jirayu_sri/stockfish
1
0
Fork 0
mirror of https://github.com/official-stockfish/Stockfish.git synced 2026-07-23 05:07:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

Improve performance on NUMA systems

Browse Source 
Allow for NUMA memory replication for NNUE weights.  Bind threads to ensure execution on a specific NUMA node.

This patch introduces NUMA memory replication, currently only utilized for the NNUE weights. Along with it comes all machinery required to identify NUMA nodes and bind threads to specific processors/nodes. It also comes with small changes to Thread and ThreadPool to allow easier execution of custom functions on the designated thread. Old thread binding (WinProcGroup) machinery is removed because it's incompatible with this patch. Small changes to unrelated parts of the code were made to ensure correctness, like some classes being made unmovable, raw pointers replaced with unique_ptr. etc.

Windows 7 and Windows 10 is partially supported. Windows 11 is fully supported. Linux is fully supported, with explicit exclusion of Android. No additional dependencies.

-----------------

A new UCI option `NumaPolicy` is introduced. It can take the following values:
```
system - gathers NUMA node information from the system (lscpu or windows api), for each threads binds it to a single NUMA node
none - assumes there is 1 NUMA node, never binds threads
auto - this is the default value, depends on the number of set threads and NUMA nodes, will only enable binding on multinode systems and when the number of threads reaches a threshold (dependent on node size and count)
[[custom]] -
  // ':'-separated numa nodes
  // ','-separated cpu indices
  // supports "first-last" range syntax for cpu indices,
  for example '0-15,32-47:16-31,48-63'
```

Setting `NumaPolicy` forces recreation of the threads in the ThreadPool, which in turn forces the recreation of the TT.

The threads are distributed among NUMA nodes in a round-robin fashion based on fill percentage (i.e. it will strive to fill all NUMA nodes evenly). Threads are bound to NUMA nodes, not specific processors, because that's our only requirement and the OS can schedule them better.

Special care is made that maximum memory usage on systems that do not require memory replication stays as previously, that is, unnecessary copies are avoided.

On linux the process' processor affinity is respected. This means that if you for example use taskset to restrict Stockfish to a single NUMA node then the `system` and `auto` settings will only see a single NUMA node (more precisely, the processors included in the current affinity mask) and act accordingly.

-----------------

We can't ensure that a memory allocation takes place on a given NUMA node without using libnuma on linux, or using appropriate custom allocators on windows (https://learn.microsoft.com/en-us/windows/win32/memory/allocating-memory-from-a-numa-node), so to avoid complications the current implementation relies on first-touch policy. Due to this we also rely on the memory allocator to give us a new chunk of untouched memory from the system. This appears to work reliably on linux, but results may vary.

MacOS is not supported, because AFAIK it's not affected, and implementation would be problematic anyway.

Windows is supported since Windows 7 (https://learn.microsoft.com/en-us/windows/win32/api/processtopologyapi/nf-processtopologyapi-setthreadgroupaffinity). Until Windows 11/Server 2022 NUMA nodes are split such that they cannot span processor groups. This is because before Windows 11/Server 2022 it's not possible to set thread affinity spanning processor groups. The splitting is done manually in some cases (required after Windows 10 Build 20348). Since Windows 11/Server 2022 we can set affinites spanning processor group so this splitting is not done, so the behaviour is pretty much like on linux.

Linux is supported, **without** libnuma requirement. `lscpu` is expected.

-----------------

Passed 60+1 @ 256t 16000MB hash: https://tests.stockfishchess.org/tests/view/6654e443a86388d5e27db0d8
```
LLR: 2.95 (-2.94,2.94) <0.00,10.00>
Total: 278 W: 110 L: 29 D: 139
Ptnml(0-2): 0, 1, 56, 82, 0
```

Passed SMP STC: https://tests.stockfishchess.org/tests/view/6654fc74a86388d5e27db1cd
```
LLR: 2.95 (-2.94,2.94) <-1.75,0.25>
Total: 67152 W: 17354 L: 17177 D: 32621
Ptnml(0-2): 64, 7428, 18408, 7619, 57
```

