This page provides the column vector buffering and distribution implementation details. The following figure shows the column vector buffering and distribution logic.
- The input parameter streams contain the information of the size of each column vector block, the minimum and maximum column vector entry indices.
- The
dispColVec
module reads the parameters and multiple column vector enties, buffers the column entires in its own on-chip memory and forward the rest parameters and vector entires to the nextdisColVec
module. If the module is the last one in the chain, the forwarding logic is omitted. - After the buffereing operation, each
dispColVec
module reads out the data from the on-chip memory and sends them to the output stream to be processed by its own computation path. - Apart from buffering and reading data operations, the
dispColVec
module also aligns the data and pads the data according toSPARSE_parEntries
and the minimumn and maximum row indices.