Skip to content

<batteries/cpu\_align.hpp>: Prevent false sharing by isolating a type in its own cache line(s).

<batteries/cpu_align.hpp>: Prevent false sharing by isolating a type in its own cache line(s).🔗

File Reference

Features🔗

Details🔗

Sometimes when objects modified by different threads are packed too closely in memory, performance can degrade significantly due to a "ping-pong" effect where each CPU/core that modifies an object in the same cache line (the minimum-sized block of memory fetched from main memory into cache) must wait to read changes made to the other objects in that cache line. This problem can be solved by padding the memory into which the objects are placed to ensure each resides within its own cache line. batt::CpuCacheLineIsolated<T> makes this very simple and easy to do.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
// More compact, but slower due to false sharing between CPUs:
//
std::array<int, 100> slow_ints;

// Each int is padded and aligned so it gets its own cache line:
//
std::array<batt::CpuCacheLineIsolated<int>, 100> fast_ints;

// The underlying `T` (int in this case) is accessed via `*` and `->`
// operators, like a pointer or `std::optional`.
//
int& i_ref = *fast_int[0];