From chk@erato.cs.rice.edu Mon Mar 16 13:07:08 1992 Received: from erato.cs.rice.edu by cs.rice.edu (AA13215); Mon, 16 Mar 92 13:07:08 CST Received: from localhost.cs.rice.edu by erato.cs.rice.edu (AA13047); Mon, 16 Mar 92 13:07:06 CST Message-Id: <9203161907.AA13047@erato.cs.rice.edu> To: hpff-subroutines@erato.cs.rice.edu Word-Of-The-Day: roustabout : (n) 1. a dock worker 2. a laborer in an oil field or refinery 3. a circus worker with a variety of jobs Subject: Welcome to Working Group 3! Date: Mon, 16 Mar 92 13:07:06 -0600 From: chk@erato.cs.rice.edu If you are receiving this message, then you are on the mailing list for working group 3. This group is charged with discussing subroutine interfaces and automatic redistribution High Performance Fortran. Chuck From chen-marina@cs.yale.edu Fri Mar 20 13:40:50 1992 Received: from SUNNY.SYSTEMSX.CS.YALE.EDU (SYSTEMSX-GW.CS.YALE.EDU) by cs.rice.edu (AA05710); Fri, 20 Mar 92 13:40:50 CST Received: by SUNNY.SYSTEMSX.CS.YALE.EDU; Fri, 20 Mar 92 14:40:48 EST Date: Fri, 20 Mar 92 14:40:48 EST From: Marina Chen Message-Id: <9203201940.AA22307@SUNNY.SYSTEMSX.CS.YALE.EDU> Received: by sunny (crystal) via WIMP-MAIL (Version 1.3/1.7) ; Fri Mar 20 14:40:46 To: hpff-subroutines@cs.rice.edu Subject: working group 3 Dear Members of Working Group 3, Below is the list of issues raised at our group meeting. If you have additional items you want to add, please send them to me. I'd also like to collect examples that help to illustrate these issues. If you feel strongly about how we should do certain things certain way, please use one or more examples to illustrate your approach. Please send the examples directly to me by April 6 if possible. I'll organize them into groups and broadcast to everyone. Hope that can get us started. --Marina Chen ================================================================== Issues of Working Group 3: (1) SUBROUTINE INTERFACES * array sections * automatic distribution upon entry and exit, specification/conversion by caller or callee * (sequence association) * inheritance of distribution * distribution of local array variables w.r.t. formal arguments (2) SCOPE OF DIRECTIVES * lexical -- block structured * subroutine boundary * executable statements * For those arrays that are aligned, what happens when the computation goes across the (procedure or block) boundary ? (3) CLASSES OF DISTRIBUTION FUNCTIONS * parameterized (e.g. if P > 16 then D1 else D2) where P is the num. of procs. and D1 and D2 are distributions * data-dependent distribution (4) DIRECTIVES SHOULD BE SEMANTIC-PRESERVING * e.g. sequential, shared-memory, distributed-memory implementations all have the same behavior. In what way the directives can be safely ignored? * automatic distribution upon entry/exit (5) IS DISTRIBUTION QUERY NECESSARY? * restriction on redistribution: returned value can only be other directives. * Is this enough * effect on block structure (6) IMPACT ON LIBRARY From halstead@crl.dec.com Wed Apr 15 14:23:45 1992 Received: from crl.dec.com by cs.rice.edu (AA10114); Wed, 15 Apr 92 14:23:45 CDT Received: by crl.dec.com; id AA17240; Wed, 15 Apr 92 15:23:46 -0400 Received: by easynet.crl.dec.com; id AA08050; Wed, 15 Apr 92 15:22:01 -0400 Message-Id: <9204151923.AA18433@seine.crl.dec.com> To: hpff-subroutines@cs.rice.edu Cc: halstead@crl.dec.com Subject: is anybody home? Date: Wed, 15 Apr 92 15:23:43 -0400 From: halstead@crl.dec.com X-Mts: smtp I have not received any messages on this list yet, though I've seen a fair amount on hpff-distribute that overlaps with what I presume is this group's mandate. Has this group effectively been merged into hpff-distribute, or has this group just been quiet because its results aren't to be discussed until the June HPFF meeting? -Bert Halstead From gls@think.com Thu May 14 15:07:14 1992 Received: from mail.think.com by cs.rice.edu (AA29796); Thu, 14 May 92 15:07:14 CDT Return-Path: Received: from Strident.Think.COM by mail.think.com; Thu, 14 May 92 16:06:48 -0400 From: Guy Steele Received: by strident.think.com (4.1/Think-1.0C) id AA11536; Thu, 14 May 92 16:07:12 EDT Date: Thu, 14 May 92 16:07:12 EDT Message-Id: <9205142007.AA11536@strident.think.com> To: hpff-distribute@cs.rice.edu, hpff-subroutines@cs.rice.edu Cc: gls@think.com Subject: Groups 2 and 3 merge Per our discussion at the April meeting, HPFF working groups 2 and 3 (disrtibute and subroutines) are being merged. Chuck will merge the E-mail distribution list for hpff-subroutines into hpff-distribute sometime next week. Meanwhile, I apologize that some of you will receive two copies of this message. One of the most important questions to be resolved about the subroutine interface is whether a subroutine call can result in a remapping of an array as seen by the caller. Just to get the discussion going, I will take the position that HPF should allow a subroutine to effect remappings visible to the caller if and only if HPF includes explicit REALIGNABLE and REDISTRIBUTABLE attributes. For if such attributes are in the language, then failure to use them (in an interface block, for example) provides a guarantee to the compiler that a given routine will not remap its arguments. Similar remarks apply to arrays in COMMON. Comments? Counterproposals? --Guy