Skip to content

Environment and world

In the upcoming sections we will get you started with programming in Bulk. Two important concepts are that of an environment and a world.

In these code examples, we will often use the short-hand s for the local processor id which is referred to as its rank, and p for the active number of processors, and we will not always define these variables explicitly.

Parallel environments

A program runs in some parallel environment. For example, this environment could be an MPI cluster, a many-core co-processor, or simply threads on a multi-core computer. This environment is accessed within the program through a bulk::environment object. This object is specialized for each backend, which is an implementation of the lower-level communication that reflects the actual environment. For example, to setup an environment on an MPI cluster, we would write:

#include <bulk/bulk.hpp>
#include <bulk/backends/mpi/mpi.hpp>

int main() {
    bulk::mpi::environment env;
}

For a list of default providers, consult the backends section of this documentation.

This environment object contains information on the parallel system, for example we can request the number of processors that are available.

We note that throughout this documentation (and in the library), processor is a general term for the entity that executes the SPMD section (more on this later) and communicates with other processors – this can be an MPI node, a core, or a thread, depending on the backend that is used, but they are all treated in the same manner.

auto processor_count = env.available_processors();

This information can be used to spawn the program on the right amount of processors. Programs written in Bulk follow the SPMD (Single Program Multiple Data) paradigm. This means that each processor executes the same code, but has its own (local) data that it manipulates. In Bulk, the SPMD section is a function object. This can be a C++ lambda, a std::function, or a C function pointer. In this documentation we will use lambda functions for our examples.

This function will run on each processor, and should take three arguments. The first, contains the world object, which we will describe in detail in the next section. The SPMD section is executed in the following way:

env.spawn(env.available_processors(), [](auto& world) {
    auto s = world.rank();
    auto p = world.available_processors();
    world.log("Hello world from processor %d / %d!", s, p);
}

The spawn function takes two arguments. The first is the total number of processors to run the SPMD section on, here we simply use all the processors that are available. The second is the SPMD function itself, that is run on the given number of processors.

The world of a processor

image

Each processor can communicate to other processors using the world object of type bulk::world. The world object contains some information on the specifics of the SPMD section, such as the number of processors executing the section, and its identifier (as we have seen, these are also provided as arguments for programmer convenience). We can also obtain indices of the neighbouring processors:

auto next = world.next_rank();
auto previous = world.prev_rank();

The next and previous processor can also be computed manually using:

next = (s + 1) % p;
previous = (s + p - 1) % p;

However, we would suggest using the appropriate methods of world to increase readability. Another important mechanism exposed through the world object is the ability to perform a bulk synchronization, which is the cornerstone of programs written in BSP style:

world.sync();

We will see the specific uses of bulk synchronization in the upcoming sections.