Fixes for omp declare target feature

test/offloading/amdgpu/swdev288446.F90

@@ -0,0 +1,49 @@
+subroutine _compute_dev(sum_dev, a_dev, b_dev, c_dev)
+    integer i,j,k
+    REAL(8), dimension(100,100,100), intent(in) :: b_dev, c_dev
+    REAL(8), dimension(100,100,100), intent(out) :: a_dev
+    REAL(8), intent(inout) :: sum_dev
+    integer :: nsize = 100
+
+!$omp declare target
+!$omp parallel do
+    do i=1,nsize
+        do j=1,nsize
+            do k=1,nsize
+                a_dev(i,j,k) = b_dev(i,j,k) * c_dev(i,j,k) * i * nsize*nsize + j * nsize + k
+            end do
+        end do
+    end do
+!$xomp end target
+end subroutine _compute_dev
+
+program main
+    integer :: nsize
+    REAL(8) :: sum_dev = 0
+    REAL(8), dimension(100,100,100) :: a_dev, b_dev, c_dev
+    integer :: nsize = 100
+!$omp declare target(_compute_dev)
+    do i=1,nsize
+        do j=1,nsize
+            do k=1,nsize
+                b_dev(i,j,k) = 1
+                c_dev(i,j,k) = 2
+            end do
+        end do
+    end do
+
+!$omp target update to(b_dev,c_dev,sum_dev,nsize)
+!$omp target
+    CALL _compute_dev(sum_dev, a_dev, b_dev, c_dev)
+!$omp end target
+!$omp target update from(a_dev,sum_dev)
+
+    do i=1,3
+        do j=1,3
+            do k=1,3
+                print *, a_dev(i,j,k)
+            end do
+        end do
+    end do
+end program main
+

tools/flang2/flang2exe/exp_rte.cpp

@@ -2157,7 +2157,7 @@ exp_end(ILM *ilmp, int curilm, bool is_func)
 #ifdef OMP_OFFLOAD_AMD
   int ilix;
   if (flg.omptarget && !is_func) {
-    if (XBIT(232, 0x40) && gbl.ompaccel_intarget) {
+    if (XBIT(232, 0x40) && gbl.ompaccel_intarget && is_gpu_output_file() ) {
       ilix = ll_make_kmpc_spmd_kernel_deinit_v2();
       iltb.callfg = 1;
       chk_block(ilix);

tools/flang2/flang2exe/llutil.cpp

@@ -1783,7 +1783,12 @@ set_llasm_output_file(FILE *fd)
 FILE *get_llasm_output_file() {
   return LLVMFIL;
 }
+
 #endif // OMP_OFFLOAD_LLVM
+
+bool is_gpu_output_file() {
+  return (LLVMFIL == gbl.ompaccfile);
+}
 // AOCC End
 
 void

tools/flang2/flang2exe/llutil.h

@@ -1620,11 +1620,16 @@ void use_cpu_output_file(void);
  */
 void use_gpu_output_file(void);
 
+
 // AOCC Begin
 /**
   \brief Return the current output file
  */
 FILE *get_llasm_output_file();
+/**
+  \brief return true if gpu file is currently written
+ */
+bool is_gpu_output_file();
 
 #ifdef OMP_OFFLOAD_AMD
 /**

tools/flang2/flang2exe/main.cpp

@@ -370,8 +370,9 @@ process_input(char *argv0, bool *need_cuda_constructor)
           gbl.ompaccel_isdevice = true;
           schedule();
           gbl.ompaccel_isdevice = orig;
-	  if (flg.omptarget && !gbl.ompaccel_isdevice)
+          if (flg.omptarget && !gbl.ompaccel_isdevice) {
             schedule();
+          }
         } else {
           schedule();
         }

tools/flang2/flang2exe/ompaccel.cpp

@@ -2748,7 +2748,7 @@ exp_ompaccel_mploop(ILM *ilmp, int curilm)
   case KMP_DISTRIBUTE_STATIC:
   case KMP_DISTRIBUTE_STATIC_CHUNKED_CHUNKONE: // AOCC
     // AOCC begin
-    if (flg.x86_64_omptarget) {
+    if (flg.x86_64_omptarget || !is_gpu_output_file() ) {
       ili = ll_make_kmpc_for_static_init(&loop_args);
     // AOCC end
     } else {
@@ -3709,6 +3709,7 @@ ompaccel_set_default_map(int maptype) {
 void
 ompaccel_set_target_declare() {
   OMPACCFUNCDEVP(gbl.currsub, 1);
+  gbl.ompaccel_intarget = true;
 }
 // AOCC End
 #endif

tools/flang2/flang2exe/ompaccel_x86.cpp

@@ -174,18 +174,20 @@ void ompaccel_x86_fix_arg_types(SPTR func_sptr) {
   // Remember all the reduction symbols of func_sptr so that we can blacklist
   // them during the type update.
   std::set<SPTR> reduc_syms;
-  for (int i = 0; i < tinfo->n_reduction_symbols; i++) {
-    OMPACCEL_RED_SYM *reduction_sym = &(tinfo->reduction_symbols[i]);
-    OMPACCEL_SYM *ompaccel_sym = get_ompaccel_sym_for(reduction_sym, tinfo);
+  if (tinfo) {
+    for (int i = 0; i < tinfo->n_reduction_symbols; i++) {
+      OMPACCEL_RED_SYM *reduction_sym = &(tinfo->reduction_symbols[i]);
+      OMPACCEL_SYM *ompaccel_sym = get_ompaccel_sym_for(reduction_sym, tinfo);
 
-    if (!ompaccel_sym)
-      continue;
-    SPTR device_sym = ompaccel_sym->device_sym;
+      if (!ompaccel_sym)
+        continue;
+      SPTR device_sym = ompaccel_sym->device_sym;
 
-    if (PASSBYVALG(device_sym))
-      continue;
+      if (PASSBYVALG(device_sym))
+        continue;
 
-    reduc_syms.insert(device_sym);
+      reduc_syms.insert(device_sym);
+    }
   }
 
   for (int i = 0; i < func_paramct; i++) {