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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! Cumsum F64 - Cumulative sum along a dimension (double precision)
//!
//! GPU-accelerated prefix sum / cumulative sum for f64 arrays.
//! Useful for numerical integration and cumulative operations.
//!
//! # Algorithm
//!
//! Per-thread sequential scan: each thread handles one (outer, inner) coordinate
//! pair and scans along the specified dimension.
//!
//! # Example
//!
//! ```ignore
//! use barracuda::ops::CumsumF64;
//! use barracuda::prelude::Tensor;
//!
//! // 1D cumsum: [1, 2, 3, 4] → [1, 3, 6, 10]
//! let result = CumsumF64::execute_1d(&device, &[1.0, 2.0, 3.0, 4.0]).await?;
//!
//! // 2D cumsum along axis 0
//! let tensor = Tensor::from_f64_data(&data, vec![3, 4], device.clone())?;
//! let result = CumsumF64::new(tensor, 0).execute().await?;
//! ```
use crate::device::WgpuDevice;
use crate::device::capabilities::WORKGROUP_SIZE_1D;
use crate::device::compute_pipeline::ComputeDispatch;
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
use bytemuck::{Pod, Zeroable};
use std::sync::Arc;
use wgpu;
/// Parameters passed to the `cumsum_f64` shader
#[repr(C)]
#[derive(Debug, Clone, Copy, Pod, Zeroable)]
struct CumsumF64Params {
size: u32, // Total number of elements
dim_size: u32, // Size of the dimension to scan along
outer_size: u32, // Product of dimensions before dim
inner_size: u32, // Product of dimensions after dim
}
/// F64 cumulative sum operation
pub struct CumsumF64 {
input: Tensor,
dim: usize,
}
impl CumsumF64 {
/// Create a new cumsum operation along the specified dimension
#[must_use]
pub fn new(input: Tensor, dim: usize) -> Self {
Self { input, dim }
}
/// WGSL shader source for f64 cumsum
fn shader() -> &'static str {
include_str!("../shaders/reduce/cumsum_f64.wgsl")
}
/// Execute the cumsum operation
/// # 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().to_vec();
let n_dims = shape.len();
// Validate dimension
if self.dim >= n_dims {
return Err(BarracudaError::InvalidInput {
message: format!(
"Dimension {} out of bounds for tensor with {} dimensions",
self.dim, n_dims
),
});
}
let size: usize = shape.iter().product();
let dim_size = shape[self.dim];
// Compute outer_size (product of dims before dim) and inner_size (product after)
let outer_size: usize = shape[..self.dim].iter().product();
let inner_size: usize = shape[self.dim + 1..].iter().product();
// Handle edge cases
if outer_size == 0 {
// No outer dimensions - treat as 1
return self.execute_with_params(size, dim_size, 1, inner_size.max(1));
}
self.execute_with_params(size, dim_size, outer_size, inner_size.max(1))
}
fn execute_with_params(
self,
size: usize,
dim_size: usize,
outer_size: usize,
inner_size: usize,
) -> Result<Tensor> {
let device = self.input.device();
// Create output buffer
let output_buffer = device.device().create_buffer(&wgpu::BufferDescriptor {
label: Some("CumsumF64 Output"),
size: (size * std::mem::size_of::<f64>()) as u64,
usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_SRC,
mapped_at_creation: false,
});
// Create params buffer
let params = CumsumF64Params {
size: size as u32,
dim_size: dim_size as u32,
outer_size: outer_size as u32,
inner_size: inner_size as u32,
};
let params_buffer = device.create_uniform_buffer("CumsumF64 Params", ¶ms);
let total_pairs = (outer_size * inner_size) as u32;
let workgroups = total_pairs.div_ceil(WORKGROUP_SIZE_1D);
ComputeDispatch::new(device.as_ref(), "CumsumF64")
.shader(Self::shader(), "main")
.f64()
.storage_read(0, self.input.buffer())
.storage_rw(1, &output_buffer)
.uniform(2, ¶ms_buffer)
.dispatch(workgroups, 1, 1)
.submit()?;
// Create output tensor
Ok(Tensor::from_buffer(
output_buffer,
self.input.shape().to_vec(),
device.clone(),
))
}
/// Execute 1D cumsum directly on a slice (convenience method)
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
/// readback fails (e.g. device lost or out of memory).
pub async fn execute_1d(device: &Arc<WgpuDevice>, data: &[f64]) -> Result<Vec<f64>> {
let n = data.len();
if n == 0 {
return Ok(Vec::new());
}
let tensor = Tensor::from_f64_data(data, vec![n], device.clone())?;
let result = Self::new(tensor, 0).execute()?;
result.to_f64_vec()
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::device::test_pool::get_test_device_if_f64_gpu_available;
#[tokio::test]
async fn test_cumsum_f64_1d() {
let Some(device) = get_test_device_if_f64_gpu_available().await else {
return;
};
let input = [1.0, 2.0, 3.0, 4.0, 5.0];
let result = CumsumF64::execute_1d(&device, &input).await.unwrap();
// Expected: [1, 3, 6, 10, 15]
let expected = [1.0, 3.0, 6.0, 10.0, 15.0];
for (i, (&got, &exp)) in result.iter().zip(expected.iter()).enumerate() {
assert!(
(got - exp).abs() < 1e-10,
"Mismatch at {i}: got {got}, expected {exp}"
);
}
}
#[tokio::test]
async fn test_cumsum_f64_large() {
let Some(device) = get_test_device_if_f64_gpu_available().await else {
return;
};
// Test with larger input
let n = 10_000;
let input: Vec<f64> = (1..=n).map(|x| x as f64).collect();
let result = CumsumF64::execute_1d(&device, &input).await.unwrap();
// Verify final element: sum(1..=n) = n*(n+1)/2
let expected_final = (n * (n + 1) / 2) as f64;
assert!(
(result[n - 1] - expected_final).abs() < 1e-6,
"Final cumsum mismatch: got {}, expected {}",
result[n - 1],
expected_final
);
// Verify a few intermediate values
assert!((result[0] - 1.0).abs() < 1e-10);
assert!((result[99] - 5050.0).abs() < 1e-6); // sum(1..=100)
}
#[tokio::test]
async fn test_cumsum_f64_precision() {
let Some(device) = get_test_device_if_f64_gpu_available().await else {
return;
};
// Test f64 precision with small increments
let n = 1000;
let input: Vec<f64> = vec![1e-10; n];
let result = CumsumF64::execute_1d(&device, &input).await.unwrap();
// Final should be n * 1e-10 = 1e-7
let expected_final = n as f64 * 1e-10;
let error = (result[n - 1] - expected_final).abs() / expected_final;
assert!(
error < 1e-10,
"Precision error too large: {error} (expected {expected_final})"
);
}
}