John Merlin, Bryan Carpenter, and Tony Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, March 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....Compiler Runtime Consortium (PCRC) 19] project. It was a high level runtime library designed to support translation of data parallel languages [16] Initial emphasis was on High Performance Fortran (HPF) and two experimental HPF translators used the library to manage their communications [38, 58]. It incorporated a built in representation of a distributed array, and a library of communication and arithmetic operations acting on these arrays. The array model supported general HPF like distribution formats, and arbitrary regular sections. Initially HPJava used a JNI wrapper interface to ....

John Merlin, Bryan Carpenter, and Tony Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, March 1996.

....The front end, transformation system and runtime have been integrated and demonstrated in the Syracuse HPF compilation system [83] As HPF has evolved to version 2.0, we have extended our HPF front end to cover HPF 2.0. The Syracuse common runtime also supports an older subset HPF system [61], has a user level C interface, and is being used as the basis of a new HPJava translator. Rochester [31, 34] and Indiana [13, 14, 17, 16] are also developing Java compilers. Rice is working on Java compiler optimizations [19] 2.9 Java and SPMD programming A signi cant new task in the revised ....

....schedules Message etc gather scatter Figure 1: NPAC PCRC runtime architecture 3 Syracuse University Report NPAC Kernel Runtime The kernel of NPAC library [28] is a C class library. It is primarily descended from the run time library of an earlier research implementation of HPF[61] with in uences from the Fortran 90D [30] run time and the CHAOS PARTI libraries. The kernel is currently implemented on top of MPI. The library design is solidly objectoriented, but eciency is maintained as a primary goal. Inlining is used extensively, and dynamic memory allocation, unnecessary ....

John Merlin, Bryan Carpenter, and Tony Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2-6, March 1996.

....construction intrinsics: Pi PACK, Pi UNPACK, 2 the intrinsic SUBROUTINEs: Pi RANDOM SEED, Pi RANDOM NUMBER Pi DATE AND TIME, Pi MVBITS, There are also restrictions associated with MINLOC and MAXLOC, MATMUL, SPREAD, RESHAPE and TRANSPOSE. The paper which accompanies the software, [13], is rather optimistic, yet at the same time truthful about the language coverage: shpf accepts the official Subset HPF language . with a few restrictions. It recognises full HPF (including full Fortran 90) and gives error messages upon encountering unsupported features. The Southampton ....

B. Carpenter J. Merlin and T. Hey. shpf: A Subset High Performance Fortran compilation system. Technical report, Department of Electronics and Computer Science, University of Southampton, Nov 1995.

....some are briefly discussed. The PARADIGM compiler [1] is a source to source parallelizing compiler based upon Parafrase 2 which accepts a sequential Fortran 77 or High Performance Fortran (HPF) program and produces an optimized message passing Fortran 77 parallel program. The sHPF system [19] transforms a Subset HPF program into a message passing Fortran 90 program with calls to the ADLIB runtime library [10] Parallel execution in sHPF is obtained from the use of Fortran 90 array syntax. ADAPTOR [8] is a source to source transformation system for HPF that generates Fortran SPMD ....

J. H. Merlin, D. B. Carpenter and A. J. G. Hey. SHPF: a Subset High Performance Fortran compilation system. Fortran Journal, pages 2-6, March/April 1996.

....The front end, transformation system and runtime have been integrated and demonstrated in the Syracuse HPF compilation system [83] As HPF has evolved to version 2.0, we have extended our HPF front end to cover HPF 2.0. The Syracuse common runtime also supports an older subset HPF system [61], has a user level C interface, and is being used as the basis of a new HPJava translator. Rochester [31, 34] and Indiana [13, 14, 17, 16] are also developing Java compilers. Rice is working on Java compiler optimizations [19] 2.9 Java and SPMD programming A significant new task in the ....

.... shift , schedules Message etc gather scatter Figure 1: NPAC PCRC runtime architecture 3 Syracuse University Report NPAC Kernel Runtime The kernel of NPAC library [28] is a C class library. It is primarily descended from the run time library of an earlier research implementation of HPF[61] with influences from the Fortran 90D [30] run time and the CHAOS PARTI libraries. The kernel is currently implemented on top of MPI. The library design is solidly objectoriented, but efficiency is maintained as a primary goal. Inlining is used extensively, and dynamic memory allocation, ....

John Merlin, Bryan Carpenter, and Tony Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, March 1996.

....added or subtracted to locations, yielding new, shifted, locations. This kind of shifted access only works if the subscripted array has suitable ghost extensions. 15 3.2 Adlib. The Adlib runtime library was initially designed to support HPF translation. Early development took place in the shpf [31] project at Southampton, UK. Subsequently the library was redesigned and reimplemented at Syracuse during in the PCRC project, and delivered as the NPAC PCRC runtime kernel [11] It has been used as a foundation of two experimental HPF compilation systems [31, 39] one in Europe and one at ....

