P4 to Git Change 1410373 by gandryey@gera-w8 on 2017/05/16 17:16:52

SWDEV-120180 - [amdgpu-pro] OpenCL support for SSG
	- Add initial support of DGMA memory under ROCr backend.
	 - The implementation requires amdgpu-pro stack initialization and memory allocation.
	 - An interop with HSA device is created for ROCr access

Affected files ...

... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/Makefile#10 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/build/Makefile.api#153 edit
... //depot/stg/opencl/drivers/opencl/api/opencl/amdocl/cl_lqdflash_amd.cpp#18 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/device.hpp#285 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/gpu/gpusettings.cpp#351 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/build/Makefile.oclrocm#16 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/pro/lnxheaders.h#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/pro/prodevice.cpp#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/pro/prodevice.hpp#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/pro/prodriver.hpp#1 add
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.cpp#51 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocdevice.hpp#21 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#19 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#18 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.cpp#38 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocvirtual.hpp#11 edit
... //depot/stg/opencl/drivers/opencl/runtime/platform/memory.hpp#101 edit
... //depot/stg/opencl/drivers/opencl/runtime/runtimedefs#41 edit

rocclr/runtime/device/rocm/rocmemory.cpp

@@ -20,6 +20,7 @@
 #include "platform/memory.hpp"
 #include "platform/sampler.hpp"
 #include "amdocl/cl_gl_amd.hpp"
+#include "pro/prodriver.hpp"
 
 namespace roc {
 
@@ -548,7 +549,12 @@ void Buffer::destroy() {
   }
 
   const cl_mem_flags memFlags = owner()->getMemFlags();
-
+#ifdef WITH_AMDGPU_PRO
+  if ((memFlags & CL_MEM_USE_PERSISTENT_MEM_AMD) && dev().ProEna()) {
+    dev().iPro().FreeDmaBuffer(deviceMemory_);
+    return;
+  }
+#endif
   if ((deviceMemory_ != nullptr) && (deviceMemory_ != owner()->getHostMem())) {
     // if they are identical, the host pointer will be
     // deallocated later on => avoid double deallocation
@@ -611,6 +617,20 @@ bool Buffer::create() {
 
   // Allocate backing storage in device local memory unless UHP or AHP are set
   const cl_mem_flags memFlags = owner()->getMemFlags();
+
+#ifdef WITH_AMDGPU_PRO
+  if ((memFlags & CL_MEM_USE_PERSISTENT_MEM_AMD) && dev().ProEna()) {
+    void* host_ptr = nullptr;
+    deviceMemory_ = dev().iPro().AllocDmaBuffer(dev().getGpuAgents()[0], size(), &host_ptr);
+    if (deviceMemory_ == nullptr) {
+      return false;
+    }
+    flags_ |= HostMemoryDirectAccess;
+    owner()->setHostMem(host_ptr);
+    return true;
+  }
+#endif
+
   if (!(memFlags & (CL_MEM_USE_HOST_PTR | CL_MEM_ALLOC_HOST_PTR))) {
     deviceMemory_ = dev().deviceLocalAlloc(size());