Merge branch 'master' of https://github.com/Microsoft/CNTK into fseide/s2s

Makefile

@@ -171,7 +171,7 @@ ifeq ("$(BUILDTYPE)","debug")
 
   CXXFLAGS += -g
   LDFLAGS += -rdynamic
-  COMMON_FLAGS += -D_DEBUG
+  COMMON_FLAGS += -D_DEBUG -DNO_SYNC
   CUFLAGS += -O0 -g -use_fast_math -lineinfo  $(GENCODE_FLAGS)
 endif
 
@@ -184,7 +184,7 @@ ifeq ("$(BUILDTYPE)","release")
 
   CXXFLAGS += -g -O4
   LDFLAGS += -rdynamic
-  COMMON_FLAGS += -DNDEBUG
+  COMMON_FLAGS += -DNDEBUG -DNO_SYNC
   CUFLAGS += -O3 -g -use_fast_math -lineinfo $(GENCODE_FLAGS)
 endif
 

Source/Math/GPUSparseMatrix.cu

@@ -402,7 +402,7 @@ void GPUSparseMatrix<ElemType>::ChangeDeviceTo(DEVICEID_TYPE to_id)
     }
     else
     {
-        ElemType* d_dst = TracingGPUMemoryAllocator::Allocate<ElemType>(to_id, m_totalBufferSizeAllocated / sizeof(ElemType));
+        ElemType* d_dst = (ElemType*)TracingGPUMemoryAllocator::Allocate<char>(to_id, m_totalBufferSizeAllocated);
 
         // first try peer access
         int canAccessPeer = false;
@@ -625,7 +625,7 @@ void GPUSparseMatrix<ElemType>::Reshape(const size_t numRows, const size_t numCo
 
     size_t bufferSizeNeeded = BufferSizeNeeded(numRows, numCols, m_elemSizeAllocated, m_format);
 
-    ElemType* pArray = TracingGPUMemoryAllocator::Allocate<ElemType>(m_computeDevice, bufferSizeNeeded / sizeof(ElemType));
+    ElemType* pArray = (ElemType*)TracingGPUMemoryAllocator::Allocate<char>(m_computeDevice, bufferSizeNeeded);
 
     if (m_pArray != nullptr)
     {
@@ -683,7 +683,7 @@ void GPUSparseMatrix<ElemType>::Resize(const size_t numRows, const size_t numCol
 
     if (reallocate)
     {
-        ElemType* pArray = TracingGPUMemoryAllocator::Allocate<ElemType>(m_computeDevice, bufferSizeNeeded / sizeof(ElemType));
+        ElemType* pArray = (ElemType*)TracingGPUMemoryAllocator::Allocate<char>(m_computeDevice, bufferSizeNeeded);
 
         if (m_pArray != nullptr)
         {