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

//! Rounding operations using NEON SIMD
//!
//! This module provides optimized floor, ceil, and round 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 Rounding Operations
// =============================================================================

impl NeonEnhancedOps {
    /// NEON vectorized floor for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_floor_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 floor_v = vrndmq_f64(v);
                vst1q_f64(result.as_mut_ptr().add(i), floor_v);
            }
        }

        for i in simd_len..len {
            result[i] = data[i].floor();
        }

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

    /// NEON vectorized ceiling for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_ceil_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 ceil_v = vrndpq_f64(v);
                vst1q_f64(result.as_mut_ptr().add(i), ceil_v);
            }
        }

        for i in simd_len..len {
            result[i] = data[i].ceil();
        }

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

    /// NEON vectorized round for f64
    #[cfg(target_arch = "aarch64")]
    pub fn vectorized_round_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 round_v = vrndnq_f64(v);
                vst1q_f64(result.as_mut_ptr().add(i), round_v);
            }
        }

        for i in simd_len..len {
            result[i] = data[i].round();
        }

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

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

#[cfg(not(target_arch = "aarch64"))]
impl NeonEnhancedOps {
    pub fn vectorized_floor_f64(input: &Array<f64>) -> Array<f64> {
        input.map(|x| x.floor())
    }

    pub fn vectorized_ceil_f64(input: &Array<f64>) -> Array<f64> {
        input.map(|x| x.ceil())
    }

    pub fn vectorized_round_f64(input: &Array<f64>) -> Array<f64> {
        input.map(|x| x.round())
    }
}