High performance reconfigurable multi-processor-based computing on FPGAs

Abstract

Multi-processor architectures are a promising solution to provide the required computational performance for applications in the area of high performance computing. Multi- and many-core Systems-on-Chip offer the possibility to host an application, partitioned in a number of tasks, on the different cores on one silicon die. Unfortunately, a partitioning of the tasks near to the performance optimum is the challenge in this domain and often a show-stopper for the success story of multi- and many-core hardware. The missing feature of these architectures is runtime adaptivity of the underlying hardware, which offers to tailor the hardware to the application in order to meet the task mapping process coming from top-down development. Especially, this Meet-in-the- Middle solution offers the novel hardware and software approach of RAMPSoC, which is described in this paper

Overview on Strategies and Approaches for FPGA Programming

“…A construction methodology perspective may concentrate on any level of abstraction defined within this process: Recent literature on VHDL programming perspectives describe a number of distinct methodologies. All of these methodologies can be grouped into three mainstreams that differ in design philosophy [26][27][28][29]: a. Top-down approach (behavioural)…”

“…The complexity of designed ICs favoured the development of new design methodologies based on abstraction. Three strategies may be distinguished in design methodologies [26][27][28][29]: -Top-Down, proceeds hierarchically from an abstract level to a more detailed one by successive decomposition into subsystems. With top-down design, the circuit can first be modelled by making a behavioural model.…”

An Efficient Evolutionary Task Scheduling/Binding Framework for Reconfigurable Systems

“…The work in [16] proposes a hardware OS, called CAP-OS, that uses run-time scheduling, task mapping, and resource management on a run-time reconfigurable multiprocessor system. The authors in [17] proposed to use the RAMPSOC with two schedulers: a static scheduler to generate a task graph and assign resources and a dynamic scheduler that uses the generated task graph produced by the static scheduler. A deadline constraint for the entire graph is imposed instead of individual tasks.…”

“…(2) ReadyQ =findAvailableTasks(DFGs)//no dependencies (3) while there are unscheduled tasks do (4) if (Reuse = true) (5) for all tasks in ReadyQ do (6) PRR# = FindAvailablePRR(task) (7) if (PRR#) then (8) ts=t (9) for all p in PredecessorOf(task) (10) ts=max(ts, tps + tpl) (11) end for (12) remove task from ReadyQ (13) end if (14) end for (15) else /⋆Reuse = false⋆/ (16) for all tasks in ReadyQ do (17) PRR# = findEmptyPRR(task) (18) if (PRR# and Reconfiguration) then (19) ts=t+trs (20) for all p in PredecessorOf(task) (21) ts=max(ts, tps + tpl) (22) end for (23) Remove task from ReadyQ (24) end if (25) end for (26) end if (27) ReadyQ=CheckforAvailableTasks /⋆ dependencies met ⋆/ (28) end while Algorithm 1: Pseudo-code for RCOffline.…”

“…(1) // The PRRS should be stored from the smallest to largest such that (2) // those with smaller reconfiguration time have smaller IDs (3) Sort PRRs in ascending order based on size. (4) Read the node with the highest task priority (5) if (All dependencies are met) (6) {add task to ready queue} (7) // SELECTION BY RCOnline versus RCOnline-Enh (8) Fetch the first task from the ReadyQueue // --used by RCOnline (9) Fetch a task that matches an existing reconfigured task // --used by RCOnline-Enh (10) switch (task->mode) (11) { case Hybrid HW : (12) { if(TaskTypePriority==0 and GPP is free) (13) {// TaskTypePriority ==0 indicates that task is preferred to run on SW (14) Migrate task from HybridHW to HybridSW (15) Add to the beginning of the ReadyQueue} (16) }else{ (17) if (there is a free PRR(s) that fits the current task) (18) {{Search Free PRRs for a match (19) against the ready task} (20) if (task found) (21) run ready task on the free PRR (22) else{ (23) currentPRR= smallest free PRR; (24) if(TaskTypePriority< currentPRR and GPP is free) (25) { // check if it's faster to reconfigure or run on SW (26) Migrate task from HybridHW to HybridSW (27) Add to the beginning of the ReadyQueue } (28) }else{ (29) Reconfigure currentPRR with the ready task bit-stream}} (30) }else // all PRRs are busy (31) {if ( GPP is free) (32)…”