HCha/Stockfish
1
0
Fork 0
mirror of https://github.com/HChaZZY/Stockfish.git synced 2025-12-20 17:16:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Advise the kernel to use huge pages (Linux)

Browse Source 
Align the TT allocation by 2M to make it huge page friendly and advise the
kernel to use huge pages.

Benchmarks on my i7-8700K (6C/12T) box: (3 runs per bench per config)

                    vanilla (nps)               hugepages (nps)              avg
==================================================================================
bench             | 3012490  3024364  3036331   3071052  3067544  3071052    +1.5%
bench 16 12 20    | 19237932 19050166 19085315  19266346 19207025 19548758   +1.1%
bench 16384 12 20 | 18182313 18371581 18336838  19381275 19738012 19620225   +7.0%

On my box, huge pages have a significant perf impact when using a big
hash size. They also speed up TT initialization big time:

                                  vanilla (s)  huge pages (s)  speed-up
=======================================================================
time stockfish bench 16384 1 1  | 5.37         1.48            3.6x

In practice, huge pages with auto-defrag may always be enabled in the
system, in which case this patch has no effect. This
depends on the values in /sys/kernel/mm/transparent_hugepage/enabled
and /sys/kernel/mm/transparent_hugepage/defrag.

closes https://github.com/official-stockfish/Stockfish/pull/2463

No functional change
This commit is contained in:
Sami Kiminki
2019-12-21 21:41:42 +02:00
committed by Joost VandeVondele
parent 6d0eabd5fe
commit 39437f4e55
4 changed files with 45 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
src/tt.h
View File

@@ -57,24 +57,22 @@ private:
};
/// A TranspositionTable consists of a power of 2 number of clusters and each
/// cluster consists of ClusterSize number of TTEntry. Each non-empty entry
/// contains information of exactly one position. The size of a cluster should
/// divide the size of a cache line size, to ensure that clusters never cross
/// cache lines. This ensures best cache performance, as the cacheline is
/// prefetched, as soon as possible.
/// A TranspositionTable is an array of Cluster, of size clusterCount. Each
/// cluster consists of ClusterSize number of TTEntry. Each non-empty TTEntry
/// contains information on exactly one position. The size of a Cluster should
/// divide the size of a cache line for best performance,
/// as the cacheline is prefetched when possible.
class TranspositionTable {
static constexpr int CacheLineSize = 64;
static constexpr int ClusterSize = 3;
struct Cluster {
TTEntry entry[ClusterSize];
char padding[2]; // Align to a divisor of the cache line size
char padding[2]; // Pad to 32 bytes
};
static_assert(CacheLineSize % sizeof(Cluster) == 0, "Cluster size incorrect");
static_assert(sizeof(Cluster) == 32, "Unexpected Cluster size");
public:
~TranspositionTable() { free(mem); }