File size: 9,096 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
// Auto-generated file. Do not edit!
//   Template: src/f16-igemm/avx2-broadcast.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 <immintrin.h>

#include <xnnpack/igemm.h>
#include <xnnpack/intrinsics-polyfill.h>


void xnn_f16_igemm_minmax_ukernel_4x16__avx2_broadcast(
    size_t mr,
    size_t nc,
    size_t kc,
    size_t ks,
    const void** restrict a,
    const void* restrict w,
    void* restrict c,
    size_t cm_stride,
    size_t cn_stride,
    size_t a_offset,
    const void* zero,
    const union xnn_f16_minmax_params params[restrict XNN_MIN_ELEMENTS(1)])
{
  assert(mr != 0);
  assert(mr <= 4);
  assert(nc != 0);
  assert(kc != 0);
  assert(kc % sizeof(uint16_t) == 0);
  assert(ks != 0);
  assert(ks % (4 * sizeof(void*)) == 0);
  assert(a_offset % sizeof(uint16_t) == 0);
  assert(a != NULL);
  assert(w != NULL);
  assert(c != NULL);

  uint16_t* c0 = c;
  uint16_t* c1 = (uint16_t*) ((uintptr_t) c0 + cm_stride);
  if XNN_UNPREDICTABLE(mr < 2) {
    c1 = c0;
  }
  uint16_t* c2 = (uint16_t*) ((uintptr_t) c1 + cm_stride);
  if XNN_UNPREDICTABLE(mr <= 2) {
    c2 = c1;
  }
  uint16_t* c3 = (uint16_t*) ((uintptr_t) c2 + cm_stride);
  if XNN_UNPREDICTABLE(mr != 4) {
    c3 = c2;
  }

  do {
    __m256 vacc0x01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w));
    __m256 vacc0x89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) ((const uint16_t*) w + 8)));
    __m256 vacc1x01234567 = vacc0x01234567;
    __m256 vacc1x89ABCDEF = vacc0x89ABCDEF;
    __m256 vacc2x01234567 = vacc0x01234567;
    __m256 vacc2x89ABCDEF = vacc0x89ABCDEF;
    __m256 vacc3x01234567 = vacc0x01234567;
    __m256 vacc3x89ABCDEF = vacc0x89ABCDEF;
    w = (const uint16_t*) w + 16;

    size_t p = ks;
    do {
      const uint16_t* restrict a0 = (const uint16_t*) a[0];
      assert(a0 != NULL);
      if XNN_UNPREDICTABLE(a0 != zero) {
        a0 = (const uint16_t*) ((uintptr_t) a0 + a_offset);
      }
      const uint16_t* restrict a1 = (const uint16_t*) a[1];
      assert(a1 != NULL);
      if XNN_UNPREDICTABLE(a1 != zero) {
        a1 = (const uint16_t*) ((uintptr_t) a1 + a_offset);
      }
      const uint16_t* restrict a2 = (const uint16_t*) a[2];
      assert(a2 != NULL);
      if XNN_UNPREDICTABLE(a2 != zero) {
        a2 = (const uint16_t*) ((uintptr_t) a2 + a_offset);
      }
      const uint16_t* restrict a3 = (const uint16_t*) a[3];
      assert(a3 != NULL);
      if XNN_UNPREDICTABLE(a3 != zero) {
        a3 = (const uint16_t*) ((uintptr_t) a3 + a_offset);
      }
      a += 4;

      size_t k = kc;
      do {
        const __m256 vb01234567 = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) w));
        const __m256 vb89ABCDEF = _mm256_cvtph_ps(_mm_load_si128((const __m128i*) ((const uint16_t*) w + 8)));
        w = (const uint16_t*) w + 16;

        const __m256 va0 = _mm256_cvtph_ps(_mm_set1_epi16((short) *a0));
        a0 += 1;
        const __m256 va1 = _mm256_cvtph_ps(_mm_set1_epi16((short) *a1));
        a1 += 1;
        const __m256 va2 = _mm256_cvtph_ps(_mm_set1_epi16((short) *a2));
        a2 += 1;
        const __m256 va3 = _mm256_cvtph_ps(_mm_set1_epi16((short) *a3));
        a3 += 1;

