StarPU Handbook
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
Basic Examples

Hello World Using The C Extension

This section shows how to implement a simple program that submits a task to StarPU using the StarPU C extension (C Extensions). The complete example, and additional examples, is available in the directory gcc-plugin/examples of the StarPU distribution. A similar example showing how to directly use the StarPU's API is shown in Hello World Using StarPU's API.

GCC from version 4.5 permit to use the StarPU GCC plug-in (C Extensions). This makes writing a task both simpler and less error-prone. In a nutshell, all it takes is to declare a task, declare and define its implementations (for CPU, OpenCL, and/or CUDA), and invoke the task like a regular C function. The example below defines my_task which has a single implementation for CPU:

#include <stdio.h>
/* Task declaration. */
static void my_task (int x) __attribute__ ((task));
/* Definition of the CPU implementation of `my_task'. */
static void my_task (int x)
{
printf ("Hello, world! With x = %d\n", x);
}
int main ()
{
/* Initialize StarPU. */
#pragma starpu initialize
/* Do an asynchronous call to `my_task'. */
my_task (42);
/* Wait for the call to complete. */
#pragma starpu wait
/* Terminate. */
#pragma starpu shutdown
return 0;
}

The code can then be compiled and linked with GCC and the flag -fplugin:

$ gcc `pkg-config starpu-1.2 --cflags` hello-starpu.c \
    -fplugin=`pkg-config starpu-1.2 --variable=gccplugin` \
    `pkg-config starpu-1.2 --libs`

The code can also be compiled without the StarPU C extension and will behave as a normal sequential code.

$ gcc hello-starpu.c
hello-starpu.c:33:1: warning: ‘task’ attribute directive ignored [-Wattributes]
$ ./a.out
Hello, world! With x = 42

As can be seen above, the C extensions allows programmers to use StarPU tasks by essentially annotating ``regular'' C code.

Hello World Using StarPU's API

This section shows how to achieve the same result as in the previous section using StarPU's standard C API.

Required Headers

The header starpu.h should be included in any code using StarPU.

#include <starpu.h>

Defining A Codelet

A codelet is a structure that represents a computational kernel. Such a codelet may contain an implementation of the same kernel on different architectures (e.g. CUDA, x86, ...). For compatibility, make sure that the whole structure is properly initialized to zero, either by using the function starpu_codelet_init(), or by letting the compiler implicitly do it as examplified below.

The field starpu_codelet::nbuffers specifies the number of data buffers that are manipulated by the codelet: here the codelet does not access or modify any data that is controlled by our data management library.

We create a codelet which may only be executed on CPUs. When a CPU core will execute a codelet, it will call the function cpu_func, which must have the following prototype:

void (*cpu_func)(void *buffers[], void *cl_arg);

In this example, we can ignore the first argument of this function which gives a description of the input and output buffers (e.g. the size and the location of the matrices) since there is none. We also ignore the second argument which is a pointer to optional arguments for the codelet.

void cpu_func(void *buffers[], void *cl_arg)
{
printf("Hello world\n");
}
struct starpu_codelet cl =
{
.nbuffers = 0
};

Submitting A Task

Before submitting any tasks to StarPU, starpu_init() must be called. The NULL argument specifies that we use the default configuration. Tasks can then be submitted until the termination of StarPU – done by a call to starpu_shutdown().

In the example below, a task structure is allocated by a call to starpu_task_create(). This function allocates and fills the task structure with its default settings, it does not submit the task to StarPU.

The field starpu_task::cl is a pointer to the codelet which the task will execute: in other words, the codelet structure describes which computational kernel should be offloaded on the different architectures, and the task structure is a wrapper containing a codelet and the piece of data on which the codelet should operate.

