git.sesse.net Git - stockfish/blob - src/nnue/nnue_feature_transformer.h

   1 /*
   2   Stockfish, a UCI chess playing engine derived from Glaurung 2.1
   3   Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
   4
   5   Stockfish is free software: you can redistribute it and/or modify
   6   it under the terms of the GNU General Public License as published by
   7   the Free Software Foundation, either version 3 of the License, or
   8   (at your option) any later version.
   9
  10   Stockfish is distributed in the hope that it will be useful,
  11   but WITHOUT ANY WARRANTY; without even the implied warranty of
  12   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13   GNU General Public License for more details.
  14
  15   You should have received a copy of the GNU General Public License
  16   along with this program.  If not, see <http://www.gnu.org/licenses/>.
  17 */
  18
  19 // A class that converts the input features of the NNUE evaluation function
  20
  21 #ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED
  22 #define NNUE_FEATURE_TRANSFORMER_H_INCLUDED
  23
  24 #include "nnue_common.h"
  25 #include "nnue_architecture.h"
  26 #include "features/index_list.h"
  27
  28 #include <cstring> // std::memset()
  29
  30 namespace Eval::NNUE {
  31
  32   // Input feature converter
  33   class FeatureTransformer {
  34
  35    private:
  36     // Number of output dimensions for one side
  37     static constexpr IndexType kHalfDimensions = kTransformedFeatureDimensions;
  38
  39    public:
  40     // Output type
  41     using OutputType = TransformedFeatureType;
  42
  43     // Number of input/output dimensions
  44     static constexpr IndexType kInputDimensions = RawFeatures::kDimensions;
  45     static constexpr IndexType kOutputDimensions = kHalfDimensions * 2;
  46
  47     // Size of forward propagation buffer
  48     static constexpr std::size_t kBufferSize =
  49         kOutputDimensions * sizeof(OutputType);
  50
  51     // Hash value embedded in the evaluation file
  52     static constexpr std::uint32_t GetHashValue() {
  53       return RawFeatures::kHashValue ^ kOutputDimensions;
  54     }
  55
  56     // Read network parameters
  57     bool ReadParameters(std::istream& stream) {
  58       stream.read(reinterpret_cast<char*>(biases_),
  59                   kHalfDimensions * sizeof(BiasType));
  60       stream.read(reinterpret_cast<char*>(weights_),
  61                   kHalfDimensions * kInputDimensions * sizeof(WeightType));
  62       return !stream.fail();
  63     }
  64
  65     // Proceed with the difference calculation if possible
  66     bool UpdateAccumulatorIfPossible(const Position& pos) const {
  67       const auto now = pos.state();
  68       if (now->accumulator.computed_accumulation) {
  69         return true;
  70       }
  71       const auto prev = now->previous;
  72       if (prev && prev->accumulator.computed_accumulation) {
  73         UpdateAccumulator(pos);
  74         return true;
  75       }
  76       return false;
  77     }
  78
  79     // Convert input features
  80     void Transform(const Position& pos, OutputType* output, bool refresh) const {
  81       if (refresh || !UpdateAccumulatorIfPossible(pos)) {
  82         RefreshAccumulator(pos);
  83       }
  84       const auto& accumulation = pos.state()->accumulator.accumulation;
  85
  86   #if defined(USE_AVX2)
  87       constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
  88       constexpr int kControl = 0b11011000;
  89       const __m256i kZero = _mm256_setzero_si256();
  90
  91   #elif defined(USE_SSSE3)
  92       constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
  93
  94   #ifdef USE_SSE41
  95       const __m128i kZero = _mm_setzero_si128();
  96   #else
  97       const __m128i k0x80s = _mm_set1_epi8(-128);
  98   #endif
  99
 100   #elif defined(USE_NEON)
 101       constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 102       const int8x8_t kZero = {0};
 103   #endif
 104
 105       const Color perspectives[2] = {pos.side_to_move(), ~pos.side_to_move()};
 106       for (IndexType p = 0; p < 2; ++p) {
 107         const IndexType offset = kHalfDimensions * p;
 108
 109   #if defined(USE_AVX2)
 110         auto out = reinterpret_cast<__m256i*>(&output[offset]);
 111         for (IndexType j = 0; j < kNumChunks; ++j) {
 112           __m256i sum0 = _mm256_loadA_si256(
 113               &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
 114           __m256i sum1 = _mm256_loadA_si256(
 115             &reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
 116           _mm256_storeA_si256(&out[j], _mm256_permute4x64_epi64(_mm256_max_epi8(
