[flang][runtime] Added Assign runtime to CUDA build closure.

flang/include/flang/ISO_Fortran_binding.h

@@ -127,8 +127,8 @@ namespace cfi_internal {
 template <typename T> struct FlexibleArray : T {
   RT_API_ATTRS T &operator[](int index) { return *(this + index); }
   const RT_API_ATTRS T &operator[](int index) const { return *(this + index); }
-  operator T *() { return this; }
-  operator const T *() const { return this; }
+  RT_API_ATTRS operator T *() { return this; }
+  RT_API_ATTRS operator const T *() const { return this; }
 };
 } // namespace cfi_internal
 #endif
@@ -182,19 +182,20 @@ template <> struct CdescStorage<0> : public CFI_cdesc_t {};
 #ifdef __cplusplus
 extern "C" {
 #endif
-void *CFI_address(const CFI_cdesc_t *, const CFI_index_t subscripts[]);
-int CFI_allocate(CFI_cdesc_t *, const CFI_index_t lower_bounds[],
+RT_API_ATTRS void *CFI_address(
+    const CFI_cdesc_t *, const CFI_index_t subscripts[]);
+RT_API_ATTRS int CFI_allocate(CFI_cdesc_t *, const CFI_index_t lower_bounds[],
     const CFI_index_t upper_bounds[], size_t elem_len);
 RT_API_ATTRS int CFI_deallocate(CFI_cdesc_t *);
 int CFI_establish(CFI_cdesc_t *, void *base_addr, CFI_attribute_t, CFI_type_t,
     size_t elem_len, CFI_rank_t, const CFI_index_t extents[]);
-int CFI_is_contiguous(const CFI_cdesc_t *);
+RT_API_ATTRS int CFI_is_contiguous(const CFI_cdesc_t *);
 RT_API_ATTRS int CFI_section(CFI_cdesc_t *, const CFI_cdesc_t *source,
     const CFI_index_t lower_bounds[], const CFI_index_t upper_bounds[],
     const CFI_index_t strides[]);
-int CFI_select_part(CFI_cdesc_t *, const CFI_cdesc_t *source,
+RT_API_ATTRS int CFI_select_part(CFI_cdesc_t *, const CFI_cdesc_t *source,
     size_t displacement, size_t elem_len);
-int CFI_setpointer(
+RT_API_ATTRS int CFI_setpointer(
     CFI_cdesc_t *, const CFI_cdesc_t *source, const CFI_index_t lower_bounds[]);
 #ifdef __cplusplus
 } // extern "C"

flang/include/flang/Runtime/api-attrs.h

@@ -45,7 +45,7 @@
 /*
  * RT_OFFLOAD_API_GROUP_BEGIN/END pair is placed around definitions
  * of functions that can be referenced in other modules of Flang
- * runtime. For OpenMP offload these functions are made "declare target"
+ * runtime. For OpenMP offload, these functions are made "declare target"
  * making sure they are compiled for the target even though direct
  * references to them from other "declare target" functions may not
  * be seen. Host-only functions should not be put in between these
@@ -54,6 +54,15 @@
 #define RT_OFFLOAD_API_GROUP_BEGIN RT_EXT_API_GROUP_BEGIN
 #define RT_OFFLOAD_API_GROUP_END RT_EXT_API_GROUP_END
 
+/*
+ * RT_OFFLOAD_VAR_GROUP_BEGIN/END pair is placed around definitions
+ * of variables (e.g. globals or static class members) that can be
+ * referenced in functions marked with RT_OFFLOAD_API_GROUP_BEGIN/END.
+ * For OpenMP offload, these variables are made "declare target".
+ */
+#define RT_OFFLOAD_VAR_GROUP_BEGIN RT_EXT_API_GROUP_BEGIN
+#define RT_OFFLOAD_VAR_GROUP_END RT_EXT_API_GROUP_END
+
 /*
  * RT_VAR_GROUP_BEGIN/END pair is placed around definitions
  * of module scope variables referenced by Flang runtime (directly

flang/include/flang/Runtime/assign.h

@@ -30,23 +30,23 @@ class Descriptor;
 