If the field starpu_task::synchronous is non-zero, task submission will be synchronous: the function starpu_task_submit() will not return until the task has been executed. Note that the function starpu_shutdown() does not guarantee that asynchronous tasks have been executed before it returns, starpu_task_wait_for_all() can be used to that effect, or data can be unregistered (starpu_data_unregister()), which will implicitly wait for all the tasks scheduled to work on it, unless explicitly disabled thanks to starpu_data_set_default_sequential_consistency_flag() or starpu_data_set_sequential_consistency_flag().

int main(int argc, char **argv)
{
/* initialize StarPU */
starpu_init(NULL);
struct starpu_task *task = starpu_task_create();
task->cl = &cl; /* Pointer to the codelet defined above */
/* starpu_task_submit will be a blocking call. If unset,
starpu_task_wait() needs to be called after submitting the task. */
task->synchronous = 1;
/* submit the task to StarPU */
/* terminate StarPU */
return 0;
}

Execution Of Hello World

$ make hello_world
cc $(pkg-config --cflags starpu-1.2) hello_world.c -o hello_world $(pkg-config --libs starpu-1.2)
$ ./hello_world
Hello world

Passing Arguments To The Codelet

The optional field starpu_task::cl_arg field is a pointer to a buffer (of size starpu_task::cl_arg_size) with some parameters for the kernel described by the codelet. For instance, if a codelet implements a computational kernel that multiplies its input vector by a constant, the constant could be specified by the means of this buffer, instead of registering it as a StarPU data. It must however be noted that StarPU avoids making copy whenever possible and rather passes the pointer as such, so the buffer which is pointed at must be kept allocated until the task terminates, and if several tasks are submitted with various parameters, each of them must be given a pointer to their own buffer.

struct params
{
int i;
float f;
};
void cpu_func(void *buffers[], void *cl_arg)
{
struct params *params = cl_arg;
printf("Hello world (params = {%i, %f} )\n", params->i, params->f);
}

As said before, the field starpu_codelet::nbuffers specifies the number of data buffers that are manipulated by the codelet. It does not count the argument — the parameter cl_arg of the function cpu_func — since it is not managed by our data management library, but just contains trivial parameters.

Be aware that this may be a pointer to a copy of the actual buffer, and not the pointer given by the programmer: if the codelet modifies this buffer, there is no guarantee that the initial buffer will be modified as well: this for instance implies that the buffer cannot be used as a synchronization medium. If synchronization is needed, data has to be registered to StarPU, see Vector Scaling Using StarPU's API.

int main(int argc, char **argv)
{
/* initialize StarPU */
starpu_init(NULL);
struct starpu_task *task = starpu_task_create();
task->cl = &cl; /* Pointer to the codelet defined above */
struct params params = { 1, 2.0f };
task->cl_arg = &params;
task->cl_arg_size = sizeof(params);
/* starpu_task_submit will be a blocking call */
task->synchronous = 1;
/* submit the task to StarPU */
/* terminate StarPU */
return 0;
}
$ make hello_world
cc $(pkg-config --cflags starpu-1.2) hello_world.c -o hello_world $(pkg-config --libs starpu-1.2)
$ ./hello_world
Hello world (params = {1, 2.000000} )

Defining A Callback

Once a task has been executed, an optional callback function starpu_task::callback_func is called when defined. While the computational kernel could be offloaded on various architectures, the callback function is always executed on a CPU. The pointer starpu_task::callback_arg is passed as an argument of the callback function. The prototype of a callback function must be:

void (*callback_function)(void *);
void callback_func(void *callback_arg)
{
printf("Callback function (arg %x)\n", callback_arg);
}
int main(int argc, char **argv)
{
/* initialize StarPU */
starpu_init(NULL);
struct starpu_task *task = starpu_task_create();
task->cl = &cl; /* Pointer to the codelet defined above */
task->callback_arg = 0x42;
/* starpu_task_submit will be a blocking call */
task->synchronous = 1;
/* submit the task to StarPU */
/* terminate StarPU */
return 0;
}
$ make hello_world
cc $(pkg-config --cflags starpu-1.2) hello_world.c -o hello_world $(pkg-config --libs starpu-1.2) 
$ ./hello_world
Hello world
Callback function (arg 42)

