Standard conformance fixes (#190)

* Make various functions impure to allow returning polymorphic allocatables

* Update tested compilers

* More on tested compilers

* Update CMake definitions to make serial default and add PARALLEL macro

Author: milancurcic

README.md

@@ -63,9 +63,10 @@ Optional dependencies are:
 
 Compilers tested include:
 
-* gfortran-9.4.0
-* ifort-2021.4
-* ifx-2021.4
+* flang-new 20.0.0
+* gfortran 13.2.0, 14.0.1
+* ifort 2021.13.1
+* ifx 2024.2.1
 
 ### Building with fpm
 
@@ -85,7 +86,7 @@ Once installed, use the compiler wrappers `caf` and `cafrun` to build and execut
 in parallel, respectively:
 
 ```
-fpm build --compiler caf --profile release
+fpm build --compiler caf --profile release --flag "-cpp -DPARALLEL"
 ```
 
 #### Testing with fpm
@@ -107,7 +108,7 @@ See the [Fortran Package Manager](https://github.com/fortran-lang/fpm) for more
 ```
 mkdir build
 cd build
-cmake .. -DSERIAL=1
+cmake ..
 make
 ```
 
@@ -122,7 +123,7 @@ in parallel, respectively:
 
 
 ```
-FC=caf cmake ..
+FC=caf cmake .. -DPARALLEL
 make
 cafrun -n 4 bin/mnist # run MNIST example on 4 cores
 ```
@@ -139,7 +140,7 @@ FC=ifort cmake ..
 for a parallel build of neural-fortran, or
 
 ```
-FC=ifort cmake .. -DSERIAL=1
+FC=ifort cmake ..
 ```
 
 for a serial build.

cmake/compilers.cmake

@@ -1,11 +1,12 @@
 # compiler flags for gfortran
 if(CMAKE_Fortran_COMPILER_ID STREQUAL "GNU")
 
-  if(SERIAL)
-    message(STATUS "Configuring to build with -fcoarray=single")
-    add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:-fcoarray=single>")
+  if(PARALLEL)
+    message(STATUS "Configuring to build with -fcoarray=shared")
+    add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:-fcoarray=shared>")
+    add_compile_definitions(PARALLEL)
   else()
-    add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:-fcoarray=lib>")
+    add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:-fcoarray=single>")
   endif()
 
   if(BLAS)
@@ -14,21 +15,22 @@ if(CMAKE_Fortran_COMPILER_ID STREQUAL "GNU")
     message(STATUS "Configuring build to use BLAS from ${BLAS}")
   endif()
 
-  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-fcheck=bounds;-fbacktrace>")
-  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-Ofast;-fno-frontend-optimize;-fno-backtrace>")
+  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-cpp;-fcheck=bounds;-fbacktrace>")
+  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-cpp;-Ofast;-fno-frontend-optimize;-fno-backtrace>")
 
 elseif(CMAKE_Fortran_COMPILER_ID MATCHES "^Intel")
   # compiler flags for ifort
 
-  if(SERIAL)
-    message(STATUS "Configuring to build with -coarray=single")
+  if(PARALLEL)
+    message(STATUS "Configuring to build with -coarray=shared")
     if(WIN32)
-      add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:/Qcoarray:single>")
-      add_link_options("$<$<COMPILE_LANGUAGE:Fortran>:/Qcoarray:single>")
+      add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:/Qcoarray:shared>")
+      add_link_options("$<$<COMPILE_LANGUAGE:Fortran>:/Qcoarray:shared>")
     else()
-      add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:-coarray=single>")
-      add_link_options("$<$<COMPILE_LANGUAGE:Fortran>:-coarray=single>")
+      add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:-coarray=shared>")
+      add_link_options("$<$<COMPILE_LANGUAGE:Fortran>:-coarray=shared>")
     endif()
+    add_compile_definitions(PARALLEL)
   else()
     if(WIN32)
       add_compile_options("$<$<COMPILE_LANGUAGE:Fortran>:/Qcoarray:shared>")
@@ -40,16 +42,16 @@ elseif(CMAKE_Fortran_COMPILER_ID MATCHES "^Intel")
   endif()
 
   if(WIN32)
-    string(APPEND CMAKE_Fortran_FLAGS " /assume:byterecl")
+    string(APPEND CMAKE_Fortran_FLAGS " /assume:byterecl /fpp")
   else()
-    string(APPEND CMAKE_Fortran_FLAGS " -assume byterecl")
+    string(APPEND CMAKE_Fortran_FLAGS " -assume byterecl -fpp")
   endif()
-  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-check;-traceback>")
-  # add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-O3>")
+  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-fpp;-check;-traceback>")
+  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-fpp;-O3>")
 
 elseif(CMAKE_Fortran_COMPILER_ID STREQUAL "Cray")
   # compiler flags for Cray ftn
   string(APPEND CMAKE_Fortran_FLAGS " -h noomp")
-  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-O0;-g>")
-  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-O3>")
+  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Debug>>:-e Z;-O0;-g>")
+  add_compile_options("$<$<AND:$<COMPILE_LANGUAGE:Fortran>,$<CONFIG:Release>>:-e Z;-O3>")
 endif()

cmake/options.cmake

@@ -1,4 +1,4 @@
-option(SERIAL "Serial execution")
+option(PARALLEL "Parallel execution")
 option(${PROJECT_NAME}_BUILD_TESTING "build ${PROJECT_NAME} tests" true)
 option(${PROJECT_NAME}_BUILD_EXAMPLES "build ${PROJECT_NAME} examples" true)
 
