J. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Trans. CAD , 12(8):1107-1122, August 1993.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....edges in the input graph, resulting in a strongly polynomial time algorithm. The reason why we should care about such an algorithm is the wide applicability of SDCs. Below we list only a few applications without being exhaustive: # scheduling, e.g. operation scheduling in behavioral synthesis [20, 23, 28, 32], software scheduling [8, 22] job task scheduling [9] or data flow scheduling [31] # interface timing verification [3, 4] # rate analysis [10, 25] # real time systems, e.g. real time scheduling [27] run time monitoring timing constraints [29] or safety analysis of realtime systems ....

NESTOR,J.A.,AND KRISHNAMOORTHY,G.SALSA:A new approach to scheduling with timing constraints. IEEE Trans. Computer-Aided Design 12, 8 (Aug. 1993), 1107--22.

....approaches, in global approaches all control steps and all operations are considered simultaneously when operations are assigned to control steps. Examples of global scheduling are found in the integer linear programming (ILP) scheduling approaches [5, 17, 18] simulated annealing approach [19, 20], simulated evolution [21] tabu search [22] and genetic algorithm [23, 24] The performance of a scheduling algorithm can be evaluated using interdependent quality metrics such as the quality of the solution produced, the complexity of the algorithm and the possibility of handling large ....

....neighborhood structure in which moves are done by swapping the two nodes on the priority list which changes their priority. Since list scheduling is very sensitive to the priority assigned to the operations, we change the priority of the operations by moves. In most of the previous approaches [19, 20, 21, 22], moves are done in the solution domain. Tabu search Algorithm: TLS Input DFG, resources, tabu search parameters. Definitions X : Set of feasible solutions f : Objective function T : Tabu list A : Aspiration function max : Maximum number of iterations between improvement maxmov : Maximum ....

Nestor, J. A. and Krishnamoorthy, G. (1993) SALSA: a new approach to scheduling with timing constraints. IEEE Trans. Computer-Aided Design Integrated Circuits Syst., 12, 1107-- 1122.

....that tend to rely on application specific logic functions. The use of logic synthesis for estimates improves the quality of the synthesized designs and avoids erroneous high level decisions due to insufficient data or inappropriate assumptions. With the exception of CADDY [6] SAW [7] and SALSA [8], most synthesis approaches do not support detailed timing constraints. That is, they support either no timing constraints at all or they support at most constraints on the overall latency. This may be inadequate to describe complicated requirements on the timing of operations. SAW, because of ....

J. Nestor and G. Krishnamoorthy, "SALSA: A new approach to scheduling with timing constraints," in Pro- ceedings' of the International Conference on Computer-Aided Design, pp. 262-265, Nov. 1990.

....will be described for defining, in a safe way, minimum and maximum timing constraints in a VHDL input specification program. 2.1. Timing Definition Coherency The general problem of checking the consistency of a specification with timing constraints has been addressed, for instance, in [3] 11] [13]. Instead, concerning the use of VHDL as specification language for synthesis, the problem of defining a synthesizable subset of the whole language has been addressed in [12] 9] 4] 14] 7] 8] In pure VHDL, the typical algorithmic constructs, composed by high level sequential statements, can ....

J. Nestor, G. Krishnamoorthy, "SALSA: a New Approach to Scheduling with Timing Constraints", IEEE Transactions on Computer-Aided Design, Vol. 12, No. 8, August 1993.

....for mixed signal applications originates in accommodating physical level aspects i.e. DSN at higher levels of abstraction. We believe that a tight integration of HLS and power net routing permits ecient high level synthesis under DSN constraints. Most of the existing digital HLS methods [5] 7] [9] [10] are heuristics guided by speci c priority cost functions i.e. mobility, urgency, force etc, acting as approximate performance models. Addressing new performances i.e. DSN, however, involves a very tedious task of de ning and validating characteristic priority functions. Moreover, heuristic ....

.... scheduling method joins advantages of heuristic and exact methods: it uses exact performance models (similar to exact methods) and heuristic techniques for model optimization (similar to heuristic methods) Better quality solutions resulted than for existing heuristic methods i.e. SALSA HLS tool [9] or FDLS algorithm [10] 3. DSN EVALUATION This section presents modeling of noise propagation paths for transient (time) analysis of DSN. In [6] we discuss how the same concept can be used for AC (frequency) analysis, also. The two major noise propagation paths in a mixed signal IC are (1) ....

