[parsec-users] vips and fluidanimate on Solaris/SPARC

Christian Bienia cbienia at CS.Princeton.EDU
Mon Jun 30 23:32:42 EDT 2008


You have to be careful with floating-point variables because the compiler
inserts some 'magic' to handle them. There is some information on the
Internet about how to handle it correctly. I found the following article:

    http://www.dmh2000.com/cpp/dswap.shtml

If that still doesn't solve it, there's also the hard core approach: Write
an extra program that converts all floats in the input file to integers (on
an x86 machine) using fixed-point arithmetic. Then adapt fluidanimate to
read in all integers using the new file format, byte swap them if necessary
and generate all floating-point variables from them.

- Chris


-----Original Message-----
From: parsec-users-bounces at lists.cs.princeton.edu
[mailto:parsec-users-bounces at lists.cs.princeton.edu] On Behalf Of Kshitij
Sudan
Sent: Monday, June 30, 2008 11:06 PM
To: PARSEC Users
Subject: Re: [parsec-users] vips and fluidanimate on Solaris/SPARC

I tried porting some of these applications for big-endian architecture
but did not make much headway since porting floating point inputs that
the application was reading from file turned out to be non-trivial.

The error mentioned here for fluidanimate arises when it reads a
floating point value. I had tracked this down when I was working and
more details about it can be found in my email to this mailing list.

On a different note, I also tried debugging canneal and the associated
bus error, but was unable to pinpoint what exactly was causing the
incorrect atomic swap function to be called. I even tried removing the
64-bit version of the function that was called but it did not work
either. There's something else going on there.

It appears at the outset that these problems are trivial and easily
fixable, but I can pretty much claim from experience that it's not the
case!

best,
-Kshitij
_______________________________________________
parsec-users mailing list
parsec-users at lists.cs.princeton.edu
https://lists.cs.princeton.edu/mailman/listinfo/parsec-users



More information about the parsec-users mailing list