Fixing compilation of CPUMatrix on Windows: heap overflow fixed

Makefile

@@ -360,7 +360,8 @@ MATH_SRC =\
 	$(SOURCEDIR)/Math/BatchNormalizationEngine.cpp \
 	$(SOURCEDIR)/Math/BlockHandlerSSE.cpp \
 	$(SOURCEDIR)/Math/CUDAPageLockedMemAllocator.cpp \
-	$(SOURCEDIR)/Math/CPUMatrix.cpp \
+	$(SOURCEDIR)/Math/CPUMatrixFloat.cpp \
+	$(SOURCEDIR)/Math/CPUMatrixDouble.cpp \
 	$(SOURCEDIR)/Math/CPURNGHandle.cpp \
 	$(SOURCEDIR)/Math/CPUSparseMatrix.cpp \
 	$(SOURCEDIR)/Math/ConvolutionEngine.cpp \

Source/Math/CPUMatrixDouble.cpp

@@ -0,0 +1,12 @@
+//
+// Copyright (c) Microsoft. All rights reserved.
+// Licensed under the MIT license. See LICENSE.md file in the project root for full license information.
+//
+#include "stdafx.h"
+#include "CPUMatrixImpl.h"
+
+namespace Microsoft { namespace MSR { namespace CNTK {
+
+    // explicit instantiations, due to CPUMatrix being too big and causing VS2015 cl crash.
+    template class MATH_API CPUMatrix<double>;
+}}}

Source/Math/CPUMatrixFloat.cpp

@@ -0,0 +1,24 @@
+//
+// Copyright (c) Microsoft. All rights reserved.
+// Licensed under the MIT license. See LICENSE.md file in the project root for full license information.
+//
+#include "stdafx.h"
+#include "CPUMatrixImpl.h"
+
+namespace Microsoft { namespace MSR { namespace CNTK {
+
+    int MATH_API TracingGPUMemoryAllocator::m_traceLevel = 0;
+
+    void TracingGPUMemoryAllocator::SetTraceLevel(int traceLevel)
+    {
+        m_traceLevel = traceLevel;
+    }
+
+    bool TracingGPUMemoryAllocator::IsTraceEnabled()
+    {
+        return (m_traceLevel > 0);
+    }
+
+    // explicit instantiations, due to CPUMatrix being too big and causing VS2015 cl crash.
+    template class MATH_API CPUMatrix<float>;
+}}}

Source/Math/CPUMatrixImpl.h

@@ -2,10 +2,11 @@
 // Copyright (c) Microsoft. All rights reserved.
 // Licensed under the MIT license. See LICENSE.md file in the project root for full license information.
 //
-// CPUMatrix.cpp : full implementation of all matrix functions on the CPU side
+// CPUMatrix.h : template implementation of all matrix functions on the CPU side
 //
 
-#include "stdafx.h"
+#pragma once
+
 #include "Basics.h"
 #include "File.h"
 
@@ -66,18 +67,6 @@
 #define IDX2C(i, j, ld) (((j) * (ld)) + (i)) // 0 based indexing
 namespace Microsoft { namespace MSR { namespace CNTK {
 
-int MATH_API TracingGPUMemoryAllocator::m_traceLevel = 0;
-
-void TracingGPUMemoryAllocator::SetTraceLevel(int traceLevel)
-{
-    m_traceLevel = traceLevel;
-}
-
-bool TracingGPUMemoryAllocator::IsTraceEnabled()
-{
-    return (m_traceLevel > 0);
-}
-
 #pragma region Helpful Enum Definitions
 enum class MatrixOrder
 {
@@ -5923,7 +5912,7 @@ CPUMatrix<ElemType>& CPUMatrix<ElemType>::AssignElementProductOfWithShift(const
 #pragma endregion Static BLAS Functions
 
 // 'double' version of LogAdd
-double LogAddD(double x, double y)
+inline double LogAddD(double x, double y)
 {
     return LogAdd(x, y);
 }
@@ -7184,12 +7173,6 @@ void CPUMatrix<ElemType>::TensorArgOp(const CPUMatrix<ElemType>& a, ElementWiseO
     }
 }
 
-// =======================================================================
-// explicit instantiations
-// =======================================================================
-template class MATH_API CPUMatrix<float>;
-template class MATH_API CPUMatrix<double>;
-
 // We use Matrix<char> as the backing store for QuantizedMatrix
 // Let's explicitly instantiate the methods we need for that purpose
 template CPUMatrix<char>::CPUMatrix(const size_t numRows, const size_t numCols);

Source/Math/Math.vcxproj

@@ -194,12 +194,15 @@
     <ClInclude Include="QuantizedMatrix.h" />
     <ClInclude Include="stdafx.h" />
     <ClInclude Include="targetver.h" />
+    <ClInclude Include="CPUMatrixImpl.h" />
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="BatchNormalizationEngine.cpp" />
     <ClCompile Include="BlockHandlerAVX.cpp" />
     <ClCompile Include="BlockHandlerSSE.cpp" />
     <ClCompile Include="ConvolutionEngine.cpp" />
+    <ClCompile Include="CPUMatrixDouble.cpp" />
+    <ClCompile Include="CPUMatrixFloat.cpp" />
     <ClCompile Include="CPURNGHandle.cpp" />
     <ClCompile Include="CPUSparseMatrix.cpp" />
     <ClCompile Include="CUDAPageLockedMemAllocator.cpp" />
@@ -209,7 +212,6 @@
       <PrecompiledHeader>
       </PrecompiledHeader>
     </ClCompile>
-    <ClCompile Include="CPUMatrix.cpp" />
     <ClCompile Include="MatrixQuantizerCPU.cpp" />
     <ClCompile Include="MatrixQuantizerImpl.cpp" />
     <ClCompile Include="NoGPU.cpp" />
@@ -231,4 +233,4 @@
     <Error Condition="'$(MathLibrary)' == 'MKL' And !Exists('$(CNTKCustomMKLPath)')" Text="CNTK custom MKL version $(CNTKCustomMKLVersion) not found. See https://github.com/Microsoft/CNTK/wiki/Setup-CNTK-on-Windows#mkl for instructions." />
     <Error Condition="!$(HasBoost)" Text="The Boost library is required to build. Please see https://github.com/Microsoft/CNTK/wiki/Setup-CNTK-on-Windows#boost for installation instructions." />
   </Target>
-</Project>
+</Project>

Source/Math/Math.vcxproj.filters

@@ -2,9 +2,6 @@
 <Project ToolsVersion="4.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
   <ItemGroup>
     <ClCompile Include="Matrix.cpp" />
-    <ClCompile Include="CPUMatrix.cpp">
-      <Filter>CPU</Filter>
-    </ClCompile>
     <ClCompile Include="CPUSparseMatrix.cpp">
       <Filter>CPU</Filter>
     </ClCompile>
@@ -49,6 +46,12 @@
       <Filter>CPU</Filter>
     </ClCompile>
     <ClCompile Include="DataTransferer.cpp" />
+    <ClCompile Include="CPUMatrixDouble.cpp">
+      <Filter>CPU</Filter>
+    </ClCompile>
+    <ClCompile Include="CPUMatrixFloat.cpp">
+      <Filter>CPU</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="CommonMatrix.h" />
@@ -131,6 +134,9 @@
     <ClInclude Include="QuantizedOperations.h" />
     <ClInclude Include="BlockMultiplierMatrixUtil.h" />
     <ClInclude Include="DataTransferer.h" />
+    <ClInclude Include="CPUMatrixImpl.h">
+      <Filter>CPU</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <None Include="GPUMatrix.h">