unimplemented serial netcdf functions

[Jianwei Li]

>
>FUNCTIONS: U == unimpplemented I == implemented
>
>ncmpi_get  text uchar schar short int long float double
>var1 [0]     I   U      U    I     I   U    I      I
>var          I   U      U    I     I   U    I      I     
>var_all      I   U      U    I     I   U    I      I     
>vara         I   U      U    I     I   U    I      I     
>vara_all     I   U      U    I     I   U    I      I     
>vars         I   U      U    I     I   U    I      I     
>vars_all     I   U      U    I     I   U    I      I     
>varm         U   U      U    U     U   U    U      U     
>varm_all     U   U      U    U     U   U    U      U     
>vars         I   U      U    I     I   U    I      I     
>vars_all     I   U      U    I     I   U    I      I     
>
>ncmpi_put  text uchar schar short int long float double
>var1 [0]     I   U      U    I     I   U    I      I
>var          I   U      U    I     I   U    I      I
>var_all      I   U      U    I     I   U    I      I
>vara         I   U      U    I     I   U    I      I
>vara_all     I   U      U    I     I   U    I      I
>vars         I   U      U    I     I   U    I      I
>vars_all     I   U      U    I     I   U    I      I
>varm         U   U      U    U     U   U    U      U
>varm_all     U   U      U    U     U   U    U      U
>vars         I   U      U    I     I   U    I      I
>vars_all     I   U      U    I     I   U    I      I

Yes, it's correct excet that ncmpi_put_var_all does not exist at all, 
as all such attemp would result in using ncmpi_put_vara_all, while 
ncmpi_get_var_all is designed for a special case of optimization so
that all processes get the whole array but each process only reads a
subarray for example.


I like this table :>

Also, the parallel netCDF use "NO_FILL" and does not include "set_fill"
support, since it is considered deprecate. 

>
>[0] the var1 functions are all independent operations
>
>
>It *might* make sense to have a get/put_var1_all, just to save the
>hassle of wrapping those calls with begin/end_indep() calls. 

Unless it won't confuse the user. I'm not sure myself.

Jianwei