CVS log for JSOC/proj/sharp/apps/sw_functions.c
	Help

Request diff between arbitrary revisions

removed debug variable countabit

removed some print statements

removed check for ( (derx_los[j * nx + i] + dery_los[j * nx + i]) == 0) in computeLOSderivative

fixed some issues with the LOS functions
added documentation

Performed a check on count_mask before assigning a value to ERRMSHA to avoid a divide by zero error.
If count_mask is indeed zero then ERRMSHA is now NaN.

Added LOS keywords

reverted to 1.35 and added comment to SAVNCPP header

modified the documentation for SAVNCPP; didn't write the units correctly (as square arcsecond) in version 1.32

out of bounds problem still remained in Function 8; fixed it

There was a counting out-of-bounds problem with the functions that produce MEANGBZ, MEANGBT, and MEANGBH. I re-coded it such that the error terms are calculated separately; further, the error arrays are initalized to zero and the edge pixels are ignored (since they will be selected out by the bitmap and conf_disambig bitmaps anyway).

changed the numerical derivative. here is a description of the problem that i fixed:

In computeJz(), the newly malloc'd jz array has its interior
points filled with values but its boundary points are left
alone:

    for (i = 1; i <= nx-2; i++)
    {
        for (j = 1; j <= ny-2; j++)
        {
            // calculate jz at all points
            ...

In all subsequent functions the array jz is used whole, including
the uninitialized border.

I ran your test case in the debugger.  On n02, jz was 0 on the
border; on solar3, I got wild values

(gdb) p jz[0]
$16 = 1.37765386e+10
(gdb) p jz[1]
$17 = 1.35754556e-19
(gdb) p jz[2]
$18 = 1.35631939e-19

If the code relies on the border being 0, then it was pure luck
that it ran correctly on x86_64.

-- modified sign error in horizontal component of lorentz force
-- changed units of SAVNCPP to amps/arcsec

added lorentz force function to calculate keys TOTFX, TOTFY, TOTFZ, TOTBSQ, EPSX, EPSY, EPSZ

jesper's gaussian convolution function left a hwidth stripe of 0's along the edge of the image. i fixed this by padding with zeros prior to convolving. this involves creating some intermediate arrays, so both sharp.c and update_sharp_keys.c also need to be updated with some mallocs and different calls to the R parameter function.

No code change. Added documentation to the header of the R function.

RE-REVERT
jesper has modified fresize.c such that there is no need to check on the dimension of the convolution kernel with respect to the array size

reverted back to v1.25

Revert the versions of several sharps source-code files to make them intercompatible.

remedied an incorrect check

more conservative comparision between convolution kernel and array size

added R parameter (adapted from Schrijver 2007)

removed some print statements

: fixed error in MEANSHR, SHRGT45, ERRBZ wherein all were giving nans at times -- this was due to some weird condition being satisfied, so now i check on that condition

updated error analysis

corrected gamma, alpha, and error in alpha
this is an incomplete version -- errors in errors remain
but i am pushing it over now so it is available on the web

added an "if jz eq nan, skip" line in SAVNCPP calculation

corrected indexing in the computation of jz_err in computeJz()

removed printf statements

Changed all sums to doubles instead of floats, since there is a significant loss of precision in summing millions of floats:

P% cat prec.c
#include <stdio.h>
#include <stdlib.h>

main()
  {
  float sum = 0.0;
  double dsum = 0.0;
  int n = 13000000;
  int i;

  for (i=0; i<n; i++)
    {
    sum += 1.333;
    dsum += 1.333;
    }

  printf("avg = %f, davg = %f\n", sum/n, dsum/n);
  }

P% cc prec.c -o prec
P% prec
avg = 1.181585, davg = 1.333000
P%

now contains a unique cvs version number
initialized some uninitalized variables

reinstated old version of calculating mean free energy; total free energy remains unchanged

Hasty last time! Clean-up and modified error in mean free energy calculation.

Corrected factor of 8*PI in total and mean free energy calculation

Oops, last commit used a 6th-order finite difference method. I replace it with a second-order finite difference method.

Rewrote module to compute errors using formal error propagation and VFISV error arrays

re-modified Jz calculation (a bit hasty last time)

corrected error in rsun_ref / rsun_obs

Only changed documentation

Use a smoothed jz array using fresize.h routines.
Modified the methodology for calculating alpha.

some nans were being propagated into the spaceweather calculations; this is fixed

Modified code to select pixels in conf_disambig map that are greater than 70 and pixels in bitmap that are greater than 30.

per Monica's request

Update functioning code sharp.c per Monica's request