Where To Execute A Codelet

struct starpu_codelet cl =
{
.cpu_funcs = { cpu_func },
.cpu_funcs_name = { "cpu_func" },
.nbuffers = 0
};

We create a codelet which may only be executed on the CPUs. The optional field starpu_codelet::where is a bitmask that defines where the codelet may be executed. Here, the value STARPU_CPU means that only CPUs can execute this codelet. When the optional field starpu_codelet::where is unset, its value is automatically set based on the availability of the different fields XXX_funcs.

TODO: explain starpu_codelet::cpu_funcs_name

Vector Scaling Using the C Extension

The previous example has shown how to submit tasks. In this section, we show how StarPU tasks can manipulate data.

We will first show how to use the C language extensions provided by the GCC plug-in (C Extensions). The complete example, and additional examples, is available in the directory gcc-plugin/examples of the StarPU distribution. These extensions map directly to StarPU's main concepts: tasks, task implementations for CPU, OpenCL, or CUDA, and registered data buffers. The standard C version that uses StarPU's standard C programming interface is given in Vector Scaling Using StarPU's API.

First of all, the vector-scaling task and its simple CPU implementation has to be defined:

/* Declare the `vector_scal' task. */
static void vector_scal (unsigned size, float vector[size],
float factor)
__attribute__ ((task));
/* Define the standard CPU implementation. */
static void
vector_scal (unsigned size, float vector[size], float factor)
{
unsigned i;
for (i = 0; i < size; i++)
vector[i] *= factor;
}

Next, the body of the program, which uses the task defined above, can be implemented:

int main (void)
{
#pragma starpu initialize
#define NX 0x100000
#define FACTOR 3.14
{
float vector[NX]
__attribute__ ((heap_allocated, registered));
size_t i;
for (i = 0; i < NX; i++)
vector[i] = (float) i;
vector_scal (NX, vector, FACTOR);
#pragma starpu wait
} /* VECTOR is automatically freed here. */
#pragma starpu shutdown
return valid ? EXIT_SUCCESS : EXIT_FAILURE;
}

The function main above does several things:

  • It initializes StarPU.
  • It allocates vector in the heap; it will automatically be freed when its scope is left. Alternatively, good old malloc and free could have been used, but they are more error-prone and require more typing.
  • It registers the memory pointed to by vector. Eventually, when OpenCL or CUDA task implementations are added, this will allow StarPU to transfer that memory region between GPUs and the main memory. Removing this pragma is an error.
  • It invokes the task vector_scal. The invocation looks the same as a standard C function call. However, it is an asynchronous invocation, meaning that the actual call is performed in parallel with the caller's continuation.
  • It waits for the termination of the asynchronous call vector_scal.
  • Finally, StarPU is shut down.

The program can be compiled and linked with GCC and the flag -fplugin:

$ gcc `pkg-config starpu-1.2 --cflags` vector_scal.c 
    -fplugin=`pkg-config starpu-1.2 --variable=gccplugin` 
    `pkg-config starpu-1.2 --libs`

And voilà!

Adding an OpenCL Task Implementation

Now, this is all fine and great, but you certainly want to take advantage of these newfangled GPUs that your lab just bought, don't you?

