minor fix

build/darknet/darknet.vcxproj

@@ -132,7 +132,7 @@
       <IntrinsicFunctions>true</IntrinsicFunctions>
       <SDLCheck>true</SDLCheck>
       <AdditionalIncludeDirectories>C:\opencv_2.4.9\opencv\build\include;..\..\3rdparty\include;%(AdditionalIncludeDirectories);$(CudaToolkitIncludeDir);$(cudnn)\include</AdditionalIncludeDirectories>
-      <PreprocessorDefinitions>OPENCV;_TIMESPEC_DEFINED;_CRT_SECURE_NO_WARNINGS;GPU;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
+      <PreprocessorDefinitions>CUDNN;OPENCV;_TIMESPEC_DEFINED;_CRT_SECURE_NO_WARNINGS;GPU;WIN32;NDEBUG;_CONSOLE;_LIB;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <CLanguageStandard>c11</CLanguageStandard>
       <CppLanguageStandard>c++1y</CppLanguageStandard>
       <PrecompiledHeaderCompileAs>CompileAsCpp</PrecompiledHeaderCompileAs>

src/blas_kernels.cu

@@ -223,6 +223,7 @@ __global__ void fast_mean_delta_kernel(float *delta, float *variance, int batch,
             local[id] += (i+id < spatial) ? delta[index] : 0;
         }
     }
+	__syncthreads();
 
     if(id == 0){
         mean_delta[filter] = 0;
@@ -251,6 +252,7 @@ __global__ void  fast_variance_delta_kernel(float *x, float *delta, float *mean,
             local[id] += (i+id < spatial) ? delta[index]*(x[index] - mean[filter]) : 0;
         }
     }
+	__syncthreads();
 
     if(id == 0){
         variance_delta[filter] = 0;
@@ -446,6 +448,7 @@ __global__ void  fast_mean_kernel(float *x, int batch, int filters, int spatial,
             local[id] += (i+id < spatial) ? x[index] : 0;
         }
     }
+	__syncthreads();
 
     if(id == 0){
         mean[filter] = 0;
@@ -474,6 +477,7 @@ __global__ void  fast_variance_kernel(float *x, float *mean, int batch, int filt
             local[id] += (i+id < spatial) ? pow((x[index] - mean[filter]), 2) : 0;
         }
     }
+	__syncthreads();
 
     if(id == 0){
         variance[filter] = 0;

src/convolutional_kernels.cu

@@ -127,6 +127,7 @@ void forward_convolutional_layer_gpu(convolutional_layer l, network_state state)
     activate_array_ongpu(l.output_gpu, l.outputs*l.batch, l.activation);
     //if(l.dot > 0) dot_error_gpu(l);
     if(l.binary || l.xnor) swap_binary(&l);
+	//cudaDeviceSynchronize();	// for correct profiling of performance
 }
 
 void backward_convolutional_layer_gpu(convolutional_layer l, network_state state)