Multithreaded render

Cargo.lock

@@ -681,12 +681,10 @@ dependencies = [
  "foldhash",
  "glam",
  "log",
+ "num_cpus",
  "obj",
  "rand 0.9.1",
- "rayon",
  "rspirv",
- "serde",
- "serde_json",
  "simplelog",
  "utf-8",
  "vulkano",
@@ -1332,6 +1330,16 @@ dependencies = [
  "autocfg",
 ]
 
+[[package]]
+name = "num_cpus"
+version = "1.17.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "91df4bbde75afed763b708b7eee1e8e7651e02d97f6d5dd763e89367e957b23b"
+dependencies = [
+ "hermit-abi",
+ "libc",
+]
+
 [[package]]
 name = "num_enum"
 version = "0.7.4"

Cargo.toml

@@ -24,13 +24,11 @@ egui = "0.31.1"
 egui_winit_vulkano = "0.28.0"
 egui_plot = "0.32.1"
 
-serde = { version = "1", features = ["derive"] }
+#serde = { version = "1", features = ["derive"] }
-serde_json = "1"
+#serde_json = "1"
 
 utf-8 = "0.7"
 
-rayon = "1.7"
-
 rand = "0.9.1"
 foldhash = "*"
 
@@ -39,6 +37,8 @@ simplelog = "0.12"
 
 rspirv = "0.12"
 
+num_cpus = "1"
+
 # using latest gits
 [patch.crates-io]
 #vulkano = { git = "https://github.com/vulkano-rs/vulkano" }

src/gui.rs

@@ -1,12 +1,10 @@
+use std::sync::{Arc, RwLock};
+
 use egui::{Color32, Frame, Id};
 use egui_plot::{Line, Plot, PlotPoints};
 use egui_winit_vulkano::Gui;
-use glam::Vec3;
 
-use crate::{
+use crate::objects::{CSG, Light, Mesh};
-    create_csg,
-    objects::{CSG, Light, Mesh},
-};
 
 fn sized_text(ui: &mut egui::Ui, text: impl Into<String>, size: f32) {
     ui.label(egui::RichText::new(text).size(size));
@@ -29,7 +27,7 @@ pub(crate) struct GState {
 
     pub(crate) meshes: Vec<Mesh>,
     pub(crate) lights: Vec<Light>,
-    pub(crate) csg:    Vec<CSG>,
+    pub(crate) csg:    Vec<Arc<RwLock<CSG>>>,
 
     pub(crate) fps: [f64; 128],
 
@@ -53,7 +51,7 @@ impl Default for GState {
     }
 }
 
-pub(crate) fn gui_up(gui: &mut Gui, state: &mut GState) {
+pub(crate) fn gui_up(gui: &mut Gui, state: &mut GState, new_csg_needed: &mut bool) {
     gui.immediate_ui(|gui| {
         let ctx = gui.context();
         egui::SidePanel::left(Id::new("main_left"))
@@ -118,6 +116,7 @@ pub(crate) fn gui_up(gui: &mut Gui, state: &mut GState) {
                         ui.heading("Implicit Surfaces");
                         let mut csgdel = vec![];
                         for csg in &mut state.csg {
+                            let mut csg = csg.write().unwrap();
                             ui.label(csg.name.clone());
                             csgdel.push(ui.small_button("remove csg").clicked());
                             ui.add(
@@ -153,13 +152,7 @@ pub(crate) fn gui_up(gui: &mut Gui, state: &mut GState) {
                             state.csg.remove(i);
                         }
                         if state.csg.len() < 32 && ui.small_button("add csg").clicked() {
-                            state.csg.push(CSG {
+                            *new_csg_needed = true;
-                                name:  "generated".to_string(),
-                                parts: create_csg(),
-                                pos:   Vec3::ZERO,
-                                rot:   Vec3::ZERO,
-                                scale: Vec3::ONE,
-                            });
                         }
                         ui.heading("Lights");
                         let mut lightdel = vec![];

src/objects.rs

@@ -8,7 +8,7 @@ use vulkano::{
     memory::allocator::{
         AllocationCreateInfo, MemoryAllocatePreference, MemoryTypeFilter, StandardMemoryAllocator,
     },
-    pipeline::graphics::vertex_input::Vertex,
+    pipeline::{GraphicsPipeline, graphics::vertex_input::Vertex},
 };
 
