From 6a0c07ab3e01e575bab60f8079aa55973455b74c Mon Sep 17 00:00:00 2001 From: firestar99 Date: Thu, 14 Aug 2025 15:35:21 +0200 Subject: [PATCH 01/17] shader-rt: initial --- node-graph/gcore/src/raster_types.rs | 2 +- .../graster-nodes/src/fullscreen_vertex.rs | 14 ++ node-graph/graster-nodes/src/lib.rs | 1 + node-graph/node-macro/src/codegen.rs | 9 +- node-graph/node-macro/src/parsing.rs | 2 +- node-graph/node-macro/src/shader_nodes/mod.rs | 9 +- .../src/shader_nodes/per_pixel_adjust.rs | 54 +++++- node-graph/wgpu-executor/src/lib.rs | 1 + .../wgpu-executor/src/shader_runtime/mod.rs | 20 +++ .../per_pixel_adjust_runtime.rs | 155 ++++++++++++++++++ 10 files changed, 260 insertions(+), 7 deletions(-) create mode 100644 node-graph/graster-nodes/src/fullscreen_vertex.rs create mode 100644 node-graph/wgpu-executor/src/shader_runtime/mod.rs create mode 100644 node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs diff --git a/node-graph/gcore/src/raster_types.rs b/node-graph/gcore/src/raster_types.rs index 97dd138145..7efae73fb8 100644 --- a/node-graph/gcore/src/raster_types.rs +++ b/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 {} diff --git a/node-graph/graster-nodes/src/fullscreen_vertex.rs b/node-graph/graster-nodes/src/fullscreen_vertex.rs new file mode 100644 index 0000000000..b8ef775b9f --- /dev/null +++ b/node-graph/graster-nodes/src/fullscreen_vertex.rs @@ -0,0 +1,14 @@ +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 +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.)); +} diff --git a/node-graph/graster-nodes/src/lib.rs b/node-graph/graster-nodes/src/lib.rs index 8dc169cea6..793b041205 100644 --- a/node-graph/graster-nodes/src/lib.rs +++ b/node-graph/graster-nodes/src/lib.rs @@ -4,6 +4,7 @@ pub mod adjust; pub mod adjustments; pub mod blending_nodes; pub mod cubic_spline; +pub mod fullscreen_vertex; #[cfg(feature = "std")] pub mod curve; diff --git a/node-graph/node-macro/src/codegen.rs b/node-graph/node-macro/src/codegen.rs index 05b8d19cc5..308f575f9b 100644 --- a/node-graph/node-macro/src/codegen.rs +++ b/node-graph/node-macro/src/codegen.rs @@ -295,7 +295,12 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result, _>(|n| Ok((n.codegen_shader_entry_point(parsed)?, n.codegen_gpu_node(parsed)?))) + .unwrap_or(Ok((TokenStream2::new(), TokenStream2::new())))?; + Ok(quote! { /// Underlying implementation for [#struct_name] #[inline] @@ -387,6 +392,8 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result, pub(crate) display_name: Option, diff --git a/node-graph/node-macro/src/shader_nodes/mod.rs b/node-graph/node-macro/src/shader_nodes/mod.rs index 0720869d01..3eff7fed14 100644 --- a/node-graph/node-macro/src/shader_nodes/mod.rs +++ b/node-graph/node-macro/src/shader_nodes/mod.rs @@ -19,7 +19,7 @@ pub fn modify_cfg(attributes: &NodeFnAttributes) -> TokenStream { } } -#[derive(Debug, VariantNames)] +#[derive(Debug, Clone, VariantNames)] pub(crate) enum ShaderNodeType { PerPixelAdjust(PerPixelAdjust), } @@ -36,6 +36,7 @@ impl Parse for ShaderNodeType { pub trait CodegenShaderEntryPoint { fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result; + fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result; } impl CodegenShaderEntryPoint for ShaderNodeType { @@ -48,4 +49,10 @@ impl CodegenShaderEntryPoint for ShaderNodeType { ShaderNodeType::PerPixelAdjust(x) => x.codegen_shader_entry_point(parsed), } } + + fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result { + match self { + ShaderNodeType::PerPixelAdjust(x) => x.codegen_gpu_node(parsed), + } + } } diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 0e220c8aae..7e3a644175 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -1,11 +1,13 @@ -use crate::parsing::{ParsedFieldType, ParsedNodeFn, RegularParsedField}; +use crate::parsing::{Input, NodeFnAttributes, ParsedField, ParsedFieldType, ParsedNodeFn, RegularParsedField}; use crate::shader_nodes::CodegenShaderEntryPoint; +use convert_case::{Case, Casing}; use proc_macro2::{Ident, TokenStream}; use quote::{ToTokens, format_ident, quote}; use std::borrow::Cow; use syn::parse::{Parse, ParseStream}; +use syn::{Path, Type, TypePath}; -#[derive(Debug)] +#[derive(Debug, Clone)] pub struct PerPixelAdjust {} impl Parse for PerPixelAdjust { @@ -17,7 +19,7 @@ impl Parse for PerPixelAdjust { impl CodegenShaderEntryPoint for PerPixelAdjust { fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result { let fn_name = &parsed.fn_name; - let gpu_mod = format_ident!("{}_gpu", parsed.fn_name); + let gpu_mod = format_ident!("{}_gpu_entry_point", parsed.fn_name); let spirv_image_ty = quote!(Image2d); // bindings for images start at 1 @@ -96,6 +98,52 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { } }) } + + fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result { + let fn_name = format_ident!("{}_gpu", parsed.fn_name); + let struct_name = format_ident!("{}", fn_name.to_string().to_case(Case::Pascal)); + let mod_name = fn_name.clone(); + + let fields = parsed + .fields + .iter() + .map(|f| match &f.ty { + ParsedFieldType::Regular(reg) => Ok(ParsedField { + ty: ParsedFieldType::Regular(RegularParsedField { gpu_image: false, ..reg.clone() }), + ..f.clone() + }), + ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(&f.pat_ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), + }) + .collect::>()?; + let body = quote! {}; + + crate::codegen::generate_node_code(&ParsedNodeFn { + vis: parsed.vis.clone(), + attributes: NodeFnAttributes { + shader_node: None, + ..parsed.attributes.clone() + }, + fn_name, + struct_name, + mod_name, + fn_generics: vec![], + where_clause: None, + input: Input { + pat_ident: parsed.input.pat_ident.clone(), + ty: Type::Path(TypePath { + path: Path::from(format_ident!("Ctx")), + qself: None, + }), + implementations: Default::default(), + }, + output_type: parsed.output_type.clone(), + is_async: true, + fields, + body, + crate_name: parsed.crate_name.clone(), + description: "".to_string(), + }) + } } struct Param<'a> { diff --git a/node-graph/wgpu-executor/src/lib.rs b/node-graph/wgpu-executor/src/lib.rs index 920a002c4e..b45db35a38 100644 --- a/node-graph/wgpu-executor/src/lib.rs +++ b/node-graph/wgpu-executor/src/lib.rs @@ -1,4 +1,5 @@ mod context; +pub mod shader_runtime; pub mod texture_upload; use anyhow::Result; diff --git a/node-graph/wgpu-executor/src/shader_runtime/mod.rs b/node-graph/wgpu-executor/src/shader_runtime/mod.rs new file mode 100644 index 0000000000..e7e0df8d94 --- /dev/null +++ b/node-graph/wgpu-executor/src/shader_runtime/mod.rs @@ -0,0 +1,20 @@ +use crate::Context; +use crate::shader_runtime::per_pixel_adjust_runtime::PerPixelAdjustShaderRuntime; + +pub mod per_pixel_adjust_runtime; + +pub const FULLSCREEN_VERTEX_SHADER_NAME: &str = "fullscreen_vertexfullscreen_vertex"; + +pub struct ShaderRuntime { + context: Context, + per_pixel_adjust: PerPixelAdjustShaderRuntime, +} + +impl ShaderRuntime { + pub fn new(context: &Context) -> Self { + Self { + context: context.clone(), + per_pixel_adjust: PerPixelAdjustShaderRuntime::new(), + } + } +} diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs new file mode 100644 index 0000000000..604a2c5bfd --- /dev/null +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -0,0 +1,155 @@ +use crate::Context; +use crate::shader_runtime::{FULLSCREEN_VERTEX_SHADER_NAME, ShaderRuntime}; +use futures::lock::Mutex; +use graphene_core::raster_types::{GPU, Raster}; +use graphene_core::table::{Table, TableRow}; +use std::borrow::Cow; +use std::collections::HashMap; +use wgpu::{ + BindGroupDescriptor, BindGroupEntry, BindingResource, ColorTargetState, Face, FragmentState, FrontFace, LoadOp, Operations, PolygonMode, PrimitiveState, PrimitiveTopology, + RenderPassColorAttachment, RenderPassDescriptor, RenderPipelineDescriptor, ShaderModuleDescriptor, ShaderSource, StoreOp, TextureDescriptor, TextureDimension, TextureFormat, + TextureViewDescriptor, VertexState, +}; + +pub struct PerPixelAdjustShaderRuntime { + // TODO: PerPixelAdjustGraphicsPipeline already contains the key as `name` + pipeline_cache: Mutex>, +} + +impl PerPixelAdjustShaderRuntime { + pub fn new() -> Self { + Self { + pipeline_cache: Mutex::new(HashMap::new()), + } + } +} + +impl ShaderRuntime { + pub async fn run_per_pixel_adjust(&self, input: Table>, info: &PerPixelAdjustInfo<'_>) -> Table> { + let mut cache = self.per_pixel_adjust.pipeline_cache.lock().await; + let pipeline = cache + .entry(info.fragment_shader_name.to_owned()) + .or_insert_with(|| PerPixelAdjustGraphicsPipeline::new(&self.context, &info)); + pipeline.run(&self.context, input) + } +} + +pub struct PerPixelAdjustInfo<'a> { + shader_wgsl: &'a str, + fragment_shader_name: &'a str, +} + +pub struct PerPixelAdjustGraphicsPipeline { + name: String, + pipeline: wgpu::RenderPipeline, +} + +impl PerPixelAdjustGraphicsPipeline { + pub fn new(context: &Context, info: &PerPixelAdjustInfo) -> Self { + let device = &context.device; + let name = info.fragment_shader_name.to_owned(); + let shader_module = device.create_shader_module(ShaderModuleDescriptor { + label: Some(&format!("PerPixelAdjust {} wgsl shader", name)), + source: ShaderSource::Wgsl(Cow::Borrowed(info.shader_wgsl)), + }); + let pipeline = device.create_render_pipeline(&RenderPipelineDescriptor { + label: Some(&format!