opencv/opencv_contrib
1
0
Fork 0
mirror of https://github.com/opencv/opencv_contrib.git synced 2025-10-16 13:57:05 +08:00
Code Issues Releases Wiki Activity

Merge pull request #3386 from r2d3:convolve_cuda

Browse Source 
keep cufftPlan2d across ConvolveImpl::convolve calls
This commit is contained in:
Alexander Smorkalov
2022-12-14 16:28:49 +03:00
committed by GitHub
parent 9d1c0f5294 53ab7c3efa
commit c0133e5a40
1 changed files with 26 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
modules/cudaarithm/src/arithm.cpp
View File

@@ -439,7 +439,8 @@ namespace
class ConvolutionImpl : public Convolution class ConvolutionImpl : public Convolution
{ {
public: public:
explicit ConvolutionImpl(Size user_block_size_) : user_block_size(user_block_size_) {} explicit ConvolutionImpl(Size user_block_size_) : user_block_size(user_block_size_), planR2C(0), planC2R(0) {}
~ConvolutionImpl();
void convolve(InputArray image, InputArray templ, OutputArray result, bool ccorr = false, Stream& stream = Stream::Null()); void convolve(InputArray image, InputArray templ, OutputArray result, bool ccorr = false, Stream& stream = Stream::Null());
@@ -452,6 +453,9 @@ namespace
Size user_block_size; Size user_block_size;
Size dft_size; Size dft_size;
cufftHandle planR2C, planC2R;
Size plan_size;
GpuMat image_spect, templ_spect, result_spect; GpuMat image_spect, templ_spect, result_spect;
GpuMat image_block, templ_block, result_data; GpuMat image_block, templ_block, result_data;
}; };
@@ -491,6 +495,27 @@ namespace
// Use maximum result matrix block size for the estimated DFT block size // 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.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); block_size.height = std::min(dft_size.height - templ_size.height + 1, result_size.height);
if (dft_size != plan_size)
{
if (planR2C != 0)
cufftSafeCall( cufftDestroy(planR2C) );
if (planC2R != 0)
cufftSafeCall( cufftDestroy(planC2R) );
cufftSafeCall( cufftPlan2d(&planC2R, dft_size.height, dft_size.width, CUFFT_C2R) );
cufftSafeCall( cufftPlan2d(&planR2C, dft_size.height, dft_size.width, CUFFT_R2C) );
plan_size = dft_size;
}
}
ConvolutionImpl::~ConvolutionImpl()
{
if (planR2C != 0)
cufftSafeCall( cufftDestroy(planR2C) );
if (planC2R != 0)
cufftSafeCall( cufftDestroy(planC2R) );
} }
Size ConvolutionImpl::estimateBlockSize(Size result_size) Size ConvolutionImpl::estimateBlockSize(Size result_size)
@@ -516,10 +541,6 @@ namespace
cudaStream_t stream = StreamAccessor::getStream(_stream); 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(planR2C, stream) );
cufftSafeCall( cufftSetStream(planC2R, stream) ); cufftSafeCall( cufftSetStream(planC2R, stream) );
@@ -559,9 +580,6 @@ namespace
} }
} }
cufftSafeCall( cufftDestroy(planR2C) );
cufftSafeCall( cufftDestroy(planC2R) );
syncOutput(result, _result, _stream); syncOutput(result, _result, _stream);
} }
} }