 extern "C" {
 // API for lowering assignment
-void RTNAME(Assign)(Descriptor &to, const Descriptor &from,
+void RTDECL(Assign)(Descriptor &to, const Descriptor &from,
     const char *sourceFile = nullptr, int sourceLine = 0);
 // This variant has no finalization, defined assignment, or allocatable
 // reallocation.
-void RTNAME(AssignTemporary)(Descriptor &to, const Descriptor &from,
+void RTDECL(AssignTemporary)(Descriptor &to, const Descriptor &from,
     const char *sourceFile = nullptr, int sourceLine = 0);
-void RTNAME(CopyOutAssign)(Descriptor &to, const Descriptor &from,
+void RTDECL(CopyOutAssign)(Descriptor &to, const Descriptor &from,
     bool skipToInit, const char *sourceFile = nullptr, int sourceLine = 0);
 // This variant is for assignments to explicit-length CHARACTER left-hand
 // sides that might need to handle truncation or blank-fill, and
 // must maintain the character length even if an allocatable array
 // is reallocated.
-void RTNAME(AssignExplicitLengthCharacter)(Descriptor &to,
+void RTDECL(AssignExplicitLengthCharacter)(Descriptor &to,
     const Descriptor &from, const char *sourceFile = nullptr,
     int sourceLine = 0);
 // This variant is assignments to whole polymorphic allocatables.
-void RTNAME(AssignPolymorphic)(Descriptor &to, const Descriptor &from,
+void RTDECL(AssignPolymorphic)(Descriptor &to, const Descriptor &from,
     const char *sourceFile = nullptr, int sourceLine = 0);
 } // extern "C"
 } // namespace Fortran::runtime

flang/include/flang/Runtime/descriptor.h

@@ -427,11 +427,13 @@ static_assert(sizeof(Descriptor) == sizeof(ISO::CFI_cdesc_t));
 template <int MAX_RANK = maxRank, bool ADDENDUM = false, int MAX_LEN_PARMS = 0>
 class alignas(Descriptor) StaticDescriptor {
 public:
+  RT_OFFLOAD_VAR_GROUP_BEGIN
   static constexpr int maxRank{MAX_RANK};
   static constexpr int maxLengthTypeParameters{MAX_LEN_PARMS};
   static constexpr bool hasAddendum{ADDENDUM || MAX_LEN_PARMS > 0};
   static constexpr std::size_t byteSize{
       Descriptor::SizeInBytes(maxRank, hasAddendum, maxLengthTypeParameters)};
+  RT_OFFLOAD_VAR_GROUP_END
 
   RT_API_ATTRS Descriptor &descriptor() {
     return *reinterpret_cast<Descriptor *>(storage_);

flang/runtime/CMakeLists.txt

@@ -150,10 +150,16 @@ option(FLANG_EXPERIMENTAL_CUDA_RUNTIME
 
 # List of files that are buildable for all devices.
 set(supported_files
+  ISO_Fortran_binding.cpp
+  assign.cpp
+  derived.cpp
   descriptor.cpp
+  stat.cpp
   terminator.cpp
+  tools.cpp
   transformational.cpp
   type-code.cpp
+  type-info.cpp
   )
 
 if (FLANG_EXPERIMENTAL_CUDA_RUNTIME)

flang/runtime/ISO_Fortran_binding.cpp

@@ -19,7 +19,9 @@
 namespace Fortran::ISO {
 extern "C" {
 
-void *CFI_address(
+RT_EXT_API_GROUP_BEGIN
+
+RT_API_ATTRS void *CFI_address(
     const CFI_cdesc_t *descriptor, const CFI_index_t subscripts[]) {
   char *p{static_cast<char *>(descriptor->base_addr)};
   const CFI_rank_t rank{descriptor->rank};
@@ -30,8 +32,9 @@ void *CFI_address(
   return p;
 }
 
-int CFI_allocate(CFI_cdesc_t *descriptor, const CFI_index_t lower_bounds[],
-    const CFI_index_t upper_bounds[], std::size_t elem_len) {
+RT_API_ATTRS int CFI_allocate(CFI_cdesc_t *descriptor,
+    const CFI_index_t lower_bounds[], const CFI_index_t upper_bounds[],
+    std::size_t elem_len) {
   if (!descriptor) {
     return CFI_INVALID_DESCRIPTOR;
   }
@@ -81,7 +84,7 @@ int CFI_allocate(CFI_cdesc_t *descriptor, const CFI_index_t lower_bounds[],
   return CFI_SUCCESS;
 }
 
