keep cufftPlan2d across ConvolveImpl::convolve calls

on some CUDA versions creating/destroying cufftPlan2d is very time consuming
we now create them in ConvolveImpl::create() and destroy them in the dtor

this solves issue #3385

Author: r2d3

modules/cudaarithm/src/arithm.cpp

@@ -440,6 +440,7 @@ namespace
     {
     public:
         explicit ConvolutionImpl(Size user_block_size_) : user_block_size(user_block_size_) {}
+        ~ConvolutionImpl();
 
         void convolve(InputArray image, InputArray templ, OutputArray result, bool ccorr = false, Stream& stream = Stream::Null());
 
@@ -452,6 +453,8 @@ namespace
         Size user_block_size;
         Size dft_size;
 
+        cufftHandle planR2C, planC2R;
+
         GpuMat image_spect, templ_spect, result_spect;
         GpuMat image_block, templ_block, result_data;
     };
@@ -491,6 +494,15 @@ namespace
         // Use maximum result matrix block size for the estimated DFT block size
         block_size.width = std::min(dft_size.width - templ_size.width + 1, result_size.width);
         block_size.height = std::min(dft_size.height - templ_size.height + 1, result_size.height);
+
+        cufftSafeCall( cufftPlan2d(&planC2R, dft_size.height, dft_size.width, CUFFT_C2R) );
+        cufftSafeCall( cufftPlan2d(&planR2C, dft_size.height, dft_size.width, CUFFT_R2C) );
+    }
+
+    ConvolutionImpl::~ConvolutionImpl()
+    {
+        cufftSafeCall( cufftDestroy(planR2C) );
+        cufftSafeCall( cufftDestroy(planC2R) );
     }
 
     Size ConvolutionImpl::estimateBlockSize(Size result_size)
@@ -516,10 +528,6 @@ namespace
 
         cudaStream_t stream = StreamAccessor::getStream(_stream);
 
-        cufftHandle planR2C, planC2R;
-        cufftSafeCall( cufftPlan2d(&planC2R, dft_size.height, dft_size.width, CUFFT_C2R) );
-        cufftSafeCall( cufftPlan2d(&planR2C, dft_size.height, dft_size.width, CUFFT_R2C) );
-
         cufftSafeCall( cufftSetStream(planR2C, stream) );
         cufftSafeCall( cufftSetStream(planC2R, stream) );
 
@@ -559,9 +567,6 @@ namespace
             }
         }
 
-        cufftSafeCall( cufftDestroy(planR2C) );
-        cufftSafeCall( cufftDestroy(planC2R) );
-
         syncOutput(result, _result, _stream);
     }
 }