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
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
// SPDX-License-Identifier: AGPL-3.0-or-later
//! Trace - Sum of diagonal elements - Pure WGSL
//!
//! **Deep Debt Principles**:
//! - ✅ Pure WGSL implementation (GPU-optimized)
//! - ✅ Safe Rust wrapper (no unsafe code)
//! - ✅ Hardware-agnostic via WebGPU
//! - ✅ Complete implementation (production-ready)
//!
//! ## Algorithm
//!
//! Computes the trace of a square matrix:
//! ```text
//! trace(A) = sum of diagonal elements
//! For [[a, b], [c, d]]: trace = a + d
//! ```
use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::Result;
use crate::tensor::Tensor;
/// f64 is the canonical source — math is universal, precision is silicon.
const SHADER_F64: &str = include_str!("../shaders/linalg/trace_f64.wgsl");
/// f32 variant derived from f64 via precision downcast.
static SHADER_F32: std::sync::LazyLock<String> =
std::sync::LazyLock::new(|| SHADER_F64.to_string());
/// Sum of diagonal elements of a square matrix.
pub struct Trace {
input: Tensor,
}
impl Trace {
/// Create a trace operation for the given square matrix.
#[must_use]
pub fn new(input: Tensor) -> Self {
Self { input }
}
fn wgsl_shader() -> &'static str {
&SHADER_F32
}
/// Execute trace (sum of diagonal elements) on a square matrix.
/// # 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> {
let device = self.input.device();
let shape = self.input.shape();
// Expect 2D square matrix
if shape.len() != 2 || shape[0] != shape[1] {
return Err(crate::error::BarracudaError::InvalidShape {
expected: vec![0, 0],
actual: shape.to_vec(),
});
}
let n = shape[0];
// Deep Debt Evolution: Capability-based dispatch
let caps = DeviceCapabilities::from_device(device);
let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::Reduction);
let workgroups = (n as u32).div_ceil(optimal_wg_size);
// Output buffer: single element for final result, or partial results if multi-workgroup
let output_size = if workgroups > 1 {
workgroups as usize
} else {
1
};
let output_buffer = device.create_buffer_f32(output_size)?;
let params_buffer = device.create_uniform_buffer("Params", &[n as u32, 0u32, 0u32, 0u32]);
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Trace BGL"),
entries: &[
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,
},
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,
},
wgpu::BindGroupLayoutEntry {
binding: 2,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("Trace BG"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.input.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: params_buffer.as_entire_binding(),
},
],
});
let shader = device.compile_shader(Self::wgsl_shader(), Some("Trace"));
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Trace PL"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Trace Pipeline"),
layout: Some(&pipeline_layout),
module: &shader,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("Trace Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Trace Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.set_bind_group(0, Some(&bind_group), &[]);
pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
// If multiple workgroups, reduce partial results in a second pass using reduce shader
let final_buffer = if workgroups > 1 {
// Second pass: reduce partial results using reduce shader
let reduce_shader_source = crate::ops::reduce::Reduce::wgsl_shader();
let reduce_shader =
device.compile_shader(reduce_shader_source, Some("Trace Reduce Shader"));
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct ReduceParams {
size: u32,
operation: u32, // 0 = Sum
_pad0: u32,
_pad1: u32,
}
let reduce_params = ReduceParams {
size: workgroups,
operation: 0u32, // Sum operation
_pad0: 0,
_pad1: 0,
};
let final_output_buffer = device.create_buffer_f32(1)?;
let reduce_params_buffer =
device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("Trace Reduce Params"),
contents: bytemuck::cast_slice(&[reduce_params]),
usage: wgpu::BufferUsages::UNIFORM,
});
let bind_group_layout_2 =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Trace Reduce BGL"),
entries: &[
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,
},
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,
},
wgpu::BindGroupLayoutEntry {
binding: 2,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
let bind_group_2 = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("Trace Reduce BG"),
layout: &bind_group_layout_2,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: final_output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: reduce_params_buffer.as_entire_binding(),
},
],
});
let pipeline_layout_2 =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Trace Reduce PL"),
bind_group_layouts: &[&bind_group_layout_2],
immediate_size: 0,
});
let pipeline_2 =
device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Trace Reduce Pipeline"),
layout: Some(&pipeline_layout_2),
module: &reduce_shader,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
let mut encoder_2 = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("Trace Reduce Encoder"),
});
{
let mut pass_2 = encoder_2.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Trace Reduce Pass"),
timestamp_writes: None,
});
pass_2.set_pipeline(&pipeline_2);
pass_2.set_bind_group(0, Some(&bind_group_2), &[]);
// Deep Debt Evolution: Capability-based dispatch for reduction pass
let caps_2 = DeviceCapabilities::from_device(device);
let optimal_wg_size_2 = caps_2.optimal_workgroup_size(WorkloadType::Reduction);
let workgroups_2 = workgroups.div_ceil(optimal_wg_size_2);
pass_2.dispatch_workgroups(workgroups_2.max(1), 1, 1);
}
device.submit_commands(Some(encoder_2.finish()));
final_output_buffer
} else {
output_buffer
};
// Return scalar tensor with trace result
Ok(Tensor::from_buffer(final_buffer, vec![1], device.clone()))
}
}
impl Tensor {
/// Compute the trace (sum of diagonal elements) of this square matrix.
/// # Errors
/// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
/// readback fails (e.g. device lost or out of memory).
pub fn trace_wgsl(self) -> Result<Self> {
Trace::new(self).execute()
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_trace_2x2() {
let device = crate::device::test_pool::get_test_device().await;
let input_data = vec![1.0, 2.0, 3.0, 4.0];
let input = Tensor::from_vec_on(input_data, vec![2, 2], device)
.await
.unwrap();
let result = input.trace_wgsl().unwrap();
let trace_result = result.to_vec().unwrap();
// Result should be scalar tensor [trace_value]
assert_eq!(trace_result.len(), 1);
// Trace = 1.0 + 4.0 = 5.0
assert!((trace_result[0] - 5.0).abs() < 1e-5);
}
#[tokio::test]
async fn test_trace_3x3() {
let device = crate::device::test_pool::get_test_device().await;
// Matrix: [[1, 2, 3], [4, 5, 6], [7, 8, 9]]
// Diagonal: [1, 5, 9], trace = 15
let input_data = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0];
let input = Tensor::from_vec_on(input_data, vec![3, 3], device)
.await
.unwrap();
let result = input.trace_wgsl().unwrap();
let trace_result = result.to_vec().unwrap();
assert_eq!(trace_result.len(), 1);
assert!((trace_result[0] - 15.0).abs() < 1e-5);
}
#[tokio::test]
async fn test_trace_large_matrix() {
let device = crate::device::test_pool::get_test_device().await;
let n = 512; // Larger than workgroup size to test multi-workgroup reduction
let mut input_data = vec![0.0; n * n];
// Fill diagonal with sequential values: 1, 2, 3, ..., n
for i in 0..n {
input_data[i * n + i] = (i + 1) as f32;
}
let input = Tensor::from_vec_on(input_data, vec![n, n], device)
.await
.unwrap();
let result = input.trace_wgsl().unwrap();
let trace_result = result.to_vec().unwrap();
assert_eq!(trace_result.len(), 1);
// Sum of 1..n = n*(n+1)/2
let expected = (n * (n + 1) / 2) as f32;
assert!((trace_result[0] - expected).abs() < 1e-4);
}
}