numrs2 0.3.3

A Rust implementation inspired by NumPy for numerical computing (NumRS2)
Documentation 
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

//! Comparison and sign operations using NEON SIMD
//!
//! This module provides optimized abs, sign, maximum, minimum, clamp,
//! and copysign functions for ARM NEON.

use crate::array::Array;

use super::core::{NeonEnhancedOps, NEON_F64_LANES};

#[cfg(target_arch = "aarch64")]
use std::arch::aarch64::*;

// =============================================================================
// NEON f64 Comparison Operations
// =============================================================================

impl NeonEnhancedOps {
    /// NEON vectorized absolute value for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_abs_f64(input: &Array<f64>) -> Array<f64> {
        let data = input.to_vec();
        let mut result = vec![0.0f64; data.len()];
        let len = data.len();
        let simd_len = len & !(NEON_F64_LANES - 1);

        unsafe {
            for i in (0..simd_len).step_by(NEON_F64_LANES) {
                let v = vld1q_f64(data.as_ptr().add(i));
                let abs_v = vabsq_f64(v);
                vst1q_f64(result.as_mut_ptr().add(i), abs_v);
            }
        }

        // Scalar fallback for remaining elements
        for i in simd_len..len {
            result[i] = data[i].abs();
        }

        Array::from_vec(result).reshape(&input.shape())
    }

    /// NEON vectorized sign for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_sign_f64(input: &Array<f64>) -> Array<f64> {
        let data = input.to_vec();
        let mut result = vec![0.0f64; data.len()];
        let len = data.len();
        let simd_len = len & !(NEON_F64_LANES - 1);

        unsafe {
            let zero = vdupq_n_f64(0.0);
            let _one = vdupq_n_f64(1.0);
            let _neg_one = vdupq_n_f64(-1.0);

            for i in (0..simd_len).step_by(NEON_F64_LANES) {
                let v = vld1q_f64(data.as_ptr().add(i));

                // Compare v > 0 and v < 0
                let gt_zero = vcgtq_f64(v, zero);
                let lt_zero = vcltq_f64(v, zero);

                // Select: if v > 0 -> 1, if v < 0 -> -1, else 0
                let pos_mask = vreinterpretq_f64_u64(gt_zero);
                let neg_mask = vreinterpretq_f64_u64(lt_zero);

                let mut temp = [0.0f64; NEON_F64_LANES];
                let mut _temp_pos = [0.0f64; NEON_F64_LANES];
                let mut _temp_neg = [0.0f64; NEON_F64_LANES];
                vst1q_f64(temp.as_mut_ptr(), v);
                vst1q_f64(_temp_pos.as_mut_ptr(), pos_mask);
                vst1q_f64(_temp_neg.as_mut_ptr(), neg_mask);

                for j in 0..NEON_F64_LANES {
                    if temp[j] > 0.0 {
                        result[i + j] = 1.0;
                    } else if temp[j] < 0.0 {
                        result[i + j] = -1.0;
                    } else {
                        result[i + j] = 0.0;
                    }
                }
            }
        }

        for i in simd_len..len {
            result[i] = if data[i] > 0.0 {
                1.0
            } else if data[i] < 0.0 {
                -1.0
            } else {
                0.0
            };
        }

        Array::from_vec(result).reshape(&input.shape())
    }

    /// NEON vectorized element-wise maximum for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_maximum_f64(a: &Array<f64>, b: &Array<f64>) -> Array<f64> {
        let data_a = a.to_vec();
        let data_b = b.to_vec();
        let len = data_a.len().min(data_b.len());
        let mut result = vec![0.0f64; len];
        let simd_len = len & !(NEON_F64_LANES - 1);

        unsafe {
            for i in (0..simd_len).step_by(NEON_F64_LANES) {
                let va = vld1q_f64(data_a.as_ptr().add(i));
                let vb = vld1q_f64(data_b.as_ptr().add(i));
                let vmax = vmaxq_f64(va, vb);
                vst1q_f64(result.as_mut_ptr().add(i), vmax);
            }
        }

        for i in simd_len..len {
            result[i] = data_a[i].max(data_b[i]);
        }

        Array::from_vec(result).reshape(&a.shape())
    }

    /// NEON vectorized element-wise minimum for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_minimum_f64(a: &Array<f64>, b: &Array<f64>) -> Array<f64> {
        let data_a = a.to_vec();
        let data_b = b.to_vec();
        let len = data_a.len().min(data_b.len());
        let mut result = vec![0.0f64; len];
        let simd_len = len & !(NEON_F64_LANES - 1);

        unsafe {
            for i in (0..simd_len).step_by(NEON_F64_LANES) {
                let va = vld1q_f64(data_a.as_ptr().add(i));
                let vb = vld1q_f64(data_b.as_ptr().add(i));
                let vmin = vminq_f64(va, vb);
                vst1q_f64(result.as_mut_ptr().add(i), vmin);
            }
        }

        for i in simd_len..len {
            result[i] = data_a[i].min(data_b[i]);
        }

