Update graphene-cli and fix no-std compilation for graphene-core (#1428)

* Initialize wgpu executor from graphene cli

* Make `graphene-core` `no-std` complient again

* Implemnt image similarity calculation

* Add nan checks

* Make image comparison optional

* Feature gate quantization to reduce the number of warnings

node-graph/gstd/Cargo.toml

@@ -21,10 +21,14 @@ wgpu = ["gpu", "wgpu-executor"]
 quantization = ["autoquant"]
 wasm = ["wasm-bindgen", "web-sys", "js-sys"]
 imaginate = ["image/png", "base64", "js-sys", "web-sys", "wasm-bindgen-futures"]
+image-compare = ["dep:image-compare"]
 wayland = []
 
 [dependencies]
-rand = { version = "0.8.5", features = ["alloc", "small_rng"], default-features = false}
+rand = { version = "0.8.5", features = [
+	"alloc",
+	"small_rng",
+], default-features = false }
 rand_chacha = { version = "0.3.1", default-features = false }
 autoquant = { git = "https://github.com/truedoctor/autoquant", optional = true, features = [
 	"fitting",
@@ -44,7 +48,10 @@ gpu-compiler-bin-wrapper = { path = "../gpu-compiler/gpu-compiler-bin-wrapper",
 compilation-client = { path = "../compilation-client", optional = true }
 bytemuck = { version = "1.13" }
 tempfile = "3"
-image = { version = "0.24", default-features = false, features = ["png", "jpeg"] }
+image = { version = "0.24", default-features = false, features = [
+	"png",
+	"jpeg",
+] }
 base64 = { version = "0.21", optional = true }
 dyn-clone = "1.0"
 
@@ -64,6 +71,7 @@ wasm-bindgen-futures = { version = "0.4.36", optional = true }
 winit = "0.28.6"
 url = "2.4.0"
 tokio = { version = "1.29.0", optional = true, features = ["fs", "io-std"] }
+image-compare = { version = "0.3.0", optional = true }
 
 [dependencies.serde]
 version = "1.0"

node-graph/gstd/src/gpu_nodes.rs

@@ -10,6 +10,9 @@ use graphene_core::raster::*;
 use graphene_core::*;
 use wgpu_executor::WgpuExecutor;
 
+#[cfg(feature = "quantization")]
+use graphene_core::quantization::PackedPixel;
+
 use std::cell::RefCell;
 use std::collections::HashMap;
 use std::sync::Arc;
@@ -74,6 +77,11 @@ async fn map_gpu<'a: 'input>(image: ImageFrame<Color>, node: DocumentNode, edito
 	let quantization = QuantizationChannels::default();
 	log::debug!("quantization: {:?}", quantization);
 
+	#[cfg(feature = "image-compare")]
+	let img: image::DynamicImage = image::Rgba32FImage::from_raw(image.image.width, image.image.height, bytemuck::cast_vec(image.image.data.clone()))
+		.unwrap()
+		.into();
+
 	#[cfg(feature = "quantization")]
 	let image = ImageFrame {
 		image: Image {
@@ -83,6 +91,7 @@ async fn map_gpu<'a: 'input>(image: ImageFrame<Color>, node: DocumentNode, edito
 		},
 		transform: image.transform,
 	};
+
 	// TODO: The cache should be based on the network topology not the node name
 	let compute_pass_descriptor = if self.cache.borrow().contains_key(&node.name) {
 		self.cache.borrow().get(&node.name).unwrap().clone()
@@ -114,6 +123,14 @@ async fn map_gpu<'a: 'input>(image: ImageFrame<Color>, node: DocumentNode, edito
 	#[cfg(not(feature = "quantization"))]
 	let colors = bytemuck::pod_collect_to_vec::<u8, Color>(result.as_slice());
 	log::debug!("first color: {:?}", colors[0]);
+
+	#[cfg(feature = "image-compare")]
+	let img2: image::DynamicImage = image::Rgba32FImage::from_raw(image.image.width, image.image.height, bytemuck::cast_vec(colors.clone())).unwrap().into();
+	#[cfg(feature = "image-compare")]
+	let score = image_compare::rgb_hybrid_compare(&img.into_rgb8(), &img2.into_rgb8()).unwrap();
+	#[cfg(feature = "image-compare")]
+	log::debug!("score: {:?}", score.score);
+
 	ImageFrame {
 		image: Image {
 			data: colors,
@@ -279,6 +296,10 @@ async fn create_compute_pass_descriptor<T: Clone + Pixel + StaticTypeSized>(
 	return frame;*/
 	log::debug!("creating buffer");
 	let width_uniform = executor.create_uniform_buffer(image.image.width).unwrap();
+	#[cfg(not(feature = "quantization"))]
+	core::hint::black_box(quantization);
+
+	#[cfg(feature = "quantization")]
 	let quantization_uniform = executor.create_uniform_buffer(quantization).unwrap();
 	let storage_buffer = executor
 		.create_storage_buffer(
@@ -292,7 +313,8 @@ async fn create_compute_pass_descriptor<T: Clone + Pixel + StaticTypeSized>(
 		)
 		.unwrap();
 	let width_uniform = Arc::new(width_uniform);
-	let _quantization_uniform = Arc::new(quantization_uniform);
+	#[cfg(feature = "quantization")]
+	let quantization_uniform = Arc::new(quantization_uniform);
 	let storage_buffer = Arc::new(storage_buffer);
 	let output_buffer = executor.create_output_buffer(len, concrete!(Color), false).unwrap();
 	let output_buffer = Arc::new(output_buffer);

node-graph/gstd/src/quantization.rs

@@ -48,7 +48,7 @@ fn generate_quantization(data: Vec<f64>, samples: usize) -> [Quantization; 4] {
 	let merged: ErrorFunction<30> = autoquant::packing::merge_error_functions(&merged, &blue_error);
 	let merged: ErrorFunction<40> = autoquant::packing::merge_error_functions(&merged, &alpha_error);
 
-	let bin_size = 32;
+	let bin_size = 8;
 	let mut distributions = [red, green, blue, alpha].into_iter();
 
 	let bits = &merged.bits[bin_size];

node-graph/gstd/src/wasm_application_io.rs

@@ -56,7 +56,7 @@ impl WasmApplicationIo {
 			None
 		};
 		#[cfg(all(feature = "wgpu", not(target_arch = "wasm32")))]
-		let executor = None;
+		let executor = WgpuExecutor::new().await;
 		let mut io = Self {
 			#[cfg(target_arch = "wasm32")]
 			ids: RefCell::new(0),