removed sync

src/kernels/activation_elu.cu

@@ -34,5 +34,4 @@ void activationELUForward(dnnType* srcData, dnnType* dstData, int size)
     int threads = 256;
     
     activation_elu<<<blocks, threads>>>(srcData, dstData, size);
-    checkCuda( cudaDeviceSynchronize() );
 }

src/kernels/activation_leaky.cu

@@ -23,7 +23,6 @@ void activationLEAKYForward(dnnType* srcData, dnnType* dstData, int size)
     int threads = 256;
     
     activation_leaky<<<blocks, threads>>>(srcData, dstData, size);
-    checkCuda( cudaDeviceSynchronize() );
 }
 
 

src/kernels/activation_logistic.cu

@@ -20,7 +20,6 @@ void activationLOGISTICForward(dnnType* srcData, dnnType* dstData, int size)
     int threads = 256;
     
     activation_logistic<<<blocks, threads>>>(srcData, dstData, size);
-    checkCuda( cudaDeviceSynchronize() );
 }
 
 

src/kernels/reorg.cu

@@ -44,7 +44,6 @@ void reorgForward(dnnType* srcData, dnnType* dstData,
     int threads = 256;
     
     reorg_kernel<<<blocks, threads>>>(size, srcData, w, h, c, n, stride, false, dstData);
-    checkCuda( cudaDeviceSynchronize() );
 }
 
 

src/kernels/softmax.cu

@@ -39,5 +39,4 @@ void softmaxForward(float *input, int n, int batch, int batch_offset,
     int threads = 256;
 
     softmax_kernel<<<blocks, threads>>>(input, n, batch, batch_offset, groups, group_offset, stride, temp, output);
-    checkCuda( cudaDeviceSynchronize() );
 }