Remove SM20 support

Makefile

@@ -169,7 +169,6 @@ endif
 
 # Set up nvcc target architectures (will generate code to support them all, i.e. fat-binary, in release mode)
 # In debug mode we will rely on JIT to create code "on the fly" for the underlying architecture
-GENCODE_SM20 := -gencode arch=compute_20,code=\"sm_20,compute_20\"
 GENCODE_SM30 := -gencode arch=compute_30,code=\"sm_30,compute_30\"
 GENCODE_SM35 := -gencode arch=compute_35,code=\"sm_35,compute_35\"
 GENCODE_SM50 := -gencode arch=compute_50,code=\"sm_50,compute_50\"
@@ -189,7 +188,7 @@ ifeq ("$(BUILDTYPE)","debug")
   ifdef CNTK_CUDA_CODEGEN_DEBUG
     GENCODE_FLAGS := $(CNTK_CUDA_CODEGEN_DEBUG)
   else
-    GENCODE_FLAGS := -gencode arch=compute_20,code=\"compute_20\" $(GENCODE_SM30)
+    GENCODE_FLAGS := $(GENCODE_SM30)
   endif
 
   CXXFLAGS += -g
@@ -202,7 +201,7 @@ ifeq ("$(BUILDTYPE)","release")
   ifdef CNTK_CUDA_CODEGEN_RELEASE
     GENCODE_FLAGS := $(CNTK_CUDA_CODEGEN_RELEASE)
   else
-    GENCODE_FLAGS := $(GENCODE_SM20) $(GENCODE_SM30) $(GENCODE_SM35) $(GENCODE_SM50)
+    GENCODE_FLAGS := $(GENCODE_SM30) $(GENCODE_SM35) $(GENCODE_SM50)
   endif
 
   CXXFLAGS += -g -O4

Source/Math/CntkBatchNormalization.cuh

@@ -108,11 +108,11 @@ __device__ __forceinline__ void StoreValues<4, float>(const float src[4], float*
 template <typename T>
 __device__ __forceinline__ T Shuffle(T input, int srcLane)
 {
-    // shfl is supported only on Kepler+. We really don't care about Fermi anymore but our build still has sm_20.
-#if __CUDA_ARCH__ >= 300
+#ifdef __CUDA_ARCH__
+    // shfl is supported only on Kepler+
+    static_assert(__CUDA_ARCH__ >= 300, "CNTK only supports only Kepler GPU architecture or newer");
     return cub::ShuffleIndex(input, srcLane);
 #else
-    // REVIEW alexeyk: make static_assert once we remove SM 2.0 support from our build.
     assert(false);
     return input;
 #endif

Source/Math/Math.vcxproj

@@ -85,7 +85,7 @@
     </PostBuildEvent>
     <CudaCompile>
       <TargetMachinePlatform>64</TargetMachinePlatform>
-      <CodeGeneration>compute_20,sm_20;compute_30,sm_30;compute_35,sm_35;compute_50,sm_50;</CodeGeneration>
+      <CodeGeneration>compute_30,sm_30;compute_35,sm_35;compute_50,sm_50;</CodeGeneration>
       <CudaRuntime>None</CudaRuntime>
       <NvccCompilation>compile</NvccCompilation>
       <GenerateLineInfo>true</GenerateLineInfo>
@@ -128,7 +128,7 @@
     <CudaCompile>
       <FastMath>true</FastMath>
       <TargetMachinePlatform>64</TargetMachinePlatform>
-      <CodeGeneration>compute_20,sm_20;compute_30,sm_30;compute_35,sm_35</CodeGeneration>
+      <CodeGeneration>compute_30,sm_30;compute_35,sm_35</CodeGeneration>
       <GPUDebugInfo>true</GPUDebugInfo>
       <GenerateLineInfo>true</GenerateLineInfo>
       <HostDebugInfo>true</HostDebugInfo>

Source/Math/MathCUDA.vcxproj

@@ -52,7 +52,7 @@
   <PropertyGroup Label="UserMacros" />
   <PropertyGroup Condition="$(DebugBuild)">
     <CudaCodeGen>$(CNTK_CUDA_CODEGEN_DEBUG)</CudaCodeGen>
-    <CudaCodeGen Condition="'$(CudaCodeGen)'==''">compute_20,compute_20;compute_30,sm_30</CudaCodeGen>
+    <CudaCodeGen Condition="'$(CudaCodeGen)'==''">compute_30,sm_30</CudaCodeGen>
   </PropertyGroup>
   <PropertyGroup Condition="$(ReleaseBuild)">
     <CudaCodeGen>$(CNTK_CUDA_CODEGEN_RELEASE)</CudaCodeGen>

Tests/UnitTests/EvalTests/EvalTests.vcxproj

@@ -91,7 +91,7 @@
     </Link>
     <CudaCompile>
       <TargetMachinePlatform>64</TargetMachinePlatform>
-      <CodeGeneration>compute_20,sm_20;compute_30,sm_30;%(CodeGeneration)</CodeGeneration>
+      <CodeGeneration>compute_30,sm_30;%(CodeGeneration)</CodeGeneration>
     </CudaCompile>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="$(ReleaseBuild)">

Tests/UnitTests/MathPerformanceTests/MathPerformanceTests.vcxproj

@@ -70,7 +70,7 @@
     </Link>
     <CudaCompile>
       <TargetMachinePlatform>64</TargetMachinePlatform>
-      <CodeGeneration>compute_20,sm_20;compute_30,sm_30;%(CodeGeneration)</CodeGeneration>
+      <CodeGeneration>compute_30,sm_30;%(CodeGeneration)</CodeGeneration>
     </CudaCompile>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="$(ReleaseBuild)">

Tests/UnitTests/MathTests/MathTests.vcxproj

@@ -90,7 +90,7 @@
     </Link>
     <CudaCompile>
       <TargetMachinePlatform>64</TargetMachinePlatform>
-      <CodeGeneration>compute_20,sm_20;compute_30,sm_30;%(CodeGeneration)</CodeGeneration>
+      <CodeGeneration>compute_30,sm_30;%(CodeGeneration)</CodeGeneration>
     </CudaCompile>
   </ItemDefinitionGroup>
   <ItemDefinitionGroup Condition="$(ReleaseBuild)">