bherbruckandClaude Opus 5.5
480fd8d517
* GPU tessellation: a pipeline the device cannot build falls back to CPU meshing instead of crashing (#1098) Build the tessellation compute pipeline inside error scopes; TessGpu::new returns an error, and the viewport logs it and meshes bodies on the CPU. A Windows CI job builds the pipeline with D3D12's FXC on the WARP adapter. Co-Authored-By: Claude Opus 5.5 <[email protected]> * Tessellation shader: basis functions in one-dimensional local arrays, so D3D12's FXC compiles it (#1098) FXC has to unroll loops that write an array inside a struct (Basis.n0/n1) or a 2-D array (ndu) by a counter, and cannot with run-time trip counts (X3511). A2.2 basis functions in a local 1-D array returned whole; derivatives from the degree p-1 functions; the pole step offsets computed rather than indexed. Co-Authored-By: Claude Opus 5.5 <[email protected]> --------- Co-authored-by: Claude Opus 5.5 <[email protected]>
1201 lines
55 KiB
Rust
1201 lines
55 KiB
Rust
//! GPU viewport: shaded triangles and screen-space wide lines with a depth buffer, drawn in an
|
||
//! `egui_wgpu` paint callback. Geometry is uploaded when the scene changes; each frame only the
|
||
//! camera uniform changes. Where the device runs compute shaders, opaque bodies are drawn
|
||
//! straight from their display plans: the plan is evaluated on the GPU into the vertex buffer
|
||
//! the pass draws (see [`PlannedBody`]), with nothing built or read back on the CPU.
|
||
|
||
use std::sync::{Arc, Mutex};
|
||
|
||
use egui_wgpu::wgpu;
|
||
use egui_wgpu::wgpu::util::DeviceExt;
|
||
|
||
/// Triangle vertex: position (3 × f32), normal (3 × f32), sRGB colour (4 × u8).
|
||
pub const TRI_SIZE: usize = 28;
|
||
/// Line instance: a, b (3 × f32 each), colour (4 × u8), width in pixels (f32).
|
||
pub const LINE_SIZE: usize = 32;
|
||
const UNIFORM_SIZE: u64 = 208;
|
||
|
||
/// A body drawn from its display plan, evaluated on the GPU: the plan, and a colour per face
|
||
/// (the plan's face numbering; faces past the end take the last colour).
|
||
#[derive(Clone)]
|
||
pub struct PlannedBody {
|
||
pub plan: Arc<solidcraft_engine::geom::tess::TessPlan>,
|
||
pub face_colors: Vec<[u8; 4]>,
|
||
}
|
||
|
||
/// CPU-side geometry waiting to be uploaded. The model scene changes with the design; the
|
||
/// highlight scene (hover, selection, the origin widget) changes often and is small.
|
||
#[derive(Default, Clone)]
|
||
pub struct GpuScene {
|
||
/// Opaque triangles. A zero normal draws the colour flat (unlit).
|
||
pub tris: Vec<u8>,
|
||
/// Translucent triangles (alpha from the colour), drawn after the opaque ones without
|
||
/// writing depth.
|
||
pub trans: Vec<u8>,
|
||
/// See-through bodies and faces (appearance opacity below 1), closed and outward-wound:
|
||
/// their far sides are drawn first, then their near sides, so they read like glass.
|
||
pub glass: Vec<u8>,
|
||
/// Depth-tested lines.
|
||
pub lines: Vec<u8>,
|
||
/// Lines drawn only where the model hides them (hidden edges).
|
||
pub hidden: Vec<u8>,
|
||
/// Lines drawn over everything (active sketch, highlights).
|
||
pub overlay: Vec<u8>,
|
||
/// Translucent triangles pushed slightly back in depth, so they only show where nothing
|
||
/// coincides with them (removed material in a preview).
|
||
pub ghost: Vec<u8>,
|
||
/// Translucent triangles drawn through everything (a cut's tool body).
|
||
pub xray: Vec<u8>,
|
||
/// Lines drawn first, under everything (the ground grid: the model hides it, it never
|
||
/// hides the model).
|
||
pub under: Vec<u8>,
|
||
/// Opaque bodies drawn from their plans on the GPU, like `tris`.
|
||
pub planned: Vec<PlannedBody>,
|
||
}
|
||
|
||
impl GpuScene {
|
||
pub fn tri(&mut self, p: [f32; 3], n: [f32; 3], c: [u8; 4]) {
|
||
for v in p.iter().chain(&n) {
|
||
self.tris.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.tris.extend_from_slice(&c);
|
||
}
|
||
pub fn trans_tri(&mut self, p: [f32; 3], n: [f32; 3], c: [u8; 4]) {
|
||
for v in p.iter().chain(&n) {
|
||
self.trans.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.trans.extend_from_slice(&c);
|
||
}
|
||
/// A line shown only where the model hides it.
|
||
pub fn hidden_line(&mut self, a: [f32; 3], b: [f32; 3], c: [u8; 4], width: f32) {
|
||
for v in a.iter().chain(&b) {
|
||
self.hidden.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.hidden.extend_from_slice(&c);
|
||
self.hidden.extend_from_slice(&width.to_le_bytes());
|
||
}
|
||
pub fn glass_tri(&mut self, p: [f32; 3], n: [f32; 3], c: [u8; 4]) {
|
||
for v in p.iter().chain(&n) {
|
||
self.glass.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.glass.extend_from_slice(&c);
|
||
}
|
||
pub fn ghost_tri(&mut self, p: [f32; 3], n: [f32; 3], c: [u8; 4]) {
|
||
for v in p.iter().chain(&n) {
|
||
self.ghost.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.ghost.extend_from_slice(&c);
|
||
}
|
||
pub fn xray_tri(&mut self, p: [f32; 3], n: [f32; 3], c: [u8; 4]) {
|
||
for v in p.iter().chain(&n) {
|
||
self.xray.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.xray.extend_from_slice(&c);
|
||
}
|
||
/// A line under everything (see [`GpuScene::under`]).
|
||
pub fn under_line(&mut self, a: [f32; 3], b: [f32; 3], c: [u8; 4], width: f32) {
|
||
for v in a.iter().chain(&b) {
|
||
self.under.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
self.under.extend_from_slice(&c);
|
||
self.under.extend_from_slice(&width.to_le_bytes());
|
||
}
|
||
pub fn line(&mut self, a: [f32; 3], b: [f32; 3], c: [u8; 4], width: f32, on_top: bool) {
|
||
let out = if on_top { &mut self.overlay } else { &mut self.lines };
|
||
for v in a.iter().chain(&b) {
|
||
out.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
out.extend_from_slice(&c);
|
||
out.extend_from_slice(&width.to_le_bytes());
|
||
}
|
||
}
|
||
|
||
pub type SceneSlot = Arc<Mutex<Option<GpuScene>>>;
|
||
|
||
/// A reference image (canvas or decal): a textured quad, depth-tested against the model.
|
||
#[derive(Clone)]
|
||
pub struct GpuImage {
|
||
pub id: u64,
|
||
/// Changes when the pixels change (the texture is re-uploaded).
|
||
pub version: u64,
|
||
pub size: [u32; 2],
|
||
/// Straight RGBA8, `size[0] * size[1] * 4` bytes.
|
||
pub rgba: Arc<Vec<u8>>,
|
||
/// Bottom-left, bottom-right, top-right, top-left of the image.
|
||
pub corners: [[f32; 3]; 4],
|
||
pub opacity: f32,
|
||
}
|
||
|
||
/// Image vertex: position (3 × f32), uv (2 × f32), opacity (f32).
|
||
const IMG_SIZE: usize = 24;
|
||
|
||
/// Per-frame parameters.
|
||
pub struct ViewportCallback {
|
||
pub key: u64,
|
||
pub slot: SceneSlot,
|
||
pub sketch_key: u64,
|
||
pub sketch_slot: SceneSlot,
|
||
/// The highlight scene and its key.
|
||
pub hl_key: u64,
|
||
pub hl_slot: SceneSlot,
|
||
/// The live preview scene and its key.
|
||
pub pv_key: u64,
|
||
pub pv_slot: SceneSlot,
|
||
/// Column-major view-projection.
|
||
pub view_proj: [[f32; 4]; 4],
|
||
/// Direction toward the eye (for lighting).
|
||
pub back: [f32; 3],
|
||
/// Viewport size in physical pixels.
|
||
pub size_px: [f32; 2],
|
||
/// Section cut: (normal, d); fragments with normal·p > d are hidden. A zero normal: none.