@@ -8,10 +8,10 @@ set(CMAKE_LIBRARY_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
 set(CMAKE_ARCHIVE_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/lib)
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 
-if(SERIAL)
-  message(STATUS "Configuring build for serial execution")
-else()
+if(PARALLEL)
   message(STATUS "Configuring build for parallel execution")
+else()
+  message(STATUS "Configuring build for serial execution; configure with -DPARALLEL=1 for a parallel build")
 endif()
 
 # --- Generally useful CMake project options

example/cnn_mnist.f90

@@ -40,9 +40,8 @@ program cnn_mnist
       optimizer=sgd(learning_rate=3.) &
     )
 
-    if (this_image() == 1) &
-      print '(a,i2,a,f5.2,a)', 'Epoch ', n, ' done, Accuracy: ', accuracy( &
-        net, validation_images, label_digits(validation_labels)) * 100, ' %'
+    print '(a,i2,a,f5.2,a)', 'Epoch ', n, ' done, Accuracy: ', accuracy( &
+      net, validation_images, label_digits(validation_labels)) * 100, ' %'
 
   end do epochs
 

example/dense_mnist.f90

@@ -24,9 +24,8 @@ program dense_mnist
 
   call net % print_info()
 
-  if (this_image() == 1) &
-    print '(a,f5.2,a)', 'Initial accuracy: ', accuracy( &
-      net, validation_images, label_digits(validation_labels)) * 100, ' %'
+  print '(a,f5.2,a)', 'Initial accuracy: ', accuracy( &
+    net, validation_images, label_digits(validation_labels)) * 100, ' %'
 
   epochs: do n = 1, num_epochs
 
@@ -44,10 +43,9 @@ program dense_mnist
       ! 2 metrics; 1st is default loss function (quadratic), other is Pearson corr.
       output_metrics = net % evaluate(validation_images, label_digits(validation_labels), metric=corr())
       mean_metrics = sum(output_metrics, 1) / size(output_metrics, 1)
-      if (this_image() == 1) &
-        print '(a,i2,3(a,f6.3))', 'Epoch ', n, ' done, Accuracy: ', &
-          accuracy(net, validation_images, label_digits(validation_labels)) * 100, &
-          '%, Loss: ', mean_metrics(1), ', Pearson correlation: ', mean_metrics(2)
+      print '(a,i2,3(a,f6.3))', 'Epoch ', n, ' done, Accuracy: ', &
+        accuracy(net, validation_images, label_digits(validation_labels)) * 100, &
+        '%, Loss: ', mean_metrics(1), ', Pearson correlation: ', mean_metrics(2)
     end block
 
   end do epochs

src/nf/nf_activation.f90

@@ -557,7 +557,7 @@ pure function eval_3d_celu_prime(self, x) result(res)
     end where
   end function
 
-  pure function get_activation_by_name(activation_name) result(res)
+  function get_activation_by_name(activation_name) result(res)
   ! Workaround to get activation_function with some
   ! hardcoded default parameters by its name.
   ! Need this function since we get only activation name

src/nf/nf_conv2d_layer.f90

@@ -41,7 +41,7 @@ module nf_conv2d_layer
   end type conv2d_layer
 
   interface conv2d_layer
-    pure module function conv2d_layer_cons(filters, kernel_size, activation) &
+    module function conv2d_layer_cons(filters, kernel_size, activation) &
       result(res)
       !! `conv2d_layer` constructor function
       integer, intent(in) :: filters

src/nf/nf_conv2d_layer_submodule.f90

@@ -7,7 +7,7 @@
 
 contains
 
-  pure module function conv2d_layer_cons(filters, kernel_size, activation) result(res)
+  module function conv2d_layer_cons(filters, kernel_size, activation) result(res)
     implicit none
     integer, intent(in) :: filters
     integer, intent(in) :: kernel_size

src/nf/nf_dense_layer.f90

@@ -42,7 +42,7 @@ module nf_dense_layer
   end type dense_layer
 
   interface dense_layer
-    elemental module function dense_layer_cons(output_size, activation) &
+    module function dense_layer_cons(output_size, activation) &
       result(res)
       !! This function returns the `dense_layer` instance.
       integer, intent(in) :: output_size

src/nf/nf_dense_layer_submodule.f90

@@ -8,7 +8,7 @@
 
 contains
 
-  elemental module function dense_layer_cons(output_size, activation) &
+  module function dense_layer_cons(output_size, activation) &
     result(res)
     integer, intent(in) :: output_size
     class(activation_function), intent(in) :: activation
@@ -129,7 +129,9 @@ module subroutine init(self, input_shape)
     self % weights = self % weights / self % input_size
 
     ! Broadcast weights to all other images, if any.
+#ifdef PARALLEL
     call co_broadcast(self % weights, 1)
+#endif
 
     allocate(self % biases(self % output_size))
     self % biases = 0