Merge branch 'main' into delete-caf-macro

Author: milancurcic

CMakeLists.txt

@@ -78,7 +78,23 @@ if(CMAKE_Fortran_COMPILER_ID MATCHES Cray)
 endif()
 
 # library to archive (libneural.a)
-add_library(neural src/mod_activation.f90 src/mod_io.f90 src/mod_kinds.f90 src/mod_layer.f90 src/mod_mnist.f90 src/mod_network.f90 src/mod_parallel.f90 src/mod_random.f90)
+add_library(neural
+  src/mod_activation.f90
+  src/mod_activation_submodule.f90
+  src/mod_io.f90
+  src/mod_io_submodule.f90
+  src/mod_kinds.f90
+  src/mod_layer.f90
+  src/mod_layer_submodule.f90
+  src/mod_mnist.f90
+  src/mod_mnist_submodule.f90
+  src/mod_network.f90
+  src/mod_network_submodule.f90
+  src/mod_parallel.f90
+  src/mod_parallel_submodule.f90
+  src/mod_random.f90
+  src/mod_random_submodule.f90
+)
 
 # Remove leading or trailing whitespace
 string(REGEX REPLACE "^ | $" "" LIBS "${LIBS}")

README.md

@@ -43,6 +43,12 @@ Dependencies:
 * OpenCoarrays (optional, for parallel execution, GFortran only)
 * BLAS, MKL (optional)
 
+Compilers tested include:
+
+* gfortran-10.3.0
+* ifort-2021.4
+* ifx-2021.4
+
 ### Building with fpm
 
 #### Building in serial mode

fpm.toml

@@ -1,6 +1,6 @@
 name = "neural-fortran"
-version = "0.1.0"
+version = "0.2.0"
 license = "MIT"
 author = "Milan Curcic"
 maintainer = "milancurcic@hey.com"
-copyright = "Copyright 2018-2021, neural-fortran contributors"
+copyright = "Copyright 2018-2022, neural-fortran contributors"

src/mod_layer.f90

@@ -31,27 +31,30 @@ module mod_layer
   end type array2d
 
   interface layer_type
-      type(layer_type) module function constructor(this_size, next_size) result(layer)
-        !! Layer class constructor. this_size is the number of neurons in the layer.
-        !! next_size is the number of neurons in the next layer, used to allocate
-        !! the weights.
-        implicit none
-        integer(ik), intent(in) :: this_size, next_size
-      end function constructor
+    module function constructor(this_size, next_size) result(layer)
+      !! Layer class constructor. this_size is the number of neurons in the layer.
+      !! next_size is the number of neurons in the next layer, used to allocate
+      !! the weights.
+      implicit none
+      integer(ik), intent(in) :: this_size, next_size
+      type(layer_type) layer
+    end function constructor
   end interface layer_type
 
   interface array1d
-    pure type(array1d) module function array1d_constructor(length) result(a)
+    pure module function array1d_constructor(length) result(a)
       !! Overloads the default type constructor.
       implicit none
       integer(ik), intent(in) :: length
+      type(array1d) :: a
     end function array1d_constructor  
   end interface array1d
 
   interface array2d  
-    pure type(array2d) module function array2d_constructor(dims) result(a)
+    pure module function array2d_constructor(dims) result(a)
       !! Overloads the default type constructor.
       integer(ik), intent(in) :: dims(2)
+      type(array2d) :: a
     end function array2d_constructor
   end interface array2d
 

src/mod_layer_submodule.f90

@@ -6,8 +6,9 @@
 
 contains
 
-  type(layer_type) module function constructor(this_size, next_size) result(layer)
+  module function constructor(this_size, next_size) result(layer)
     integer(ik), intent(in) :: this_size, next_size
+    type(layer_type) :: layer
     allocate(layer % a(this_size))
     allocate(layer % z(this_size))
     layer % a = 0
@@ -16,14 +17,16 @@ type(layer_type) module function constructor(this_size, next_size) result(layer)
     layer % b = randn(this_size)
   end function constructor
 
-  pure type(array1d) module function array1d_constructor(length) result(a)
+  pure module function array1d_constructor(length) result(a)
     integer(ik), intent(in) :: length
+    type(array1d) :: a
     allocate(a % array(length))
     a % array = 0
   end function array1d_constructor
 
-  pure type(array2d) module function array2d_constructor(dims) result(a)
+  pure module function array2d_constructor(dims) result(a)
     integer(ik), intent(in) :: dims(2)
+    type(array2d) :: a
     allocate(a % array(dims(1), dims(2)))
     a % array = 0
   end function array2d_constructor

src/mod_network.f90

@@ -40,13 +40,14 @@ module mod_network
 
   interface network_type
 
-    type(network_type) module function net_constructor(dims, activation) result(net)
+    module function net_constructor(dims, activation) result(net)
       !! Network class constructor. Size of input array dims indicates the total
       !! number of layers (input + hidden + output), and the value of its elements
       !! corresponds the size of each layer.
       implicit none
       integer(ik), intent(in) :: dims(:)
       character(len=*), intent(in), optional :: activation
+      type(network_type) :: net
     end function net_constructor
 
   end interface network_type
@@ -62,7 +63,6 @@ pure real(rk) module function accuracy(self, x, y)
       real(rk), intent(in) :: x(:,:), y(:,:)
     end function accuracy
 
-
     pure module subroutine backprop(self, y, dw, db)
       !! Applies a backward propagation through the network
       !! and returns the weight and bias gradients.
@@ -82,7 +82,6 @@ pure module subroutine fwdprop(self, x)
       real(rk), intent(in) :: x(:)
     end subroutine fwdprop
 
-
     module subroutine init(self, dims)
       !! Allocates and initializes the layers with given dimensions dims.
       implicit none

src/mod_network_submodule.f90

@@ -8,9 +8,10 @@
 
 contains
 
-  type(network_type) module function net_constructor(dims, activation) result(net)
+  module function net_constructor(dims, activation) result(net)
     integer(ik), intent(in) :: dims(:)
     character(len=*), intent(in), optional :: activation
+    type(network_type) :: net
     call net % init(dims)
     if (present(activation)) then
       call net % set_activation(activation)
@@ -20,7 +21,6 @@ type(network_type) module function net_constructor(dims, activation) result(net)
     call net % sync(1)
   end function net_constructor
 
-
   pure real(rk) module function accuracy(self, x, y)
     class(network_type), intent(in) :: self
     real(rk), intent(in) :: x(:,:), y(:,:)