Files
Stockfish/src/nnue/nnue_feature_transformer.h

516 lines
18 KiB
C++

/*
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
Stockfish is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
Stockfish is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
// A class that converts the input features of the NNUE evaluation function
#ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED
#define NNUE_FEATURE_TRANSFORMER_H_INCLUDED
#include "nnue_common.h"
#include "nnue_architecture.h"
#include "features/index_list.h"
#include <cstring>
#include <string>
namespace Eval::NNUE {
// If vector instructions are enabled, we update and refresh the
// accumulator tile by tile such that each tile fits in the CPU's
// vector registers.
#define VECTOR
#ifdef USE_AVX512
typedef __m512i vec_t;
#define vec_load(a) _mm512_load_si512(a)
#define vec_store(a,b) _mm512_store_si512(a,b)
#define vec_add_16(a,b) _mm512_add_epi16(a,b)
#define vec_sub_16(a,b) _mm512_sub_epi16(a,b)
#define vec_zero _mm512_setzero_si512()
static constexpr IndexType kNumRegs = 8; // only 8 are needed
#elif USE_AVX2
typedef __m256i vec_t;
#define vec_load(a) _mm256_load_si256(a)
#define vec_store(a,b) _mm256_store_si256(a,b)
#define vec_add_16(a,b) _mm256_add_epi16(a,b)
#define vec_sub_16(a,b) _mm256_sub_epi16(a,b)
#define vec_zero _mm256_setzero_si256()
static constexpr IndexType kNumRegs = 16;
#elif USE_SSE2
typedef __m128i vec_t;
#define vec_load(a) (*(a))
#define vec_store(a,b) *(a)=(b)
#define vec_add_16(a,b) _mm_add_epi16(a,b)
#define vec_sub_16(a,b) _mm_sub_epi16(a,b)
#define vec_zero _mm_setzero_si128()
static constexpr IndexType kNumRegs = Is64Bit ? 16 : 8;
#elif USE_MMX
typedef __m64 vec_t;
#define vec_load(a) (*(a))
#define vec_store(a,b) *(a)=(b)
#define vec_add_16(a,b) _mm_add_pi16(a,b)
#define vec_sub_16(a,b) _mm_sub_pi16(a,b)
#define vec_zero _mm_setzero_si64()
static constexpr IndexType kNumRegs = 8;
#elif USE_NEON
typedef int16x8_t vec_t;
#define vec_load(a) (*(a))
#define vec_store(a,b) *(a)=(b)
#define vec_add_16(a,b) vaddq_s16(a,b)
#define vec_sub_16(a,b) vsubq_s16(a,b)
#define vec_zero {0}
static constexpr IndexType kNumRegs = 16;
#else
#undef VECTOR
#endif
// Input feature converter
class FeatureTransformer {
private:
// Number of output dimensions for one side
static constexpr IndexType kHalfDimensions = kTransformedFeatureDimensions;
#ifdef VECTOR
static constexpr IndexType kTileHeight = kNumRegs * sizeof(vec_t) / 2;
static_assert(kHalfDimensions % kTileHeight == 0, "kTileHeight must divide kHalfDimensions");
#endif
public:
// Output type
using OutputType = TransformedFeatureType;
// Number of input/output dimensions
static constexpr IndexType kInputDimensions = RawFeatures::kDimensions;
static constexpr IndexType kOutputDimensions = kHalfDimensions * 2;
// Size of forward propagation buffer
static constexpr std::size_t kBufferSize =
kOutputDimensions * sizeof(OutputType);
static constexpr int kLayerIndex = 0;
// Hash value embedded in the evaluation file
static constexpr std::uint32_t GetHashValue() {
return RawFeatures::kHashValue ^ kOutputDimensions;
}
static std::string get_name() {
return RawFeatures::get_name() + "[" +
std::to_string(kInputDimensions) + "->" +
std::to_string(kHalfDimensions) + "x2]";
}
// a string representing the structure
static std::string get_structure_string() {
return get_name();
}
static std::string get_layers_info() {
std::string info = " - ";
info += std::to_string(kLayerIndex);
info += " - ";
info += get_name();
return info;
}
// Read network parameters
bool ReadParameters(std::istream& stream) {
for (std::size_t i = 0; i < kHalfDimensions; ++i)
biases_[i] = read_little_endian<BiasType>(stream);
for (std::size_t i = 0; i < kHalfDimensions * kInputDimensions; ++i)
