1use crate::core::renderer::Vertex;
2use crate::core::scene::GpuVertexBuffer;
3use crate::gpu::axis::{axis_storage_buffer, AxisData};
4use crate::gpu::shaders;
5use crate::gpu::{tuning, ScalarType};
6use glam::Vec4;
7use std::sync::Arc;
8use wgpu::util::DeviceExt;
9
10pub struct QuiverGpuInputs<'a> {
11 pub x_data: AxisData<'a>,
12 pub y_data: AxisData<'a>,
13 pub u_buffer: Arc<wgpu::Buffer>,
14 pub v_buffer: Arc<wgpu::Buffer>,
15 pub count: u32,
16 pub rows: u32,
17 pub cols: u32,
18 pub xy_mode: u32,
19 pub scalar: ScalarType,
20}
21
22pub struct QuiverGpuParams {
23 pub color: Vec4,
24 pub scale: f32,
25 pub head_size: f32,
26}
27
28#[repr(C)]
29#[derive(Clone, Copy, bytemuck::Pod, bytemuck::Zeroable)]
30struct QuiverUniforms {
31 color: [f32; 4],
32 count: u32,
33 rows: u32,
34 cols: u32,
35 xy_mode: u32,
36 scale: f32,
37 head_size: f32,
38 _pad: f32,
39}
40
41pub fn pack_vertices(
42 device: &Arc<wgpu::Device>,
43 queue: &Arc<wgpu::Queue>,
44 inputs: &QuiverGpuInputs<'_>,
45 params: &QuiverGpuParams,
46) -> Result<GpuVertexBuffer, String> {
47 let workgroup_size = tuning::effective_workgroup_size();
48 let shader = compile_shader(device, workgroup_size, inputs.scalar);
49 let x_buffer = axis_storage_buffer(device, "quiver-x", &inputs.x_data, inputs.scalar)?;
50 let y_buffer = axis_storage_buffer(device, "quiver-y", &inputs.y_data, inputs.scalar)?;
51 let vertex_count = inputs.count as u64 * 6;
52 let output_buffer = Arc::new(device.create_buffer(&wgpu::BufferDescriptor {
53 label: Some("quiver-gpu-vertices"),
54 size: vertex_count * std::mem::size_of::<Vertex>() as u64,
55 usage: wgpu::BufferUsages::STORAGE
56 | wgpu::BufferUsages::VERTEX
57 | wgpu::BufferUsages::COPY_DST,
58 mapped_at_creation: false,
59 }));
60 let uniforms = QuiverUniforms {
61 color: params.color.to_array(),
62 count: inputs.count,
63 rows: inputs.rows,
64 cols: inputs.cols,
65 xy_mode: inputs.xy_mode,
66 scale: params.scale,
67 head_size: params.head_size,
68 _pad: 0.0,
69 };
70 let uniform_buffer = device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
71 label: Some("quiver-pack-uniforms"),
72 contents: bytemuck::bytes_of(&uniforms),
73 usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
74 });
75 let bind_group_layout = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
76 label: Some("quiver-pack-bind-layout"),
77 entries: &[
78 storage_entry(0, true),
79 storage_entry(1, true),
80 storage_entry(2, true),
81 storage_entry(3, true),
82 storage_entry(4, false),
83 uniform_entry(5),
84 ],
85 });
86 let pipeline_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
87 label: Some("quiver-pack-pipeline-layout"),
88 bind_group_layouts: &[&bind_group_layout],
89 push_constant_ranges: &[],
90 });
91 let pipeline = device.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
92 label: Some("quiver-pack-pipeline"),
93 layout: Some(&pipeline_layout),
94 module: &shader,
95 entry_point: "main",
96 });
97 let bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
98 label: Some("quiver-pack-bind-group"),
99 layout: &bind_group_layout,
100 entries: &[
101 wgpu::BindGroupEntry {
102 binding: 0,
103 resource: x_buffer.as_entire_binding(),
104 },
105 wgpu::BindGroupEntry {
106 binding: 1,
107 resource: y_buffer.as_entire_binding(),
108 },
109 wgpu::BindGroupEntry {
110 binding: 2,
111 resource: inputs.u_buffer.as_entire_binding(),
112 },
113 wgpu::BindGroupEntry {
114 binding: 3,
115 resource: inputs.v_buffer.as_entire_binding(),
116 },
117 wgpu::BindGroupEntry {
118 binding: 4,
119 resource: output_buffer.as_entire_binding(),
120 },
121 wgpu::BindGroupEntry {
122 binding: 5,
123 resource: uniform_buffer.as_entire_binding(),
124 },
125 ],
126 });
127 let mut encoder = device.create_command_encoder(&wgpu::CommandEncoderDescriptor {
