Hugging Face's logo

Androidonnxfork
/
test

Model card Files Community
5
test
File size: 6,142 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
168
169
170
171
 
// 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.

$assert BATCH_TILE % 8 == 0
$assert BATCH_TILE >= 8
$SIMD_TILE = BATCH_TILE // 8
#include <assert.h>

#include <arm_neon.h>

#include <xnnpack/common.h>
#include <xnnpack/vcvt.h>


void xnn_f32_f16_vcvt_ukernel__neon_x${BATCH_TILE}(
    size_t batch,
    const float* input,
    void* output,
    const union xnn_f32_f16_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
{
  assert(batch != 0);
  assert(batch % sizeof(float) == 0);
  assert(input != NULL);
  assert(output != NULL);

  const uint32x4_t vexp_bias = vld1q_dup_u32(&params->neon.exp_bias);
  const float32x4_t vscale_to_inf = vld1q_dup_f32(&params->neon.scale_to_inf);
  const uint32x4_t vexpw_max = vld1q_dup_u32(&params->neon.expw_max);
  const float32x4_t vscale_to_zero = vld1q_dup_f32(&params->neon.scale_to_zero);
  const uint32x4_t vbias_min = vdupq_n_u32(UINT32_C(0x40000000));
  const uint16x8_t vexph_mask = vdupq_n_u16(UINT16_C(0x7C00));
  const uint16x8_t vmanth_mask = vdupq_n_u16(UINT16_C(0x0FFF));
  const uint16x8_t vsignh_mask = vdupq_n_u16(UINT16_C(0x8000));
  const uint16x8_t vnanh = vdupq_n_u16(UINT16_C(0x7E00));

  uint16_t* o = (uint16_t*) output;
  for (; batch >= ${BATCH_TILE} * sizeof(float); batch -= ${BATCH_TILE} * sizeof(float)) {
    $for N in range(2*SIMD_TILE):
      const float32x4_t vx${N} = vld1q_f32(input); input += 4;

    $for N in range(2*SIMD_TILE):
      const float32x4_t vabsx${N} = vabsq_f32(vx${N});

    $for N in range(2*SIMD_TILE):
      uint32x4_t vbias${N} = vaddq_u32(vreinterpretq_u32_f32(vabsx${N}), vexp_bias);

    $for N in range(2*SIMD_TILE):
      float32x4_t vf${N} = vmulq_f32(vabsx${N}, vscale_to_inf);
    $for N in range(2*SIMD_TILE):
      const uint32x4_t vnanmaskw${N} = vcgtq_u32(vreinterpretq_u32_f32(vabsx${N}), vexpw_max);

    $for N in range(2*SIMD_TILE):
      vbias${N} = vandq_u32(vbias${N}, vexpw_max);
    $for N in range(2*SIMD_TILE):
      vf${N} = vmulq_f32(vf${N}, vscale_to_zero);

    $for N in range(SIMD_TILE):
      const uint16x8_t vnanmaskh${N} = vcombine_u16(vmovn_u32(vnanmaskw${2*N}), vmovn_u32(vnanmaskw${2*N+1}));
    $for N in range(2*SIMD_TILE):
      vbias${N} = vmaxq_u32(vbias${N}, vbias_min);

    $for N in range(2*SIMD_TILE):
      vf${N} = vaddq_f32(vf${N}, vreinterpretq_f32_u32(vbias${N}));

    $for N in range(SIMD_TILE):
      uint16x8_t vexph${N} = vcombine_u16(vshrn_n_u32(vreinterpretq_u32_f32(vf${2*N}), 13), vshrn_n_u32(vreinterpretq_u32_f32(vf${2*N+1}), 13));
    $for N in range(SIMD_TILE):
      uint16x8_t vmanth${N} = vcombine_u16(vmovn_u32(vreinterpretq_u32_f32(vf${2*N})), vmovn_u32(vreinterpretq_u32_f32(vf${2*N+1})));
    $for N in range(SIMD_TILE):
      uint16x8_t vsignh${N} = vcombine_u16(vshrn_n_u32(vreinterpretq_u32_f32(vx${2*N}), 16), vshrn_n_u32(vreinterpretq_u32_f32(vx${2*N+1}), 16));

    $for N in range(SIMD_TILE):
      vexph${N} = vandq_u16(vexph${N}, vexph_mask);
    $for N in range(SIMD_TILE):
      vmanth${N} = vandq_u16(vmanth${N}, vmanth_mask);
    $for N in range(SIMD_TILE):
      vsignh${N} = vandq_u16(vsignh${N}, vsignh_mask);

