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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! FHE Polynomial Subtraction Operation
//!
//! **Purpose**: Subtract two FHE ciphertext polynomials on GPU
//!
//! **Deep Debt Compliance**:
//! - ✅ Pure Rust + WGSL (no unsafe)
//! - ✅ Hardware-agnostic (wgpu backend selection)
//! - ✅ Numerically precise (modular subtraction)
//! - ✅ Production-ready (full error handling)
//! - ✅ Canonical pattern: Tensor inputs/outputs, device from runtime
use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
use std::sync::Arc;
/// FHE polynomial subtraction operation
///
/// Subtracts two polynomials coefficient-wise with modular reduction.
///
/// ## Mathematical Operation
///
/// Given polynomials a(X) and b(X) over `Z_q`[X]/(X^N + 1):
/// ```text
/// result(X) = a(X) - b(X) mod q
/// ```
///
/// Where each coefficient is reduced modulo q.
pub struct FhePolySub {
poly_a: Tensor,
poly_b: Tensor,
pipeline: wgpu::ComputePipeline,
bind_group_layout: wgpu::BindGroupLayout,
degree: u32,
modulus: u64,
}
impl FhePolySub {
/// Create a new FHE polynomial subtraction operation
/// ## Parameters
/// - `poly_a`: First polynomial tensor (u32 pairs representing u64 coefficients)
/// - `poly_b`: Second polynomial tensor (u32 pairs representing u64 coefficients)
/// - `degree`: Polynomial degree (N), typically 2048, 4096, or 8192
/// - `modulus`: Modulus q (large prime, e.g., 2^60)
/// # Errors
/// Returns [`Err`] if polynomial lengths do not match `degree*2`, tensors are on different
/// devices, or modulus is zero.
pub fn new(poly_a: Tensor, poly_b: Tensor, degree: u32, modulus: u64) -> Result<Self> {
// Validate inputs
let expected_size = (degree as usize) * 2; // u32 pairs for u64
if poly_a.len() != expected_size {
return Err(BarracudaError::Device(format!(
"poly_a length {} doesn't match expected {} (degree {} * 2)",
poly_a.len(),
expected_size,
degree
)));
}
if poly_b.len() != expected_size {
return Err(BarracudaError::Device(format!(
"poly_b length {} doesn't match expected {} (degree {} * 2)",
poly_b.len(),
expected_size,
degree
)));
}
// Ensure both tensors are on same device
if !std::ptr::eq(poly_a.device().as_ref(), poly_b.device().as_ref()) {
return Err(BarracudaError::Device(
"poly_a and poly_b must be on the same device".to_string(),
));
}
if modulus == 0 {
return Err(BarracudaError::Device(
"Modulus must be non-zero".to_string(),
));
}
let device = poly_a.device();
// Load WGSL shader
let shader = device.compile_shader(
include_str!("fhe_poly_sub.wgsl"),
Some("FHE Polynomial Subtraction Shader"),
);
// Create bind group layout
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("FHE Poly Sub 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 layout
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("FHE Poly Sub Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
// Create compute pipeline
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("FHE Poly Sub Pipeline"),
layout: Some(&pipeline_layout),
module: &shader,
entry_point: Some("fhe_poly_sub"),
cache: None,
compilation_options: Default::default(),
});
Ok(Self {
poly_a,
poly_b,
pipeline,
bind_group_layout,
degree,
modulus,
})
}
/// Execute polynomial subtraction on GPU
/// ## Returns
/// Result tensor: (`poly_a` - `poly_b`) mod q
/// Data stays on GPU (no CPU readback)
/// # Errors
/// Returns [`Err`] if buffer allocation fails, GPU dispatch fails, or the device is lost.
pub fn execute(self) -> Result<Tensor> {
let device = self.poly_a.device();
// Create output buffer (u32 pairs for u64 coefficients)
let result_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
label: Some("FHE Result Buffer"),
size: (self.degree as u64 * 2 * std::mem::size_of::<u32>() as u64),
usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_SRC,
mapped_at_creation: false,
});
// Create params buffer
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct Params {
degree: u32,
modulus_lo: u32,
modulus_hi: u32,
_padding: [u32; 5],
}
let params = Params {
degree: self.degree,
modulus_lo: self.modulus as u32,
modulus_hi: (self.modulus >> 32) as u32,
_padding: [0; 5],
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("FHE Params Buffer"),
contents: bytemuck::bytes_of(¶ms),
usage: wgpu::BufferUsages::UNIFORM,
});
// Create bind group
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("FHE Poly Sub Bind Group"),
layout: &self.bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: self.poly_a.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.poly_b.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: result_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 3,
resource: params_buffer.as_entire_binding(),
},
],
});
// Execute compute shader
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("FHE Poly Sub Encoder"),
});
{
let mut cpass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("FHE Poly Sub Pass"),
timestamp_writes: None,
});
cpass.set_pipeline(&self.pipeline);
cpass.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::FHE);
let workgroups = self.degree.div_ceil(optimal_wg_size);
cpass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
// Return tensor (data stays on GPU)
Ok(Tensor::from_buffer(
result_buffer,
vec![self.degree as usize * 2], // u32 pairs
device.clone(),
))
}
}
/// Helper: Create FHE polynomial tensor from u64 coefficients
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation or data upload fails (e.g. device lost).
pub async fn create_fhe_poly_tensor(
poly: &[u64],
device: Arc<crate::device::WgpuDevice>,
) -> Result<Tensor> {
let poly_u32: Vec<u32> = poly
.iter()
.flat_map(|&val| vec![val as u32, (val >> 32) as u32])
.collect();
Tensor::from_data_pod(&poly_u32, vec![poly_u32.len()], device)
}
#[cfg(test)]
mod tests {
use super::*;
use crate::ops::fhe_poly_add::create_fhe_poly_tensor;
#[tokio::test]
async fn test_fhe_poly_sub_basic() {
let device = crate::device::test_pool::get_test_device().await;
let degree = 8;
let modulus = 97;
let poly_a_data = vec![50u64, 60, 70, 80, 90, 85, 75, 65];
let poly_b_data = vec![10u64, 20, 30, 40, 50, 60, 70, 80];
let poly_a = create_fhe_poly_tensor(&poly_a_data, device.clone())
.await
.unwrap();
let poly_b = create_fhe_poly_tensor(&poly_b_data, device.clone())
.await
.unwrap();
let op = FhePolySub::new(poly_a, poly_b, degree, modulus).unwrap();
let result_tensor = op.execute().unwrap();
// Read back for testing
let size = result_tensor.len();
let staging_buffer = device.device.create_buffer(&wgpu::BufferDescriptor {
label: Some("Test Staging"),
size: (size * std::mem::size_of::<u32>()) as u64,
usage: wgpu::BufferUsages::MAP_READ | wgpu::BufferUsages::COPY_DST,
mapped_at_creation: false,
});
let mut encoder = device.create_encoder_guarded(&Default::default());
encoder.copy_buffer_to_buffer(
result_tensor.buffer(),
0,
&staging_buffer,
0,
(size * std::mem::size_of::<u32>()) as u64,
);
device.submit_commands(Some(encoder.finish()));
let result_u32: Vec<u32> = device.map_staging_buffer(&staging_buffer, size).unwrap();
let result: Vec<u64> = result_u32
.chunks(2)
.map(|pair| (pair[0] as u64) | ((pair[1] as u64) << 32))
.collect();
let expected: Vec<u64> = vec![40, 40, 40, 40, 40, 25, 5, 82];
assert_eq!(result, expected);
}
}