("PerPixelAdjust {} Pipeline", name)), + layout: None, + vertex: VertexState { + module: &shader_module, + entry_point: Some(FULLSCREEN_VERTEX_SHADER_NAME), + compilation_options: Default::default(), + buffers: &[], + }, + primitive: PrimitiveState { + topology: PrimitiveTopology::TriangleList, + strip_index_format: None, + front_face: FrontFace::Ccw, + cull_mode: Some(Face::Back), + unclipped_depth: false, + polygon_mode: PolygonMode::Fill, + conservative: false, + }, + depth_stencil: None, + multisample: Default::default(), + fragment: Some(FragmentState { + module: &shader_module, + entry_point: Some(&name), + compilation_options: Default::default(), + targets: &[Some(ColorTargetState { + format: TextureFormat::Rgba32Float, + blend: None, + write_mask: Default::default(), + })], + }), + multiview: None, + cache: None, + }); + Self { pipeline, name } + } + + pub fn run(&self, context: &Context, input: Table>) -> Table> { + let device = &context.device; + let name = self.name.as_str(); + + let mut cmd = device.create_command_encoder(&wgpu::CommandEncoderDescriptor { label: Some("gpu_invert") }); + let out = input + .iter() + .map(|instance| { + let tex_in = &instance.element.texture; + let view_in = tex_in.create_view(&TextureViewDescriptor::default()); + let format = tex_in.format(); + + let bind_group = device.create_bind_group(&BindGroupDescriptor { + label: Some(&format!("{name} bind group")), + // `get_bind_group_layout` allocates unnecessary memory, we could create it manually to not do that + layout: &self.pipeline.get_bind_group_layout(0), + entries: &[BindGroupEntry { + binding: 0, + resource: BindingResource::TextureView(&view_in), + }], + }); + + let tex_out = device.create_texture(&TextureDescriptor { + label: Some(&format!("{name} texture out")), + size: tex_in.size(), + mip_level_count: 1, + sample_count: 1, + dimension: TextureDimension::D2, + format, + usage: wgpu::TextureUsages::TEXTURE_BINDING | wgpu::TextureUsages::COPY_DST | wgpu::TextureUsages::COPY_SRC | wgpu::TextureUsages::RENDER_ATTACHMENT, + view_formats: &[format], + }); + + let view_out = tex_out.create_view(&TextureViewDescriptor::default()); + let mut rp = cmd.begin_render_pass(&RenderPassDescriptor { + label: Some(&format!("{name} render pipeline")), + color_attachments: &[Some(RenderPassColorAttachment { + view: &view_out, + resolve_target: None, + ops: Operations { + // should be dont_care but wgpu doesn't expose that + load: LoadOp::Clear(wgpu::Color::BLACK), + store: StoreOp::Store, + }, + })], + depth_stencil_attachment: None, + timestamp_writes: None, + occlusion_query_set: None, + }); + rp.set_pipeline(&self.pipeline); + rp.set_bind_group(0, Some(&bind_group), &[]); + rp.draw(0..3, 0..1); + + TableRow { + element: Raster::new(GPU { texture: tex_out }), + transform: *instance.transform, + alpha_blending: *instance.alpha_blending, + source_node_id: *instance.source_node_id, + } + }) + .collect::>(); + context.queue.submit([cmd.finish()]); + out + } +} From 5981022549e730f0c6f12679f2e426ac643cc113 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Tue, 19 Aug 2025 12:41:56 +0200 Subject: [PATCH 02/17] shader-rt: fix recursion when generating shader node --- node-graph/node-macro/src/shader_nodes/mod.rs | 15 +++++++++-- .../src/shader_nodes/per_pixel_adjust.rs | 26 +++++++++++++------ 2 files changed, 31 insertions(+), 10 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/mod.rs b/node-graph/node-macro/src/shader_nodes/mod.rs index 3eff7fed14..26e1ebba87 100644 --- a/node-graph/node-macro/src/shader_nodes/mod.rs +++ b/node-graph/node-macro/src/shader_nodes/mod.rs @@ -21,6 +21,10 @@ pub fn modify_cfg(attributes: &NodeFnAttributes) -> TokenStream { #[derive(Debug, Clone, VariantNames)] pub(crate) enum ShaderNodeType { + /// 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. + GpuNode, PerPixelAdjust(PerPixelAdjust), } @@ -41,17 +45,24 @@ pub trait CodegenShaderEntryPoint { impl CodegenShaderEntryPoint for ShaderNodeType { fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result { - if parsed.is_async { - return Err(Error::new_spanned(&parsed.fn_name, "Shader nodes must not be async")); + match self { + ShaderNodeType::GpuNode => (), + _ => { + if parsed.is_async { + return Err(Error::new_spanned(&parsed.fn_name, "Shader nodes must not be async")); + } + } } match self { + ShaderNodeType::GpuNode => Ok(TokenStream::new()), ShaderNodeType::PerPixelAdjust(x) => x.codegen_shader_entry_point(parsed), } } fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result { match self { + ShaderNodeType::GpuNode => Ok(TokenStream::new()), ShaderNodeType::PerPixelAdjust(x) => x.codegen_gpu_node(parsed), } } diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 7e3a644175..36cbff50b1 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -1,11 +1,12 @@ use crate::parsing::{Input, NodeFnAttributes, ParsedField, ParsedFieldType, ParsedNodeFn, RegularParsedField}; -use crate::shader_nodes::CodegenShaderEntryPoint; +use crate::shader_nodes::{CodegenShaderEntryPoint, ShaderNodeType}; use convert_case::{Case, Casing}; -use proc_macro2::{Ident, TokenStream}; +use proc_macro2::{Ident, Span, TokenStream}; use quote::{ToTokens, format_ident, quote}; use std::borrow::Cow; use syn::parse::{Parse, ParseStream}; -use syn::{Path, Type, TypePath}; +use syn::punctuated::Punctuated; +use syn::{Path, Token, TraitBound, TraitBoundModifier, Type, TypeImplTrait, TypeParamBound}; #[derive(Debug, Clone)] pub struct PerPixelAdjust {} @@ -115,12 +116,16 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(&f.pat_ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), }) .collect::>()?; - let body = quote! {}; + let body = quote! { + { + + } + }; crate::codegen::generate_node_code(&ParsedNodeFn { vis: parsed.vis.clone(), attributes: NodeFnAttributes { - shader_node: None, + shader_node: Some(ShaderNodeType::GpuNode), ..parsed.attributes.clone() }, fn_name, @@ -130,9 +135,14 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { where_clause: None, input: Input { pat_ident: parsed.input.pat_ident.clone(), - ty: Type::Path(TypePath { - path: Path::from(format_ident!("Ctx")), - qself: None, + ty: Type::ImplTrait(TypeImplTrait { + impl_token: Token![impl](Span::call_site()), + bounds: Punctuated::from_iter([TypeParamBound::Trait(TraitBound { + paren_token: None, + modifier: TraitBoundModifier::None, + lifetimes: None, + path: Path::from(format_ident!("Ctx")), + })]), }), implementations: Default::default(), }, From 17f2da0c59e36c89940cc2a591337fda000e3545 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Tue, 19 Aug 2025 15:46:53 +0200 Subject: [PATCH 03/17] shader-rt: replace gpu node's args and ret types with `Raster` --- .../src/shader_nodes/per_pixel_adjust.rs | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 36cbff50b1..b263a2c157 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -1,6 +1,7 @@ use crate::parsing::{Input, NodeFnAttributes, ParsedField, ParsedFieldType, ParsedNodeFn, RegularParsedField}; use crate::shader_nodes::{CodegenShaderEntryPoint, ShaderNodeType}; use convert_case::{Case, Casing}; +use proc_macro_crate::FoundCrate; use proc_macro2::{Ident, Span, TokenStream}; use quote::{ToTokens, format_ident, quote}; use std::borrow::Cow; @@ -105,14 +106,24 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { let struct_name = format_ident!("{}", fn_name.to_string().to_case(Case::Pascal)); let mod_name = fn_name.clone(); + let gcore = match &parsed.crate_name { + FoundCrate::Itself => format_ident!("crate"), + FoundCrate::Name(name) => format_ident!("{name}"), + }; + let raster_gpu = syn::parse2::(quote!(#gcore::table::Table<#gcore::raster_types::Raster<#gcore::raster_types::GPU>>))?; + let fields = parsed .fields .iter() .map(|f| match &f.ty { - ParsedFieldType::Regular(reg) => Ok(ParsedField { - ty: ParsedFieldType::Regular(RegularParsedField { gpu_image: false, ..reg.clone() }), + ParsedFieldType::Regular(reg @ RegularParsedField { gpu_image: true, .. }) => Ok(ParsedField { + ty: ParsedFieldType::Regular(RegularParsedField { + ty: raster_gpu.clone(), + ..reg.clone() + }), ..f.clone() }), + ParsedFieldType::Regular(RegularParsedField { gpu_image: false, .. }) => Ok(f.clone()), ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(&f.pat_ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), }) .collect::>()?; @@ -146,7 +157,7 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { }), implementations: Default::default(), }, - output_type: parsed.output_type.clone(), + output_type: raster_gpu, is_async: true, fields, body, From 2e660be093e24be0e5e720b764394092f63c463c Mon Sep 17 00:00:00 2001 From: firestar99 Date: Tue, 19 Aug 2025 16:08:39 +0200 Subject: [PATCH 04/17] shader-rt: properly cfg out the gpu node --- node-graph/node-macro/src/codegen.rs | 10 ++---- node-graph/node-macro/src/shader_nodes/mod.rs | 24 ++++++------- .../src/shader_nodes/per_pixel_adjust.rs | 35 ++++++++++++++----- 3 files changed, 41 insertions(+), 28 deletions(-) diff --git a/node-graph/node-macro/src/codegen.rs b/node-graph/node-macro/src/codegen.rs index 308f575f9b..b3a1b28be3 100644 --- a/node-graph/node-macro/src/codegen.rs +++ b/node-graph/node-macro/src/codegen.rs @@ -295,11 +295,7 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result, _>(|n| Ok((n.codegen_shader_entry_point(parsed)?, n.codegen_gpu_node(parsed)?))) - .unwrap_or(Ok((TokenStream2::new(), TokenStream2::new())))?; + let ShaderTokens { shader_entry_point, gpu_node } = attributes.shader_node.as_ref().map(|n| n.codegen(parsed, &cfg)).unwrap_or(Ok(ShaderTokens::default()))?; Ok(quote! { /// Underlying implementation for [#struct_name] @@ -393,7 +389,7 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result syn::Result; - fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result; +pub trait ShaderCodegen { + fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result; } -impl CodegenShaderEntryPoint for ShaderNodeType { - fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result { +impl ShaderCodegen for ShaderNodeType { + fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { match self { ShaderNodeType::GpuNode => (), _ => { @@ -55,15 +54,14 @@ impl CodegenShaderEntryPoint for ShaderNodeType { } match self { - ShaderNodeType::GpuNode => Ok(TokenStream::new()), - ShaderNodeType::PerPixelAdjust(x) => x.codegen_shader_entry_point(parsed), + ShaderNodeType::GpuNode => Ok(ShaderTokens::default()), + ShaderNodeType::PerPixelAdjust(x) => x.codegen(parsed, node_cfg), } } +} - fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result { - match self { - ShaderNodeType::GpuNode => Ok(TokenStream::new()), - ShaderNodeType::PerPixelAdjust(x) => x.codegen_gpu_node(parsed), - } - } +#[derive(Clone, Default)] +pub struct ShaderTokens { + pub shader_entry_point: TokenStream, + pub gpu_node: TokenStream, } diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index b263a2c157..e21e3cdfb6 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -1,5 +1,5 @@ use crate::parsing::{Input, NodeFnAttributes, ParsedField, ParsedFieldType, ParsedNodeFn, RegularParsedField}; -use crate::shader_nodes::{CodegenShaderEntryPoint, ShaderNodeType}; +use crate::shader_nodes::{ShaderCodegen, ShaderNodeType, ShaderTokens}; use convert_case::{Case, Casing}; use proc_macro_crate::FoundCrate; use proc_macro2::{Ident, Span, TokenStream}; @@ -7,7 +7,7 @@ use quote::{ToTokens, format_ident, quote}; use std::borrow::Cow; use syn::parse::{Parse, ParseStream}; use syn::punctuated::Punctuated; -use syn::{Path, Token, TraitBound, TraitBoundModifier, Type, TypeImplTrait, TypeParamBound}; +use syn::{Token, TraitBound, TraitBoundModifier, Type, TypeImplTrait, TypeParamBound}; #[derive(Debug, Clone)] pub struct PerPixelAdjust {} @@ -18,10 +18,19 @@ impl Parse for PerPixelAdjust { } } -impl CodegenShaderEntryPoint for PerPixelAdjust { +impl ShaderCodegen for PerPixelAdjust { + fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { + Ok(ShaderTokens { + shader_entry_point: self.codegen_shader_entry_point(parsed)?, + gpu_node: self.codegen_gpu_node(parsed, node_cfg)?, + }) + } +} + +impl PerPixelAdjust { fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result { let fn_name = &parsed.fn_name; - let gpu_mod = format_ident!("{}_gpu_entry_point", parsed.fn_name); + let gpu_mod = format_ident!("{}_gpu_entry_point", fn_name); let spirv_image_ty = quote!(Image2d); // bindings for images start at 1 @@ -101,7 +110,7 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { }) } - fn codegen_gpu_node(&self, parsed: &ParsedNodeFn) -> syn::Result { + fn codegen_gpu_node(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { let fn_name = format_ident!("{}_gpu", parsed.fn_name); let struct_name = format_ident!("{}", fn_name.to_string().to_case(Case::Pascal)); let mod_name = fn_name.clone(); @@ -127,13 +136,14 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(&f.pat_ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), }) .collect::>()?; + let body = quote! { { } }; - crate::codegen::generate_node_code(&ParsedNodeFn { + let gpu_node = crate::codegen::generate_node_code(&ParsedNodeFn { vis: parsed.vis.clone(), attributes: NodeFnAttributes { shader_node: Some(ShaderNodeType::GpuNode), @@ -141,7 +151,7 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { }, fn_name, struct_name, - mod_name, + mod_name: mod_name.clone(), fn_generics: vec![], where_clause: None, input: Input { @@ -152,7 +162,7 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { paren_token: None, modifier: TraitBoundModifier::None, lifetimes: None, - path: Path::from(format_ident!("Ctx")), + path: syn::parse2(quote!(#gcore::context::Ctx))?, })]), }), implementations: Default::default(), @@ -163,6 +173,15 @@ impl CodegenShaderEntryPoint for PerPixelAdjust { body, crate_name: parsed.crate_name.clone(), description: "".to_string(), + })?; + + Ok(quote! { + #node_cfg + mod #mod_name { + use super::*; + + #gpu_node + } }) } } From 716dcba9968b12d58072b8beb6c3ec81b4c4b733 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Tue, 19 Aug 2025 16:37:36 +0200 Subject: [PATCH 05/17] shader-rt: fix `impl Context` in the wrong places --- node-graph/node-macro/src/parsing.rs | 4 ++-- .../src/shader_nodes/per_pixel_adjust.rs | 23 ++++++++----------- 2 files changed, 11 insertions(+), 16 deletions(-) diff --git a/node-graph/node-macro/src/parsing.rs b/node-graph/node-macro/src/parsing.rs index 498cee8eab..69e035d9f3 100644 --- a/node-graph/node-macro/src/parsing.rs +++ b/node-graph/node-macro/src/parsing.rs @@ -144,7 +144,7 @@ pub struct NodeParsedField { pub implementations: Punctuated, } -#[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; diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index e21e3cdfb6..d1f4c99a1d 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -2,12 +2,12 @@ use crate::parsing::{Input, NodeFnAttributes, ParsedField, ParsedFieldType, Pars use crate::shader_nodes::{ShaderCodegen, ShaderNodeType, ShaderTokens}; use convert_case::{Case, Casing}; use proc_macro_crate::FoundCrate; -use proc_macro2::{Ident, Span, TokenStream}; +use proc_macro2::{Ident, TokenStream}; use quote::{ToTokens, format_ident, quote}; use std::borrow::Cow; use syn::parse::{Parse, ParseStream}; use syn::punctuated::Punctuated; -use syn::{Token, TraitBound, TraitBoundModifier, Type, TypeImplTrait, TypeParamBound}; +use syn::{Type, parse_quote}; #[derive(Debug, Clone)] pub struct PerPixelAdjust {} @@ -119,7 +119,7 @@ impl PerPixelAdjust { FoundCrate::Itself => format_ident!("crate"), FoundCrate::Name(name) => format_ident!("{name}"), }; - let raster_gpu = syn::parse2::(quote!(#gcore::table::Table<#gcore::raster_types::Raster<#gcore::raster_types::GPU>>))?; + let raster_gpu: Type = parse_quote!(#gcore::table::Table<#gcore::raster_types::Raster<#gcore::raster_types::GPU>>); let fields = parsed .fields @@ -128,6 +128,7 @@ impl PerPixelAdjust { ParsedFieldType::Regular(reg @ RegularParsedField { gpu_image: true, .. }) => Ok(ParsedField { ty: ParsedFieldType::Regular(RegularParsedField { ty: raster_gpu.clone(), + implementations: Punctuated::default(), ..reg.clone() }), ..f.clone() @@ -143,7 +144,7 @@ impl PerPixelAdjust { } }; - let gpu_node = crate::codegen::generate_node_code(&ParsedNodeFn { + let mut parsed_node_fn = ParsedNodeFn { vis: parsed.vis.clone(), attributes: NodeFnAttributes { shader_node: Some(ShaderNodeType::GpuNode), @@ -156,15 +157,7 @@ impl PerPixelAdjust { where_clause: None, input: Input { pat_ident: parsed.input.pat_ident.clone(), - ty: Type::ImplTrait(TypeImplTrait { - impl_token: Token![impl](Span::call_site()), - bounds: Punctuated::from_iter([TypeParamBound::Trait(TraitBound { - paren_token: None, - modifier: TraitBoundModifier::None, - lifetimes: None, - path: syn::parse2(quote!(#gcore::context::Ctx))?, - })]), - }), + ty: parse_quote!(impl #gcore::context::Ctx), implementations: Default::default(), }, output_type: raster_gpu, @@ -173,7 +166,9 @@ impl PerPixelAdjust { body, crate_name: parsed.crate_name.clone(), description: "".to_string(), - })?; + }; + parsed_node_fn.replace_impl_trait_in_input(); + let gpu_node = crate::codegen::generate_node_code(&parsed_node_fn)?; Ok(quote! { #node_cfg From e9ae52d95202af989adb583186533fc67f14c487 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Tue, 19 Aug 2025 16:40:08 +0200 Subject: [PATCH 06/17] shader-rt: disable gpu blend node, needs two images --- node-graph/graster-nodes/src/blending_nodes.