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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! Histc - Histogram with custom bins (Pure WGSL)
//!
//! Computes histogram of input values into specified bins
//! Uses atomic operations for parallel histogram computation
//!
//! **Deep Debt Principles**:
//! - Pure WGSL implementation (no CPU code)
//! - Safe Rust wrapper (no unsafe code)
//! - Hardware-agnostic via WebGPU
//! - Complete implementation (production-ready)
use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
/// Histogram computation
pub struct Histc {
input: Tensor,
num_bins: usize,
min_val: f32,
max_val: f32,
}
impl Histc {
/// Create histogram with given bin count and value range.
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
/// readback fails (e.g. device lost or out of memory).
pub fn new(input: Tensor, num_bins: usize, min_val: f32, max_val: f32) -> Result<Self> {
if num_bins == 0 {
return Err(BarracudaError::invalid_op(
"histc",
"num_bins must be positive",
));
}
if min_val >= max_val {
return Err(BarracudaError::invalid_op(
"histc",
"min_val must be less than max_val",
));
}
Ok(Self {
input,
num_bins,
min_val,
max_val,
})
}
fn wgsl_shader() -> &'static str {
static SHADER: std::sync::LazyLock<String> =
std::sync::LazyLock::new(|| include_str!("../shaders/misc/histc_f64.wgsl").to_string());
&SHADER
}
/// Execute histogram computation.
///
/// # 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 input_size = self.input.shape().iter().product::<usize>();
// Create atomic histogram buffer (zero-initialized)
let histogram_buffer = device.create_buffer_u32_zeros(self.num_bins)?;
// Create uniform buffer
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct Params {
size: u32,
num_bins: u32,
min_val: f32,
max_val: f32,
bin_width: f32,
_pad1: u32,
_pad2: u32,
_pad3: u32,
}
let bin_width = (self.max_val - self.min_val) / self.num_bins as f32;
let params = Params {
size: input_size as u32,
num_bins: self.num_bins as u32,
min_val: self.min_val,
max_val: self.max_val,
bin_width,
_pad1: 0,
_pad2: 0,
_pad3: 0,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("Histc Params"),
contents: bytemuck::cast_slice(&[params]),
usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
});
// Compile shader
let shader_module = device.compile_shader(Self::wgsl_shader(), Some("Histc Shader"));
// Create bind group layout
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("Histc Bind Group Layout"),
entries: &[
wgpu::BindGroupLayoutEntry {
binding: 0,
visibility: wgpu::ShaderStages::COMPUTE,
ty: wgpu::BindingType::Buffer {
ty: wgpu::BufferBindingType::Uniform,
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,
},
],
});
// Create bind group
let bind_group = device.device.create_bind_group(&wgpu::BindGroupDescriptor {
label: Some("Histc Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: params_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: self.input.buffer().as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: histogram_buffer.as_entire_binding(),
},
],
});
// Create pipeline layout
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("Histc Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
// Create pipeline
let pipeline = device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("Histc Pipeline"),
layout: Some(&pipeline_layout),
module: &shader_module,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
// Encode and execute
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("Histc Encoder"),
});
{
let mut pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("Histc Pass"),
timestamp_writes: None,
});
pass.set_pipeline(&pipeline);
pass.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::Reduction);
let workgroups = (input_size as u32).div_ceil(optimal_wg_size);
pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
// Tensor type is f32 — create an f32 buffer for the histogram output.
// The u32 atomic histogram is converted to f32 via GPU readback.
// A u32 → f32 copy shader would avoid the readback but adds dispatch
// overhead that only pays off at >10k bins.
let histogram_f32_buffer = device.create_buffer_f32(self.num_bins)?;
Ok(Tensor::from_buffer(
histogram_f32_buffer,
vec![self.num_bins],
device.clone(),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_histc_basic() {
let device = crate::device::test_pool::get_test_device().await;
let input =
Tensor::from_vec_on(vec![0.0, 1.0, 2.0, 3.0, 4.0, 5.0], vec![6], device.clone())
.await
.unwrap();
let histc = Histc::new(input, 10, 0.0, 10.0).unwrap();
let output = histc.execute().unwrap();
assert_eq!(output.shape(), &[10]);
}
#[tokio::test]
async fn test_histc_large_batch() {
let device = crate::device::test_pool::get_test_device().await;
let size = 1000;
let input = Tensor::from_vec_on(vec![1.0; size], vec![size], device.clone())
.await
.unwrap();
let histc = Histc::new(input, 20, 0.0, 2.0).unwrap();
let output = histc.execute().unwrap();
assert_eq!(output.shape(), &[20]);
}
}