The call routines built for proc fcmp are extremely constraining because array size must be predefined AND you cannot use variables to set array sizes unless using macro variables. At compilation, values in the brackets for array name{ } must be numeric constants. Here is a way around
proc fcmp outlib=sasuser.funcs.temp;
subroutine returntheta(a[*,*],b[*,*], transa[*,*], result[1,1]);
outargs transa, result;
call transpose(a, transa);
call mult(transa, b, result);
endsub;
quit;
options cmplib=sasuser.funcs;
data _null_;
array a{6,1} _temporary_ (1,2,3,4,5,6);
array b{6,1} _temporary_ (1,2,3,4,5,6);
array transa{1,6} _temporary_;
array test{1,1} _temporary_;
call returntheta(a,b,transa, test);
put test[1,1]=;
run;
I'm no expert but from what I picked up on reviewing programs I made 2-3 months ago,
The matrix call routine mult requires an array output hence test{1,1} instead of a simple numeric variable. You could make it a numeric variable but since you need to pass array transa and modify it inside, it requires outargs statement and thus a call routine regardless. Still, you could define the 1*1 array directly inside the subroutine and use an actual numeric variable output if you don't want to have to reference test[1,1] in your data step.
proc iml will give you a much prettier result and far more flexibility if you ever need any somewhat more complex matrix manipulation and supports matrix dimension input as variables and not only numeric constants.
Because of the limitation on array definition, even if you wanted to move towards using macro variables and dim1 & dim2 functions, your subroutine would need to be re-compiled each time the array size changes so it's not really a good option.
Hope this helps.
Vincent
*Edited the code to reflect the message below, fixing the 2 errors.