 use crate::ssa::SSATape;
@@ -41,6 +41,8 @@ pub(crate) struct CSG {
     pub(crate) pos:               Vec3,
     pub(crate) rot:               Vec3,
     pub(crate) scale:             Vec3,
+    pub(crate) deferred_pipeline: Arc<GraphicsPipeline>,
+    pub(crate) normals_pipeline:  Arc<GraphicsPipeline>,
 }
 
 pub(crate) fn load_obj(

src/ssa.rs

@@ -1,3 +1,5 @@
+use std::sync::Arc;
+
 use egui::ahash::HashMapExt;
 use foldhash::HashMap;
 use rspirv::{dr::Module, spirv};

src/threads.rs

@@ -0,0 +1,467 @@
+use std::{
+    fs::File,
+    io::Write,
+    simd::{StdFloat, cmp::SimdPartialOrd},
+    sync::{Arc, RwLock, mpmc, mpsc},
+    time::Instant,
+};
+
+use foldhash::HashSet;
+use glam::{EulerRot, Mat4, Vec3};
+use log::info;
+use rspirv::{binary::Assemble, dr::Module};
+use vulkano::{
+    device::Device,
+    pipeline::{
+        DynamicState, GraphicsPipeline, PipelineCreateFlags, PipelineLayout,
+        PipelineShaderStageCreateInfo,
+        cache::PipelineCache,
+        graphics::{
+            GraphicsPipelineCreateInfo,
+            color_blend::{ColorBlendAttachmentState, ColorBlendState},
+            depth_stencil::{DepthState, DepthStencilState},
+            input_assembly::InputAssemblyState,
+            multisample::MultisampleState,
+            rasterization::{CullMode, FrontFace, PolygonMode, RasterizationState},
+            vertex_input::{Vertex, VertexDefinition, VertexInputState},
+        },
+        layout::PipelineDescriptorSetLayoutCreateInfo,
+    },
+    render_pass::{RenderPass, Subpass},
+    shader::{ShaderModule, ShaderModuleCreateInfo},
+};
+
+use crate::{
+    DUMP_SPV_TO_FILE, IVertex, ShaderModules, get_spec_constants,
+    gui::PreviousDebug,
+    implicit_vs::PushConstantData,
+    interpreter::{self, IntervalInterpreter, PointInterpreter, VALUE_0},
+    objects::CSG,
+    ssa::{SSAInput, SSAOpcode, SSAOpcodeSized, SSATape},
+    types::Interval,
+};
+
+#[derive(Debug)]
+pub enum WorkItem {
+    CreateCSG(
+        Arc<Device>,
+        Module,
+        Module,
+        Arc<RenderPass>,
+        Arc<PipelineCache>,
+        ShaderModules,
+        PreviousDebug,
+        isize,
+    ),
+    GetPushConstants(Arc<RwLock<CSG>>, f32, usize),
+}
+
+#[derive(Debug)]
+pub enum WorkComplete {
+    CreateCSG(Arc<RwLock<CSG>>, isize),
+    GetPushConstants(PushConstantData, usize),
+}
+
+pub fn thread_loop(recv: mpmc::Receiver<WorkItem>, send: mpsc::SyncSender<WorkComplete>) {
+    for work in recv.into_iter() {
+        match work {
+            WorkItem::CreateCSG(
+                device,
+                trace_module,
+                normals_module,
+                render_pass,
+                cache,
+                modules,
+                debug,
+                index,
+            ) => {
+                let csg_start = Instant::now();
+                let parts = create_csg();
+
+                let trace_shader_module =
+                    sdf_specialize_module(device.clone(), &parts, trace_module, &format!("trace"));
+                let normals_shader_module = sdf_specialize_module(
+                    device.clone(),
+                    &parts,
+                    normals_module,
+                    &format!("normals"),
+                );
+
+                let (deferred_pipeline, normals_pipeline) = deferred_pipelines_recompile(
+                    device,
+                    render_pass,
+                    cache,
+                    modules,
+                    trace_shader_module,
+                    normals_shader_module,
+                    debug,
+                );
+
+                let csg = Arc::new(RwLock::new(CSG {
+                    name: "example".to_string(),
+                    parts,
+                    pos: Vec3::ZERO,
+                    rot: Vec3::ZERO,
+                    scale: Vec3::ONE,
+                    deferred_pipeline,
+                    normals_pipeline,
+                }));
+
+                send.send(WorkComplete::CreateCSG(csg, index)).unwrap();
+
+                let csg_end = Instant::now();
+                info!(
+                    "CSG compile took {} milliseconds",
+                    (csg_end - csg_start).as_secs_f64() * 1000.0
+                );
