/*
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
- Copyright (C) 2004-2020 The Stockfish developers (see AUTHORS file)
+ Copyright (C) 2004-2023 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
#ifndef NNUE_COMMON_H_INCLUDED
#define NNUE_COMMON_H_INCLUDED
+#include <algorithm>
+#include <cassert>
+#include <cstdint>
+#include <cstring>
+#include <iostream>
+#include <type_traits>
+
+#include "../misc.h"
+
#if defined(USE_AVX2)
#include <immintrin.h>
#elif defined(USE_SSE2)
#include <emmintrin.h>
-#elif defined(USE_MMX)
-#include <mmintrin.h>
-
#elif defined(USE_NEON)
#include <arm_neon.h>
#endif
-// HACK: Use _mm256_loadu_si256() instead of _mm256_load_si256. Otherwise a binary
-// compiled with older g++ crashes because the output memory is not aligned
-// even though alignas is specified.
-#if defined(USE_AVX2)
-#if defined(__GNUC__ ) && (__GNUC__ < 9)
-#define _mm256_loadA_si256 _mm256_loadu_si256
-#define _mm256_storeA_si256 _mm256_storeu_si256
-#else
-#define _mm256_loadA_si256 _mm256_load_si256
-#define _mm256_storeA_si256 _mm256_store_si256
-#endif
-#endif
-
-#if defined(USE_AVX512)
-#if defined(__GNUC__ ) && (__GNUC__ < 9)
-#define _mm512_loadA_si512 _mm512_loadu_si512
-#define _mm512_storeA_si512 _mm512_storeu_si512
-#else
-#define _mm512_loadA_si512 _mm512_load_si512
-#define _mm512_storeA_si512 _mm512_store_si512
-#endif
-#endif
-
-namespace Eval::NNUE {
+namespace Stockfish::Eval::NNUE {
// Version of the evaluation file
- constexpr std::uint32_t kVersion = 0x7AF32F16u;
+ constexpr std::uint32_t Version = 0x7AF32F20u;
// Constant used in evaluation value calculation
- constexpr int FV_SCALE = 16;
- constexpr int kWeightScaleBits = 6;
+ constexpr int OutputScale = 16;
+ constexpr int WeightScaleBits = 6;
// Size of cache line (in bytes)
- constexpr std::size_t kCacheLineSize = 64;
+ constexpr std::size_t CacheLineSize = 64;
+
+ constexpr const char Leb128MagicString[] = "COMPRESSED_LEB128";
+ constexpr const std::size_t Leb128MagicStringSize = sizeof(Leb128MagicString) - 1;
// SIMD width (in bytes)
#if defined(USE_AVX2)
- constexpr std::size_t kSimdWidth = 32;
+ constexpr std::size_t SimdWidth = 32;
#elif defined(USE_SSE2)
- constexpr std::size_t kSimdWidth = 16;
-
- #elif defined(USE_MMX)
- constexpr std::size_t kSimdWidth = 8;
+ constexpr std::size_t SimdWidth = 16;
#elif defined(USE_NEON)
- constexpr std::size_t kSimdWidth = 16;
+ constexpr std::size_t SimdWidth = 16;
#endif
- constexpr std::size_t kMaxSimdWidth = 32;
+ constexpr std::size_t MaxSimdWidth = 32;
// Type of input feature after conversion
using TransformedFeatureType = std::uint8_t;
// Round n up to be a multiple of base
template <typename IntType>
- constexpr IntType CeilToMultiple(IntType n, IntType base) {
- return (n + base - 1) / base * base;
+ constexpr IntType ceil_to_multiple(IntType n, IntType base) {
+ return (n + base - 1) / base * base;
+ }
+
+
+ // read_little_endian() is our utility to read an integer (signed or unsigned, any size)
+ // from a stream in little-endian order. We swap the byte order after the read if
+ // necessary to return a result with the byte ordering of the compiling machine.
+ template <typename IntType>
+ inline IntType read_little_endian(std::istream& stream) {
+ IntType result;
+
+ if (IsLittleEndian)
+ stream.read(reinterpret_cast<char*>(&result), sizeof(IntType));
+ else
+ {
+ std::uint8_t u[sizeof(IntType)];
+ std::make_unsigned_t<IntType> v = 0;
+
+ stream.read(reinterpret_cast<char*>(u), sizeof(IntType));
+ for (std::size_t i = 0; i < sizeof(IntType); ++i)
+ v = (v << 8) | u[sizeof(IntType) - i - 1];
+
+ std::memcpy(&result, &v, sizeof(IntType));
+ }
+
+ return result;
+ }
+
+
+ // write_little_endian() is our utility to write an integer (signed or unsigned, any size)
+ // to a stream in little-endian order. We swap the byte order before the write if
+ // necessary to always write in little endian order, independently of the byte
+ // ordering of the compiling machine.
+ template <typename IntType>
+ inline void write_little_endian(std::ostream& stream, IntType value) {
+
+ if (IsLittleEndian)
+ stream.write(reinterpret_cast<const char*>(&value), sizeof(IntType));
+ else
+ {
+ std::uint8_t u[sizeof(IntType)];
+ std::make_unsigned_t<IntType> v = value;
+
+ std::size_t i = 0;
+ // if constexpr to silence the warning about shift by 8
+ if constexpr (sizeof(IntType) > 1)
+ {
+ for (; i + 1 < sizeof(IntType); ++i)
+ {
+ u[i] = (std::uint8_t)v;
+ v >>= 8;
+ }
+ }
+ u[i] = (std::uint8_t)v;
+
+ stream.write(reinterpret_cast<char*>(u), sizeof(IntType));
+ }
+ }
+
+
+ // read_little_endian(s, out, N) : read integers in bulk from a little indian stream.