 117               _mm256_packs_epi16(sum0, sum1), kZero), kControl));
 118         }
 119
 120   #elif defined(USE_SSSE3)
 121         auto out = reinterpret_cast<__m128i*>(&output[offset]);
 122         for (IndexType j = 0; j < kNumChunks; ++j) {
 123           __m128i sum0 = _mm_load_si128(&reinterpret_cast<const __m128i*>(
 124               accumulation[perspectives[p]][0])[j * 2 + 0]);
 125           __m128i sum1 = _mm_load_si128(&reinterpret_cast<const __m128i*>(
 126               accumulation[perspectives[p]][0])[j * 2 + 1]);
 127       const __m128i packedbytes = _mm_packs_epi16(sum0, sum1);
 128
 129           _mm_store_si128(&out[j],
 130
 131   #ifdef USE_SSE41
 132             _mm_max_epi8(packedbytes, kZero)
 133   #else
 134             _mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
 135   #endif
 136
 137           );
 138         }
 139
 140   #elif defined(USE_NEON)
 141         const auto out = reinterpret_cast<int8x8_t*>(&output[offset]);
 142         for (IndexType j = 0; j < kNumChunks; ++j) {
 143           int16x8_t sum = reinterpret_cast<const int16x8_t*>(
 144               accumulation[perspectives[p]][0])[j];
 145           out[j] = vmax_s8(vqmovn_s16(sum), kZero);
 146         }
 147
 148   #else
 149         for (IndexType j = 0; j < kHalfDimensions; ++j) {
 150           BiasType sum = accumulation[static_cast<int>(perspectives[p])][0][j];
 151           output[offset + j] = static_cast<OutputType>(
 152               std::max<int>(0, std::min<int>(127, sum)));
 153         }
 154   #endif
 155
 156       }
 157     }
 158
 159    private:
 160     // Calculate cumulative value without using difference calculation
 161     void RefreshAccumulator(const Position& pos) const {
 162       auto& accumulator = pos.state()->accumulator;
 163       IndexType i = 0;
 164       Features::IndexList active_indices[2];
 165       RawFeatures::AppendActiveIndices(pos, kRefreshTriggers[i],
 166                                        active_indices);
 167       for (Color perspective : { WHITE, BLACK }) {
 168         std::memcpy(accumulator.accumulation[perspective][i], biases_,
 169                    kHalfDimensions * sizeof(BiasType));
 170         for (const auto index : active_indices[perspective]) {
 171           const IndexType offset = kHalfDimensions * index;
 172   #if defined(USE_AVX512)
 173           auto accumulation = reinterpret_cast<__m512i*>(
 174               &accumulator.accumulation[perspective][i][0]);
 175           auto column = reinterpret_cast<const __m512i*>(&weights_[offset]);
 176           constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
 177           for (IndexType j = 0; j < kNumChunks; ++j)
 178             _mm512_storeA_si512(&accumulation[j], _mm512_add_epi16(_mm512_loadA_si512(&accumulation[j]), column[j]));
 179
 180   #elif defined(USE_AVX2)
 181           auto accumulation = reinterpret_cast<__m256i*>(
 182               &accumulator.accumulation[perspective][i][0]);
 183           auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
 184           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 185           for (IndexType j = 0; j < kNumChunks; ++j)
 186             _mm256_storeA_si256(&accumulation[j], _mm256_add_epi16(_mm256_loadA_si256(&accumulation[j]), column[j]));
 187
 188   #elif defined(USE_SSE2)
 189           auto accumulation = reinterpret_cast<__m128i*>(
 190               &accumulator.accumulation[perspective][i][0]);
 191           auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
 192           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 193           for (IndexType j = 0; j < kNumChunks; ++j)
 194             accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
 195
 196   #elif defined(USE_NEON)
 197           auto accumulation = reinterpret_cast<int16x8_t*>(
 198               &accumulator.accumulation[perspective][i][0]);
 199           auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
 200           constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 201           for (IndexType j = 0; j < kNumChunks; ++j)
 202             accumulation[j] = vaddq_s16(accumulation[j], column[j]);
 203
 204   #else
 205           for (IndexType j = 0; j < kHalfDimensions; ++j)
 206             accumulator.accumulation[perspective][i][j] += weights_[offset + j];
 207   #endif
 208
 209         }
 210       }
 211
 212       accumulator.computed_accumulation = true;
 213       accumulator.computed_score = false;
 214     }
 215
 216     // Calculate cumulative value using difference calculation
 217     void UpdateAccumulator(const Position& pos) const {