    $for N in range(SIMD_TILE):
      uint16x8_t vh${N} = vaddq_u16(vmanth${N}, vexph${N});

    $for N in range(SIMD_TILE):
      vh${N} = vbslq_u16(vnanmaskh${N}, vnanh, vh${N});

    $for N in range(SIMD_TILE):
      vh${N} = vorrq_u16(vh${N}, vsignh${N});

    $for N in range(SIMD_TILE):
      vst1q_u16(o, vh${N}); o += 8;
  }
  for (; batch >= 4 * sizeof(float); batch -= 4 * sizeof(float)) {
    const float32x4_t vx = vld1q_f32(input); input += 4;

    const float32x4_t vabsx = vabsq_f32(vx);

    uint32x4_t vbias = vaddq_u32(vreinterpretq_u32_f32(vabsx), vexp_bias);

    float32x4_t vf = vmulq_f32(vabsx, vscale_to_inf);
    const uint32x4_t vnanmaskw = vcgtq_u32(vreinterpretq_u32_f32(vabsx), vexpw_max);

    vbias = vandq_u32(vbias, vexpw_max);
    vf = vmulq_f32(vf, vscale_to_zero);

    const uint16x4_t vnanmaskh = vmovn_u32(vnanmaskw);
    vbias = vmaxq_u32(vbias, vbias_min);

    vf = vaddq_f32(vf, vreinterpretq_f32_u32(vbias));

    uint16x4_t vexph = vshrn_n_u32(vreinterpretq_u32_f32(vf), 13);
    uint16x4_t vmanth = vmovn_u32(vreinterpretq_u32_f32(vf));
    uint16x4_t vsignh = vshrn_n_u32(vreinterpretq_u32_f32(vx), 16);

    vexph = vand_u16(vexph, vget_low_u16(vexph_mask));
    vmanth = vand_u16(vmanth, vget_low_u16(vmanth_mask));
    vsignh = vand_u16(vsignh, vget_low_u16(vsignh_mask));

    uint16x4_t vh = vadd_u16(vmanth, vexph);

    vh = vbsl_u16(vnanmaskh, vget_low_u16(vnanh), vh);

    vh = vorr_u16(vh, vsignh);

    vst1_u16(o, vh); o += 4;
  }
  if XNN_UNLIKELY(batch != 0) {
    assert(batch % sizeof(float) == 0);
    assert(batch >= 1 * sizeof(float));
    assert(batch <= 3 * sizeof(float));
    const float32x4_t vx = vld1q_f32(input);

    const float32x4_t vabsx = vabsq_f32(vx);

    uint32x4_t vbias = vaddq_u32(vreinterpretq_u32_f32(vabsx), vexp_bias);

    float32x4_t vf = vmulq_f32(vabsx, vscale_to_inf);
    const uint32x4_t vnanmaskw = vcgtq_u32(vreinterpretq_u32_f32(vabsx), vexpw_max);

    vbias = vandq_u32(vbias, vexpw_max);
    vf = vmulq_f32(vf, vscale_to_zero);

    const uint16x4_t vnanmaskh = vmovn_u32(vnanmaskw);
    vbias = vmaxq_u32(vbias, vbias_min);

    vf = vaddq_f32(vf, vreinterpretq_f32_u32(vbias));

    uint16x4_t vexph = vshrn_n_u32(vreinterpretq_u32_f32(vf), 13);
    uint16x4_t vmanth = vmovn_u32(vreinterpretq_u32_f32(vf));
    uint16x4_t vsignh = vshrn_n_u32(vreinterpretq_u32_f32(vx), 16);

    vexph = vand_u16(vexph, vget_low_u16(vexph_mask));
    vmanth = vand_u16(vmanth, vget_low_u16(vmanth_mask));
    vsignh = vand_u16(vsignh, vget_low_u16(vsignh_mask));

    uint16x4_t vh = vadd_u16(vmanth, vexph);

    vh = vbsl_u16(vnanmaskh, vget_low_u16(vnanh), vh);

    vh = vorr_u16(vh, vsignh);

    if (batch & (2 * sizeof(float))) {
      vst1_lane_u32((void*) o, vreinterpret_u32_u16(vh), 0); o += 2;
      vh = vext_u16(vh, vh, 2);
    }
    if (batch & (1 * sizeof(float))) {
      vst1_lane_u16(o, vh, 0);
    }
  }
}