1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
use crate::prep_readback::{GpuReadbackData, ReadbackData};
use crate::{PhysicsState, RunState, Stage};
use slosh::rapier::na;
use slang_hal::backend::Backend;
#[cfg(feature = "webgpu")]
use slang_hal::GpuTimingResult;
use slosh::solver::{GpuParticleModelData, SimulationParams};
#[derive(Default)]
pub struct SimulationTimes {
pub total_step_time: f32,
pub encoding_time: f32,
pub readback_time: f32,
#[cfg(feature = "webgpu")]
pub gpu_passes: Vec<GpuTimingResult>,
}
#[derive(Default)]
pub struct SimulationStepResult {
pub instances: Vec<ReadbackData>,
pub timings: SimulationTimes,
}
impl<GpuModel: GpuParticleModelData> Stage<GpuModel> {
// TODO PERF: don’t reallocate the result buffer each time.
pub async fn step_simulation(&mut self) -> bool {
if self.app_state.run_state == RunState::Paused {
return false;
}
// Run the simulation.
let physics = &mut self.physics;
let prev_particle_count = physics.data.particles.len();
for callback in &mut physics.callbacks {
let mut phx = PhysicsState {
backend: &self.gpu,
data: &mut physics.data,
results: &self.step_result,
step_id: self.step_id,
};
callback.update(&mut phx);
}
// Check if the particle size changed. If it did, adjust the instance buffers.
let new_particle_count = physics.data.particles.len();
if prev_particle_count != new_particle_count {
// TODO: resize buffers instead of recreating.
self.readback =
GpuReadbackData::new(&self.gpu, new_particle_count, self.render_mode).unwrap();
self.step_result
.instances
.resize(new_particle_count, ReadbackData::default());
println!("Adjust readback buffers: {}", new_particle_count);
}
let t_total = web_time::Instant::now();
let base_dt = physics.data.base_dt;
let prev_num_substeps = self.app_state.num_substeps;
if self.app_state.min_num_substeps < self.app_state.max_num_substeps {
// Adaptive stepping.
let bounds = self
.app_state
.pipeline
.timestep_bounds
.compute_bounds(
&self.gpu,
&physics.data.grid,
&physics.data.particles,
&physics.data.timestep_bounds,
&mut physics.data.timestep_bounds_staging,
)
.await
.unwrap();
let num_substeps_estimated = (base_dt / bounds).ceil() as u32;
let num_substeps = num_substeps_estimated.clamp(
self.app_state.min_num_substeps,
self.app_state.max_num_substeps,
);
self.app_state.num_substeps = num_substeps;
println!(
"Found timestep bounds: {:?}. Estimated substeps: {}. Actual: {}",
bounds, num_substeps_estimated, num_substeps
);
} else if self.app_state.num_substeps != self.app_state.max_num_substeps {
// No adaptive stepping, but we need to update the number of substeps on the gpu.
self.app_state.num_substeps = self.app_state.max_num_substeps;
}
if prev_num_substeps != self.app_state.num_substeps {
let gravity = physics.data.gravity;
let params = SimulationParams {
gravity,
dt: base_dt / self.app_state.num_substeps as f32,
#[cfg(feature = "dim2")]
padding: 0.0,
};
println!("Updated GPU sim params to: {:?}", params);
let gpu_params = physics.data.sim_params.params.buffer_mut();
self.gpu.write_buffer(gpu_params, 0, &[params]).unwrap();
}
let t_encoding = web_time::Instant::now();
let mut encoder = self.gpu.begin_encoding();
// Send updated bodies information to the gpu.
// PERF: don’t reallocate the buffers at each step.
// let poses_data: Vec<GpuSim> = physics
// .data
// .coupling()
// .iter()
// .map(|coupling| {
// let c = &physics.rapier_data.colliders[coupling.collider];
// #[cfg(feature = "dim2")]
// return (*c.position()).into();
// #[cfg(feature = "dim3")]
// return GpuSim::from_isometry(*c.position(), 1.0);
// })
// .collect();
// println!("poses: {:?}", poses_data);
// compute_queue.write_buffer(
// physics.data.bodies.poses().buffer(),
// 0,
// bytemuck::cast_slice(&poses_data),
// );
// let divisor = 1.0; // self.app_state.num_substeps as f32;
// let gravity = Vector::y() * -9.81;
// let vels_data: Vec<_> = physics
// .data
// .coupling()
// .iter()
// .map(|coupling| {
// let rb = &physics.rapier_data.bodies[coupling.body];
// GpuVelocity {
// linear: *rb.linvel()
// + gravity * physics.rapier_data.params.dt * (rb.is_dynamic() as u32 as f32)
// / (self.app_state.num_substeps as f32),
// #[allow(clippy::clone_on_copy)] // Needed for the 2d/3d switch.
