File size: 11,511 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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
// Auto-generated file. Do not edit!
//   Template: src/f32-f16-vcvt/wasmsimd.c.in
//   Generator: tools/xngen
//
// 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 <wasm_simd128.h>

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


void xnn_f32_f16_vcvt_ukernel__wasmsimd_x24(
    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 v128_t vexp_bias = wasm_v128_load64_splat(params->wasmsimd.exp_bias);
  const v128_t vscale_to_inf = wasm_v128_load64_splat(params->wasmsimd.scale_to_inf);
  const v128_t vexpw_max = wasm_v128_load64_splat(params->wasmsimd.expw_max);
  const v128_t vscale_to_zero = wasm_v128_load64_splat(params->wasmsimd.scale_to_zero);
  const v128_t vbias_min = wasm_v128_load64_splat(params->wasmsimd.bias_min);
  const v128_t vmanth_mask = wasm_v128_load64_splat(params->wasmsimd.manth_mask);
  const v128_t vexph_mask = wasm_v128_load64_splat(params->wasmsimd.exph_mask);
  const v128_t vnanh = wasm_v128_load64_splat(params->wasmsimd.nanh);

  uint16_t* o = (uint16_t*) output;
  for (; batch >= 24 * sizeof(float); batch -= 24 * sizeof(float)) {
    const v128_t vx0 = wasm_v128_load(input);
    const v128_t vx1 = wasm_v128_load(input + 4);
    const v128_t vx2 = wasm_v128_load(input + 8);
    const v128_t vx3 = wasm_v128_load(input + 12);
    const v128_t vx4 = wasm_v128_load(input + 16);
    const v128_t vx5 = wasm_v128_load(input + 20);
    input += 24;

    const v128_t vabsx0 = wasm_f32x4_abs(vx0);
    const v128_t vabsx1 = wasm_f32x4_abs(vx1);
    const v128_t vabsx2 = wasm_f32x4_abs(vx2);
    const v128_t vabsx3 = wasm_f32x4_abs(vx3);
    const v128_t vabsx4 = wasm_f32x4_abs(vx4);
    const v128_t vabsx5 = wasm_f32x4_abs(vx5);

    const v128_t vsignx0 = wasm_v128_xor(vx0, vabsx0);
    const v128_t vsignx1 = wasm_v128_xor(vx1, vabsx1);
    const v128_t vsignx2 = wasm_v128_xor(vx2, vabsx2);
    const v128_t vsignx3 = wasm_v128_xor(vx3, vabsx3);
    const v128_t vsignx4 = wasm_v128_xor(vx4, vabsx4);
    const v128_t vsignx5 = wasm_v128_xor(vx5, vabsx5);

    v128_t vbias0 = wasm_i32x4_add(vabsx0, vexp_bias);
    v128_t vbias1 = wasm_i32x4_add(vabsx1, vexp_bias);
    v128_t vbias2 = wasm_i32x4_add(vabsx2, vexp_bias);
    v128_t vbias3 = wasm_i32x4_add(vabsx3, vexp_bias);
    v128_t vbias4 = wasm_i32x4_add(vabsx4, vexp_bias);
    v128_t vbias5 = wasm_i32x4_add(vabsx5, vexp_bias);

    v128_t vf0 = wasm_f32x4_mul(vabsx0, vscale_to_inf);
    v128_t vf1 = wasm_f32x4_mul(vabsx1, vscale_to_inf);
    v128_t vf2 = wasm_f32x4_mul(vabsx2, vscale_to_inf);
    v128_t vf3 = wasm_f32x4_mul(vabsx3, vscale_to_inf);
    v128_t vf4 = wasm_f32x4_mul(vabsx4, vscale_to_inf);
    v128_t vf5 = wasm_f32x4_mul(vabsx5, vscale_to_inf);

    const v128_t vnanmaskw0 = wasm_i32x4_gt(vabsx0, vexpw_max);
    const v128_t vnanmaskw1 = wasm_i32x4_gt(vabsx1, vexpw_max);
    const v128_t vnanmaskw2 = wasm_i32x4_gt(vabsx2, vexpw_max);
    const v128_t vnanmaskw3 = wasm_i32x4_gt(vabsx3, vexpw_max);
    const v128_t vnanmaskw4 = wasm_i32x4_gt(vabsx4, vexpw_max);
    const v128_t vnanmaskw5 = wasm_i32x4_gt(vabsx5, vexpw_max);