        Array::from_vec(result).reshape(&a.shape())
    }

    /// NEON vectorized clamp for f64: clamp values to [min_val, max_val]
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_clamp_f64(input: &Array<f64>, min_val: f64, max_val: f64) -> Array<f64> {
        let data = input.to_vec();
        let len = data.len();
        let mut result = vec![0.0f64; len];
        let simd_len = len & !(NEON_F64_LANES - 1);

        unsafe {
            let vmin = vdupq_n_f64(min_val);
            let vmax = vdupq_n_f64(max_val);
            for i in (0..simd_len).step_by(NEON_F64_LANES) {
                let v = vld1q_f64(data.as_ptr().add(i));
                // Clamp: max(min_val, min(max_val, v))
                let v_clamped = vmaxq_f64(vmin, vminq_f64(vmax, v));
                vst1q_f64(result.as_mut_ptr().add(i), v_clamped);
            }
        }

        for i in simd_len..len {
            result[i] = data[i].clamp(min_val, max_val);
        }

        Array::from_vec(result).reshape(&input.shape())
    }

    /// NEON vectorized copysign for f64: magnitude from a, sign from b
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_copysign_f64(magnitude: &Array<f64>, sign: &Array<f64>) -> Array<f64> {
        let data_mag = magnitude.to_vec();
        let data_sign = sign.to_vec();
        let len = data_mag.len().min(data_sign.len());
        let mut result = vec![0.0f64; len];
        let simd_len = len & !(NEON_F64_LANES - 1);

        unsafe {
            for i in (0..simd_len).step_by(NEON_F64_LANES) {
                let vm = vld1q_f64(data_mag.as_ptr().add(i));
                let vs = vld1q_f64(data_sign.as_ptr().add(i));

                // Get absolute value of magnitude
                let abs_m = vabsq_f64(vm);

                // Apply sign from sign array
                // For each lane, copysign
                let lane0 = vgetq_lane_f64(abs_m, 0).copysign(vgetq_lane_f64(vs, 0));
                let lane1 = vgetq_lane_f64(abs_m, 1).copysign(vgetq_lane_f64(vs, 1));

                result[i] = lane0;
                result[i + 1] = lane1;
            }
        }

        for i in simd_len..len {
            result[i] = data_mag[i].abs().copysign(data_sign[i]);
        }

        Array::from_vec(result).reshape(&magnitude.shape())
    }
}

// =============================================================================
// Non-aarch64 Fallback Implementations
// =============================================================================

#[cfg(not(target_arch = "aarch64"))]
impl NeonEnhancedOps {
    pub fn vectorized_abs_f64(input: &Array<f64>) -> Array<f64> {
        input.map(|x| x.abs())
    }

    pub fn vectorized_sign_f64(input: &Array<f64>) -> Array<f64> {
        input.map(|x| {
            if x > 0.0 {
                1.0
            } else if x < 0.0 {
                -1.0
            } else {
                0.0
            }
        })
    }

    pub fn vectorized_maximum_f64(a: &Array<f64>, b: &Array<f64>) -> Array<f64> {
        let data_a = a.to_vec();
        let data_b = b.to_vec();
        let len = data_a.len().min(data_b.len());
        let result: Vec<f64> = (0..len).map(|i| data_a[i].max(data_b[i])).collect();
        Array::from_vec(result).reshape(&a.shape())
    }

    pub fn vectorized_minimum_f64(a: &Array<f64>, b: &Array<f64>) -> Array<f64> {
        let data_a = a.to_vec();
        let data_b = b.to_vec();
        let len = data_a.len().min(data_b.len());
        let result: Vec<f64> = (0..len).map(|i| data_a[i].min(data_b[i])).collect();
        Array::from_vec(result).reshape(&a.shape())
    }

    pub fn vectorized_clamp_f64(input: &Array<f64>, min_val: f64, max_val: f64) -> Array<f64> {
        input.map(|x| x.clamp(min_val, max_val))
    }

    pub fn vectorized_copysign_f64(magnitude: &Array<f64>, sign: &Array<f64>) -> Array<f64> {
        let data_mag = magnitude.to_vec();
        let data_sign = sign.to_vec();
        let len = data_mag.len().min(data_sign.len());
        let result: Vec<f64> = (0..len)
            .map(|i| data_mag[i].abs().copysign(data_sign[i]))
            .collect();
        Array::from_vec(result).reshape(&magnitude.shape())
    }
}