StarPU Handbook
|
When GCC plug-in support is available, StarPU builds a plug-in for the GNU Compiler Collection (GCC), which defines extensions to languages of the C family (C, C++, Objective-C) that make it easier to write StarPU code. This feature is only available for GCC 4.5 and later; it is known to work with GCC 4.5, 4.6, and 4.7. You may need to install a specific -dev
package of your distro, such as gcc-4.6-plugin-dev
on Debian and derivatives. In addition, the plug-in's test suite is only run when GNU Guile (http://www.gnu.org/software/guile/) is found at configure
-time. Building the GCC plug-in can be disabled by configuring with --disable-gcc-extensions.
Those extensions include syntactic sugar for defining tasks and their implementations, invoking a task, and manipulating data buffers. Use of these extensions can be made conditional on the availability of the plug-in, leading to valid C sequential code when the plug-in is not used (Using C Extensions Conditionally).
When StarPU has been installed with its GCC plug-in, programs that use these extensions can be compiled this way:
$ gcc -c -fplugin=`pkg-config starpu-1.2 --variable=gccplugin` foo.c
When the plug-in is not available, the above pkg-config
command returns the empty string.
In addition, the -fplugin-arg-starpu-verbose
flag can be used to obtain feedback from the compiler as it analyzes the C extensions used in source files.
This section describes the C extensions implemented by StarPU's GCC plug-in. It does not require detailed knowledge of the StarPU library.
Note: this is still an area under development and subject to change.
The StarPU GCC plug-in views tasks as ``extended'' C functions:
Tasks and their implementations must be declared. These declarations are annotated with attributes (http://gcc.gnu.org/onlinedocs/gcc/Attribute-Syntax.html#Attribute-Syntax): the declaration of a task is a regular C function declaration with an additional task
attribute, and task implementations are declared with a task_implementation
attribute.
The following function attributes are provided:
task
Declare the given function as a StarPU task. Its return type must be void
. When a function declared as task
has a user-defined body, that body is interpreted as the implicit definition of the task's CPU implementation (see example below). In all cases, the actual definition of a task's body is automatically generated by the compiler.
Under the hood, declaring a task leads to the declaration of the corresponding codelet
(Codelet and Tasks). If one or more task implementations are declared in the same compilation unit, then the codelet and the function itself are also defined; they inherit the scope of the task.
Scalar arguments to the task are passed by value and copied to the target device if need be—technically, they are passed as the buffer starpu_task::cl_arg (Codelet and Tasks).
Pointer arguments are assumed to be registered data buffers—the handles argument of a task (starpu_task::handles) ; const
-qualified pointer arguments are viewed as read-only buffers (STARPU_R), and non-const
-qualified buffers are assumed to be used read-write (STARPU_RW). In addition, the output
type attribute can be as a type qualifier for output pointer or array parameters (STARPU_W).
task_implementation (target, task)
Declare the given function as an implementation of task
to run on target
. target
must be a string, currently one of "cpu"
, "opencl"
, or "cuda"
.
Here is an example:
A matmult
task is defined; it has only one implementation, matmult_cpu
, which runs on the CPU. Variables A
and B
are input buffers, whereas C
is considered an input/output buffer.
For convenience, when a function declared with the task
attribute has a user-defined body, that body is assumed to be that of the CPU implementation of a task, which we call an implicit task CPU implementation. Thus, the above snippet can be simplified like this:
Use of implicit CPU task implementations as above has the advantage that the code is valid sequential code when StarPU's GCC plug-in is not used (Using C Extensions Conditionally).
CUDA and OpenCL implementations can be declared in a similar way:
The CUDA and OpenCL implementations typically either invoke a kernel written in CUDA or OpenCL (for similar code, CUDA Kernel, and OpenCL Kernel), or call a library function that uses CUDA or OpenCL under the hood, such as CUBLAS functions:
A task can be invoked like a regular C function:
This leads to an asynchronous invocation, whereby matmult
's implementation may run in parallel with the continuation of the caller.
The next section describes how memory buffers must be handled in StarPU-GCC code. For a complete example, see the gcc-plugin/examples
directory of the source distribution, and Vector Scaling Using the C Extension.
The following pragmas allow user code to control StarPU's life time and to synchronize with tasks.
#pragma starpu initialize
Initialize StarPU. This call is compulsory and is never added implicitly. One of the reasons this has to be done explicitly is that it provides greater control to user code over its resource usage.
#pragma starpu shutdown
Shut down StarPU, giving it an opportunity to write profiling info to a file on disk, for instance (Off-line Performance Feedback).
#pragma starpu wait
Data buffers such as matrices and vectors that are to be passed to tasks must be registered. Registration allows StarPU to handle data transfers among devices—e.g., transferring an input buffer from the CPU's main memory to a task scheduled to run a GPU (StarPU Data Management Library).
The following pragmas are provided:
#pragma starpu register ptr [size]
Register ptr
as a size
-element buffer. When ptr
has an array type whose size is known, size
may be omitted. Alternatively, the registered
attribute can be used (see below.)
#pragma starpu unregister ptr
Unregister the previously-registered memory area pointed to by ptr
. As a side-effect, ptr
points to a valid copy in main memory.
#pragma starpu acquire ptr
Acquire in main memory an up-to-date copy of the previously-registered memory area pointed to by ptr
, for read-write access.
#pragma starpu release ptr
ptr
, making it available to the tasks. Additionally, the following attributes offer a simple way to allocate and register storage for arrays:
registered
This attributes applies to local variables with an array type. Its effect is to automatically register the array's storage, as per #pragma starpu register
. The array is automatically unregistered when the variable's scope is left. This attribute is typically used in conjunction with the heap_allocated
attribute, described below.
heap_allocated
The following example illustrates use of the heap_allocated
attribute:
The C extensions described in this chapter are only available when GCC and its StarPU plug-in are in use. Yet, it is possible to make use of these extensions when they are available—leading to hybrid CPU/GPU code—and discard them when they are not available—leading to valid sequential code.
To that end, the GCC plug-in defines the C preprocessor macro — STARPU_GCC_PLUGIN
— when it is being used. When defined, this macro expands to an integer denoting the version of the supported C extensions.
The code below illustrates how to define a task and its implementations in a way that allows it to be compiled without the GCC plug-in:
The above program is a valid StarPU program when StarPU's GCC plug-in is used; it is also a valid sequential program when the plug-in is not used.
Note that attributes such as task
as well as starpu
pragmas are simply ignored by GCC when the StarPU plug-in is not loaded. However, gcc -Wall
emits a warning for unknown attributes and pragmas, which can be inconvenient. In addition, other compilers may be unable to parse the attribute syntax (In practice, Clang and several proprietary compilers implement attributes.), so you may want to wrap attributes in macros like this: