A sharded hash map implementation for parallel processing written for C++20.
The central type is ShardedMap<K, V, SeqHashMapType, UpdateFn>
.
K
is the type of the keys.V
is the type of the values.SeqHashMapType
is the type of the hash map each thread will have locally. Anystd::unordered_map
compatible map should work.UpdateFn
is an update function denoting what happens, if a new value is inserted or an existing value is updated. For examples, check thesharded_map::update_functions
namespace or theexamples/
folder in this repository.
Documentation is available here.
This is a header-only library, so you can just use the sharded_map.hpp
header as is.
If you use CMake, you can for example add this repo as a git submodule to your project, and add this to your CMakeLists.txt
:
add_subdirectory(path/to/this/repo/sharded_map)
target_link_libraries(your_target PRIVATE sharded_map)
You can build and run tests using these commands:
cmake --preset=test
cmake --build --preset=test --config=Release
ctest --preset=default
#include <sharded_map/sharded_map.hpp>
#include <atomic>
#include <omp.h>
using namespace sharded_map;
int main() {
using Map = ShardedMap<size_t, size_t, std::unordered_map, update_functions::Overwrite<size_t, size_t>>;
Map map(4, 128);
// Iterate in parallel from 0 to 100'000 and insert the pair (i, 2*i) for each index
map.batch_insert(0, 100'000, [&s](size_t i) { return std::pair(i, 2*i); });
}
For a more extensive and commented example, check the examples/
folder.
You can build the examples using these commands:
cmake --preset=examples
cmake --build --preset=examples --config=Release
Then you can find the built binaries in ./build/examples/Release/
.