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
340
341
342
343
344
345
346
347
348
349
350
351
352
353
//! Vector transformation operations
//!
//! This module provides element-wise transformation methods:
//! - `abs()` - Element-wise absolute value
//! - `clamp()` / `clip()` - Clamp values to a range
//! - `lerp()` - Linear interpolation between two vectors
//! - `sqrt()` - Element-wise square root (in `math` submodule)
//! - `recip()` - Element-wise reciprocal (1/x) (in `math` submodule)
//! - `pow()` - Element-wise power (in `math` submodule)
mod math;
#[cfg(target_arch = "x86_64")]
use crate::backends::avx2::Avx2Backend;
#[cfg(any(target_arch = "aarch64", target_arch = "arm"))]
use crate::backends::neon::NeonBackend;
use crate::backends::scalar::ScalarBackend;
#[cfg(target_arch = "x86_64")]
use crate::backends::sse2::Sse2Backend;
#[cfg(target_arch = "wasm32")]
use crate::backends::wasm::WasmBackend;
use crate::backends::VectorBackend;
use crate::{Backend, Result, TruenoError, Vector};
impl Vector<f32> {
/// Compute element-wise absolute value
///
/// Returns a new vector where each element is the absolute value of the corresponding input element.
///
/// # Examples
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let v = Vector::from_slice(&[3.0, -4.0, 5.0, -2.0]);
/// let result = v.abs()?;
///
/// assert_eq!(result.as_slice(), &[3.0, 4.0, 5.0, 2.0]);
/// # Ok(())
/// # }
/// ```
///
/// # Empty Vector
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let v: Vector<f32> = Vector::from_slice(&[]);
/// let result = v.abs()?;
/// assert_eq!(result.len(), 0);
/// # Ok(())
/// # }
/// ```
pub fn abs(&self) -> Result<Vector<f32>> {
// Uninit: backend writes every element before any read.
let n = self.len();
let mut result_data: Vec<f32> = Vec::with_capacity(n);
// SAFETY: Backend writes all elements before any read.
unsafe {
result_data.set_len(n);
}
if !self.as_slice().is_empty() {
// SAFETY: Unsafe block delegates to backend implementation which maintains safety invariants
unsafe {
match self.backend() {
Backend::Scalar => ScalarBackend::abs(self.as_slice(), &mut result_data),
#[cfg(target_arch = "x86_64")]
Backend::SSE2 | Backend::AVX => {
Sse2Backend::abs(self.as_slice(), &mut result_data)
}
#[cfg(target_arch = "x86_64")]
Backend::AVX2 | Backend::AVX512 => {
Avx2Backend::abs(self.as_slice(), &mut result_data)
}
#[cfg(any(target_arch = "aarch64", target_arch = "arm"))]
Backend::NEON => NeonBackend::abs(self.as_slice(), &mut result_data),
#[cfg(target_arch = "wasm32")]
Backend::WasmSIMD => WasmBackend::abs(self.as_slice(), &mut result_data),
Backend::GPU => return Err(TruenoError::UnsupportedBackend(Backend::GPU)),
Backend::Auto => {
return Err(TruenoError::UnsupportedBackend(Backend::Auto));
}
#[cfg(not(target_arch = "x86_64"))]
Backend::SSE2 | Backend::AVX | Backend::AVX2 | Backend::AVX512 => {
ScalarBackend::abs(self.as_slice(), &mut result_data)
}
#[cfg(not(any(target_arch = "aarch64", target_arch = "arm")))]
Backend::NEON => ScalarBackend::abs(self.as_slice(), &mut result_data),
#[cfg(not(target_arch = "wasm32"))]
Backend::WasmSIMD => ScalarBackend::abs(self.as_slice(), &mut result_data),
}
}
}
// Construct directly (no copy) — from_slice_with_backend would copy 4MB!