|
||
pub clip: [f32; 4],
|
||
/// Colour of the inside of cut bodies (straight sRGBA, 0..1).
|
||
pub cap: [f32; 4],
|
||
/// Canvases and decals.
|
||
pub images: Vec<GpuImage>,
|
||
/// Surface analysis: (mode: 0 none, 1 zebra, 2 draft, 3 curvature, 4 environment,
|
||
/// 5 accessibility; parameter; scale; size) and the pull / access direction.
|
||
pub analysis: [f32; 8],
|
||
/// Accessibility: the depth map seen from the access direction.
|
||
pub access: Option<AccessMap>,
|
||
/// Draw silhouette outlines around the model (visual styles with edges).
|
||
pub outline: bool,
|
||
/// Eye position and 1 for a perspective camera (0: orthographic), for the section cap's depth
|
||
/// and the depth bias.
|
||
pub eye: [f32; 4],
|
||
}
|
||
|
||
/// Heights of the model seen from a direction (an orthographic depth map): a point is reachable
|
||
/// from that direction when nothing lies above it.
|
||
#[derive(Clone)]
|
||
pub struct AccessMap {
|
||
pub version: u64,
|
||
pub size: u32,
|
||
/// Row-major heights along the direction (−∞ where nothing is).
|
||
pub depth: Arc<Vec<f32>>,
|
||
/// Map axes: pixel = (p·u + u.w, p·v + v.w) × scale.
|
||
pub u: [f32; 4],
|
||
pub v: [f32; 4],
|
||
pub scale: f32,
|
||
}
|
||
|
||
/// Vertices (or line instances) on the GPU, in as many buffers as the device's buffer size
|
||
/// limit needs: a large import's triangles can exceed one buffer (#883).
|
||
struct Batch {
|
||
/// Each buffer and its vertex (instance) count.
|
||
parts: Vec<(wgpu::Buffer, u32)>,
|
||
/// Bytes per buffer (all but the last are full).
|
||
chunk: usize,
|
||
}
|
||
|
||
/// A chunk of planned bodies on the GPU: their evaluated mesh and a colour per vertex.
|
||
struct PlannedDraw {
|
||
mesh: Arc<solidcraft_gpu::GpuMesh>,
|
||
colors: wgpu::Buffer,
|
||
}
|
||
|
||
#[derive(Default)]
|
||
struct Batches {
|
||
key: Option<u64>,
|
||
tris: Option<Batch>,
|
||
trans: Option<Batch>,
|
||
glass: Option<Batch>,
|
||
/// The glass triangles as built, and the view direction they were last sorted for.
|
||
glass_cpu: Vec<u8>,
|
||
glass_sorted_for: Option<[f32; 3]>,
|
||
lines: Option<Batch>,
|
||
hidden: Option<Batch>,
|
||
overlays: Option<Batch>,
|
||
ghost: Option<Batch>,
|
||
xray: Option<Batch>,
|
||
under: Option<Batch>,
|
||
planned: Vec<PlannedDraw>,
|
||
/// Evaluated chunks by their plans' addresses, holding the plans so the addresses stay
|
||
/// theirs: a chunk whose plans are unchanged is not evaluated again when the scene is rebuilt.
|
||
resident: std::collections::HashMap<Vec<usize>, (Vec<Arc<solidcraft_engine::geom::tess::TessPlan>>, Arc<solidcraft_gpu::GpuMesh>)>,
|
||
}
|
||
|
||
impl Batches {
|
||
/// Upload a new scene. Planned bodies not yet on the GPU are evaluated there: their
|
||
/// dispatches go into `encoder`, ahead of this frame's render pass.
|
||
fn take(&mut self, device: &wgpu::Device, key: u64, slot: &SceneSlot, tess: Option<(&solidcraft_gpu::TessGpu, &mut wgpu::CommandEncoder)>) {
|
||
if self.key == Some(key) {
|
||
return;
|
||
}
|
||
if let Some(sc) = slot.lock().ok().and_then(|mut s| s.take()) {
|
||
self.planned = match tess {
|
||
Some((tess, encoder)) => self.planned(device, tess, encoder, &sc.planned),
|
||
None => Vec::new(),
|
||
};
|
||
self.tris = upload(device, "sc_tris", &sc.tris, TRI_SIZE);
|
||
self.trans = upload(device, "sc_trans", &sc.trans, TRI_SIZE);
|
||
self.glass = upload_with(device, "sc_glass", &sc.glass, TRI_SIZE, wgpu::BufferUsages::VERTEX | wgpu::BufferUsages::COPY_DST);
|
||
self.glass_cpu = sc.glass;
|
||
self.glass_sorted_for = None;
|
||
self.lines = upload(device, "sc_lines", &sc.lines, LINE_SIZE);
|
||
self.hidden = upload(device, "sc_hidden", &sc.hidden, LINE_SIZE);
|
||
self.overlays = upload(device, "sc_overlay", &sc.overlay, LINE_SIZE);
|
||
self.ghost = upload(device, "sc_ghost", &sc.ghost, TRI_SIZE);
|
||
self.xray = upload(device, "sc_xray", &sc.xray, TRI_SIZE);
|
||
self.under = upload(device, "sc_under", &sc.under, LINE_SIZE);
|
||
self.key = Some(key);
|
||
}
|
||
}
|
||
|
||
/// The planned bodies on the GPU: in chunks of at most `MAX_BATCH` vertices (one for any
|
||
/// ordinary scene), each uploaded and evaluated in one dispatch and drawn in one call. A
|
||
/// chunk whose plans are unchanged is not evaluated again; colours are rebuilt each time.
|
||
fn planned(
|
||
&mut self,
|
||
device: &wgpu::Device,
|
||
tess: &solidcraft_gpu::TessGpu,
|
||
encoder: &mut wgpu::CommandEncoder,
|
||
bodies: &[PlannedBody],
|
||
) -> Vec<PlannedDraw> {
|
||
let mut chunks: Vec<Vec<&PlannedBody>> = Vec::new();
|
||
let mut filled = 0usize;
|
||
for b in bodies {
|
||
let n = b.plan.vertex_count();
|
||
match chunks.last_mut() {
|
||
Some(c) if filled + n <= solidcraft_gpu::MAX_BATCH => c.push(b),
|
||
_ => {
|
||
chunks.push(vec![b]);
|
||
filled = 0;
|
||
}
|
||
}
|
||
filled += n;
|
||
}
|
||
let key = |c: &[&PlannedBody]| c.iter().map(|b| Arc::as_ptr(&b.plan) as usize).collect::<Vec<usize>>();
|
||
self.resident.retain(|k, _| chunks.iter().any(|c| key(c) == *k));
|
||
let mut out = Vec::with_capacity(chunks.len());
|
||
for chunk in &chunks {
|
||
let mesh = match self.resident.get(&key(chunk)) {
|
||
Some((_, m)) => m.clone(),
|
||
None => {
|
||
let plans: Vec<&solidcraft_engine::geom::tess::TessPlan> = chunk.iter().map(|b| &*b.plan).collect();
|
||
match tess.prepare_many(&plans) {
|
||
Ok(m) => {
|
||
tess.record(encoder, &m);
|
||
let m = Arc::new(m);
|
||
self.resident.insert(key(chunk), (chunk.iter().map(|b| b.plan.clone()).collect(), m.clone()));
|
||
m
|
||
}
|
||
Err(e) => {
|
||
log::warn!("bodies are not drawn on the GPU: {e}");
|
||
continue;
|
||
}
|
||
}
|
||
}
|
||
};
|
||
let mut bytes = Vec::with_capacity(mesh.vertex_face.len() * 4);
|
||
for (part, b) in mesh.parts.iter().zip(chunk) {
|
||
let last = b.face_colors.last().copied().unwrap_or([200, 200, 200, 255]);
|
||
let faces = mesh.vertex_face.get(part.vertices.start as usize..part.vertices.end as usize).unwrap_or_default();
|
||
bytes.extend(faces.iter().flat_map(|f| b.face_colors.get(*f as usize).copied().unwrap_or(last)));
|
||
}
|
||
if let Some((colors, _)) = upload(device, "sc_planned_colors", &bytes, 4).and_then(|c| c.parts.into_iter().next()) {
|
||
out.push(PlannedDraw { mesh, colors });
|
||
}
|
||
}
|
||
out
|
||
}
|
||
}
|
||
|
||
struct Resources {
|
||
tri: wgpu::RenderPipeline,
|
||
/// `tri` and `tri_cut` for planned bodies (positions and normals from the GPU evaluation,
|
||
/// colours per vertex, indexed).