weights_[i] = read_little_endian<WeightType>(stream);
return !stream.fail();
}
// write parameters
bool WriteParameters(std::ostream& stream) const {
stream.write(reinterpret_cast<const char*>(biases_),
kHalfDimensions * sizeof(BiasType));
stream.write(reinterpret_cast<const char*>(weights_),
kHalfDimensions * kInputDimensions * sizeof(WeightType));
return !stream.fail();
}
// Proceed with the difference calculation if possible
bool update_accumulator_if_possible(const Position& pos) const {
const auto now = pos.state();
if (now->accumulator.computed_accumulation)
return true;
const auto prev = now->previous;
if (prev && prev->accumulator.computed_accumulation) {
update_accumulator(pos);
return true;
}
return false;
}
// Convert input features
void Transform(const Position& pos, OutputType* output) const {
if (!update_accumulator_if_possible(pos))
refresh_accumulator(pos);
const auto& accumulation = pos.state()->accumulator.accumulation;
#if defined(USE_AVX512)
constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth * 2);
static_assert(kHalfDimensions % (kSimdWidth * 2) == 0);
const __m512i kControl = _mm512_setr_epi64(0, 2, 4, 6, 1, 3, 5, 7);
const __m512i kZero = _mm512_setzero_si512();
#elif defined(USE_AVX2)
constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
constexpr int kControl = 0b11011000;
const __m256i kZero = _mm256_setzero_si256();
#elif defined(USE_SSE2)
constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
#ifdef USE_SSE41
const __m128i kZero = _mm_setzero_si128();
#else
const __m128i k0x80s = _mm_set1_epi8(-128);
#endif
#elif defined(USE_MMX)
constexpr IndexType kNumChunks = kHalfDimensions / kSimdWidth;
const __m64 k0x80s = _mm_set1_pi8(-128);
#elif defined(USE_NEON)
constexpr IndexType kNumChunks = kHalfDimensions / (kSimdWidth / 2);
const int8x8_t kZero = {0};
#endif
const Color perspectives[2] = {pos.side_to_move(), ~pos.side_to_move()};
for (IndexType p = 0; p < 2; ++p) {
const IndexType offset = kHalfDimensions * p;
#if defined(USE_AVX512)
auto out = reinterpret_cast<__m512i*>(&output[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
__m512i sum0 = _mm512_load_si512(
&reinterpret_cast<const __m512i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
__m512i sum1 = _mm512_load_si512(
&reinterpret_cast<const __m512i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
for (IndexType i = 1; i < kRefreshTriggers.size(); ++i) {
sum0 = _mm512_add_epi16(sum0, reinterpret_cast<const __m512i*>(
accumulation[perspectives[p]][i])[j * 2 + 0]);
sum1 = _mm512_add_epi16(sum1, reinterpret_cast<const __m512i*>(
accumulation[perspectives[p]][i])[j * 2 + 1]);
}
_mm512_store_si512(&out[j], _mm512_permutexvar_epi64(kControl,
_mm512_max_epi8(_mm512_packs_epi16(sum0, sum1), kZero)));
}
#elif defined(USE_AVX2)
auto out = reinterpret_cast<__m256i*>(&output[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
__m256i sum0 = _mm256_load_si256(
&reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 0]);
__m256i sum1 = _mm256_load_si256(
&reinterpret_cast<const __m256i*>(accumulation[perspectives[p]][0])[j * 2 + 1]);
for (IndexType i = 1; i < kRefreshTriggers.size(); ++i) {
sum0 = _mm256_add_epi16(sum0, reinterpret_cast<const __m256i*>(
accumulation[perspectives[p]][i])[j * 2 + 0]);
sum1 = _mm256_add_epi16(sum1, reinterpret_cast<const __m256i*>(
accumulation[perspectives[p]][i])[j * 2 + 1]);
}
_mm256_store_si256(&out[j], _mm256_permute4x64_epi64(_mm256_max_epi8(
_mm256_packs_epi16(sum0, sum1), kZero), kControl));
}
#elif defined(USE_SSE2)
auto out = reinterpret_cast<__m128i*>(&output[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
__m128i sum0 = _mm_load_si128(&reinterpret_cast<const __m128i*>(
accumulation[perspectives[p]][0])[j * 2 + 0]);