        vacc0x01234567 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va0, vb01234567, vacc0x01234567), _MM_FROUND_TO_NEAREST_INT));
        vacc0x89ABCDEF = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va0, vb89ABCDEF, vacc0x89ABCDEF), _MM_FROUND_TO_NEAREST_INT));
        vacc1x01234567 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va1, vb01234567, vacc1x01234567), _MM_FROUND_TO_NEAREST_INT));
        vacc1x89ABCDEF = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va1, vb89ABCDEF, vacc1x89ABCDEF), _MM_FROUND_TO_NEAREST_INT));
        vacc2x01234567 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va2, vb01234567, vacc2x01234567), _MM_FROUND_TO_NEAREST_INT));
        vacc2x89ABCDEF = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va2, vb89ABCDEF, vacc2x89ABCDEF), _MM_FROUND_TO_NEAREST_INT));
        vacc3x01234567 = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va3, vb01234567, vacc3x01234567), _MM_FROUND_TO_NEAREST_INT));
        vacc3x89ABCDEF = _mm256_cvtph_ps(_mm256_cvtps_ph(_mm256_fmadd_ps(va3, vb89ABCDEF, vacc3x89ABCDEF), _MM_FROUND_TO_NEAREST_INT));

        k -= sizeof(uint16_t);
      } while (k != 0);
      p -= 4 * sizeof(void*);
    } while (p != 0);

    const __m256 vmin = _mm256_load_ps(params->avx.min);
    vacc0x01234567 = _mm256_max_ps(vacc0x01234567, vmin);
    vacc1x01234567 = _mm256_max_ps(vacc1x01234567, vmin);
    vacc2x01234567 = _mm256_max_ps(vacc2x01234567, vmin);
    vacc3x01234567 = _mm256_max_ps(vacc3x01234567, vmin);
    vacc0x89ABCDEF = _mm256_max_ps(vacc0x89ABCDEF, vmin);
    vacc1x89ABCDEF = _mm256_max_ps(vacc1x89ABCDEF, vmin);
    vacc2x89ABCDEF = _mm256_max_ps(vacc2x89ABCDEF, vmin);
    vacc3x89ABCDEF = _mm256_max_ps(vacc3x89ABCDEF, vmin);

    const __m256 vmax = _mm256_load_ps(params->avx.max);
    vacc0x01234567 = _mm256_min_ps(vacc0x01234567, vmax);
    vacc1x01234567 = _mm256_min_ps(vacc1x01234567, vmax);
    vacc2x01234567 = _mm256_min_ps(vacc2x01234567, vmax);
    vacc3x01234567 = _mm256_min_ps(vacc3x01234567, vmax);
    vacc0x89ABCDEF = _mm256_min_ps(vacc0x89ABCDEF, vmax);
    vacc1x89ABCDEF = _mm256_min_ps(vacc1x89ABCDEF, vmax);
    vacc2x89ABCDEF = _mm256_min_ps(vacc2x89ABCDEF, vmax);
    vacc3x89ABCDEF = _mm256_min_ps(vacc3x89ABCDEF, vmax);

    if XNN_LIKELY(nc >= 16) {
      _mm_storeu_si128((__m128i*) c3, _mm256_cvtps_ph(vacc3x01234567, _MM_FROUND_TO_NEAREST_INT));
      _mm_storeu_si128((__m128i*) (c3 + 8), _mm256_cvtps_ph(vacc3x89ABCDEF, _MM_FROUND_TO_NEAREST_INT));
      c3 = (uint16_t*) ((uintptr_t) c3 + cn_stride);
      _mm_storeu_si128((__m128i*) c2, _mm256_cvtps_ph(vacc2x01234567, _MM_FROUND_TO_NEAREST_INT));
      _mm_storeu_si128((__m128i*) (c2 + 8), _mm256_cvtps_ph(vacc2x89ABCDEF, _MM_FROUND_TO_NEAREST_INT));
      c2 = (uint16_t*) ((uintptr_t) c2 + cn_stride);
      _mm_storeu_si128((__m128i*) c1, _mm256_cvtps_ph(vacc1x01234567, _MM_FROUND_TO_NEAREST_INT));
      _mm_storeu_si128((__m128i*) (c1 + 8), _mm256_cvtps_ph(vacc1x89ABCDEF, _MM_FROUND_TO_NEAREST_INT));
      c1 = (uint16_t*) ((uintptr_t) c1 + cn_stride);
      _mm_storeu_si128((__m128i*) c0, _mm256_cvtps_ph(vacc0x01234567, _MM_FROUND_TO_NEAREST_INT));
      _mm_storeu_si128((__m128i*) (c0 + 8), _mm256_cvtps_ph(vacc0x89ABCDEF, _MM_FROUND_TO_NEAREST_INT));
      c0 = (uint16_t*) ((uintptr_t) c0 + cn_stride);