|
||
tri_planned: wgpu::RenderPipeline,
|
||
tri_cut_planned: Option<(wgpu::RenderPipeline, wgpu::RenderPipeline)>,
|
||
/// The tessellation compute pipeline, where the device runs compute shaders.
|
||
tess: Option<solidcraft_gpu::TessGpu>,
|
||
/// While a section is shown (when the depth buffer has a stencil): `tri` marking where the
|
||
/// cap shows, then the cap again at the section plane's depth, so it hides the inside.
|
||
tri_cut: Option<(wgpu::RenderPipeline, wgpu::RenderPipeline)>,
|
||
/// Highlight triangles over the model (depth test ≤, no depth write).
|
||
tri_hl: wgpu::RenderPipeline,
|
||
trans: wgpu::RenderPipeline,
|
||
/// Glass: far sides (front faces culled), then near sides (back faces culled).
|
||
glass_far: wgpu::RenderPipeline,
|
||
glass_near: wgpu::RenderPipeline,
|
||
ghost: wgpu::RenderPipeline,
|
||
xray: wgpu::RenderPipeline,
|
||
line: wgpu::RenderPipeline,
|
||
/// Lines behind the model (depth test >).
|
||
line_hidden: wgpu::RenderPipeline,
|
||
overlay: wgpu::RenderPipeline,
|
||
uniform: wgpu::Buffer,
|
||
bind: wgpu::BindGroup,
|
||
linear_out: bool,
|
||
model: Batches,
|
||
sketch: Batches,
|
||
highlight: Batches,
|
||
preview: Batches,
|
||
image: wgpu::RenderPipeline,
|
||
image_bgl: wgpu::BindGroupLayout,
|
||
sampler: wgpu::Sampler,
|
||
/// Textures by image id: (version, bind group).
|
||
textures: std::collections::HashMap<u64, (u64, wgpu::BindGroup)>,
|
||
/// This frame's quads: (image id, vertices).
|
||
quads: Vec<(u64, Batch)>,
|
||
access_bgl: wgpu::BindGroupLayout,
|
||
/// The accessibility depth map (a 1 × 1 stand-in when there is none) and its version.
|
||
access_bind: wgpu::BindGroup,
|
||
access_version: u64,
|
||
}
|
||
|
||
fn access_texture(device: &wgpu::Device, queue: Option<&wgpu::Queue>, bgl: &wgpu::BindGroupLayout, size: u32, depth: &[f32]) -> wgpu::BindGroup {
|
||
let extent = wgpu::Extent3d { width: size.max(1), height: size.max(1), depth_or_array_layers: 1 };
|
||
let tex = device.create_texture(&wgpu::TextureDescriptor {
|
||
label: Some("sc_access"),
|
||
size: extent,
|
||
mip_level_count: 1,
|
||
sample_count: 1,
|
||
dimension: wgpu::TextureDimension::D2,
|
||
format: wgpu::TextureFormat::R32Float,
|
||
usage: wgpu::TextureUsages::TEXTURE_BINDING | wgpu::TextureUsages::COPY_DST,
|
||
view_formats: &[],
|
||
});
|
||
if let Some(q) = queue
|
||
&& depth.len() == (size as usize) * (size as usize)
|
||
{
|
||
let bytes: Vec<u8> = depth.iter().flat_map(|x| x.to_le_bytes()).collect();
|
||
q.write_texture(
|
||
wgpu::TexelCopyTextureInfo { texture: &tex, mip_level: 0, origin: wgpu::Origin3d::ZERO, aspect: wgpu::TextureAspect::All },
|
||
&bytes,
|
||
wgpu::TexelCopyBufferLayout { offset: 0, bytes_per_row: Some(4 * size), rows_per_image: Some(size) },
|
||
extent,
|
||
);
|
||
}
|
||
let view = tex.create_view(&Default::default());
|
||
device.create_bind_group(&wgpu::BindGroupDescriptor {
|
||
label: Some("sc_access"),
|
||
layout: bgl,
|
||
entries: &[wgpu::BindGroupEntry { binding: 2, resource: wgpu::BindingResource::TextureView(&view) }],
|
||
})
|
||
}
|
||
|
||
const SHADER: &str = include_str!(concat!(env!("CARGO_MANIFEST_DIR"), "/../../assets/shaders/viewport.wgsl"));
|
||
|
||
fn tri_layout() -> wgpu::VertexBufferLayout<'static> {
|
||
const A: [wgpu::VertexAttribute; 3] = wgpu::vertex_attr_array![0 => Float32x3, 1 => Float32x3, 2 => Unorm8x4];
|
||
wgpu::VertexBufferLayout { array_stride: TRI_SIZE as u64, step_mode: wgpu::VertexStepMode::Vertex, attributes: &A }
|
||
}
|
||
|
||
/// A planned body's vertices as the GPU evaluation writes them: the position, then the normal
|
||
/// at [`solidcraft_gpu::NORMAL_OFFSET`] (each padded to four floats).
|
||
fn planned_layout() -> wgpu::VertexBufferLayout<'static> {
|
||
const A: [wgpu::VertexAttribute; 2] = [
|
||
wgpu::VertexAttribute { format: wgpu::VertexFormat::Float32x3, offset: 0, shader_location: 0 },
|
||
wgpu::VertexAttribute { format: wgpu::VertexFormat::Float32x3, offset: solidcraft_gpu::NORMAL_OFFSET, shader_location: 1 },
|
||
];
|
||
wgpu::VertexBufferLayout { array_stride: solidcraft_gpu::VERTEX_STRIDE, step_mode: wgpu::VertexStepMode::Vertex, attributes: &A }
|
||
}
|
||
|
||
/// A planned body's colours, one per vertex.
|
||
fn color_layout() -> wgpu::VertexBufferLayout<'static> {
|
||
const A: [wgpu::VertexAttribute; 1] = wgpu::vertex_attr_array![2 => Unorm8x4];
|
||
wgpu::VertexBufferLayout { array_stride: 4, step_mode: wgpu::VertexStepMode::Vertex, attributes: &A }
|
||
}
|
||
|
||
fn line_layout() -> wgpu::VertexBufferLayout<'static> {
|
||
const A: [wgpu::VertexAttribute; 4] = wgpu::vertex_attr_array![0 => Float32x3, 1 => Float32x3, 2 => Unorm8x4, 3 => Float32];
|
||
wgpu::VertexBufferLayout { array_stride: LINE_SIZE as u64, step_mode: wgpu::VertexStepMode::Instance, attributes: &A }
|
||
}
|
||
|
||
/// The target configuration: colour format, depth format and MSAA sample count of the app.
|
||
#[derive(Clone, Copy, Debug)]
|
||
pub struct GpuTarget {
|
||
pub format: wgpu::TextureFormat,
|
||
pub depth: Option<wgpu::TextureFormat>,
|
||
pub samples: u32,
|
||
/// The device evaluates display plans (compute shaders): bodies may be drawn from them.
