shaders: add trait BufferStruct and derive macro

Author: Firestar99

Cargo.lock

@@ -84,7 +84,7 @@ graphite-proc-macros = { path = "proc-macros" }
 
 # Workspace dependencies
 rustc-hash = "2.0"
-bytemuck = { version = "1.13", features = ["derive"] }
+bytemuck = { version = "1.13", features = ["derive", "min_const_generics"] }
 serde = { version = "1.0", features = ["derive", "rc"] }
 serde_json = "1.0"
 serde-wasm-bindgen = "0.6"
@@ -154,7 +154,7 @@ parley = "0.5.0"
 skrifa = "0.32.0"
 pretty_assertions = "1.4.1"
 fern = { version = "0.7", features = ["colored"] }
-num_enum = "0.7"
+num_enum = { version = "0.7", default-features = false }
 num-derive = "0.4"
 num-traits = { version = "0.2", default-features = false, features = ["libm"] }
 specta = { version = "2.0.0-rc.22", features = [
@@ -192,18 +192,12 @@ tracing = "0.1.41"
 rfd = "0.15.4"
 open = "5.3.2"
 poly-cool = "0.2.0"
-spirv-std = { git = "https://github.com/rust-gpu/rust-gpu", rev = "c12f216121820580731440ee79ebc7403d6ea04f" }
+spirv-std = { git = "https://github.com/rust-gpu/rust-gpu", rev = "c12f216121820580731440ee79ebc7403d6ea04f", features = ["bytemuck"] }
 cargo-gpu = { git = "https://github.com/rust-gpu/cargo-gpu", rev = "f969528e87baa17a7d48eecf4a6fcfdcaaf30566" }
 
 [workspace.lints.rust]
 unexpected_cfgs = { level = "allow", check-cfg = ['cfg(target_arch, values("spirv"))'] }
 
-[workspace.lints.rust]
-unexpected_cfgs = { level = "allow", check-cfg = ['cfg(target_arch, values("spirv"))'] }
-
-[workspace.lints.rust]
-unexpected_cfgs = { level = "allow", check-cfg = ['cfg(target_arch, values("spirv"))'] }
-
 [profile.dev]
 opt-level = 1
 

Cargo.toml

@@ -22,10 +22,14 @@ std = [
 	"glam/serde",
 	"half/std",
 	"half/serde",
-	"num-traits/std"
+	"num-traits/std",
+	"num_enum/std",
 ]
 
 [dependencies]
+# Local dependencies
+node-macro = { workspace = true }
+
 # Local std dependencies
 dyn-any = { workspace = true, optional = true }
 
@@ -35,6 +39,7 @@ glam = { workspace = true }
 half = { workspace = true, default-features = false }
 num-derive = { workspace = true }
 num-traits = { workspace = true }
+num_enum = { workspace = true }
 spirv-std = { workspace = true }
 
 # Workspace std dependencies

node-graph/gcore-shaders/Cargo.toml

@@ -1,9 +1,11 @@
 use core::fmt::Display;
 use core::hash::{Hash, Hasher};
+use node_macro::BufferStruct;
+use num_enum::{FromPrimitive, IntoPrimitive};
 #[cfg(not(feature = "std"))]
 use num_traits::float::Float;
 
-#[derive(Debug, Clone, Copy, PartialEq)]
+#[derive(Debug, Clone, Copy, PartialEq, BufferStruct)]
 #[cfg_attr(feature = "std", derive(dyn_any::DynAny, specta::Type, serde::Serialize, serde::Deserialize))]
 #[cfg_attr(feature = "std", serde(default))]
 pub struct AlphaBlending {
@@ -66,7 +68,7 @@ impl AlphaBlending {
 }
 
 #[repr(i32)]
-#[derive(Debug, Default, Clone, Copy, Eq, PartialEq, Hash, bytemuck::NoUninit)]
+#[derive(Debug, Default, Clone, Copy, Eq, PartialEq, Hash, BufferStruct, FromPrimitive, IntoPrimitive)]
 #[cfg_attr(feature = "std", derive(dyn_any::DynAny, specta::Type, serde::Serialize, serde::Deserialize))]
 pub enum BlendMode {
 	// Basic group