      a = (const void**restrict) ((uintptr_t) a - ks);
      nc -= 16;
    } else {
      __m128i vh3x01234567 = _mm256_cvtps_ph(vacc3x01234567, _MM_FROUND_TO_NEAREST_INT);
      __m128i vh2x01234567 = _mm256_cvtps_ph(vacc2x01234567, _MM_FROUND_TO_NEAREST_INT);
      __m128i vh1x01234567 = _mm256_cvtps_ph(vacc1x01234567, _MM_FROUND_TO_NEAREST_INT);
      __m128i vh0x01234567 = _mm256_cvtps_ph(vacc0x01234567, _MM_FROUND_TO_NEAREST_INT);
      if (nc & 8) {
        _mm_storeu_si128((__m128i*) c3, vh3x01234567);
        _mm_storeu_si128((__m128i*) c2, vh2x01234567);
        _mm_storeu_si128((__m128i*) c1, vh1x01234567);
        _mm_storeu_si128((__m128i*) c0, vh0x01234567);

        vh3x01234567 = _mm256_cvtps_ph(vacc3x89ABCDEF, _MM_FROUND_TO_NEAREST_INT);
        vh2x01234567 = _mm256_cvtps_ph(vacc2x89ABCDEF, _MM_FROUND_TO_NEAREST_INT);
        vh1x01234567 = _mm256_cvtps_ph(vacc1x89ABCDEF, _MM_FROUND_TO_NEAREST_INT);
        vh0x01234567 = _mm256_cvtps_ph(vacc0x89ABCDEF, _MM_FROUND_TO_NEAREST_INT);

        c3 += 8;
        c2 += 8;
        c1 += 8;
        c0 += 8;
      }
      if (nc & 4) {
        _mm_storel_epi64((__m128i*) c3, vh3x01234567);
        _mm_storel_epi64((__m128i*) c2, vh2x01234567);
        _mm_storel_epi64((__m128i*) c1, vh1x01234567);
        _mm_storel_epi64((__m128i*) c0, vh0x01234567);

        vh3x01234567 = _mm_unpackhi_epi64(vh3x01234567, vh3x01234567);
        vh2x01234567 = _mm_unpackhi_epi64(vh2x01234567, vh2x01234567);
        vh1x01234567 = _mm_unpackhi_epi64(vh1x01234567, vh1x01234567);
        vh0x01234567 = _mm_unpackhi_epi64(vh0x01234567, vh0x01234567);

        c3 += 4;
        c2 += 4;
        c1 += 4;
        c0 += 4;
      }
      if (nc & 2) {
        _mm_storeu_si32(c3, vh3x01234567);
        _mm_storeu_si32(c2, vh2x01234567);
        _mm_storeu_si32(c1, vh1x01234567);
        _mm_storeu_si32(c0, vh0x01234567);

        vh3x01234567 = _mm_srli_epi64(vh3x01234567, 32);
        vh2x01234567 = _mm_srli_epi64(vh2x01234567, 32);
        vh1x01234567 = _mm_srli_epi64(vh1x01234567, 32);
        vh0x01234567 = _mm_srli_epi64(vh0x01234567, 32);

        c3 += 2;
        c2 += 2;
        c1 += 2;
        c0 += 2;
      }
      if (nc & 1) {
        *c3 = _mm_extract_epi16(vh3x01234567, 0);
        *c2 = _mm_extract_epi16(vh2x01234567, 0);
        *c1 = _mm_extract_epi16(vh1x01234567, 0);
        *c0 = _mm_extract_epi16(vh0x01234567, 0);
      }

      nc = 0;
    }
  } while (nc != 0);
}