High Performance FORTRAN

How NUMA is handled by an automatic parallelizing compiler that hides all complexity, by a data-driven model in which distribution of data between processors is made explicit but all data are referenced with the same language constructs (e.g.. High Performance FORTRAN), or by a subroutine interface to access distributed or remote data (e.g., Linda tuples) or it may be that no direct access is provided to remote data (e.g., message passing). 

Rather than review the extensive literature of parallel programming environments, we examine a few that are of most utility in computational chemistry or are in common use, or represent current directions of research. 

Automatic parallelizing compilers have been a goal of some computer science researchers for many years (see, e.g., Refs. 51-56). Many others regard this as an impossible ambition, and the assertion by some vendors that dusty deck sequential FORTRAN codes may run efficiently on their MPP though the magic of their compiler should be regarded with skepticism. 

The proposed standard may be obtained using anonymous file transfer from titan.cs.rice.edu (in public/HPFF/draft). An effort commenced January 1994 to extend the HPF proposal to accommodate irregular data distributions and to support parallel 1/O and more complex data structures, as well as task-driven MIMD computation. Additional information is available by sending electronic mail to hpff-info cs.rice.edu. 

The High Performance FORTRAN draft standard acknowledges the limitations of current compiler technology by restricting itself (with the exception of extrinsic and PURE functions) to a data-parallel model issues of data place- 

---

HPF High Performance FORTRAN a language specification defined by industry and research community leaders as an extension to FORTRAN-90. HPF allows for explicit data-parallel distribution of data structures. 

Apart from MPl and PVM, there are programming models that attempt to hide the disttibuted nature of the machine at hand fi om the user. One of these is HPF (High Performance Fortran). In this Fortran 90-like language one can specify how the data are spread out over the processors, and all communication that results fl om this is taken care of by the HPF compiler and run time system. HPF, however, can be applied only in relatively easy and very regular cases. Such applications are characterized as data parallel. Therefore, the use of HPF is fairly limited. 

Ab initio calculations for large organometallic and other compounds. FORTRAN programs designed for high performance vector and parallel processing machines. 

Tel. 33-88-41-61-42, fax 33-88-61-20-85, e-mail rohmer frccsc21.bitnet Ab initio calculations for large organometallic and other compounds. FORTRAN programs designed for high-performance vector and parallel processing machines. 

The VMC and DMC methods are implemented in the casino code [44] (version 3.0), with which we intend to carry out this project. CASINO is written in Fortran 95 and parallelised using MPI with a master/slave program model, casino has been in existence for more than thirteen years and has been used on a wide variety of high-performance computer platforms. There are 360 registered users of the code. CASINO requires only the MPI library. 

The core of the current version of Candid is implemented in standard Fortran-77 and has been built upon the data structures and into the framework of the user interface of the program Dyana. The standard schedule and parameters for a complete automated structure determination with Candid and Dyana are specified in a script written in the interpreted command language Inclan that gives the user high flexibility in the way automated structure determination is performed without the need to modify the compiled core part of Candid [26]. 

The compiler is the primary interface to the processor. High level languages like Fortran and C are translated into the level instructions that can be executed on the processor. Compiler optimization and instruction generation decisions can be influenced with compiler switches. These switches may correspond directly to individual modules for optimization, or influence settings of several modules. The compiler performs many optimizations on the source code based on various assumptions about the program. Typical assumptions are ... 

---