X-Git-Url: https://git.sesse.net/?p=stockfish;a=blobdiff_plain;f=src%2Fmisc.cpp;h=d9c0679bf4e3704de73bcd99c94ffe54db39351a;hp=04c3c939f76d98229db4c570e089f92375657d31;hb=a24f28be8567c2527b154ef981090368a2bd8f76;hpb=5066e17eba4b9d4217a8d173de56e40e0e05fc6d diff --git a/src/misc.cpp b/src/misc.cpp index 04c3c939..d9c0679b 100644 --- a/src/misc.cpp +++ b/src/misc.cpp @@ -2,6 +2,7 @@ Stockfish, a UCI chess playing engine derived from Glaurung 2.1 Copyright (C) 2004-2008 Tord Romstad (Glaurung author) Copyright (C) 2008-2015 Marco Costalba, Joona Kiiski, Tord Romstad + Copyright (C) 2015-2020 Marco Costalba, Joona Kiiski, Gary Linscott, Tord Romstad Stockfish is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by @@ -17,10 +18,39 @@ along with this program. If not, see . */ +#ifdef _WIN32 +#if _WIN32_WINNT < 0x0601 +#undef _WIN32_WINNT +#define _WIN32_WINNT 0x0601 // Force to include needed API prototypes +#endif + +#ifndef NOMINMAX +#define NOMINMAX +#endif + +#include +// The needed Windows API for processor groups could be missed from old Windows +// versions, so instead of calling them directly (forcing the linker to resolve +// the calls at compile time), try to load them at runtime. To do this we need +// first to define the corresponding function pointers. +extern "C" { +typedef bool(*fun1_t)(LOGICAL_PROCESSOR_RELATIONSHIP, + PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, PDWORD); +typedef bool(*fun2_t)(USHORT, PGROUP_AFFINITY); +typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY); +} +#endif + #include #include #include #include +#include + +#if defined(__linux__) && !defined(__ANDROID__) +#include +#include +#endif #include "misc.h" #include "thread.h" @@ -39,14 +69,14 @@ const string Version = ""; /// usual I/O functionality, all without changing a single line of code! /// Idea from http://groups.google.com/group/comp.lang.c++/msg/1d941c0f26ea0d81 -struct Tie: public streambuf { // MSVC requires splitted streambuf for cin and cout +struct Tie: public streambuf { // MSVC requires split streambuf for cin and cout Tie(streambuf* b, streambuf* l) : buf(b), logBuf(l) {} - int sync() { return logBuf->pubsync(), buf->pubsync(); } - int overflow(int c) { return log(buf->sputc((char)c), "<< "); } - int underflow() { return buf->sgetc(); } - int uflow() { return log(buf->sbumpc(), ">> "); } + int sync() override { return logBuf->pubsync(), buf->pubsync(); } + int overflow(int c) override { return log(buf->sputc((char)c), "<< "); } + int underflow() override { return buf->sgetc(); } + int uflow() override { return log(buf->sbumpc(), ">> "); } streambuf *buf, *logBuf; @@ -64,23 +94,30 @@ struct Tie: public streambuf { // MSVC requires splitted streambuf for cin and c class Logger { Logger() : in(cin.rdbuf(), file.rdbuf()), out(cout.rdbuf(), file.rdbuf()) {} - ~Logger() { start(false); } + ~Logger() { start(""); } ofstream file; Tie in, out; public: - static void start(bool b) { + static void start(const std::string& fname) { static Logger l; - if (b && !l.file.is_open()) + if (!fname.empty() && !l.file.is_open()) { - l.file.open("io_log.txt", ifstream::out); + l.file.open(fname, ifstream::out); + + if (!l.file.is_open()) + { + cerr << "Unable to open debug log file " << fname << endl; + exit(EXIT_FAILURE); + } + cin.rdbuf(&l.in); cout.rdbuf(&l.out); } - else if (!b && l.file.is_open()) + else if (fname.empty() && l.file.is_open()) { cout.rdbuf(l.out.buf); cin.rdbuf(l.in.buf); @@ -108,6 +145,7 @@ const string engine_info(bool to_uci) { { date >> month >> day >> year; ss << setw(2) << day << setw(2) << (1 + months.find(month) / 4) << year.substr(2); + ss << "-asn"; } ss << (Is64Bit ? " 64" : "") @@ -119,8 +157,79 @@ const string engine_info(bool to_uci) { } +/// compiler_info() returns a string trying to describe the compiler we use + +const std::string compiler_info() { + + #define stringify2(x) #x + #define stringify(x) stringify2(x) + #define make_version_string(major, minor, patch) stringify(major) "." stringify(minor) "." stringify(patch) + +/// Predefined macros hell: +/// +/// __GNUC__ Compiler is gcc, Clang or Intel on Linux +/// __INTEL_COMPILER Compiler is Intel +/// _MSC_VER Compiler is MSVC or Intel on Windows +/// _WIN32 Building on Windows (any) +/// _WIN64 Building on Windows 64 bit + + std::string compiler = "\nCompiled by "; + + #ifdef __clang__ + compiler += "clang++ "; + compiler += make_version_string(__clang_major__, __clang_minor__, __clang_patchlevel__); + #elif __INTEL_COMPILER + compiler += "Intel compiler "; + compiler += "(version "; + compiler += stringify(__INTEL_COMPILER) " update " stringify(__INTEL_COMPILER_UPDATE); + compiler += ")"; + #elif _MSC_VER + compiler += "MSVC "; + compiler += "(version "; + compiler += stringify(_MSC_FULL_VER) "." stringify(_MSC_BUILD); + compiler += ")"; + #elif __GNUC__ + compiler += "g++ (GNUC) "; + compiler += make_version_string(__GNUC__, __GNUC_MINOR__, __GNUC_PATCHLEVEL__); + #else + compiler += "Unknown compiler "; + compiler += "(unknown version)"; + #endif + + #if defined(__APPLE__) + compiler += " on Apple"; + #elif defined(__CYGWIN__) + compiler += " on Cygwin"; + #elif defined(__MINGW64__) + compiler += " on MinGW64"; + #elif defined(__MINGW32__) + compiler += " on MinGW32"; + #elif defined(__ANDROID__) + compiler += " on Android"; + #elif defined(__linux__) + compiler += " on Linux"; + #elif defined(_WIN64) + compiler += " on Microsoft Windows 64-bit"; + #elif defined(_WIN32) + compiler += " on Microsoft Windows 32-bit"; + #else + compiler += " on unknown system"; + #endif + + compiler += "\n __VERSION__ macro expands to: "; + #ifdef __VERSION__ + compiler += __VERSION__; + #else + compiler += "(undefined macro)"; + #endif + compiler += "\n"; + + return compiler; +} + + /// Debug functions used mainly to collect run-time statistics -static int64_t hits[2], means[2]; +static std::atomic hits[2], means[2]; void dbg_hit_on(bool b) { ++hits[0]; if (b) ++hits[1]; } void dbg_hit_on(bool c, bool b) { if (c) dbg_hit_on(b); } @@ -143,7 +252,7 @@ void dbg_print() { std::ostream& operator<<(std::ostream& os, SyncCout sc) { - static Mutex m; + static std::mutex m; if (sc == IO_LOCK) m.lock(); @@ -156,7 +265,7 @@ std::ostream& operator<<(std::ostream& os, SyncCout sc) { /// Trampoline helper to avoid moving Logger to misc.h -void start_logger(bool b) { Logger::start(b); } +void start_logger(const std::string& fname) { Logger::start(fname); } /// prefetch() preloads the given address in L1/L2 cache. This is a non-blocking @@ -184,3 +293,236 @@ void prefetch(void* addr) { } #endif + + +/// aligned_ttmem_alloc() will return suitably aligned memory, and if possible use large pages. +/// The returned pointer is the aligned one, while the mem argument is the one that needs +/// to be passed to free. With c++17 some of this functionality could be simplified. + +#if defined(__linux__) && !defined(__ANDROID__) + +void* aligned_ttmem_alloc(size_t allocSize, void*& mem) { + + constexpr size_t alignment = 2 * 1024 * 1024; // assumed 2MB page sizes + size_t size = ((allocSize + alignment - 1) / alignment) * alignment; // multiple of alignment + if (posix_memalign(&mem, alignment, size)) + mem = nullptr; + madvise(mem, allocSize, MADV_HUGEPAGE); + return mem; +} + +#elif defined(_WIN64) + +static void* aligned_ttmem_alloc_large_pages(size_t allocSize) { + + HANDLE hProcessToken { }; + LUID luid { }; + void* mem = nullptr; + + const size_t largePageSize = GetLargePageMinimum(); + if (!largePageSize) + return nullptr; + + // We need SeLockMemoryPrivilege, so try to enable it for the process + if (!OpenProcessToken(GetCurrentProcess(), TOKEN_ADJUST_PRIVILEGES | TOKEN_QUERY, &hProcessToken)) + return nullptr; + + if (LookupPrivilegeValue(NULL, SE_LOCK_MEMORY_NAME, &luid)) + { + TOKEN_PRIVILEGES tp { }; + TOKEN_PRIVILEGES prevTp { }; + DWORD prevTpLen = 0; + + tp.PrivilegeCount = 1; + tp.Privileges[0].Luid = luid; + tp.Privileges[0].Attributes = SE_PRIVILEGE_ENABLED; + + // Try to enable SeLockMemoryPrivilege. Note that even if AdjustTokenPrivileges() succeeds, + // we still need to query GetLastError() to ensure that the privileges were actually obtained. + if (AdjustTokenPrivileges( + hProcessToken, FALSE, &tp, sizeof(TOKEN_PRIVILEGES), &prevTp, &prevTpLen) && + GetLastError() == ERROR_SUCCESS) + { + // Round up size to full pages and allocate + allocSize = (allocSize + largePageSize - 1) & ~size_t(largePageSize - 1); + mem = VirtualAlloc( + NULL, allocSize, MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES, PAGE_READWRITE); + + // Privilege no longer needed, restore previous state + AdjustTokenPrivileges(hProcessToken, FALSE, &prevTp, 0, NULL, NULL); + } + } + + CloseHandle(hProcessToken); + + return mem; +} + +void* aligned_ttmem_alloc(size_t allocSize, void*& mem) { + + static