Shaders: runtime and shader node codegen

node-graph/gcore-shaders/src/blending.rs

@@ -66,7 +66,7 @@ impl AlphaBlending {
 }
 
 #[repr(i32)]
-#[derive(Debug, Default, Clone, Copy, Eq, PartialEq, Hash)]
+#[derive(Debug, Default, Clone, Copy, Eq, PartialEq, Hash, bytemuck::NoUninit)]
 #[cfg_attr(feature = "std", derive(dyn_any::DynAny, specta::Type, serde::Serialize, serde::Deserialize))]
 pub enum BlendMode {
 	// Basic group

node-graph/gcore/src/raster_types.rs

@@ -137,7 +137,7 @@ mod gpu {
 
 	#[derive(Clone, Debug, PartialEq, Hash)]
 	pub struct GPU {
-		texture: wgpu::Texture,
+		pub texture: wgpu::Texture,
 	}
 
 	impl Sealed for Raster<GPU> {}

node-graph/graster-nodes/Cargo.toml

@@ -17,6 +17,7 @@ default = ["std"]
 shader-nodes = [
 	"std",
 	"dep:graphene-raster-nodes-shaders",
+	"dep:wgpu-executor",
 ]
 std = [
 	"dep:graphene-core",
@@ -39,6 +40,7 @@ node-macro = { workspace = true }
 # Local std dependencies
 dyn-any = { workspace = true, optional = true }
 graphene-core = { workspace = true, optional = true }
+wgpu-executor = { workspace = true, optional = true }
 graphene-raster-nodes-shaders = { path = "./shaders", optional = true }
 
 # Workspace dependencies

node-graph/graster-nodes/src/adjustments.rs

@@ -30,7 +30,7 @@ use num_traits::float::Float;
 // https://www.adobe.com/devnet-apps/photoshop/fileformatashtml/#:~:text=%27clrL%27%20%3D%20Color%20Lookup
 // https://www.adobe.com/devnet-apps/photoshop/fileformatashtml/#:~:text=Color%20Lookup%20(Photoshop%20CS6
 
-#[derive(Debug, Default, Clone, Copy, Eq, PartialEq, Hash, node_macro::ChoiceType)]
+#[derive(Debug, Default, Clone, Copy, Eq, PartialEq, Hash, node_macro::ChoiceType, bytemuck::NoUninit)]
 #[cfg_attr(feature = "std", derive(dyn_any::DynAny, specta::Type, serde::Serialize, serde::Deserialize))]
 #[widget(Dropdown)]
 #[repr(u32)]
@@ -560,7 +560,7 @@ pub enum RedGreenBlue {
 }
 
 /// Color Channel
-#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Hash, node_macro::ChoiceType)]
+#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Hash, node_macro::ChoiceType, bytemuck::NoUninit)]
 #[cfg_attr(feature = "std", derive(dyn_any::DynAny, specta::Type, serde::Serialize, serde::Deserialize))]
 #[widget(Radio)]
 #[repr(u32)]

node-graph/graster-nodes/src/blending_nodes.rs

@@ -132,7 +132,7 @@ pub fn apply_blend_mode(foreground: Color, background: Color, blend_mode: BlendM
 	}
 }
 
-#[node_macro::node(category("Raster"), shader_node(PerPixelAdjust))]
+#[node_macro::node(category("Raster"), cfg(feature = "std"))]
 fn blend<T: Blend<Color> + Send>(
 	_: impl Ctx,
 	#[implementations(

