|
graph
|
A COO matrix backed by CUDA Unified Memory, with a CSR rowPtr. More...
#include <coo.hpp>
Public Member Functions | |
| COO () | |
| empty CSR More... | |
| HOST DEVICE uint64_t | nnz () const |
| number of non-zeros More... | |
| uint64_t | num_nodes () const |
| number of unique row/col indices More... | |
| HOST DEVICE uint64_t | num_rows () const |
| number of matrix rows More... | |
| COOView< Index > | view () const |
| create a COOView for this COO More... | |
| const Index * | row_ptr () |
| row offset array More... | |
| const Index * | col_ind () |
| column index array More... | |
| const Index * | row_ind () |
Static Public Member Functions | |
| static COO< Index > | from_edgelist (const EdgeList &es, bool(*edgeFilter)(const Edge &)=nullptr) |
Public Attributes | |
| Vector< Index > | rowPtr_ |
| offset in col_ that each row starts at More... | |
| Vector< Index > | colInd_ |
| non-zero column indices More... | |
| Vector< Index > | rowInd_ |
| non-zero row indices More... | |
Private Attributes | |
| Index | maxCol_ |
A COO matrix backed by CUDA Unified Memory, with a CSR rowPtr.
Copying to a GPU kernel by value will cause the underling memory to be copied as well. For read-only GPU access, use the view() method to get a lightweight reference to the data.
|
inline |
column index array
|
static |
Build a COO from an EdgeList
Do not include edges where edgeFilter(edge) returns true
number of non-zeros
| uint64_t COO< Index >::num_nodes | ( | ) | const |
number of unique row/col indices
number of matrix rows
|
inline |
|
inline |
row offset array
| Vector<Index> COO< Index >::colInd_ |
non-zero column indices
|
private |
| Vector<Index> COO< Index >::rowInd_ |
non-zero row indices
| Vector<Index> COO< Index >::rowPtr_ |
offset in col_ that each row starts at
1.8.13