__m128i sum1 = _mm_load_si128(&reinterpret_cast<const __m128i*>(
accumulation[perspectives[p]][0])[j * 2 + 1]);
for (IndexType i = 1; i < kRefreshTriggers.size(); ++i) {
sum0 = _mm_add_epi16(sum0, reinterpret_cast<const __m128i*>(
accumulation[perspectives[p]][i])[j * 2 + 0]);
sum1 = _mm_add_epi16(sum1, reinterpret_cast<const __m128i*>(
accumulation[perspectives[p]][i])[j * 2 + 1]);
}
const __m128i packedbytes = _mm_packs_epi16(sum0, sum1);
_mm_store_si128(&out[j],
#ifdef USE_SSE41
_mm_max_epi8(packedbytes, kZero)
#else
_mm_subs_epi8(_mm_adds_epi8(packedbytes, k0x80s), k0x80s)
#endif
);
}
#elif defined(USE_MMX)
auto out = reinterpret_cast<__m64*>(&output[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
__m64 sum0 = *(&reinterpret_cast<const __m64*>(
accumulation[perspectives[p]][0])[j * 2 + 0]);
__m64 sum1 = *(&reinterpret_cast<const __m64*>(
accumulation[perspectives[p]][0])[j * 2 + 1]);
for (IndexType i = 1; i < kRefreshTriggers.size(); ++i) {
sum0 = _mm_add_pi16(sum0, reinterpret_cast<const __m64*>(
accumulation[perspectives[p]][i])[j * 2 + 0]);
sum1 = _mm_add_pi16(sum1, reinterpret_cast<const __m64*>(
accumulation[perspectives[p]][i])[j * 2 + 1]);
}
const __m64 packedbytes = _mm_packs_pi16(sum0, sum1);
out[j] = _mm_subs_pi8(_mm_adds_pi8(packedbytes, k0x80s), k0x80s);
}
#elif defined(USE_NEON)
const auto out = reinterpret_cast<int8x8_t*>(&output[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
int16x8_t sum = reinterpret_cast<const int16x8_t*>(
accumulation[perspectives[p]][0])[j];
for (IndexType i = 1; i < kRefreshTriggers.size(); ++i) {
sum = vaddq_s16(sum, reinterpret_cast<const int16x8_t*>(
accumulation[perspectives[p]][i])[j]);
}
out[j] = vmax_s8(vqmovn_s16(sum), kZero);
}
#else
for (IndexType j = 0; j < kHalfDimensions; ++j) {
BiasType sum = accumulation[static_cast<int>(perspectives[p])][0][j];
for (IndexType i = 1; i < kRefreshTriggers.size(); ++i) {
sum += accumulation[static_cast<int>(perspectives[p])][i][j];
}
output[offset + j] = static_cast<OutputType>(
std::max<int>(0, std::min<int>(127, sum)));
}
#endif
}
#if defined(USE_MMX)
_mm_empty();
#endif
}
private:
// Calculate cumulative value without using difference calculation
void refresh_accumulator(const Position& pos) const {
#ifdef VECTOR
// Gcc-10.2 unnecessarily spills AVX2 registers if this array
// is defined in the VECTOR code below, once in each branch
vec_t acc[kNumRegs];
#endif
auto& accumulator = pos.state()->accumulator;
for (IndexType i = 0; i < kRefreshTriggers.size(); ++i) {
Features::IndexList active_indices[2];
RawFeatures::append_active_indices(pos, kRefreshTriggers[i],
active_indices);
for (Color perspective : { WHITE, BLACK }) {
#ifdef VECTOR
for (IndexType j = 0; j < kHalfDimensions / kTileHeight; ++j) {
auto accTile = reinterpret_cast<vec_t*>(
&accumulator.accumulation[perspective][i][j * kTileHeight]);
if (i == 0) {
auto biasesTile = reinterpret_cast<const vec_t*>(
&biases_[j * kTileHeight]);
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = biasesTile[k];
} else {
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = vec_zero;
}
for (const auto index : active_indices[perspective]) {
const IndexType offset = kHalfDimensions * index + j * kTileHeight;
auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = vec_add_16(acc[k], column[k]);
}
for (IndexType k = 0; k < kNumRegs; k++)
vec_store(&accTile[k], acc[k]);
}
#else
if (i == 0) {
std::memcpy(accumulator.accumulation[perspective][i], biases_,
kHalfDimensions * sizeof(BiasType));
} else {
std::memset(accumulator.accumulation[perspective][i], 0,
kHalfDimensions * sizeof(BiasType));
}
for (const auto index : active_indices[perspective]) {