    vbias0 = wasm_v128_and(vbias0, vexpw_max);
    vbias1 = wasm_v128_and(vbias1, vexpw_max);
    vbias2 = wasm_v128_and(vbias2, vexpw_max);
    vbias3 = wasm_v128_and(vbias3, vexpw_max);
    vbias4 = wasm_v128_and(vbias4, vexpw_max);
    vbias5 = wasm_v128_and(vbias5, vexpw_max);

    vf0 = wasm_f32x4_mul(vf0, vscale_to_zero);
    vf1 = wasm_f32x4_mul(vf1, vscale_to_zero);
    vf2 = wasm_f32x4_mul(vf2, vscale_to_zero);
    vf3 = wasm_f32x4_mul(vf3, vscale_to_zero);
    vf4 = wasm_f32x4_mul(vf4, vscale_to_zero);
    vf5 = wasm_f32x4_mul(vf5, vscale_to_zero);

    const v128_t vnanmaskh0 = wasm_i16x8_narrow_i32x4(vnanmaskw0, vnanmaskw1);
    const v128_t vnanmaskh1 = wasm_i16x8_narrow_i32x4(vnanmaskw2, vnanmaskw3);
    const v128_t vnanmaskh2 = wasm_i16x8_narrow_i32x4(vnanmaskw4, vnanmaskw5);

    const v128_t vsignh0 = wasm_i16x8_narrow_i32x4(vsignx0, vsignx1);
    const v128_t vsignh1 = wasm_i16x8_narrow_i32x4(vsignx2, vsignx3);
    const v128_t vsignh2 = wasm_i16x8_narrow_i32x4(vsignx4, vsignx5);

    vbias0 = wasm_i16x8_max(vbias0, vbias_min);
    vbias1 = wasm_i16x8_max(vbias1, vbias_min);
    vbias2 = wasm_i16x8_max(vbias2, vbias_min);
    vbias3 = wasm_i16x8_max(vbias3, vbias_min);
    vbias4 = wasm_i16x8_max(vbias4, vbias_min);
    vbias5 = wasm_i16x8_max(vbias5, vbias_min);

    vf0 = wasm_f32x4_add(vf0, vbias0);
    vf1 = wasm_f32x4_add(vf1, vbias1);
    vf2 = wasm_f32x4_add(vf2, vbias2);
    vf3 = wasm_f32x4_add(vf3, vbias3);
    vf4 = wasm_f32x4_add(vf4, vbias4);
    vf5 = wasm_f32x4_add(vf5, vbias5);

    v128_t vexpw0 = wasm_i32x4_shr(vf0, 13);
    v128_t vexpw1 = wasm_i32x4_shr(vf1, 13);
    v128_t vexpw2 = wasm_i32x4_shr(vf2, 13);
    v128_t vexpw3 = wasm_i32x4_shr(vf3, 13);
    v128_t vexpw4 = wasm_i32x4_shr(vf4, 13);
    v128_t vexpw5 = wasm_i32x4_shr(vf5, 13);

    const v128_t vmantw0 = wasm_v128_and(vf0, vmanth_mask);
    const v128_t vmantw1 = wasm_v128_and(vf1, vmanth_mask);
    const v128_t vmantw2 = wasm_v128_and(vf2, vmanth_mask);
    const v128_t vmantw3 = wasm_v128_and(vf3, vmanth_mask);
    const v128_t vmantw4 = wasm_v128_and(vf4, vmanth_mask);
    const v128_t vmantw5 = wasm_v128_and(vf5, vmanth_mask);

    vexpw0 = wasm_v128_and(vexpw0, vexph_mask);
    vexpw1 = wasm_v128_and(vexpw1, vexph_mask);
    vexpw2 = wasm_v128_and(vexpw2, vexph_mask);
    vexpw3 = wasm_v128_and(vexpw3, vexph_mask);
    vexpw4 = wasm_v128_and(vexpw4, vexph_mask);
    vexpw5 = wasm_v128_and(vexpw5, vexph_mask);