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/node-graph/graster-nodes/src/blending_nodes.rs b/node-graph/graster-nodes/src/blending_nodes.rs index 182f6a802d..13f93773d4 100644 --- a/node-graph/graster-nodes/src/blending_nodes.rs +++ b/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 + Send>( _: impl Ctx, #[implementations( From b25b059b5f0cc1913fcd478bce248a3fd61bfd4f Mon Sep 17 00:00:00 2001 From: firestar99 Date: Thu, 21 Aug 2025 11:08:14 +0200 Subject: [PATCH 07/17] shader-rt: connect shader runtime --- Cargo.lock | 1 + node-graph/graster-nodes/Cargo.toml | 2 + node-graph/graster-nodes/src/lib.rs | 4 + .../src/shader_nodes/per_pixel_adjust.rs | 75 ++++++++++++++++--- node-graph/wgpu-executor/src/lib.rs | 3 + .../per_pixel_adjust_runtime.rs | 11 ++- 6 files changed, 80 insertions(+), 16 deletions(-) diff --git a/Cargo.lock b/Cargo.lock index 97edb1e0c7..7189e30817 100644 --- a/Cargo.lock +++ b/Cargo.lock @@ -2191,6 +2191,7 @@ dependencies = [ "specta", "spirv-std", "tokio", + "wgpu-executor", ] [[package]] diff --git a/node-graph/graster-nodes/Cargo.toml b/node-graph/graster-nodes/Cargo.toml index a66dbb5eb3..7ed5357f00 100644 --- a/node-graph/graster-nodes/Cargo.toml +++ b/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 diff --git a/node-graph/graster-nodes/src/lib.rs b/node-graph/graster-nodes/src/lib.rs index 793b041205..d5383df034 100644 --- a/node-graph/graster-nodes/src/lib.rs +++ b/node-graph/graster-nodes/src/lib.rs @@ -6,6 +6,10 @@ pub mod blending_nodes; pub mod cubic_spline; pub mod fullscreen_vertex; +/// required by shader macro +#[cfg(feature = "std")] +pub use graphene_raster_nodes_shaders::WGSL_SHADER; + #[cfg(feature = "std")] pub mod curve; #[cfg(feature = "std")] diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index d1f4c99a1d..84780538d2 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -7,7 +7,7 @@ use quote::{ToTokens, format_ident, quote}; use std::borrow::Cow; use syn::parse::{Parse, ParseStream}; use syn::punctuated::Punctuated; -use syn::{Type, parse_quote}; +use syn::{PatIdent, Type, parse_quote}; #[derive(Debug, Clone)] pub struct PerPixelAdjust {} @@ -20,15 +20,14 @@ impl Parse for PerPixelAdjust { impl ShaderCodegen for PerPixelAdjust { fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { - Ok(ShaderTokens { - shader_entry_point: self.codegen_shader_entry_point(parsed)?, - gpu_node: self.codegen_gpu_node(parsed, node_cfg)?, - }) + let (shader_entry_point, entry_point_name) = self.codegen_shader_entry_point(parsed)?; + let gpu_node = self.codegen_gpu_node(parsed, node_cfg, &entry_point_name)?; + Ok(ShaderTokens { shader_entry_point, gpu_node }) } } impl PerPixelAdjust { - fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result { + fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result<(TokenStream, TokenStream)> { let fn_name = &parsed.fn_name; let gpu_mod = format_ident!("{}_gpu_entry_point", fn_name); let spirv_image_ty = quote!(Image2d); @@ -82,7 +81,10 @@ impl PerPixelAdjust { .collect::>(); let context = quote!(()); - Ok(quote! { + let entry_point_name = format_ident!("ENTRY_POINT_NAME"); + let entry_point_sym = quote!(#gpu_mod::#entry_point_name); + + let shader_entry_point = quote! { pub mod #gpu_mod { use super::*; use graphene_core_shaders::color::Color; @@ -91,6 +93,8 @@ impl PerPixelAdjust { use spirv_std::image::{Image2d, ImageWithMethods}; use spirv_std::image::sample_with::lod; + pub const #entry_point_name: &str = core::concat!(core::module_path!(), "::entry_point"); + pub struct Uniform { #(#uniform_members),* } @@ -107,10 +111,11 @@ impl PerPixelAdjust { *color_out = color.to_vec4(); } } - }) + }; + Ok((shader_entry_point, entry_point_sym)) } - fn codegen_gpu_node(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { + fn codegen_gpu_node(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream, entry_point_name: &TokenStream) -> syn::Result { let fn_name = format_ident!("{}_gpu", parsed.fn_name); let struct_name = format_ident!("{}", fn_name.to_string().to_case(Case::Pascal)); let mod_name = fn_name.clone(); @@ -121,7 +126,8 @@ impl PerPixelAdjust { }; let raster_gpu: Type = parse_quote!(#gcore::table::Table<#gcore::raster_types::Raster<#gcore::raster_types::GPU>>); - let fields = parsed + // adapt fields for gpu node + let mut fields = parsed .fields .iter() .map(|f| match &f.ty { @@ -136,11 +142,55 @@ impl PerPixelAdjust { ParsedFieldType::Regular(RegularParsedField { gpu_image: false, .. }) => Ok(f.clone()), ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(&f.pat_ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), }) - .collect::>()?; + .collect::>>()?; + + // wgpu_executor field + let wgpu_executor = format_ident!("__wgpu_executor"); + fields.push(ParsedField { + pat_ident: PatIdent { + attrs: vec![], + by_ref: None, + mutability: None, + ident: parse_quote!(#wgpu_executor), + subpat: None, + }, + name: None, + description: "".to_string(), + widget_override: Default::default(), + ty: ParsedFieldType::Regular(RegularParsedField { + ty: parse_quote!(WgpuExecutor), + exposed: false, + value_source: Default::default(), + number_soft_min: None, + number_soft_max: None, + number_hard_min: None, + number_hard_max: None, + number_mode_range: None, + implementations: Default::default(), + gpu_image: false, + }), + number_display_decimal_places: None, + number_step: None, + unit: None, + }); + + // exactly one gpu_image field, may be expanded later + let gpu_image_field = { + let mut iter = fields.iter().filter(|f| matches!(f.ty, ParsedFieldType::Regular(RegularParsedField { gpu_image: true, .. }))); + match (iter.next(), iter.next()) { + (Some(v), None) => Ok(v), + (Some(_), Some(more)) => Err(syn::Error::new_spanned(&more.pat_ident, "No more than one parameter must be annotated with `#[gpu_image]`")), + (None, _) => Err(syn::Error::new_spanned(&parsed.fn_name, "At least one parameter must be annotated with `#[gpu_image]`")), + }? + }; + let gpu_image = &gpu_image_field.pat_ident.ident; let body = quote! { { - + #wgpu_executor.shader_runtime.run_per_pixel_adjust(#gpu_image, &::wgpu_executor::shader_runtime::per_pixel_adjust_runtime::PerPixelAdjustInfo { + wgsl_shader: crate::WGSL_SHADER, + fragment_shader_name: super::#entry_point_name, + }).await } }; @@ -174,6 +224,7 @@ impl PerPixelAdjust { #node_cfg mod #mod_name { use super::*; + use wgpu_executor::WgpuExecutor; #gpu_node } diff --git a/node-graph/wgpu-executor/src/lib.rs b/node-graph/wgpu-executor/src/lib.rs index b45db35a38..0b42dd631e 100644 --- a/node-graph/wgpu-executor/src/lib.rs +++ b/node-graph/wgpu-executor/src/lib.rs @@ -2,6 +2,7 @@ mod context; pub mod shader_runtime; pub mod texture_upload; +use crate::shader_runtime::ShaderRuntime; use anyhow::Result; pub use context::Context; use dyn_any::StaticType; @@ -19,6 +20,7 @@ use wgpu::{Origin3d, SurfaceConfiguration, TextureAspect}; pub struct WgpuExecutor { pub context: Context, vello_renderer: Mutex, + pub shader_runtime: ShaderRuntime, } impl std::fmt::Debug for WgpuExecutor { @@ -196,6 +198,7 @@ impl WgpuExecutor { .ok()?; Some(Self { + shader_runtime: ShaderRuntime::new(&context), context, vello_renderer: vello_renderer.into(), }) diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs index 604a2c5bfd..2b01bfc705 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -35,8 +35,8 @@ impl ShaderRuntime { } pub struct PerPixelAdjustInfo<'a> { - shader_wgsl: &'a str, - fragment_shader_name: &'a str, + pub wgsl_shader: &'a str, + pub fragment_shader_name: &'a str, } pub struct PerPixelAdjustGraphicsPipeline { @@ -48,9 +48,12 @@ impl PerPixelAdjustGraphicsPipeline { pub fn new(context: &Context, info: &PerPixelAdjustInfo) -> Self { let device = &context.device; let name = info.fragment_shader_name.to_owned(); + // TODO workaround to naga removing `:` + let fragment_name = name.replace(":", ""); + let shader_module = device.create_shader_module(ShaderModuleDescriptor { label: Some(&format!("PerPixelAdjust {} wgsl shader", name)), - source: ShaderSource::Wgsl(Cow::Borrowed(info.shader_wgsl)), + source: ShaderSource::Wgsl(Cow::Borrowed(info.wgsl_shader)), }); let pipeline = device.create_render_pipeline(&RenderPipelineDescriptor { label: Some(&format!("PerPixelAdjust {} Pipeline", name)), @@ -74,7 +77,7 @@ impl PerPixelAdjustGraphicsPipeline { multisample: Default::default(), fragment: Some(FragmentState { module: &shader_module, - entry_point: Some(&name), + entry_point: Some(&fragment_name), compilation_options: Default::default(), targets: &[Some(ColorTargetState { format: TextureFormat::Rgba32Float, From 6a116fdb2bc04ccb5bb161fb103e3a8357186445 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Wed, 20 Aug 2025 15:51:22 +0200 Subject: [PATCH 08/17] shader-rt: pass WgpuExecutor by reference --- node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 84780538d2..8dc97eba2e 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -158,7 +158,7 @@ impl PerPixelAdjust { description: "".to_string(), widget_override: Default::default(), ty: ParsedFieldType::Regular(RegularParsedField { - ty: parse_quote!(WgpuExecutor), + ty: parse_quote!(&'a WgpuExecutor), exposed: false, value_source: Default::default(), number_soft_min: None, @@ -203,7 +203,7 @@ impl PerPixelAdjust { fn_name, struct_name, mod_name: mod_name.clone(), - fn_generics: vec![], + fn_generics: vec![parse_quote!('a: 'n)], where_clause: None, input: Input { pat_ident: parsed.input.pat_ident.clone(), From 1c89d7a5f3ddd00143dd157960233d793917ee88 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Wed, 20 Aug 2025 16:44:54 +0200 Subject: [PATCH 09/17] shader-rt: correct bindings with derpy arg buffer --- .../src/shader_nodes/per_pixel_adjust.rs | 4 +- .../wgpu-executor/src/shader_runtime/mod.rs | 5 ++ .../per_pixel_adjust_runtime.rs | 57 ++++++++++++------- 3 files changed, 44 insertions(+), 22 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 8dc97eba2e..06a4508015 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -187,10 +187,10 @@ impl PerPixelAdjust { let body = quote! { { - #wgpu_executor.shader_runtime.run_per_pixel_adjust(#gpu_image, &::wgpu_executor::shader_runtime::per_pixel_adjust_runtime::PerPixelAdjustInfo { + #wgpu_executor.shader_runtime.run_per_pixel_adjust(&::wgpu_executor::shader_runtime::Shaders { wgsl_shader: crate::WGSL_SHADER, fragment_shader_name: super::#entry_point_name, - }).await + }, #gpu_image, &()).await } }; diff --git a/node-graph/wgpu-executor/src/shader_runtime/mod.rs b/node-graph/wgpu-executor/src/shader_runtime/mod.rs index e7e0df8d94..2745d5bda8 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/mod.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/mod.rs @@ -18,3 +18,8 @@ impl ShaderRuntime { } } } + +pub struct Shaders<'a> { + pub wgsl_shader: &'a str, + pub fragment_shader_name: &'a str, +} diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs index 2b01bfc705..352adb7e9e 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -1,13 +1,15 @@ use crate::Context; -use crate::shader_runtime::{FULLSCREEN_VERTEX_SHADER_NAME, ShaderRuntime}; +use crate::shader_runtime::{FULLSCREEN_VERTEX_SHADER_NAME, ShaderRuntime, Shaders}; +use bytemuck::NoUninit; use futures::lock::Mutex; use graphene_core::raster_types::{GPU, Raster}; use graphene_core::table::{Table, TableRow}; use std::borrow::Cow; use std::collections::HashMap; +use wgpu::util::{BufferInitDescriptor, DeviceExt}; use wgpu::{ - BindGroupDescriptor, BindGroupEntry, BindingResource, ColorTargetState, Face, FragmentState, FrontFace, LoadOp, Operations, PolygonMode, PrimitiveState, PrimitiveTopology, - RenderPassColorAttachment, RenderPassDescriptor, RenderPipelineDescriptor, ShaderModuleDescriptor, ShaderSource, StoreOp, TextureDescriptor, TextureDimension, TextureFormat, + BindGroupDescriptor, BindGroupEntry, BindingResource, Buffer, BufferBinding, BufferUsages, ColorTargetState, Face, FragmentState, FrontFace, LoadOp, Operations, PolygonMode, PrimitiveState, + PrimitiveTopology, RenderPassColorAttachment, RenderPassDescriptor, RenderPipelineDescriptor, ShaderModuleDescriptor, ShaderSource, StoreOp, TextureDescriptor, TextureDimension, TextureFormat, TextureViewDescriptor, VertexState, }; @@ -25,18 +27,20 @@ impl PerPixelAdjustShaderRuntime { } impl ShaderRuntime { - pub async fn run_per_pixel_adjust(&self, input: Table>, info: &PerPixelAdjustInfo<'_>) -> Table> { + pub async fn run_per_pixel_adjust(&self, shaders: &Shaders<'_>, textures: Table>, args: &T) -> Table> { let mut cache = self.per_pixel_adjust.pipeline_cache.lock().await; let pipeline = cache - .entry(info.fragment_shader_name.to_owned()) - .or_insert_with(|| PerPixelAdjustGraphicsPipeline::new(&self.context, &info)); - pipeline.run(&self.context, input) - } -} + .entry(shaders.fragment_shader_name.to_owned()) + .or_insert_with(|| PerPixelAdjustGraphicsPipeline::new(&self.context, &shaders)); -pub struct PerPixelAdjustInfo<'a> { - pub wgsl_shader: &'a str, - pub fragment_shader_name: &'a str, + let device = &self.context.device; + let arg_buffer = device.create_buffer_init(&BufferInitDescriptor { + label: Some(&format!("{} arg buffer", pipeline.name.as_str())), + usage: BufferUsages::STORAGE, + contents: bytemuck::bytes_of(args), + }); + pipeline.dispatch(&self.context, textures, &arg_buffer) + } } pub struct PerPixelAdjustGraphicsPipeline { @@ -45,11 +49,14 @@ pub struct PerPixelAdjustGraphicsPipeline { } impl PerPixelAdjustGraphicsPipeline { - pub fn new(context: &Context, info: &PerPixelAdjustInfo) -> Self { + pub fn new(context: &Context, info: &Shaders) -> Self { let device = &context.device; let name = info.fragment_shader_name.to_owned(); + // TODO workaround to naga removing `:` - let fragment_name = name.replace(":", ""); + let fragment_name = &name; + let fragment_name = &fragment_name[(fragment_name.find("::").unwrap() + 2)..]; + let fragment_name = fragment_name.replace(":", ""); let shader_module = device.create_shader_module(ShaderModuleDescriptor { label: Some(&format!("PerPixelAdjust {} wgsl shader", name)), @@ -91,12 +98,12 @@ impl PerPixelAdjustGraphicsPipeline { Self { pipeline, name } } - pub fn run(&self, context: &Context, input: Table>) -> Table> { + pub fn dispatch(&self, context: &Context, textures: Table>, arg_buffer: &Buffer) -> Table> { let device = &context.device; let name = self.name.as_str(); let mut cmd = device.create_command_encoder(&wgpu::CommandEncoderDescriptor { label: Some("gpu_invert") }); - let out = input + let out = textures .iter() .map(|instance| { let tex_in = &instance.element.texture; @@ -107,10 +114,20 @@ impl PerPixelAdjustGraphicsPipeline { label: Some(&format!("{name} bind group")), // `get_bind_group_layout` allocates unnecessary memory, we could create it manually to not do that layout: &self.pipeline.get_bind_group_layout(0), - entries: &[BindGroupEntry { - binding: 0, - resource: BindingResource::TextureView(&view_in), - }], + entries: &[ + BindGroupEntry { + binding: 0, + resource: BindingResource::Buffer(BufferBinding { + buffer: arg_buffer, + offset: 0, + size: None, + }), + }, + BindGroupEntry { + binding: 1, + resource: BindingResource::TextureView(&view_in), + }, + ], }); let tex_out = device.create_texture(&TextureDescriptor { From 24954219903bf8dd1955beb3852c5033b113b61f Mon Sep 17 00:00:00 2001 From: firestar99 Date: Wed, 20 Aug 2025 17:18:03 +0200 Subject: [PATCH 10/17] shader-rt: manual pipeline layout, fixing errors when bindings got DCE'd --- .../src/shader_nodes/per_pixel_adjust.rs | 2 +- .../per_pixel_adjust_runtime.rs | 42 ++++++++++++++++--- 2 files changed, 37 insertions(+), 7 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 06a4508015..3203e4328d 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -190,7 +190,7 @@ impl PerPixelAdjust { #wgpu_executor.shader_runtime.run_per_pixel_adjust(&::wgpu_executor::shader_runtime::Shaders { wgsl_shader: crate::WGSL_SHADER, fragment_shader_name: super::#entry_point_name, - }, #gpu_image, &()).await + }, #gpu_image, &1u32).await } }; diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs index 352adb7e9e..746e1a900e 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -8,9 +8,9 @@ use std::borrow::Cow; use std::collections::HashMap; use wgpu::util::{BufferInitDescriptor, DeviceExt}; use wgpu::{ - BindGroupDescriptor, BindGroupEntry, BindingResource, Buffer, BufferBinding, BufferUsages, ColorTargetState, Face, FragmentState, FrontFace, LoadOp, Operations, PolygonMode, PrimitiveState, - PrimitiveTopology, RenderPassColorAttachment, RenderPassDescriptor, RenderPipelineDescriptor, ShaderModuleDescriptor, ShaderSource, StoreOp, TextureDescriptor, TextureDimension, TextureFormat, - TextureViewDescriptor, VertexState, + BindGroupDescriptor, BindGroupEntry, BindGroupLayoutDescriptor, BindGroupLayoutEntry, BindingResource, BindingType, Buffer, BufferBinding, BufferBindingType, BufferUsages, ColorTargetState, Face, + FragmentState, FrontFace, LoadOp, Operations, PipelineLayoutDescriptor, PolygonMode, PrimitiveState, PrimitiveTopology, RenderPassColorAttachment, RenderPassDescriptor, RenderPipelineDescriptor, + ShaderModuleDescriptor, ShaderSource, ShaderStages, StoreOp, TextureDescriptor, TextureDimension, TextureFormat, TextureSampleType, TextureViewDescriptor, TextureViewDimension, VertexState, }; pub struct PerPixelAdjustShaderRuntime { @@ -53,18 +53,48 @@ impl PerPixelAdjustGraphicsPipeline { let device = &context.device; let name = info.fragment_shader_name.to_owned(); - // TODO workaround to naga removing `:` let fragment_name = &name; let fragment_name = &fragment_name[(fragment_name.find("::").unwrap() + 2)..]; + // TODO workaround to naga removing `:` let fragment_name = fragment_name.replace(":", ""); - let shader_module = device.create_shader_module(ShaderModuleDescriptor { label: Some(&format!("PerPixelAdjust {} wgsl shader", name)), source: ShaderSource::Wgsl(Cow::Borrowed(info.wgsl_shader)), }); + + let pipeline_layout = device.create_pipeline_layout(&PipelineLayoutDescriptor { + label: Some(&format!("PerPixelAdjust {} PipelineLayout", name)), + bind_group_layouts: &[&device.create_bind_group_layout(&BindGroupLayoutDescriptor { + label: Some(&format!("PerPixelAdjust {} BindGroupLayout 0", name)), + entries: &[ + BindGroupLayoutEntry { + binding: 0, + visibility: ShaderStages::FRAGMENT, + ty: BindingType::Buffer { + ty: BufferBindingType::Storage { read_only: true }, + has_dynamic_offset: false, + min_binding_size: None, + }, + count: None, + }, + BindGroupLayoutEntry { + binding: 1, + visibility: ShaderStages::FRAGMENT, + ty: BindingType::Texture { + sample_type: TextureSampleType::Float { filterable: false }, + view_dimension: TextureViewDimension::D2, + multisampled: false, + }, + count: None, + }, + ], + })], + push_constant_ranges: &[], + }); + let pipeline = device.create_render_pipeline(&RenderPipelineDescriptor { label: Some(&format!("PerPixelAdjust {} Pipeline", name)), - layout: None, + layout: Some(&pipeline_layout), vertex: VertexState { module: &shader_module, entry_point: Some(FULLSCREEN_VERTEX_SHADER_NAME), From 8fc9eb7e57b168d0c93d42329458a25db2890b11 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Wed, 20 Aug 2025 17:22:10 +0200 Subject: [PATCH 11/17] shader-rt: correct RT format, working invert gpu node --- .../src/shader_runtime/per_pixel_adjust_runtime.rs | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs index 746e1a900e..119d346956 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -117,7 +117,7 @@ impl PerPixelAdjustGraphicsPipeline { entry_point: Some(&fragment_name), compilation_options: Default::default(), targets: &[Some(ColorTargetState { - format: TextureFormat::Rgba32Float, + format: TextureFormat::Rgba8UnormSrgb, blend: None, write_mask: Default::default(), })], From 16c67ecc684100144182bc7606de9926f7566658 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Thu, 21 Aug 2025 11:53:53 +0200 Subject: [PATCH 12/17] shader-rt: cleanup codegen with common sym struct --- .../src/shader_nodes/per_pixel_adjust.rs | 162 +++++++++++------- 1 file changed, 97 insertions(+), 65 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 3203e4328d..1096d88daa 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -20,59 +20,90 @@ impl Parse for PerPixelAdjust { impl ShaderCodegen for PerPixelAdjust { fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { - let (shader_entry_point, entry_point_name) = self.codegen_shader_entry_point(parsed)?; - let gpu_node = self.codegen_gpu_node(parsed, node_cfg, &entry_point_name)?; - Ok(ShaderTokens { shader_entry_point, gpu_node }) - } -} - -impl PerPixelAdjust { - fn codegen_shader_entry_point(&self, parsed: &ParsedNodeFn) -> syn::Result<(TokenStream, TokenStream)> { let fn_name = &parsed.fn_name; - let gpu_mod = format_ident!