Years ago, I heard one explanation for the seeming contradiction between
the C order and
the Fortran order that the C order is computer science based while the
Fortran order is
mathematics based. Here is what it means.
The C order, aka row major order, is "last index changes first". So, a
10 by 10 two dimensional
array is stored in memory like this:
(0,0), (0,1), ..., (0,9), (1,0), (1,1),..., (9,9)
which is as "natural" as "00, 01, ..., 09, 10, 11, ..., 99".
The Fortran order, aka column major order, is "first index changes
first". So, a 10 by 10 two dimensional
array is stored in memory like this:
(1,1), (2,1), ..., (10,1), (1, 2), (2,2), ..., (10, 10)
which seems unnatural. But if you recall when you learned
the Cartesian coordination in high school mathematics.
The x-axis is horizontal and y-axis is vertical. Points on the
Cartesian plain are (x, y):
(1,1), (2,1), ..., (10, 1), ...
Since Fortran, was first known as FORTRAN, was derived from
"The IBM Mathematical Formula Translating System", the Fortran
order is "natural" too.
One more comment, the two ordering, though seem like a simple
rotation at two dimensions, are not really the same in physics for higher
dimensions. The C order is in left-hand rule, while the Fortran order is in
right-hand rule. If mixed up, you may attempt to operate the human
heart in the wrong side. :-)
-Albert
On 3/14/13 11:57 PM, Elena Pourmal wrote:
John,
I think you explained it well.
Looks like Jason is not along. Let me try to explain one more time....
To address Jason's
"I've read the section of the hdf5 page that Pradeep linked to (C vs
Fortran ordering) several times, and for the life of me, I can't
figure out whether the hdf5 format ordering is transparent to the
application or not"
Yes, it is transparent.
Think about N-dim array in any language as 1-dim flatten array.
Transparency means that one always gets the same 1-dim flatten array
of data in any programming language used.
Dimensions of the user's array (dataset dimensions) are stored in HDF5
file as 1-dim array following a convention that the last element of
that array has the size of the fastest changing dimension of the
user's array. File format is INDEPENDENT from which language an array
is written (and yes, it is turned to be a C convention :-)
When an array is written from Fortran, the value of its first
dimension is stored in the last element of the 1-dim array in the HDF5
file.
When Fortran library queries dimensions for an array (dataset
dimensions), it knows as does the C library, that the last element of
the 1-dim array in the file is the size of the fastest changing
dimension of the user's array. It will be the first dimension for the
Fortran array and the last dimension for the C array. I.e., Fortran
wrappers flip dimensions, but NEVER touch user's data.
If one reads data written by a Fortran application using a C
application (and vice versa), data will appear to be transposed, but
it will be same data if you think about it as 1-dim flatten array!
Is it even more confusing now? :-)
Elena
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Elena Pourmal The HDF Group http://hdfgroup.org
1800 So. Oak St., Suite 203, Champaign IL 61820
217.531.6112
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
On Mar 13, 2013, at 2:28 PM, Biddiscombe, John A. wrote:
I'm forwarding this to the hdf mailing list just in case someone can
explain it better, or correct me
You write a Fortran array with dimensions say arr[10,20,30]
{x=10,y=20,z=30}, and this goes onto disk as an array with dimension
sizes
0 = 10
1 = 20
2 = 30
The convention is simply that the dimensions are listed in row major
order, so if you read the data in c into an array
arr[z][y][x]
then everything will look fine. nine times out of ten, I've found
that fortran arrays are declared as arr[x,y,z] and c arrays as
arr[z][y][x] so the programmer has already flipped the orders of the
dimensions and everything works out. All hdf does is say that the
dimensions are listed in row major order, you can interpret the data
how you like. Hdf doesn't say 'this is X, this is Y, this is Z' - it
only says, dim 0 is size 10, dim 1 size 20, dim 2 size 30 - you may
add metadata yourself to tell the user which 'should' be X/Y/Z if you
wish.
