2 Stockfish, a UCI chess playing engine derived from Glaurung 2.1
3 Copyright (C) 2004-2008 Tord Romstad (Glaurung author)
4 Copyright (C) 2008-2015 Marco Costalba, Joona Kiiski, Tord Romstad
5 Copyright (C) 2015-2019 Marco Costalba, Joona Kiiski, Gary Linscott, Tord Romstad
7 Stockfish is free software: you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation, either version 3 of the License, or
10 (at your option) any later version.
12 Stockfish is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
22 #if _WIN32_WINNT < 0x0601
24 #define _WIN32_WINNT 0x0601 // Force to include needed API prototypes
32 // The needed Windows API for processor groups could be missed from old Windows
33 // versions, so instead of calling them directly (forcing the linker to resolve
34 // the calls at compile time), try to load them at runtime. To do this we need
35 // first to define the corresponding function pointers.
37 typedef bool(*fun1_t)(LOGICAL_PROCESSOR_RELATIONSHIP,
38 PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, PDWORD);
39 typedef bool(*fun2_t)(USHORT, PGROUP_AFFINITY);
40 typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
57 /// Version number. If Version is left empty, then compile date in the format
58 /// DD-MM-YY and show in engine_info.
59 const string Version = "";
61 /// Our fancy logging facility. The trick here is to replace cin.rdbuf() and
62 /// cout.rdbuf() with two Tie objects that tie cin and cout to a file stream. We
63 /// can toggle the logging of std::cout and std:cin at runtime whilst preserving
64 /// usual I/O functionality, all without changing a single line of code!
65 /// Idea from http://groups.google.com/group/comp.lang.c++/msg/1d941c0f26ea0d81
67 struct Tie: public streambuf { // MSVC requires split streambuf for cin and cout
69 Tie(streambuf* b, streambuf* l) : buf(b), logBuf(l) {}
71 int sync() override { return logBuf->pubsync(), buf->pubsync(); }
72 int overflow(int c) override { return log(buf->sputc((char)c), "<< "); }
73 int underflow() override { return buf->sgetc(); }
74 int uflow() override { return log(buf->sbumpc(), ">> "); }
76 streambuf *buf, *logBuf;
78 int log(int c, const char* prefix) {
80 static int last = '\n'; // Single log file
83 logBuf->sputn(prefix, 3);
85 return last = logBuf->sputc((char)c);
91 Logger() : in(cin.rdbuf(), file.rdbuf()), out(cout.rdbuf(), file.rdbuf()) {}
92 ~Logger() { start(""); }
98 static void start(const std::string& fname) {
102 if (!fname.empty() && !l.file.is_open())
104 l.file.open(fname, ifstream::out);
108 else if (fname.empty() && l.file.is_open())
110 cout.rdbuf(l.out.buf);
119 /// engine_info() returns the full name of the current Stockfish version. This
120 /// will be either "Stockfish <Tag> DD-MM-YY" (where DD-MM-YY is the date when
121 /// the program was compiled) or "Stockfish <Version>", depending on whether
122 /// Version is empty.
124 const string engine_info(bool to_uci) {
126 const string months("Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec");
127 string month, day, year;
128 stringstream ss, date(__DATE__); // From compiler, format is "Sep 21 2008"
130 ss << "Stockfish " << Version << setfill('0');
134 date >> month >> day >> year;
135 ss << setw(2) << day << setw(2) << (1 + months.find(month) / 4) << year.substr(2);
138 ss << (Is64Bit ? " 64" : "")
139 << (HasPext ? " BMI2" : (HasPopCnt ? " POPCNT" : ""))
140 << (to_uci ? "\nid author ": " by ")
141 << "T. Romstad, M. Costalba, J. Kiiski, G. Linscott";
147 /// Debug functions used mainly to collect run-time statistics
148 static int64_t hits[2], means[2];
150 void dbg_hit_on(bool b) { ++hits[0]; if (b) ++hits[1]; }
151 void dbg_hit_on(bool c, bool b) { if (c) dbg_hit_on(b); }
152 void dbg_mean_of(int v) { ++means[0]; means[1] += v; }
157 cerr << "Total " << hits[0] << " Hits " << hits[1]
158 << " hit rate (%) " << 100 * hits[1] / hits[0] << endl;
161 cerr << "Total " << means[0] << " Mean "
162 << (double)means[1] / means[0] << endl;
166 /// Used to serialize access to std::cout to avoid multiple threads writing at
169 std::ostream& operator<<(std::ostream& os, SyncCout sc) {
183 /// Trampoline helper to avoid moving Logger to misc.h
184 void start_logger(const std::string& fname) { Logger::start(fname); }
187 /// prefetch() preloads the given address in L1/L2 cache. This is a non-blocking
188 /// function that doesn't stall the CPU waiting for data to be loaded from memory,
189 /// which can be quite slow.
