[OpenMP] Avoid zero size copies to the device

[jdoerfert]

Summary

This unblocks one of the XFAIL tests for AMD, though we need to work
around the missing printf still.

[jhuber6]

LG

openmp/libomptarget/src/device.cpp

325

nit. size != 0 makes the intent clearer.

openmp/libomptarget/src/device.cpp

if (notifyDataMapped(HstPtrBegin, Size))
               nullptr /* TargetPointer */};
   } else {
     // This entry is not present and we did not create a new entry for it.
     IsPresent = false;
   }
 
   // If the target pointer is valid, and we need to transfer data, issue the
   // data transfer.
-  if (TargetPointer && !IsHostPtr && HasFlagTo && (IsNew || HasFlagAlways)) {
+  if (TargetPointer && !IsHostPtr && HasFlagTo && (IsNew || HasFlagAlways) &&
+      Size != 0) {
     // Lock the entry before releasing the mapping table lock such that another
     // thread that could issue data movement will get the right result.
     std::lock_guard<decltype(*Entry)> LG(*Entry);
     // Release the mapping table lock right after the entry is locked.
     HDTTMap.destroy();
 
     DP("Moving %" PRId64 " bytes (hst:" DPxMOD ") -> (tgt:" DPxMOD ")\n", Size,
        DPxPTR(HstPtrBegin), DPxPTR(TargetPointer));

openmp/libomptarget/src/omptarget.cpp

for (int32_t I = ArgNum - 1; I >= 0; --I) {
     // present in the device data environment on exit from the region then the
     // list item is ignored."
     if (!TPR.isPresent())
       continue;
 
     // Move data back to the host
     const bool HasAlways = ArgTypes[I] & OMP_TGT_MAPTYPE_ALWAYS;
     const bool HasFrom = ArgTypes[I] & OMP_TGT_MAPTYPE_FROM;
-    if (HasFrom && (HasAlways || IsLast) && !IsHostPtr) {
+    if (HasFrom && (HasAlways || IsLast) && !IsHostPtr && DataSize != 0) {
       DP("Moving %" PRId64 " bytes (tgt:" DPxMOD ") -> (hst:" DPxMOD ")\n",
          DataSize, DPxPTR(TgtPtrBegin), DPxPTR(HstPtrBegin));
 
       std::lock_guard<decltype(*TPR.Entry)> LG(*TPR.Entry);
       // Wait for any previous transfer if an event is present.
       if (void *Event = TPR.Entry->getEvent()) {
         if (Device.waitEvent(Event, AsyncInfo) != OFFLOAD_SUCCESS) {
           REPORT("Failed to wait for event " DPxMOD ".\n", DPxPTR(Event));

openmp/libomptarget/test/mapping/data_member_ref.cpp

 // RUN: %libomptarget-compilexx-run-and-check-generic
 
-// Wrong results on amdgpu
-// XFAIL: amdgcn-amd-amdhsa
-
 #include <stdio.h>
 
 struct View {
   int Data;
 };
 
 struct ViewPtr {
   int *Data;
 };
 
 template <typename T> struct Foo {
   Foo(T &V) : VRef(V) {}
   T &VRef;
 };
 
 int main() {
   View V;
   V.Data = 123456;
   Foo<View> Bar(V);
   ViewPtr V1;
   int Data = 123456;
   V1.Data = &Data;
   Foo<ViewPtr> Baz(V1);
+  int D1, D2;
 
   // CHECK: Host 123456.
   printf("Host %d.\n", Bar.VRef.Data);
-#pragma omp target map(Bar.VRef)
+#pragma omp target map(Bar.VRef) map(from : D1, D2)
   {
     // CHECK: Device 123456.
-    printf("Device %d.\n", Bar.VRef.Data);
+    D1 = Bar.VRef.Data;
+    printf("Device %d.\n", D1);
     V.Data = 654321;
     // CHECK: Device 654321.
-    printf("Device %d.\n", Bar.VRef.Data);
+    D2 = Bar.VRef.Data;
+    printf("Device %d.\n", D2);
   }
+  printf("Device %d.\n", D1);
+  printf("Device %d.\n", D2);
   // CHECK: Host 654321 654321.
   printf("Host %d %d.\n", Bar.VRef.Data, V.Data);
   V.Data = 123456;
   // CHECK: Host 123456.
   printf("Host %d.\n", Bar.VRef.Data);
-#pragma omp target map(Bar) map(Bar.VRef)
+#pragma omp target map(Bar) map(Bar.VRef) map(from : D1, D2)
   {
     // CHECK: Device 123456.
-    printf("Device %d.\n", Bar.VRef.Data);
+    D1 = Bar.VRef.Data;
+    printf("Device %d.\n", D1);
     V.Data = 654321;
     // CHECK: Device 654321.
-    printf("Device %d.\n", Bar.VRef.Data);
+    D2 = Bar.VRef.Data;
+    printf("Device %d.\n", D2);
   }
+  printf("Device %d.\n", D1);
+  printf("Device %d.\n", D2);
   // CHECK: Host 654321 654321.
   printf("Host %d %d.\n", Bar.VRef.Data, V.Data);
   // CHECK: Host 123456.
   printf("Host %d.\n", *Baz.VRef.Data);
-#pragma omp target map(*Baz.VRef.Data)
+#pragma omp target map(*Baz.VRef.Data) map(from : D1, D2)
   {
     // CHECK: Device 123456.
-    printf("Device %d.\n", *Baz.VRef.Data);
+    D1 = *Baz.VRef.Data;
+    printf("Device %d.\n", D1);
     *V1.Data = 654321;
     // CHECK: Device 654321.
-    printf("Device %d.\n", *Baz.VRef.Data);
+    D2 = *Baz.VRef.Data;
+    printf("Device %d.\n", D2);
   }
+  printf("Device %d.\n", D1);
+  printf("Device %d.\n", D2);
   // CHECK: Host 654321 654321 654321.
   printf("Host %d %d %d.\n", *Baz.VRef.Data, *V1.Data, Data);
   return 0;
 }