+ // This reads N integers from stream s and put them in array out.
+ template <typename IntType>
+ inline void read_little_endian(std::istream& stream, IntType* out, std::size_t count) {
+ if (IsLittleEndian)
+ stream.read(reinterpret_cast<char*>(out), sizeof(IntType) * count);
+ else
+ for (std::size_t i = 0; i < count; ++i)
+ out[i] = read_little_endian<IntType>(stream);
+ }
+
+
+ // write_little_endian(s, values, N) : write integers in bulk to a little indian stream.
+ // This takes N integers from array values and writes them on stream s.
+ template <typename IntType>
+ inline void write_little_endian(std::ostream& stream, const IntType* values, std::size_t count) {
+ if (IsLittleEndian)
+ stream.write(reinterpret_cast<const char*>(values), sizeof(IntType) * count);
+ else
+ for (std::size_t i = 0; i < count; ++i)
+ write_little_endian<IntType>(stream, values[i]);
+ }
+
+
+ // read_leb_128(s, out, N) : read N signed integers from the stream s, putting them in
+ // the array out. The stream is assumed to be compressed using the signed LEB128 format.
+ // See https://en.wikipedia.org/wiki/LEB128 for a description of the compression scheme.
+ template <typename IntType>
+ inline void read_leb_128(std::istream& stream, IntType* out, std::size_t count) {
+
+ // Check the presence of our LEB128 magic string
+ char leb128MagicString[Leb128MagicStringSize];
+ stream.read(leb128MagicString, Leb128MagicStringSize);
+ assert(strncmp(Leb128MagicString, leb128MagicString, Leb128MagicStringSize) == 0);
+
+ static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
+
+ const std::uint32_t BUF_SIZE = 4096;
+ std::uint8_t buf[BUF_SIZE];
+
+ auto bytes_left = read_little_endian<std::uint32_t>(stream);
+
+ std::uint32_t buf_pos = BUF_SIZE;
+ for (std::size_t i = 0; i < count; ++i)
+ {
+ IntType result = 0;
+ size_t shift = 0;
+ do
+ {
+ if (buf_pos == BUF_SIZE)
+ {
+ stream.read(reinterpret_cast<char*>(buf), std::min(bytes_left, BUF_SIZE));
+ buf_pos = 0;
+ }
+
+ std::uint8_t byte = buf[buf_pos++];
+ --bytes_left;
+ result |= (byte & 0x7f) << shift;
+ shift += 7;
+
+ if ((byte & 0x80) == 0)
+ {
+ out[i] = (sizeof(IntType) * 8 <= shift || (byte & 0x40) == 0) ? result
+ : result | ~((1 << shift) - 1);
+ break;
+ }
+ }
+ while (shift < sizeof(IntType) * 8);
+ }
+
+ assert(bytes_left == 0);
+ }
+
+
+ // write_leb_128(s, values, N) : write signed integers to a stream with LEB128 compression.
+ // This takes N integers from array values, compress them with the LEB128 algorithm and
+ // writes the result on the stream s.
+ // See https://en.wikipedia.org/wiki/LEB128 for a description of the compression scheme.
+ template <typename IntType>
+ inline void write_leb_128(std::ostream& stream, const IntType* values, std::size_t count) {
+
+ // Write our LEB128 magic string
+ stream.write(Leb128MagicString, Leb128MagicStringSize);
+
+ static_assert(std::is_signed_v<IntType>, "Not implemented for unsigned types");
+
+ std::uint32_t byte_count = 0;
+ for (std::size_t i = 0; i < count; ++i)
+ {
+ IntType value = values[i];
+ std::uint8_t byte;
+ do
+ {
+ byte = value & 0x7f;
+ value >>= 7;
+ ++byte_count;
+ }
+ while ((byte & 0x40) == 0 ? value != 0 : value != -1);
+ }
+
+ write_little_endian(stream, byte_count);
+
+ const std::uint32_t BUF_SIZE = 4096;
+ std::uint8_t buf[BUF_SIZE];
+ std::uint32_t buf_pos = 0;
+
+ auto flush = [&]() {
+ if (buf_pos > 0)
+ {
+ stream.write(reinterpret_cast<char*>(buf), buf_pos);
+ buf_pos = 0;
+ }
+ };
+
+ auto write = [&](std::uint8_t byte) {
+ buf[buf_pos++] = byte;
+ if (buf_pos == BUF_SIZE)
+ flush();
+ };
+
+ for (std::size_t i = 0; i < count; ++i)
+ {
+ IntType value = values[i];
+ while (true)
+ {
+ std::uint8_t byte = value & 0x7f;
+ value >>= 7;
+ if ((byte & 0x40) == 0 ? value == 0 : value == -1)
+ {
+ write(byte);
+ break;
+ }
+ write(byte | 0x80);
+ }
+ }
+
+ flush();
}
-} // namespace Eval::NNUE
+} // namespace Stockfish::Eval::NNUE
#endif // #ifndef NNUE_COMMON_H_INCLUDED
projects / stockfish / blobdiff