W. B. Ackerman. Data flow languages. In N. Gehani and A. D. McGettrick, editors, Concurrent Programming, chapter 3, pages 163--181. Addison-Wesley, 1988.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... issue of software reusability [10] 2 The Paralex Programming Paradigm The choices made for programming paradigm and notation are fundamental in harnessing parallelism in a particular application domain [11] The programming paradigm supported by Paralex can be classified as static data flow [1]. A Paralex program is composed of nodes and links. Nodes correspond to computations (functions, procedures, programs) and the links indicate flow of (typed) data. Thus, Paralex programs can be thought of as directed graphs (and indeed are visualized as such on the screen) representing the data ....

W. B. Ackerman. Data Flow Languages. IEEE Computer, February 1982, pp. 15-22.

.... languages, including for example the computation graphs of Karp and Miller [24] the synchronous dataflow graphs of Lee and Messerschmitt [28] the cyclo static dataflow model of Lauwereins, et al. 26] 5] the Processing Graph Method (PGM) of Kaplan, et al. 23] Granular Lucid [19] and others [1][9] 12] 36] Many of these limit expressiveness in exchange for considerable advantages such as compile time predictability. Synchronous dataflow (SDF) and cyclo static dataflow both have the particularly useful property that a finite static schedule can always be found that will return the graph ....

W. B. Ackerman, "Data Flow Languages," Computer, Vol. 15, No. 2, February 1982.

.... of dataflow coordination languages, including Karp and Miller s computation graphs [34] Lee and Messerschmitt s synchronous dataflow graphs [35] Lauwereins et al. s cyclo static dataflow model [36] 37] Kaplan et al. s Processing Graph Method (PGM) 38] Granular Lucid [39] and others [40], 41] 42] 43] Many of these limit expressiveness in exchange for formal properties (e.g. provable liveness and bounded memory) Synchronous dataflow (SDF) and cyclo static dataflow require processes to consume and produce a fixed number of tokens for each firing. Both have the useful ....

W. B. Ackerman, "Data flow languages," Computer, vol. 15, no. 2, 1982.

.... can also be utilized in producing optimized parallel executables [3] Additionally, the compiler uses a multi language precompiler (Parafrase 2 [5] and a multi platform post compiler (Optimizing SISAL Compiler [4] in order to simplify the translation process and to provide multi platform support [1]. We are strong believers in the do not reinvent the wheel principle, so we decided to utilize existing tools as much as possible. For the front end we choose Parafrase 2 a source to source multilingual restructuring compiler that translates Fortran to Cedar (parallel) Fortran. We chose P2 ....

....machines like Sequent, SGI, HP, Cray, Convex and most UNIX based workstations [2] 3 Fine and Medium Grain Parallelism The problem when generating code on the basic (expression) level is the single assignment restriction. This is a concern when Fortran variables are translated into IF1 edges [1]. In Fortran a variable is a memory location that can have many values stored in it over time whereas IF1 edges carry single values. An IF1 edge is a single assignment. Our system handles this problem on the basic block level using a structure called Variable to Edge Translating Table (VETT) 3] ....

Ackerman, W. B., "Dataflow languages." In S. S. Thakkar, editor, Selected Reprints on Dataflow and Reduction Architectures, Computer Society Press, Washington D.C., 1987.

....run time architecture. Still, side effect free languages are likely to contain more implicit parallelism than imperative languages because they lack artificial time dependencies. Two examples of languages and architectures that support implicit parallelism are dataflow languages and architectures [1, 7] and multi processor graph reduction of pure functional programs [105] In both cases, the languages are side effect free. However, Gajski et al. point out that parallelising compilers can in fact perform better than single assignment languages [49] Implicit parallelism is also a key ingredient ....

....w (x0,x1) x0 t, x0 t) where t = w x1 2.2.3 Patterns A pattern is formed when a data constructor appears on the left hand side of a binding. The result of evaluating the right hand side is bound to the corresponding identifiers in the pattern. For example, the binding (x:xs) [1,2,3,4] will cause the integer 1 to be bound to x, and the list [2; 3; 4] to be bound to xs. Patterns can be nested arbitrarily, as in CHAPTER 2. BACKGROUND MATERIAL 23 ( x,y) z : zs) Arguments to function bindings are often patterns; in this case, they also serve to select one of possibly ....

