performance - Fortran strange behavior when timing subroutines -

the strangest behavior happens when i'm timing code shown below. it's long it's doing matrix multiplication different ways of passing arguments. matmul0 use explicit shape, matmul1 use assumed shape, , matmul2 use assumed shape plus use pointers inside subroutine point matrices (don't ask me why this, doesn't matter). thing is, when time 3 subroutines, like:

time explicit shape: 3.712099 time assumed shape: 12.55620 time assumed shape + pointer: 3.821299 

now, if comment third subroutine (the 1 pointers), time like:

time explicit shape: 3.712099 time assumed shape: 3.824401 time assumed shape + pointer: 0.00000 

why happening? , why doesn't happen first subroutine too? i'm running on intel core i3, intel compiler, no optimization flags, ifort test.f90 -fpp (the fpp preprocessing timer macro). full code below.

#define timer(func, store) call system_clock(start_t, rate); call func; call system_clock(end_t); store  = store + real(end_t - start_t)/real(rate); program test     interface         subroutine matmul1(a, b, c)             real :: a(:,:), b(:,:), c(:,:)         end subroutine         subroutine matmul2(a, b, c)             real, target :: a(:,:), b(:,:), c(:,:)         end subroutine     end interface     real, allocatable, dimension(:,:) :: a, b, c     integer, parameter :: m = 500, n = 500, o = 500     integer, parameter :: loops = 100     integer :: start_t, end_t, rate     real :: time     allocate(a(m,n), b(n,o), c(m,o))     a(:,:) = 1; b(:,:) = 2; c(:,:) = 0      time = 0     = 1, loops     timer(matmul0(a, b, c, m, n, o), time)     end     print*, 'time explicit shape:', time      time = 0     = 1, loops     timer(matmul1(a, b, c), time)     end     print*, 'time assumed shape:', time      time = 0     = 1, loops     ! timer(matmul2(a, b, c), time)     end     print*, 'time assumed shape + pointer:', time  end program  subroutine matmul0(a, b, c, m, n, o)     integer :: m, n, o     real :: a(m,n), b(n,o), c(m,o)     = 1, m         j = 1, o             k = 1, n                 c(i,j) = c(i,j) + a(i,k)*b(k,j)             end         end     end end subroutine  subroutine matmul1(a, b, c)     real :: a(:,:), b(:,:), c(:,:)     = 1, size(c,1)         j = 1, size(c,2)             k = 1, size(a,2)                 c(i,j) = c(i,j) + a(i,k)*b(k,j)             end         end     end end subroutine  subroutine matmul2(a, b, c)     real, target :: a(:,:), b(:,:), c(:,:)     real, pointer :: a0(:,:), b0(:,:), c0(:,:)     a0 => a; b0 => b; c0 => c     = 1, size(c,1)         j = 1, size(c,2)             k = 1, size(a,2)                 c0(i,j) = c0(i,j) + a0(i,k)*b0(k,j)             end         end     end end subroutine 

this seems issue ifort , size() function calls inside loops...

on machine, code yields:

 time explicit shape:   3.145800  time assumed shape:   14.98891  time assumed shape + pointer:   14.82460 

i rewrote subroutines re-use array shapes:

subroutine matmul1(a, b, c)     real,intent(in) :: a(:,:), b(:,:)     real,intent(out) :: c(:,:)     integer :: m, n, o      m = size(c,1)     n = size(c,2)     o = size(a,2)     = 1, m         j = 1, n             k = 1, o                 c(i,j) = c(i,j) + a(i,k)*b(k,j)             end         end     end end subroutine 

[matmul2 adjusted similarly...]

now, get:

 time explicit shape:   3.159100  time assumed shape:   3.009802  time assumed shape + pointer:   3.567401 

interestingly, gfortran seems not have problem.

commenting out third subroutine call has no effect on either compilers on machine, can't there :(