File size: 12,684 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
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
// Copyright 2023 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 P == H + 1 or P == H + 2
$assert not PS or (P, H) == (4, 3)
$assert DIV in ["DIV", "RCP"]
$assert SAT in ["MINMAX", "SELECT"]
$assert AVX != 2 or FMA == 3
$assert BATCH_TILE % 8 == 0
$assert BATCH_TILE >= 8
$SIMD_TILE = BATCH_TILE // 8
#include <assert.h>
#include <stddef.h>
#include <stdint.h>

#include <immintrin.h>

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

$POLY_SUFFIX = "p%dh%d%s" % (P, H, "ps" if PS else "ts")
$DIV_SUFFIX = DIV.lower()
$PARAMS_STRUCT = "avx_expm1minus_rr1_" + ("p%dh%d" % (P, H))
$ISA = "avx2" if AVX == 2 else "fma3" if FMA == 3 else "f16c"
void xnn_f16_vtanh_ukernel__${ISA}_expm1minus_rr1_${POLY_SUFFIX}_${DIV_SUFFIX}_x${BATCH_TILE}(
    size_t batch,
    const void* input,
    void* output,
    const union xnn_f16_tanh_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 __m128i vsign_mask = _mm_load_si128((const __m128i*) params->${PARAMS_STRUCT}.sign_mask);
  const __m256 vsat_cutoff = _mm256_load_ps(params->${PARAMS_STRUCT}.sat_cutoff);
  const __m256 vlog2e = _mm256_load_ps(params->${PARAMS_STRUCT}.log2e);
  const __m256 vmagic_bias = _mm256_load_ps(params->${PARAMS_STRUCT}.magic_bias);
  const __m256 vminus_ln2 = _mm256_load_ps(params->${PARAMS_STRUCT}.minus_ln2);
  $for i in reversed(range(2, P+1)):
    const __m256 vc${i} = _mm256_load_ps(params->${PARAMS_STRUCT}.c${i});
  $if P != H + 1:
    const __m256 vminus_one = _mm256_load_ps(params->${PARAMS_STRUCT}.minus_one);
  const __m256 vtwo = _mm256_load_ps(params->${PARAMS_STRUCT}.two);
  $if P == H + 1:
    const __m256 vminus_one = _mm256_load_ps(params->${PARAMS_STRUCT}.minus_one);

  const uint16_t* i = (const uint16_t*) input;
  uint16_t* o = (uint16_t*) output;
  $if BATCH_TILE > 8:
    for (; batch >= ${BATCH_TILE} * sizeof(uint16_t); batch -= ${BATCH_TILE} * sizeof(uint16_t)) {
      const __m128i vx0 = _mm_loadu_si128((const __m128i*) i);
      $for N in range(1, SIMD_TILE):
        const __m128i vx${N} = _mm_loadu_si128((const __m128i*) (i + ${N * 8}));
      i += ${BATCH_TILE};

      $for N in range(SIMD_TILE):
        const __m128i vabsx${N} = _mm_or_si128(vx${N}, vsign_mask);

      $for N in range(SIMD_TILE):
        __m256 vz${N} = _mm256_cvtph_ps(vabsx${N});
        const __m128i vinvsignx${N} = _mm_xor_si128(vx${N}, vabsx${N});

      $for N in range(SIMD_TILE):
        $if SAT == "MINMAX":
          vz${N} = _mm256_max_ps(vsat_cutoff, vz${N});
        $elif SAT == "SELECT":
          const __m256 vm${N} = _mm256_cmp_ps(vz${N}, vsat_cutoff, _CMP_LE_OS);
        $if FMA == 3:
          __m256 vn${N} = _mm256_fmadd_ps(vz${N}, vlog2e, vmagic_bias);
        $else:
          __m256 vn${N} = _mm256_add_ps(_mm256_mul_ps(vz${N}, vlog2e), vmagic_bias);

      $if AVX == 1:
        $for N in range(SIMD_TILE):
          const __m128 vn${N}_hi = _mm256_extractf128_ps(vn${N}, 1);
          __m256 vs${N} = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn${N})), 23)));
          vn${N} = _mm256_sub_ps(vn${N}, vmagic_bias);