("{}_gpu_entry_point", fn_name); - let spirv_image_ty = quote!(Image2d); + // categorize params and assign image bindings // bindings for images start at 1 - let mut binding_cnt = 0; - let params = parsed - .fields - .iter() - .map(|f| { - let ident = &f.pat_ident; - match &f.ty { - ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), - ParsedFieldType::Regular(RegularParsedField { gpu_image: false, ty, .. }) => Ok(Param { - ident: Cow::Borrowed(&ident.ident), - ty: Cow::Owned(ty.to_token_stream()), - param_type: ParamType::Uniform, - }), - ParsedFieldType::Regular(RegularParsedField { gpu_image: true, .. }) => { - binding_cnt += 1; - Ok(Param { - ident: Cow::Owned(format_ident!("image_{}", &ident.ident)), - ty: Cow::Borrowed(&spirv_image_ty), - param_type: ParamType::Image { binding: binding_cnt }, - }) + let params = { + let mut binding_cnt = 0; + parsed + .fields + .iter() + .map(|f| { + let ident = &f.pat_ident; + match &f.ty { + ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), + ParsedFieldType::Regular(RegularParsedField { gpu_image: false, ty, .. }) => Ok(Param { + ident: Cow::Borrowed(&ident.ident), + ty: ty.to_token_stream(), + param_type: ParamType::Uniform, + }), + ParsedFieldType::Regular(RegularParsedField { gpu_image: true, .. }) => { + binding_cnt += 1; + Ok(Param { + ident: Cow::Owned(format_ident!("image_{}", &ident.ident)), + ty: quote!(Image2d), + param_type: ParamType::Image { binding: binding_cnt }, + }) + } } - } - }) - .collect::>>()?; + }) + .collect::>>()? + }; + + let entry_point_mod = format_ident!("{}_gpu_entry_point", fn_name); + let entry_point_name_ident = format_ident!("ENTRY_POINT_NAME"); + let entry_point_name = quote!(#entry_point_mod::#entry_point_name_ident); + let gpu_node_mod = format_ident!("{}_gpu", fn_name); + + let codegen = PerPixelAdjustCodegen { + parsed, + node_cfg, + params, + entry_point_mod, + entry_point_name_ident, + entry_point_name, + gpu_node_mod, + }; - let uniform_members = params + Ok(ShaderTokens { + shader_entry_point: codegen.codegen_shader_entry_point()?, + gpu_node: codegen.codegen_gpu_node()?, + }) + } +} + +pub struct PerPixelAdjustCodegen<'a> { + parsed: &'a ParsedNodeFn, + node_cfg: &'a TokenStream, + params: Vec>, + entry_point_mod: Ident, + entry_point_name_ident: Ident, + entry_point_name: TokenStream, + gpu_node_mod: Ident, +} + +impl PerPixelAdjustCodegen<'_> { + fn codegen_shader_entry_point(&self) -> syn::Result { + let fn_name = &self.parsed.fn_name; + let uniform_members = self + .params .iter() .filter_map(|Param { ident, ty, param_type }| match param_type { ParamType::Image { .. } => None, ParamType::Uniform => Some(quote! {#ident: #ty}), }) .collect::>(); - let image_params = params + let image_params = self + .params .iter() .filter_map(|Param { ident, ty, param_type }| match param_type { ParamType::Image { binding } => Some(quote! {#[spirv(descriptor_set = 0, binding = #binding)] #ident: &#ty}), ParamType::Uniform => None, }) .collect::>(); - let call_args = params + let call_args = self + .params .iter() .map(|Param { ident, param_type, .. }| match param_type { ParamType::Image { .. } => quote!(Color::from_vec4(#ident.fetch_with(texel_coord, lod(0)))), @@ -81,11 +112,10 @@ impl PerPixelAdjust { .collect::>(); let context = quote!(()); - let entry_point_name = format_ident!("ENTRY_POINT_NAME"); - let entry_point_sym = quote!(#gpu_mod::#entry_point_name); - - let shader_entry_point = quote! { - pub mod #gpu_mod { + let entry_point_mod = &self.entry_point_mod; + let entry_point_name = &self.entry_point_name_ident; + Ok(quote! { + pub mod #entry_point_mod { use super::*; use graphene_core_shaders::color::Color; use spirv_std::spirv; @@ -111,23 +141,19 @@ impl PerPixelAdjust { *color_out = color.to_vec4(); } } - }; - Ok((shader_entry_point, entry_point_sym)) + }) } - fn codegen_gpu_node(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream, entry_point_name: &TokenStream) -> syn::Result { - let fn_name = format_ident!("{}_gpu", parsed.fn_name); - let struct_name = format_ident!("{}", fn_name.to_string().to_case(Case::Pascal)); - let mod_name = fn_name.clone(); - - let gcore = match &parsed.crate_name { + fn codegen_gpu_node(&self) -> syn::Result { + let gcore = match &self.parsed.crate_name { FoundCrate::Itself => format_ident!("crate"), FoundCrate::Name(name) => format_ident!("{name}"), }; - let raster_gpu: Type = parse_quote!(#gcore::table::Table<#gcore::raster_types::Raster<#gcore::raster_types::GPU>>); // adapt fields for gpu node - let mut fields = parsed + let raster_gpu: Type = parse_quote!(#gcore::table::Table<#gcore::raster_types::Raster<#gcore::raster_types::GPU>>); + let mut fields = self + .parsed .fields .iter() .map(|f| match &f.ty { @@ -144,7 +170,7 @@ impl PerPixelAdjust { }) .collect::>>()?; - // wgpu_executor field + // insert wgpu_executor field let wgpu_executor = format_ident!("__wgpu_executor"); fields.push(ParsedField { pat_ident: PatIdent { @@ -174,17 +200,19 @@ impl PerPixelAdjust { unit: None, }); - // exactly one gpu_image field, may be expanded later + // find exactly one gpu_image field, runtime doesn't support more than 1 atm let gpu_image_field = { let mut iter = fields.iter().filter(|f| matches!(f.ty, ParsedFieldType::Regular(RegularParsedField { gpu_image: true, .. }))); match (iter.next(), iter.next()) { (Some(v), None) => Ok(v), (Some(_), Some(more)) => Err(syn::Error::new_spanned(&more.pat_ident, "No more than one parameter must be annotated with `#[gpu_image]`")), - (None, _) => Err(syn::Error::new_spanned(&parsed.fn_name, "At least one parameter must be annotated with `#[gpu_image]`")), + (None, _) => Err(syn::Error::new_spanned(&self.parsed.fn_name, "At least one parameter must be annotated with `#[gpu_image]`")), }? }; let gpu_image = &gpu_image_field.pat_ident.ident; + // node function body + let entry_point_name = &self.entry_point_name; let body = quote! { { #wgpu_executor.shader_runtime.run_per_pixel_adjust(&::wgpu_executor::shader_runtime::Shaders { @@ -194,19 +222,20 @@ impl PerPixelAdjust { } }; + // call node codegen let mut parsed_node_fn = ParsedNodeFn { - vis: parsed.vis.clone(), + vis: self.parsed.vis.clone(), attributes: NodeFnAttributes { shader_node: Some(ShaderNodeType::GpuNode), - ..parsed.attributes.clone() + ..self.parsed.attributes.clone() }, - fn_name, - struct_name, - mod_name: mod_name.clone(), + fn_name: self.gpu_node_mod.clone(), + struct_name: format_ident!("{}", self.gpu_node_mod.to_string().to_case(Case::Pascal)), + mod_name: self.gpu_node_mod.clone(), fn_generics: vec![parse_quote!('a: 'n)], where_clause: None, input: Input { - pat_ident: parsed.input.pat_ident.clone(), + pat_ident: self.parsed.input.pat_ident.clone(), ty: parse_quote!(impl #gcore::context::Ctx), implementations: Default::default(), }, @@ -214,19 +243,22 @@ impl PerPixelAdjust { is_async: true, fields, body, - crate_name: parsed.crate_name.clone(), + crate_name: self.parsed.crate_name.clone(), description: "".to_string(), }; parsed_node_fn.replace_impl_trait_in_input(); - let gpu_node = crate::codegen::generate_node_code(&parsed_node_fn)?; + let gpu_node_impl = crate::codegen::generate_node_code(&parsed_node_fn)?; + // wrap node in `mod #gpu_node_mod` + let node_cfg = self.node_cfg; + let gpu_node_mod = &self.gpu_node_mod; Ok(quote! { #node_cfg - mod #mod_name { + mod #gpu_node_mod { use super::*; use wgpu_executor::WgpuExecutor; - #gpu_node + #gpu_node_impl } }) } @@ -234,7 +266,7 @@ impl PerPixelAdjust { struct Param<'a> { ident: Cow<'a, Ident>, - ty: Cow<'a, TokenStream>, + ty: TokenStream, param_type: ParamType, } From 1128271b4162c45a928835bf2666fd07988f1b3a Mon Sep 17 00:00:00 2001 From: firestar99 Date: Thu, 21 Aug 2025 12:38:23 +0200 Subject: [PATCH 13/17] shader-rt: correct arg buffer handling --- node-graph/gcore-shaders/src/blending.rs | 2 +- node-graph/graster-nodes/src/adjustments.rs | 4 +- .../src/shader_nodes/per_pixel_adjust.rs | 91 +++++++++++--- .../wgpu-executor/src/shader_runtime/mod.rs | 5 - .../per_pixel_adjust_runtime.rs | 112 ++++++++++++------ 5 files changed, 152 insertions(+), 62 deletions(-) diff --git a/node-graph/gcore-shaders/src/blending.rs b/node-graph/gcore-shaders/src/blending.rs index c3701e2cc0..b305dd0910 100644 --- a/node-graph/gcore-shaders/src/blending.rs +++ b/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 diff --git a/node-graph/graster-nodes/src/adjustments.rs b/node-graph/graster-nodes/src/adjustments.rs index 35de34c76b..dc3bd15b82 100644 --- a/node-graph/graster-nodes/src/adjustments.rs +++ b/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)] diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 1096d88daa..57e4fcefa9 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -22,11 +22,11 @@ impl ShaderCodegen for PerPixelAdjust { fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { let fn_name = &parsed.fn_name; - // categorize params and assign image bindings - // bindings for images start at 1 - let params = { - let mut binding_cnt = 0; - parsed + let mut params; + let has_uniform; + { + // categorize params + params = parsed .fields .iter() .map(|f| { @@ -39,30 +39,50 @@ impl ShaderCodegen for PerPixelAdjust { param_type: ParamType::Uniform, }), ParsedFieldType::Regular(RegularParsedField { gpu_image: true, .. }) => { - binding_cnt += 1; - Ok(Param { + let param = Param { ident: Cow::Owned(format_ident!