[Article contains additional citation context not shown here]

J. Nestor, G. Krishnamoorthy, \SALSA: A New Approachto Scheduling with Timing Constraints", #### ###### ## ######,May1993.

....of functional units simultaneously while minimizing the necessary numbers of registers and buses. However, his technique handled separately the most interdependent of all the allocation sub tasks, register binding and interconnection binding. SALSA also used simulated annealing for scheduling [8], and for allocation of ACM Transactions on Design Automation of Electronic Systems, Vol. vv, No. nn, Month Year. A Robust Datapath Allocation Method for Realistic System Design Delta 5 functional units and registers [6] During allocation improvement by simulated annealing, a cost function that ....

Nestor, J. A. and Krishnamoorthy, G., "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Trans. on Computer-Aided Design, vol.12, no.8, pp.1107-1122, Aug. 1993.

.... We have also found that the difference in area between the two solutions, yielding the two limits, is small (two to three registers) In our experiments, we used three well known benchmarks: the differential equation [DeM94] the AR lattice filter [Jai92] and the discrete cosine transform (DCT) [Nes93], with the characteristics presented in Table 4.2, for operands represented in 2 s complement. It is also assumed that the ALU mapped operations always take one clock cycle, and the multiplications take two clock cycles to execute. Table 4.2: The Benchmarks Used Design Operations Input Signals ....

Nestor, J.; Krishnamoorthy, G., "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Transactions on Computer- Aided Design of Integrated Circuits and Systems, vol. 12 8, Aug. 1993, pp 1107 - 1122.

....where the total FU cost is represented by the sum of the area of each FU weighted by the number of FUs of each type. It has a very large solution space and finding an optimal solution may require exhaustive search. Therefore, most existing scheduling methods such as FDS [3] FAMOS [4] SALSA [5], and SEHWA [6] use various heuristics to find good designs quickly. Although they find optimal solutions for some benchmarks, they may generate suboptimal results for others. On the other hand, an optimal scheduling approach which is based on Integer Linear Programming formulation is proposed in ....

....Results Our scheduling algorithm (LBS) has been implemented in C language under UNIX environments running on SUN workstations. Tables 1, 2, 3, and 4 show the experimental results of our method as well as the published results of the existing scheduling methods such as FDS [3] FAMOS [4] SALSA [5], SEHWA [6] and ALPS [7] Note that only ALPS and our system LBS guarantee the optimal scheduling unlike the other systems. Experiments of FDS, FAMOS, SALSA, SEHWA, ALPS, and LBS are performed on Xerox 1108 lisp machine, SUN 4 280, SUN IPC, VAX 11 750, VAX 11 8800, and 5 S1 NS1 optimal solution ....

[Article contains additional citation context not shown here]

J. A. Nestor and Ganesh Krishnamoorthy, "SALSA: A new Approach to Scheduling with Timing Constraints," IEEE Transactions on CAD, vol. 12, no. 8, pp.1107-1122, August 1993.

....of the estimates To demonstrate the quality of our estimation, we compared our results with actual designs obtained through conventional scheduling and allocation processes. We compared our results with actual designs generated by HAL [17] InSyn [18] ALPS [19] OASIC [12] ILP [20] and SALSA [22]. Tables 9 through 14 show the comparison results. Note that some of these systems did not optimize any resources other than Functional Units, in many cases resulting in designs which may be suboptimal with respect to registers. In such cases, we attempted to improve the FU optimal designs by ....

J. A. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Trans. CAD, vol. 12, no. 8, pp. 1107-1122, August 1993.

....maximum TCs between operation pairs [11] and global TCs to specify a maximum delay time to process a set of data. The term TCs has been previously used to refer to both local and global TCs, despite they are completely different. Approaches to solve scheduling with local TCs can be found in [7, 10, 11]. On the other hand, some Integer Linear Programming (ILP) approaches have been proposed to solve scheduling (not loop pipelining) with global TCs [1, 5] Force Directed Scheduling [12] solves both local and global TCs. This paper addresses loop pipelining with global TCs. Henceforth, we will ....

J. Nestor and G. Krishnamoorthy. SALSA: A new approach to scheduling with timing constraints. In Proc. Int. Conf. Computer-Aided Design, pages 262--265, November1990.

