M. Philippsen and W. Tichy. Compiling for massively par- allel machines. In First International Conference of the Austrian Center for Parallel Computation, Salzburg, Austria, September 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....eliminate the need for advanced compile time analysis and optimization. 6 Related Work The Fortran D compiler is a second generation distributedmemory compiler that incorporates and extends features from previous compilation systems [6, 12, 19, 21, 25] Compared with other contemporary systems [2, 3, 7, 8, 18, 24], The Fortran D compiler is less flexible but performs deeper compile time analysis, many more advanced optimizations, requires fewer language extensions, and relies on less runtime support. Few researchers have published experimental results for large programs. Pingali Rogers apply message ....

M. Philippsen and W. Tichy. Compiling for massively par- allel machines. In First International Conference of the Austrian Center for Parallel Computation, Salzburg, Austria, September 1991.

....passing or shared memory type. There have been several approaches, which tried to support a pure PRAM programming style on networks, e.g. in [Sei90] or in the language pm2 [Juv92] A general purpose data parallelism language with a logically shared data space called Modula 2 was proposed in [PT91] Another idea is BSP Occam [All91] which supplements occam 2 with a library of procedures for bulksynchronous parallel computers suggested in [Val89] This concept does not integrate the shared memory features into occam 2, additionally it prohibits the standard use of channel communication ....

M. Philippsen and W.F. Tichy. Compiling for massively parallel machines. In Proc. Workshop on Code Generation, Dagstuhl Castle, 20-24 May 1991, WICS. Springer Verlag, 1991.

....eliminate the need for advanced compile time analysis and optimization. 6 Related Work The Fortran D compiler is a second generation distributed memory compiler that incorporates and extends features from previous compilation systems [8, 9, 20, 21, 10] Compared with other contemporary systems [22, 23, 24, 25, 26], it is less flexible but performs deeper compile time analysis, many more advanced optimizations, requires fewer language extensions, and relies on less run time support. The Fortran D compiler computes information to support the D Editor, a interactive parallel programming tool for presenting ....

M. Philippsen and W. Tichy. Compiling for massively parallel machines. In First International Conference of the Austrian Center for Parallel Computation, Salzburg, Austria, September 1991.

....short, this programmingmodel is closely related to the popular PRAM and BSP models but offers, in our view, a good compromise, as it is closer to real parallel machines than PRAMs and easier to program than BSP. The PRAM programming model, as supported e.g. by Fork95 [15] ll [19] and Modula 2 [21], offers deterministic write conflict resolution and operator level synchronous execution: there are no race conditions at all, data dependencies need not be protected by locks or barriers. Unfortunately, this ideal parallel programming model leads to very inefficient code when compiled to ....

M. Philippsen, W. F. Tichy. Compiling for Massively Parallel Machines. In Code Generation -- Concepts, Tools, Techniques, pp. 92--111. Springer Workshops in Computing, 1991.

....software to manage the conceptual control dynamism. In order to program such an abstract machine one should use the technique of MSPMD (Multiple SPMD) This means that the problem may be expressed with arbitrary control internesting, such as in NESL [Ble92, BCH 93] or Modula 2 [TH, Phi, TPHL, PT92] The abstract machine s memory area is fc, where fc means flexibly coupled. This means that the machine s memory organisation may vary between being totally tightly coupled and being totally loosely coupled, including combinations of the two. Chapter 4. CLUMPS Definition 62 Specialised ....

Michael Philippsen and Walter F. Tichy. Compiling for massively parallel machines. In Proceedings of the Workshops on Code Generation--- Concepts, Tools, Techniques, pages 92--111. Springer Verlag, 1992.

....on lists are converted to vector instructions (by flattening) and executed on SIMD machines. In contrast, the MIMD oriented Fork95 also allows for asynchronous and task parallelism, lowlevel PRAM programming and direct access to shared memory locations. Dataparallel variants of Modula [9, 16] support a subset of Fork95 s functionality. The main constructs to express parallelism are synchronous and asynchronous parallel loops. However no strict synchronicity is supported, and there is no group concept. 9] compiles to a PRAM simulator while [16] offers back ends for several existing ....

.... Dataparallel variants of Modula [9, 16] support a subset of Fork95 s functionality. The main constructs to express parallelism are synchronous and asynchronous parallel loops. However no strict synchronicity is supported, and there is no group concept. 9] compiles to a PRAM simulator while [16] offers back ends for several existing machines. ll [14] a similar approach, uses Pascal as base language. Further dataparallel languages in this tradition (see e.g. 17] are C , Dataparallel C [6] and dataparallel Fortran dialects such as HPF. The latter ones are mainly targeted towards ....

M. Philippsen and W.F. Tichy. Compiling for Massively Parallel Machines. In Code Generation -- Concepts, Tools, Techniques, pp 92--111. Springer, 1991.

....between control synchronous and asynchronous code regions, automatic group splitting for unstable branches, more flexible explicit group splitting, shared local variables, group IDs, or the join statement. PRAM languages are those MIMD languages which support exact synchronicity: Modula 2 [31], FORK [19] Fork95 [24] and ll [29] Although there are still some syntactical similarities between ForkLight and Fork95, there are fundamental differences in semantics: The PRAM programming model, as supported by Fork95, allows the programmer to exploit deterministic write conflict resolution ....

M. Philippsen and W. F. Tichy. Compiling for Massively Parallel Machines. In R. Giegerich and S. Graham, editors, Code Generation -- Concepts, Tools, Techniques, pages 92--111. Springer Workshops in Computing Series, 1991.

....passing or shared memory type. There have been several approaches which tried to support a pure PRAM programming style on networks, e.g. in [Sei90] or in the language pm2 [Juv92] A general purpose data parallelism language with a logically shared data space called Modula 2 was proposed in [PT91] Another idea is BSP Occam [All91] which supplements occam 2 with a library of procedures for bulk synchronous parallel computers suggested in [Val89] This concept does not integrate the shared memory features into occam 2, additionally it prohibits the standard use of channel communication ....

M. Philippsen and W.F. Tichy. Compiling for massively parallel machines. In Proc. Workshop on Code Generation, Dagstuhl Castle, 20-24 May 1991, WICS. Springer Verlag, 1991.

....list. Elementwise operations on lists are converted to vector instructions for execution on SIMD machines. In contrast, the MIMD oriented Fork95 also allows for asynchronous and task parallelism, low level PRAM programming and direct shared memory access. Dataparallel variants of Modula, e.g. [15], support a subset of Fork95 s functionality. The main parallel constructs are synchronous and asynchronous parallel loops; there is no group concept. Other PRAM oriented dataparallel languages are Dataparallel C and C [7] 3 PAD: A library of basic PRAM algorithms PAD is an attempt to ....

M. Philippsen and W. F. Tichy. Compiling for Massively Parallel Machines. In Code Generation -- Concepts, Tools, Techniques, pages 92--111. Springer, 1991.

....eliminate the need for advanced compile time analysis and optimization. 6 Related Work The Fortran D compiler is a second generation distributedmemory compiler that incorporates and extends features from previous compilation systems [6, 12, 19, 21, 25] Compared with other contemporary systems [2, 3, 7, 8, 18, 24], The Fortran D compiler is less flexible but performs deeper compile time analysis, many more advanced optimizations, requires fewer language extensions, and relies on less runtime support. Few researchers have published experimental results for large programs. Pingali Rogers apply message ....

M. Philippsen and W. Tichy. Compiling for massively parallel machines. In First International Conference of the Austrian Center for Parallel Computation, Salzburg, Austria, September 1991.

....forall, plus perhaps its asynchronous form. For data structures, one needs to introduce true multidimensional arrays plus pragmas that specify how to lay out the data. Such extensions have been implemented successfully in a compiler for the language Modula 2, targeting the Connection Machine[4, 3]. A simple and consistent extension of Modula 2 avoids all the problems mentioned above, without loss of efficiency. One might, however, call this work an unconfirmed experiment in language design and compiler construction. But this is exactly our point: More time is needed before we can ....

Michael Philippsen and Walter F. Tichy. Compiling for massively parallel machines. In Proc. of the Workshop on Code Generation, Schloss Dagstuhl. Springer Verlag, May 20-24 1991. to appear.

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