|
||
pub tess: bool,
|
||
}
|
||
|
||
impl Resources {
|
||
fn new(device: &wgpu::Device, t: GpuTarget, tess: Option<solidcraft_gpu::TessGpu>) -> Self {
|
||
let module =
|
||
device.create_shader_module(wgpu::ShaderModuleDescriptor { label: Some("sc_viewport"), source: wgpu::ShaderSource::Wgsl(SHADER.into()) });
|
||
let bgl = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
|
||
label: Some("sc_viewport"),
|
||
entries: &[wgpu::BindGroupLayoutEntry {
|
||
binding: 0,
|
||
visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
|
||
ty: wgpu::BindingType::Buffer { ty: wgpu::BufferBindingType::Uniform, has_dynamic_offset: false, min_binding_size: None },
|
||
count: None,
|
||
}],
|
||
});
|
||
let layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
|
||
label: Some("sc_viewport"),
|
||
bind_group_layouts: &[Some(&bgl)],
|
||
immediate_size: 0,
|
||
});
|
||
let uniform = device.create_buffer(&wgpu::BufferDescriptor {
|
||
label: Some("sc_viewport_u"),
|
||
size: UNIFORM_SIZE,
|
||
usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
|
||
mapped_at_creation: false,
|
||
});
|
||
let bind = device.create_bind_group(&wgpu::BindGroupDescriptor {
|
||
label: Some("sc_viewport"),
|
||
layout: &bgl,
|
||
entries: &[wgpu::BindGroupEntry { binding: 0, resource: uniform.as_entire_binding() }],
|
||
});
|
||
let depth = |compare: wgpu::CompareFunction, write: bool| -> Option<wgpu::DepthStencilState> {
|
||
t.depth.map(|format| wgpu::DepthStencilState {
|
||
format,
|
||
depth_write_enabled: Some(write),
|
||
depth_compare: Some(compare),
|
||
stencil: Default::default(),
|
||
bias: Default::default(),
|
||
})
|
||
};
|
||
let pipeline = |label: &str,
|
||
vs: &str,
|
||
fs: &str,
|
||
buf: wgpu::VertexBufferLayout<'static>,
|
||
ds: Option<wgpu::DepthStencilState>,
|
||
blend: Option<wgpu::BlendState>| {
|
||
device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
|
||
label: Some(label),
|
||
layout: Some(&layout),
|
||
vertex: wgpu::VertexState { module: &module, entry_point: Some(vs), buffers: &[Some(buf)], compilation_options: Default::default() },
|
||
primitive: wgpu::PrimitiveState { topology: wgpu::PrimitiveTopology::TriangleList, ..Default::default() },
|
||
depth_stencil: ds,
|
||
multisample: wgpu::MultisampleState { count: t.samples.max(1), mask: !0, alpha_to_coverage_enabled: false },
|
||
fragment: Some(wgpu::FragmentState {
|
||
module: &module,
|
||
entry_point: Some(fs),
|
||
targets: &[Some(wgpu::ColorTargetState { format: t.format, blend, write_mask: wgpu::ColorWrites::ALL })],
|
||
compilation_options: Default::default(),
|
||
}),
|
||
multiview_mask: None,
|
||
cache: None,
|
||
})
|
||
};
|
||
let alpha = Some(wgpu::BlendState::ALPHA_BLENDING);
|
||
let glass = |label: &str, cull: wgpu::Face| {
|
||
device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
|
||
label: Some(label),
|
||
layout: Some(&layout),
|
||
vertex: wgpu::VertexState {
|
||
module: &module,
|
||
entry_point: Some("vs_tri"),
|
||
buffers: &[Some(tri_layout())],
|
||
compilation_options: Default::default(),
|
||
},
|
||
primitive: wgpu::PrimitiveState { topology: wgpu::PrimitiveTopology::TriangleList, cull_mode: Some(cull), ..Default::default() },
|
||
depth_stencil: depth(wgpu::CompareFunction::LessEqual, false),
|
||
multisample: wgpu::MultisampleState { count: t.samples.max(1), mask: !0, alpha_to_coverage_enabled: false },
|
||
fragment: Some(wgpu::FragmentState {
|
||
module: &module,
|
||
entry_point: Some("fs_tri"),
|
||
targets: &[Some(wgpu::ColorTargetState { format: t.format, blend: alpha, write_mask: wgpu::ColorWrites::ALL })],
|
||
compilation_options: Default::default(),
|
||
}),
|
||
multiview_mask: None,
|
||
cache: None,
|
||
})
|
||
};
|
||
let image_bgl = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
|
||
label: Some("sc_image"),
|
||
entries: &[
|
||
wgpu::BindGroupLayoutEntry {
|
||
binding: 0,
|
||
visibility: wgpu::ShaderStages::FRAGMENT,
|
||
ty: wgpu::BindingType::Texture {
|
||
sample_type: wgpu::TextureSampleType::Float { filterable: true },
|
||
view_dimension: wgpu::TextureViewDimension::D2,
|
||
multisampled: false,
|
||
},
|
||
count: None,
|
||
},
|
||
wgpu::BindGroupLayoutEntry {
|
||
binding: 1,
|
||
visibility: wgpu::ShaderStages::FRAGMENT,
|
||
ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::Filtering),
|
||
count: None,
|
||
},
|
||
],
|
||
});
|
||
let image_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
|
||
label: Some("sc_image"),
|
||
bind_group_layouts: &[Some(&bgl), Some(&image_bgl)],
|
||
immediate_size: 0,
|
||
});
|
||
const IA: [wgpu::VertexAttribute; 3] = wgpu::vertex_attr_array![0 => Float32x3, 1 => Float32x2, 2 => Float32];
|
||
let image = device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
|
||
label: Some("sc_image"),
|
||
layout: Some(&image_layout),
|
||
vertex: wgpu::VertexState {
|
||
module: &module,
|
||
entry_point: Some("vs_img"),
|
||
buffers: &[Some(wgpu::VertexBufferLayout {
|
||
array_stride: IMG_SIZE as u64,
|
||
step_mode: wgpu::VertexStepMode::Vertex,
|
||
attributes: &IA,
|
||
})],
|
||
compilation_options: Default::default(),
|
||
},
|
||
primitive: wgpu::PrimitiveState { topology: wgpu::PrimitiveTopology::TriangleList, ..Default::default() },
|
||
depth_stencil: depth(wgpu::CompareFunction::LessEqual, false),
|
||
multisample: wgpu::MultisampleState { count: t.samples.max(1), mask: !0, alpha_to_coverage_enabled: false },
|
||
fragment: Some(wgpu::FragmentState {
|
||
module: &module,
|
||
entry_point: Some("fs_img"),
|
||
targets: &[Some(wgpu::ColorTargetState { format: t.format, blend: alpha, write_mask: wgpu::ColorWrites::ALL })],
|
||
compilation_options: Default::default(),
|
||
}),
|
||
multiview_mask: None,
|
||
cache: None,
|
||
});
|
||
let access_bgl = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
|
||
label: Some("sc_access"),
|
||
entries: &[wgpu::BindGroupLayoutEntry {
|
||
binding: 2,
|
||
visibility: wgpu::ShaderStages::FRAGMENT,
|
||
ty: wgpu::BindingType::Texture {
|
||
sample_type: wgpu::TextureSampleType::Float { filterable: false },
|
||
view_dimension: wgpu::TextureViewDimension::D2,
|
||
multisampled: false,
|
||
},
|
||
count: None,
|
||
}],
|
||
});
|
||
let solid_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
|
||
label: Some("sc_solid"),
|
||
bind_group_layouts: &[Some(&bgl), Some(&access_bgl)],
|
||
immediate_size: 0,
|
||
});
|
||
let built = [Some(tri_layout())];
|
||
let evaluated = [Some(planned_layout()), Some(color_layout())];
|
||
let solid = |label: &str, fs: &str, cull: Option<wgpu::Face>, ds: Option<wgpu::DepthStencilState>, planned: bool| {
|
||
device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
|
||
label: Some(label),
|
||
layout: Some(&solid_layout),
|
||
vertex: wgpu::VertexState {
|
||
module: &module,
|
||
entry_point: Some("vs_tri"),
|
||
buffers: if planned { &evaluated[..] } else { &built[..] },
|
||
compilation_options: Default::default(),
|
||
},
|
||
primitive: wgpu::PrimitiveState { topology: wgpu::PrimitiveTopology::TriangleList, cull_mode: cull, ..Default::default() },
|
||
depth_stencil: ds,
|
||
multisample: wgpu::MultisampleState { count: t.samples.max(1), mask: !0, alpha_to_coverage_enabled: false },
|
||
fragment: Some(wgpu::FragmentState {
|
||
module: &module,
|
||
entry_point: Some(fs),
|
||
targets: &[Some(wgpu::ColorTargetState { format: t.format, blend: None, write_mask: wgpu::ColorWrites::ALL })],
|
||
compilation_options: Default::default(),
|
||
}),
|
||
multiview_mask: None,
|
||
cache: None,
|
||
})
|
||
};
|
||
// While a section is shown: the stencil marking and the cap (see `Resources::tri_cut`).
|
||
let cut = |planned: bool| {
|
||
t.depth.filter(|f| f.has_stencil_aspect()).map(|format| {
|
||
// The visible surface marks the stencil: 1 where it is a back face (the cap).
|
||
use wgpu::StencilOperation::{Keep, Replace, Zero};
|
||
let face = |pass_op| wgpu::StencilFaceState { compare: wgpu::CompareFunction::Always, fail_op: Keep, depth_fail_op: Keep, pass_op };
|
||
let mark = wgpu::DepthStencilState {
|
||
format,
|
||
depth_write_enabled: Some(true),
|
||
depth_compare: Some(wgpu::CompareFunction::Less),
|
||
stencil: wgpu::StencilState { front: face(Zero), back: face(Replace), read_mask: 0xff, write_mask: 0xff },
|
||
bias: Default::default(),
|
||
};
|
||
// Then the cap's back faces again, only there, at the section plane's depth.