    const v128_t vnonsignw0 = wasm_i32x4_add(vmantw0, vexpw0);
    const v128_t vnonsignw1 = wasm_i32x4_add(vmantw1, vexpw1);
    const v128_t vnonsignw2 = wasm_i32x4_add(vmantw2, vexpw2);
    const v128_t vnonsignw3 = wasm_i32x4_add(vmantw3, vexpw3);
    const v128_t vnonsignw4 = wasm_i32x4_add(vmantw4, vexpw4);
    const v128_t vnonsignw5 = wasm_i32x4_add(vmantw5, vexpw5);

    const v128_t vnonsignh0 = wasm_i16x8_narrow_i32x4(vnonsignw0, vnonsignw1);
    const v128_t vnonsignh1 = wasm_i16x8_narrow_i32x4(vnonsignw2, vnonsignw3);
    const v128_t vnonsignh2 = wasm_i16x8_narrow_i32x4(vnonsignw4, vnonsignw5);

    const v128_t vabsh0 = wasm_v128_bitselect(vnanh, vnonsignh0, vnanmaskh0);
    const v128_t vabsh1 = wasm_v128_bitselect(vnanh, vnonsignh1, vnanmaskh1);
    const v128_t vabsh2 = wasm_v128_bitselect(vnanh, vnonsignh2, vnanmaskh2);

    const v128_t vh0 = wasm_v128_or(vabsh0, vsignh0);
    const v128_t vh1 = wasm_v128_or(vabsh1, vsignh1);
    const v128_t vh2 = wasm_v128_or(vabsh2, vsignh2);

    wasm_v128_store(o, vh0);
    wasm_v128_store(o + 8, vh1);
    wasm_v128_store(o + 16, vh2);
    o += 24;
  }
  for (; batch >= 8 * sizeof(float); batch -= 8 * sizeof(float)) {
    const v128_t vx_lo = wasm_v128_load(input);
    const v128_t vx_hi = wasm_v128_load(input + 4);
    input += 8;

    const v128_t vabsx_lo = wasm_f32x4_abs(vx_lo);
    const v128_t vabsx_hi = wasm_f32x4_abs(vx_hi);

    const v128_t vsignx_lo = wasm_v128_xor(vx_lo, vabsx_lo);
    const v128_t vsignx_hi = wasm_v128_xor(vx_hi, vabsx_hi);
    v128_t vbias_lo = wasm_i32x4_add(vabsx_lo, vexp_bias);
    v128_t vbias_hi = wasm_i32x4_add(vabsx_hi, vexp_bias);
    v128_t vf_lo = wasm_f32x4_mul(vabsx_lo, vscale_to_inf);
    v128_t vf_hi = wasm_f32x4_mul(vabsx_hi, vscale_to_inf);
    const v128_t vnanmaskw_lo = wasm_i32x4_gt(vabsx_lo, vexpw_max);
    const v128_t vnanmaskw_hi = wasm_i32x4_gt(vabsx_hi, vexpw_max);

    vbias_lo = wasm_v128_and(vbias_lo, vexpw_max);
    vbias_hi = wasm_v128_and(vbias_hi, vexpw_max);
    vf_lo = wasm_f32x4_mul(vf_lo, vscale_to_zero);
    vf_hi = wasm_f32x4_mul(vf_hi, vscale_to_zero);
    const v128_t vnanmaskh = wasm_i16x8_narrow_i32x4(vnanmaskw_lo, vnanmaskw_hi);
    const v128_t vsignh = wasm_i16x8_narrow_i32x4(vsignx_lo, vsignx_hi);

    vbias_lo = wasm_i16x8_max(vbias_lo, vbias_min);
    vbias_hi = wasm_i16x8_max(vbias_hi, vbias_min);

    vf_lo = wasm_f32x4_add(vf_lo, vbias_lo);
    vf_hi = wasm_f32x4_add(vf_hi, vbias_hi);

    v128_t vexpw_lo = wasm_i32x4_shr(vf_lo, 13);
    v128_t vexpw_hi = wasm_i32x4_shr(vf_hi, 13);
    const v128_t vmantw_lo = wasm_v128_and(vf_lo, vmanth_mask);
    const v128_t vmantw_hi = wasm_v128_and(vf_hi, vmanth_mask);

    vexpw_lo = wasm_v128_and(vexpw_lo, vexph_mask);
    vexpw_hi = wasm_v128_and(vexpw_hi, vexph_mask);

