SWDEV-498474 - APU: when total allocated is greater than 75% of invisible

Only when memory type is Local and the invisible memory is +ve
Should also fix SWDEV-490991

Change-Id: I78a4925a234ba90c63909bde5b7dc217568b4de3


[ROCm/clr commit: 7d763fb803]

projects/clr/rocclr/device/pal/paldevice.hpp

@@ -680,6 +680,13 @@ class Device : public NullDevice {
     return heaps_[Pal::GpuHeapInvisible];
   }
 
+  Pal::gpusize TotalAlloc() const {
+    Pal::gpusize local = allocedMem[Pal::GpuHeapLocal] - resourceCache().persistentCacheSize();
+    Pal::gpusize invisible = allocedMem[Pal::GpuHeapInvisible] - resourceCache().lclCacheSize();
+    Pal::gpusize total_alloced = local + invisible;
+    return total_alloced;
+  }
+
  private:
   static void PAL_STDCALL PalDeveloperCallback(void* pPrivateData, const Pal::uint32 deviceIndex,
                                                Pal::Developer::CallbackType type, void* pCbData);

projects/clr/rocclr/device/pal/palmemory.cpp

@@ -113,18 +113,21 @@ bool Memory::create(Resource::MemoryType memType, Resource::CreateParams* params
         dev().info().largeBar_) {
       memType = Persistent;
     }
+
+    if (dev().settings().apuSystem_) {
+      const Pal::GpuMemoryHeapProperties& invisibleHeap = dev().GetGpuHeapInvisible();
+      Pal::gpusize totalAlloc = dev().TotalAlloc();
+      if (invisibleHeap.logicalSize > 0 && memType == Local &&
+          (totalAlloc > (invisibleHeap.logicalSize * 0.75))) {
+        memType = RemoteUSWC;
+      }
+    }
     // Create a resource in PAL
     result = Resource::create(memType, params, forceLinear);
     if (!result) {
       size_t freeMemory[2];
       // if requested memory is greater than available then exit the loop
       dev().globalFreeMemory(freeMemory);
-
-      const Pal::GpuMemoryHeapProperties& invisibleHeap = dev().GetGpuHeapInvisible();
-      if (dev().settings().apuSystem_ && (size() > (invisibleHeap.logicalSize * 2))) {
-        memType = RemoteUSWC;
-        break;
-      }
       // Local to Persistent
       if (memoryType() == Local) {
         // For dgpu freeMemory[0] reports a sum of visible+invisible fb