|
||
let only = wgpu::StencilFaceState { compare: wgpu::CompareFunction::Equal, ..face(Keep) };
|
||
let cap = wgpu::DepthStencilState {
|
||
format,
|
||
depth_write_enabled: Some(true),
|
||
depth_compare: Some(wgpu::CompareFunction::Always),
|
||
stencil: wgpu::StencilState { front: only, back: only, read_mask: 0xff, write_mask: 0 },
|
||
bias: Default::default(),
|
||
};
|
||
(
|
||
solid("sc_tris_mark", "fs_solid", None, Some(mark), planned),
|
||
solid("sc_tris_cap", "fs_solid_cut", Some(wgpu::Face::Front), Some(cap), planned),
|
||
)
|
||
})
|
||
};
|
||
let access_bind = access_texture(device, None, &access_bgl, 1, &[]);
|
||
let sampler = device.create_sampler(&wgpu::SamplerDescriptor {
|
||
label: Some("sc_image"),
|
||
mag_filter: wgpu::FilterMode::Linear,
|
||
min_filter: wgpu::FilterMode::Linear,
|
||
..Default::default()
|
||
});
|
||
Resources {
|
||
image,
|
||
image_bgl,
|
||
sampler,
|
||
textures: Default::default(),
|
||
quads: Vec::new(),
|
||
tri: solid("sc_tris", "fs_solid", None, depth(wgpu::CompareFunction::Less, true), false),
|
||
tri_cut: cut(false),
|
||
tri_planned: solid("sc_tris_planned", "fs_solid", None, depth(wgpu::CompareFunction::Less, true), true),
|
||
tri_cut_planned: cut(true),
|
||
tess,
|
||
access_bgl,
|
||
access_bind,
|
||
access_version: 0,
|
||
tri_hl: pipeline("sc_tris_hl", "vs_tri", "fs_tri", tri_layout(), highlight_depth(t.depth), alpha),
|
||
trans: pipeline("sc_trans", "vs_tri", "fs_tri", tri_layout(), depth(wgpu::CompareFunction::LessEqual, false), alpha),
|
||
glass_far: glass("sc_glass_far", wgpu::Face::Front),
|
||
glass_near: glass("sc_glass_near", wgpu::Face::Back),
|
||
ghost: pipeline("sc_ghost", "vs_ghost", "fs_tri", tri_layout(), depth(wgpu::CompareFunction::LessEqual, false), alpha),
|
||
xray: pipeline("sc_xray", "vs_tri", "fs_tri", tri_layout(), depth(wgpu::CompareFunction::Always, false), alpha),
|
||
line: pipeline("sc_lines", "vs_line", "fs_line", line_layout(), depth(wgpu::CompareFunction::LessEqual, false), alpha),
|
||
line_hidden: pipeline("sc_lines_hidden", "vs_line", "fs_line", line_layout(), depth(wgpu::CompareFunction::Greater, false), alpha),
|
||
overlay: pipeline("sc_overlay", "vs_line", "fs_line_all", line_layout(), depth(wgpu::CompareFunction::Always, false), alpha),
|
||
uniform,
|
||
bind,
|
||
linear_out: t.format.is_srgb(),
|
||
model: Batches::default(),
|
||
sketch: Batches::default(),
|
||
highlight: Batches::default(),
|
||
preview: Batches::default(),
|
||
}
|
||
}
|
||
}
|
||
|
||
/// Depth state of faces lit over the model (hover, selection): tested, not written, and pulled a
|
||
/// little toward the eye. They lie on the model's own faces, but the GPU evaluates a body's plan
|
||
/// in f32 while the highlight's copy of the same triangles is evaluated on the CPU, so the two
|
||
/// depths differ by rounding and an unbiased equal test lights the face in speckles and wedges.
|
||
/// The bias is a few depth steps plus the face's own slope: far less than any real gap.
|
||
fn highlight_depth(format: Option<wgpu::TextureFormat>) -> Option<wgpu::DepthStencilState> {
|
||
format.map(|format| wgpu::DepthStencilState {
|
||
format,
|
||
depth_write_enabled: Some(false),
|
||
depth_compare: Some(wgpu::CompareFunction::LessEqual),
|
||
stencil: Default::default(),
|
||
bias: HIGHLIGHT_BIAS,
|
||
})
|
||
}
|
||
|
||
/// See [`highlight_depth`]. Negative: toward the eye (depth grows away from it).
|
||
const HIGHLIGHT_BIAS: wgpu::DepthBiasState = wgpu::DepthBiasState { constant: -8, slope_scale: -1.0, clamp: 0.0 };
|
||
|
||
fn upload(device: &wgpu::Device, label: &str, bytes: &[u8], stride: usize) -> Option<Batch> {
|
||
upload_with(device, label, bytes, stride, wgpu::BufferUsages::VERTEX)
|
||
}
|
||
|
||
fn upload_with(device: &wgpu::Device, label: &str, bytes: &[u8], stride: usize, usage: wgpu::BufferUsages) -> Option<Batch> {
|
||
if bytes.len() < stride || stride == 0 {
|
||
return None;
|
||
}
|
||
let chunk = buffer_chunk(stride, device.limits().max_buffer_size);
|
||
let parts = bytes
|
||
.chunks(chunk)
|
||
.filter(|part| part.len() >= stride)
|
||
.map(|part| {
|
||
let buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor { label: Some(label), contents: part, usage });
|
||
(buffer, u32::try_from(part.len() / stride).unwrap_or(0))
|
||
})
|
||
.collect();
|
||
Some(Batch { parts, chunk })
|
||
}
|
||
|
||
/// Bytes per GPU buffer for vertices of `stride` bytes: within the device's limit and a
|
||
/// whole number of triangles and line pairs, so no primitive straddles two buffers.
|
||
fn buffer_chunk(stride: usize, max_buffer_size: u64) -> usize {
|
||
let unit = stride.max(1) * 6;
|
||
let max = usize::try_from(max_buffer_size).unwrap_or(usize::MAX).min(1 << 30);
|
||
(max / unit).max(1) * unit
|
||
}
|
||
|
||
/// Triangles (3 vertices of `TRI_SIZE` bytes each, position first) ordered far to near along
|
||
/// the view direction `back` (toward the eye), so blended glass draws back to front.
|
||
pub fn sort_back_to_front(bytes: &[u8], back: [f32; 3]) -> Vec<u8> {
|
||
let tri = TRI_SIZE * 3;
|
||
let depth = |t: &[u8]| -> f32 {
|
||
let f = |o: usize| t.get(o..o + 4).and_then(|b| b.try_into().ok()).map_or(0.0, f32::from_le_bytes);
|
||
(0..3).map(|v| (0..3).map(|k| f(v * TRI_SIZE + k * 4) * back[k]).sum::<f32>()).sum::<f32>()
|
||
};
|
||
let mut tris: Vec<(f32, &[u8])> = bytes.chunks_exact(tri).map(|t| (depth(t), t)).collect();
|
||
tris.sort_by(|a, b| a.0.total_cmp(&b.0));
|
||
tris.into_iter().flat_map(|(_, t)| t.iter().copied()).collect()
|
||
}
|
||
|
||
impl Batches {
|
||
/// Re-sort the glass for a new view direction (when it turned more than about 2 degrees).
|
||
fn sort_glass(&mut self, queue: &wgpu::Queue, back: [f32; 3]) {
|
||
let (Some(g), false) = (&self.glass, self.glass_cpu.is_empty()) else { return };
|
||
let turned = self.glass_sorted_for.is_none_or(|s| s[0] * back[0] + s[1] * back[1] + s[2] * back[2] < 0.9994);
|
||
if turned {
|
||
let sorted = sort_back_to_front(&self.glass_cpu, back);
|
||
for ((buffer, _), bytes) in g.parts.iter().zip(sorted.chunks(g.chunk.max(1))) {
|
||
queue.write_buffer(buffer, 0, bytes);
|
||
}
|
||
self.glass_sorted_for = Some(back);
|
||
}
|
||
}
|
||
}
|
||
|
||
/// The tessellation pipeline (`shader`: its source), where display plans are evaluated: on a
|
||
/// device that runs compute shaders with the storage buffers the evaluation binds (not WebGL2)
|
||
/// and builds the pipeline. Elsewhere `None`, and bodies are built on the CPU: a device whose
|
||
/// shader compiler rejects the shader (D3D12's FXC once did) only loses GPU evaluation.