[Article contains additional citation context not shown here]

William B. Ackerman. Data flow languages. IEEE Computer, pages 15--24, February 1982.

....run time architecture. Still, side effect free languages are likely to contain more implicit parallelism than imperative languages because they lack artificial timedependencies. Two examples of languages and architectures that support implicit parallelism are dataflow languages and architectures [1, 7] and multi processor graph reduction of pure functional programs [105] In both cases, the languages are side effect free. However, Gajski et al. point out that parallelising compilers can in fact perform better than single assignment languages [49] Implicit parallelism is also a key ingredient ....

....t) CHAPTER 2. BACKGROUND MATERIAL 23 where t = w x1 2.2.3 Patterns A pattern is formed when a data constructor appears on the left hand side of a binding. The result of evaluating the right hand side is bound to the corresponding identifiers in the pattern. For example, the binding (x:xs) [1,2,3,4] will cause the integer 1 to be bound to x, and the list [2; 3; 4] to be bound to xs. Patterns can be nested arbitrarily, as in ( x,y) z : zs) Arguments to function bindings are often patterns; in this case, they also serve to select one of possibly multiple clauses of a function ....

[Article contains additional citation context not shown here]

William B. Ackerman. Data flow languages. IEEE Computer, pages 15--24, February 1982.

....to hold the entire processor waiting for a particular data, leading to an adaptive overlap between communication and computation. 22] was probably one of the first to use data driven execution on commercial parallel computers, although the idea itself existed in the earlier data flow approach[2], and the Rediflow[24] project for parallel functional languages. More recently, handler based approaches such as active messages [29] have been implemented. An active message encodes the name of a handler a user defined function of one parameter. The message is sent to a remote processor ....

W. B. Ackerman. Data Flow Languages. IEEE Computer, 15(2):15--25, February 1982.

....blocks, a set of functions defining the DMM Interface and an example of using the DMM functions. The paper is summarized by some conclusions. 2. The Vesicular Dataflow Model Program execution in the basic dataflow (DF) model is controlled only by the availability of arguments. The DF graph [1] consists of nodes representing operations. Arcs connecting the nodes show data dependencies between the operations. Availability of data is indicated by presence of tokens on the arcs. The tokens convey data between the nodes along the arcs. The program execution is performed by firing of ....

W. B. Ackerman., "Data Flow Languages", Computer, vol. 15, no. 2, pp. 15-25, Feb. 1982.

....of the Ada naming system distinguishes it not only from CSP, but from all naming systems that allow only private names, regardless of how the private names property is realized. Examples of systems in which communication among objects is specified using private names include dataflow languages[2] (streams) parallel logic languages[61] single assignment variables) and point to point channels. The Ada naming system is further distinguished from that of CSP by the dynamic domain property. This property is required for the implementation of reactive systems, where events in the ....

....Since the copies of a component never synchronize and replication is symmetric, increasing the number of copies of a component may improve performance by reducing communication delays. 8.4 Definitions 8.4. 1 Partitionable Programs Partitionable programs are a generalization of dataflow programs [2] in that they consist of a set of components, each of which encapsulates code and data, and which communicate by asynchronously reading and writing streams of data values. Unlike dataflow programs, values written to an output stream may appear in more than one input stream, and input streams may ....

W. Ackerman, Dataflow languages, IEEE Computer, vol. 15, no. 2, Feb. 1982

....Environment) is a user friendly environment which provides a easy to use visual interface for design, task mapping, execution control, monitoring and debugging of graphically expressed parallel applications. The programming paradigm adopted by PARADE refers to the notion of coarse grain data flow[2], traduced here in a directed acyclic graph (DAG) where the nodes are either computation nodes or control nodes, and the arcs represent the flow of data and the synchronization between data flows. Each node executes a sequential program, activated by a simple procedure call performed by the node ....

W. Ackerman, "Data Flow Languages", IEEE Computer, pp. 15-22, February 1982.

....are free of side effects the evaluation order of their arguments is arbitrary, i.e. the order does not affect the result. For example in the expression f(g(1) g(2) the compiler can execute both invocations of g concurrently. Several dataflow languages are based on this principle, e.g. VAL [3] and Id90 [80] And Or Parallelism Logic programs consist of sequences of clauses as shown in the following example: A : B, C A : D, E Each of the clauses has a declarative meaning which results from the underlying Horn logic. In the example, the declarative meaning of the first clause is ....

W. B. Ackerman. Data flow languages. Computer, 15(2):15--25, February 1982.

.... coordination languages, including Karp and Miller s computation graphs [KM66] Lee and Messerschmitt s synchronous dataflow graphs [LM87] Lauwereins et al. s cyclo static dataflow model [LWAP94, BELP94] Kaplan et al. s Processing Graph Method (PGM) K 87] Granular Lucid [Jag92] and others [Ack82, CG89, CH71, SBKB90] Many of these limit expressiveness in exchange for for24 A C B D (a) A B C D (b) Figure 4: a) A dataflow process network (b) A single processor static schedule for it mal properties (e.g. provable liveness and bounded memory) Synchronous dataflow (SDF) and ....

W. B. Ackerman. Data flow languages. Computer, 15(2), 1982.

....composing synchronous programs requires a procedure checking well definedness. This procedure is in the general case NP complete but, thanks to particular features of each of the synchronous languages, it is in practice quite simple. For instance, Lustre and Signal are data flow languages [1] thus satisfying the single assignment principle. Hence, this procedure becomes purely syntactical as the union over disjoint sets of variables is closed. As performed by Esterel, another simplifying choice is to extend the composition operator by enabling the specification of combination ....

W. B. Ackerman. Data flow languages. IEEE Computer, 15(2):15--25, February 1982.

.... Most of these applications are written today in conventional procedural languages and there is no trend to write them in nonconventional ones (like logical, applicative or data flow languages) that might provide better opportunities for parallel execution, as is believed by some researchers [2, 3, 4]. Streams have been studied extensively, and code written with streams was shown to have the following good qualities: 1 ffl Designing and programming by means of flow of data: the stream enhances the transformation, even automatic one, from clear design as a set of individual modules connected ....

.... simulate a change of state without using assignments [8] hence it may be used in applicative and single assignment languages where multiple assignments are not allowed [7] Thus streams were introduced to many languages Prolog [6, 7] lisp [8, 10] functional programming [2] data flow languages [3, 4] APL and even in unix. There are even languages like Lucid [5] whose main data structure is the stream. The only experiment with introducing streams to a conventional language was done in [9] This paper shows that introducing streams in languages can achieve parallel executions of programs ....

W. Ackerman. Data Flow Languages. comp. IEEE, vol 15, feb. 1982. pp 15-25.

....the above graph is equivalent to the graph plus(u,v; g) plus(v,w; h) mult(g,h; c) An executing application is represented as a DAG in order to make explicit the dependencies between tasks. Previous dataflow models assume that explicit dependencies are incompatible with inouts [Architectures] [languages]. As demonstrated in this presentation, the dataflow model can support inouts. Among other benefits, inouts allow the modification of large data structures such as arrays. An example using inouts is given by the following graph. f( x; g( x; The task f( x; must execute before the task ....

....11 Input Output, Global Items and Other Such Effects Previous dataflow models prohibit most effects. a data flow computer imposes much stricter prohibitions against side effects a procedure may not even modify its own arguments. In fact, in a sense nothing may ever be modified at all. [languages] In previous dataflow models, the effect of a task is the production of a value. In contrast, TSIA places no restrictions on the effects of task, provided that all effects are declared. The dependencies between tasks thus are explicit and can be obeyed by a TSIA executing the application. Up ....

William B. Ackerman. "Data flow languages", at National Computer Conference, June 4-7, 1979, New York, USA. Editor Richard E. Merwin, AFIPS PRESS.

....from video coding, are given throughout the article to illustrate key points. 2 Synchronous Data Flow For applications which meet its strict constraints on periodicity, the Synchronous Data Flow graph offers a powerful model for static analysis and simplified implementation. Data Flow graphs [1] [7] are often used in signal processing to model real time applications because they offer an intuitive and visually oriented approach [6] SDF is a variant of the traditional Data Flow graph; the amount of data consumed and produced by each node for all its inputs and outputs, is fixed in ....

W.B. Ackerman. Data flow languages. Computer, 15:15--25, February 1982.

