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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! GPU compute operations for `AdamW` Optimizer
//!
//! This module contains the single-pass GPU execution for `AdamW` optimizer
//! with decoupled weight decay.
use super::{AdamW, AdamWParams};
use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::Result;
use crate::tensor::Tensor;
impl AdamW {
/// Execute `AdamW` optimizer step (GPU single-pass with decoupled weight decay)
///
/// **Deep Debt**: Efficient single-pass update with decoupled weight decay
///
/// Returns: (`new_params`, `new_m`, `new_v`)
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
/// readback fails (e.g. device lost or out of memory).
pub fn execute(self) -> Result<(Tensor, Tensor, Tensor)> {
let device = self.params().device();
let size = self.params().len();
// Create parameters
let params = AdamWParams {
num_params: size as u32,
learning_rate: self.learning_rate(),
beta1: self.beta1(),
beta2: self.beta2(),
epsilon: self.epsilon(),
weight_decay: self.weight_decay(),
step: self.step(),
};
let params_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
label: Some("AdamW Params"),
size: std::mem::size_of::<AdamWParams>() as u64,
usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
mapped_at_creation: false,
});
device
.queue
.write_buffer(¶ms_buffer, 0, bytemuck::bytes_of(¶ms));
// Output buffers
let params_out_buffer = device.create_buffer_f32(size)?;
let m_out_buffer = device.create_buffer_f32(size)?;
let v_out_buffer = device.create_buffer_f32(size)?;
// Copy initial params, m, v to output buffers (will be updated in-place)
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("AdamW Copy Encoder"),
});
encoder.copy_buffer_to_buffer(
self.params().buffer(),
0,
¶ms_out_buffer,
0,
(size * std::mem::size_of::<f32>()) as u64,
);
encoder.copy_buffer_to_buffer(
self.m().buffer(),
0,
&m_out_buffer,
0,
(size * std::mem::size_of::<f32>()) as u64,
);
encoder.copy_buffer_to_buffer(
self.v().buffer(),
0,
&v_out_buffer,
0,
(size * std::mem::size_of::<f32>()) as u64,
);
device.submit_commands(Some(encoder.finish()));
// Compile shader
let shader = device.compile_shader(Self::shader(), Some("AdamW"));
// Create bind group layout (5 bindings)
let bgl = device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("AdamW BGL"),
entries: &[
// gradients (read)
wgpu::BindGroupLayoutEntry {
binding: 0,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Storage { read_only: true },
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
// params (read_write)
wgpu::BindGroupLayoutEntry {
binding: 1,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Storage { read_only: false },
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
// m (read_write)
wgpu::BindGroupLayoutEntry {
binding: 2,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Storage { read_only: false },
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
// v (read_write)
wgpu::BindGroupLayoutEntry {
binding: 3,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Storage { read_only: false },
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
// params (uniform)
wgpu::BindGroupLayoutEntry {
binding: 4,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
// Create bind group
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("AdamW BG"),
layout: &bgl,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.gradients().buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: params_out_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: m_out_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: v_out_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 4,
resource: params_buffer.as_entire_binding(),
},
],
});
// Create pipeline
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("AdamW Pipeline Layout"),
bind_group_layouts: &[&bgl],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("AdamW Pipeline"),
layout: Some(&pipeline_layout),
module: &shader,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
// Execute
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("AdamW Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("AdamW Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.set_bind_group(0, Some(&bind_group), &[]);
// Deep Debt Evolution: Capability-based dispatch
let caps = DeviceCapabilities::from_device(device);
let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::ElementWise);
let workgroups = (size as u32).div_ceil(optimal_wg_size);
pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
// Return all three outputs
Ok((
Tensor::from_buffer(
params_out_buffer,
self.params().shape().to_vec(),
device.clone(),
),
Tensor::from_buffer(m_out_buffer, self.m().shape().to_vec(), device.clone()),
Tensor::from_buffer(v_out_buffer, self.v().shape().to_vec(), device.clone()),
))
}
}