|
||
fn tessellation(adapter: &wgpu::Adapter, device: &wgpu::Device, queue: &wgpu::Queue, shader: &str) -> Option<solidcraft_gpu::TessGpu> {
|
||
let compute = adapter.get_downlevel_capabilities().flags.contains(wgpu::DownlevelFlags::COMPUTE_SHADERS)
|
||
&& device.limits().max_storage_buffers_per_shader_stage >= 4;
|
||
if !compute {
|
||
return None;
|
||
}
|
||
let info = adapter.get_info();
|
||
match solidcraft_gpu::TessGpu::with_shader(info.name.clone(), device.clone(), queue.clone(), shader) {
|
||
Ok(t) => Some(t),
|
||
Err(e) => {
|
||
log::warn!("bodies are built on the CPU on {} ({:?}): {e}", info.name, info.backend);
|
||
None
|
||
}
|
||
}
|
||
}
|
||
|
||
/// Create the pipelines. Call once with the app's render state and its depth/stencil/MSAA
|
||
/// settings. Without a stencil the section cap does not hide the inside of cut bodies.
|
||
pub fn install(rs: &egui_wgpu::RenderState, depth_bits: u8, stencil_bits: u8, samples: u32) -> GpuTarget {
|
||
let tess = tessellation(&rs.adapter, &rs.device, &rs.queue, solidcraft_gpu::SHADER);
|
||
let t = GpuTarget { format: rs.target_format, depth: egui_wgpu::depth_format_from_bits(depth_bits, stencil_bits), samples, tess: tess.is_some() };
|
||
let res = Resources::new(&rs.device, t, tess);
|
||
rs.renderer.write().callback_resources.insert(res);
|
||
t
|
||
}
|
||
|
||
impl egui_wgpu::CallbackTrait for ViewportCallback {
|
||
fn prepare(
|
||
&self,
|
||
device: &wgpu::Device,
|
||
queue: &wgpu::Queue,
|
||
_screen: &egui_wgpu::ScreenDescriptor,
|
||
encoder: &mut wgpu::CommandEncoder,
|
||
resources: &mut egui_wgpu::CallbackResources,
|
||
) -> Vec<wgpu::CommandBuffer> {
|
||
let Some(res) = resources.get_mut::<Resources>() else { return Vec::new() };
|
||
res.model.take(device, self.key, &self.slot, res.tess.as_ref().map(|t| (t, &mut *encoder)));
|
||
res.sketch.take(device, self.sketch_key, &self.sketch_slot, None);
|
||
res.highlight.take(device, self.hl_key, &self.hl_slot, None);
|
||
res.preview.take(device, self.pv_key, &self.pv_slot, None);
|
||
res.model.sort_glass(queue, self.back);
|
||
queue.write_buffer(&res.uniform, 0, &uniform_bytes(self, self.size_px[0], self.size_px[1], res.linear_out));
|
||
if let Some(a) = &self.access
|
||
&& a.version != res.access_version
|
||
{
|
||
res.access_bind = access_texture(device, Some(queue), &res.access_bgl, a.size, &a.depth);
|
||
res.access_version = a.version;
|
||
}
|
||
// Images: upload new or changed textures, drop removed ones, rebuild the quads.
|
||
res.textures.retain(|id, _| self.images.iter().any(|i| i.id == *id));
|
||
res.quads.clear();
|
||
for img in &self.images {
|
||
let [w, h] = img.size;
|
||
if w == 0 || h == 0 || img.rgba.len() != (w as usize) * (h as usize) * 4 {
|
||
continue;
|
||
}
|
||
if res.textures.get(&img.id).is_none_or(|t| t.0 != img.version) {
|
||
let size = wgpu::Extent3d { width: w, height: h, depth_or_array_layers: 1 };
|
||
let tex = device.create_texture(&wgpu::TextureDescriptor {
|
||
label: Some("sc_image"),
|
||
size,
|
||
mip_level_count: 1,
|
||
sample_count: 1,
|
||
dimension: wgpu::TextureDimension::D2,
|
||
format: wgpu::TextureFormat::Rgba8Unorm,
|
||
usage: wgpu::TextureUsages::TEXTURE_BINDING | wgpu::TextureUsages::COPY_DST,
|
||
view_formats: &[],
|
||
});
|
||
queue.write_texture(
|
||
wgpu::TexelCopyTextureInfo { texture: &tex, mip_level: 0, origin: wgpu::Origin3d::ZERO, aspect: wgpu::TextureAspect::All },
|
||
&img.rgba,
|
||
wgpu::TexelCopyBufferLayout { offset: 0, bytes_per_row: Some(4 * w), rows_per_image: Some(h) },
|
||
size,
|
||
);
|
||
let view = tex.create_view(&Default::default());
|
||
let bind = device.create_bind_group(&wgpu::BindGroupDescriptor {
|
||
label: Some("sc_image"),
|
||
layout: &res.image_bgl,
|
||
entries: &[
|
||
wgpu::BindGroupEntry { binding: 0, resource: wgpu::BindingResource::TextureView(&view) },
|
||
wgpu::BindGroupEntry { binding: 1, resource: wgpu::BindingResource::Sampler(&res.sampler) },
|
||
],
|
||
});
|
||
res.textures.insert(img.id, (img.version, bind));
|
||
}
|
||
let c = img.corners;
|
||
let uv = [[0.0_f32, 1.0], [1.0, 1.0], [1.0, 0.0], [0.0, 0.0]];
|
||
let mut bytes = Vec::with_capacity(6 * IMG_SIZE);
|
||
for k in [0, 1, 2, 0, 2, 3] {
|
||
for v in c[k].iter().chain(&uv[k]).chain(std::iter::once(&img.opacity)) {
|
||
bytes.extend_from_slice(&v.to_le_bytes());
|
||
}
|
||
}
|
||
if let Some(b) = upload(device, "sc_image_quad", &bytes, IMG_SIZE) {
|
||
res.quads.push((img.id, b));
|
||
}
|
||
}
|
||
Vec::new()
|
||
}
|
||
|
||
fn paint(&self, info: egui::PaintCallbackInfo, pass: &mut wgpu::RenderPass<'static>, resources: &egui_wgpu::CallbackResources) {
|
||
let Some(res) = resources.get::<Resources>() else { return };
|
||
let vp = info.viewport_in_pixels();
|
||
if vp.width_px <= 0 || vp.height_px <= 0 {
|
||
return;
|
||
}
|
||
pass.set_viewport(vp.left_px as f32, vp.top_px as f32, vp.width_px as f32, vp.height_px as f32, 0.0, 1.0);
|
||
pass.set_bind_group(0, &res.bind, &[]);
|
||
let tris = |pass: &mut wgpu::RenderPass<'static>, p: &wgpu::RenderPipeline, b: &Option<Batch>| {
|
||
if let Some(b) = b {
|
||
pass.set_pipeline(p);
|
||
for (buffer, count) in &b.parts {
|
||
pass.set_vertex_buffer(0, buffer.slice(..));
|
||
pass.draw(0..*count, 0..1);
|
||
}
|
||
}
|
||
};
|
||
let lines = |pass: &mut wgpu::RenderPass<'static>, p: &wgpu::RenderPipeline, b: &Option<Batch>| {
|
||
if let Some(b) = b {
|
||
pass.set_pipeline(p);
|
||
for (buffer, count) in &b.parts {
|
||
pass.set_vertex_buffer(0, buffer.slice(..));
|
||
pass.draw(0..6, 0..*count);
|
||
}
|
||
}
|
||
};
|
||
let (m, sk, h, pv) = (&res.model, &res.sketch, &res.highlight, &res.preview);
|
||
lines(pass, &res.overlay, &m.under);
|
||
lines(pass, &res.overlay, &sk.under);
|
||
pass.set_bind_group(1, &res.access_bind, &[]);
|
||
let planned = |pass: &mut wgpu::RenderPass<'static>, p: &wgpu::RenderPipeline| {
|
||
for d in &m.planned {
|
||
pass.set_pipeline(p);
|
||
pass.set_vertex_buffer(0, d.mesh.vertices.slice(..));
|
||
pass.set_vertex_buffer(1, d.colors.slice(..));
|
||
pass.set_index_buffer(d.mesh.indices.slice(..), wgpu::IndexFormat::Uint32);
|
||
pass.draw_indexed(0..d.mesh.index_count, 0, 0..1);
|
||
}
|
||
};
|
||
let sectioned = self.clip[..3].iter().map(|c| c * c).sum::<f32>() > 0.25;
|
||
match (res.tri_cut.as_ref().filter(|_| sectioned), res.tri_cut_planned.as_ref()) {
|
||
(Some((mark, cap)), Some((mark_planned, cap_planned))) => {
|
||
pass.set_stencil_reference(1);
|
||
tris(pass, mark, &m.tris);
|
||
tris(pass, mark, &pv.tris);
|
||
planned(pass, mark_planned);
|
||
tris(pass, cap, &m.tris);
|
||
tris(pass, cap, &pv.tris);
|
||
planned(pass, cap_planned);
|
||
}
|
||
_ => {
|
||
tris(pass, &res.tri, &m.tris);
|
||
tris(pass, &res.tri, &pv.tris);
|
||
planned(pass, &res.tri_planned);
|
||
}
|
||
}
|
||
tris(pass, &res.tri_hl, &h.tris);
|
||
lines(pass, &res.line, &m.lines);
|
||
lines(pass, &res.line, &sk.lines);
|
||
lines(pass, &res.line_hidden, &m.hidden);
|
||
lines(pass, &res.line_hidden, &h.hidden);
|
||
lines(pass, &res.line, &pv.lines);
|
||
lines(pass, &res.line, &h.lines);
|
||
for (id, q) in &res.quads {
|
||
if let Some((_, bind)) = res.textures.get(id) {
|
||
pass.set_pipeline(&res.image);
|
||
pass.set_bind_group(1, bind, &[]);
|
||
for (buffer, count) in &q.parts {
|
||
pass.set_vertex_buffer(0, buffer.slice(..));
|
||
pass.draw(0..*count, 0..1);
|
||
}
|
||
}
|
||
}
|
||
tris(pass, &res.ghost, &pv.ghost);
|
||
tris(pass, &res.glass_far, &m.glass);
|
||
tris(pass, &res.glass_near, &m.glass);
|
||
tris(pass, &res.trans, &m.trans);
|
||
tris(pass, &res.trans, &sk.trans);
|
||
tris(pass, &res.trans, &h.trans);
|
||
tris(pass, &res.xray, &pv.xray);
|
||
lines(pass, &res.overlay, &m.overlays);
|
||
lines(pass, &res.overlay, &sk.overlays);
|
||
lines(pass, &res.overlay, &h.overlays);
|
||
}
|
||
}
|
||
|
||
/// Render `scene` offscreen on a device of its own, through the viewport's real resources and
|
||
/// frame callback (`prepare`, then `paint`), and read the pixels back: straight RGBA, `size` ×
|
||
/// `size`. `None` without a GPU adapter. With `tess` (the tessellation shader), planned bodies
|
||
/// are evaluated on the GPU where the device builds that pipeline, as [`install`] sets it up.
|
||
/// `highlight` is drawn as the viewport's highlight scene (hover and selection).
|
||
#[cfg(all(test, not(target_arch = "wasm32")))]
|
||
fn offscreen_device() -> Option<(wgpu::Adapter, wgpu::Device, wgpu::Queue)> {
|
||
let instance = wgpu::Instance::new(wgpu::InstanceDescriptor::new_without_display_handle_from_env());
|
||
let adapter = pollster::block_on(instance.request_adapter(&Default::default())).ok()?;
|
||
let (device, queue) = pollster::block_on(adapter.request_device(&wgpu::DeviceDescriptor {
|
||
required_limits: wgpu::Limits::downlevel_defaults().using_resolution(adapter.limits()),
|
||
..Default::default()
|
||
}))
|
||
.ok()?;
|
||
Some((adapter, device, queue))
|
||
}
|
||
|
||
/// Whether [`install`] would evaluate plans on the GPU with `shader`, on an offscreen device
|
||
/// (`None` without a GPU adapter).
|
||
#[cfg(all(test, not(target_arch = "wasm32")))]
|
||
pub(crate) fn offscreen_tessellates(shader: &str) -> Option<bool> {
|
||
let (adapter, device, queue) = offscreen_device()?;
|
||
Some(tessellation(&adapter, &device, &queue, shader).is_some())
|
||
}
|
||
|
||
#[cfg(all(test, not(target_arch = "wasm32")))]
|
||
pub(crate) fn render_offscreen(
|
||
scene: GpuScene,
|
||
highlight: Option<GpuScene>,
|
||
view_proj: [[f32; 4]; 4],
|
||
back: [f32; 3],
|
||
eye: [f32; 4],
|
||
size: u32,
|
||
tess: Option<&str>,
|
||
) -> Option<Vec<u8>> {
|
||
let (adapter, device, queue) = offscreen_device()?;
|
||
let (format, depth) = (wgpu::TextureFormat::Rgba8Unorm, wgpu::TextureFormat::Depth24PlusStencil8);
|
||
let compute = tess.and_then(|shader| tessellation(&adapter, &device, &queue, shader));
|
||
let target = GpuTarget { format, depth: Some(depth), samples: 1, tess: compute.is_some() };
|
||
let mut resources = egui_wgpu::CallbackResources::default();
|
||
resources.insert(Resources::new(&device, target, compute));
|
||
let none = || Arc::new(Mutex::new(None));
|
||
let cb = ViewportCallback {
|
||
key: 1,
|
||
slot: Arc::new(Mutex::new(Some(scene))),
|
||
sketch_key: 0,
|
||
sketch_slot: none(),
|
||
hl_key: u64::from(highlight.is_some()),
|
||
hl_slot: Arc::new(Mutex::new(highlight)),
|
||
pv_key: 0,
|
||
pv_slot: none(),
|
||
view_proj,
|
||
back,
|
||
size_px: [size as f32; 2],
|
||
clip: [0.0; 4],
|
||
cap: [0.0; 4],
|
||
images: Vec::new(),
|
||
analysis: [0.0; 8],
|
||
access: None,
|
||
outline: false,
|
||
eye,
|
||
};
|
||
let texture = |format, usage| {
|
||
device.create_texture(&wgpu::TextureDescriptor {
|
||
label: Some("offscreen"),
|
||
size: wgpu::Extent3d { width: size, height: size, depth_or_array_layers: 1 },
|
||
mip_level_count: 1,
|
||
sample_count: 1,
|
||
dimension: wgpu::TextureDimension::D2,
|
||
format,
|
||
usage,
|
||
view_formats: &[],
|
||
})
|
||
};
|
||
let color = texture(format, wgpu::TextureUsages::RENDER_ATTACHMENT | wgpu::TextureUsages::COPY_SRC);
|
||
let depth_tex = texture(depth, wgpu::TextureUsages::RENDER_ATTACHMENT);
|
||
let (color_view, depth_view) = (color.create_view(&Default::default()), depth_tex.create_view(&Default::default()));
|
||
let mut encoder = device.create_command_encoder(&Default::default());
|
||
let screen = egui_wgpu::ScreenDescriptor { size_in_pixels: [size, size], pixels_per_point: 1.0 };
|
||
let extra = egui_wgpu::CallbackTrait::prepare(&cb, &device, &queue, &screen, &mut encoder, &mut resources);
|
||
{
|
||
let mut pass = encoder
|
||
.begin_render_pass(&wgpu::RenderPassDescriptor {
|
||
label: Some("offscreen"),
|
||
color_attachments: &[Some(wgpu::RenderPassColorAttachment {
|
||
view: &color_view,
|
||
depth_slice: None,
|
||
resolve_target: None,
|
||
ops: wgpu::Operations { load: wgpu::LoadOp::Clear(wgpu::Color::WHITE), store: wgpu::StoreOp::Store },
|
||
})],
|
||
depth_stencil_attachment: Some(wgpu::RenderPassDepthStencilAttachment {
|
||
view: &depth_view,
|
||
depth_ops: Some(wgpu::Operations { load: wgpu::LoadOp::Clear(1.0), store: wgpu::StoreOp::Store }),
|
||
stencil_ops: Some(wgpu::Operations { load: wgpu::LoadOp::Clear(0), store: wgpu::StoreOp::Store }),
|
||
}),
|
||
timestamp_writes: None,
|
||
occlusion_query_set: None,
|
||
multiview_mask: None,
|
||
})
|
||
.forget_lifetime();
|
||
let rect = egui::Rect::from_min_size(egui::Pos2::ZERO, egui::vec2(size as f32, size as f32));
|
||
let info = egui::PaintCallbackInfo { viewport: rect, clip_rect: rect, pixels_per_point: 1.0, screen_size_px: [size, size] };
|
||
egui_wgpu::CallbackTrait::paint(&cb, info, &mut pass, &resources);
|
||
}
|
||
let read = device.create_buffer(&wgpu::BufferDescriptor {
|
||
label: Some("offscreen read"),
|
||
size: u64::from(size * size * 4),
|
||
usage: wgpu::BufferUsages::MAP_READ | wgpu::BufferUsages::COPY_DST,
|
||
mapped_at_creation: false,
|
||
});
|
||
encoder.copy_texture_to_buffer(
|
||
color.as_image_copy(),
|
||
wgpu::TexelCopyBufferInfo {
|
||
buffer: &read,
|
||
layout: wgpu::TexelCopyBufferLayout { offset: 0, bytes_per_row: Some(size * 4), rows_per_image: Some(size) },
|
||
},
|
||
wgpu::Extent3d { width: size, height: size, depth_or_array_layers: 1 },
|
||
);
|
||
queue.submit(extra.into_iter().chain([encoder.finish()]));
|
||
let slice = read.slice(..);
|
||
slice.map_async(wgpu::MapMode::Read, |_| {});
|
||
device.poll(wgpu::PollType::wait_indefinitely()).ok()?;
|
||
let bytes = slice.get_mapped_range().ok()?.to_vec();
|
||
Some(bytes)
|
||
}
|
||
|
||
/// Uniform bytes for a frame.
