ViennaCL - The Vienna Computing Library
1.5.2
|
A STL-type const-iterator for vector elements. Elements can be accessed, but cannot be manipulated. VERY SLOW!! More...
#include <forwards.h>
Public Types | |
typedef scalar< SCALARTYPE > | value_type |
typedef vcl_ptrdiff_t | difference_type |
typedef viennacl::backend::mem_handle | handle_type |
Public Member Functions | |
const_vector_iterator (vector_base< SCALARTYPE > const &vec, vcl_size_t index, vcl_size_t start=0, vcl_ptrdiff_t stride=1) | |
Constructor. More... | |
const_vector_iterator (handle_type const &elements, vcl_size_t index, vcl_size_t start=0, vcl_ptrdiff_t stride=1) | |
Constructor for vector-like treatment of arbitrary buffers. More... | |
value_type | operator* (void) const |
Dereferences the iterator and returns the value of the element. For convenience only, performance is poor due to OpenCL overhead! More... | |
self_type | operator++ (void) |
self_type | operator++ (int) |
bool | operator== (self_type const &other) const |
bool | operator!= (self_type const &other) const |
difference_type | operator- (self_type const &other) const |
self_type | operator+ (difference_type diff) const |
vcl_size_t | offset () const |
Offset of the current element index with respect to the beginning of the buffer. More... | |
vcl_size_t | stride () const |
Index increment in the underlying buffer when incrementing the iterator to the next element. More... | |
handle_type const & | handle () const |
Protected Attributes | |
handle_type const & | elements_ |
The index of the entry the iterator is currently pointing to. More... | |
vcl_size_t | index_ |
vcl_size_t | start_ |
vcl_ptrdiff_t | stride_ |
A STL-type const-iterator for vector elements. Elements can be accessed, but cannot be manipulated. VERY SLOW!!
Every dereference operation initiates a transfer from the GPU to the CPU. The overhead of such a transfer is around 50us, so 20.000 dereferences take one second. This is four orders of magnitude slower than similar dereferences on the CPU. However, increments and comparisons of iterators is as fast as for CPU types. If you need a fast iterator, copy the whole vector to the CPU first and iterate over the CPU object, e.g. std::vector<float> temp; copy(gpu_vector, temp); for (std::vector<float>::const_iterator iter = temp.begin(); iter != temp.end(); ++iter) { //do something } Note that you may obtain inconsistent data if entries of gpu_vector are manipulated elsewhere in the meanwhile.
SCALARTYPE | The underlying floating point type (either float or double) |
ALIGNMENT | Alignment of the underlying vector, |
typedef vcl_ptrdiff_t difference_type |
typedef scalar<SCALARTYPE> value_type |
|
inline |
Constructor.
vec | The vector over which to iterate |
index | The starting index of the iterator |
start | First index of the element in the vector pointed to be the iterator (for vector_range and vector_slice) |
stride | Stride for the support of vector_slice |
|
inline |
Constructor for vector-like treatment of arbitrary buffers.
elements | The buffer over which to iterate |
index | The starting index of the iterator |
start | First index of the element in the vector pointed to be the iterator (for vector_range and vector_slice) |
stride | Stride for the support of vector_slice |
|
inline |
|
inline |
Offset of the current element index with respect to the beginning of the buffer.
|
inline |
|
inline |
Dereferences the iterator and returns the value of the element. For convenience only, performance is poor due to OpenCL overhead!
|
inline |
|
inline |
|
inline |
|
inline |
|
inline |
|
inline |
Index increment in the underlying buffer when incrementing the iterator to the next element.
|
protected |
The index of the entry the iterator is currently pointing to.
|
protected |
|
protected |
|
protected |