Ok(Vector { data: result_data, backend: self.backend() })
}
/// Clip values to a specified range [min_val, max_val]
///
/// Constrains each element to be within the specified range:
/// - Values below min_val become min_val
/// - Values above max_val become max_val
/// - Values within range stay unchanged
///
/// This is useful for outlier handling, gradient clipping in neural networks,
/// and ensuring values stay within valid bounds.
///
/// # Examples
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let v = Vector::from_slice(&[-5.0, 0.0, 5.0, 10.0, 15.0]);
/// let clipped = v.clip(0.0, 10.0)?;
///
/// // Values: [-5, 0, 5, 10, 15] → [0, 0, 5, 10, 10]
/// assert_eq!(clipped.as_slice(), &[0.0, 0.0, 5.0, 10.0, 10.0]);
/// # Ok(())
/// # }
/// ```
///
/// # Invalid range
///
/// Returns InvalidInput error if min_val > max_val.
///
/// ```
/// use trueno::{Vector, TruenoError};
///
/// let v = Vector::from_slice(&[1.0, 2.0, 3.0]);
/// let result = v.clip(10.0, 5.0); // min > max
/// assert!(matches!(result, Err(TruenoError::InvalidInput(_))));
/// ```
pub fn clip(&self, min_val: f32, max_val: f32) -> Result<Self> {
if min_val > max_val {
return Err(TruenoError::InvalidInput(format!(
"min_val ({}) must be <= max_val ({})",
min_val, max_val
)));
}
// Scalar fallback: Element-wise clamp
let data: Vec<f32> = self.as_slice().iter().map(|&x| x.max(min_val).min(max_val)).collect();
Ok(Vector::from_vec(data))
}
/// Clamp elements to range [min_val, max_val]
///
/// Returns a new vector where each element is constrained to the specified range.
/// Elements below min_val become min_val, elements above max_val become max_val.
///
/// # Examples
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let v = Vector::from_slice(&[-5.0, 0.0, 5.0, 10.0, 15.0]);
/// let result = v.clamp(0.0, 10.0)?;
///
/// assert_eq!(result.as_slice(), &[0.0, 0.0, 5.0, 10.0, 10.0]);
/// # Ok(())
/// # }
/// ```
///
/// # Negative Range
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let v = Vector::from_slice(&[-10.0, -5.0, 0.0, 5.0]);
/// let result = v.clamp(-8.0, -2.0)?;
/// assert_eq!(result.as_slice(), &[-8.0, -5.0, -2.0, -2.0]);
/// # Ok(())
/// # }
/// ```
///
/// # Errors
///
/// Returns `InvalidInput` if min_val > max_val.
pub fn clamp(&self, min_val: f32, max_val: f32) -> Result<Vector<f32>> {
// Validate range
if min_val > max_val {
return Err(TruenoError::InvalidInput(format!(
"Invalid clamp range: min ({}) > max ({})",
min_val, max_val
)));
}
// Uninit: backend writes every element before any read.
let n = self.len();
let mut result_data: Vec<f32> = Vec::with_capacity(n);
// SAFETY: Backend writes all elements before any read.
