Devices.md

Using Device tasks

Nanos6 is designed with the capacity to be extended to handle tasks running on heterogeneous systems, using Accelerators.

For this purpose, a device clause specifying the accelerator type can be used in a task directive, to annotate code written for an Accelerator so that the runtime handles it properly.

Device tasks should be written as outlined tasks; this means that the whole task code is encapsulated in a dedicated function. Then the prototype of this function is annotated with an oss task directive.

Example:

In a header file:

#pragma oss task device(<type>) in(...) out(...) ... /* rest of clauses */
void function_foo(<params>);

In a code file:

void function_foo(<params>)
{
	/* function code */
}

This will result in every call to function_foo() in the program to implicitly create and launch a new task.

Supported Devices

Currently, the supported devices in Nanos6 are:

1. CUDA
2. OpenACC

To use device tasks in Nanos6, the user must explicitly enable the specific devices they intend to use, during configuration as described in README

For device-specific information about usage and compilation, refer to the respective guides.