So, let's add an OpenCL implementation of the task vector_scal. We assume that the OpenCL kernel is available in a file, vector_scal_opencl_kernel.cl, not shown here. The OpenCL task implementation is similar to that used with the standard C API (Definition of the OpenCL Kernel). It is declared and defined in our C file like this:

/* The OpenCL programs, loaded from 'main' (see below). */
static struct starpu_opencl_program cl_programs;
static void vector_scal_opencl (unsigned size, float vector[size],
float factor)
__attribute__ ((task_implementation ("opencl", vector_scal)));
static void
vector_scal_opencl (unsigned size, float vector[size], float factor)
{
int id, devid, err;
cl_kernel kernel;
cl_command_queue queue;
cl_event event;
/* VECTOR is GPU memory pointer, not a main memory pointer. */
cl_mem val = (cl_mem) vector;
/* Prepare to invoke the kernel. In the future, this will be largely automated. */
err = starpu_opencl_load_kernel (&kernel, &queue, &cl_programs,
"vector_mult_opencl", devid);
if (err != CL_SUCCESS)
err = clSetKernelArg (kernel, 0, sizeof (size), &size);
err |= clSetKernelArg (kernel, 1, sizeof (val), &val);
err |= clSetKernelArg (kernel, 2, sizeof (factor), &factor);
if (err)
size_t global = 1, local = 1;
err = clEnqueueNDRangeKernel (queue, kernel, 1, NULL, &global,
&local, 0, NULL, &event);
if (err != CL_SUCCESS)
clFinish (queue);
clReleaseEvent (event);
/* Done with KERNEL. */
}

The OpenCL kernel itself must be loaded from main, sometime after the pragma initialize:

starpu_opencl_load_opencl_from_file ("vector_scal_opencl_kernel.cl",
&cl_programs, "");

And that's it. The task vector_scal now has an additional implementation, for OpenCL, which StarPU's scheduler may choose to use at run-time. Unfortunately, the vector_scal_opencl above still has to go through the common OpenCL boilerplate; in the future, additional extensions will automate most of it.

Adding a CUDA Task Implementation

Adding a CUDA implementation of the task is very similar, except that the implementation itself is typically written in CUDA, and compiled with nvcc. Thus, the C file only needs to contain an external declaration for the task implementation:

extern void vector_scal_cuda (unsigned size, float vector[size],
float factor)
__attribute__ ((task_implementation ("cuda", vector_scal)));

The actual implementation of the CUDA task goes into a separate compilation unit, in a .cu file. It is very close to the implementation when using StarPU's standard C API (Definition of the CUDA Kernel).

/* CUDA implementation of the `vector_scal' task, to be compiled with `nvcc'. */
#include <starpu.h>
#include <stdlib.h>
static __global__ void
vector_mult_cuda (unsigned n, float *val, float factor)
{
unsigned i = blockIdx.x * blockDim.x + threadIdx.x;
if (i < n)
val[i] *= factor;
}
/* Definition of the task implementation declared in the C file. */
extern "C" void
vector_scal_cuda (size_t size, float vector[], float factor)
{
unsigned threads_per_block = 64;
unsigned nblocks = (size + threads_per_block - 1) / threads_per_block;
vector_mult_cuda <<< nblocks, threads_per_block, 0,
starpu_cuda_get_local_stream () >>> (size, vector, factor);
cudaStreamSynchronize (starpu_cuda_get_local_stream ());
}

The complete source code, in the directory gcc-plugin/examples/vector_scal of the StarPU distribution, also shows how an SSE-specialized CPU task implementation can be added.

For more details on the C extensions provided by StarPU's GCC plug-in, see C Extensions.

Vector Scaling Using StarPU's API

This section shows how to achieve the same result as explained in the previous section using StarPU's standard C API.

The full source code for this example is given in Full source code for the ’Scaling a Vector’ example.

Source Code of Vector Scaling

Programmers can describe the data layout of their application so that StarPU is responsible for enforcing data coherency and availability across the machine. Instead of handling complex (and non-portable) mechanisms to perform data movements, programmers only declare which piece of data is accessed and/or modified by a task, and StarPU makes sure that when a computational kernel starts somewhere (e.g. on a GPU), its data are available locally.

Before submitting those tasks, the programmer first needs to declare the different pieces of data to StarPU using the functions starpu_*_data_register. To ease the development of applications for StarPU, it is possible to describe multiple types of data layout. A type of data layout is called an interface. There are different predefined interfaces available in StarPU: here we will consider the vector interface.