node-graph/graster-nodes/src/fullscreen_vertex.rs

@@ -0,0 +1,29 @@
+use glam::{Vec2, Vec4};
+use spirv_std::spirv;
+
+/// webgpu NDC is like OpenGL: (-1.0 .. 1.0, -1.0 .. 1.0, 0.0 .. 1.0)
+/// https://www.w3.org/TR/webgpu/#coordinate-systems
+///
+/// So to make a fullscreen triangle around a box at (-1..1):
+///
+/// ```norun
+/// 3  +
+///    |\
+/// 2  |  \
+///    |    \
+/// 1  +-----+
+///    |     |\
+/// 0  |  0  |  \
+///    |     |    \
+/// -1 +-----+-----+
+///   -1  0  1  2  3
+/// ```
+const FULLSCREEN_VERTICES: [Vec2; 3] = [Vec2::new(-1., -1.), Vec2::new(-1., 3.), Vec2::new(3., -1.)];
+
+#[spirv(vertex)]
+pub fn fullscreen_vertex(#[spirv(vertex_index)] vertex_index: u32, #[spirv(position)] gl_position: &mut Vec4) {
+	// broken on edition 2024 branch
+	// let vertex = unsafe { *FULLSCREEN_VERTICES.index_unchecked(vertex_index as usize) };
+	let vertex = FULLSCREEN_VERTICES[vertex_index as usize];
+	*gl_position = Vec4::from((vertex, 0., 1.));
+}

node-graph/graster-nodes/src/lib.rs

@@ -4,6 +4,11 @@ pub mod adjust;
 pub mod adjustments;
 pub mod blending_nodes;
 pub mod cubic_spline;
+pub mod fullscreen_vertex;
+
+/// required by shader macro
+#[cfg(feature = "shader-nodes")]
+pub use graphene_raster_nodes_shaders::WGSL_SHADER;
 
 #[cfg(feature = "std")]
 pub mod curve;

node-graph/node-macro/src/codegen.rs

@@ -295,7 +295,8 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result<TokenStre
 
 	let cfg = crate::shader_nodes::modify_cfg(attributes);
 	let node_input_accessor = generate_node_input_references(parsed, fn_generics, &field_idents, &graphene_core, &identifier, &cfg);
-	let shader_entry_point = attributes.shader_node.as_ref().map(|n| n.codegen_shader_entry_point(parsed)).unwrap_or(Ok(TokenStream2::new()))?;
+	let ShaderTokens { shader_entry_point, gpu_node } = attributes.shader_node.as_ref().map(|n| n.codegen(parsed)).unwrap_or(Ok(ShaderTokens::default()))?;
+
 	Ok(quote! {
 		/// Underlying implementation for [#struct_name]
 		#[inline]
@@ -387,6 +388,8 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result<TokenStre
 		}
 
 		#shader_entry_point
+
+		#gpu_node
 	})
 }
 
@@ -589,7 +592,7 @@ fn generate_register_node_impl(parsed: &ParsedNodeFn, field_names: &[&Ident], st
 	})
 }
 
-use crate::shader_nodes::CodegenShaderEntryPoint;
+use crate::shader_nodes::{ShaderCodegen, ShaderTokens};
 use syn::visit_mut::VisitMut;
 use syn::{GenericArgument, Lifetime, Type};
 

node-graph/node-macro/src/parsing.rs

@@ -39,7 +39,7 @@ pub(crate) struct ParsedNodeFn {
 	pub(crate) description: String,
 }
 
-#[derive(Debug, Default)]
+#[derive(Debug, Default, Clone)]
 pub(crate) struct NodeFnAttributes {
 	pub(crate) category: Option<LitStr>,
 	pub(crate) display_name: Option<LitStr>,
@@ -144,7 +144,7 @@ pub struct NodeParsedField {
 	pub implementations: Punctuated<Implementation, Comma>,
 }
 
-#[derive(Debug)]
+#[derive(Clone, Debug)]
 pub(crate) struct Input {
 	pub(crate) pat_ident: PatIdent,
 	pub(crate) ty: Type,
@@ -663,7 +663,7 @@ pub fn new_node_fn(attr: TokenStream2, item: TokenStream2) -> TokenStream2 {
 }
 
 impl ParsedNodeFn {
-	fn replace_impl_trait_in_input(&mut self) {
+	pub fn replace_impl_trait_in_input(&mut self) {
 		if let Type::ImplTrait(impl_trait) = self.input.ty.clone() {
 			let ident = Ident::new("_Input", impl_trait.span());
 			let mut bounds = impl_trait.bounds;

node-graph/node-macro/src/shader_nodes/mod.rs

@@ -3,49 +3,76 @@ use crate::shader_nodes::per_pixel_adjust::PerPixelAdjust;
 use proc_macro2::{Ident, TokenStream};
 use quote::quote;
 use strum::VariantNames;
-use syn::Error;
 use syn::parse::{Parse, ParseStream};
+use syn::punctuated::Punctuated;
+use syn::{Error, Token};
 