....an example, components for motion video image compression. 2 Synchronous Data Flow The Synchronous Data Flow (SDF) graph offers an elegant block level data flow representation for certain real time Digital Signal Processing (DSP) applications. SDF is a variant of the traditional data flow graph [1][8] where the amount of data consumed and produced by each node for all its inputs and outputs, is fixed in advance. Data flow graphs are often used in signal processing to model real time applications, because they offer an intuitive, visually oriented approach [7] An application expressed as ....

W.B. Ackerman. Data flow languages. Computer, 15:15--25, February 1982.

....works and introduces our proposed strategy. The detailed techniques will be explained in sections 4 and 5. Two real life examples are discussed in section 6. Section 7 wraps up the paper with conclusions and future works. 2. Synchronous Dataflow: Scheduling and Code Generation Dataflow model [12] has been used to describe algorithms as directed graphs where the nodes (actors) represent computations or function modules and the arcs represent data paths. Synchronous dataflow (SDF) is a special case of dataflow, where the number of data samples produced or consumed by each node on each ....

....be shared. The Ramp block in figure 3 is an example. 15 5. Adjusting Single Appearance Schedule 5. 1 Motivation A B C 3 7 5 6 1 = 2 ( 7A 3B ) 5C A B C 7 3 2 5 G [0,0,3] 7,0,5] 6,0,0] 30,0,0] 0,30,0] 21,0,15] 0,0,21] 0,21,30] A B C 7 3 1 2 2 [0,0,3] 7,0,5] 6,0,0] [12,0,0] [0,12,6] 21,0,15] 0,0,21] 0,21,15] C 1 G [6,0,0] 6,0,0] 0,6,0] 2 = 2 ( 7A 3B 2C) 1C buffer memroy(S 1 ) 51 buffer memroy(S 2 ) 45 (a) b) c) Figure 5. a) An example SDF graph, b) Schedule tree (s tree) data structure for the graph, and (c) s tree after schedule ....

[Article contains additional citation context not shown here]

W. B. Ackerman, "Data flow languages",Computer, vol. 15, no. 2, pp. 15-25, Feb. 1982.

....(VAR A, B: Vec) VAR i: INTEGER; BEGIN FOR i: 1 TO N DO B[i] A[N i 1] END; END reverse1; type Vec is array (1. N) of real; procedure reverse2 (A: in Vec; B: out Vec) is begin for i in A range loop B(i) A (A last i 1) end loop; end reverse2; Fig. 1. A procedure with potential aliasing [Ack82] in Modula 2 and Ada A: T; procedure P(f: in out T) is begin A: Const2; end P; begin A: Const1; P(A) Fig. 2. Aliasing between a non local variable and a formal parameter variable A after calling P is Const2 if T is passed by reference, and Const1 in case of passing by copy in copy out. ....

.... copying out unspecified. In the following, aliasing will include these cases. 3 Variables Without Addresses It has been stated that the ability to change a variable s value is the principal problem in procedural languages and that one should have so called single assignment variables instead [Ack82,KMW84,Ske91,Can92] Such variables can be bound to a value but that binding then remains immutable during program execution, thus preventing any kind of mutable state. This approach is often considered overly restrictive [Hoa73,KMW84,Yue91,Fin95] In contrast, our approach is based on the idea of ....

W.B. Ackerman. Data Flow Languages. IEEE Computer, 15(2):14--25, February 1982.

No context found.

W. B. Ackerman. Data flow languages. In N. Gehani and A. D. McGettrick, editors, Concurrent Programming, chapter 3, pages 163--181. Addison-Wesley, 1988.

No context found.

Ackerman, W. B. Data Flow Languages. Computer (February 1982), 15--25.

No context found.

W. B. Ackerman, "Data Flow Languages," Computer , pp. 15--25, February 1982.

No context found.

ACK82 Ackerman, W. B., Data Flow Languages, IEEE Computer, Vol. 15, No. 2, February 1982.

No context found.

ACK82 Ackerman, W. B., Data Flow Languages, IEEE Computer, Vol. 15, No. 2, February 1982.

No context found.

W.B. Ackerman (1982). Data flow languages. IEEE Computer 15(2), 15-25.

First 50 documents

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