Add Fortran examples

1_square/Makefile

@@ -6,5 +6,7 @@ clean:
 %.o: %.c
 	../dockerscript.sh clang-12 /host/$^ -O3 -Xclang -load -Xclang /Enzyme/enzyme/build/Enzyme/ClangEnzyme-12.so -ffast-math -o /host/$@
 
+# TODO: Fortran build once Flang plugin available
+
 run-%: %.o
 	../dockerscript.sh /host/$^ 3.14

1_square/square.f90

@@ -0,0 +1,24 @@
+program main
+  use enzyme, only: enzyme_autodiff
+  implicit none
+  real :: x, y, dy
+
+  ! Evaluate the function for some arbitrary input
+  x = 3.0
+  y = square(x)
+  write(*, "('square(',f0.0,')=',f0.0)") x, y
+
+  ! Evaluate its derivative using reverse mode
+  dy = 0.0
+  call enzyme_autodiff(square, x, dy)
+  write(*, "('Gradient square(',f0.0,')=',f0.0)") x, dy
+
+  contains
+
+    ! Function we seek to differentiate
+    real function square(x)
+      real, intent(in) :: x
+      square = x ** 2
+    end function
+
+end program

2_norm/Makefile

@@ -15,5 +15,7 @@ clean:
 %.o: %.ll
 	../dockerscript.sh clang-12 -O2 /host/$^ -o /host/$@ -lm
 
+# TODO: Fortran builds once Flang plugin available
+
 run-%: %.o
 	../dockerscript.sh /host/$^ 2000

2_norm/norm.f90

@@ -0,0 +1,43 @@
+program main
+  use enzyme, only: enzyme_const, enzyme_dup, enzyme_autodiff
+  implicit none
+  integer, parameter :: n = 1000000
+  integer, parameter :: initial_value = 20
+  real :: x(n), dx(n)
+  real :: y(n), dy(n)
+  real :: start, finish, time
+
+  ! Fill the input array with an arbitrary value
+  x(:) = initial_value
+
+  ! Time the primal computation
+  call cpu_time(start)
+  call norm(n, x, y)
+  call cpu_time(finish)
+  time = finish - start
+  write(*, "('Serial normalize ',f0.6,1x,es11.4)") time, y(n)
+
+  ! Zero the gradients and specify the seed
+  dx(:) = 0.0
+  dy(:) = 1.0
+
+  ! Time the reverse mode computation
+  call cpu_time(start)
+  call enzyme_autodiff(norm, enzyme_const, n, &
+                       enzyme_dup, x, dx, &
+                       enzyme_dup, y, dy)
+  call cpu_time(finish)
+  time = finish - start
+  write(*, "('Gradient normalize ',f0.6,2(1x,es11.4))") time, y(n), dx(1)
+
+contains
+
+  ! Function we seek to differentiate
+  subroutine norm(n, x, y)
+    integer, intent(in) :: n
+    real, dimension(n), intent(in) :: x
+    real, dimension(n), intent(out) :: y
+    y(:) = x / sum(x)
+  end subroutine
+
+end program

3_dot/Makefile

@@ -6,5 +6,7 @@ clean:
 %.o: %.c
 	../dockerscript.sh clang-12 /host/$^ -O3 -Xclang -load -Xclang /Enzyme/enzyme/build/Enzyme/ClangEnzyme-12.so -ffast-math -o /host/$@
 
+# TODO: Fortran builds once Flang plugin available
+
 run-%: %.o
 	../dockerscript.sh /host/$^

3_dot/dot.f90

@@ -0,0 +1,57 @@
+program main
+  use enzyme, only: enzyme_const, enzyme_dup, enzyme_autodiff
+  implicit none
+
+  integer, parameter :: n = 20000000
+  integer, parameter :: x = 20
+  real :: a(n), b(n), c = 1 / x
+  real :: da(n), db(n), dc
+  real :: start, finish, time
+  integer :: i, dn
+
+  ! Specify input
+  do i = 1, n
+    a(i) = x / i
+    b(i) = x + i - 1 ! NOTE: Minus one to account for 1-indexing
+  end do
+
+  ! Time gradient computation with respect to all variables (A, B, and C)
+  da(:) = 0.0
+  db(:) = 0.0
+  dc = 0.0
+  call cpu_time(start)
+  call enzyme_autodiff(dot, enzyme_const, n, &
+                       enzyme_dup, a, da, &
+                       enzyme_dup, b, db, &
+                       enzyme_dup, c, dc)
+  call cpu_time(finish)
+  time = finish - start
+  write(*, "('Grad Normalize time=',f0.6)", advance="no") time
+  write(*, "(' A`(1)=',f4.1,' B`(1)=',f4.1,' C`=',f4.1)") da(1), db(1), dc
+
+  ! Time gradient computation with respect to just B
+  da(:) = 0.0
+  db(:) = 0.0
+  dc = 0.0
+  call cpu_time(start)
+  call enzyme_autodiff(dot, enzyme_const, n, &
+                       enzyme_const, a, &
+                       enzyme_dup, b, db, &
+                       enzyme_const, c)
+  call cpu_time(finish)
+  time = finish - start
+  write(*, "('Constant A time=',f0.6)", advance="no") time
+  write(*, "(' A`(1)=',f4.1,' B`(1)=',f4.1,' C`=',f4.1)") da(1), db(1), dc
+
+contains
+
+  ! Function we seek to differentiate
+  real function dot(n, x, y, z)
+    integer, intent(in) :: n
+    real, dimension(n), intent(in) :: x
+    real, dimension(n), intent(in) :: y
+    real, intent(in) :: z
+    dot = dot_product(x, y) + z
+  end function
+
+end program

7_fwd/Makefile

@@ -6,5 +6,7 @@ clean:
 %.o: %.c
 	../dockerscript.sh clang-12 /host/$^ -O3 -Xclang -load -Xclang /Enzyme/enzyme/build/Enzyme/ClangEnzyme-12.so -ffast-math -o /host/$@
 
+# TODO: Fortran build once Flang plugin available
+
 run-%: %.o
 	../dockerscript.sh /host/$^ 3.14

7_fwd/fwd.f90

@@ -0,0 +1,37 @@
+program main
+  use enzyme, only: enzyme_fwddiff
+  implicit none
+
+  real :: a, b, c
+  real :: da, db, dc
+
+  ! Specify input
+  a = 2.0
+  b = 3.0
+
+  ! Specify seeds
+  da = 1.0
+  db = 1.0
+
+  ! Compute forward mode derivative
+  call enzyme_fwddiff(my_loop, a, da, b, db, c, dc)
+  write(unit=6, fmt="('  a=',f2.0,'  b=',f2.0,'  c=',f4.0)") a, b, c
+  write(unit=6, fmt="(' da=',f2.0,' db=',f2.0,' dc=',f4.0)") da, db, dc
+
+contains
+
+  ! Function we seek to evaluate
+  subroutine my_loop(x, y, z)
+    real, intent(in)  :: x
+    real, intent(in)  :: y
+    real, intent(out) :: z
+    real :: s
+    integer :: i
+    s = 0.0
+    do i = 1, 100
+      s = s + x + y
+    end do
+    z = s
+  end subroutine
+
+end program