Utilization of multiport memories in data path synthesis

In this paper, we present a new scheduling algorithm that generates area-efficient register transfer level datapaths with multiport memories. The proposed scheduling algorithm assigns an operation to a specific control step such that maximal sharing of functional units can be achieved with minimal number of memory ports, while satisfying given constraints. We propose a measure of multiport memory cost, MAV (Multiple Access Variable) which is defined as a variable accessed at several control steps, and overall memory cost is reduced by equally distributing the MAVs throughout all the control steps. When compared with previous approaches for several benchmarks available from the literature, the proposed algorithm generates the datapaths with less memory modules and interconnection structures by reflecting the memory cost in the scheduling process.

At present, datapath synthesis techniques produce a design with a large number of isolated registers. Allocation of memory modules to implement these resisters are usually left to the designer. This paper presents new approach to the allocation of multiport memories which minimizes hardware costs in ASIC datapath synthesis. The proposed approach, AMD, considers not only the access requirements of registers but also the lifetime of registers. The objective is to minimize the requirement of registers and multiport memory modules under a given resource constraints simultaneously. The minimization problem has been modeled as a 0-1 integer linear programming problem. This approach is illustrated with an example.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Grouping variables into multiport memories is an essential step for multiport memory based data path synthesis. The proposed system, GMD, not only groups variables into a minimum number of multiport memory modules, but also simultaneously minimizes the number of registers in each memory module. The minimization problem is formulated as a 0-1 integer linear programming (ILP) problem. Experiments on benchmarks are promising. >

In order to optimally allocate multiport memories in datapath synthesis, registers are simultaneously assigned to a configuration of several memories; this gives a more uniform distribution of register activity across the memories and usually provides a more compact assignment, allowing even fewer ports and fewer module interconnections. The LP model is extended, and a fast heuristic approach that searches for an allocation is presented. This allocation algorithm is part of a more general design tool that generates alternative multiport memory configurations and allows their rapid exploration. The designer specifies a range of design parameters, and can control the thoroughness of the search. Within the limits set by the designer, the program finds the minimum cost configuration having a feasible register allocation. The components of this system are described, and some examples of its use are presented.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

A limiting performance factor for soft cores that exploit high instruction-level parallelism (ILP) is the need of a high performance multiport memory to provide data at the rate it can be consumed by the available functional units. At the same time, FPGAs contain block RAMs (BRAMs) that provide enormous bandwidth, especially when data is spread through multiple blocks of independent address spaces. We present a novel technique to cope with the multiport memory problem and increase performance in very long instruction word (VLIW) FPGA-based soft-core processors. We emulate the use of a multi-ported memory system through the use of multiple single-ported memories, which are controlled at the software level. Experimental results demonstrate that our technique can improve performance on data-driven and control-driven applications where data reuse is found. Measurements show performance improvements of up to 2.15x for a Filter benchmark, a data-driven application, and 1.25x for a Knuth-Morris-Pratt (KMP) string search, a control-driven application. These improvements are achieved while maintaining the same area of the baseline soft processor core.

Along with the functional units, storage and interconnects also contribute significantly to the data path cost. This paper addresses the issue of reducing the storage and interconnect cost by allocating queues for storing variables. In contrast to the earlier works, we support regular cdfgs and multi-cycle functional units for queue synthesis. Initial results on HLS benchmark examples have been encouraging and show the potential of data path cost reduction using queue synthesis. A novel feature of our work is the formulation of the problem for a variety of FIFO structures with their own queueing criteria.

A combinatorial optimization technique has been developed and applied to the scheduling problem in data path synthesis. The cost function is minimized using a gradient-like method, and constraints are satisfied by a new technique analogous to the combination of the penalty method and the feasible direction method used for nonlinear optimization. To overcome the drawbacks of assigning operations to control steps one at a time, this technique assigns all the operations to control steps simultaneously. Experimental results show that this method is as good as or better than other published methods.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

ShiftQs are hardware structures consisting of registers and switches which buffer and transport operands among function units within custom hardware loop accelerators. ShiftQs help minimize buffering and interconnect costs by customizing the hardware to the given schedule and by intelligent sharing of register and interconnect resources. This paper describes the ShiftQ schema and a method to automatically synthesize them from modulo-scheduled loops. We also evaluate the cost savings by comparing them against traditional storage and interconnect mechanisms.

