Fixes

05/?/94
log-detrminant incorrect on some computers
tol in mkldet should be declared as real*4

06/02/94
second should be declared as real*4

06/15/94
checkb return too conservative estimate when the matrix was less than full 
rank.

11/18/94
tol of 1d-6 may be too high. replaced by 1d-9.
Ja in option 61 is permuted and thus not in original order; no changes
made but manual modified.

11/21/94
sparse inverse in option 60 generated elements with column index 0 for each
inverse element close to 0; inversion caused segmentation error especially with
large data sets;

11/23/94
option 61 computed incorrectly off-diagonals for a small fraction of
inverse elements and only with selected data sets; memory reserved
was excessive

12/27/94
real*8 variables were not aligned, which created problems especially on DEC 
Alpha computers; real*8 are now aligned, however, the work vector in fspak 
needs to be aligned; in the easiest case that vector can be
equivalenced with any real*8 variable, which is aligned automatically by the 
compiler.  

1/17/95
FSPAK crashes when some equations are empty. 

10/1/97
FSPAK returns with insufficient memory message when memory available 
is sufficient but smaller than the largest memory available in previous calls
to FSPAK.

5/25/07
Operations to multiply by factor (s_original) or its inverse (s_inverse) added by Juan Pablo Sanchez. Some redundant declarations deleted

1/14/08
do while(k.ne.0 and found(k)) in SEGPATH generated errors with bound checking or with some compilers. Code rewritten.