+            },
+            WorkItem::GetPushConstants(csg, time, index) => {
+                let mut push_constants = PushConstantData {
+                    world:         Mat4::IDENTITY.to_cols_array_2d(),
+                    dimensions:    Vec3::ONE.to_array().into(),
+                    lowest_corner: Vec3::ZERO.into(),
+                    enable:        [0; 16],
+                    material:      0,
+                };
+
+                let csg = csg.read().unwrap();
+
+                push_constants.world = (Mat4::from_translation(csg.pos * 0.01)
+                    * Mat4::from_euler(
+                        EulerRot::XYZ,
+                        csg.rot.x.to_radians(),
+                        csg.rot.y.to_radians(),
+                        csg.rot.z.to_radians(),
+                    )
+                    * Mat4::from_scale(csg.scale * 2.0))
+                .to_cols_array_2d();
+
+                interval_check(&csg, &mut push_constants, time);
+
+                push_constants.material = index as _;
+
+                send.send(WorkComplete::GetPushConstants(push_constants, index))
+                    .unwrap();
+            },
+        }
+    }
+}
+
+fn deferred_pipelines_recompile(
+    device: Arc<Device>,
+    render_pass: Arc<RenderPass>,
+    cache: Arc<PipelineCache>,
+    shader_modules: ShaderModules,
+    trace_shader_module: Arc<ShaderModule>,
+    normals_shader_module: Arc<ShaderModule>,
+    debug: PreviousDebug,
+) -> (Arc<GraphicsPipeline>, Arc<GraphicsPipeline>) {
+    let specs = get_spec_constants(&debug);
+
+    let dynamic_state = [DynamicState::Viewport].into_iter().collect::<HashSet<_>>();
+
+    let implicit_vs_entry = shader_modules
+        .implicit_vs
+        .specialize(specs.clone())
+        .unwrap()
+        .single_entry_point()
+        .unwrap();
+
+    let vertex_input_state = [IVertex::per_vertex()]
+        .definition(&implicit_vs_entry)
+        .unwrap();
+
+    let implicit_vs_info = PipelineShaderStageCreateInfo::new(implicit_vs_entry);
+
+    let implicit_fs_info = PipelineShaderStageCreateInfo::new(
+        trace_shader_module
+            .specialize(specs.clone())
+            .unwrap()
+            .single_entry_point()
+            .unwrap(),
+    );
+
+    let layout = PipelineLayout::new(
+        device.clone(),
+        PipelineDescriptorSetLayoutCreateInfo::from_stages([&implicit_vs_info, &implicit_fs_info])
+            .into_pipeline_layout_create_info(device.clone())
+            .unwrap(),
+    )
+    .unwrap();
+
+    let stages = [implicit_vs_info.clone(), implicit_fs_info]
+        .into_iter()
+        .collect();
+
+    let deferred_subpass = Subpass::from(render_pass.clone(), 0).unwrap();
+
+    let implicit_pipeline = GraphicsPipeline::new(
+        device.clone(),
+        Some(cache.clone()),
+        GraphicsPipelineCreateInfo {
+            flags: PipelineCreateFlags::DISABLE_OPTIMIZATION,
+            stages,
+            vertex_input_state: Some(vertex_input_state.clone()),
+            input_assembly_state: Some(InputAssemblyState::default()),
+            dynamic_state: dynamic_state.clone(),
+            viewport_state: Some(Default::default()),
+            rasterization_state: Some(RasterizationState {
+                front_face: FrontFace::Clockwise,
+                cull_mode: CullMode::Back,
+                polygon_mode: PolygonMode::Fill,
+                ..RasterizationState::default()
+            }),
+            depth_stencil_state: Some(DepthStencilState {
+                depth: Some(DepthState::simple()),
+                ..Default::default()
+            }),
+            multisample_state: Some(MultisampleState::default()),
+            color_blend_state: Some(ColorBlendState::with_attachment_states(
+                deferred_subpass.num_color_attachments(),
+                ColorBlendAttachmentState::default(),
+            )),
+            subpass: Some(deferred_subpass.into()),
+            ..GraphicsPipelineCreateInfo::layout(layout.clone())
+        },
+    )
+    .unwrap();
+
+    let fullscreen_vs_entry = shader_modules
+        .fullscreen_vs
+        .specialize(specs.clone())
+        .unwrap()
+        .single_entry_point()
