opelly/Stockfish
1
0
Fork 0
mirror of https://github.com/opelly27/Stockfish.git synced 2026-05-20 08:37:44 +00:00
Code Issues Packages Projects Releases Wiki Activity

Cleanup and optimize SSE/AVX code

Browse Source 
AVX512 +4% faster
AVX2 +1% faster
SSSE3 +5% faster

passed non-regression STC:
STC https://tests.stockfishchess.org/tests/view/5f31249f90816720665374f6
LLR: 2.96 (-2.94,2.94) {-1.50,0.50}
Total: 17576 W: 2344 L: 2245 D: 12987
Ptnml(0-2): 127, 1570, 5292, 1675, 124

closes https://github.com/official-stockfish/Stockfish/pull/2962

No functional change
This commit is contained in:
mstembera
2020-08-09 16:23:33 -07:00
committed by Joost VandeVondele
parent cb0504028e
commit f948cd008d
4 changed files with 41 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/nnue/nnue_feature_transformer.h
+12 -9
View File
@@ -169,38 +169,41 @@ namespace Eval::NNUE {
kHalfDimensions * sizeof(BiasType));
for (const auto index : active_indices[perspective]) {
const IndexType offset = kHalfDimensions * index;
#if defined(USE_AVX512)
auto accumulation = reinterpret_cast<__m512i*>(
&accumulator.accumulation[perspective][i][0]);
auto column = reinterpret_cast<const __m512i*>(&weights_[offset]);
constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
for (IndexType j = 0; j < kNumChunks; ++j)
_mm512_storeA_si512(&accumulation[j], _mm512_add_epi16(_mm512_loadA_si512(&accumulation[j]), column[j]));
#if defined(USE_AVX2)
#elif defined(USE_AVX2)
auto accumulation = reinterpret_cast<__m256i*>(
&accumulator.accumulation[perspective][i][0]);
auto column = reinterpret_cast<const __m256i*>(&weights_[offset]);
constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
for (IndexType j = 0; j < kNumChunks; ++j) {
for (IndexType j = 0; j < kNumChunks; ++j)
_mm256_storeA_si256(&accumulation[j], _mm256_add_epi16(_mm256_loadA_si256(&accumulation[j]), column[j]));
}
#elif defined(USE_SSE2)
auto accumulation = reinterpret_cast<__m128i*>(
&accumulator.accumulation[perspective][i][0]);
auto column = reinterpret_cast<const __m128i*>(&weights_[offset]);
constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
for (IndexType j = 0; j < kNumChunks; ++j) {
for (IndexType j = 0; j < kNumChunks; ++j)
accumulation[j] = _mm_add_epi16(accumulation[j], column[j]);
}
#elif defined(USE_NEON)
auto accumulation = reinterpret_cast<int16x8_t*>(
&accumulator.accumulation[perspective][i][0]);
auto column = reinterpret_cast<const int16x8_t*>(&weights_[offset]);
constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
for (IndexType j = 0; j < kNumChunks; ++j) {
for (IndexType j = 0; j < kNumChunks; ++j)
accumulation[j] = vaddq_s16(accumulation[j], column[j]);
}
#else
for (IndexType j = 0; j < kHalfDimensions; ++j) {
for (IndexType j = 0; j < kHalfDimensions; ++j)
accumulator.accumulation[perspective][i][j] += weights_[offset + j];
}
#endif
}