P4 to Git Change 1458325 by gandryey@gera-w8 on 2017/09/13 11:33:32

SWDEV-132238 - [CQE OCL][Vega10][DTB-Blocker][QR] 'Allocation (Single)' test of WF Conformance is failing; Faulty CL# 1451444
	- Disable reporting extra HBCC memory by default. Reporting extra memory can be reenabled with GPU_ADD_HBCC_SIZE=1

Affected files ...

... //depot/stg/opencl/drivers/opencl/runtime/device/pal/paldevice.cpp#60 edit
... //depot/stg/opencl/drivers/opencl/runtime/utils/flags.hpp#280 edit

rocclr/runtime/device/pal/paldevice.cpp

@@ -213,7 +213,8 @@ bool NullDevice::create(Pal::AsicRevision asicRevision, Pal::GfxIpLevel ipLevel,
 
   // Report 512MB for all offline devices
   Pal::GpuMemoryHeapProperties heaps[Pal::GpuHeapCount];
-  heaps[Pal::GpuHeapLocal].heapSize = 512 * Mi;
+  heaps[Pal::GpuHeapLocal].heapSize = 
+  heaps[Pal::GpuHeapLocal].physicalHeapSize = 512 * Mi;
 
   Pal::WorkStationCaps wscaps = {};
 
@@ -301,7 +302,12 @@ void NullDevice::fillDeviceInfo(const Pal::DeviceProperties& palProp,
     info_.globalMemCacheType_ = CL_NONE;
   }
 
-  uint64_t localRAM = heaps[Pal::GpuHeapLocal].heapSize + heaps[Pal::GpuHeapInvisible].heapSize;
+  uint64_t localRAM;
+  if (GPU_ADD_HBCC_SIZE) {
+    localRAM = heaps[Pal::GpuHeapLocal].heapSize + heaps[Pal::GpuHeapInvisible].heapSize;
+  } else {
+    localRAM = heaps[Pal::GpuHeapLocal].physicalHeapSize + heaps[Pal::GpuHeapInvisible].physicalHeapSize;
+  }
 
   info_.globalMemSize_ = (static_cast<cl_ulong>(std::min(GPU_MAX_HEAP_SIZE, 100u)) *
                           static_cast<cl_ulong>(localRAM) / 100u);
@@ -316,8 +322,13 @@ void NullDevice::fillDeviceInfo(const Pal::DeviceProperties& palProp,
   }
 
   // Find the largest heap form FB memory
-  info_.maxMemAllocSize_ = std::max(cl_ulong(heaps[Pal::GpuHeapLocal].heapSize),
-                                    cl_ulong(heaps[Pal::GpuHeapInvisible].heapSize));
+  if (GPU_ADD_HBCC_SIZE) {
+    info_.maxMemAllocSize_ = std::max(cl_ulong(heaps[Pal::GpuHeapLocal].heapSize),
+      cl_ulong(heaps[Pal::GpuHeapInvisible].heapSize));
+  } else {
+    info_.maxMemAllocSize_ = std::max(cl_ulong(heaps[Pal::GpuHeapLocal].physicalHeapSize),
+      cl_ulong(heaps[Pal::GpuHeapInvisible].physicalHeapSize));
+  }
 
 #if defined(ATI_OS_WIN)
   if (settings().apuSystem_) {

rocclr/runtime/utils/flags.hpp

@@ -187,6 +187,8 @@ release(bool, GPU_ANALYZE_HANG, false,                                        \
         "1 = Enables GPU hang analysis")                                      \
 release(uint, GPU_MAX_REMOTE_MEM_SIZE, 2,                                     \
         "Maximum size (in Ki) that allows device memory substitution with system") \
+release(bool, GPU_ADD_HBCC_SIZE, false,                                       \
+        "Add HBCC size to the reported device memory")                        \
 release_on_stg(uint, GPU_WAVE_LIMIT_CU_PER_SH, 0,                             \
         "Assume the number of CU per SH for wave limiter")                    \
 release_on_stg(uint, GPU_WAVE_LIMIT_MAX_WAVE, 10,                             \