Hugging Face's logo

Androidonnxfork
/
test

Model card Files Community
6
test
File size: 6,531 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
 
// 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
$ABC = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ"
#include <assert.h>

#include <arm_neon.h>

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


void xnn_f16_f32_vcvt_ukernel__neon_int16_x${BATCH_TILE}(
    size_t batch,
    const void* input,
    float* output,
    const union xnn_f16_f32_cvt_params params[restrict XNN_MIN_ELEMENTS(1)]) XNN_OOB_READS
{
  assert(batch != 0);
  assert(batch % sizeof(uint16_t) == 0);
  assert(input != NULL);
  assert(output != NULL);

  const uint16x8_t vsign_mask = vmovq_n_u16(0x8000);
  const uint16x8_t vexp_offset = vmovq_n_u16(0x7000);
  const float32x4_t vexp_scale = vld1q_dup_f32(&params->neon.exp_scale);
  const uint32x4_t vmagic_bias = vmovq_n_u32(0x3F000000);
  const uint16x8_t vdenorm_cutoff = vmovq_n_u16(0x0400);

  const uint16_t* i = (const uint16_t*) input;
  $if BATCH_TILE > 8:
    for (; batch >= ${BATCH_TILE} * sizeof(uint16_t); batch -= ${BATCH_TILE} * sizeof(uint16_t)) {
      $for N in range(SIMD_TILE):
        const uint16x8_t vh${N} = vld1q_u16(i); i += 8;

      $for N in range(SIMD_TILE):
        const uint16x8_t vsign${N} = vandq_u16(vh${N}, vsign_mask);

      $for N in range(SIMD_TILE):
        const uint16x8_t vnonsign${N} = veorq_u16(vh${N}, vsign${N});

      $for N in range(SIMD_TILE):
        const uint16x8x2_t vprenorm${N} = vzipq_u16(vshlq_n_u16(vnonsign${N}, 13), vsraq_n_u16(vexp_offset, vnonsign${N}, 3));

      $for N in range(SIMD_TILE):
        const float32x4_t vnorm${2*N} = vmulq_f32(vreinterpretq_f32_u16(vprenorm${N}.val[0]), vexp_scale);
        const float32x4_t vnorm${2*N+1} = vmulq_f32(vreinterpretq_f32_u16(vprenorm${N}.val[1]), vexp_scale);

      $for N in range(SIMD_TILE):
        const float32x4_t vdenorm${2*N} = vsubq_f32(vreinterpretq_f32_u32(vaddw_u16(vmagic_bias, vget_low_u16(vnonsign${N}))), vreinterpretq_f32_u32(vmagic_bias));
        const float32x4_t vdenorm${2*N+1} = vsubq_f32(vreinterpretq_f32_u32(vaddw_u16(vmagic_bias, vget_high_u16(vnonsign${N}))), vreinterpretq_f32_u32(vmagic_bias));

      $for N in range(SIMD_TILE):
        const uint16x8_t vmask${N} = vcgtq_u16(vnonsign${N}, vdenorm_cutoff);

      $for N in range(SIMD_TILE):
        const uint32x4_t vxmask${2*N} = vreinterpretq_u32_s32(vmovl_s16(vreinterpret_s16_u16(vget_low_u16(vmask${N}))));
        const uint32x4_t vf${2*N} = vorrq_u32(vshll_n_u16(vget_low_u16(vsign${N}), 16),
          vreinterpretq_u32_f32(vbslq_f32(vxmask${2*N}, vnorm${2*N}, vdenorm${2*N})));

      $for N in range(SIMD_TILE):
        const uint32x4_t vxmask${2*N+1} = vreinterpretq_u32_s32(vmovl_s16(vreinterpret_s16_u16(vget_high_u16(vmask${N}))));
        const uint32x4_t vf${2*N+1} = vorrq_u32(vshll_n_u16(vget_high_u16(vsign${N}), 16),
          vreinterpretq_u32_f32(vbslq_f32(vxmask${2*N+1}, vnorm${2*N+1}, vdenorm${2*N+1})));

