P4 to Git Change 1503207 by wchau@wchau_OCL_boltzmann on 2018/01/12 14:51:08

SWDEV-140733 - OpenCL Runtime Support for Raven BU in ROCm
	- Fix OCLMemoryInfo issue of incorrect free memory info
	- Fix OCLPerfMemCreate failure due to memory leak

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocmemory.cpp#31 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.cpp#30 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/rocm/rocsettings.hpp#12 edit


[ROCm/clr commit: 074f37ac4d]

projects/clr/rocclr/runtime/device/rocm/rocmemory.cpp

@@ -579,17 +579,25 @@ void Buffer::destroy() {
     return;
   }
 #endif
-  if ((deviceMemory_ != nullptr) && (deviceMemory_ != owner()->getHostMem())) {
+  if (deviceMemory_ != nullptr) {
-    // if they are identical, the host pointer will be
+    if (deviceMemory_ != owner()->getHostMem()) {
-    // deallocated later on => avoid double deallocation
+      // if they are identical, the host pointer will be
-    if (isHostMemDirectAccess()) {
+      // deallocated later on => avoid double deallocation
-      if (memFlags & (CL_MEM_USE_HOST_PTR | CL_MEM_ALLOC_HOST_PTR)) {
+      if (isHostMemDirectAccess()) {
-        if (dev().agent_profile() != HSA_PROFILE_FULL) {
+        if (memFlags & (CL_MEM_USE_HOST_PTR | CL_MEM_ALLOC_HOST_PTR)) {
-          hsa_amd_memory_unlock(owner()->getHostMem());
+          if (dev().agent_profile() != HSA_PROFILE_FULL) {
+            hsa_amd_memory_unlock(owner()->getHostMem());
+          }
         }
+      } else {
+        dev().memFree(deviceMemory_, size());
+        const_cast<Device&>(dev()).updateFreeMemory(size(), true);
+      }
+    }
+    else if (dev().settings().apuSystem_) {
+      if (!(memFlags & (CL_MEM_USE_HOST_PTR | CL_MEM_ALLOC_HOST_PTR | CL_MEM_COPY_HOST_PTR))) {
+        dev().memFree(deviceMemory_, size());
       }
-    } else {
-      dev().memFree(deviceMemory_, size());
       const_cast<Device&>(dev()).updateFreeMemory(size(), true);
     }
   }
@@ -672,6 +680,10 @@ bool Buffer::create() {
 
       deviceMemory_ = dev().hostAlloc(size(), 1, false);
       owner()->setHostMem(deviceMemory_);
+
+      if (dev().settings().apuSystem_) {
+        const_cast<Device&>(dev()).updateFreeMemory(size(), false);
+      }
     }
     else {
       const_cast<Device&>(dev()).updateFreeMemory(size(), false);
@@ -938,6 +950,9 @@ bool Image::create() {
 
   if (originalDeviceMemory_ == nullptr) {
     originalDeviceMemory_ = dev().hostAlloc(alloc_size, 1, false);
+    if (dev().settings().apuSystem_) {
+      const_cast<Device&>(dev()).updateFreeMemory(alloc_size, false);
+    }
   }
   else {
     const_cast<Device&>(dev()).updateFreeMemory(alloc_size, false);

projects/clr/rocclr/runtime/device/rocm/rocsettings.cpp

@@ -72,6 +72,8 @@ Settings::Settings() {
 
   // Don't support Denormals for single precision by default
   singleFpDenorm_ = false;
+
+  apuSystem_ = false;
 }
 
 bool Settings::create(bool fullProfile, int gfxipVersion) {
@@ -124,6 +126,10 @@ bool Settings::create(bool fullProfile, int gfxipVersion) {
   }
 #endif  // WITH_LIGHTNING_COMPILER
 
+  if (gfxipVersion == 902) {
+    apuSystem_ = true;
+  }
+
   // Override current device settings
   override();
 

projects/clr/rocclr/runtime/device/rocm/rocsettings.hpp

@@ -29,7 +29,8 @@ class Settings : public device::Settings {
       uint stagedXferRead_ : 1;         //!< Uses a staged buffer read
       uint stagedXferWrite_ : 1;        //!< Uses a staged buffer write
       uint singleFpDenorm_ : 1;         //!< Support Single FP Denorm
-      uint reserved_ : 21;
+      uint apuSystem_ : 1;              //!< APU system
+      uint reserved_ : 20;
     };
     uint value_;
   };