On-line testable data path synthesis for minimizing testing time

We address the problem of minimizing power consumption in behavioral synthesis of data-dominated circuits. The complex nature of power as a cost function implies that the effects of several behavioral synthesis tasks like module selection, clock selection, scheduling, and resource sharing on supply voltage and switched capacitance need to be considered simultaneously to fully derive the benefits of design space exploration at the behavior level. Recent work has established the importance of behavioral synthesis in low power VLSI design. However, most of the algorithms that have been proposed separate these tasks and perform them sequentially, and are hence not able to explore the tradeoffs possible due to their interaction. We present an efficient algorithm for performing scheduling, clock selection, module selection, and resource allocation and assignment simultaneously with an aim of reducing the power consumption in the synthesized data path. The algorithm, which is based on an iterative improvement strategy, is capable of escaping local minima in its search for a low power solution. The algorithm considers diverse module libraries and complex scheduling constructs such as multicycling, chaining, and structural pipelining. We describe supply voltage and clock pruning strategies that significantly improve the efficiency of our algorithm by cutting down on the computational effort involved in exploring candidate supply voltages and clock periods that are unlikely to lead to the best solution. Experimental results are reported to demonstrate the effectiveness of the algorithm. Our techniques can be combined with other known methods of behavioral power optimization like data path replication and transformations, to result in a complete data path synthesis system for low power applications.

Recently there has been a trend for the designer to group registers into register files for efficiently implementing large VLSI chips. Multiport memories provide an effective way for such an implementation. Interconnection minimization (such as multiplexers and tristate buffers) has become more difficult with the use of multiport memories. A heuristic is presented which performs functional units and connection allocation tasks simultaneously to get better results for application specific designs assuming that registers have already been grouped to multiport memories. Experiments on benchmarks show very promising results. >

In this paper we examine the multi-criteria optimization involved in scheduling for data path synthesis (DPS). We present a method to find non-dominated schedules using a combination of restricted search and heuristic scheduling techniques. Our method supports design with architectural constraints such as the total number of functional units, buses, etc. The schedules produced have been taken to completion using GA-BIND and the results are promising.

FPGAs (Field Programmable Gate Arrays) are making their way into data centers (DC). They are used as accelerators to boost the compute power of individual server nodes and to improve the overall power efficiency. Meanwhile, DC infrastructures are being redesigned to pack ever more compute capacity into the same volume and power envelops. This redesign leads to the disaggregation of the server and its resources into a collection of standalone computing, memory, and storage modules. To embrace this evolution, we propose an architecture that decouples the FPGA from the CPU of the server by connecting the FPGA directly to the DC network. This proposal turns the FPGA into a network-attached computing resource that can be incorporated with disaggregated servers into these emerging data centers. We implemented a prototype and compared its network performance with that obtained from bare metal servers (Native), virtual machines (VM), and containers (CT). The results show that standalone network-attached FPGAs outperform them in terms of network latency and throughput by a factor of up to 35x and 73x, respectively. We also observed that the proposed architecture consumes only 14% of the total FPGA resources.

OASIS is a cell-based silicon compiler system that supports rapid implementation of digital Application specific ICs (ASICs). The system automates the design process by taking a high-level language description and producing a layout of a testable IC. OASIS facilitates design of large digital systems on silicon and gives the designer sufficient control to enable exploration of the speed-versus-area design tradeoffs. The front end of OASIS is the LOGIC-III compiler that translates a high-level language description into a hierarchical netlist of gates. The LOGIC-III language is a PASCAL-like language that allows the designer to specify his design completely and in a uniform, consistent manner. The language supports both the logic synthesis for FSMs and combinational logic, and the structural/algorithmic synthesis of data paths and regular structures such as adders or multipliers. The OASIS system comprises an integrated set of tools for complete design, simulation, verification, layout and testing of ASICs. Experience with the system shows a marked increase in design productivity when compared with more traditional design techniques. The authors describe the system with its unique features and present some of the results of using OASIS in designing semi-custom ASICs. >

A multiport memory allocation problem for hierarchical, i.e. multi-dimensional, data streams is described. Memory allocation techniques are used in high level synthesis for foreground and background memory allocation, the design of data format converters, and the design of synchronous inter-processor communication hardware. The techniques presented in this paper differ from other approaches in the sense that data streams are considered to be design entities and are not expanded to individual samples. A formal model for hierarchical data streams is given and a memory allocation algorithm is presented. The algorithm comprises two steps: data routing and assignment of signal delays to memories. A number of sub-problems are formulated as ILP programs. In the presented form, the allocation algorithm only considers interconnect costs, but memory size and other cost factors can be taken into account. The presented work is implemented in the memory allocation tool MEDEA which is part of the PHIDEO synthesis system.