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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! Complex Addition Operation
//!
//! **Operation**: (a + bi) + (c + di) = (a+c) + (b+d)i
//! **Complexity**: O(1) - trivial (native vec2 addition)
//! **Performance**: 1 SIMD operation
//!
//! **Deep Debt Compliance**:
//! - ✅ Pure WGSL (no unsafe)
//! - ✅ Hardware-agnostic
//! - ✅ Numerically exact (IEEE 754)
use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
/// Complex addition operation
///
/// Adds two complex tensors element-wise.
/// Complex numbers are stored as `vec2<f32>` where:
/// - x component = real part
/// - y component = imaginary part
pub struct ComplexAdd {
input_a: Tensor,
input_b: Tensor,
pipeline: wgpu::ComputePipeline,
bind_group_layout: wgpu::BindGroupLayout,
}
impl ComplexAdd {
/// Create a new complex addition operation
///
/// # Arguments
/// * `input_a` - First complex tensor (shape [..., 2])
/// * `input_b` - Second complex tensor (shape [..., 2])
///
/// Both tensors must:
/// - Have same shape
/// - Last dimension must be 2 (real, imag)
/// - Be on the same device
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
/// readback fails (e.g. device lost or out of memory).
pub fn new(input_a: Tensor, input_b: Tensor) -> Result<Self> {
// Validate tensors
if input_a.shape() != input_b.shape() {
return Err(BarracudaError::Device(
"Complex tensors must have same shape".to_string(),
));
}
let shape = input_a.shape();
if shape.last() != Some(&2) {
return Err(BarracudaError::Device(
"Complex tensors must have last dimension = 2 (real, imag)".to_string(),
));
}
if !std::ptr::eq(input_a.device().as_ref(), input_b.device().as_ref()) {
return Err(BarracudaError::Device(
"Tensors must be on the same device".to_string(),
));
}
let device = input_a.device();
// Load shader
let shader = device.compile_shader(include_str!("add.wgsl"), Some("Complex Add Shader"));
// Create bind group layout
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Complex Add Bind Group Layout"),
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: true },
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::Storage { read_only: false },
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
wgpu::BindGroupLayoutEntry {
binding: 3,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
has_dynamic_offset: false,
min_binding_size: None,
},
count: None,
},
],
});
// Create pipeline
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Complex Add Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Complex Add Pipeline"),
layout: Some(&pipeline_layout),
module: &shader,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
Ok(Self {
input_a,
input_b,
pipeline,
bind_group_layout,
})
}
/// Execute complex addition on GPU
///
/// Returns: New tensor with element-wise complex sum
///
/// # 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_a.device();
let num_elements = self.input_a.len();
// Create output buffer
let output_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
label: Some("Complex Add Output"),
size: (num_elements * std::mem::size_of::<f32>()) as u64,
usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_SRC,
mapped_at_creation: false,
});
// Create params buffer (size)
let params = [num_elements as u32 / 2]; // Divide by 2 (complex pairs)
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("Complex Add Params"),
contents: bytemuck::cast_slice(¶ms),
usage: wgpu::BufferUsages::UNIFORM,
});
// Create bind group
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("Complex Add Bind Group"),
layout: &self.bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.input_a.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.input_b.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: params_buffer.as_entire_binding(),
},
],
});
// Dispatch shader
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("Complex Add Encoder"),
});
{
let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Complex Add Compute Pass"),
timestamp_writes: None,
});
compute_pass.set_pipeline(&self.pipeline);
compute_pass.set_bind_group(0, Some(&bind_group), &[]);
// Capability-based workgroup sizing
let caps = DeviceCapabilities::from_device(device);
let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::ElementWise);
let num_complex = num_elements / 2;
let workgroups = (num_complex as u32).div_ceil(optimal_wg_size);
compute_pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
// Return output tensor
Ok(Tensor::from_buffer(
output_buffer,
self.input_a.shape().to_vec(),
device.clone(),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_complex_add_simple() {
let device = crate::device::test_pool::get_test_device().await;
// (3+4i) + (1+2i) = 4+6i
let data_a = vec![3.0f32, 4.0];
let data_b = vec![1.0f32, 2.0];
let tensor_a = Tensor::from_data(&data_a, vec![1, 2], device.clone()).unwrap();
let tensor_b = Tensor::from_data(&data_b, vec![1, 2], device).unwrap();
let op = ComplexAdd::new(tensor_a, tensor_b).unwrap();
let result = op.execute().unwrap();
let result_data = result.to_vec().unwrap();
assert!((result_data[0] - 4.0).abs() < 1e-6); // Real part
assert!((result_data[1] - 6.0).abs() < 1e-6); // Imag part
}
#[tokio::test]
async fn test_complex_add_batch() {
let device = crate::device::test_pool::get_test_device().await;
// Multiple complex numbers
let data_a = vec![1.0f32, 2.0, 3.0, 4.0, 5.0, 6.0];
let data_b = vec![1.0f32, 1.0, 1.0, 1.0, 1.0, 1.0];
let tensor_a = Tensor::from_data(&data_a, vec![3, 2], device.clone()).unwrap();
let tensor_b = Tensor::from_data(&data_b, vec![3, 2], device).unwrap();
let op = ComplexAdd::new(tensor_a, tensor_b).unwrap();
let result = op.execute().unwrap();
let result_data = result.to_vec().unwrap();
// (1+2i) + (1+1i) = 2+3i
assert!((result_data[0] - 2.0).abs() < 1e-6);
assert!((result_data[1] - 3.0).abs() < 1e-6);
// (3+4i) + (1+1i) = 4+5i
assert!((result_data[2] - 4.0).abs() < 1e-6);
assert!((result_data[3] - 5.0).abs() < 1e-6);
}
}