int main()
{
typedef float ScalarType;
//
// Initialize OpenCL vectors:
//
unsigned int vector_size = 10;
viennacl::scalar<ScalarType> s = 1.0; //dummy
viennacl::vector<ScalarType> vec1(vector_size);
viennacl::vector<ScalarType> vec2(vector_size);
viennacl::vector<ScalarType> result_mul(vector_size);
viennacl::vector<ScalarType> result_div(vector_size);
//
// fill the operands vec1 and vec2:
//
for (unsigned int i=0; i<vector_size; ++i)
{
vec1[i] = static_cast<ScalarType>(i);
vec2[i] = static_cast<ScalarType>(vector_size-i);
}
//
// Set up the OpenCL program given in my_compute_kernel:
// A program is one compilation unit and can hold many different compute kernels.
//
viennacl::ocl::program & my_prog = viennacl::ocl::current_context().add_program(my_compute_program, "my_compute_program");
my_prog.add_kernel("elementwise_prod"); //register elementwise product kernel
my_prog.add_kernel("elementwise_div"); //register elementwise division kernel
//
// Now we can get the kernels from the program 'my_program'.
// (Note that first all kernels need to be registered via add_kernel() before get_kernel() can be called,
// otherwise existing references might be invalidated)
//
viennacl::ocl::kernel & my_kernel_mul = my_prog.get_kernel("elementwise_prod");
viennacl::ocl::kernel & my_kernel_div = my_prog.get_kernel("elementwise_div");
//
// Launch the kernel with 'vector_size' threads in one work group
// Note that size_t might differ between host and device. Thus, a cast to cl_uint is necessary for the forth argument.
//
viennacl::ocl::enqueue(my_kernel_mul(vec1, vec2, result_mul, static_cast<cl_uint>(vec1.size())));
viennacl::ocl::enqueue(my_kernel_div(vec1, vec2, result_div, static_cast<cl_uint>(vec1.size())));
//
// Print the result:
//
std::cout << " vec1: " << vec1 << std::endl;
std::cout << " vec2: " << vec2 << std::endl;
std::cout << "vec1 .* vec2: " << result_mul << std::endl;
std::cout << "vec1 /* vec2: " << result_div << std::endl;
std::cout << "norm_2(vec1 .* vec2): " << viennacl::linalg::norm_2(result_mul) << std::endl;
std::cout << "norm_2(vec1 /* vec2): " << viennacl::linalg::norm_2(result_div) << std::endl;
//
// That's it.
//
std::cout << "!!!! TUTORIAL COMPLETED SUCCESSFULLY !!!!" << std::endl;
return 0;
}
/**
* Since no auxiliary routines are needed, we can directly start with main().
**/
int main()
{
typedef float ScalarType;
/**
* Initialize OpenCL vectors:
**/
unsigned int vector_size = 10;
viennacl::vector<ScalarType> vec1(vector_size);
viennacl::vector<ScalarType> vec2(vector_size);
viennacl::vector<ScalarType> result_mul(vector_size);
viennacl::vector<ScalarType> result_div(vector_size);
/**
* Fill the operands vec1 and vec2 with some numbers.
**/
for (unsigned int i=0; i<vector_size; ++i)
{
vec1[i] = static_cast<ScalarType>(i);
vec2[i] = static_cast<ScalarType>(vector_size-i);
}
/**
* Set up the OpenCL program given in my_compute_kernel:
* A program is one compilation unit and can hold many different compute kernels.
**/
viennacl::ocl::program & my_prog = viennacl::ocl::current_context().add_program(my_compute_program, "my_compute_program");
// Note: Releases older than ViennaCL 1.5.0 required calls to add_kernel(). This is no longer needed, the respective interface has been removed.
/**
* Now we can get the kernels from the program 'my_program'.
* (Note that first all kernels need to be registered via add_kernel() before get_kernel() can be called,
* otherwise existing references might be invalidated)
**/
viennacl::ocl::kernel & my_kernel_mul = my_prog.get_kernel("elementwise_prod");
viennacl::ocl::kernel & my_kernel_div = my_prog.get_kernel("elementwise_div");
/**
* Launch the kernel with 'vector_size' threads in one work group
* Note that std::size_t might differ between host and device. Thus, a cast to cl_uint is necessary for the forth argument.
**/
viennacl::ocl::enqueue(my_kernel_mul(vec1, vec2, result_mul, static_cast<cl_uint>(vec1.size())));
viennacl::ocl::enqueue(my_kernel_div(vec1, vec2, result_div, static_cast<cl_uint>(vec1.size())));
/**
* Print the result:
**/
std::cout << " vec1: " << vec1 << std::endl;
std::cout << " vec2: " << vec2 << std::endl;
std::cout << "vec1 .* vec2: " << result_mul << std::endl;
std::cout << "vec1 /* vec2: " << result_div << std::endl;
std::cout << "norm_2(vec1 .* vec2): " << viennacl::linalg::norm_2(result_mul) << std::endl;
std::cout << "norm_2(vec1 /* vec2): " << viennacl::linalg::norm_2(result_div) << std::endl;
/**
* We are already done. We only needed a few lines of code by letting ViennaCL deal with the details :-)
**/
std::cout << "!!!! TUTORIAL COMPLETED SUCCESSFULLY !!!!" << std::endl;
return EXIT_SUCCESS;
}