128 label: Some("quiver-pack-encoder"),
129 });
130 {
131 let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
132 label: Some("quiver-pack-pass"),
133 timestamp_writes: None,
134 });
135 pass.set_pipeline(&pipeline);
136 pass.set_bind_group(0, &bind_group, &[]);
137 pass.dispatch_workgroups(inputs.count.div_ceil(workgroup_size), 1, 1);
138 }
139 queue.submit(Some(encoder.finish()));
140 Ok(GpuVertexBuffer::new(output_buffer, vertex_count as usize))
141}
142
143fn compile_shader(
144 device: &Arc<wgpu::Device>,
145 workgroup_size: u32,
146 scalar: ScalarType,
147) -> wgpu::ShaderModule {
148 let template = match scalar {
149 ScalarType::F32 => shaders::quiver::F32,
150 ScalarType::F64 => shaders::quiver::F64,
151 };
152 let source = template.replace("{{WORKGROUP_SIZE}}", &workgroup_size.to_string());
153 device.create_shader_module(wgpu::ShaderModuleDescriptor {
154 label: Some("quiver-pack-shader"),
155 source: wgpu::ShaderSource::Wgsl(source.into()),
156 })
157}
158
159fn storage_entry(binding: u32, read_only: bool) -> wgpu::BindGroupLayoutEntry {
160 wgpu::BindGroupLayoutEntry {
161 binding,
162 visibility: wgpu::ShaderStages::COMPUTE,
163 ty: wgpu::BindingType::Buffer {
164 ty: wgpu::BufferBindingType::Storage { read_only },
165 has_dynamic_offset: false,
166 min_binding_size: None,
167 },
168 count: None,
169 }
170}
171
172fn uniform_entry(binding: u32) -> wgpu::BindGroupLayoutEntry {
173 wgpu::BindGroupLayoutEntry {
174 binding,
175 visibility: wgpu::ShaderStages::COMPUTE,
176 ty: wgpu::BindingType::Buffer {
177 ty: wgpu::BufferBindingType::Uniform,
178 has_dynamic_offset: false,
179 min_binding_size: None,
180 },
181 count: None,
182 }
183}
184
185#[cfg(test)]
186mod tests {
187 use super::*;
188 use pollster::FutureExt;
189 fn maybe_device() -> Option<(Arc<wgpu::Device>, Arc<wgpu::Queue>)> {
190 if std::env::var("RUNMAT_PLOT_SKIP_GPU_TESTS").is_ok()
191 || std::env::var("RUNMAT_PLOT_FORCE_GPU_TESTS").is_err()
192 {
193 return None;
194 }
195 let instance = wgpu::Instance::default();
196 let adapter = instance
197 .request_adapter(&wgpu::RequestAdapterOptions {
198 power_preference: wgpu::PowerPreference::HighPerformance,
199 compatible_surface: None,
200 force_fallback_adapter: false,
201 })
202 .block_on()?;
203 let (device, queue) = adapter
204 .request_device(
205 &wgpu::DeviceDescriptor {
206 label: Some("runmat-plot-quiver-test-device"),
207 required_features: wgpu::Features::empty(),
208 required_limits: adapter.limits(),
209 },
210 None,
211 )
212 .block_on()
213 .ok()?;
214 Some((Arc::new(device), Arc::new(queue)))
215 }
216
217 #[test]
218 fn gpu_packer_generates_quiver_vertices() {
219 let Some((device, queue)) = maybe_device() else {
220 return;
221 };
222 let x = [1.0f32, 2.0f32];
223 let y = [1.0f32, 2.0f32];
224 let u = Arc::new(
225 device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
226 label: Some("quiver-test-u"),
227 contents: bytemuck::cast_slice(&[0.5f32, -0.25f32]),
228 usage: wgpu::BufferUsages::STORAGE,
229 }),
230 );
231 let v = Arc::new(
232 device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
233 label: Some("quiver-test-v"),
234 contents: bytemuck::cast_slice(&[1.0f32, 0.75f32]),
235 usage: wgpu::BufferUsages::STORAGE,
236 }),
237 );
238 let packed = pack_vertices(
239 &device,
240 &queue,
241 &QuiverGpuInputs {
242 x_data: AxisData::F32(&x),
243 y_data: AxisData::F32(&y),
244 u_buffer: u,
245 v_buffer: v,
246 count: 2,
247 rows: 2,
248 cols: 1,
249 xy_mode: 0,
250 scalar: ScalarType::F32,
251 },
252 &QuiverGpuParams {
253 color: Vec4::ONE,
254 scale: 1.0,
255 head_size: 0.2,
256 },
257 )
258 .expect("quiver pack should succeed");
259 assert_eq!(packed.vertex_count, 12);
260 }
261}