The following lines show how to declare an array of NX elements of type float using the vector interface:

float vector[NX];
starpu_data_handle_t vector_handle;
starpu_vector_data_register(&vector_handle, STARPU_MAIN_RAM, (uintptr_t)vector, NX,
sizeof(vector[0]));

The first argument, called the data handle, is an opaque pointer which designates the array within StarPU. This is also the structure which is used to describe which data is used by a task. The second argument is the node number where the data originally resides. Here it is STARPU_MAIN_RAM since the array vector is in the main memory. Then comes the pointer vector where the data can be found in main memory, the number of elements in the vector and the size of each element. The following shows how to construct a StarPU task that will manipulate the vector and a constant factor.

float factor = 3.14;
task->cl = &cl; /* Pointer to the codelet defined below */
task->handles[0] = vector_handle; /* First parameter of the codelet */
task->cl_arg = &factor;
task->cl_arg_size = sizeof(factor);
task->synchronous = 1;

Since the factor is a mere constant float value parameter, it does not need a preliminary registration, and can just be passed through the pointer starpu_task::cl_arg like in the previous example. The vector parameter is described by its handle. starpu_task::handles should be set with the handles of the data, the access modes for the data are defined in the field starpu_codelet::modes (STARPU_R for read-only, STARPU_W for write-only and STARPU_RW for read and write access).

The definition of the codelet can be written as follows:

void scal_cpu_func(void *buffers[], void *cl_arg)
{
unsigned i;
float *factor = cl_arg;
/* length of the vector */
unsigned n = STARPU_VECTOR_GET_NX(buffers[0]);
/* CPU copy of the vector pointer */
float *val = (float *)STARPU_VECTOR_GET_PTR(buffers[0]);
for (i = 0; i < n; i++)
val[i] *= *factor;
}
struct starpu_codelet cl =
{
.cpu_funcs = { scal_cpu_func },
.cpu_funcs_name = { "scal_cpu_func" },
.nbuffers = 1,
.modes = { STARPU_RW }
};

The first argument is an array that gives a description of all the buffers passed in the array starpu_task::handles. The size of this array is given by the field starpu_codelet::nbuffers. For the sake of genericity, this array contains pointers to the different interfaces describing each buffer. In the case of the vector interface, the location of the vector (resp. its length) is accessible in the starpu_vector_interface::ptr (resp. starpu_vector_interface::nx) of this interface. Since the vector is accessed in a read-write fashion, any modification will automatically affect future accesses to this vector made by other tasks.

The second argument of the function scal_cpu_func contains a pointer to the parameters of the codelet (given in starpu_task::cl_arg), so that we read the constant factor from this pointer.

Execution of Vector Scaling

$ make vector_scal
cc $(pkg-config --cflags starpu-1.2) vector_scal.c -o vector_scal $(pkg-config --libs starpu-1.2)
$ ./vector_scal
0.000000 3.000000 6.000000 9.000000 12.000000

Vector Scaling on an Hybrid CPU/GPU Machine

Contrary to the previous examples, the task submitted in this example may not only be executed by the CPUs, but also by a CUDA device.

Definition of the CUDA Kernel

The CUDA implementation can be written as follows. It needs to be compiled with a CUDA compiler such as nvcc, the NVIDIA CUDA compiler driver. It must be noted that the vector pointer returned by STARPU_VECTOR_GET_PTR is here a pointer in GPU memory, so that it can be passed as such to the kernel call vector_mult_cuda.

