David Manegold

I have a 15x51 matrix on which this algorithm fails. the JV handles it.

can I send you the matrix

I haven't seen such a brute collection of EVAL commands ever before. Almost all commands are converted to strings and delivered to the evil EVAL without any need to do so. David, why do you choose this strange, inefficient and error prone method to write a Matlab program?
Some improvements:
if isstruct(eval(['S.' names{i}])) ==>
if isstruct(S.(names{i}))
and
for ic = 1:cols, eval(['fwrite(fid,num2str(S.',names{i},'(ir,ic)));']); ... ==>
data = S.(names{i});
for ic = 1:cols, fprintf(fid, '%g', data(ir, ic)); ...

I do not dare to guess, what this command creates:
eval(['fwrite(fid,['''''''',S.',names{i},','''''''']);']);
Some comments in the code would be useful, especially because programs in the FEX are thought to be useful for 1000th of other users.
Checking the success of an FOPEN is strongly recommended.

This submission contains a lean help text, but neither the inputs nor the outputs are well defined. Without trying it it is hard to guess, if this function works e.g. for multi-dimensional output. The code is due to the EVALisation very slow and hard to read. Therefore I would not use this function for productive work or for eductation. While the idea is useful, the implementation is not usable.