If the OP has data physically stored as fortran array[z,y,x] then the
data will be written out transposed relative to what we expect when
reading into paraview, then transposing will be necessary (we won't
go into coordinate transforms to achieve the same flipping at the
graphics end).
Did I get that right?
JB
-----Original Message-----
From: [email protected]
<mailto:[email protected]>
[mailto:[email protected]] On Behalf Of Jason Fleming
Sent: 13 March 2013 19:51
To: [email protected] <mailto:[email protected]>
Subject: Re: [Paraview] Fortran wrapper of HDF5
I've read the section of the hdf5 page that Pradeep linked to (C vs
Fortran ordering) several times, and for the life of me, I can't
figure out whether the hdf5 format ordering is transparent to the
application or not.
It seems really silly that hdf5 can take care of endianness so that
app developers don't have to worry about it, but on the other hand,
app developers now have to know whether a particular hdf5 file was
written by a C or Fortran app in order to be able to read it
properly. And yet
hdf5 seems to work that way. Is that right?
Cheers
Jason
On Thu, 2013-03-14 at 00:08 +0900, Pradeep Jha wrote:
Are you implying that if I use the HDF5 fortran wrapper to convert the
fortran binary data in h5 format and then visualize this h5 file using
Paraview, I am looking at the the actual data with correct dimensions?
Or I have to make some modifications so that I see the data correctly
in Paraview?
I dont want to transpose the data. I just want to visualize what I
wrote using Fortran without any alterations.
Pradeep
2013/3/13 Biddiscombe, John A. <[email protected]
<mailto:[email protected]>>
"How do I write the h5 file data in exactly the same way as it
was written in original binary file written by Fortran?"
It is writing the file the same, the problem is that fortran
stores arrays in column major, and C in row major order. You
state "it automatically transposes the matrix" -- not true -- it
transposes the dimensions so that the data is still stored the
same, but when you write array[z,y,x] from fortran, you want
to read it as array[x,y,z] from C. The actual data on disk is
the same as your binary fortran data, but the dimensions are
reversed compared to the same data from C.
Does that help? -- the short answer is just swap the order of
the dimensions in your read function in the C version and then
things should appear the same. (but you must declare your
arrays with the dimensions flipped).
If you want to actually transpose the data, then I'm sure
google will provide a code snippet
I hope I'm not remembering this wrong.
JB
From: [email protected]
<mailto:[email protected]>
[mailto:[email protected]] On Behalf Of Pradeep
Jha
Sent: 13 March 2013 10:58
To: [email protected] <mailto:[email protected]>
Subject: [Paraview] Fortran wrapper of HDF5
I recently noticed that when I am using the fortran wrapper of
HDF5 to convert a binary file written by fortran into the "h5"
format, it automatically transposes the matrix. Apparently,
this is because HDF5 uses the C convention for writing binary
files, as explained in section: 7.3.2.5. of this page.
Is anyone aware of this situation? And any solutions for this
problem? How do I write the h5 file data in exactly the same
way as it was written in original binary file written by
Fortran?
_______________________________________________
Powered by www.kitware.com <http://www.kitware.com>
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Please keep messages on-topic and check the ParaView Wiki at:
http://paraview.org/Wiki/ParaView
Follow this link to subscribe/unsubscribe:
http://www.paraview.org/mailman/listinfo/paraview
--
Dr. Jason G. Fleming
Chief Engineer, Seahorse Coastal Consulting
3103 Mandy Ln
Morehead City, NC 28557
Tel: (252) 726-6323
Mobile: (252) 269-0962
Web: http://www.seahorsecoastal.com
_______________________________________________
Powered by www.kitware.com <http://www.kitware.com>
Visit other Kitware open-source projects at
http://www.kitware.com/opensource/opensource.html
Please keep messages on-topic and check the ParaView Wiki at:
http://paraview.org/Wiki/ParaView
Follow this link to subscribe/unsubscribe:
http://www.paraview.org/mailman/listinfo/paraview
_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
_______________________________________________
Hdf-forum is for HDF software users discussion.
[email protected]
http://mail.hdfgroup.org/mailman/listinfo/hdf-forum_hdfgroup.org
