Fig. 5

Details of Dataflow Kernel in the Pipeline CU. The Data Loading Stage loads reads and events from DRAM using vector access. Events/reads are decomposed into individual events as float type or bases as char type by the Event Vector Distributor and the Read Vector Distributor, respectively. The float event is converted into a fixed-point representation and sent to the Compute Stage or the Post Stage by Event Generator. the Kmer Generator converts consecutive bases into k-mers. The Compute Stage performs alignment and generates score and trace tables. The max score collected by the Max Stage and the reverse trace table produced by the Trace Out Stage are transferred to the Post Stage for performing backtracking. All inputs and outputs are connected via stream FIFOs, allowing all stages and sub-functions to run concurrently