|
||
pub fn uniform_bytes(cb: &ViewportCallback, w: f32, h: f32, linear: bool) -> Vec<u8> {
|
||
let mut v: Vec<f32> = cb.view_proj.iter().flatten().copied().collect();
|
||
v.extend([cb.back[0], cb.back[1], cb.back[2], if linear { 1.0 } else { 0.0 }]);
|
||
v.extend([w.max(1.0), h.max(1.0), 2e-4, if cb.outline { 1.0 } else { 0.0 }]);
|
||
v.extend(cb.clip);
|
||
v.extend(cb.cap);
|
||
v.extend(cb.analysis);
|
||
match &cb.access {
|
||
Some(a) => {
|
||
v.extend(a.u);
|
||
v.extend(a.v);
|
||
}
|
||
None => v.extend([0.0; 8]),
|
||
}
|
||
v.extend(cb.eye);
|
||
v.iter().flat_map(|x| x.to_le_bytes()).collect()
|
||
}
|
||
|
||
#[cfg(test)]
|
||
mod tests {
|
||
use egui_wgpu::wgpu::naga;
|
||
|
||
/// The viewport shader parses and validates (a bad shader would only fail at run time).
|
||
#[test]
|
||
fn shader_is_valid() {
|
||
let module = naga::front::wgsl::parse_str(super::SHADER).unwrap_or_else(|e| panic!("{}", e.emit_to_string(super::SHADER)));
|
||
naga::valid::Validator::new(naga::valid::ValidationFlags::all(), naga::valid::Capabilities::empty()).validate(&module).unwrap();
|
||
}
|
||
|
||
/// Derivatives and implicit-lod sampling run only in uniform control flow: browsers' WGSL
|
||
/// compilers (Tint) reject them after a branch, and the whole viewport shader then fails,
|
||
/// a black screen on the web (naga accepts it). Every fragment entry point makes such calls
|
||
/// (directly or through a helper) before its first branch or return.
|
||
#[test]
|
||
fn derivatives_only_in_uniform_control_flow() {
|
||
// Without comments (their words are not code).
|
||
let src: String = super::SHADER.lines().map(|l| l.split("//").next().unwrap_or("")).collect::<Vec<_>>().join("\n");
|
||
let src = src.as_str();
|
||
let derivs = ["dpdx(", "dpdy(", "fwidth(", "textureSample("];
|
||
// Functions: name -> body.
|
||
let mut fns: Vec<(String, String, bool)> = Vec::new();
|
||
let mut rest = src;
|
||
let mut fragment = false;
|
||
while let Some(at) = rest.find("fn ") {
|
||
let before = &rest[..at];
|
||
if before.contains("@fragment") {
|
||
fragment = true;
|
||
}
|
||
let Some(open) = rest[at..].find('{') else { break };
|
||
let name = rest[at + 3..].split('(').next().unwrap_or("").trim().to_string();
|
||
let start = at + open;
|
||
let mut depth = 0;
|
||
let mut end = start;
|
||
for (k, ch) in rest[start..].char_indices() {
|
||
match ch {
|
||
'{' => depth += 1,
|
||
'}' => {
|
||
depth -= 1;
|
||
if depth == 0 {
|
||
end = start + k;
|
||
break;
|
||
}
|
||
}
|
||
_ => {}
|
||
}
|
||
}
|
||
fns.push((name, rest[start..=end].to_string(), fragment));
|
||
fragment = false;
|
||
rest = &rest[end + 1..];
|
||
}
|
||
// Helpers that take derivatives (directly or through another helper).
|
||
let mut uses: Vec<String> = Vec::new();
|
||
for _ in 0..8 {
|
||
for (n, body, _) in &fns {
|
||
if !uses.contains(n) && (derivs.iter().any(|d| body.contains(d)) || uses.iter().any(|u| body.contains(&format!("{u}(")))) {
|
||
uses.push(n.clone());
|
||
}
|
||
}
|
||
}
|
||
for (n, body, frag) in &fns {
|
||
let calls: Vec<usize> = derivs
|
||
.iter()
|
||
.map(|d| d.to_string())
|
||
.chain(uses.iter().map(|u| format!("{u}(")))
|
||
.flat_map(|c| body.match_indices(&c).map(|(i, _)| i).collect::<Vec<_>>())
|
||
.collect();
|
||
if calls.is_empty() {
|
||
continue;
|
||
}
|
||
if !*frag {
|
||
// A helper: its own derivatives must come before any branch or return in it.
|
||
let first = ["if (", "return", "discard"].iter().filter_map(|k| body.find(k)).min().unwrap_or(usize::MAX);
|
||
for d in derivs {
|
||
if let Some(i) = body.find(d) {
|
||
assert!(i < first, "{n}: {d} after a branch");
|
||
}
|
||
}
|
||
continue;
|
||
}
|
||
let first = ["if (", "return", "discard"].iter().filter_map(|k| body.find(k)).min().unwrap_or(usize::MAX);
|
||
for c in calls {
|
||
assert!(c < first, "fragment {n}: a derivative after a branch or return: {}", &body[c..(c + 40).min(body.len())]);
|
||
}
|
||
}
|
||
}
|
||
}
|
||
|
||
#[cfg(test)]
|
||
mod glass_tests {
|
||
use super::*;
|
||
|
||
/// A large import's triangles once went into one buffer past the device's limit (524 MB
|
||
/// against 256 MB) and the app crashed (#883): they are split into whole triangles.
|
||
#[test]
|
||
fn large_scenes_are_split_into_buffers_within_the_device_limit() {
|
||
let limit = 256u64 << 20;
|
||
for stride in [TRI_SIZE, LINE_SIZE, 4] {
|
||
let chunk = buffer_chunk(stride, limit);
|
||
assert!(chunk as u64 <= limit && chunk > limit as usize / 2);
|
||
assert_eq!(chunk % (stride * 3), 0, "whole triangles");
|
||
assert_eq!(chunk % (stride * 2), 0, "whole line pairs");
|
||
}
|
||
let scene = 524_434_680usize;
|
||
assert_eq!(scene.div_ceil(buffer_chunk(TRI_SIZE, limit)), 2);
|
||
assert!(buffer_chunk(TRI_SIZE, 1) >= TRI_SIZE * 6, "a tiny limit still holds one unit");
|
||
}
|
||
|
||
#[test]
|
||
fn glass_is_drawn_far_to_near() {
|
||
// Two triangles at z = 5 and z = -5, given near first; looking down from +Z.
|
||
let mut sc = GpuScene::default();
|
||
for z in [5.0f32, -5.0] {
|
||
for p in [[0.0, 0.0, z], [1.0, 0.0, z], [0.0, 1.0, z]] {
|
||
sc.glass_tri(p, [0.0, 0.0, 1.0], [255, 0, 0, 100]);
|
||
}
|
||
}
|
||
let sorted = sort_back_to_front(&sc.glass, [0.0, 0.0, 1.0]);
|
||
let z0 = f32::from_le_bytes(sorted[8..12].try_into().unwrap());
|
||
assert_eq!(z0, -5.0, "the far triangle first");
|
||
assert_eq!(sorted.len(), sc.glass.len());
|
||
// Seen from below, the order flips.
|
||
let sorted = sort_back_to_front(&sc.glass, [0.0, 0.0, -1.0]);
|
||
assert_eq!(f32::from_le_bytes(sorted[8..12].try_into().unwrap()), 5.0);
|
||
}
|
||
}
|