....module allocation and binding are interdependent with scheduling this restricts the possibilities to optimize the design and, moreover, removes the possibility to do trade offs between speed and cost (silicon area) during the synthesis process. 2.3. 4 Nestor et al. In the work of Nestor et al. Nes 93] and [Nes 93b] another approach for scheduling under timing constraints is described. In this case, the algorithm tries to minimize the hardware resource cost during the scheduling process. An extended control data flow graph called SALSA is used to represent the designs. The scheduling process ....

....and binding are interdependent with scheduling this restricts the possibilities to optimize the design and, moreover, removes the possibility to do trade offs between speed and cost (silicon area) during the synthesis process. 2.3. 4 Nestor et al. In the work of Nestor et al. Nes 93] and [Nes 93b] another approach for scheduling under timing constraints is described. In this case, the algorithm tries to minimize the hardware resource cost during the scheduling process. An extended control data flow graph called SALSA is used to represent the designs. The scheduling process is divided ....

[Article contains additional citation context not shown here]

J. Nestor, G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, vol. 12, no. 8, Aug. 1993, pp. 1107-22.

....Results Without Chaining VII. Experimental Results To demonstrate the accuracy and performance of Voyager s 3D scheduling methodology, we conducted a series of experiments using the well known AR lattice Filter (AR) 19] Elliptic Wave Filter (EWF) 6, p. 206] and Discrete Cosine Transform (DCT) [43] benchmarks. We used the VDP100 module library from [7] 8] giving a datapath delay of 48ns for addition, 56ns for subtraction, and 163ns for multiplication. For each benchmark, we performed Time Constrained 3D Scheduling (TCS 3D) and Resource Constrained 3D Scheduling (RCS 3D) with and without ....

J. A. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Transactions on Computer-Aided Design, vol. 12, pp. 1107--1122, Aug. 1993.

....of all possible precedence relaxations in polynomial time. 7 Experimental Results To demonstrate the accuracy and performance of Voyager s 3D scheduling methodology, we conducted a series of experiments using the well known Elliptic Wave Filter (EWF) 23, p. 206] and Discrete Cosine Transform [17] (DCT) benchmarks. We used the VDP100 module library from [16, 24] giving a datapath delay of 48ns for addition, 56ns for subtraction, and 163ns for multiplication. For each benchmark, we performed Time Constrained 3D Scheduling (TCS 3D) and Resource Constrained 3D Scheduling (RCS 3D) using the ....

J. A. Nestor and G. Krishnamoorthy. SALSA: A New Approach to Scheduling with Timing Constraints. IEEE Transactions on Computer-Aided Design, 12(8):1107--1122, Aug. 1993.

....Design Automation Conference, pp. 61 66, Chiba, Japan, 1995 critical path scheduling [Park86] and force directed (list) scheduling [Paul89] Verh92] are faster, but may produce unsatisfactory results because of their greedy characteristics. Methods based on approximation schemes [Deva89] [Nest90], Wehn91] may produce better results, but in general these algorithms suffer from large run times. In this article a new scheduling method is presented which is based on genetic algorithms. Some previous work about genetic algorithms used for high level synthesis scheduling will be presented, and ....

J.A. NESTOR AND G. KRISHNAMOORTHY. SALSA: A New Approach to Scheduling with Timing Constraints. In Digest of TechnicalPapersof the IEEE International Conference on Computer-Aided Design, pages 262--265, Santa Clara, November 1990. IEEE Computer Society Press.

....(2,1) 19 19 (2,1) 2,1) 2,1) 19 17 (2,1) 2,1) 2,1) 21 21 (2,1) 2,1) 2,1) 21 19 (2,1) 2,1) 2,1) TABLE III Number of FUs for the EWF (Using Pipelined Multiplier) VI. Experiments and Results Experiments were run on the Elliptic Wave Filter [16, p. 206] EWF) and Discrete Cosine Transform [17] (DCT) benchmarks. The results are listed in Table II, III, and IV, respectively. Given the time constraint c listed in the first column, the bounds computed by Voyager s FU lowerbounding method are reported in the column labeled Voyager, and the minimum number of FUs for which a feasible schedule ....

J. A. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Transactions on Computer-Aided Design, vol. 12, pp. 1107--1122, Aug. 1993.