+        .unwrap();
+
+    let vertex_input_state = VertexInputState::new();
+
+    let fullscreen_vs_info = PipelineShaderStageCreateInfo::new(fullscreen_vs_entry);
+
+    let normals_fs_info = PipelineShaderStageCreateInfo::new(
+        normals_shader_module
+            .specialize(specs.clone())
+            .unwrap()
+            .single_entry_point()
+            .unwrap(),
+    );
+
+    let layout = PipelineLayout::new(
+        device.clone(),
+        PipelineDescriptorSetLayoutCreateInfo::from_stages([&fullscreen_vs_info, &normals_fs_info])
+            .into_pipeline_layout_create_info(device.clone())
+            .unwrap(),
+    )
+    .unwrap();
+
+    let stages = [fullscreen_vs_info.clone(), normals_fs_info]
+        .into_iter()
+        .collect();
+    let normals_subpass = Subpass::from(render_pass.clone(), 1).unwrap();
+
+    let normals_pipeline = GraphicsPipeline::new(
+        device.clone(),
+        Some(cache.clone()),
+        GraphicsPipelineCreateInfo {
+            flags: PipelineCreateFlags::DISABLE_OPTIMIZATION,
+            stages,
+            vertex_input_state: Some(vertex_input_state.clone()),
+            input_assembly_state: Some(InputAssemblyState::default()),
+            dynamic_state: dynamic_state.clone(),
+            viewport_state: Some(Default::default()),
+            rasterization_state: Some(RasterizationState {
+                front_face: FrontFace::Clockwise,
+                cull_mode: CullMode::Back,
+                polygon_mode: PolygonMode::Fill,
+                ..RasterizationState::default()
+            }),
+            depth_stencil_state: None,
+            multisample_state: Some(MultisampleState::default()),
+            color_blend_state: Some(ColorBlendState::with_attachment_states(
+                normals_subpass.num_color_attachments(),
+                ColorBlendAttachmentState::default(),
+            )),
+            subpass: Some(normals_subpass.into()),
+            ..GraphicsPipelineCreateInfo::layout(layout.clone())
+        },
+    )
+    .unwrap();
+
+    (implicit_pipeline, normals_pipeline)
+}
+
+fn create_csg() -> SSATape {
+    let mut tape = SSATape::default();
+    let pos = tape.push_instruction(
+        SSAOpcodeSized {
+            opcode: SSAOpcode::SSAPosition,
+            size:   1,
+        },
+        vec![],
+    );
+    let offset = tape.push_instruction(
+        SSAOpcodeSized {
+            opcode: SSAOpcode::SSAAdd,
+            size:   3,
+        },
+        vec![
+            pos[0],
+            pos[1],
+            pos[2],
+            SSAInput::Constant(rand::random_range(0.0..3.0)),
+            SSAInput::Constant(rand::random_range(0.0..3.0)),
+            SSAInput::Constant(rand::random_range(0.0..3.0)),
+        ],
+    );
+    let sphere = tape.push_instruction(
+        SSAOpcodeSized {
+            opcode: SSAOpcode::SSASDFSphere,
+            size:   1,
+        },
+        vec![
+            offset[0],
+            offset[1],
+            offset[2],
+            SSAInput::Constant(rand::random_range(0.0..1.0)),
+        ],
+    );
+    let torus = tape.push_instruction(
+        SSAOpcodeSized {
+            opcode: SSAOpcode::SSASDFTorus,
+            size:   1,
+        },
+        vec![
+            pos[0],
+            pos[1],
+            pos[2],
+            SSAInput::Constant(rand::random_range(0.0..1.0)),
+            SSAInput::Constant(rand::random_range(0.0..1.0)),
+        ],
+    );
+    let min = tape.push_instruction(
+        SSAOpcodeSized {
+            opcode: SSAOpcode::SSAMin,
+            size:   1,
+        },
+        vec![sphere[0], torus[0]],
+    );
+    tape.push_instruction(
+        SSAOpcodeSized {
+            opcode: SSAOpcode::SSAReturn,
+            size:   1,
+        },
+        min,
+    );
+    tape
+}
+
+fn interval_check(csg: &CSG, push_constants: &mut PushConstantData, time: f32) {
+    const INTERPRET_INPUT_X: interpreter::Value =
+        interpreter::Value::from_array([10000.0, 0.0, 0.0, -10000.0, 0.0, 0.0, 0.0, 0.0]);
+    const INTERPRET_INPUT_Y: interpreter::Value =
+        interpreter::Value::from_array([0.0, 10000.0, 0.0, 0.0, -10000.0, 0.0, 0.0, 0.0]);
+    const INTERPRET_INPUT_Z: interpreter::Value =