        $for N in range(SIMD_TILE):
          const __m128 vs${N}_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn${N}_hi), 23));

        $for N in range(SIMD_TILE):
          vs${N} = _mm256_insertf128_ps(vs${N}, vs${N}_hi, 1);
      $else:
        $for N in range(SIMD_TILE):
          const __m256 vs${N} = _mm256_castsi256_ps(_mm256_slli_epi32(_mm256_castps_si256(vn${N}), 23));
          vn${N} = _mm256_sub_ps(vn${N}, vmagic_bias);

      $for N in range(SIMD_TILE):
        $if FMA == 3:
          const __m256 vt${N} = _mm256_fmadd_ps(vn${N}, vminus_ln2, vz${N});
        $else:
          const __m256 vt${N} = _mm256_add_ps(_mm256_mul_ps(vn${N}, vminus_ln2), vz${N});

      $if FMA == 3:
        $for N in range(SIMD_TILE):
          __m256 vp${N} = vc${P};
        $for i in reversed(range(2, P)):
          $for N in range(SIMD_TILE):
            vp${N} = _mm256_fmadd_ps(vp${N}, vt${N}, vc${i});
      $else:
        $for N in range(SIMD_TILE):
          __m256 vp${N} = _mm256_add_ps(_mm256_mul_ps(vc${P}, vt${N}), vc${P-1});
        $for i in reversed(range(2, P-1)):
          $for N in range(SIMD_TILE):
            vp${N} = _mm256_add_ps(_mm256_mul_ps(vp${N}, vt${N}), vc${i});
      $if P == H + 1:
        $for N in range(SIMD_TILE):
          $if FMA == 3:
            vp${N} = _mm256_fmadd_ps(vp${N}, vt${N}, vtwo);
          $else:
            vp${N} = _mm256_add_ps(_mm256_mul_ps(vp${N}, vt${N}), vtwo);
      $else:
        $for N in range(SIMD_TILE):
          vp${N} = _mm256_mul_ps(vp${N}, vt${N});

      $for N in range(SIMD_TILE):
        const __m256 vts${N} = _mm256_mul_ps(vt${N}, vs${N});
        const __m256 vsmo${N} = _mm256_add_ps(vs${N}, vminus_one);
      $if P == H + 1:
        $for N in range(SIMD_TILE):
          $if FMA == 3:
            const __m256 vemo${N} = _mm256_fmadd_ps(vp${N}, vts${N}, vsmo${N});
          $else:
            const __m256 vemo${N} = _mm256_add_ps(_mm256_mul_ps(vp${N}, vts${N}), vsmo${N});
      $else:
        $if FMA == 3:
          $for N in range(SIMD_TILE):
            vp${N} = _mm256_fmadd_ps(vp${N}, vts${N}, vts${N});
          $for N in range(SIMD_TILE):
            const __m256 vemo${N} = _mm256_fmadd_ps(vp${N}, vtwo${N}, vsmo${N});
        $else:
          $for N in range(SIMD_TILE):
            vp${N} = _mm256_add_ps(_mm256_mul_ps(vp${N}, vts${N}), vts${N});
          $for N in range(SIMD_TILE):
            const __m256 vemo${N} = _mm256_add_ps(_mm256_mul_ps(vp${N}, vtwo), vsmo${N});

      $for N in range(SIMD_TILE):
        const __m256 vepo${N} = _mm256_add_ps(vemo${N}, vtwo);

      $if DIV == "DIV":
        $for N in range(SIMD_TILE):
          __m256 vy${N} = _mm256_div_ps(vemo${N}, vepo${N});
      $else:
        $for N in range(SIMD_TILE):
          __m256 vrepo${N} = _mm256_rcp_ps(vepo${N});

        $for N in range(SIMD_TILE):
          __m256 vy${N} = _mm256_mul_ps(vemo${N}, vrepo${N});

      $if SAT == "SELECT":
        $for N in range(SIMD_TILE):
          vy${N} = _mm256_blendv_ps(vy${N}, vminus_one, vm${N});

      $for N in range(SIMD_TILE):
        __m128i vh${N} = _mm256_cvtps_ph(vy${N}, _MM_FROUND_TO_NEAREST_INT);
      $for N in range(SIMD_TILE):
        vh${N} = _mm_xor_si128(vh${N}, vinvsignx${N});

