Hugging Face's logo

Androidonnxfork
/
test

Model card Files Community
5
test
File size: 4,380 Bytes 
8b7c501
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
// Copyright 2021 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.

#include <assert.h>
#include <string.h>

#include <arm_fp16.h>

#include <xnnpack/common.h>
#include <xnnpack/intrinsics-polyfill.h>
#include <xnnpack/math.h>
#include <xnnpack/vbinary.h>


void xnn_f16_vdiv_minmax_ukernel__fp16arith_x2(
    size_t batch,
    const void* restrict input_a,
    const void* restrict input_b,
    void* restrict output,
    const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
  assert(batch != 0);
  assert(batch % sizeof(float16_t) == 0);
  assert(input_a != NULL);
  assert(input_b != NULL);
  assert(output != NULL);

  const float16_t* a = (const float16_t*) input_a;
  const float16_t* b = (const float16_t*) input_b;
  float16_t* o = (float16_t*) output;

  float16_t vy_min, vy_max;
  memcpy(&vy_min, &params->fp16arith.min, sizeof(vy_min));
  memcpy(&vy_max, &params->fp16arith.max, sizeof(vy_max));

  for (; batch >= 2 * sizeof(float16_t); batch -= 2 * sizeof(float16_t)) {
    const float16_t va0 = *a++;
    const float16_t va1 = *a++;

    const float16_t vb0 = *b++;
    const float16_t vb1 = *b++;

    float16_t vacc0 = vdivh_f16(va0, vb0);
    float16_t vacc1 = vdivh_f16(va1, vb1);


    vacc0 = vmaxnmh_f16(vacc0, vy_min);
    vacc1 = vmaxnmh_f16(vacc1, vy_min);

    vacc0 = vminnmh_f16(vacc0, vy_max);
    vacc1 = vminnmh_f16(vacc1, vy_max);

    *o++ = vacc0;
    *o++ = vacc1;
  }
  if XNN_UNLIKELY(batch != 0) {
    const float16_t va = *a;
    const float16_t vb = *b;
    float16_t vacc = vdivh_f16(va, vb);
    vacc = vmaxnmh_f16(vacc, vy_min);
    vacc = vminnmh_f16(vacc, vy_max);
    *o = vacc;
  }
}

void xnn_f16_vdivc_minmax_ukernel__fp16arith_x2(
    size_t batch,
    const void* restrict input_a,
    const void* restrict input_b,
    void* restrict output,
    const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
  assert(batch != 0);
  assert(batch % sizeof(float16_t) == 0);
  assert(input_a != NULL);
  assert(input_b != NULL);
  assert(output != NULL);

  const float16_t* a = (const float16_t*) input_a;
  const float16_t* b = (const float16_t*) input_b;
  float16_t* o = (float16_t*) output;

  float16_t vy_min, vy_max;
  memcpy(&vy_min, &params->fp16arith.min, sizeof(vy_min));
  memcpy(&vy_max, &params->fp16arith.max, sizeof(vy_max));

  const float16_t vb = *b;
  for (; batch >= 2 * sizeof(float16_t); batch -= 2 * sizeof(float16_t)) {
    float16_t vacc0 = a[0];
    float16_t vacc1 = a[1];
    a += 2;

    vacc0 = vdivh_f16(vacc0, vb);
    vacc1 = vdivh_f16(vacc1, vb);


    vacc0 = vmaxnmh_f16(vacc0, vy_min);
    vacc1 = vmaxnmh_f16(vacc1, vy_min);

    vacc0 = vminnmh_f16(vacc0, vy_max);
    vacc1 = vminnmh_f16(vacc1, vy_max);

    o[0] = vacc0;
    o[1] = vacc1;
    o += 2;
  }
  if XNN_UNLIKELY(batch != 0) {
    float16_t vacc = *a;
    vacc = vdivh_f16(vacc, vb);
    vacc = vmaxnmh_f16(vacc, vy_min);
    vacc = vminnmh_f16(vacc, vy_max);
    *o = vacc;
  }
}

void xnn_f16_vrdivc_minmax_ukernel__fp16arith_x2(
    size_t batch,
    const void* restrict input_a,
    const void* restrict input_b,
    void* restrict output,
    const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
  assert(batch != 0);
  assert(batch % sizeof(float16_t) == 0);
  assert(input_a != NULL);
  assert(input_b != NULL);
  assert(output != NULL);

  const float16_t* a = (const float16_t*) input_a;
  const float16_t* b = (const float16_t*) input_b;
  float16_t* o = (float16_t*) output;

  float16_t vy_min, vy_max;
  memcpy(&vy_min, &params->fp16arith.min, sizeof(vy_min));
  memcpy(&vy_max, &params->fp16arith.max, sizeof(vy_max));

  const float16_t vb = *b;
  for (; batch >= 2 * sizeof(float16_t); batch -= 2 * sizeof(float16_t)) {
    float16_t vacc0 = a[0];
    float16_t vacc1 = a[1];
    a += 2;

    vacc0 = vdivh_f16(vb, vacc0);
    vacc1 = vdivh_f16(vb, vacc1);


    vacc0 = vmaxnmh_f16(vacc0, vy_min);
    vacc1 = vmaxnmh_f16(vacc1, vy_min);

    vacc0 = vminnmh_f16(vacc0, vy_max);
    vacc1 = vminnmh_f16(vacc1, vy_max);

    o[0] = vacc0;
    o[1] = vacc1;
    o += 2;
  }
  if XNN_UNLIKELY(batch != 0) {
    float16_t vacc = *a;
    vacc = vdivh_f16(vb, vacc);
    vacc = vmaxnmh_f16(vacc, vy_min);
    vacc = vminnmh_f16(vacc, vy_max);
    *o = vacc;
  }
}