Content

Example 2: Vector

The main vector type in ViennaCL is vector<T, alignment>, representing a chunk of memory on the compute device. T is the underlying scalar type (either float or double, complex types are not supported in ViennaCL 1.0.x) and alignment denotes the memory the vector is aligned to (in multiples of sizeof(T)). For example, a vector initialized to have 55 entries and an alignment of 16 will reside in a block of memory equal to 64 entries. However, memory alignment is fully transparent, so from the end-user's point of view, alignment allows to tune the library for maximum speed on the available compute device.

Handling vectors in ViennaCL

typedef float        ScalarType;
//typedef double    ScalarType; //use this if your GPU supports double precision
 
// Define a few CPU vectors using the STL
std::vector<ScalarType>      std_vec1(10);
std::vector<ScalarType>      std_vec2(10);
std::vector<ScalarType>      std_vec3(10);
 
// Define a few GPU vectors using ViennaCL
viennacl::vector<ScalarType> vcl_vec1(10);
viennacl::vector<ScalarType> vcl_vec2(10);
viennacl::vector<ScalarType> vcl_vec3(10);
 
// Fill the CPU vectors with random values:
// (random<> is a helper function defined elsewhere)
  
for (unsigned int i = 0; i < 10; ++i)
{
  std_vec1[i] = random<ScalarType>(); 
  vcl_vec2[i] = random<ScalarType>();  //also works for GPU vectors, but is slow!
  std_vec3[i] = random<ScalarType>(); 
}
  
// Copy the CPU vectors to the GPU vectors and vice versa
copy(std_vec1.begin(), std_vec1.end(), vcl_vec1.begin()); //either the STL way
copy(vcl_vec2.begin(), vcl_vec2.end(), std_vec2.begin());
copy(std_vec3, vcl_vec3); //or using the short hand notation
copy(vcl_vec2, std_vec2);
 
// Compute the inner product of two GPU vectors and write the result to either CPU or GPU
vcl_s1 = viennacl::linalg::inner_prod(vcl_vec1, vcl_vec2);
s1     = viennacl::linalg::inner_prod(vcl_vec1, vcl_vec2);
 
// Compute norms:
s1     = viennacl::linalg::norm_1(vcl_vec1);
vcl_s2 = viennacl::linalg::norm_2(vcl_vec2);
s3     = viennacl::linalg::norm_inf(vcl_vec3);
  
// Use viennacl::vector via the overloaded operators just as you would write it on paper:
vcl_vec1 = vcl_s1 * vcl_vec2 / vcl_s3;
vcl_vec1 = vcl_vec2 / vcl_s1 + vcl_s2 * (vcl_vec1 - vcl_s2 * vcl_vec2);