("image_{}", &ident.ident)), ty: quote!(Image2d), - param_type: ParamType::Image { binding: binding_cnt }, - }) + param_type: ParamType::Image { binding: 0 }, + }; + Ok(param) } } }) - .collect::>>()? - }; + .collect::>>()?; + + has_uniform = params.iter().any(|p| matches!(p.param_type, ParamType::Uniform)); + + // assign image bindings + // if an arg_buffer exists, bindings for images start at 1 to leave 0 for arg buffer + let mut binding_cnt = if has_uniform { 1 } else { 0 }; + for p in params.iter_mut() { + match &mut p.param_type { + ParamType::Image { binding } => { + *binding = binding_cnt; + binding_cnt += 1; + } + ParamType::Uniform => {} + } + } + } let entry_point_mod = format_ident!("{}_gpu_entry_point", fn_name); let entry_point_name_ident = format_ident!("ENTRY_POINT_NAME"); let entry_point_name = quote!(#entry_point_mod::#entry_point_name_ident); + let uniform_struct_ident = format_ident!("Uniform"); + let uniform_struct = quote!(#entry_point_mod::#uniform_struct_ident); let gpu_node_mod = format_ident!("{}_gpu", fn_name); let codegen = PerPixelAdjustCodegen { parsed, node_cfg, params, + has_uniform, entry_point_mod, entry_point_name_ident, entry_point_name, + uniform_struct_ident, + uniform_struct, gpu_node_mod, }; @@ -77,9 +97,12 @@ pub struct PerPixelAdjustCodegen<'a> { parsed: &'a ParsedNodeFn, node_cfg: &'a TokenStream, params: Vec>, + has_uniform: bool, entry_point_mod: Ident, entry_point_name_ident: Ident, entry_point_name: TokenStream, + uniform_struct_ident: Ident, + uniform_struct: TokenStream, gpu_node_mod: Ident, } @@ -114,6 +137,7 @@ impl PerPixelAdjustCodegen<'_> { let entry_point_mod = &self.entry_point_mod; let entry_point_name = &self.entry_point_name_ident; + let uniform_struct_ident = &self.uniform_struct_ident; Ok(quote! { pub mod #entry_point_mod { use super::*; @@ -125,8 +149,10 @@ impl PerPixelAdjustCodegen<'_> { pub const #entry_point_name: &str = core::concat!(core::module_path!(), "::entry_point"); - pub struct Uniform { - #(#uniform_members),* + #[repr(C)] + #[derive(Copy, Clone, bytemuck::NoUninit)] + pub struct #uniform_struct_ident { + #(pub #uniform_members),* } #[spirv(fragment)] @@ -158,6 +184,11 @@ impl PerPixelAdjustCodegen<'_> { .iter() .map(|f| match &f.ty { ParsedFieldType::Regular(reg @ RegularParsedField { gpu_image: true, .. }) => Ok(ParsedField { + pat_ident: PatIdent { + mutability: None, + by_ref: None, + ..f.pat_ident.clone() + }, ty: ParsedFieldType::Regular(RegularParsedField { ty: raster_gpu.clone(), implementations: Punctuated::default(), @@ -165,7 +196,14 @@ impl PerPixelAdjustCodegen<'_> { }), ..f.clone() }), - ParsedFieldType::Regular(RegularParsedField { gpu_image: false, .. }) => Ok(f.clone()), + ParsedFieldType::Regular(RegularParsedField { gpu_image: false, .. }) => Ok(ParsedField { + pat_ident: PatIdent { + mutability: None, + by_ref: None, + ..f.pat_ident.clone() + }, + ..f.clone() + }), ParsedFieldType::Node { .. } => Err(syn::Error::new_spanned(&f.pat_ident, "PerPixelAdjust shader nodes cannot accept other nodes as generics")), }) .collect::>>()?; @@ -211,14 +249,35 @@ impl PerPixelAdjustCodegen<'_> { }; let gpu_image = &gpu_image_field.pat_ident.ident; + // uniform buffer struct construction + let has_uniform = self.has_uniform; + let uniform_buffer = if has_uniform { + let uniform_struct = &self.uniform_struct; + let uniform_members = self + .params + .iter() + .filter_map(|p| match p.param_type { + ParamType::Image { .. } => None, + ParamType::Uniform => Some(p.ident.as_ref()), + }) + .collect::>(); + quote!(Some(&super::#uniform_struct { + #(#uniform_members),* + })) + } else { + // explicit generics placed here cause it's easier than explicitly writing `run_per_pixel_adjust::<()>` + quote!(Option::<&()>::None) + }; + // node function body let entry_point_name = &self.entry_point_name; let body = quote! { { - #wgpu_executor.shader_runtime.run_per_pixel_adjust(&::wgpu_executor::shader_runtime::Shaders { + #wgpu_executor.shader_runtime.run_per_pixel_adjust(&::wgpu_executor::shader_runtime::per_pixel_adjust_runtime::Shaders { wgsl_shader: crate::WGSL_SHADER, fragment_shader_name: super::#entry_point_name, - }, #gpu_image, &1u32).await + has_uniform: #has_uniform, + }, #gpu_image, #uniform_buffer).await } }; diff --git a/node-graph/wgpu-executor/src/shader_runtime/mod.rs b/node-graph/wgpu-executor/src/shader_runtime/mod.rs index 2745d5bda8..e7e0df8d94 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/mod.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/mod.rs @@ -18,8 +18,3 @@ impl ShaderRuntime { } } } - -pub struct Shaders<'a> { - pub wgsl_shader: &'a str, - pub fragment_shader_name: &'a str, -} diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs index 119d346956..d958e0650d 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -1,5 +1,5 @@ use crate::Context; -use crate::shader_runtime::{FULLSCREEN_VERTEX_SHADER_NAME, ShaderRuntime, Shaders}; +use crate::shader_runtime::{FULLSCREEN_VERTEX_SHADER_NAME, ShaderRuntime}; use bytemuck::NoUninit; use futures::lock::Mutex; use graphene_core::raster_types::{GPU, Raster}; @@ -27,24 +27,33 @@ impl PerPixelAdjustShaderRuntime { } impl ShaderRuntime { - pub async fn run_per_pixel_adjust(&self, shaders: &Shaders<'_>, textures: Table>, args: &T) -> Table> { + pub async fn run_per_pixel_adjust(&self, shaders: &Shaders<'_>, textures: Table>, args: Option<&T>) -> Table> { let mut cache = self.per_pixel_adjust.pipeline_cache.lock().await; let pipeline = cache .entry(shaders.fragment_shader_name.to_owned()) .or_insert_with(|| PerPixelAdjustGraphicsPipeline::new(&self.context, &shaders)); - let device = &self.context.device; - let arg_buffer = device.create_buffer_init(&BufferInitDescriptor { - label: Some(&format!("{} arg buffer", pipeline.name.as_str())), - usage: BufferUsages::STORAGE, - contents: bytemuck::bytes_of(args), + let arg_buffer = args.map(|args| { + let device = &self.context.device; + device.create_buffer_init(&BufferInitDescriptor { + label: Some(&format!("{} arg buffer", pipeline.name.as_str())), + usage: BufferUsages::STORAGE, + contents: bytemuck::bytes_of(args), + }) }); - pipeline.dispatch(&self.context, textures, &arg_buffer) + pipeline.dispatch(&self.context, textures, arg_buffer) } } +pub struct Shaders<'a> { + pub wgsl_shader: &'a str, + pub fragment_shader_name: &'a str, + pub has_uniform: bool, +} + pub struct PerPixelAdjustGraphicsPipeline { name: String, + has_uniform: bool, pipeline: wgpu::RenderPipeline, } @@ -62,32 +71,46 @@ impl PerPixelAdjustGraphicsPipeline { source: ShaderSource::Wgsl(Cow::Borrowed(info.wgsl_shader)), }); + let entries: &[_] = if info.has_uniform { + &[ + BindGroupLayoutEntry { + binding: 0, + visibility: ShaderStages::FRAGMENT, + ty: BindingType::Buffer { + ty: BufferBindingType::Storage { read_only: true }, + has_dynamic_offset: false, + min_binding_size: None, + }, + count: None, + }, + BindGroupLayoutEntry { + binding: 1, + visibility: ShaderStages::FRAGMENT, + ty: BindingType::Texture { + sample_type: TextureSampleType::Float { filterable: false }, + view_dimension: TextureViewDimension::D2, + multisampled: false, + }, + count: None, + }, + ] + } else { + &[BindGroupLayoutEntry { + binding: 0, + visibility: ShaderStages::FRAGMENT, + ty: BindingType::Texture { + sample_type: TextureSampleType::Float { filterable: false }, + view_dimension: TextureViewDimension::D2, + multisampled: false, + }, + count: None, + }] + }; let pipeline_layout = device.create_pipeline_layout(&PipelineLayoutDescriptor { label: Some(&format!("PerPixelAdjust {} PipelineLayout", name)), bind_group_layouts: &[&device.create_bind_group_layout(&BindGroupLayoutDescriptor { label: Some(&format!("PerPixelAdjust {} BindGroupLayout 0", name)), - entries: &[ - BindGroupLayoutEntry { - binding: 0, - visibility: ShaderStages::FRAGMENT, - ty: BindingType::Buffer { - ty: BufferBindingType::Storage { read_only: true }, - has_dynamic_offset: false, - min_binding_size: None, - }, - count: None, - }, - BindGroupLayoutEntry { - binding: 1, - visibility: ShaderStages::FRAGMENT, - ty: BindingType::Texture { - sample_type: TextureSampleType::Float { filterable: false }, - view_dimension: TextureViewDimension::D2, - multisampled: false, - }, - count: None, - }, - ], + entries, })], push_constant_ranges: &[], }); @@ -125,10 +148,15 @@ impl PerPixelAdjustGraphicsPipeline { multiview: None, cache: None, }); - Self { pipeline, name } + Self { + pipeline, + name, + has_uniform: info.has_uniform, + } } - pub fn dispatch(&self, context: &Context, textures: Table>, arg_buffer: &Buffer) -> Table> { + pub fn dispatch(&self, context: &Context, textures: Table>, arg_buffer: Option) -> Table> { + assert_eq!