192 void prefetch(void*) {}
196 void prefetch(void* addr) {
198 # if defined(__INTEL_COMPILER)
199 // This hack prevents prefetches from being optimized away by
200 // Intel compiler. Both MSVC and gcc seem not be affected by this.
204 # if defined(__INTEL_COMPILER) || defined(_MSC_VER)
205 _mm_prefetch((char*)addr, _MM_HINT_T0);
207 __builtin_prefetch(addr);
213 void prefetch2(void* addr) {
216 prefetch((uint8_t*)addr + 64);
219 namespace WinProcGroup {
223 void bindThisThread(size_t) {}
227 /// best_group() retrieves logical processor information using Windows specific
228 /// API and returns the best group id for the thread with index idx. Original
229 /// code from Texel by Peter Ă–sterlund.
231 int best_group(size_t idx) {
236 DWORD returnLength = 0;
237 DWORD byteOffset = 0;
239 // Early exit if the needed API is not available at runtime
240 HMODULE k32 = GetModuleHandle("Kernel32.dll");
241 auto fun1 = (fun1_t)(void(*)())GetProcAddress(k32, "GetLogicalProcessorInformationEx");
245 // First call to get returnLength. We expect it to fail due to null buffer
246 if (fun1(RelationAll, nullptr, &returnLength))
249 // Once we know returnLength, allocate the buffer
250 SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX *buffer, *ptr;
251 ptr = buffer = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)malloc(returnLength);
253 // Second call, now we expect to succeed
254 if (!fun1(RelationAll, buffer, &returnLength))
260 while (byteOffset < returnLength)
262 if (ptr->Relationship == RelationNumaNode)
265 else if (ptr->Relationship == RelationProcessorCore)
268 threads += (ptr->Processor.Flags == LTP_PC_SMT) ? 2 : 1;
272 byteOffset += ptr->Size;
273 ptr = (SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX*)(((char*)ptr) + ptr->Size);
278 std::vector<int> groups;
280 // Run as many threads as possible on the same node until core limit is
281 // reached, then move on filling the next node.
282 for (int n = 0; n < nodes; n++)
283 for (int i = 0; i < cores / nodes; i++)
286 // In case a core has more than one logical processor (we assume 2) and we
287 // have still threads to allocate, then spread them evenly across available
289 for (int t = 0; t < threads - cores; t++)
290 groups.push_back(t % nodes);
292 // If we still have more threads than the total number of logical processors
293 // then return -1 and let the OS to decide what to do.
294 return idx < groups.size() ? groups[idx] : -1;
298 /// bindThisThread() set the group affinity of the current thread
300 void bindThisThread(size_t idx) {
302 // Use only local variables to be thread-safe
303 int group = best_group(idx);
308 // Early exit if the needed API are not available at runtime
309 HMODULE k32 = GetModuleHandle("Kernel32.dll");
310 auto fun2 = (fun2_t)(void(*)())GetProcAddress(k32, "GetNumaNodeProcessorMaskEx");
311 auto fun3 = (fun3_t)(void(*)())GetProcAddress(k32, "SetThreadGroupAffinity");
316 GROUP_AFFINITY affinity;
317 if (fun2(group, &affinity))
318 fun3(GetCurrentThread(), &affinity, nullptr);
323 } // namespace WinProcGroup