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

@@ -5,6 +5,7 @@ use core::fmt::Debug;
 use core::hash::Hash;
 use glam::Vec4;
 use half::f16;
+use node_macro::BufferStruct;
 #[cfg(not(feature = "std"))]
 use num_traits::Euclid;
 #[cfg(not(feature = "std"))]
@@ -215,7 +216,7 @@ impl Pixel for Luma {}
 /// The other components (RGB) are stored as `f32` that range from `0.0` up to `f32::MAX`,
 /// the values encode the brightness of each channel proportional to the light intensity in cd/m² (nits) in HDR, and `0.0` (black) to `1.0` (white) in SDR color.
 #[repr(C)]
-#[derive(Debug, Default, Clone, Copy, PartialEq, Pod, Zeroable)]
+#[derive(Debug, Default, Clone, Copy, PartialEq, Pod, Zeroable, BufferStruct)]
 #[cfg_attr(feature = "std", derive(dyn_any::DynAny, specta::Type, serde::Serialize, serde::Deserialize))]
 pub struct Color {
 	red: f32,

node-graph/gcore-shaders/src/color/color_types.rs

@@ -0,0 +1,114 @@
+use crate::shaders::buffer_struct::BufferStruct;
+
+macro_rules! glam_array {
+	($t:ty, $a:ty) => {
+		unsafe impl BufferStruct for $t {
+			type Buffer = $a;
+
+			#[inline]
+			fn write(from: Self) -> Self::Buffer {
+				<$t>::to_array(&from)
+			}
+
+			#[inline]
+			fn read(from: Self::Buffer) -> Self {
+				<$t>::from_array(from)
+			}
+		}
+	};
+}
+
+macro_rules! glam_cols_array {
+	($t:ty, $a:ty) => {
+		unsafe impl BufferStruct for $t {
+			type Buffer = $a;
+
+			#[inline]
+			fn write(from: Self) -> Self::Buffer {
+				<$t>::to_cols_array(&from)
+			}
+
+			#[inline]
+			fn read(from: Self::Buffer) -> Self {
+				<$t>::from_cols_array(&from)
+			}
+		}
+	};
+}
+
+glam_array!(glam::Vec2, [f32; 2]);
+glam_array!(glam::Vec3, [f32; 3]);
+// glam_array!(Vec3A, [f32; 4]);
+glam_array!(glam::Vec4, [f32; 4]);
+glam_array!(glam::Quat, [f32; 4]);
+glam_cols_array!(glam::Mat2, [f32; 4]);
+glam_cols_array!(glam::Mat3, [f32; 9]);
+// glam_cols_array!(Mat3A, [f32; 4]);
+glam_cols_array!(glam::Mat4, [f32; 16]);
+glam_cols_array!(glam::Affine2, [f32; 6]);
+glam_cols_array!(glam::Affine3A, [f32; 12]);
+
+glam_array!(glam::DVec2, [f64; 2]);
+glam_array!(glam::DVec3, [f64; 3]);
+glam_array!(glam::DVec4, [f64; 4]);
+glam_array!(glam::DQuat, [f64; 4]);
+glam_cols_array!(glam::DMat2, [f64; 4]);
+glam_cols_array!(glam::DMat3, [f64; 9]);
+glam_cols_array!(glam::DMat4, [f64; 16]);
+glam_cols_array!(glam::DAffine2, [f64; 6]);
+glam_cols_array!(glam::DAffine3, [f64; 12]);
+
+glam_array!(glam::I16Vec2, [i16; 2]);
+glam_array!(glam::I16Vec3, [i16; 3]);
+glam_array!(glam::I16Vec4, [i16; 4]);
+
+glam_array!(glam::U16Vec2, [u16; 2]);
+glam_array!(glam::U16Vec3, [u16; 3]);
+glam_array!(glam::U16Vec4, [u16; 4]);
+
+glam_array!(glam::IVec2, [i32; 2]);
+glam_array!(glam::IVec3, [i32; 3]);
+glam_array!(glam::IVec4, [i32; 4]);
+
+glam_array!(glam::UVec2, [u32; 2]);
+glam_array!(glam::UVec3, [u32; 3]);
+glam_array!(glam::UVec4, [u32; 4]);
+
+glam_array!(glam::I64Vec2, [i64; 2]);
+glam_array!(glam::I64Vec3, [i64; 3]);
+glam_array!(glam::I64Vec4, [i64; 4]);
+
+glam_array!(glam::U64Vec2, [u64; 2]);
+glam_array!(glam::U64Vec3, [u64; 3]);
+glam_array!(glam::U64Vec4, [u64; 4]);
+
+unsafe impl BufferStruct for glam::Vec3A {
+	type Buffer = [f32; 4];
+
+	#[inline]
+	fn write(from: Self) -> Self::Buffer {
+		glam::Vec4::to_array(&from.extend(0.))
+	}
+
+	#[inline]
+	fn read(from: Self::Buffer) -> Self {
+		glam::Vec3A::from_vec4(glam::Vec4::from_array(from))
+	}
+}
+
+/// do NOT use slices, otherwise spirv will fail to compile
+unsafe impl BufferStruct for glam::Mat3A {
+	type Buffer = [f32; 12];
+
+	#[inline]
+	fn write(from: Self) -> Self::Buffer {
+		let a = from.to_cols_array();
+		[a[0], a[1], a[2], a[3], a[4], a[5], a[6], a[7], a[8], 0., 0., 0.]
+	}
+
+	#[inline]
+	fn read(from: Self::Buffer) -> Self {
+		let a = from;
+		glam::Mat3A::from_cols_array(&[a[0], a[1], a[2], a[3], a[4], a[5], a[6], a[7], a[8]])
+	}
+}

