API Functions of xf::sparse¶
xBarCol¶
#include "xf_sparse/cscmv.hpp"
template < unsigned int t_LogParEntries, typename t_DataType, typename t_IndexType = unsigned int, unsigned int t_DataBits = 32, unsigned int t_IndexBits = 32 > void xBarCol ( const unsigned int p_colPtrBlocks, const unsigned int p_nnzBlocks )
xBarCol function that distributes input col values to the dedicated banks according to their col index pointers
Parameters:
| t_LogParEntries | log2 of the parallelly processed entries in the input/output vector stream |
| t_DataType | the data type of the matrix and vector entries |
| t_IndexType | the data type of the indicies |
| t_DataBits | the number of bits for storing the data |
| t_IndexBits | the number of bits for storing the indices |
| p_colPtrBlocks | the number of col index pointer blocks |
| p_nnzBlocks | the number of NNZ blocks |
| p_colPtrStr | the input col pointer vector stream |
| p_colValStr | the input col value vector stream |
| p_nnzColValStr | the output banked col value vector stream |
xBarRow¶
#include "xf_sparse/cscmv.hpp"
template < unsigned int t_LogParEntries, typename t_DataType, typename t_IndexType = unsigned int, unsigned int t_DataBits = 32, unsigned int t_IndexBits = 32 > void xBarRow (const unsigned int p_nnzBlocks)
xBarRow function that multiplies input NNZs’ values with input vectors and distributes the results to the dedicated banks according to their row indices
Parameters:
| t_LogParEntries | log2 of the parallelly processed entries in the input/output vector stream |
| t_DataType | the data type of the matrix and vector entries |
| t_IndexType | the data type of the indicies |
| t_DataBits | the number of bits for storing the data |
| t_IndexBits | the number of bits for storing the indices |
| p_nnzBlocks | the number of NNZ blocks |
| p_nnzValStr | the input NNZ value stream |
| p_nnzColValStr | the input col value stream |
| p_rowIndexStr | the inpuut NNZ row index stream |
| p_rowEntryStr | the output banked multiplication results stream array |
| p_isEndStr | the output control stream |
rowAgg¶
#include "xf_sparse/cscmv.hpp"
template < unsigned int t_ParEntries, unsigned int t_ParGroups, typename t_DataType, typename t_IndexType, unsigned int t_DataBits = 32 > void rowAgg ( const unsigned int p_rowBlocks, hls::stream <ap_uint <t_DataBits>> p_rowValStr [t_ParEntries], hls::stream <ap_uint <t_DataBits*t_ParEntries>>& p_rowAggStr )
rowAgg function that aggregates multiple row entry streams into one row entry stream
Parameters:
| t_ParEntries | the parallelly processed entries in the input/output vector stream |
| t_ParGroups | the number of parallel accumulation paths |
| t_DataType | the data type of the matrix and vector entries |
| t_IndexType | the data type of the indicies |
| t_DataBits | the number of bits for storing the data |
| p_rowBlocks | the number of row blocks |
| p_rowValStr | the iutput row entry stream array |
| p_rowAggStr | the output aggregated row entry stream |
cscRow¶
#include "xf_sparse/cscmv.hpp"
template < unsigned int t_MaxRowBlocks, unsigned int t_LogParEntries, unsigned int t_LogParGroups, typename t_DataType, typename t_IndexType = unsigned int, unsigned int t_DataBits = 32, unsigned int t_IndexBits = 32 > void cscRow ( const unsigned int p_nnzBlocks, const unsigned int p_rowBlocks )
cscRow function that returns the multiplication results of a sparse matrix and a dense vector
Parameters:
| t_MaxRowBlocks | the maximum number of row entrie blocks buffered onchip per PE |
| t_LogParEntries | log2 of the parallelly processed entries in the input/output vector stream |
| t_LogParGroups | log2 of the number of parallel accumulation paths |
| t_DataType | the data type of the matrix and vector entries |
| t_IndexType | the data type of the indicies |
| t_DataBits | the number of bits for storing the data |
| t_IndexBits | the number of bits for storing the indices |
| p_nnzBlocks | the number of NNZ vector blocks |
| p_rowBlocks | the number of result row vector blocks |
| p_nnzValStr | the input NNZ value vector stream |
| p_nnzColValStr | the input col vector stream |
| p_rowIndexStr | the input NNZ index vector stream |
| p_rowAggStr | the output row vector stream |
dispColVec¶
#include "xf_sparse/moverL1.hpp"
template < unsigned int t_MaxColParBlocks, unsigned int t_ParBlocks4Param, unsigned int t_HbmChannels, unsigned int t_ParEntries, unsigned int t_DataBits > void dispColVec ( const unsigned int t_chId, hls::stream <ap_uint <t_DataBits*t_ParEntries>>& p_datStr, hls::stream <ap_uint <t_DataBits*t_ParEntries>>& p_datFwdStr, hls::stream <ap_uint <t_DataBits*t_ParEntries>>& p_datOutStr )
dispColVec function that forward and copy input column vector and parameters
Parameters:
| t_MaxColParBlocks | the maximum number of parallel processed column blocks buffered in on-chip memory |
| t_ParBlocks4Param | the number of parallelly processed parameter blocks |
| t_HbmChannels | number of HBM channels |
| t_ParEntries | parallelly processed entries |
| t_DataBits | number of bits used to store each entry |
| t_chId | constant HBM channel ID |
| p_datStr | input vector stream |
| p_datFwdStr | an forwarded parameter and column vector streams |
| p_datOutStr | an copied parameter and column vector streams |
dispCol¶
#include "xf_sparse/moverL1.hpp"
template < unsigned int t_MaxColParBlocks, unsigned int t_ParBlocks4Param, unsigned int t_HbmChannels, unsigned int t_ParEntries, unsigned int t_DataBits > void dispCol ( hls::stream <ap_uint <t_DataBits*t_ParEntries>>& p_datStr, hls::stream <ap_uint <t_DataBits*t_ParEntries>> p_datOutStr [t_HbmChannels] )
dispCol function that dispatchs input column vectors accross parallel CUs for computing SpMV simultaneously
Parameters:
| t_MaxColParBlocks | the maximum number of parallelly processed column vector entries in the on-chip buffer |
| t_ParBlocks4Param | the number of parallelly processed parameter blocks |
| t_HbmChannels | number of HBM channels |
| t_ParEntries | parallelly processed entries |
| t_DataBits | number of bits used to store each entry |
| p_datStr | input vector stream |
| p_datOutStr | an output array of column vector streams |
dispNnzCol¶
#include "xf_sparse/moverL1.hpp"
template < unsigned int t_MaxColParBlocks, unsigned int t_ParBlocks4Param, unsigned int t_HbmChannels, unsigned int t_ParEntries, unsigned int t_DataBits > void dispNnzCol ( hls::stream <ap_uint <t_DataBits*t_ParEntries>>& p_datStr, hls::stream <ap_uint <t_DataBits*t_ParEntries>> p_datOutStr [t_HbmChannels] )
dispNnzCol function that dispatchs NNZ Col pointer entries accross parallel compute CUs
Parameters:
| t_MaxColParBlocks | the maximum number of parallelly processed column entries in the on-chip buffer |
| t_ParBlocks4Param | the number of parallelly processed parameter blocks |
| t_HbmChannels | number of HBM channels |
| t_ParEntries | parallelly processed entries |
| t_DataBits | number of bits used to store each entry |
| p_datStr | input vector stream |
| p_datOutStr | an output array of vector streams |