-int CFI_deallocate(CFI_cdesc_t *descriptor) {
+RT_API_ATTRS int CFI_deallocate(CFI_cdesc_t *descriptor) {
   if (!descriptor) {
     return CFI_INVALID_DESCRIPTOR;
   }
@@ -101,7 +104,7 @@ int CFI_deallocate(CFI_cdesc_t *descriptor) {
   return CFI_SUCCESS;
 }
 
-int CFI_establish(CFI_cdesc_t *descriptor, void *base_addr,
+RT_API_ATTRS int CFI_establish(CFI_cdesc_t *descriptor, void *base_addr,
     CFI_attribute_t attribute, CFI_type_t type, std::size_t elem_len,
     CFI_rank_t rank, const CFI_index_t extents[]) {
   int cfiStatus{VerifyEstablishParameters(descriptor, base_addr, attribute,
@@ -121,7 +124,7 @@ int CFI_establish(CFI_cdesc_t *descriptor, void *base_addr,
   return CFI_SUCCESS;
 }
 
-int CFI_is_contiguous(const CFI_cdesc_t *descriptor) {
+RT_API_ATTRS int CFI_is_contiguous(const CFI_cdesc_t *descriptor) {
   CFI_index_t bytes = descriptor->elem_len;
   for (int j{0}; j < descriptor->rank; ++j) {
     if (bytes != descriptor->dim[j].sm) {
@@ -132,7 +135,7 @@ int CFI_is_contiguous(const CFI_cdesc_t *descriptor) {
   return 1;
 }
 
-int CFI_section(CFI_cdesc_t *result, const CFI_cdesc_t *source,
+RT_API_ATTRS int CFI_section(CFI_cdesc_t *result, const CFI_cdesc_t *source,
     const CFI_index_t lower_bounds[], const CFI_index_t upper_bounds[],
     const CFI_index_t strides[]) {
   CFI_index_t extent[CFI_MAX_RANK];
@@ -208,7 +211,7 @@ int CFI_section(CFI_cdesc_t *result, const CFI_cdesc_t *source,
   return CFI_SUCCESS;
 }
 
-int CFI_select_part(CFI_cdesc_t *result, const CFI_cdesc_t *source,
+RT_API_ATTRS int CFI_select_part(CFI_cdesc_t *result, const CFI_cdesc_t *source,
     std::size_t displacement, std::size_t elem_len) {
   if (!result || !source) {
     return CFI_INVALID_DESCRIPTOR;
@@ -243,7 +246,7 @@ int CFI_select_part(CFI_cdesc_t *result, const CFI_cdesc_t *source,
   return CFI_SUCCESS;
 }
 
-int CFI_setpointer(CFI_cdesc_t *result, const CFI_cdesc_t *source,
+RT_API_ATTRS int CFI_setpointer(CFI_cdesc_t *result, const CFI_cdesc_t *source,
     const CFI_index_t lower_bounds[]) {
   if (!result) {
     return CFI_INVALID_DESCRIPTOR;
@@ -285,5 +288,7 @@ int CFI_setpointer(CFI_cdesc_t *result, const CFI_cdesc_t *source,
   }
   return CFI_SUCCESS;
 }
+
+RT_EXT_API_GROUP_END
 } // extern "C"
 } // namespace Fortran::ISO

flang/runtime/assign-impl.h

@@ -17,7 +17,8 @@ class Terminator;
 // Note that if allocate object and source expression have the same rank, the
 // value of the allocate object becomes the value provided; otherwise the value
 // of each element of allocate object becomes the value provided (9.7.1.2(7)).
-void DoFromSourceAssign(Descriptor &, const Descriptor &, Terminator &);
+RT_API_ATTRS void DoFromSourceAssign(
+    Descriptor &, const Descriptor &, Terminator &);
 
 } // namespace Fortran::runtime
 #endif // FORTRAN_RUNTIME_ASSIGN_IMPL_H_