Skip to content
Tutorials Center
My learning

Topics

Lesson 16 of 39 ยท Arrays and Numerics

Array Intrinsic Functions

Master Fortran's array intrinsic functions: sum, product, maxval, minloc, count, any, all, dot_product, matmul, transpose, pack, spread, cshift and the dim and mask arguments.

Fortran ships with a rich library of functions that take a whole array and return a single answer, or a new array. Many programs that need a loop in other languages become a one-liner here. This lesson tours the most useful intrinsics, shows the dim and mask arguments, and ends with a few realistic examples.

Reductions: one answer from many

Function Result
sum(a) sum of all elements
product(a) product of all elements
maxval(a) / minval(a) largest / smallest element
maxloc(a) / minloc(a) position of the largest / smallest, as an array
count(mask) how many elements of a logical array are true
any(mask) / all(mask) is at least one / is every element true
size(a) number of elements
norm2(a) Euclidean length (Fortran 2008)
dot_product(a, b) sum of a(i) * b(i)
FORTRAN
program reductions
  implicit none
  real :: a(5) = [3.0, 8.0, 1.0, 9.0, 4.0]

  print *, sum(a)               ! 25
  print *, sum(a) / size(a)     ! mean: 5
  print *, maxval(a), minval(a) ! 9 and 1
  print *, maxloc(a)            ! 4  (as a one-element array)
  print *, count(a > 3.0)       ! 3
  print *, any(a > 8.5), all(a > 0.0)    ! T T
  print *, dot_product(a, a)    ! 171
end program reductions

Notice that maxloc returns an array (one index per dimension). For a 1D array that is one number, and you take it with maxloc(a, dim=1) to get a scalar, or just maxloc(a, 1). Positions count from the array's lower bound... almost: maxloc of an array declared a(0:4) still reports 1 for the first element, since positions are relative to the start of the array. Remember this when your bounds do not start at 1.

Python Stand-in (Python): reductions

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The dim argument

For a matrix, sum(m) adds everything. Add dim= to reduce along one dimension and get an array back:

FORTRAN
integer :: m(2, 3) = reshape([1, 2, 3, 4, 5, 6], [2, 3])
! m = | 1 3 5 |
!     | 2 4 6 |

print *, sum(m)             ! 21
print *, sum(m, dim=1)      ! 3 7 11   sum down each column
print *, sum(m, dim=2)      ! 9 12     sum across each row
print *, maxval(m, dim=2)   ! 5 6

dim=1 collapses the first index: you get one value per column. This surprises people who think of it as "sum by row". Work it out: the first dimension disappears.

Python Stand-in (Python): dim=1 and dim=2

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The mask argument

Most reductions accept mask=, a logical array of the same shape; only elements where the mask is true take part:

FORTRAN
real :: t(6) = [12.0, -3.0, 25.0, 7.0, -8.0, 30.0]

print *, sum(t, mask=t > 0.0)           ! 74: sum of positives
print *, count(t < 0.0)                 ! 2 negatives
print *, maxval(t, mask=t < 20.0)       ! 12
print *, minval(t, mask=t > 0.0)        ! 7

If the mask selects nothing, sum is 0 and product is 1; maxval gives the most negative number of the type.

Matrices: matmul and transpose

matmul(a, b) multiplies matrices (rows times columns). transpose(m) swaps rows and columns:

FORTRAN
real :: a(2, 2) = reshape([1, 3, 2, 4], [2, 2])     ! | 1 2 |
real :: b(2, 2) = reshape([0, 1, 1, 0], [2, 2])     ! | 3 4 |  and | 0 1 | swap matrix
real :: c(2, 2)                                      !              | 1 0 |

c = matmul(a, b)        ! | 2 1 |
                        ! | 4 3 |   (columns of a swapped)

matmul also handles matrix-vector products, matmul(a, x). For large matrices the compiler can call a fast library (BLAS) behind the scenes. dot_product(x, y) is the vector special case.

Python Stand-in (Python): matmul, transpose, dot product

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Building and rearranging arrays

Function Purpose Example
reshape(a, shape) change shape reshape(v, [3, 4])
spread(a, dim, n) repeat along a new dimension spread([1,2,3], 1, 2) gives 2 rows
merge(t, f, mask) choose elementwise merge(a, 0.0, a > 0.0)
pack(a, mask) gather selected elements into a vector pack(a, a > 0)
unpack(v, mask, field) scatter a vector back
cshift(a, k) circular shift cshift([1,2,3,4], 1) = [2,3,4,1]
eoshift(a, k) end-off shift, filling with zero eoshift([1,2,3,4], 1) = [2,3,4,0]
product, iall, iany bitwise or product reductions

pack with a mask is the Fortran way to filter. merge is an elementwise conditional (like the ternary operator): merge(x, y, cond) takes x where cond is true and y otherwise, so abs_a = merge(a, -a, a >= 0.0) is a one-line absolute value.

Python Stand-in (Python): pack, merge, cshift, eoshift

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

A worked example: statistics in four lines

FORTRAN
program stats
  implicit none
  real :: x(8) = [4.0, 8.0, 15.0, 16.0, 23.0, 42.0, 5.0, 9.0]
  real :: mean, var

  mean = sum(x) / size(x)
  var  = sum((x - mean) ** 2) / (size(x) - 1)
  print *, "mean =", mean
  print *, "sd   =", sqrt(var)
  print *, "range:", minval(x), "to", maxval(x)
end program stats

The mean is 15.25; the variance and standard deviation come from the same array expressions. (x - mean) ** 2 is an array of squared deviations, and sum collapses it. There is not a single loop.

Python Stand-in (Python): the statistics example

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Which call returns the sum of each column of a matrix m?

    dim=1 collapses the first (row) index, leaving one sum per column.
  2. 2. What does count(a > 3.0) return?

    count counts the true elements of a logical array.
  3. 3. What does merge(a, 0.0, a > 0.0) compute?

    merge chooses the first argument where the mask is true and the second elsewhere, keeping the shape.
  4. 4. What does pack(a, a > 0.0) return?

    pack gathers the selected elements into a shorter vector.

Summary

  • Reductions such as sum, product, maxval, minval, count, any and all replace loops.
  • dim= reduces along one dimension; mask= limits which elements take part.
  • maxloc/minloc return positions as arrays, relative to the array start.
  • matmul, transpose and dot_product do linear algebra; reshape, spread, merge, pack, cshift rearrange data.
  • Array expressions like sum((x - mean) ** 2) give statistics without a loop.

Next: where and Masking