....development took place in the shpf [31] project at Southampton, UK. Subsequently the library was redesigned and reimplemented at Syracuse during in the PCRC project, and delivered as the NPAC PCRC runtime kernel [11] It has been used as a foundation of two experimental HPF compilation systems [31, 39], one in Europe and one at Syracuse) and is currently being used as a basis of the HPJava translator. The Adlib kernel is C class library, built on MPI. Fortran, C and Java interfaces are available or under development. It provides communication primitives similar to Multiblock PARTI, plus ....

John Merlin, Bryan Carpenter, and Tony Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2-6, March 1996. 19

....array descriptors and collective operations needed for straightforward, ecient translation of parallel constructs like FORALL and array assignments. This goal was achieved quite successfully, and two experimental subset HPF translators have used the library to manage their communications [13, 7]. Unfortunately it is evident that implementing compilers for a language as complex as full HPF is a formidable task. On the other hand the runtimes that have evolved to support HPF and its kin are powerful and have an underlying elegance, using them without a compiler through direct calls from ....

....data parallel SPMD programming models can achieve a level of elegance and expressivity comparable to full HPF. 2 Background: runtime kernel The kernel of NPAC library is a C class library. It is most directly descended from the run time library of an earlier research implementation of HPF [7] with in uences from the Fortran 90D run time and the CHAOS PARTI libraries [1, 11, 5] The kernel is currently implemented on top of MPI. The library design is solidly object oriented, but eciency is maintained as a primary goal. The overall architecture of the library is illustrated in gure 1. ....

J. Merlin, B. Carpenter, and T. Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2-6, Mar. 1996.

....of DADs are major functions of PCRC runtime. For a thorough discussion of the DAD design, the reader is referred to [8] or Chapter 6 of [6] 3.2 Runtime kernel The kernel of NPAC library is a C class library. It is descended from the run time library of an earlier research implementation of HPF[5] with in uences from the Fortran 90D run time and the CHAOS PARTI libraries. The kernel is currently implemented on top of MPI. The library design is solidly object oriented, but eciency is maintained as a primary goal. Inlining is used extensively, and dynamic memory allocation, unnecessary ....

John Merlin, Bryan Carpenter and Tony Hey, \shpf: a Subset High Performance Fortran compilation system," Fortran Journal, pp 2-6, March 1996.

....on each processor. Here we have built a simple interface from KeLP to HPF. This allows KeLP to invoke multiple data parallel HPF tasks concurrently, each operating on an individual distributed grid. Our implementation of the interface works in conjunction with the public domain HPF compiler SHPF [8]. There are good reasons for using the KeLP HPF approach for implementing multiple data parallelism , quite apart from the current difficulties of using pure HPF for this purpose. First, it combines the strengths of both languages: KeLP for concisely expressing block structured computations and ....

....is, the same instructions are executed on every processor, except for code enclosed in a for all end for all construct. Data layout and communication operations must be performed in the single threaded code. 1 Actually for all and end for all are macros. John Merlin et al. 2. 2 SHPF SHPF [8] is a public domain HPF 2.0 compilation system. It comprises a translator that translates HPF into Fortran 90 SPMD code that runs on each node of the target computer, and a runtime library, ADLIB, that is called by the Fortran 90 code to perform communications. ADLIB [10] like KeLP, is a C ....

Merlin, J. H., Carpenter, D. B., Hey, A. J. G.: SHPF: a Subset High Performance Fortran Compilation System, Fortran Journal (March / April 1996), 2--6.

....or assume any particular implementation model (for example, threads) One of the primary design goals was to allow wide portability of the library. Another one was to allow an implementation to exploit the most efficient mechanisms available on a given platform, which might involve active messages [11], native put get, shared memory, and or threads. For example, depending on whatever solution delivers the best performance, ARMCI accumulate might or might not be implemented using the owner computes rule. In particular, this rule is used on IBM SP where accumulate is executed inside the Active ....

....the Parallel Compiler Runtime Consortium project [14] project. It is a high level runtime library designed to support translation of data parallel languages [4] Initial emphasis was on High Performance Fortran, and two experimental HPF translators used the library to manage their communications [17, 11]. Currently the library is being used in the HPspmd project at NPAC [3] It incorporates a built in representation of a distributed array, and a library of communication and arithmetic operations acting on these arrays. The array model is more general than GA, supporting general HPF like ....