    const v128_t vnonsignw_lo = wasm_i32x4_add(vmantw_lo, vexpw_lo);
    const v128_t vnonsignw_hi = wasm_i32x4_add(vmantw_hi, vexpw_hi);

    const v128_t vnonsignh = wasm_i16x8_narrow_i32x4(vnonsignw_lo, vnonsignw_hi);

    const v128_t vabsh = wasm_v128_bitselect(vnanh, vnonsignh, vnanmaskh);

    const v128_t vh = wasm_v128_or(vabsh, vsignh);

    wasm_v128_store(o, vh);
    o += 8;
  }
  if XNN_UNPREDICTABLE(batch != 0) {
    const v128_t vx_lo = wasm_v128_load(input);
    const float* input_hi = (const float*) ((uintptr_t) input + (batch & (4 * sizeof(float))));
    const v128_t vx_hi = wasm_v128_load(input_hi);

    const v128_t vabsx_lo = wasm_f32x4_abs(vx_lo);
    const v128_t vabsx_hi = wasm_f32x4_abs(vx_hi);

    const v128_t vsignx_lo = wasm_v128_xor(vx_lo, vabsx_lo);
    const v128_t vsignx_hi = wasm_v128_xor(vx_hi, vabsx_hi);
    v128_t vbias_lo = wasm_i32x4_add(vabsx_lo, vexp_bias);
    v128_t vbias_hi = wasm_i32x4_add(vabsx_hi, vexp_bias);
    v128_t vf_lo = wasm_f32x4_mul(vabsx_lo, vscale_to_inf);
    v128_t vf_hi = wasm_f32x4_mul(vabsx_hi, vscale_to_inf);
    const v128_t vnanmaskw_lo = wasm_i32x4_gt(vabsx_lo, vexpw_max);
    const v128_t vnanmaskw_hi = wasm_i32x4_gt(vabsx_hi, vexpw_max);

    vbias_lo = wasm_v128_and(vbias_lo, vexpw_max);
    vbias_hi = wasm_v128_and(vbias_hi, vexpw_max);
    vf_lo = wasm_f32x4_mul(vf_lo, vscale_to_zero);
    vf_hi = wasm_f32x4_mul(vf_hi, vscale_to_zero);
    const v128_t vnanmaskh = wasm_i16x8_narrow_i32x4(vnanmaskw_lo, vnanmaskw_hi);
    const v128_t vsignh = wasm_i16x8_narrow_i32x4(vsignx_lo, vsignx_hi);

    vbias_lo = wasm_i16x8_max(vbias_lo, vbias_min);
    vbias_hi = wasm_i16x8_max(vbias_hi, vbias_min);

    vf_lo = wasm_f32x4_add(vf_lo, vbias_lo);
    vf_hi = wasm_f32x4_add(vf_hi, vbias_hi);

    v128_t vexpw_lo = wasm_i32x4_shr(vf_lo, 13);
    v128_t vexpw_hi = wasm_i32x4_shr(vf_hi, 13);
    const v128_t vmantw_lo = wasm_v128_and(vf_lo, vmanth_mask);
    const v128_t vmantw_hi = wasm_v128_and(vf_hi, vmanth_mask);

    vexpw_lo = wasm_v128_and(vexpw_lo, vexph_mask);
    vexpw_hi = wasm_v128_and(vexpw_hi, vexph_mask);

    const v128_t vnonsignw_lo = wasm_i32x4_add(vmantw_lo, vexpw_lo);
    const v128_t vnonsignw_hi = wasm_i32x4_add(vmantw_hi, vexpw_hi);

    const v128_t vnonsignh = wasm_i16x8_narrow_i32x4(vnonsignw_lo, vnonsignw_hi);

    const v128_t vabsh = wasm_v128_bitselect(vnanh, vnonsignh, vnanmaskh);

    v128_t vh = wasm_v128_or(vabsh, vsignh);

    if (batch & (4 * sizeof(float))) {
      wasm_v128_store64_lane(o, vh, 0);
      vh = wasm_v64x2_shuffle(vh, vh, 1, 1);
      o += 4;
    }
    if (batch & (2 * sizeof(float))) {
      wasm_v128_store32_lane(o, vh, 0);
      vh = wasm_i64x2_shr(vh, 32);
      o += 2;
    }
    if (batch & (1 * sizeof(float))) {
      wasm_v128_store16_lane(o, vh, 0);
    }
  }
}