 218       const auto prev_accumulator = pos.state()->previous->accumulator;
 219       auto& accumulator = pos.state()->accumulator;
 220       IndexType i = 0;
 221       Features::IndexList removed_indices[2], added_indices[2];
 222       bool reset[2];
 223       RawFeatures::AppendChangedIndices(pos, kRefreshTriggers[i],
 224                                         removed_indices, added_indices, reset);
 225       for (Color perspective : { WHITE, BLACK }) {
 226
 227   #if defined(USE_AVX2)
 228         constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 229         auto accumulation = reinterpret_cast<__m256i*>(
 230             &accumulator.accumulation[perspective][i][0]);
 231
 232   #elif defined(USE_SSE2)
 233         constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 234         auto accumulation = reinterpret_cast<__m128i*>(
 235             &accumulator.accumulation[perspective][i][0]);
 236
 237   #elif defined(USE_NEON)
 238         constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
 239         auto accumulation = reinterpret_cast<int16x8_t*>(
 240             &accumulator.accumulation[perspective][i][0]);
 241   #endif
 242
 243         if (reset[perspective]) {
 244           std::memcpy(accumulator.accumulation[perspective][i], biases_,
 245                       kHalfDimensions * sizeof(BiasType));
 246         } else {
 247           std::memcpy(accumulator.accumulation[perspective][i],
 248                       prev_accumulator.accumulation[perspective][i],
 249                       kHalfDimensions * sizeof(BiasType));
 250           // Difference calculation for the deactivated features
 251           for (const auto index : removed_indices[perspective]) {
 252             const IndexType offset = kHalfDimensions * index;
 253
 254   #if defined(USE_AVX2)
 255             auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
 256             for (IndexType j = 0; j < kNumChunks; ++j) {
 257               accumulation[j] = _mm256_sub_epi16(accumulation[j], column[j]);
 258             }
 259
 260   #elif defined(USE_SSE2)
 261             auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
 262             for (IndexType j = 0; j < kNumChunks; ++j) {
 263               accumulation[j] = _mm_sub_epi16(accumulation[j], column[j]);
 264             }
 265
 266   #elif defined(USE_NEON)
 267             auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
 268             for (IndexType j = 0; j < kNumChunks; ++j) {
 269               accumulation[j] = vsubq_s16(accumulation[j], column[j]);
 270             }
 271
 272   #else
 273             for (IndexType j = 0; j < kHalfDimensions; ++j) {
 274               accumulator.accumulation[perspective][i][j] -=
 275                   weights_[offset + j];
 276             }
 277   #endif
 278
 279           }
 280         }
 281         { // Difference calculation for the activated features
 282           for (const auto index : added_indices[perspective]) {
 283             const IndexType offset = kHalfDimensions * index;
 284
 285   #if defined(USE_AVX2)
 286             auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
 287             for (IndexType j = 0; j < kNumChunks; ++j) {
 288               accumulation[j] = _mm256_add_epi16(accumulation[j], column[j]);
 289             }
 290
 291   #elif defined(USE_SSE2)
 292             auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
 293             for (IndexType j = 0; j < kNumChunks; ++j) {
 294               accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
 295             }
 296
 297   #elif defined(USE_NEON)
 298             auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
 299             for (IndexType j = 0; j < kNumChunks; ++j) {
 300               accumulation[j] = vaddq_s16(accumulation[j], column[j]);
 301             }
 302
 303   #else
 304             for (IndexType j = 0; j < kHalfDimensions; ++j) {
 305               accumulator.accumulation[perspective][i][j] +=
 306                   weights_[offset + j];
 307             }
 308   #endif
 309
 310           }
 311         }
 312       }
 313
 314       accumulator.computed_accumulation = true;
 315       accumulator.computed_score = false;
 316     }
 317
 318     using BiasType = std::int16_t;
 319     using WeightType = std::int16_t;
 320
 321     alignas(kCacheLineSize) BiasType biases_[kHalfDimensions];
 322     alignas(kCacheLineSize)
 323         WeightType weights_[kHalfDimensions * kInputDimensions];
 324   };
 325
 326 }  // namespace Eval::NNUE
 327
 328 #endif // #ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED