Input matrix data must be provided in column-major order. TP_DIM_ROWS must be a multiple of vecSampleNum and divisible by TP_CASC_LEN; TP_DIM_ROWS / TP_CASC_LEN must also be a multiple of vecSampleNum. Not all combinations of TP_DIM_ROWS, TP_DIM_COLS, and TP_CASC_LEN are valid; the per-kernel data memory footprint must fit within the AIE tile data memory. If this constraint is violated, increase TP_CASC_LEN or reduce TP_DIM_ROWS or TP_DIM_COLS.