hole:
+* So is it even running?
+  - Yes, logging shows that it's running.
+* Could it be it needs to be run in its own render pass?
+* Before I even check that, I see that the skybox's vertex shader uses the `instance_index` as the `Id` of the camera, and I'm passing `0..1` as the instance range in the draw call.
+  - So we need a way to pass the camera's `Id` to the skybox.
+    - I just added it as a field on `Skybox`
+    - Using that new field fixed that issue. Now I have an issue with bloom.
+
+After fixing the skybox rendering it seems bloom isn't running.
+
+Debugging rabbit hole:
+* So is it even running?
+  - Yes, logging shows that it's running.
+* Is the result being used downstream during tonemapping?
+  - It seems to be.
+* Let's check to see that there isn't something funky when configuring the graph.
+  - Nothing I can tell there.
+* Maybe print out the brightness texture and make sure it's populated?
+* Losing steam here, especially since bloom needs to be re-done as "physically based".
+
+### Physically Based Bloom
+
+## Thu Dec 21, 2023
+
+It's the solstice! My Dad's birthday, and another bug hunt in `renderling`.
+
+### Porting gltf_images test
+The test `gltf_images` tests our image decoding by loading a GLTF file and then
+creating a new staged object that uses the image's texture.
+
+It's currently coming out all black, and it should come out like
+![gltf_images test](test_img/gltf_images.png).
+
+I recently got rid of the distinction between "native" vertex data and GLTF vertex
+data. Now there is only GLTF vertex data and the "native" `Vertex` meshes can be
+conveniently staged (marshalled to the GPU) using a helper function that creates
+a `GltfPrimitive` complete with `GltfAccessors` etc.
+
+Debbuging rabbit hole:
+* Let's compare old vs new vertex shaders
+  - It doesn't seem to be the vertices, because the staged vertices (read from the GPU) are equal to the original mesh.
+  - The staged vertices are equal to the original CPU-side mesh, but the computed vertex values are different from legacy.
+    - It looks like transforms on `RenderUnits` are not transforming their child primitive's geometry
+      - Got it! It was because `GltfNode`'s `Default` instance was setting `scale` to `Vec3::ZERO`.
+
+## Wed Dec 20, 2023
+
+I think I'm going to keep going with this idea of making GLTF the internal representation of the
+renderer.
+
+## Tue Dec 19, 2023
+
+### Thoughts on GLTF
+GLTF on-the-slab has been a boon to this project and I'm tempted to make it the main way we do
+rendering. I just want to write this down somewhere so I don't forget. Currently when loading
+a GLTF file we traverse the GLTF document and store the whole thing on the GPU's slab. Then
+the user has to specify which nodes (or a scene) to draw, which traverses one more time, linking
+the `RenderUnit`s to the primitives within the GLTF. I _think_ it might be cognitively easier
+to have GLTF nodes somehow be the base unit of rendering ... but I also have plans for supporting
+SDFs and I'm not sure how that all fits together.
+
+* [At least one other person is thinking about putting SDFs in GLTF using an extension](https://community.khronos.org/t/signed-distance-field-representation-of-geometry-extension/109575)
+
+Anyway - I'll keep going with the momentum I have and think about refactoring towards this in the future.
+
+## Mon Dec 18, 2023
+
+### Simple Texture GLTF Example
+* The `simple_texture` test is rendering the texture upside-down.
+* There are _no rotation transformations_ in its node's hierarchy.
+* What does the atlas look like?
+  - It's not the atlas, the two tests (slabbed and the previous non-slabbed) have
+    identical atlas images.
+* So what about UV coords?
+  - Comparing runs of the vertex shaders shows that the UV coords' Y components are flipped.
+  - So, 0.0 is 1.0 and 1.0 is 0.0
+* So is there something doing this intentionally?
+  - Nothing that I can easily see in the `gltf_support` modules...
+  - It has something to do with the accessor.
+  - I can see in the GLTF file that the accessor's byte offset is 48, but somehow in
+    my code it comes out 12...
+  - It was because the accessor's offset was not being taken into account.
+
+### Analytical Directional Lights
+I got analytical lighting working (at least for directional lights) on the stage.
+The problem I was having was that the shaders use `Camera.position` in lighting
+equations, but that was defaulting to `Vec3::ZERO`. Previously in the "scene"
+version of the renderer (which I'm porting over to "stage") the camera's position
+was set automatically when setting the projection and/or view.
+I had to run both versions of the vertex AND fragement shaders to track this down. Ugh!
+
+## Fri Dec 8, 2023
+
+I've been having trouble getting the new GLTF files on-the-slab method to pass my
+previous tests. Mainly because of little things I had forgotten. Little bits of
+state that need to be updated to run the shaders. The most recent was that the
+size of the atlas needs to be updated on the GPU when the atlas changes.
+
+I'm moving over tests from `renderling/scene/gltf_support.rs` to
+`renderling/stage/gltf_support.rs` one at a time.
+
+## Thu Dec 7, 2023
+
+Ongoing work to get GLTF files on-the-slab working. When this work is done GLTF
+file imports should be lightening fast.
+
 ## Wed Nov 15, 2023
 
 I resubmitted the NLNet grant proposal with expanded scope to take care of [the

NOTES.md

@@ -17,6 +17,7 @@ Just pro-cons on tech choices and little things I don't want to forget whil impl
   - using cargo and Rust module system
   - expressions!
   - type checking!
+  - traits!
   - editor tooling!
 
 ## cons / limititions
@@ -28,6 +29,7 @@ Just pro-cons on tech choices and little things I don't want to forget whil impl
 * for loops are hit or miss, sometimes they work and sometimes they don't
   - see [this rust-gpu issue](https://github.com/EmbarkStudios/rust-gpu/issues/739)
   - see [conversation with eddyb on discord](https://discord.com/channels/750717012564770887/750717499737243679/threads/1092283362217046066)
+* can't use `.max` or `.min`
 * meh, but no support for dynamically sized arrays (how would that work in no-std?)
   - see [conversation on discord](https://discord.com/channels/750717012564770887/750717499737243679/1091813590400516106)
 * can't use bitwise rotate_left or rotate_right

crates/renderling-derive/Cargo.toml → crates/crabslab-derive/Cargo.toml

@@ -1,5 +1,5 @@
 [package]
-name = "renderling-derive"
+name = "crabslab-derive"
 version = "0.1.0"
 edition = "2021"
 

crates/renderling-derive/src/lib.rs → crates/crabslab-derive/src/lib.rs

@@ -31,7 +31,7 @@ fn get_params(input: &DeriveInput) -> syn::Result<Params> {
         _ => {
             return Err(syn::Error::new(
                 name.span(),
-                "deriving Slabbed only supports structs".to_string(),
+                "deriving SlabItem only supports structs".to_string(),
             ))
         }
     };
@@ -41,7 +41,7 @@ fn get_params(input: &DeriveInput) -> syn::Result<Params> {
         .map(|field| {
             let ty = &field.ty;
             quote! {
-                <#ty as renderling_shader::slab::Slabbed>::slab_size()
+                <#ty as crabslab::SlabItem>::slab_size()
             }
         })
         .collect();
@@ -69,7 +69,7 @@ fn get_params(input: &DeriveInput) -> syn::Result<Params> {
     })
 }
 