Passed STC: https://tests.stockfishchess.org/tests/view/6654fb27a86388d5e27db15c
```
LLR: 2.94 (-2.94,2.94) <-1.75,0.25>
Total: 131648 W: 34155 L: 34045 D: 63448
Ptnml(0-2): 426, 13878, 37096, 14008, 416
```

fixes #5253
closes https://github.com/official-stockfish/Stockfish/pull/5285

No functional change
This commit is contained in:
Tomasz Sobczyk
2024-05-28 18:34:15 +02:00
committed by Joost VandeVondele
parent b0287dcb1c
commit a169c78b6d
19 changed files with 1418 additions and 289 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/misc.cpp
+11 -123
View File
@@ -48,6 +48,7 @@ using fun8_t = bool (*)(HANDLE, BOOL, PTOKEN_PRIVILEGES, DWORD, PTOKEN_PRIVILEGE
#endif
#include <atomic>
#include <charconv>
#include <cmath>
#include <cstdlib>
#include <fstream>
@@ -56,6 +57,7 @@ using fun8_t = bool (*)(HANDLE, BOOL, PTOKEN_PRIVILEGES, DWORD, PTOKEN_PRIVILEGE
#include <mutex>
#include <sstream>
#include <string_view>
#include <system_error>
#include "types.h"
@@ -592,129 +594,6 @@ void aligned_large_pages_free(void* mem) { std_aligned_free(mem); }
#endif
namespace WinProcGroup {
#ifndef _WIN32
void bind_this_thread(size_t) {}
#else
namespace {
// Retrieves logical processor information using Windows-specific
// API and returns the best node id for the thread with index idx. Original
// code from Texel by Peter Österlund.
int best_node(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(TEXT("Kernel32.dll"));
auto fun1 = (fun1_t) (void (*)()) GetProcAddress(k32, "GetLogicalProcessorInformationEx");
if (!fun1)
return -1;
// First call to GetLogicalProcessorInformationEx() to get returnLength.
// We expect the call 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 to GetLogicalProcessorInformationEx(), 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<int> groups;
// Run as many threads as possible on the same node until the core limit is
// reached, then move on to 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
// still have threads to allocate, 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;
}
}
// Sets the group affinity of the current thread
void bind_this_thread(size_t idx) {
// Use only local variables to be thread-safe
int node = best_node(idx);
if (node == -1)
return;
// Early exit if the needed API are not available at runtime
HMODULE k32 = GetModuleHandle(TEXT("Kernel32.dll"));
auto fun2 = fun2_t((void (*)()) GetProcAddress(k32, "GetNumaNodeProcessorMaskEx"));
auto fun3 = fun3_t((void (*)()) GetProcAddress(k32, "SetThreadGroupAffinity"));
auto fun4 = fun4_t((void (*)()) GetProcAddress(k32, "GetNumaNodeProcessorMask2"));
auto fun5 = fun5_t((void (*)()) GetProcAddress(k32, "GetMaximumProcessorGroupCount"));
if (!fun2 || !fun3)
return;
if (!fun4 || !fun5)
{
GROUP_AFFINITY affinity;
if (fun2(node, &affinity)) // GetNumaNodeProcessorMaskEx
fun3(GetCurrentThread(), &affinity, nullptr); // SetThreadGroupAffinity
}
else
{
// If a numa node has more than one processor group, we assume they are
// sized equal and we spread threads evenly across the groups.
USHORT elements, returnedElements;
elements = fun5(); // GetMaximumProcessorGroupCount
GROUP_AFFINITY* affinity = (GROUP_AFFINITY*) malloc(elements * sizeof(GROUP_AFFINITY));
if (fun4(node, affinity, elements, &returnedElements)) // GetNumaNodeProcessorMask2
fun3(GetCurrentThread(), &affinity[idx % returnedElements],
nullptr); // SetThreadGroupAffinity
free(affinity);
}
}
#endif
} // namespace WinProcGroup
#ifdef _WIN32
#include <direct.h>
#define GETCWD _getcwd
@@ -723,6 +602,15 @@ void bind_this_thread(size_t idx) {
#define GETCWD getcwd
#endif
size_t str_to_size_t(const std::string& s) {
size_t value;
auto result = std::from_chars(s.data(), s.data() + s.size(), value);
if (result.ec != std::errc())
std::exit(EXIT_FAILURE);
return value;
}
std::string CommandLine::get_binary_directory(std::string argv0) {
std::string pathSeparator;