--- Begin Message ---
- To: Camm Maguire <camm@enhanced.com>
- Subject: Re: LAM: Re: Bug#51871: lam: does not compile for sparc
- From: Jeff Squyres <jsquyres@lsc.nd.edu>
- Date: Mon, 27 Dec 1999 12:31:54 -0500 (EST)
- Message-id: <Pine.LNX.4.21.9912271201520.1783-100000@pokey.lsc.nd.edu>
On 27 Dec 1999, Camm Maguire wrote:
> > I'm looking around for references to xdr_quadruple (the problem function
> > in question -- we call it in lib/otb/t/etc/ieee.c, and apparently it
> > doesn't exist on some architectures), and finding very little other than
> > "quadruple" being specified in RFC 1832 (XDR: External Data Representation
> > Standard).
>
> > Question 1: I do see it in libnsl.a under Solaris 2.6. Is Solaris the
> > only OS with it? Is it a Solaris-ism, or has just no one else implemented
> > it yet?
>
> I haven't read the standard myself, so I'm not fully qualified to
> answer. We could forward this to the libc devel people if we'd like
> to pursue this question. One thing for sure is that many other xdr
That might be a good thing -- I only really briefly scanned the XDR
docs; I couldn't find any reference to xdr_quadruple anywhere except on
Solaris, so it may or may not be standard. Asking someone who would know
would be a Good Thing. :-)
> functions exist both on Linux glibc6 systems and Solaris. I have
> access to a Solaris box, and therefore also to the relevant man pages
> which describe the xdr structure and functions. But it seems as
> though you do too, so forwarding these to you may not be of any help.
Our primary development platform is Solaris. We port from there.
> I also have the Debian sparc man pages available, if these would be of
> assistance.
They're likely to be similar, if not the same, to the Solaris pages. But
if they've got more/better info, feel free to send them along.
> > Question 2: Is the assumption that on big-endian systems (specifically,
> > Debian running on a Sparc) that xdr_quadruple can be effectively a no-op
> > (i.e., copy src to destination)?
>
> This is probaby essentially correct, although the xdr structure has a
> number of auxiliary data elements that most likely should be filled
> in. The structure is defined in the rpc man page. Should I forward
> it to you?
No, I've got that. That's actually what I meant -- fill in the XDR data
structures as appropriate. I seem to recall that we could do some kind of
trick to get the XDR functions to fill in the data themselves (so that we
don't have to muck around in the underlying XDR -- kind of hazardous), and
then copy the actual data in manually.
However, I do notice that we don't ever seem to call this particular
function in question right now. This may or may not be a bug -- I'll have
to investigate this, because I seem to recall that we *do* support
MPI_LONG_DOUBLE -- but I don't see the harm in #if 0'ing out the following
subroutines in lib/otb/t/etc/ieee.c:
ltoieeeld
mieeetolld
mltoieeeld
Indeed, I have #if 0'ed out these routines under Solaris 2.6 (where
sizeof(long double) == 16), and LAM seems to still compile fine -- I can
compile/run a simple "hello world" program with no problems. I'll have to
check into support of MPI_LONG_DOUBLE, though.
So that seems to be a short term solution. If we can answer the questions
about xdr_quadratic properly, we can get the Right long term fix in the
LAM source code.
{+} Jeff Squyres
{+} squyres@cse.nd.edu
{+} Perpetual Obsessive Notre Dame Student Craving Utter Madness
{+} "I came to ND for 4 years and ended up staying for a decade"
--
Camm Maguire camm@enhanced.com
==========================================================================
"The earth is but one country, and mankind its citizens." -- Baha'u'llah
--- End Message ---