 pub mod per_pixel_adjust;
 
 pub const STD_FEATURE_GATE: &str = "std";
+pub const SHADER_NODES_FEATURE_GATE: &str = "shader-nodes";
 
 pub fn modify_cfg(attributes: &NodeFnAttributes) -> TokenStream {
-	match (&attributes.cfg, &attributes.shader_node) {
-		(Some(cfg), Some(_)) => quote!(#[cfg(all(#cfg, feature = #STD_FEATURE_GATE))]),
-		(Some(cfg), None) => quote!(#[cfg(#cfg)]),
-		(None, Some(_)) => quote!(#[cfg(feature = #STD_FEATURE_GATE)]),
-		(None, None) => quote!(),
-	}
+	let feature_gate = match &attributes.shader_node {
+		// shader node cfg is done on the mod
+		Some(ShaderNodeType::ShaderNode) => quote!(),
+		Some(_) => quote!(feature = #STD_FEATURE_GATE),
+		None => quote!(),
+	};
+	let cfgs: Punctuated<_, Token![,]> = match &attributes.cfg {
+		None => [&feature_gate].into_iter().collect(),
+		Some(cfg) => [cfg, &feature_gate].into_iter().collect(),
+	};
+	quote!(#[cfg(all(#cfgs))])
 }
 
-#[derive(Debug, VariantNames)]
+#[derive(Debug, Clone, VariantNames)]
 pub(crate) enum ShaderNodeType {
+	/// Marker for this node being in a gpu node crate, but not having a gpu implementation. This is distinct from not
+	/// declaring `shader_node` at all, as it will wrap the CPU node with a `#[cfg(feature = "std")]` feature gate.
+	None,
+	/// Marker for this node being a generated gpu node implementation, that should not emit anything to prevent
+	/// recursively generating more gpu nodes. But it still counts as a gpu node and will get the
+	/// `#[cfg(feature = "std")]` feature gate around it's impl.
+	ShaderNode,
 	PerPixelAdjust(PerPixelAdjust),
 }
 
 impl Parse for ShaderNodeType {
 	fn parse(input: ParseStream) -> syn::Result<Self> {
 		let ident: Ident = input.parse()?;
 		Ok(match ident.to_string().as_str() {
+			"None" => ShaderNodeType::None,
 			"PerPixelAdjust" => ShaderNodeType::PerPixelAdjust(PerPixelAdjust::parse(input)?),
 			_ => return Err(Error::new_spanned(&ident, format!("attr 'shader_node' must be one of {:?}", Self::VARIANTS))),
 		})
 	}
 }
 
-pub trait CodegenShaderEntryPoint {
-	fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result<TokenStream>;
+pub trait ShaderCodegen {
+	fn codegen(&self, parsed: &ParsedNodeFn) -> syn::Result<ShaderTokens>;
 }
 
-impl CodegenShaderEntryPoint for ShaderNodeType {
-	fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result<TokenStream> {
-		if parsed.is_async {
-			return Err(Error::new_spanned(&parsed.fn_name, "Shader nodes must not be async"));
+impl ShaderCodegen for ShaderNodeType {
+	fn codegen(&self, parsed: &ParsedNodeFn) -> syn::Result<ShaderTokens> {
+		match self {
+			ShaderNodeType::None | ShaderNodeType::ShaderNode => (),
+			_ => {
+				if parsed.is_async {
+					return Err(Error::new_spanned(&parsed.fn_name, "Shader nodes must not be async"));
+				}
+			}
 		}
 
 		match self {
-			ShaderNodeType::PerPixelAdjust(x) => x.codegen_shader_entry_point(parsed),
+			ShaderNodeType::None | ShaderNodeType::ShaderNode => Ok(ShaderTokens::default()),
+			ShaderNodeType::PerPixelAdjust(x) => x.codegen(parsed),
 		}
 	}
 }
+
+#[derive(Clone, Default)]
+pub struct ShaderTokens {
+	pub shader_entry_point: TokenStream,
+	pub gpu_node: TokenStream,
+}