+        interpreter::Value::from_array([0.0, 0.0, 10000.0, 0.0, 0.0, -10000.0, 0.0, 0.0]);
+    const INTERPRET_MUL: interpreter::Value =
+        interpreter::Value::from_array([-1.0, -1.0, -1.0, 1.0, 1.0, 1.0, 0.0, 0.0]);
+    const INTERPRET_ADD: interpreter::Value = interpreter::Value::from_array([
+        10000.0, 10000.0, 10000.0, -10000.0, -10000.0, -10000.0, 0.0, 0.0,
+    ]);
+
+    let mut interpreter = PointInterpreter::new(csg);
+    let interpreter_out = interpreter
+        .scene(
+            INTERPRET_INPUT_X,
+            INTERPRET_INPUT_Y,
+            INTERPRET_INPUT_Z,
+            interpreter::Value::splat(time),
+        )
+        .mul_add(INTERPRET_MUL, INTERPRET_ADD)
+        .to_array();
+
+    let highest_corner = Vec3::new(interpreter_out[0], interpreter_out[1], interpreter_out[2]);
+    let lowest_corner = Vec3::new(interpreter_out[3], interpreter_out[4], interpreter_out[5]);
+
+    push_constants.dimensions = (highest_corner - lowest_corner).to_array().into();
+    push_constants.lowest_corner = lowest_corner.into();
+
+    push_constants.enable = [0; 16];
+
+    for x in 0..8 {
+        for y in 0..8 {
+            let interval_input_z: Interval = Interval::new_unchecked(
+                interpreter::Value::from_array([
+                    (0.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (1.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (2.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (3.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (4.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (5.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (6.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (7.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                ]),
+                interpreter::Value::from_array([
+                    (1.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (2.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (3.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (4.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (5.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (6.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (7.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                    (8.0 / 8.0) * push_constants.dimensions[2] + push_constants.lowest_corner[2],
+                ]),
+            );
+
+            let mut interpreter = IntervalInterpreter::new(csg);
+            let output = interpreter.scene(
+                Interval::const_splat2(
+                    (x as f32 / 8.0) * push_constants.dimensions[0]
+                        + push_constants.lowest_corner[0],
+                    ((x + 1) as f32 / 8.0) * push_constants.dimensions[0]
+                        + push_constants.lowest_corner[0],
+                ),
+                Interval::const_splat2(
+                    (y as f32 / 8.0) * push_constants.dimensions[1]
+                        + push_constants.lowest_corner[1],
+                    ((y + 1) as f32 / 8.0) * push_constants.dimensions[1]
+                        + push_constants.lowest_corner[1],
+                ),
+                interval_input_z,
+                Interval::const_splat(time),
+            );
+
+            let output = ((output.lower().simd_le(VALUE_0) & output.upper().simd_ge(VALUE_0))
+                .to_bitmask() as u32)
+                << ((y & 0b11) * 8);
+            push_constants.enable[(x << 1) | (y >> 2)] |= output;
+        }
+    }
+}
+
+fn sdf_specialize_module(
+    device: Arc<Device>,
+    parts: &SSATape,
+    module: Module,
+    name: &str,
+) -> Arc<ShaderModule> {
+    let new_module = parts.compile_to_spirv(Some(module));
+
+    let assembled_module = new_module.assemble();
+    let module_as_bytes = assembled_module
+        .iter()
+        .cloned()
+        .map(u32::to_le_bytes)
+        .flatten()
+        .collect::<Vec<_>>();
+
+    if DUMP_SPV_TO_FILE {
+        File::create(format!("{name}.out.spv"))
+            .unwrap()
+            .write(&module_as_bytes)
+            .unwrap();
+    }
+
+    unsafe {
+        ::vulkano::shader::ShaderModule::new(device, ShaderModuleCreateInfo::new(&assembled_module))
+            .unwrap()
+    }
+}