(self.has_uniform, arg_buffer.is_some()); let device = &context.device; let name = self.name.as_str(); @@ -140,11 +168,8 @@ impl PerPixelAdjustGraphicsPipeline { let view_in = tex_in.create_view(&TextureViewDescriptor::default()); let format = tex_in.format(); - let bind_group = device.create_bind_group(&BindGroupDescriptor { - label: Some(&format!("{name} bind group")), - // `get_bind_group_layout` allocates unnecessary memory, we could create it manually to not do that - layout: &self.pipeline.get_bind_group_layout(0), - entries: &[ + let entries: &[_] = if let Some(arg_buffer) = arg_buffer.as_ref() { + &[ BindGroupEntry { binding: 0, resource: BindingResource::Buffer(BufferBinding { @@ -157,7 +182,18 @@ impl PerPixelAdjustGraphicsPipeline { binding: 1, resource: BindingResource::TextureView(&view_in), }, - ], + ] + } else { + &[BindGroupEntry { + binding: 0, + resource: BindingResource::TextureView(&view_in), + }] + }; + let bind_group = device.create_bind_group(&BindGroupDescriptor { + label: Some(&format!("{name} bind group")), + // `get_bind_group_layout` allocates unnecessary memory, we could create it manually to not do that + layout: &self.pipeline.get_bind_group_layout(0), + entries, }); let tex_out = device.create_texture(&TextureDescriptor { From 676aea33feabc53c443fec27a99f315999b9485b Mon Sep 17 00:00:00 2001 From: firestar99 Date: Fri, 29 Aug 2025 12:42:37 +0200 Subject: [PATCH 14/17] shader-nodes feature: put shader nodes behind feature gate --- node-graph/graster-nodes/src/lib.rs | 2 +- node-graph/node-macro/src/codegen.rs | 2 +- node-graph/node-macro/src/shader_nodes/mod.rs | 37 ++++++++++++------- .../src/shader_nodes/per_pixel_adjust.rs | 11 ++---- 4 files changed, 30 insertions(+), 22 deletions(-) diff --git a/node-graph/graster-nodes/src/lib.rs b/node-graph/graster-nodes/src/lib.rs index d5383df034..080504f9ea 100644 --- a/node-graph/graster-nodes/src/lib.rs +++ b/node-graph/graster-nodes/src/lib.rs @@ -7,7 +7,7 @@ pub mod cubic_spline; pub mod fullscreen_vertex; /// required by shader macro -#[cfg(feature = "std")] +#[cfg(feature = "shader-nodes")] pub use graphene_raster_nodes_shaders::WGSL_SHADER; #[cfg(feature = "std")] diff --git a/node-graph/node-macro/src/codegen.rs b/node-graph/node-macro/src/codegen.rs index b3a1b28be3..98bb5588a9 100644 --- a/node-graph/node-macro/src/codegen.rs +++ b/node-graph/node-macro/src/codegen.rs @@ -295,7 +295,7 @@ pub(crate) fn generate_node_code(parsed: &ParsedNodeFn) -> syn::Result 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, 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. - GpuNode, + ShaderNode, PerPixelAdjust(PerPixelAdjust), } @@ -32,6 +42,7 @@ impl Parse for ShaderNodeType { fn parse(input: ParseStream) -> syn::Result { 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))), }) @@ -39,13 +50,13 @@ impl Parse for ShaderNodeType { } pub trait ShaderCodegen { - fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result; + fn codegen(&self, parsed: &ParsedNodeFn) -> syn::Result; } impl ShaderCodegen for ShaderNodeType { - fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { + fn codegen(&self, parsed: &ParsedNodeFn) -> syn::Result { match self { - ShaderNodeType::GpuNode => (), + ShaderNodeType::None | ShaderNodeType::ShaderNode => (), _ => { if parsed.is_async { return Err(Error::new_spanned(&parsed.fn_name, "Shader nodes must not be async")); @@ -54,8 +65,8 @@ impl ShaderCodegen for ShaderNodeType { } match self { - ShaderNodeType::GpuNode => Ok(ShaderTokens::default()), - ShaderNodeType::PerPixelAdjust(x) => x.codegen(parsed, node_cfg), + ShaderNodeType::None | ShaderNodeType::ShaderNode => Ok(ShaderTokens::default()), + ShaderNodeType::PerPixelAdjust(x) => x.codegen(parsed), } } } diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index 57e4fcefa9..cd0d911694 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -1,5 +1,5 @@ use crate::parsing::{Input, NodeFnAttributes, ParsedField, ParsedFieldType, ParsedNodeFn, RegularParsedField}; -use crate::shader_nodes::{ShaderCodegen, ShaderNodeType, ShaderTokens}; +use crate::shader_nodes::{SHADER_NODES_FEATURE_GATE, ShaderCodegen, ShaderNodeType, ShaderTokens}; use convert_case::{Case, Casing}; use proc_macro_crate::FoundCrate; use proc_macro2::{Ident, TokenStream}; @@ -19,7 +19,7 @@ impl Parse for PerPixelAdjust { } impl ShaderCodegen for PerPixelAdjust { - fn codegen(&self, parsed: &ParsedNodeFn, node_cfg: &TokenStream) -> syn::Result { + fn codegen(&self, parsed: &ParsedNodeFn) -> syn::Result { let fn_name = &parsed.fn_name; let mut params; @@ -75,7 +75,6 @@ impl ShaderCodegen for PerPixelAdjust { let codegen = PerPixelAdjustCodegen { parsed, - node_cfg, params, has_uniform, entry_point_mod, @@ -95,7 +94,6 @@ impl ShaderCodegen for PerPixelAdjust { pub struct PerPixelAdjustCodegen<'a> { parsed: &'a ParsedNodeFn, - node_cfg: &'a TokenStream, params: Vec>, has_uniform: bool, entry_point_mod: Ident, @@ -285,7 +283,7 @@ impl PerPixelAdjustCodegen<'_> { let mut parsed_node_fn = ParsedNodeFn { vis: self.parsed.vis.clone(), attributes: NodeFnAttributes { - shader_node: Some(ShaderNodeType::GpuNode), + shader_node: Some(ShaderNodeType::ShaderNode), ..self.parsed.attributes.clone() }, fn_name: self.gpu_node_mod.clone(), @@ -309,10 +307,9 @@ impl PerPixelAdjustCodegen<'_> { let gpu_node_impl = crate::codegen::generate_node_code(&parsed_node_fn)?; // wrap node in `mod #gpu_node_mod` - let node_cfg = self.node_cfg; let gpu_node_mod = &self.gpu_node_mod; Ok(quote! { - #node_cfg + #[cfg(feature = #SHADER_NODES_FEATURE_GATE)] mod #gpu_node_mod { use super::*; use wgpu_executor::WgpuExecutor; From e3e0fbe75cc479eaa289481f9e9657d814bbf155 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Fri, 29 Aug 2025 12:43:05 +0200 Subject: [PATCH 15/17] shader-nodes feature: rename any `gpu_node` to `shader-node` --- .../src/shader_nodes/per_pixel_adjust.rs | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs index cd0d911694..f01c364223 100644 --- a/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs +++ b/node-graph/node-macro/src/shader_nodes/per_pixel_adjust.rs @@ -71,7 +71,7 @@ impl ShaderCodegen for PerPixelAdjust { let entry_point_name = quote!(#entry_point_mod::#entry_point_name_ident); let uniform_struct_ident = format_ident!("Uniform"); let uniform_struct = quote!(#entry_point_mod::#uniform_struct_ident); - let gpu_node_mod = format_ident!("{}_gpu", fn_name); + let shader_node_mod = format_ident!("{}_shader_node", fn_name); let codegen = PerPixelAdjustCodegen { parsed, @@ -82,7 +82,7 @@ impl ShaderCodegen for PerPixelAdjust { entry_point_name, uniform_struct_ident, uniform_struct, - gpu_node_mod, + shader_node_mod, }; Ok(ShaderTokens { @@ -101,7 +101,7 @@ pub struct PerPixelAdjustCodegen<'a> { entry_point_name: TokenStream, uniform_struct_ident: Ident, uniform_struct: TokenStream, - gpu_node_mod: Ident, + shader_node_mod: Ident, } impl PerPixelAdjustCodegen<'_> { @@ -286,9 +286,9 @@ impl PerPixelAdjustCodegen<'_> { shader_node: Some(ShaderNodeType::ShaderNode), ..self.parsed.attributes.clone() }, - fn_name: self.gpu_node_mod.clone(), - struct_name: format_ident!("{}", self.gpu_node_mod.to_string().to_case(Case::Pascal)), - mod_name: self.gpu_node_mod.clone(), + fn_name: self.shader_node_mod.clone(), + struct_name: format_ident!("{}", self.shader_node_mod.to_string().to_case(Case::Pascal)), + mod_name: self.shader_node_mod.clone(), fn_generics: vec![parse_quote!('a: 'n)], where_clause: None, input: Input { @@ -307,10 +307,10 @@ impl PerPixelAdjustCodegen<'_> { let gpu_node_impl = crate::codegen::generate_node_code(&parsed_node_fn)?; // wrap node in `mod #gpu_node_mod` - let gpu_node_mod = &self.gpu_node_mod; + let shader_node_mod = &self.shader_node_mod; Ok(quote! { #[cfg(feature = #SHADER_NODES_FEATURE_GATE)] - mod #gpu_node_mod { + mod #shader_node_mod { use super::*; use wgpu_executor::WgpuExecutor; From 7c7fea2e32dd4d7de99370f371b70723a393b2ea Mon Sep 17 00:00:00 2001 From: firestar99 Date: Fri, 5 Sep 2025 00:18:10 +0200 Subject: [PATCH 16/17] shaders-rt: fix wgpu label name --- .../src/shader_runtime/per_pixel_adjust_runtime.rs | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs index d958e0650d..928f35a0b0 100644 --- a/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs +++ b/node-graph/wgpu-executor/src/shader_runtime/per_pixel_adjust_runtime.rs @@ -160,7 +160,9 @@ impl PerPixelAdjustGraphicsPipeline { let device = &context.device; let name = self.name.as_str(); - let mut cmd = device.create_command_encoder(&wgpu::CommandEncoderDescriptor { label: Some("gpu_invert") }); + let mut cmd = device.create_command_encoder(&wgpu::CommandEncoderDescriptor { + label: Some(&format!("{name} cmd encoder")), + }); let out = textures .iter() .map(|instance| { From aba44eb676a46eb4c49b42d58df99857331c4963 Mon Sep 17 00:00:00 2001 From: firestar99 Date: Fri, 5 Sep 2025 00:23:23 +0200 Subject: [PATCH 17/17] shaders-rt: explain fullscreen_vertex coordinates with a drawing --- node-graph/graster-nodes/src/fullscreen_vertex.rs | 15 +++++++++++++++ 1 file changed, 15 insertions(+) diff --git a/node-graph/graster-nodes/src/fullscreen_vertex.rs b/node-graph/graster-nodes/src/fullscreen_vertex.rs index b8ef775b9f..533efb03b3 100644 --- a/node-graph/graster-nodes/src/fullscreen_vertex.rs +++ b/node-graph/graster-nodes/src/fullscreen_vertex.rs @@ -3,6 +3,21 @@ 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)]