bool firstCall = true; + + // Try to allocate large pages + mem = aligned_ttmem_alloc_large_pages(allocSize); + + // Suppress info strings on the first call. The first call occurs before 'uci' + // is received and in that case this output confuses some GUIs. + if (!firstCall) + { + if (mem) + sync_cout << "info string Hash table allocation: Windows large pages used." << sync_endl; + else + sync_cout << "info string Hash table allocation: Windows large pages not used." << sync_endl; + } + firstCall = false; + + // Fall back to regular, page aligned, allocation if necessary + if (!mem) + mem = VirtualAlloc(NULL, allocSize, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE); + + return mem; +} + +#else + +void* aligned_ttmem_alloc(size_t allocSize, void*& mem) { + + constexpr size_t alignment = 64; // assumed cache line size + size_t size = allocSize + alignment - 1; // allocate some extra space + mem = malloc(size); + void* ret = reinterpret_cast((uintptr_t(mem) + alignment - 1) & ~uintptr_t(alignment - 1)); + return ret; +} + +#endif + + +/// aligned_ttmem_free() will free the previously allocated ttmem + +#if defined(_WIN64) + +void aligned_ttmem_free(void* mem) { + + if (mem && !VirtualFree(mem, 0, MEM_RELEASE)) + { + DWORD err = GetLastError(); + std::cerr << "Failed to free transposition table. Error code: 0x" << + std::hex << err << std::dec << std::endl; + exit(EXIT_FAILURE); + } +} + +#else + +void aligned_ttmem_free(void *mem) { + free(mem); +} + +#endif + + +namespace WinProcGroup { + +#ifndef _WIN32 + +void bindThisThread(size_t) {} + +#else + +/// best_group() retrieves logical processor information using Windows specific +/// API and returns the best group id for the thread with index idx. Original +/// code from Texel by Peter Österlund. + +int best_group(size_t idx) { + + int threads = 0; + int nodes = 0; + int cores = 0; + DWORD returnLength = 0; + DWORD byteOffset = 0; + + // Early exit if the needed API is not available at runtime + HMODULE k32 = GetModuleHandle("Kernel32.dll"); + auto fun1 = (fun1_t)(void(*)())GetProcAddress(k32, "GetLogicalProcessorInformationEx"); + if (!fun1) + return -1; + + // First call to get returnLength. We expect it to fail due to null buffer + if (fun1(RelationAll, nullptr, &returnLength)) + return -1; + + // Once we know returnLength, allocate the buffer + SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *buffer, *ptr; + ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)malloc(returnLength); + + // Second call, now we expect to succeed + if (!fun1(RelationAll, buffer, &returnLength)) + { + free(buffer); + return -1; + } + + while (byteOffset < returnLength) + { + if (ptr->Relationship == RelationNumaNode) + nodes++; + + else if (ptr->Relationship == RelationProcessorCore) + { + cores++; + threads += (ptr->Processor.Flags == LTP_PC_SMT) ? 2 : 1; + } + + assert(ptr->Size); + byteOffset += ptr->Size; + ptr = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)(((char*)ptr) + ptr->Size); + } + + free(buffer); + + std::vector groups; + + // Run as many threads as possible on the same node until core limit is + // reached, then move on filling the next node. + for (int n = 0; n < nodes; n++) + for (int i = 0; i < cores / nodes; i++) + groups.push_back(n); + + // In case a core has more than one logical processor (we assume 2) and we + // have still threads to allocate, then spread them evenly across available + // nodes. + for (int t = 0; t < threads - cores; t++) + groups.push_back(t % nodes); + + // If we still have more threads than the total number of logical processors + // then return -1 and let the OS to decide what to do. + return idx < groups.size() ? groups[idx] : -1; +} + + +/// bindThisThread() set the group affinity of the current thread + +void bindThisThread(size_t idx) { + + // Use only local variables to be thread-safe + int group = best_group(idx); + + if (group == -1) + return; + + // Early exit if the needed API are not available at runtime + HMODULE k32 = GetModuleHandle("Kernel32.dll"); + auto fun2 = (fun2_t)(void(*)())GetProcAddress(k32, "GetNumaNodeProcessorMaskEx"); + auto fun3 = (fun3_t)(void(*)())GetProcAddress(k32, "SetThreadGroupAffinity"); + + if (!fun2 || !fun3) + return; + + GROUP_AFFINITY affinity; + if (fun2(group, &affinity)) + fun3(GetCurrentThread(), &affinity, nullptr); +} + +#endif + +} // namespace WinProcGroup