P4 to Git Change 1451494 by gandryey@gera-w8 on 2017/08/24 17:34:58

SWDEV-129129 - [[CQE OCL][Vega vs Fiji] Upto 12% Performance drop observed on VEGA10 compared to FIJI while running BlackMagic Davinci Resolve
	- Remove some debug logic (resident_ field) from CL#1451293. It caused some CPU overhead
	- Use a template for waifForFence() to avoid some calls when unnecessary

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palmemory.cpp#15 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#31 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.hpp#14 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.cpp#53 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#29 edit

rocclr/runtime/device/pal/palresource.cpp

@@ -46,6 +46,7 @@ GpuMemoryReference* GpuMemoryReference::Create(const Device& dev,
   }
   // Update free memory size counters
   const_cast<Device&>(dev).updateFreeMemory(createInfo.heaps[0], createInfo.size, false);
+  dev.addResource(memRef);
   return memRef;
 }
 
@@ -68,6 +69,7 @@ GpuMemoryReference* GpuMemoryReference::Create(const Device& dev,
   }
   // Update free memory size counters
   const_cast<Device&>(dev).updateFreeMemory(Pal::GpuHeap::GpuHeapGartCacheable, createInfo.size, false);
+  dev.addResource(memRef);
   return memRef;
 }
 
@@ -90,6 +92,7 @@ GpuMemoryReference* GpuMemoryReference::Create(const Device& dev,
   // Update free memory size counters
   const_cast<Device&>(dev).updateFreeMemory(Pal::GpuHeap::GpuHeapGartCacheable, createInfo.size,
                                             false);
+  dev.addResource(memRef);
   return memRef;
 }
 
@@ -111,6 +114,7 @@ GpuMemoryReference* GpuMemoryReference::Create(const Device& dev,
       return nullptr;
     }
   }
+  dev.addResource(memRef);
   return memRef;
 }
 
@@ -137,11 +141,12 @@ GpuMemoryReference* GpuMemoryReference::Create(const Device& dev,
       return nullptr;
     }
   }
+  dev.addResource(memRef);
   return memRef;
 }
 
 GpuMemoryReference::GpuMemoryReference(const Device& dev)
-  : gpuMem_(nullptr), cpuAddress_(nullptr), events_(dev.numOfVgpus()), device_(dev), gpu_(nullptr), resident_(0) {}
+  : gpuMem_(nullptr), cpuAddress_(nullptr), events_(dev.numOfVgpus()), device_(dev), gpu_(nullptr) {}
 
 GpuMemoryReference::~GpuMemoryReference() {
   if (gpu_ == nullptr) {
@@ -162,10 +167,6 @@ GpuMemoryReference::~GpuMemoryReference() {
     device_.vgpus()[0]->releaseMemory(this, &events_[0]);
   }
 
-  if (resident_ != 0) {
-    LogError("Residency counter isn't 0 on memory destroy!");
-  }
-
   {
     amd::ScopedLock lk(device_.lockPAL());
     if (cpuAddress_ != nullptr) {