Chiebot-Mirror
/
opencv
8
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

added more types support for gpu separable filters

Browse Source
pull/328/head
Vladislav Vinogradov 12 years ago
parent 281d036fcf
commit e446903aac
35 changed files with 2396 additions and 880 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 53
      modules/gpu/src/cuda/column_filter.0.cu
  2. 53
      modules/gpu/src/cuda/column_filter.1.cu
  3. 53
      modules/gpu/src/cuda/column_filter.10.cu
  4. 53
      modules/gpu/src/cuda/column_filter.11.cu
  5. 53
      modules/gpu/src/cuda/column_filter.12.cu
  6. 53
      modules/gpu/src/cuda/column_filter.13.cu
  7. 53
      modules/gpu/src/cuda/column_filter.14.cu
  8. 53
      modules/gpu/src/cuda/column_filter.2.cu
  9. 53
      modules/gpu/src/cuda/column_filter.3.cu
  10. 53
      modules/gpu/src/cuda/column_filter.4.cu
  11. 53
      modules/gpu/src/cuda/column_filter.5.cu
  12. 53
      modules/gpu/src/cuda/column_filter.6.cu
  13. 53
      modules/gpu/src/cuda/column_filter.7.cu
  14. 53
      modules/gpu/src/cuda/column_filter.8.cu
  15. 53
      modules/gpu/src/cuda/column_filter.9.cu
  16. 391
      modules/gpu/src/cuda/column_filter.cu
  17. 373
      modules/gpu/src/cuda/column_filter.h
  18. 53
      modules/gpu/src/cuda/row_filter.0.cu
  19. 53
      modules/gpu/src/cuda/row_filter.1.cu
  20. 53
      modules/gpu/src/cuda/row_filter.10.cu
  21. 53
      modules/gpu/src/cuda/row_filter.11.cu
  22. 53
      modules/gpu/src/cuda/row_filter.12.cu
  23. 53
      modules/gpu/src/cuda/row_filter.13.cu
  24. 53
      modules/gpu/src/cuda/row_filter.14.cu
  25. 53
      modules/gpu/src/cuda/row_filter.2.cu
  26. 53
      modules/gpu/src/cuda/row_filter.3.cu
  27. 53
      modules/gpu/src/cuda/row_filter.4.cu
  28. 53
      modules/gpu/src/cuda/row_filter.5.cu
  29. 53
      modules/gpu/src/cuda/row_filter.6.cu
  30. 53
      modules/gpu/src/cuda/row_filter.7.cu
  31. 53
      modules/gpu/src/cuda/row_filter.8.cu
  32. 53
      modules/gpu/src/cuda/row_filter.9.cu
  33. 390
      modules/gpu/src/cuda/row_filter.cu
  34. 372
      modules/gpu/src/cuda/row_filter.h
  35. 160
      modules/gpu/src/filtering.cpp

53
modules/gpu/src/cuda/column_filter.0.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float, uchar>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.1.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float3, uchar3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.10.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float, unsigned short>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.11.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float3, ushort3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.12.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float4, ushort4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.13.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float3, int3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.14.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float4, int4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.2.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float4, uchar4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.3.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float3, short3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.4.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float, int>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.5.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float, float>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.6.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.7.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.8.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float, short>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/column_filter.9.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "column_filter.h"
namespace filter
{
template void linearColumn<float4, short4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

