X-Git-Url: https://git.sesse.net/?p=stockfish;a=blobdiff_plain;f=src%2Fmisc.cpp;h=2a5bc603b8af0ebff5c4e3e3b0a2a376009f8063;hp=7075dd3cdc8498192cd8f256740fcc497cb63827;hb=384bff4264f199ded8fa28d241ce0e7dc021a97c;hpb=0d9a9f5e985c13852cf9f29767e95f295bb29575 diff --git a/src/misc.cpp b/src/misc.cpp index 7075dd3c..2a5bc603 100644 --- a/src/misc.cpp +++ b/src/misc.cpp @@ -2,7 +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) 2015-2016 Marco Costalba, Joona Kiiski, Gary Linscott, 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 @@ -21,9 +21,24 @@ #ifdef _WIN32 #if _WIN32_WINNT < 0x0601 #undef _WIN32_WINNT -#define _WIN32_WINNT 0x0601 // Force to include newest API (Win 7 or later) +#define _WIN32_WINNT 0x0601 // Force to include needed API prototypes #endif -#include // For processor groups + +#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 @@ -53,10 +68,10 @@ 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; @@ -87,6 +102,13 @@ public: if (!fname.empty() && !l.file.is_open()) { 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); } @@ -130,7 +152,7 @@ const string engine_info(bool to_uci) { /// 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); } @@ -153,7 +175,7 @@ void dbg_print() { std::ostream& operator<<(std::ostream& os, SyncCout sc) { - static Mutex m; + static std::mutex m; if (sc == IO_LOCK) m.lock(); @@ -203,11 +225,11 @@ void bindThisThread(size_t) {} #else -/// get_group() retrieves logical processor information using Windows specific +/// 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 get_group(size_t idx) { +int best_group(size_t idx) { int threads = 0; int nodes = 0; @@ -215,15 +237,14 @@ int get_group(size_t idx) { DWORD returnLength = 0; DWORD byteOffset = 0; - // Early exit if the needed API are not available at runtime + // Early exit if the needed API is not available at runtime HMODULE k32 = GetModuleHandle("Kernel32.dll"); - if ( !GetProcAddress(k32, "GetLogicalProcessorInformationEx") - || !GetProcAddress(k32, "GetNumaNodeProcessorMaskEx") - || !GetProcAddress(k32, "SetThreadGroupAffinity")) + 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 (GetLogicalProcessorInformationEx(RelationAll, nullptr, &returnLength)) + if (fun1(RelationAll, nullptr, &returnLength)) return -1; // Once we know returnLength, allocate the buffer @@ -231,13 +252,13 @@ int get_group(size_t idx) { ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)malloc(returnLength); // Second call, now we expect to succeed - if (!GetLogicalProcessorInformationEx(RelationAll, buffer, &returnLength)) + if (!fun1(RelationAll, buffer, &returnLength)) { free(buffer); return -1; } - while (ptr->Size > 0 && byteOffset + ptr->Size <= returnLength) + while (byteOffset < returnLength) { if (ptr->Relationship == RelationNumaNode) nodes++; @@ -248,6 +269,7 @@ int get_group(size_t idx) { 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); } @@ -278,15 +300,23 @@ int get_group(size_t idx) { void bindThisThread(size_t idx) { - // Use a local variable instead of a static: slower but thread-safe - int group = get_group(idx); + // Use only local variables to be thread-safe + int group = best_group(idx); if (group == -1) return; - GROUP_AFFINITY mask; - if (GetNumaNodeProcessorMaskEx(group, &mask)) - SetThreadGroupAffinity(GetCurrentThread(), &mask, nullptr); + // 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