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
// SPDX-License-Identifier: AGPL-3.0-or-later
//! MFCC - Mel-Frequency Cepstral Coefficients
//!
//! Extracts MFCC features from audio.
//! Standard features for speech recognition.
//!
//! Deep Debt Principles:
//! - Self-knowledge: Operation knows its computation
//! - Zero hardcoding: Hardware-agnostic implementation
//! - Modern idiomatic Rust: Safe, zero unsafe code
//! - Complete implementation: Production-ready, no mocks
//! - Hardware-agnostic: Pure WGSL for universal compute
use crate::device::{DeviceCapabilities, WorkloadType};
use crate::error::{BarracudaError, Result};
use crate::tensor::Tensor;
/// f64 is the canonical source — math is universal, precision is silicon.
static SHADER_F64: &str = include_str!("../shaders/audio/mfcc_f64.wgsl");
static SHADER_F32: std::sync::LazyLock<String> =
std::sync::LazyLock::new(|| SHADER_F64.to_string());
/// MFCC operation
pub struct MFCC {
mel_spectrogram: Tensor,
n_frames: usize,
n_mels: usize,
n_mfcc: usize,
}
impl MFCC {
/// Create a new MFCC operation
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation, GPU dispatch, or buffer
/// readback fails (e.g. device lost or out of memory).
pub fn new(
mel_spectrogram: Tensor,
n_frames: usize,
n_mels: usize,
n_mfcc: usize,
) -> Result<Self> {
if n_mfcc > n_mels {
return Err(BarracudaError::InvalidInput {
message: format!("n_mfcc ({n_mfcc}) cannot exceed n_mels ({n_mels})"),
});
}
let mel_size: usize = mel_spectrogram.shape().iter().product();
if mel_size != n_frames * n_mels {
return Err(BarracudaError::InvalidInput {
message: format!(
"Mel spectrogram size ({}) must equal n_frames * n_mels ({})",
mel_size,
n_frames * n_mels
),
});
}
Ok(Self {
mel_spectrogram,
n_frames,
n_mels,
n_mfcc,
})
}
/// Get the WGSL shader source
fn wgsl_shader() -> &'static str {
&SHADER_F32
}
/// Execute the MFCC 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.mel_spectrogram.device();
let output_size = self.n_frames * self.n_mfcc;
// Access input buffer directly (zero-copy)
let mel_buffer = self.mel_spectrogram.buffer();
// Create output buffer
let output_buffer = device.create_buffer_f32(output_size)?;
// Create uniform buffer for parameters
#[repr(C)]
#[derive(Copy, Clone, bytemuck::Pod, bytemuck::Zeroable)]
struct Params {
n_frames: u32,
n_mels: u32,
n_mfcc: u32,
}
let params = Params {
n_frames: self.n_frames as u32,
n_mels: self.n_mels as u32,
n_mfcc: self.n_mfcc as u32,
};
let params_buffer = device
.device
.create_buffer_init(&wgpu::util::BufferInitDescriptor {
label: Some("MFCC 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("MFCC Shader"));
// Create bind group layout
let bind_group_layout =
device
.device
.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
label: Some("MFCC 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: false },
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::Uniform,
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("MFCC Bind Group"),
layout: &bind_group_layout,
entries: &[
wgpu::BindGroupEntry {
binding: 0,
resource: mel_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 1,
resource: output_buffer.as_entire_binding(),
},
wgpu::BindGroupEntry {
binding: 2,
resource: params_buffer.as_entire_binding(),
},
],
});
// Create compute pipeline
let pipeline_layout =
device
.device
.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
label: Some("MFCC Pipeline Layout"),
bind_group_layouts: &[&bind_group_layout],
immediate_size: 0,
});
let compute_pipeline =
device
.device
.create_compute_pipeline(&wgpu::ComputePipelineDescriptor {
label: Some("MFCC Pipeline"),
layout: Some(&pipeline_layout),
module: &shader_module,
entry_point: Some("main"),
cache: None,
compilation_options: Default::default(),
});
// Execute compute shader
let mut encoder = device.create_encoder_guarded(&wgpu::CommandEncoderDescriptor {
label: Some("MFCC Encoder"),
});
{
let mut compute_pass = encoder.begin_compute_pass(&wgpu::ComputePassDescriptor {
label: Some("MFCC Pass"),
timestamp_writes: None,
});
compute_pass.set_pipeline(&compute_pipeline);
compute_pass.set_bind_group(0, Some(&bind_group), &[]);
// Deep Debt Evolution: Capability-based dispatch
let caps = DeviceCapabilities::from_device(device);
let size = self.n_frames * self.n_mfcc;
let optimal_wg_size = caps.optimal_workgroup_size(WorkloadType::MatMul);
let workgroups = (size as u32).div_ceil(optimal_wg_size);
compute_pass.dispatch_workgroups(workgroups, 1, 1);
}
device.submit_commands(Some(encoder.finish()));
// Output shape: [n_frames, n_mfcc]
let output_shape = vec![self.n_frames, self.n_mfcc];
// Return tensor without reading back (zero-copy)
Ok(Tensor::from_buffer(
output_buffer,
output_shape,
device.clone(),
))
}
}
#[cfg(test)]
mod tests {
use super::*;
#[tokio::test]
async fn test_mfcc_basic() {
let device = crate::device::test_pool::get_test_device().await;
let mel_spec = Tensor::from_vec_on(vec![1.0; 100 * 80], vec![100, 80], device.clone())
.await
.unwrap();
let mfcc_features = MFCC::new(mel_spec, 100, 80, 13).unwrap().execute().unwrap();
assert_eq!(mfcc_features.shape(), &[100, 13]);
}
}