....would block an operator move. Simple adaptations of these four moves support chaining. Analysis of the SALSA representation and the space of possible schedules has shown that from any starting schedule, every other legal schedule can be reached using this move set. This result, which is proven in [NES93], is important since it shows that a transformational search cannot be prevented from reaching the best possible schedule by the structure of the schedule space. The original SALSA scheduler used this move set to search the schedule space for lowcost schedules. In this approach, ASAP and ALAP ....

J. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints", to appear in IEEE Trans. CAD, 1993.

....(312) 567 3386 Fax: 312) 567 8976 email: jnestor salsa.iit.edu TOPIC: CAD of Integrated Circuits and Systems PREFERRED FORMAT: Lecture Abstract This paper describes an improved transformational approach to the scheduling problem in high level synthesis. Based on an existing approach called SALSA [NES90], it uses more powerful transformations and lower bounds on scheduling cost to quickly find highquality schedules of data oriented Control Data Flow Graphs. Results show the ability to find very good schedules for difficult scheduling problems in very small amounts of CPU time. In addition, ....

....in reasonable execution times. Several approaches to scheduling have been proposed, including simple listscheduling heuristics [PAN87,GOO89,THO90] Force directed scheduling (FDS) PAU89] integer linear programming (ILP) HWA91,GEB91] and iterative approaches such as simulated annealing (SA) [DEV89, NES90] and simulated evolution (SE) LY90] Simple list scheduling heuristics are very fast but often fail to find the best schedule. More powerful approaches such as FDS give much better results at the expense of an execution time that grows as the square of both number of nodes and schedule length ....

[Article contains additional citation context not shown here]

J. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints", Proceedings ICCAD-90, pp. 262-265, Nov. 1990.

....of node to the CDFG called the slack node. Slack nodes explicitly represent slack in the data flow, control flow, and timing constraints implied by CDFG edges. Figure 2(a) shows a CDFG with added slack nodes, each marked with an S . This explicit representation of slack can be used in scheduling [16] used to rapidly evaluate whether an operator node may be rescheduled without violating data or timing constraints. However, slack nodes on data edges also prove useful in allocation. V5 V6.1 V7 V8 V9 V10 X1 X2 X3 X4 X5 X6 X3 X1 X2 X4 X5 X6 V2 V3 V5 V7 V8 V9 V10 (a) SALSA CDFG (b) FU Allocation ....

....as possible. Then, another multiplexer is selected and merged with other compatible multiplexers. This process continues until merging has been attempted with all multiplexers. 5. Results The SALSA binding model and allocator have been implemented in C and work with the existing SALSA scheduler [16]. This implementation has been tested with a number of examples. Table 2 summarizes the results of an experiment allocating the Elliptic Wave Filter (EWF) benchmark [2,17] under a wide variety of conditions. Assumptions about operator delay are equivalent to those in [2] adders require one ....

J. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints", Proceedings ICCAD-90, pp. 262-265, Nov. 1990.

....graph (CDFG) representation that provides a partial ordering of operations in the original behavioral description. During synthesis, this partial ordering provides great flexibility in scheduling operations that can be used to reduce the cost of the implementation. Many scheduling algorithms (e.g. [PAR86, PAN87, PAU89, NES90]) exploit this freedom to reduce datapath cost. However, only limited efforts have been made to use scheduling freedom to reduce controller cost, especially when the controller is implemented as a single finite state machine (FSM) One reason for this is that few synthesis tools use controller ....

....can be solved using constraint solution techniques that start with all vertices scheduled in step 0 and iteratively correct individual constraint 5 violations by moving vertices into later control steps until all constraints are satisfied. This approach has been applied in high level synthesis [KU90,NES90] and results in an as soon as possible (ASAP) schedule that meets all timing and ordering constraints. An as late as possible (ALAP) schedule can be found in a similar way placing all nodes at the end of the schedule and moving nodes to previous control steps until all constraints are ....

[Article contains additional citation context not shown here]

J. A. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints", Proceedings ICCAD-90, pp. 262-265, Nov. 1990.

No context found.

J. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Trans. CAD , 12(8):1107-1122, August 1993.

No context found.

J. Nestor and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Trans. CAD,Vol.#2, No. 8, #993.

No context found.

Nestor 93 J. A. Nestor, and G. Krishnamoorthy, "SALSA: A New Approach to Scheduling with Timing Constraints," IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, Vol. 12, No. 8, pp. 1107-1122, Aug. 1993.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC