File size: 1,775 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
// Auto-generated file. Do not edit!
//   Template: src/f16-vbinary/vop-fp16arith.c.in
//   Generator: tools/xngen
//
// Copyright 2022 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 <arm_fp16.h>

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


void xnn_f16_vmax_ukernel__fp16arith_x4(
    size_t batch,
    const void* restrict input_a,
    const void* restrict input_b,
    void* restrict output,
    const union xnn_f16_default_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;


  for (; batch >= 4 * sizeof(float16_t); batch -= 4 * sizeof(float16_t)) {
    const float16_t va0 = *a++;
    const float16_t va1 = *a++;
    const float16_t va2 = *a++;
    const float16_t va3 = *a++;

    const float16_t vb0 = *b++;
    const float16_t vb1 = *b++;
    const float16_t vb2 = *b++;
    const float16_t vb3 = *b++;

    float16_t vacc0 = vmaxnmh_f16(va0, vb0);
    float16_t vacc1 = vmaxnmh_f16(va1, vb1);
    float16_t vacc2 = vmaxnmh_f16(va2, vb2);
    float16_t vacc3 = vmaxnmh_f16(va3, vb3);



    *o++ = vacc0;
    *o++ = vacc1;
    *o++ = vacc2;
    *o++ = vacc3;
  }
  if XNN_UNLIKELY(batch != 0) {
    do {
      const float16_t va = *a++;
      const float16_t vb = *b++;
      float16_t vacc = vmaxnmh_f16(va, vb);
      *o++ = vacc;
      batch -= sizeof(float16_t);
    } while (batch != 0);
  }
}