// angular: rb.angvel().clone(),
// }
// })
// .collect();
//
// let mut vels_bytes = vec![];
// let mut buffer = StorageBuffer::new(&mut vels_bytes);
// buffer.write(&vels_data).unwrap();
// compute_queue.write_buffer(physics.data.bodies.vels().buffer(), 0, &vels_bytes);
//// Step the simulation.
let mut no_state = Box::new(());
let hooks_state = physics.hooks_state.as_deref_mut().unwrap_or(&mut no_state);
#[cfg(feature = "webgpu")]
let mut timestamps = slang_hal::GpuTimestamps::new(
self.gpu.device(),
self.gpu.queue(),
self.app_state.num_substeps * 10,
);
for _ in 0..self.app_state.num_substeps {
self.app_state
.pipeline
.launch_step(
&self.gpu,
&mut encoder,
&mut physics.data,
&mut *self.hooks,
hooks_state,
#[cfg(feature = "webgpu")]
Some(&mut timestamps),
#[cfg(not(feature = "webgpu"))]
None,
)
.await
.unwrap();
}
#[cfg(feature = "webgpu")]
timestamps.resolve(&mut encoder);
// physics
// .data
// .poses_staging
// .copy_from(&mut encoder, physics.data.bodies.poses());
// physics
// .data
// .particles_pos_staging
// .copy_from(&mut encoder, &physics.data.particles.positions);
// Prepare the vertex buffer for rendering the particles.
/*
if let Ok(instances_buffer) = particles.get_single() {
queue.clear();
self.app_state.prep_vertex_buffer.queue(
&mut queue,
&self.app_state.gpu_render_config,
&physics.data.particles,
&physics.data.rigid_particles,
&physics.data.grid,
&physics.data.sim_params,
&instances_buffer.buffer.buffer,
rigid_particles
.get_single()
.ok()
.map(|b| &**b.buffer.buffer),
);
queue.encode(&mut encoder, None);
}
*/
// Submit.
self.readback_shader
.launch(
&self.gpu,
&mut encoder,
&mut self.readback,
&physics.data.sim_params,
&physics.data.grid,
&physics.data.particles,
)
.unwrap();
self.gpu.submit(encoder).unwrap();
let t_encoding = t_encoding.elapsed().as_secs_f32() * 1000.0;
self.gpu.synchronize().unwrap();
let t_total_step = t_total.elapsed().as_secs_f32() * 1000.0;
// TODO: reuse the `physics.data.particles_pos_staging` buffer.
let t_readback = web_time::Instant::now();
self.gpu
.read_buffer(
self.readback.instances_staging.buffer(),
self.step_result.instances.as_mut_slice(),
)
.await
.unwrap();
let t_readback = t_readback.elapsed().as_secs_f32() * 1000.0;
// Step rapier to update kinematic bodies.
let rapier = &mut self.physics.rapier_data;
rapier.physics_pipeline.step(
&na::zero(),
&rapier.params,
&mut rapier.islands,
&mut rapier.broad_phase,
&mut rapier.narrow_phase,
&mut rapier.bodies,
&mut rapier.colliders,
&mut rapier.impulse_joints,
&mut rapier.multibody_joints,
&mut rapier.ccd_solver,
&(),
&(),
);
if self.app_state.run_state == RunState::Step {
self.app_state.run_state = RunState::Paused;
}
#[cfg(feature = "webgpu")]
let gpu_passes = timestamps.read_results(self.gpu.device()).await.unwrap_or_default();
self.step_result.timings = SimulationTimes {
total_step_time: t_total_step,
encoding_time: t_encoding,
readback_time: t_readback,
#[cfg(feature = "webgpu")]
gpu_passes,
};
self.step_id += 1;
true
}
}