P4 to Git Change 1318397 by gandryey@gera-rcf-lnx on 2016/09/23 17:46:56

SWDEV-86035 - Add PAL backend to OpenCL
	- Fix more strict compilation rules in C++11

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/build/Makefile.pal#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palcounters.hpp#7 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#23 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevicegl.cpp#2 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palresource.cpp#12 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palsettings.hpp#5 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palthreadtrace.hpp#4 edit
... //depot/stg/opencl/drivers/opencl/runtime/device/pal/palvirtual.hpp#11 edit

rocclr/runtime/device/pal/palcounters.hpp

@@ -21,11 +21,11 @@ public:
     PalCounterReference(
         VirtualGPU&     gpu //!< Virtual GPU device object
         )
-        : perfExp_(nullptr)
-        , gpu_(gpu)
+        : gpu_(gpu)
+        , perfExp_(nullptr)
+        , layout_(nullptr)
         , memory_(nullptr)
         , cpuAddr_(nullptr)
-        , layout_(nullptr)
         , numExpCounters_(0) {}
 
     //! Get PAL counter

rocclr/runtime/device/pal/paldevice.cpp

@@ -264,17 +264,11 @@ void NullDevice::fillDeviceInfo(
 
     uint64_t localRAM = heaps[Pal::GpuHeapLocal].heapSize +
          heaps[Pal::GpuHeapInvisible].heapSize;
-#if defined(ATI_OS_LINUX)
-    info_.globalMemSize_   =
-        (static_cast<cl_ulong>(std::min(GPU_MAX_HEAP_SIZE, 100u)) *
-        // globalMemSize is the actual available size for app on Linux
-        // Because Linux base driver doesn't support paging
-        static_cast<cl_ulong>(memInfo.cardMemAvailableBytes + memInfo.cardExtMemAvailableBytes) / 100u);
-#else
+
     info_.globalMemSize_   =
         (static_cast<cl_ulong>(std::min(GPU_MAX_HEAP_SIZE, 100u)) *
         static_cast<cl_ulong>(localRAM) / 100u);
-#endif
+
     if (settings().apuSystem_) {
         info_.globalMemSize_   +=
             (static_cast<cl_ulong>(heaps[Pal::GpuHeapGartUswc].heapSize) * Mi * 75)/100;

rocclr/runtime/device/pal/paldevicegl.cpp

@@ -76,9 +76,6 @@ Device::initGLInteropPrivateExt(void* GLplatformContext, void* GLdeviceContext)
 
     if (!glXBeginCLInteropAMD || !glXEndCLInteropAMD || !glXResourceAttachAMD ||
         !glXResourceDetachAMD
-#ifndef BRAHMA
-        || !glXGetContextMVPUInfoAMD
-#endif
         ) {
         return false;
     }
@@ -132,22 +129,7 @@ Device::glCanInterop(void* GLplatformContext, void* GLdeviceContext) const
             ((1 << properties().gpuIndex) == glChainBitMask);
     }
 #else
-#ifdef BRAHMA
     canInteroperate = true;
-#else
-    GLuint glDeviceId = 0 ;
-    GLuint glChainMask = 0 ;
-    GLXContext ctx = (GLXContext)GLplatformContext;
-    
-    if (glXGetContextMVPUInfoAMD(ctx, &glDeviceId, &glChainMask)) {
-        // we allow intoperability only with GL context reside on a single GPU
-        canInteroperate =
-            (properties().deviceId == glDeviceId) &&
-            ((1 << properties().gpuIndex) == glChainBitMask);
-
-        }
-    }
-#endif
 #endif
     return canInteroperate;
 }
@@ -219,7 +201,7 @@ Device::resGLAssociate(
 #ifdef ATI_OS_LINUX
     GLXContext ctx = (GLXContext)GLContext;
     if (glXResourceAttachAMD(ctx, &hRes, &hData)) {
-        attribs.dynamicSharedBufferID = hData->sharedBufferID;
+        //attribs.dynamicSharedBufferID = hData->sharedBufferID;
         status = true;
     }
 #else

rocclr/runtime/device/pal/palresource.cpp

@@ -283,6 +283,9 @@ Resource::~Resource()
     case Local:
         heap = Pal::GpuHeapInvisible;
         break;
+    default:
+        heap = Pal::GpuHeapLocal;
+        break;
     }
     if ((memRef_ != nullptr) && (heap != Pal::GpuHeapCount)) {
         // Update free memory size counters
@@ -379,6 +382,9 @@ Resource::memTypeToHeap(Pal::GpuMemoryCreateInfo* createInfo)
         createInfo->heaps[0] = Pal::GpuHeapInvisible;
         createInfo->heaps[1] = Pal::GpuHeapLocal;
         break;
+    default:
+        createInfo->heaps[0] = Pal::GpuHeapLocal;
+        break;    
     }
 }
 

rocclr/runtime/device/pal/palsettings.hpp

@@ -5,7 +5,7 @@
 
 #include "top.hpp"
 #include "library.hpp"
-#include "inc\core\palDevice.h"
+#include "inc/core/palDevice.h"
 
 /*! \addtogroup pal PAL Resource Implementation
  *  @{

rocclr/runtime/device/pal/palthreadtrace.hpp

@@ -22,10 +22,11 @@ public:
     PalThreadTraceReference(
         VirtualGPU&     gpu             //!< Virtual GPU device object
         )
-        : perfExp_(nullptr)
-        , gpu_(gpu)
+        : gpu_(gpu)
+        , perfExp_(nullptr)
+        , layout_(nullptr)
         , memory_(nullptr)
-        , layout_(nullptr) {}
+        {}
 
     //! Get PAL thread race object
     Pal::IPerfExperiment* iPerf() const { return perfExp_; }
@@ -70,16 +71,14 @@ public:
         )
         : gpuDevice_(device)
         , palRef_(palRef)
-        , memObj_(memObjs)
         , numSe_(numSe)
-    {
-
-    }
+        , memObj_(memObjs)
+        {}
 
     //! Destructor for the GPU ThreadTrace object
     virtual ~ThreadTrace();
 
-    //! Creates the current object
+    //! Creates the current object4
     bool create();
 
     // Populate ThreadTrace memory with PerfExperiment memory

rocclr/runtime/device/pal/palvirtual.hpp

@@ -54,7 +54,7 @@ public:
             );
 
         Queue(Pal::IDevice* palDev)
-            : iDev_(palDev), iQueue_(NULL),
+            : iQueue_(NULL), iDev_(palDev),
             cmdBufIdSlot_(StartCmdBufIdx), cmdBufIdCurrent_(StartCmdBufIdx),
             cmbBufIdRetired_(0), cmdCnt_(0), vlAlloc_(64 * Ki)
         {