hodu_cpu_kernels 0.2.1

hodu cpu kernels
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

#ifndef HODU_CPU_KERNELS_MATH_UTILS_H
#define HODU_CPU_KERNELS_MATH_UTILS_H

#include "constants.h"
#include <math.h>
#include <stdint.h>

#ifdef __cplusplus
extern "C" {
#endif

// ============================================================================
// GENERIC MATH HELPERS
// ============================================================================

// Integer power function for integer types
static inline int64_t ipow_i64(int64_t base, int64_t exp) {
    if (exp < 0)
        return 0;
    int64_t result = 1;
    while (exp > 0) {
        if (exp & 1)
            result *= base;
        base *= base;
        exp >>= 1;
    }
    return result;
}

static inline uint64_t ipow_u64(uint64_t base, uint64_t exp) {
    uint64_t result = 1;
    while (exp > 0) {
        if (exp & 1)
            result *= base;
        base *= base;
        exp >>= 1;
    }
    return result;
}

// Float power with optimization
static inline float powf_opt(float base, float exponent) {
    if (exponent == 0.0f)
        return 1.0f;
    if (base == 0.0f)
        return (exponent > 0.0f) ? 0.0f : INFINITY;
    if (base == 1.0f)
        return 1.0f;
    if (exponent == 1.0f)
        return base;

    // Check if exponent is an integer
    if (floorf(exponent) == exponent) {
        if (exponent >= 0.0f && exponent < 32.0f) {
            int iexp = (int)exponent;
            float result = 1.0f;
            while (iexp > 0) {
                if (iexp & 1)
                    result *= base;
                base *= base;
                iexp >>= 1;
            }
            return result;
        } else if (exponent < 0.0f && exponent > -32.0f) {
            return 1.0f / powf_opt(base, -exponent);
        }
    }

    if (base < 0.0f)
        return NAN;

    return powf(base, exponent);
}

static inline double pow_opt(double base, double exponent) {
    if (exponent == 0.0)
        return 1.0;
    if (base == 0.0)
        return (exponent > 0.0) ? 0.0 : INFINITY;
    if (base == 1.0)
        return 1.0;
    if (exponent == 1.0)
        return base;

    if (floor(exponent) == exponent) {
        if (exponent >= 0.0 && exponent < 32.0) {
            int iexp = (int)exponent;
            double result = 1.0;
            while (iexp > 0) {
                if (iexp & 1)
                    result *= base;
                base *= base;
                iexp >>= 1;
            }
            return result;
        } else if (exponent < 0.0 && exponent > -32.0) {
            return 1.0 / pow_opt(base, -exponent);
        }
    }

    if (base < 0.0)
        return NAN;

    return pow(base, exponent);
}

// Optimized tan
static inline float tanf_opt(float x) {
    x = fmodf(x, 2.0f * (float)M_PI);
    if (x > (float)M_PI)
        x -= 2.0f * (float)M_PI;
    else if (x < -(float)M_PI)
        x += 2.0f * (float)M_PI;

    float halfPi = (float)M_PI / 2.0f;
    float eps = 1e-6f;

    if (fabsf(fabsf(x) - halfPi) < eps) {
        return x > 0 ? 1e6f : -1e6f;
    }

    return sinf(x) / cosf(x);
}

static inline double tan_opt(double x) {
    x = fmod(x, 2.0 * M_PI);
    if (x > M_PI)
        x -= 2.0 * M_PI;
    else if (x < -M_PI)
        x += 2.0 * M_PI;

    double halfPi = M_PI / 2.0;
    double eps = 1e-14;

    if (fabs(fabs(x) - halfPi) < eps) {
        return x > 0 ? 1e14 : -1e14;
    }

    return sin(x) / cos(x);
}

// Base-10 exponential
static inline float exp10f_opt(float x) { return expf(x * (float)M_LN10); }

static inline double exp10_opt(double x) { return exp(x * M_LN10); }

// ============================================================================
// ACTIVATION FUNCTION HELPERS
// ============================================================================

static inline float gelu_helper_f32(float x) {
    return 0.5f * x * (1.0f + tanhf(0.7978845608f * (x + 0.044715f * x * x * x)));
}

static inline double gelu_helper_f64(double x) {
    return 0.5 * x * (1.0 + tanh(0.7978845608 * (x + 0.044715 * x * x * x)));
}

static inline float softplus_helper_f32(float x) { return logf(1.0f + expf(x)); }

static inline double softplus_helper_f64(double x) { return log(1.0 + exp(x)); }

static inline float silu_helper_f32(float x) { return x / (1.0f + expf(-x)); }

static inline double silu_helper_f64(double x) { return x / (1.0 + exp(-x)); }

static inline float mish_helper_f32(float x) { return x * tanhf(softplus_helper_f32(x)); }

static inline double mish_helper_f64(double x) { return x * tanh(softplus_helper_f64(x)); }

#ifdef __cplusplus
}
#endif

#endif // HODU_CPU_KERNELS_MATH_UTILS_H