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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! Diagonal matrix operations - Pure WGSL implementation
//!
//! **Deep Debt Principles**:
//! - ✅ Pure WGSL implementation
//! - ✅ Safe Rust wrapper
//! - ✅ Two modes: extract diagonal or create diagonal matrix
//!
//! ## Usage
//!
//! ```rust,ignore
//! // Extract diagonal from matrix
//! let matrix = Tensor::from_vec(vec![1.0, 2.0, 3.0, 4.0], vec![2, 2]).await?;
//! let diagonal = matrix.diag_extract()?; // Returns [1.0, 4.0]
//!
//! // Create diagonal matrix from vector
//! let vector = Tensor::from_vec(vec![1.0, 2.0, 3.0], vec![3]).await?;
//! let matrix = vector.diag_create()?; // Returns 3x3 matrix with diagonal [1, 2, 3]
//! ```
use crate::device::compute_pipeline::ComputeDispatch;
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
/// f64 is the canonical source — math is universal, precision is silicon.
const SHADER_F64: &str = include_str!("../shaders/linalg/diag_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());
#[repr(C)]
#[derive(Copy, Clone, Debug, bytemuck::Pod, bytemuck::Zeroable)]
struct DiagParams {
size: u32,
output_size: u32, // elements to process: size for extract, size*size for create
mode: u32, // 0 = extract, 1 = create
_pad: u32,
}
/// Diagonal extract or create operation.
pub struct Diag {
input: Tensor,
mode: DiagMode,
}
/// Mode for diagonal operation.
#[derive(Debug, Clone, Copy)]
pub enum DiagMode {
/// Extract diagonal from matrix to vector.
Extract,
/// Create diagonal matrix from vector.
Create,
}
impl Diag {
/// Create a new diagonal operation.
/// # Errors
/// Returns [`Err`] if extract mode requires non-square matrix, or create
/// mode requires non-1D input.
pub fn new(input: Tensor, mode: DiagMode) -> Result<Self> {
let shape = input.shape();
match mode {
DiagMode::Extract => {
// Must be square matrix
if shape.len() < 2 {
return Err(BarracudaError::invalid_op(
"diag",
"Extract mode requires 2D matrix",
));
}
let rows = shape[shape.len() - 2];
let cols = shape[shape.len() - 1];
if rows != cols {
return Err(BarracudaError::invalid_op(
"diag",
format!("Extract mode requires square matrix, got {rows}x{cols}"),
));
}
}
DiagMode::Create => {
// Must be 1D vector
if shape.len() != 1 {
return Err(BarracudaError::invalid_op(
"diag",
"Create mode requires 1D vector",
));
}
}
}
Ok(Self { input, mode })
}
fn wgsl_shader() -> &'static str {
&SHADER_F32
}
/// Execute the diagonal operation and return the output tensor.
/// # 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();
let (size, output_size, output_shape) = match self.mode {
DiagMode::Extract => {
let n = shape[shape.len() - 1];
(n, n, vec![n])
}
DiagMode::Create => {
let n = shape[0];
(n, n * n, vec![n, n])
}
};
let output_buffer = device.create_buffer_f32(output_size)?;
let params = DiagParams {
size: size as u32,
output_size: output_size as u32,
mode: match self.mode {
DiagMode::Extract => 0,
DiagMode::Create => 1,
},
_pad: 0,
};
let params_buffer = device.create_uniform_buffer("Diag Params", ¶ms);
let dispatch_size = match self.mode {
DiagMode::Extract => size,
DiagMode::Create => output_size,
};
ComputeDispatch::new(device, "Diag")
.shader(Self::wgsl_shader(), "main")
.storage_read(0, self.input.buffer())
.storage_rw(1, &output_buffer)
.uniform(2, ¶ms_buffer)
.dispatch_1d(dispatch_size as u32)
.submit()?;
let output_data = crate::utils::read_buffer(device, &output_buffer, output_size)?;
Ok(Tensor::new(output_data, output_shape, device.clone()))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_diag_extract() {
let device = crate::device::test_pool::get_test_device().await;
// Matrix: [[1, 2], [3, 4]]
let matrix = Tensor::from_vec_on(vec![1.0, 2.0, 3.0, 4.0], vec![2, 2], device)
.await
.unwrap();
let diag = Diag::new(matrix, DiagMode::Extract)
.unwrap()
.execute()
.unwrap();
let result = diag.to_vec().unwrap();
assert_eq!(result, vec![1.0, 4.0]);
}
#[tokio::test]
async fn test_diag_create() {
let device = crate::device::test_pool::get_test_device().await;
let vector = Tensor::from_vec_on(vec![1.0, 2.0, 3.0], vec![3], device)
.await
.unwrap();
let matrix = Diag::new(vector, DiagMode::Create)
.unwrap()
.execute()
.unwrap();
let result = matrix.to_vec().unwrap();
// Should be: [[1, 0, 0], [0, 2, 0], [0, 0, 3]]
let expected = vec![1.0, 0.0, 0.0, 0.0, 2.0, 0.0, 0.0, 0.0, 3.0];
for (r, e) in result.iter().zip(expected.iter()) {
assert!((r - e).abs() < 1e-6);
}
}
#[tokio::test]
async fn test_diag_extract_3x3() {
let device = crate::device::test_pool::get_test_device().await;
let matrix = Tensor::from_vec_on(
vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0],
vec![3, 3],
device,
)
.await
.unwrap();
let diag = Diag::new(matrix, DiagMode::Extract)
.unwrap()
.execute()
.unwrap();
let result = diag.to_vec().unwrap();
assert_eq!(result, vec![1.0, 5.0, 9.0]);
}
}