#include <starpu.h>
static __global__ void vector_mult_cuda(unsigned n, float *val,
float factor)
{
unsigned i = blockIdx.x*blockDim.x + threadIdx.x;
if (i < n)
val[i] *= factor;
}
extern "C" void scal_cuda_func(void *buffers[], void *_args)
{
float *factor = (float *)_args;
/* length of the vector */
unsigned n = STARPU_VECTOR_GET_NX(buffers[0]);
/* local copy of the vector pointer */
float *val = (float *)STARPU_VECTOR_GET_PTR(buffers[0]);
unsigned threads_per_block = 64;
unsigned nblocks = (n + threads_per_block-1) / threads_per_block;
vector_mult_cuda<<<nblocks,threads_per_block, 0, starpu_cuda_get_local_stream()>>>
(n, val, *factor);
cudaStreamSynchronize(starpu_cuda_get_local_stream());
}

Definition of the OpenCL Kernel

The OpenCL implementation can be written as follows. StarPU provides tools to compile a OpenCL kernel stored in a file.

__kernel void vector_mult_opencl(int nx, __global float* val, float factor)
{
const int i = get_global_id(0);
if (i < nx) {
val[i] *= factor;
}
}

Contrary to CUDA and CPU, STARPU_VECTOR_GET_DEV_HANDLE has to be used, which returns a cl_mem (which is not a device pointer, but an OpenCL handle), which can be passed as such to the OpenCL kernel. The difference is important when using partitioning, see Partitioning Data.

#include <starpu.h>
void scal_opencl_func(void *buffers[], void *_args)
{
float *factor = _args;
int id, devid, err; /* OpenCL specific code */
cl_kernel kernel; /* OpenCL specific code */
cl_command_queue queue; /* OpenCL specific code */
cl_event event; /* OpenCL specific code */
/* length of the vector */
unsigned n = STARPU_VECTOR_GET_NX(buffers[0]);
/* OpenCL copy of the vector pointer */
cl_mem val = (cl_mem)STARPU_VECTOR_GET_DEV_HANDLE(buffers[0]);
{ /* OpenCL specific code */
err = starpu_opencl_load_kernel(&kernel, &queue,
"vector_mult_opencl", /* Name of the codelet */
devid);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
err = clSetKernelArg(kernel, 0, sizeof(n), &n);
err |= clSetKernelArg(kernel, 1, sizeof(val), &val);
err |= clSetKernelArg(kernel, 2, sizeof(*factor), factor);
}
{ /* OpenCL specific code */
size_t global=n;
size_t local;
size_t s;
cl_device_id device;
starpu_opencl_get_device(devid, &device);
err = clGetKernelWorkGroupInfo (kernel, device, CL_KERNEL_WORK_GROUP_SIZE,
sizeof(local), &local, &s);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
if (local > global) local=global;
err = clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global, &local, 0,
NULL, &event);
if (err != CL_SUCCESS) STARPU_OPENCL_REPORT_ERROR(err);
}
{ /* OpenCL specific code */
clFinish(queue);
clReleaseEvent(event);
}
}

Definition of the Main Code

The CPU implementation is the same as in the previous section.

Here is the source of the main application. You can notice that the fields starpu_codelet::cuda_funcs and starpu_codelet::opencl_funcs are set to define the pointers to the CUDA and OpenCL implementations of the task.