J. Merlin, B. Carpenter, and A. Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, March 1996. Lecture Notes in Computer Science 15

....Parallel Compiler Runtime Consortium project [14] project. It is a high level runtime library designed to support translation of data parallel languages [4] Initial emphasis was on High Performance Fortran (HPF) and two experimental HPF translators used the library to manage their communications [16, 10]. Currently the library is being used in the HPspmd project at NPAC [3] It incorporates a built in representation of a distributed array, and a library of communication and arithmetic operations acting on these arrays. The array model is more general than GA, supporting general HPF like ....

J. Merlin, B. Carpenter, and A. Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, March 1996.

....be added or subtracted to locations, yielding new, shifted, locations. This kind of shifted access only works if the subscripted array has suitable ghost extensions. 3.2 Adlib. The Adlib runtime library was initially designed to support HPF translation. Early development took place in the shpf [31] project at Southampton, UK. Subsequently the library was redesigned and reimplemented at Syracuse during in the PCRC project, and delivered as the NPAC PCRC runtime kernel [11] It has been used as a foundation of two experimental HPF compilation systems [31, 39] one in Europe and one at ....

....development took place in the shpf [31] project at Southampton, UK. Subsequently the library was redesigned and reimplemented at Syracuse during in the PCRC project, and delivered as the NPAC PCRC runtime kernel [11] It has been used as a foundation of two experimental HPF compilation systems [31, 39], one in Europe and one at Syracuse) and is currently being used as a basis of the HPJava translator. The Adlib kernel is C class library, built on MPI. Fortran, C and Java interfaces are available or under development. It provides communication primitives similar to Multiblock PARTI, plus ....

John Merlin, Bryan Carpenter, and Tony Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, March 1996.

....of DADs are major functions of PCRC runtime. For a thorough discussion of the DAD design, the reader is referred to [8] or Chapter 6 of [6] 3.2 Runtime kernel The kernel of NPAC library is a C class library. It is descended from the run time library of an earlier research implementation of HPF[5] with influences from the Fortran 90D run time and the CHAOS PARTI libraries. The kernel is currently implemented on top of MPI. The library design is solidly object oriented, but efficiency is maintained as a primary goal. Inlining is used extensively, and dynamic memory allocation, unnecessary ....

John Merlin, Bryan Carpenter and Tony Hey, "shpf: a Subset High Performance Fortran compilation system," Fortran Journal, pp 2-6, March 1996.

....on each processor. Here we have built a simple interface from KeLP to HPF. This allows KeLP to invoke multiple data parallel HPF tasks concurrently, each operating on an individual distributed grid. Our implementation of the interface works in conjunction with the public domain HPF compiler SHPF [13]. There are good reasons for using the KeLP HPF approach for implementing multiple data parallelism , quite apart from the current difficulties of using pure HPF for this purpose. First, it combines the strengths of both languages: KeLP for concisely expressing block structured computations and ....

....a for all end for all construct. Data layout and communication operations must be performed in the single threaded code. The KeLP source code and documentation can be obtained from http: www cse.ucsd.edu groups hpcl scg kelp.html. 5 Actually for all and end for all are macros. 2. 2 SHPF SHPF [13] is a public domain HPF 2.0 compilation system. It comprises a translator that translates HPF into Fortran 90 SPMD code that runs on each node of the target computer, and a runtime library, ADLIB, that is called by the Fortran 90 code to perform communications. ADLIB [16] like KeLP, is a C ....

J.H. Merlin, D.B. Carpenter and A.J.G. Hey, SHPF: a Subset High Performance Fortran Compilation System, Fortran Journal (March / April 1996) 2--6.

....array descriptors and collective operations needed for straightforward, efficient translation of parallel constructs like FORALL and array assignments. This goal was achieved quite successfully, and two experimental subset HPF translators have used the library to manage their communications [13, 7]. Unfortunately it is evident that implementing compilers for a language as complex as full HPF is a formidable task. On the other hand the runtimes that have evolved to support HPF and its kin are powerful and have an underlying elegance, using them without a compiler through direct calls from ....

....hybrid data parallel SPMD programming models can achieve a level of elegance and expressivity comparable to full HPF. 2 Background: runtime kernel The kernel of NPAC library is a C class library. It is most directly descended from the run time library of an earlier research implementation of HPF [7] with influences from the Fortran 90D run time and the CHAOS PARTI libraries [1, 11, 5] The kernel is currently implemented on top of MPI. The library design is solidly object oriented, but efficiency is maintained as a primary goal. The overall architecture of the library is illustrated in ....

J. Merlin, B. Carpenter, and T. Hey. shpf: a subset High Performance Fortran compilation system. Fortran Journal, pages 2--6, Mar. 1996.

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