To do array copy from A(:,:,2) to A(:,:,1),
which of these results in better performance?
A(:,:,1) = A(:,:,2)
OR
DO 10 J=1,JMAX
DO 10 I=1,IMAX
A(I,J,1)=A(I,J,2)
10 CONTINUE
Hi Mr.Smith,
Assuming you compile at -O2 or above, the first will be faster since the implied do loop will be replaced by a memcpy.
Hope this helps,
Mat
$ cat cpy.f90
program test
real :: A(100,100,2)
A(:,:,1) = A (:,:,2)
print *, A(1,1,1)
end program test
$ pgf90 -O2 -Minfo cpy.f90
test:
5, Memory copy idiom, loop replaced by call to __c_mcopy4