      _mm_storeu_si128((__m128i*) o, vh0);
      $for N in range(1, SIMD_TILE):
        _mm_storeu_si128((__m128i*) (o + ${N * 8}), vh${N});
      o += ${BATCH_TILE};
    }
  for (; batch >= 8 * sizeof(uint16_t); batch -= 8 * sizeof(uint16_t)) {
    const __m128i vx = _mm_loadu_si128((const __m128i*) i);
    i += 8;

    const __m128i vabsx = _mm_or_si128(vx, vsign_mask);
    __m256 vz = _mm256_cvtph_ps(vabsx);

    const __m128i vinvsignx = _mm_xor_si128(vx, vabsx);
    $if SAT == "MINMAX":
      vz = _mm256_max_ps(vsat_cutoff, vz);
    $elif SAT == "SELECT":
      const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS);

    $if FMA == 3:
      __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias);
    $else:
      __m256 vn = _mm256_add_ps(_mm256_mul_ps(vz, vlog2e), vmagic_bias);

    $if AVX == 1:
      const __m128 vn_hi = _mm256_extractf128_ps(vn, 1);
      __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23)));
      const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23));
      vs = _mm256_insertf128_ps(vs, vs_hi, 1);
    $else:
      const __m256 vs = _mm256_castsi256_ps(_mm256_slli_epi32(_mm256_castps_si256(vn), 23));

    vn = _mm256_sub_ps(vn, vmagic_bias);

    $if FMA == 3:
      const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz);
    $else:
      const __m256 vt = _mm256_add_ps(_mm256_mul_ps(vn, vminus_ln2), vz);

    $if FMA == 3:
      __m256 vp = vc${P};
      $for i in reversed(range(2, P)):
        vp = _mm256_fmadd_ps(vp, vt, vc${i});
    $else:
      __m256 vp = _mm256_add_ps(_mm256_mul_ps(vc${P}, vt), vc${P-1});
      $for i in reversed(range(2, P-1)):
        vp = _mm256_add_ps(_mm256_mul_ps(vp, vt), vc${i});
    $if P == H + 1:
      $if FMA == 3:
        vp = _mm256_fmadd_ps(vp, vt, vtwo);
      $else:
        vp = _mm256_add_ps(_mm256_mul_ps(vp, vt), vtwo);
    $else:
      vp = _mm256_mul_ps(vp, vt);

    const __m256 vts = _mm256_mul_ps(vt, vs);
    const __m256 vsmo = _mm256_add_ps(vs, vminus_one);
    $if P == H + 1:
      $if FMA == 3:
        const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo);
      $else:
        const __m256 vemo = _mm256_add_ps(_mm256_mul_ps(vp, vts), vsmo);
    $else:
      $if FMA == 3:
        vp = _mm256_fmadd_ps(vp, vts, vts);
        const __m256 vemo = _mm256_fmadd_ps(vp, vtwo, vsmo);
      $else:
        vp = _mm256_add_ps(_mm256_mul_ps(vp, vts), vts);
        const __m256 vemo = _mm256_add_ps(_mm256_mul_ps(vp, vtwo), vsmo);

    const __m256 vepo = _mm256_add_ps(vemo, vtwo);

    $if DIV == "DIV":
      __m256 vy = _mm256_div_ps(vemo, vepo);
    $else:
      __m256 vrepo = _mm256_rcp_ps(vepo);

      __m256 vy = _mm256_mul_ps(vemo, vrepo);

    $if SAT == "SELECT":
      vy = _mm256_blendv_ps(vy, vminus_one, vm);

    __m128i vh = _mm256_cvtps_ph(vy, _MM_FROUND_TO_NEAREST_INT);
    vh = _mm_xor_si128(vh, vinvsignx);

    _mm_storeu_si128((__m128i*) o, vh);
    o += 8;
  }
  if (batch != 0) {
    const __m128i vx = _mm_loadu_si128((const __m128i*) i);

    const __m128i vabsx = _mm_or_si128(vx, vsign_mask);
    __m256 vz = _mm256_cvtph_ps(vabsx);

    const __m128i vinvsignx = _mm_xor_si128(vx, vabsx);
    $if SAT == "MINMAX":
      vz = _mm256_max_ps(vsat_cutoff, vz);
    $elif SAT == "SELECT":
      const __m256 vm = _mm256_cmp_ps(vz, vsat_cutoff, _CMP_LE_OS);

