Welcome to the High Performance Fortran Forum! WHAT IS HPFF? ============= Since its introduction over three decades ago, Fortran has been the language of choice for scientific programming for sequential computers. Exploiting the full capability of modern architectures, however, increasingly requires more information than ordinary Fortran 77 or Fortran 90 programs provide. This information applies to such areas as 1. Opportunities for parallel execution 2. Type of available parallelism - MIMD, SIMD, or some combination 3. Allocation of data among individual processor memories 4. Placement of data within a single processor The High Performance Fortran Forum is a coalition of industrial and academic groups working to identify the basic issues and suggest a set of standard extensions to Fortran to provide the necessary information. To date, the group includes most vendors currently delivering parallel machines, government labs, and many university research groups. Our intent is to develop extensions to Fortran which provide support for high performance programming on a wide variety of machines, including massively parallel SIMD and MIMD systems and vector processors. Some of the topics considered will be 1. Data distribution and alignment 2. Explicit parallelism, particularly for data-parallel problems 3. Computation distribution 4. I/O support The first draft of the High Performance Fortran language specification, which attempts to address these issues, is now available; see below for details of getting it. Anybody with an interest in using Fortran on modern architectures is welcome to participate in HPFF. To allow wide participation, all HPFF materials will be available electronically, and we welcome electronic comments (send mail to hpff-comments@rice.edu). WHERE CAN I GET THE HPF LANGUAGE SPECIFICATION? Electronic copies of HPF draft specification are available by anonymous FTP from the following sources: Machine name File name titan.cs.rice.edu public/HPFF/draft/hpf-v10-final.tar titan.cs.rice.edu public/HPFF/draft/hpf-v10-final.tar.Z titan.cs.rice.edu public/HPFF/draft/hpf-v10-final.ps titan.cs.rice.edu public/HPFF/draft/hpf-v10-final.ps.Z think.com public/HPFF/hpf-v10-final.ps.Z ftp.gmd.de hpf-europe/hpf-v10-final.ps.Z theory.tc.cornell.edu pub/hpf-v10-final.ps.Z minerva.npac.syr.edu public/hpf-v10-final.tar.Z minerva.npac.syr.edu public/hpf-v10-final.ps.Z File names ending in .ps are Postscript(TM) format. File names ending in .tar are tar files of LaTeX source. File names ending in .Z are compressed (UNIX compress(1)). A script for retrieving one of these files is given below (things you type are underscored with ^): % ftp titan.cs.rice.edu ^^^^^^^^^^^^^^^^^^^^^ Connected to cs.rice.edu. 220 titan FTP server (SunOS 4.1) ready. Name (titan.cs.rice.edu:chk): anonymous ^^^^^^^^^ 331 Guest login ok, send ident as password. Password: your name ^^^^^^^^^ 230 Guest login ok, access restrictions apply. ftp> cd public/HPFF/draft ^^^^^^^^^^^^^^^^^^^^ 250 CWD command successful. ftp> binary on ^^^^^^^^^ 200 Type set to I. ftp> get hpf-v10-final.ps.Z ^^^^^^^^^^^^^^^^^^^^^^ 200 PORT command successful. 150 Binary data connection for hpf-v10-final.ps.Z (128.42.1.69,1265) (322893 bytes). 226 Binary Transfer complete. local: hpf-v10-final.ps.Z remote: hpf-v10-final.ps.Z 322893 bytes received in 13 seconds (24 Kbytes/s) ftp> quit ^^^^ 221 Goodbye. Several of the FTP sites also have other documents related to HPFF. See the following files for more information: Machine name File name titan.cs.rice.edu public/HPFF/README ftp.gmd.de hpf-europe/README minerva.npac.syr.edu public/README In addition to anonymous FTP, the draft is also available by electronic mail from the following archive servers: Machine name First line of message softlib@cs.rice.edu send hpf-v10-final.ps softlib@cs.rice.edu send hpf-v10-final.tar.Z netlib@research.att.com send hpf-v10-final.ps from hpff netlib@ornl.gov send hpf-v10-final.ps from hpf To retrieve a file from an archive server, send a message with the given line in the message body (some servers will also recognize the request in the "Subject:" line.) The archive servers will send the requested file in several pieces to avoid various mailer limitations; concatenate these pieces to get the whole file. Finally, the HPF language specification is available in hardcopy format as technical report CRPC-TR 92225 from the Center for Research on Parallel Computation at Rice University. Send requests to Theresa Chapman CITI/CRPC, Box 1892 Rice University Houston, TX 77251 There is a charge of $50.00 for this report to cover copying and mailing costs. HOW DO I PRINT OUT THESE FILES? The HPF language specification is nearly 200 pages long. We strongly recommend that you check with other users and administrators on your system to make sure printing such a document will not inconvenience anyone. Compressed files must be processed before printing. On UNIX systems, the appropriate command is uncompress -v filename.Z This will create the file "filename" and delete "filename.Z". Tar files must be extracted before being used. On UNIX systems, the appropriate command is tar xvf hpf-v10-final.tar This will extract the contents of the archive file. In the case of the HPF language specification, it will create a directory named "hpf-report" in the current directory, and a collection of LaTeX files within that subdirectory. LaTeX files must be formatted before printing. On UNIX systems, the appropriate commands are latex hpf-report.tex latex hpf-report.tex Note that latex must be run twice to generate the correct cross-references in the document. This process generates several files, including hpf-report.dvi (which is the one to be printed). Consult your local system administrators for the correct command to print this file; some common commands are dvips hpf-report.dvi dvipr hpf-report.dvi The .dvi file is not a text file; DO NOT try to print (or read) it directly! Postscript files can be printed on Postscript printers, such as laser printers. Consult your local systems administrators for the correct command to use; a popular one is lpr hpf-report.ps Note that a Postscript file consists of commands represented as text; DO NOT print it as a text file! HOW CAN I PARTICIPATE IN HPFF? We highly encourage feedback on the current language specification. Please send any comments, either technical or editorial to hpff-comments@cs.rice.edu. To ease sorting of the messages, please send a separate message for comments related to each chapter, and identify the chapter in the "Subject:" line. If you cannot send comments electronically, then hardcopy comments can be sent to HPFF COMMENTS c/o Theresa Chatman CITI/CRPC, Box 1892 Rice University Houston, TX 77251 We will make our best efforts to respond to any questions received. Another way to participate is to join one or more of the HPFF mailing lists. The possible lists are hpff Main HPFF list (meeting minutes, pointers to language drafts, miscellaneous announcements) hpff-bench Discussion of benchmarking for HPF hpff-io Discussion of parallel I/O for HPF hpff-interpret Interpretations of HPF draft hpff-irreg Discussion of irregular computations in HPF hpff-task Discussion of task parallelism in HPF All mailing lists are kept at cs.rice.edu. To add your mail address to one of the HPFF mailing lists, mail a message to hpff-request@cs.rice.edu with a subject line containing the word "add" and the name of the list. This will add the E-mail address of the message originator to the requested mailing list (if no list name is given, the address is added to "hpff"). For example, to receive the Fortran 90 discussions, send a message with "Subject: add hpff-f90". If you would like to attend future HPFF technical meetings, please contact Theresa Chatman (tlc@cs.rice.edu) for information about the "core" mailing list.