The row-wise accumulator is implemened by the L1 primitive cscRow. This primitive basically multiplies the values of multiple NNZ entries with their correponding dense column vector values, and accumulates the results according to the row indices. The basic functions used by this primitive include xBarRow, rowMemAcc and rowAgg. The xBarRow primitive includes formRowEntry logic for multiplying the NNZ values with the corresponding input column vector entries and the split, merge logic for distributing the multiplication results to the corresponding row banks. The rowMemAcc primitives accumulates the intermediate results in on-chip memories. Multiple on-chip memory buffers are provided to remove the floating pointer accumulation bubbles. The rowAgg primitive collects the results from all accumulators and outputs the results in sequence.
For more information, see Row-wise Accumulator Implementation.