    $if FMA == 3:
      __m256 vn = _mm256_fmadd_ps(vz, vlog2e, vmagic_bias);
    $else:
      __m256 vn = _mm256_add_ps(_mm256_mul_ps(vz, vlog2e), vmagic_bias);

    $if AVX == 1:
      const __m128 vn_hi = _mm256_extractf128_ps(vn, 1);
      __m256 vs = _mm256_castps128_ps256(_mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 23)));
      const __m128 vs_hi = _mm_castsi128_ps(_mm_slli_epi32(_mm_castps_si128(vn_hi), 23));
      vs = _mm256_insertf128_ps(vs, vs_hi, 1);
    $else:
      const __m256 vs = _mm256_castsi256_ps(_mm256_slli_epi32(_mm256_castps_si256(vn), 23));

    vn = _mm256_sub_ps(vn, vmagic_bias);

    $if FMA == 3:
      const __m256 vt = _mm256_fmadd_ps(vn, vminus_ln2, vz);
    $else:
      const __m256 vt = _mm256_add_ps(_mm256_mul_ps(vn, vminus_ln2), vz);

    $if FMA == 3:
      __m256 vp = vc${P};
      $for i in reversed(range(2, P)):
        vp = _mm256_fmadd_ps(vp, vt, vc${i});
    $else:
      __m256 vp = _mm256_add_ps(_mm256_mul_ps(vc${P}, vt), vc${P-1});
      $for i in reversed(range(2, P-1)):
        vp = _mm256_add_ps(_mm256_mul_ps(vp, vt), vc${i});
    $if P == H + 1:
      $if FMA == 3:
        vp = _mm256_fmadd_ps(vp, vt, vtwo);
      $else:
        vp = _mm256_add_ps(_mm256_mul_ps(vp, vt), vtwo);
    $else:
      vp = _mm256_mul_ps(vp, vt);

    const __m256 vts = _mm256_mul_ps(vt, vs);
    const __m256 vsmo = _mm256_add_ps(vs, vminus_one);
    $if P == H + 1:
      $if FMA == 3:
        const __m256 vemo = _mm256_fmadd_ps(vp, vts, vsmo);
      $else:
        const __m256 vemo = _mm256_add_ps(_mm256_mul_ps(vp, vts), vsmo);
    $else:
      $if FMA == 3:
        vp = _mm256_fmadd_ps(vp, vts, vts);
        const __m256 vemo = _mm256_fmadd_ps(vp, vtwo, vsmo);
      $else:
        vp = _mm256_add_ps(_mm256_mul_ps(vp, vts), vts);
        const __m256 vemo = _mm256_add_ps(_mm256_mul_ps(vp, vtwo), vsmo);

    const __m256 vepo = _mm256_add_ps(vemo, vtwo);

    $if DIV == "DIV":
      __m256 vy = _mm256_div_ps(vemo, vepo);
    $else:
      __m256 vrepo = _mm256_rcp_ps(vepo);

      __m256 vy = _mm256_mul_ps(vemo, vrepo);

    $if SAT == "SELECT":
      vy = _mm256_blendv_ps(vy, vminus_one, vm);

    __m128i vh = _mm256_cvtps_ph(vy, _MM_FROUND_TO_NEAREST_INT);
    vh = _mm_xor_si128(vh, vinvsignx);

    if (batch & (4 * sizeof(uint16_t))) {
      _mm_storel_epi64((__m128i*) o, vh);
      vh = _mm_unpackhi_epi64(vh, vh);
      o += 4;
    }
    if (batch & (2 * sizeof(uint16_t))) {
      _mm_storeu_si32(o, vh);
      vh = _mm_srli_epi64(vh, 32);
      o += 2;
    }
    if (batch & (1 * sizeof(uint16_t))) {
      *o = (uint16_t) _mm_extract_epi16(vh, 0);
    }
  }
}