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 =
92 device.create_compute_pipeline(&crate::wgpu_compat::wgpu_compute_pipeline_descriptor! {
93 label: Some("quiver-pack-pipeline"),
94 layout: Some(&pipeline_layout),
95 module: &shader,
96 entry_point: "main",
97 });
98 let bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
99 label: Some("quiver-pack-bind-group"),
100 layout: &bind_group_layout,
101 entries: &[
102 wgpu::BindGroupEntry {
103 binding: 0,
104 resource: x_buffer.as_entire_binding(),
105 },
106 wgpu::BindGroupEntry {
107 binding: 1,
108 resource: y_buffer.as_entire_binding(),
109 },
110 wgpu::BindGroupEntry {
111 binding: 2,
112 resource: inputs.u_buffer.as_entire_binding(),
113 },
114 wgpu::BindGroupEntry {
115 binding: 3,
116 resource: inputs.v_buffer.as_entire_binding(),
117 },
118 wgpu::BindGroupEntry {
119 binding: 4,
120 resource: output_buffer.as_entire_binding(),
121 },
122 wgpu::BindGroupEntry {
123 binding: 5,
124 resource: uniform_buffer.as_entire_binding(),
125 },
126 ],
127 });
128 let mut encoder = device.create_command_encoder(&wgpu::CommandEncoderDescriptor {
129 label: Some("quiver-pack-encoder"),
130 });
131 {
132 let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
133 label: Some("quiver-pack-pass"),
134 timestamp_writes: None,
135 });
136 pass.set_pipeline(&pipeline);
137 pass.set_bind_group(0, &bind_group, &[]);
138 pass.dispatch_workgroups(inputs.count.div_ceil(workgroup_size), 1, 1);
139 }
140 queue.submit(Some(encoder.finish()));
141 Ok(GpuVertexBuffer::new(output_buffer, vertex_count as usize))
142}
143
144fn compile_shader(
145 device: &Arc<wgpu::Device>,
146 workgroup_size: u32,
147 scalar: ScalarType,
148) -> wgpu::ShaderModule {
149 let template = match scalar {
150 ScalarType::F32 => shaders::quiver::F32,
151 ScalarType::F64 => shaders::quiver::F64,
152 };
153 let source = template.replace("{{WORKGROUP_SIZE}}", &workgroup_size.to_string());
154 device.create_shader_module(wgpu::ShaderModuleDescriptor {
155 label: Some("quiver-pack-shader"),
156 source: wgpu::ShaderSource::Wgsl(source.into()),
157 })
158}
159
160fn storage_entry(binding: u32, read_only: bool) -> wgpu::BindGroupLayoutEntry {
161 wgpu::BindGroupLayoutEntry {
162 binding,
163 visibility: wgpu::ShaderStages::COMPUTE,
164 ty: wgpu::BindingType::Buffer {
165 ty: wgpu::BufferBindingType::Storage { read_only },
166 has_dynamic_offset: false,
167 min_binding_size: None,
168 },
169 count: None,
170 }
171}
172
173fn uniform_entry(binding: u32) -> wgpu::BindGroupLayoutEntry {
174 wgpu::BindGroupLayoutEntry {
175 binding,
176 visibility: wgpu::ShaderStages::COMPUTE,
177 ty: wgpu::BindingType::Buffer {
178 ty: wgpu::BufferBindingType::Uniform,
179 has_dynamic_offset: false,
180 min_binding_size: None,
181 },
182 count: None,
183 }
184}
185
186#[cfg(test)]
187mod tests {
188 use super::*;
189 use pollster::FutureExt;
190 fn maybe_device() -> Option<(Arc<wgpu::Device>, Arc<wgpu::Queue>)> {
191 if std::env::var("RUNMAT_PLOT_SKIP_GPU_TESTS").is_ok()
192 || std::env::var("RUNMAT_PLOT_FORCE_GPU_TESTS").is_err()
193 {
194 return None;
195 }
196 let instance = wgpu::Instance::default();
197 let adapter = instance
198 .request_adapter(&wgpu::RequestAdapterOptions {
199 power_preference: wgpu::PowerPreference::HighPerformance,
200 compatible_surface: None,
201 force_fallback_adapter: false,
202 })
203 .block_on()?;
204 let (device, queue) = adapter
205 .request_device(
206 &crate::wgpu_compat::device_descriptor(
207 Some("runmat-plot-quiver-test-device"),
208 wgpu::Features::empty(),
209 adapter.limits(),
210 ),
211 None,
212 )
213 .block_on()
214 .ok()?;
215 Some((Arc::new(device), Arc::new(queue)))
216 }
217
218 #[test]
219 fn gpu_packer_generates_quiver_vertices() {
220 let Some((device, queue)) = maybe_device() else {
221 return;
222 };
223 let x = [1.0f32, 2.0f32];
224 let y = [1.0f32, 2.0f32];
225 let u = Arc::new(
226 device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
227 label: Some("quiver-test-u"),
228 contents: bytemuck::cast_slice(&[0.5f32, -0.25f32]),
229 usage: wgpu::BufferUsages::STORAGE,
230 }),
231 );
232 let v = Arc::new(
233 device.create_buffer_init(&wgpu::util::BufferInitDescriptor {
234 label: Some("quiver-test-v"),
235 contents: bytemuck::cast_slice(&[1.0f32, 0.75f32]),
236 usage: wgpu::BufferUsages::STORAGE,
237 }),
238 );
239 let packed = pack_vertices(
240 &device,
241 &queue,
242 &QuiverGpuInputs {
243 x_data: AxisData::F32(&x),
244 y_data: AxisData::F32(&y),
245 u_buffer: u,
246 v_buffer: v,
247 count: 2,
248 rows: 2,
249 cols: 1,
250 xy_mode: 0,
251 scalar: ScalarType::F32,
252 },
253 &QuiverGpuParams {
254 color: Vec4::ONE,
255 scale: 1.0,
256 head_size: 0.2,
257 },
258 )
259 .expect("quiver pack should succeed");
260 assert_eq!(packed.vertex_count, 12);
261 }
262}