/*
* This example demonstrates how to use StarPU to scale an array by a factor.
* It shows how to manipulate data with StarPU's data management library.
* 1- how to declare a piece of data to StarPU (starpu_vector_data_register)
* 2- how to describe which data are accessed by a task (task->handles[0])
* 3- how a kernel can manipulate the data (buffers[0].vector.ptr)
*/
#include <starpu.h>
#define NX 2048
extern void scal_cpu_func(void *buffers[], void *_args);
extern void scal_sse_func(void *buffers[], void *_args);
extern void scal_cuda_func(void *buffers[], void *_args);
extern void scal_opencl_func(void *buffers[], void *_args);
static struct starpu_codelet cl = {
/* CPU implementation of the codelet */
.cpu_funcs = { scal_cpu_func, scal_sse_func },
.cpu_funcs_name = { "scal_cpu_func", "scal_sse_func" },
#ifdef STARPU_USE_CUDA
/* CUDA implementation of the codelet */
.cuda_funcs = { scal_cuda_func },
#endif
#ifdef STARPU_USE_OPENCL
/* OpenCL implementation of the codelet */
.opencl_funcs = { scal_opencl_func },
#endif
.nbuffers = 1,
.modes = { STARPU_RW }
};
#ifdef STARPU_USE_OPENCL
#endif
int main(int argc, char **argv)
{
/* We consider a vector of float that is initialized just as any of C
* data */
float vector[NX];
unsigned i;
for (i = 0; i < NX; i++)
vector[i] = 1.0f;
fprintf(stderr, "BEFORE: First element was %f\n", vector[0]);
/* Initialize StarPU with default configuration */
starpu_init(NULL);
#ifdef STARPU_USE_OPENCL
"examples/basic_examples/vector_scal_opencl_kernel.cl", &programs, NULL);
#endif
/* Tell StaPU to associate the "vector" vector with the "vector_handle"
* identifier. When a task needs to access a piece of data, it should
* refer to the handle that is associated to it.
* In the case of the "vector" data interface:
* - the first argument of the registration method is a pointer to the
* handle that should describe the data
* - the second argument is the memory node where the data (ie. "vector")
* resides initially: STARPU_MAIN_RAM stands for an address in main memory, as
* opposed to an adress on a GPU for instance.
* - the third argument is the adress of the vector in RAM
* - the fourth argument is the number of elements in the vector
* - the fifth argument is the size of each element.
*/
starpu_data_handle_t vector_handle;
starpu_vector_data_register(&vector_handle, STARPU_MAIN_RAM, (uintptr_t)vector,
NX, sizeof(vector[0]));
float factor = 3.14;
/* create a synchronous task: any call to starpu_task_submit will block
* until it is terminated */
struct starpu_task *task = starpu_task_create();
task->synchronous = 1;
task->cl = &cl;
/* the codelet manipulates one buffer in RW mode */
task->handles[0] = vector_handle;
/* an argument is passed to the codelet, beware that this is a
* READ-ONLY buffer and that the codelet may be given a pointer to a
* COPY of the argument */
task->cl_arg = &factor;
task->cl_arg_size = sizeof(factor);
/* execute the task on any eligible computational ressource */
/* StarPU does not need to manipulate the array anymore so we can stop
* monitoring it */
starpu_data_unregister(vector_handle);
#ifdef STARPU_USE_OPENCL
#endif
/* terminate StarPU, no task can be submitted after */
fprintf(stderr, "AFTER First element is %f\n", vector[0]);
return 0;
}

Execution of Hybrid Vector Scaling

The Makefile given at the beginning of the section must be extended to give the rules to compile the CUDA source code. Note that the source file of the OpenCL kernel does not need to be compiled now, it will be compiled at run-time when calling the function starpu_opencl_load_opencl_from_file().

CFLAGS  += $(shell pkg-config --cflags starpu-1.2)
LDFLAGS += $(shell pkg-config --libs starpu-1.2)
CC       = gcc

vector_scal: vector_scal.o vector_scal_cpu.o vector_scal_cuda.o vector_scal_opencl.o

%.o: %.cu
       nvcc $(CFLAGS) $< -c $@

clean:
       rm -f vector_scal *.o
$ make

and to execute it, with the default configuration:

$ ./vector_scal
0.000000 3.000000 6.000000 9.000000 12.000000

or for example, by disabling CPU devices:

$ STARPU_NCPU=0 ./vector_scal
0.000000 3.000000 6.000000 9.000000 12.000000

or by disabling CUDA devices (which may permit to enable the use of OpenCL, see Enabling OpenCL) :

$ STARPU_NCUDA=0 ./vector_scal
0.000000 3.000000 6.000000 9.000000 12.000000