      $for N in range(2*SIMD_TILE):
        vst1q_f32(output, vreinterpretq_f32_u32(vf${N})); output += 4;
    }
  for (; batch >= 8 * sizeof(uint16_t); batch -= 8 * sizeof(uint16_t)) {
    const uint16x8_t vh = vld1q_u16(i); i += 8;

    const uint16x8_t vsign = vandq_u16(vh, vsign_mask);

    const uint16x8_t vnonsign = veorq_u16(vh, vsign);

    const uint16x8x2_t vprenorm = vzipq_u16(vshlq_n_u16(vnonsign, 13), vsraq_n_u16(vexp_offset, vnonsign, 3));
    const float32x4_t vnorm_lo = vmulq_f32(vreinterpretq_f32_u16(vprenorm.val[0]), vexp_scale);
    const float32x4_t vnorm_hi = vmulq_f32(vreinterpretq_f32_u16(vprenorm.val[1]), vexp_scale);

    const float32x4_t vdenorm_lo = vsubq_f32(vreinterpretq_f32_u32(vaddw_u16(vmagic_bias, vget_low_u16(vnonsign))), vreinterpretq_f32_u32(vmagic_bias));
    const float32x4_t vdenorm_hi = vsubq_f32(vreinterpretq_f32_u32(vaddw_u16(vmagic_bias, vget_high_u16(vnonsign))), vreinterpretq_f32_u32(vmagic_bias));

    const uint16x8_t vmask = vcgtq_u16(vnonsign, vdenorm_cutoff);

    const uint32x4_t vxmask_lo = vreinterpretq_u32_s32(vmovl_s16(vreinterpret_s16_u16(vget_low_u16(vmask))));
    const uint32x4_t vf_lo = vorrq_u32(vshll_n_u16(vget_low_u16(vsign), 16),
      vreinterpretq_u32_f32(vbslq_f32(vxmask_lo, vnorm_lo, vdenorm_lo)));

    const uint32x4_t vxmask_hi = vreinterpretq_u32_s32(vmovl_s16(vreinterpret_s16_u16(vget_high_u16(vmask))));
    const uint32x4_t vf_hi = vorrq_u32(vshll_n_u16(vget_high_u16(vsign), 16),
      vreinterpretq_u32_f32(vbslq_f32(vxmask_hi, vnorm_hi, vdenorm_hi)));

    vst1q_f32(output, vreinterpretq_f32_u32(vf_lo)); output += 4;
    vst1q_f32(output, vreinterpretq_f32_u32(vf_hi)); output += 4;
  }
  if XNN_UNPREDICTABLE(batch != 0) {
    const uint16x8_t vh = vld1q_u16(i); i += 8;

    const uint16x8_t vsign = vandq_u16(vh, vsign_mask);

    const uint16x8_t vnonsign = veorq_u16(vh, vsign);

    const uint16x8x2_t vprenorm = vzipq_u16(vshlq_n_u16(vnonsign, 13), vsraq_n_u16(vexp_offset, vnonsign, 3));
    const float32x4_t vnorm_lo = vmulq_f32(vreinterpretq_f32_u16(vprenorm.val[0]), vexp_scale);
    const float32x4_t vnorm_hi = vmulq_f32(vreinterpretq_f32_u16(vprenorm.val[1]), vexp_scale);

    const float32x4_t vdenorm_lo = vsubq_f32(vreinterpretq_f32_u32(vaddw_u16(vmagic_bias, vget_low_u16(vnonsign))), vreinterpretq_f32_u32(vmagic_bias));
    const float32x4_t vdenorm_hi = vsubq_f32(vreinterpretq_f32_u32(vaddw_u16(vmagic_bias, vget_high_u16(vnonsign))), vreinterpretq_f32_u32(vmagic_bias));

    const uint16x8_t vmask = vcgtq_u16(vnonsign, vdenorm_cutoff);

    const uint32x4_t vxmask_lo = vreinterpretq_u32_s32(vmovl_s16(vreinterpret_s16_u16(vget_low_u16(vmask))));
    uint32x4_t vf = vorrq_u32(vshll_n_u16(vget_low_u16(vsign), 16),
      vreinterpretq_u32_f32(vbslq_f32(vxmask_lo, vnorm_lo, vdenorm_lo)));

    if (batch & (4 * sizeof(uint16_t))) {
      vst1q_f32(output, vreinterpretq_f32_u32(vf)); output += 4;

      const uint32x4_t vxmask_hi = vreinterpretq_u32_s32(vmovl_s16(vreinterpret_s16_u16(vget_high_u16(vmask))));
      vf = vorrq_u32(vshll_n_u16(vget_high_u16(vsign), 16),
        vreinterpretq_u32_f32(vbslq_f32(vxmask_hi, vnorm_hi, vdenorm_hi)));
    }
    uint32x2_t vf_lo = vget_low_u32(vf);
    if (batch & (2 * sizeof(uint16_t))) {
      vst1_f32(output, vreinterpret_f32_u32(vf_lo)); output += 2;
      vf_lo = vget_high_u32(vf);
    }
    if (batch & (1 * sizeof(uint16_t))) {
      vst1_lane_f32(output, vreinterpret_f32_u32(vf_lo), 0);
    }
  }
}