I am generating C code from Matlab coder. I have a function as follows
function C = foo( A, B ) %#codegen
for j=1:100,
C = C+A(j);
end
end
The code for this function in c generated is
void foo(float A[100],B,float* C){
for(j=0;j<100;j++){
*C+=A[j];
}
}
I want the code to be efficient and generated in the following way:
void foo(float* A,B,float* C){
//here B is the length of the array
for(j=0;j<B;j++){
*C+=*(A+j);
}
}
Do you have an idea?
I don't understand the whole story, but why don't you change your matlab code to actually use B for a start and tell us what happens then, like
function C = foo( A, B ) %#codegen
for j=1:B,
C = C+A(j);
end
end
You probably need to optimize your Matlab code to be more like what you want in C. The Matlab code is accessing 100 elements in A; which is what the generated C is replicating. If you want it to only go to B then you would have to make that in Matlab.
% MATLAB
function C = foo( A, B ) %#codegen
for j=1:B,
C = C+A(j);
end
end
My guess is that Matlab Coder expects vectorized code:
Try again on this one:
function C = foo( A, B ) %#codegen
C = sum(A(1:100));
end
C is scalar... it seems he wants C=sum(A(1:B));