-#[proc_macro_derive(Slabbed)]
+#[proc_macro_derive(SlabItem)]
 pub fn derive_from_slab(input: proc_macro::TokenStream) -> proc_macro::TokenStream {
     let input: DeriveInput = syn::parse_macro_input!(input);
     let name = &input.ident;
@@ -88,8 +88,7 @@ pub fn derive_from_slab(input: proc_macro::TokenStream) -> proc_macro::TokenStre
         /// Adds a `CanFetch<'lt>` bound on each of the system data types.
         fn constrain_system_data_types(clause: &mut WhereClause, tys: &[Type]) {
             for ty in tys.iter() {
-                let where_predicate: WherePredicate =
-                    syn::parse_quote!(#ty : renderling_shader::slab::Slabbed);
+                let where_predicate: WherePredicate = syn::parse_quote!(#ty : crabslab::SlabItem);
                 clause.predicates.push(where_predicate);
             }
         }
@@ -129,9 +128,11 @@ pub fn derive_from_slab(input: proc_macro::TokenStream) -> proc_macro::TokenStre
             FieldName::Ident(field) => Ident::new(&format!("offset_of_{}", field), field.span()),
         };
         offsets.push(quote! {
-            pub fn #ident() -> usize {
-                #(<#offset_tys as renderling_shader::slab::Slabbed>::slab_size()+)*
-                0
+            pub fn #ident() -> crabslab::Offset<#ty> {
+                crabslab::Offset::new(
+                    #(<#offset_tys as crabslab::SlabItem>::slab_size()+)*
+                    0
+                )
             }
         });
         offset_tys.push(ty.clone());
@@ -145,7 +146,7 @@ pub fn derive_from_slab(input: proc_macro::TokenStream) -> proc_macro::TokenStre
         }
 
         #[automatically_derived]
-        impl #impl_generics renderling_shader::slab::Slabbed for #name #ty_generics #where_clause
+        impl #impl_generics crabslab::SlabItem for #name #ty_generics #where_clause
         {
             fn slab_size() -> usize {
                 #(#sizes)+*

crates/crabslab/Cargo.toml

@@ -0,0 +1,33 @@
+[package]
+name = "crabslab"
+version = "0.1.0"
+edition = "2021"
+description = "Slab allocator focused on GPU compute (rust-gpu)"
+repository = "https://github.com/schell/renderling"
+license = "MIT OR Apache-2.0"
+keywords = ["game", "graphics", "shader", "rendering"]
+categories = ["rendering", "game-development", "graphics"]
+readme = "README.md"
+
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
+
+[features]
+default = ["wgpu", "glam", "futures-lite"]
+futures-lite = ["dep:futures-lite"]
+glam = ["dep:glam"]
+wgpu = ["dep:wgpu", "dep:bytemuck", "dep:snafu", "dep:async-channel", "dep:log"]
+
+[dependencies]
+async-channel = {workspace=true, optional=true}
+bytemuck = {workspace=true, optional=true}
+futures-lite = {workspace=true, optional=true}
+log = {workspace=true, optional=true}
+crabslab-derive = { version = "0.1.0", path = "../crabslab-derive" }
+snafu = {workspace=true, optional=true}
+wgpu = {workspace=true, optional=true}
+
+[target.'cfg(not(target_arch = "spirv"))'.dependencies]
+glam = { workspace = true, features = ["std"], optional = true }
+
+[target.'cfg(target_arch = "spirv")'.dependencies]
+glam = { version = "0.24.2", default-features = false, features = ["libm"], optional = true }

crates/crabslab/README.md

@@ -0,0 +1,34 @@
+<div align="center">
+    <img src="crabslab.png" alt="slabcraft for crabs" width="512" />
+</div>
+
+## what
+`crabslab` is a slab implementation focused on marshalling data from CPUs to GPUs.
+
+## why
+### Opinion
+Working with shaders is much easier using a slab.
+
+### rust-gpu
+This crate was made to work with [`rust-gpu`](https://github.com/EmbarkStudios/rust-gpu/).
+Specifically, using this crate it is possible to pack your types into a buffer on the CPU
+and then read your types from the slab on the GPU (in Rust).
+
+### Other no-std platforms
+Even though this crate was written with `rust-gpu` in mind, it should work in other `no-std`
+contexts.
+
+## how
+`crabslab` includes:
+* a few traits:
+  - `Slab`
+  - `GrowableSlab`
+  - `SlabItem`
+* a derive macro for `SlabItem`
+* a few structs for working with various slabs
+  - `Id`
+  - `Array`
+  - `Offset`
+* a helper struct `CpuSlab`
+* a feature-gated helper for using slabs with `wgpu` - `WgpuBuffer`
+  - [example](src/wgpu_slab.rs#L344)