node-graph/gcore-shaders/src/shaders/buffer_struct/glam.rs

@@ -0,0 +1,63 @@
+//! I (@firestar99) copied this entire mod from one of my projects, as I haven't uploaded that lib to crates. Hopefully
+//! rust-gpu improves and this entire thing becomes unnecessary in the future.
+//!
+//! https://github.com/Firestar99/nanite-at-home/tree/008dac8df656959c71efeddd2d3ddabcb801771c/rust-gpu-bindless/crates/buffer-content
+
+use bytemuck::Pod;
+
+mod glam;
+mod primitive;
+
+/// A BufferStruct is a "parallel representation" of the original struct with some fundamental types remapped. This
+/// struct hierarchy represents how data is stored in GPU Buffers, where all types must be [`Pod`] to allow
+/// transmuting them to `&[u8]` with [`bytemuck`].
+///
+/// Notable type remappings (original: buffer):
+/// * bool: u32 of 0 or 1
+/// * any repr(u32) enum: u32 with remapping via [`num_enum`]
+///
+/// By adding `#[derive(ShaderStruct)]` to your struct (or enum), a parallel `{name}Buffer` struct is created with all
+/// the members of the original struct, but with their types using the associated remapped types as specified by this
+/// trait.
+///
+/// # Origin
+/// I (@firestar99) copied this entire mod from my [Nanite-at-home] project, specifically the [buffer-content] crate
+/// and the [buffer_struct] proc macro. The variant here has quite some modifications, to both cleaned up some of the
+/// mistakes my implementation has and to customize it a bit for graphite.
+///
+/// Hopefully rust-gpu improves to the point where this remapping becomes unnecessary.
+///
+/// [Nanite-at-home]: https://github.com/Firestar99/nanite-at-home
+/// [buffer-content]: https://github.com/Firestar99/nanite-at-home/tree/008dac8df656959c71efeddd2d3ddabcb801771c/rust-gpu-bindless/crates/buffer-content
+/// [buffer_struct]: https://github.com/Firestar99/nanite-at-home/blob/008dac8df656959c71efeddd2d3ddabcb801771c/rust-gpu-bindless/crates/macros/src/buffer_struct.rs
+///
+/// # Safety
+/// The associated type Transfer must be the same on all targets. Writing followed by reading back a value must result
+/// in the same value.
+pub unsafe trait BufferStruct: Copy + Send + Sync + 'static {
+	type Buffer: Pod + Send + Sync;
+
+	fn write(from: Self) -> Self::Buffer;
+
+	fn read(from: Self::Buffer) -> Self;
+}
+
+/// Trait marking all [`BufferStruct`] whose read and write methods are identity. While [`BufferStruct`] only
+/// requires `t == read(write(t))`, this trait additionally requires `t == read(t) == write(t)`. As this removes the
+/// conversion requirement for writing to or reading from a buffer, one can acquire slices from buffers created of these
+/// types.
+///
+/// Implementing this type is completely safe due to the [`Pod`] requirement.
+pub trait BufferStructIdentity: Pod + Send + Sync {}
+
+unsafe impl<T: BufferStructIdentity> BufferStruct for T {
+	type Buffer = Self;
+
+	fn write(from: Self) -> Self::Buffer {
+		from
+	}
+
+	fn read(from: Self::Buffer) -> Self {
+		from
+	}
+}