unsafe {
result_data.set_len(n);
}
if !self.as_slice().is_empty() {
// SAFETY: Unsafe block delegates to backend implementation which maintains safety invariants
unsafe {
match self.backend() {
Backend::Scalar => {
ScalarBackend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
#[cfg(target_arch = "x86_64")]
Backend::SSE2 | Backend::AVX => {
Sse2Backend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
#[cfg(target_arch = "x86_64")]
Backend::AVX2 | Backend::AVX512 => {
Avx2Backend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
#[cfg(any(target_arch = "aarch64", target_arch = "arm"))]
Backend::NEON => {
NeonBackend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
#[cfg(target_arch = "wasm32")]
Backend::WasmSIMD => {
WasmBackend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
Backend::GPU => return Err(TruenoError::UnsupportedBackend(Backend::GPU)),
Backend::Auto => {
return Err(TruenoError::UnsupportedBackend(Backend::Auto));
}
#[cfg(not(target_arch = "x86_64"))]
Backend::SSE2 | Backend::AVX | Backend::AVX2 | Backend::AVX512 => {
ScalarBackend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
#[cfg(not(any(target_arch = "aarch64", target_arch = "arm")))]
Backend::NEON => {
ScalarBackend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
#[cfg(not(target_arch = "wasm32"))]
Backend::WasmSIMD => {
ScalarBackend::clamp(self.as_slice(), min_val, max_val, &mut result_data)
}
}
}
}
Ok(Vector { data: result_data, backend: self.backend() })
}
/// Linear interpolation between two vectors
///
/// Computes element-wise linear interpolation: `result\[i\] = a\[i\] + t * (b\[i\] - a\[i\])`
///
/// - When `t = 0.0`, returns `self`
/// - When `t = 1.0`, returns `other`
/// - Values outside `[0, 1]` perform extrapolation
///
/// # Examples
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let a = Vector::from_slice(&[0.0, 10.0, 20.0]);
/// let b = Vector::from_slice(&[100.0, 110.0, 120.0]);
/// let result = a.lerp(&b, 0.5)?;
///
/// assert_eq!(result.as_slice(), &[50.0, 60.0, 70.0]);
/// # Ok(())
/// # }
/// ```
///
/// # Extrapolation
///
/// ```
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// use trueno::Vector;
///
/// let a = Vector::from_slice(&[0.0, 10.0]);
/// let b = Vector::from_slice(&[10.0, 20.0]);
///
/// // t > 1.0 extrapolates beyond b
/// let result = a.lerp(&b, 2.0)?;
/// assert_eq!(result.as_slice(), &[20.0, 30.0]);
/// # Ok(())
/// # }
/// ```
///
/// # Errors
///
/// Returns `SizeMismatch` if vectors have different lengths.
pub fn lerp(&self, other: &Vector<f32>, t: f32) -> Result<Vector<f32>> {
if self.len() != other.len() {
return Err(TruenoError::SizeMismatch { expected: self.len(), actual: other.len() });
}
// Uninit: backend writes every element before any read.
let n = self.len();
let mut result_data: Vec<f32> = Vec::with_capacity(n);
// SAFETY: Backend writes all elements before any read.
unsafe {
result_data.set_len(n);
}
if !self.as_slice().is_empty() {
// SAFETY: Unsafe block delegates to backend implementation which maintains safety invariants
unsafe {
match self.backend() {
Backend::Scalar => {
ScalarBackend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
#[cfg(target_arch = "x86_64")]
Backend::SSE2 | Backend::AVX => {
Sse2Backend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
#[cfg(target_arch = "x86_64")]
Backend::AVX2 | Backend::AVX512 => {
Avx2Backend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
#[cfg(any(target_arch = "aarch64", target_arch = "arm"))]
Backend::NEON => {
NeonBackend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
#[cfg(target_arch = "wasm32")]
Backend::WasmSIMD => {
WasmBackend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
Backend::GPU => return Err(TruenoError::UnsupportedBackend(Backend::GPU)),
Backend::Auto => {
return Err(TruenoError::UnsupportedBackend(Backend::Auto));
}
#[cfg(not(target_arch = "x86_64"))]
Backend::SSE2 | Backend::AVX | Backend::AVX2 | Backend::AVX512 => {
ScalarBackend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
#[cfg(not(any(target_arch = "aarch64", target_arch = "arm")))]
Backend::NEON => {
ScalarBackend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
#[cfg(not(target_arch = "wasm32"))]
Backend::WasmSIMD => {
ScalarBackend::lerp(self.as_slice(), other.as_slice(), t, &mut result_data)
}
}
}
}
Ok(Vector { data: result_data, backend: self.backend() })
}
}
#[cfg(test)]
mod tests;