Files
stockfish/src/memory.cpp
T
69a01b88f3 Use shared memory for network weights
This enables different Stockfish processes that use the same weights to use the
same memory. The approach establishes equivalence by memory content, and is
compatible with NUMA replication.  The benefit of sharing is reduced memory usage
and a speedup thanks to improved (inter-process) caching of the network in the
CPUs cache, and thus reduced bandwidth usage to main memory. Even though this
change doesn't benefit a user running a single process, this helps on fishtest
or e.g.  for Lichess, when multiple games run concurrently, or multiple
positions are analyzed in parallel.

This concept was probably first introduced in the Monty engine
(https://github.com/official-monty/Monty/pull/62), after a discussion in
https://github.com/official-stockfish/fishtest/issues/2077 on the issue of
memory pressure. Measurements based on Torch
(https://github.com/user-attachments/files/21386224/verbatim.pdf) further
suggested that large gains were possible. Multiple other engines have
adopted this 'verbatim' format as well.

The implementation here adds the flexibility needed for SF, for example, retains
the ability to bundle compressed networks with the binary, to load nets by uci
option, and to distribute the shared nets to the proper NUMA region. This
flexibility comes with a fair amount of complexity in the implementation, such
as OS specific code, and fallback code.

For most users this should be transparent. However, for example, those running
docker containers should ensure the `--ipc` flag is set correctly, and
`--shm-size` is sufficiently large.

The benefits of this patch significantly depend on hardware, with systems with
many cores and a large (O(150MB), the net size) L3 cache benefitting typically
most.  On such systems SF speedups (as measured via nps playing games with
large concurrency but just 1 thread) can be 38%, which results in master vs.
patch Elo which gains about 25 Elo.

```
   # PLAYER             :  RATING  ERROR   POINTS  PLAYED   (%)
   1 shared_memoryPR    :    24.8    1.9  39432.0   73728    53
   2 master             :     0.0   ----  34296.0   73728    47
```

In a multithreaded setup, where weights are already shared, that benefit is smaller,
for example on the same HW as above, but with 8t for each side.
```
   # PLAYER             :  RATING  ERROR  POINTS  PLAYED   (%)
   1 shared_memoryPR    :     5.2    3.5  9351.0   18432    51
   2 master             :     0.0   ----  9081.0   18432    49
```

On fishtest with a typical hardware mix of our contributors, the following was measured:

STC, 60k games
https://tests.stockfishchess.org/tests/view/69074a49ea4b268f1fac236c
Elo: 4.69 ± 1.4 (95%) LOS: 100.0%
Total: 60000 W: 16085 L: 15275 D: 28640
Ptnml(0-2): 154, 6440, 16053, 7148, 205
nElo: 9.38 ± 2.8 (95%) PairsRatio: 1.12

To verify correctness with a single process on a NUMA architecture,
speedtest was used, confirming near equivalence:
```
master:        Average (over 10):  296236186
shared_memory: Average (over 10):  295769332
```
Currently, using large pages for the shared network weights is not always possible,
which can lead to a small slowdown (1-2%), in case a single process is run.

closes https://github.com/official-stockfish/Stockfish/pull/6173

No functional change

Co-authored-by: disservin <disservin.social@gmail.com>
Co-authored-by: Joost VandeVondele <Joost.VandeVondele@gmail.com>
2025-11-02 16:04:09 +01:00

200 lines
5.3 KiB
C++

/*
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
Copyright (C) 2004-2025 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/>.
*/
#include "memory.h"
#include <cstdlib>
#if __has_include("features.h")
#include <features.h>
#endif
#if defined(__linux__) && !defined(__ANDROID__)
#include <sys/mman.h>
#endif
#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) \
|| (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32)) \
|| defined(__e2k__)
#define POSIXALIGNEDALLOC
#include <stdlib.h>
#endif
#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 <ios> // std::hex, std::dec
#include <iostream> // std::cerr
#include <ostream> // std::endl
#include <windows.h>
// 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.
#endif
namespace Stockfish {
// Wrappers for systems where the c++17 implementation does not guarantee the
// availability of aligned_alloc(). Memory allocated with std_aligned_alloc()
// must be freed with std_aligned_free().
void* std_aligned_alloc(size_t alignment, size_t size) {
#if defined(_ISOC11_SOURCE)
return aligned_alloc(alignment, size);
#elif defined(POSIXALIGNEDALLOC)
void* mem = nullptr;
posix_memalign(&mem, alignment, size);
return mem;
#elif defined(_WIN32) && !defined(_M_ARM) && !defined(_M_ARM64)
return _mm_malloc(size, alignment);
#elif defined(_WIN32)
return _aligned_malloc(size, alignment);
#else
return std::aligned_alloc(alignment, size);
#endif
}
void std_aligned_free(void* ptr) {
#if defined(POSIXALIGNEDALLOC)
free(ptr);
#elif defined(_WIN32) && !defined(_M_ARM) && !defined(_M_ARM64)
_mm_free(ptr);
#elif defined(_WIN32)
_aligned_free(ptr);
#else
free(ptr);
#endif
}
// aligned_large_pages_alloc() will return suitably aligned memory,
// if possible using large pages.
#if defined(_WIN32)
static void* aligned_large_pages_alloc_windows([[maybe_unused]] size_t allocSize) {
return windows_try_with_large_page_priviliges(
[&](size_t largePageSize) {
// Round up size to full pages and allocate
allocSize = (allocSize + largePageSize - 1) & ~size_t(largePageSize - 1);
return VirtualAlloc(nullptr, allocSize, MEM_RESERVE | MEM_COMMIT | MEM_LARGE_PAGES,
PAGE_READWRITE);
},
[]() { return (void*) nullptr; });
}
void* aligned_large_pages_alloc(size_t allocSize) {
// Try to allocate large pages
void* mem = aligned_large_pages_alloc_windows(allocSize);
// Fall back to regular, page-aligned, allocation if necessary
if (!mem)
mem = VirtualAlloc(nullptr, allocSize, MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE);
return mem;
}
#else
void* aligned_large_pages_alloc(size_t allocSize) {
#if defined(__linux__)
constexpr size_t alignment = 2 * 1024 * 1024; // 2MB page size assumed
#else
constexpr size_t alignment = 4096; // small page size assumed
#endif
// Round up to multiples of alignment
size_t size = ((allocSize + alignment - 1) / alignment) * alignment;
void* mem = std_aligned_alloc(alignment, size);
#if defined(MADV_HUGEPAGE)
madvise(mem, size, MADV_HUGEPAGE);
#endif
return mem;
}
#endif
bool has_large_pages() {
#if defined(_WIN32)
constexpr size_t page_size = 2 * 1024 * 1024; // 2MB page size assumed
void* mem = aligned_large_pages_alloc_windows(page_size);
if (mem == nullptr)
{
return false;
}
else
{
aligned_large_pages_free(mem);
return true;
}
#elif defined(__linux__)
#if defined(MADV_HUGEPAGE)
return true;
#else
return false;
#endif
#else
return false;
#endif
}
// aligned_large_pages_free() will free the previously memory allocated
// by aligned_large_pages_alloc(). The effect is a nop if mem == nullptr.
#if defined(_WIN32)
void aligned_large_pages_free(void* mem) {
if (mem && !VirtualFree(mem, 0, MEM_RELEASE))
{
DWORD err = GetLastError();
std::cerr << "Failed to free large page memory. Error code: 0x" << std::hex << err
<< std::dec << std::endl;
exit(EXIT_FAILURE);
}
}
#else
void aligned_large_pages_free(void* mem) { std_aligned_free(mem); }
#endif
} // namespace Stockfish