const IndexType offset = kHalfDimensions * index;
for (IndexType j = 0; j < kHalfDimensions; ++j)
accumulator.accumulation[perspective][i][j] += weights_[offset + j];
}
#endif
}
}
#if defined(USE_MMX)
_mm_empty();
#endif
accumulator.computed_accumulation = true;
}
// Calculate cumulative value using difference calculation
void update_accumulator(const Position& pos) const {
#ifdef VECTOR
// Gcc-10.2 unnecessarily spills AVX2 registers if this array
// is defined in the VECTOR code below, once in each branch
vec_t acc[kNumRegs];
#endif
const auto& prev_accumulator = pos.state()->previous->accumulator;
auto& accumulator = pos.state()->accumulator;
for (IndexType i = 0; i < kRefreshTriggers.size(); ++i) {
Features::IndexList removed_indices[2], added_indices[2];
bool reset[2] = { false, false };
RawFeatures::append_changed_indices(pos, kRefreshTriggers[i],
removed_indices, added_indices, reset);
#ifdef VECTOR
for (IndexType j = 0; j < kHalfDimensions / kTileHeight; ++j) {
for (Color perspective : { WHITE, BLACK }) {
auto accTile = reinterpret_cast<vec_t*>(
&accumulator.accumulation[perspective][i][j * kTileHeight]);
if (reset[perspective]) {
if (i == 0) {
auto biasesTile = reinterpret_cast<const vec_t*>(
&biases_[j * kTileHeight]);
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = biasesTile[k];
} else {
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = vec_zero;
}
} else {
auto prevAccTile = reinterpret_cast<const vec_t*>(
&prev_accumulator.accumulation[perspective][i][j * kTileHeight]);
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = vec_load(&prevAccTile[k]);
// Difference calculation for the deactivated features
for (const auto index : removed_indices[perspective]) {
const IndexType offset = kHalfDimensions * index + j * kTileHeight;
auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = vec_sub_16(acc[k], column[k]);
}
}
{ // Difference calculation for the activated features
for (const auto index : added_indices[perspective]) {
const IndexType offset = kHalfDimensions * index + j * kTileHeight;
auto column = reinterpret_cast<const vec_t*>(&weights_[offset]);
for (IndexType k = 0; k < kNumRegs; ++k)
acc[k] = vec_add_16(acc[k], column[k]);
}
}
for (IndexType k = 0; k < kNumRegs; ++k)
vec_store(&accTile[k], acc[k]);
}
}
#if defined(USE_MMX)
_mm_empty();
#endif
#else
for (Color perspective : { WHITE, BLACK }) {
if (reset[perspective]) {
if (i == 0) {
std::memcpy(accumulator.accumulation[perspective][i], biases_,
kHalfDimensions * sizeof(BiasType));
} else {
std::memset(accumulator.accumulation[perspective][i], 0,
kHalfDimensions * sizeof(BiasType));
}
} else {
std::memcpy(accumulator.accumulation[perspective][i],
prev_accumulator.accumulation[perspective][i],
kHalfDimensions * sizeof(BiasType));
// Difference calculation for the deactivated features
for (const auto index : removed_indices[perspective]) {
const IndexType offset = kHalfDimensions * index;
for (IndexType j = 0; j < kHalfDimensions; ++j)
accumulator.accumulation[perspective][i][j] -= weights_[offset + j];
}
}
{ // Difference calculation for the activated features
for (const auto index : added_indices[perspective]) {
const IndexType offset = kHalfDimensions * index;
for (IndexType j = 0; j < kHalfDimensions; ++j)
accumulator.accumulation[perspective][i][j] += weights_[offset + j];
}
}
}
#endif
}
accumulator.computed_accumulation = true;
}
using BiasType = std::int16_t;
using WeightType = std::int16_t;
// Make the learning class a friend
friend class Trainer<FeatureTransformer>;
alignas(kCacheLineSize) BiasType biases_[kHalfDimensions];
alignas(kCacheLineSize)
WeightType weights_[kHalfDimensions * kInputDimensions];
};
} // namespace Eval::NNUE
#endif // #ifndef NNUE_FEATURE_TRANSFORMER_H_INCLUDED