From bug-octave-request at bevo dot che dot wisc dot edu  Thu Jan 29 16:58:09 1998
Subject: row/column major ordering problem
From: "John W. Eaton" <jwe at bevo dot che dot wisc dot edu>
To: Michael Hohn <hohn at math dot utah dot edu>
Cc: bug-octave at bevo dot che dot wisc dot edu
Date: Thu, 29 Jan 1998 16:57:49 -0600

On 29-Jan-1998, Michael Hohn <hohn at math dot utah dot edu> wrote:

| This is a problem I thought fixed a long time ago, but here it it
| again: 
| 
| I have a program writing files in matlab binary format, in either row
| or column major order:
| 
| ./a.out
| Matrix: 2 by 2
| row 0:    0.846186767    0.468314366 
| row 1:    0.236917463     0.65696803 
| 
| 
| In matlab:
| >> load foo.mat
| >> mat
| mat =
|     0.8462    0.4683
|     0.2369    0.6570
|  
| 
| Octave, version 2.0.9 (sparc-sun-solaris2.5).
| octave:1> load foo.mat
| octave:2> mat
| mat =
|  
|   0.84619  0.23692
|   0.46831  0.65697
| 
| Oops!! Apparently, the order flag is ignored...
| 
| Fix:
| ---
| (from an old message):
| 
| The first thing written to a matlab binary file should be a long int
| (4 bytes); it contains info of the form
| 
| ... + 1000*?? + 100*ORDER + 10*?? + ...
| 
| The ORDER flag is 0 for column major order, 1 for row major. 

I could never find this documented.  All I ever found was a statement
that the third digit was ignored.

In any case, if the third digit is supposed to be a flag that says
that the ordering is row or column major, here is a patch that should
fix the problem.  It's relative to 2.0.9.97, but I think it should
work ok for 2.0.9 too.

Thanks,

jwe


Thu Jan 29 16:25:46 1998  John W. Eaton  <jwe at bevo dot che dot wisc dot edu>

	* load-save.cc (read_mat_binary_data): Handle third digit of MOPT
	as flag indicating row or column major ordering.


*** src/load-save.cc~	Wed Jan 28 00:41:27 1998
--- src/load-save.cc	Thu Jan 29 16:52:47 1998
***************
*** 1235,1241 ****
    Matrix re;
    oct_mach_info::float_format flt_fmt = oct_mach_info::unknown;
    char *name = 0;
!   int swap = 0, type = 0, prec = 0, mach = 0, dlen = 0;
  
    FOUR_BYTE_INT mopt, nr, nc, imag, len;
  
--- 1235,1241 ----
    Matrix re;
    oct_mach_info::float_format flt_fmt = oct_mach_info::unknown;
    char *name = 0;
!   int swap = 0, type = 0, prec = 0, order = 0, mach = 0, dlen = 0;
  
    FOUR_BYTE_INT mopt, nr, nc, imag, len;
  
***************
*** 1248,1258 ****
  	return 0;
      }
  
!   type = mopt % 10; // Full, sparse, etc.
!   mopt /= 10;       // Eliminate first digit.
!   prec = mopt % 10; // double, float, int, etc.
!   mopt /= 100;      // Skip unused third digit too.
!   mach = mopt % 10; // IEEE, VAX, etc.
  
    flt_fmt = mopt_digit_to_float_format (mach);
  
--- 1248,1260 ----
  	return 0;
      }
  
!   type = mopt % 10;  // Full, sparse, etc.
!   mopt /= 10;        // Eliminate first digit.
!   prec = mopt % 10;  // double, float, int, etc.
!   mopt /= 10;        // Eliminate second digit.
!   order = mopt % 10; // Row or column major ordering.
!   mopt /= 10;        // Eliminate third digit.
!   mach = mopt % 10;  // IEEE, VAX, etc.
  
    flt_fmt = mopt_digit_to_float_format (mach);
  
***************
*** 1287,1292 ****
--- 1289,1301 ----
    if (dlen < 0)
      goto data_read_error;
  
+   if (order)
+     {
+       int tmp = nr;
+       nr = nc;
+       nc = tmp;
+     }
+ 
    re.resize (nr, nc);
  
    read_mat_binary_data (is, re.fortran_vec (), prec, dlen, swap, flt_fmt);
***************
*** 1315,1324 ****
  	for (int i = 0; i < nr; i++)
  	  ctmp (i, j) = Complex (re (i, j), im (i, j));
  
!       tc = ctmp;
      }
    else
!     tc = re;
  
    if (type == 1)
      tc = tc.convert_to_str ();
--- 1324,1333 ----
  	for (int i = 0; i < nr; i++)
  	  ctmp (i, j) = Complex (re (i, j), im (i, j));
  
!       tc = order ? ctmp.transpose () : ctmp;
      }
    else
!     tc = order ? re.transpose () : re;
  
    if (type == 1)
      tc = tc.convert_to_str ();