data type on matlab performance

Hi all,

I am analyzing the performance of my code using profile. Here is part of my code:

global NX1 NX2 LU_nr dx1 dx2 u1 u2 strain du11 du12 du21 du22 index;
j=1:1:LU_nr ;  
%left
idx1=(rem(j,NX1)==1);
idx2=(rem(j,NX1)==0);
idx3=(~(idx1|idx2));
index(idx1)=LU_nr+(j(idx1)-1)/NX1+1;
du11(idx1)=(u1(j(idx1)+1)-u1(index(idx1)))/(2.*dx1);
du21(idx1)=(u2(j(idx1)+1)-u2(index(idx1)))/(2.*dx1);
%right
index(idx2)=LU_nr+NX2+j(idx2)/NX1;
du11(idx2)=(u1(index(idx2))-u1(j(idx2)-1))/(2.*dx1);
du21(idx2)=(u2(index(idx2))-u2(j(idx2)-1))/(2.*dx1);
%middle
du11(idx3)=(u1(j(idx3)+1)-u1(j(idx3)-1))/(2.*dx1);
du21(idx3)=(u2(j(idx3)+1)-u2(j(idx3)-1))/(2.*dx1);

I found that the global variables on line 1 cost more than half time. I have defined the variables in the main function and use them in this subroutine. Can anyone suggest a more efficient way? Thank you.