391
modules/gpu/src/cuda/column_filter.cu
Unescape View File

@ -1,391 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "internal_shared.hpp"
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/vec_math.hpp"
#include "opencv2/gpu/device/limits.hpp"
#include "opencv2/gpu/device/border_interpolate.hpp"
#include "opencv2/gpu/device/static_check.hpp"
namespace cv { namespace gpu { namespace device
{
namespace column_filter
{
#define MAX_KERNEL_SIZE 32
__constant__ float c_kernel[MAX_KERNEL_SIZE];
void loadKernel(const float* kernel, int ksize, cudaStream_t stream)
{
if (stream == 0)
cudaSafeCall( cudaMemcpyToSymbol(c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice) );
else
cudaSafeCall( cudaMemcpyToSymbolAsync(c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice, stream) );
}
template <int KSIZE, typename T, typename D, typename B>
__global__ void linearColumnFilter(const PtrStepSz<T> src, PtrStep<D> dst, const int anchor, const B brd)
{
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
const int BLOCK_DIM_X = 16;
const int BLOCK_DIM_Y = 16;
const int PATCH_PER_BLOCK = 4;
const int HALO_SIZE = KSIZE <= 16 ? 1 : 2;
#else
const int BLOCK_DIM_X = 16;
const int BLOCK_DIM_Y = 8;
const int PATCH_PER_BLOCK = 2;
const int HALO_SIZE = 2;
#endif
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type sum_t;
__shared__ sum_t smem[(PATCH_PER_BLOCK + 2 * HALO_SIZE) * BLOCK_DIM_Y][BLOCK_DIM_X];
const int x = blockIdx.x * BLOCK_DIM_X + threadIdx.x;
if (x >= src.cols)
return;
const T* src_col = src.ptr() + x;
const int yStart = blockIdx.y * (BLOCK_DIM_Y * PATCH_PER_BLOCK) + threadIdx.y;
if (blockIdx.y > 0)
{
//Upper halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(src(yStart - (HALO_SIZE - j) * BLOCK_DIM_Y, x));
}
else
{
//Upper halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(brd.at_low(yStart - (HALO_SIZE - j) * BLOCK_DIM_Y, src_col, src.step));
}
if (blockIdx.y + 2 < gridDim.y)
{
//Main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y + HALO_SIZE * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(src(yStart + j * BLOCK_DIM_Y, x));
//Lower halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(src(yStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_Y, x));
}
else
{
//Main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y + HALO_SIZE * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(brd.at_high(yStart + j * BLOCK_DIM_Y, src_col, src.step));
//Lower halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(brd.at_high(yStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_Y, src_col, src.step));
}
__syncthreads();
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
{
const int y = yStart + j * BLOCK_DIM_Y;
if (y < src.rows)
{
sum_t sum = VecTraits<sum_t>::all(0);
#pragma unroll
for (int k = 0; k < KSIZE; ++k)
sum = sum + smem[threadIdx.y + HALO_SIZE * BLOCK_DIM_Y + j * BLOCK_DIM_Y - anchor + k][threadIdx.x] * c_kernel[k];
dst(y, x) = saturate_cast<D>(sum);
}
}
}
template <int KSIZE, typename T, typename D, template<typename> class B>
void linearColumnFilter_caller(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream)
{
int BLOCK_DIM_X;
int BLOCK_DIM_Y;
int PATCH_PER_BLOCK;
if (cc >= 20)
{
BLOCK_DIM_X = 16;
BLOCK_DIM_Y = 16;
PATCH_PER_BLOCK = 4;
}
else
{
BLOCK_DIM_X = 16;
BLOCK_DIM_Y = 8;
PATCH_PER_BLOCK = 2;
}
const dim3 block(BLOCK_DIM_X, BLOCK_DIM_Y);
const dim3 grid(divUp(src.cols, BLOCK_DIM_X), divUp(src.rows, BLOCK_DIM_Y * PATCH_PER_BLOCK));
B<T> brd(src.rows);
linearColumnFilter<KSIZE, T, D><<<grid, block, 0, stream>>>(src, dst, anchor, brd);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template <typename T, typename D>
void linearColumnFilter_gpu(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream)
{
typedef void (*caller_t)(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream);
static const caller_t callers[5][33] =
{
{
0,
linearColumnFilter_caller< 1, T, D, BrdColReflect101>,
linearColumnFilter_caller< 2, T, D, BrdColReflect101>,
linearColumnFilter_caller< 3, T, D, BrdColReflect101>,
linearColumnFilter_caller< 4, T, D, BrdColReflect101>,
linearColumnFilter_caller< 5, T, D, BrdColReflect101>,
linearColumnFilter_caller< 6, T, D, BrdColReflect101>,
linearColumnFilter_caller< 7, T, D, BrdColReflect101>,
linearColumnFilter_caller< 8, T, D, BrdColReflect101>,
linearColumnFilter_caller< 9, T, D, BrdColReflect101>,
linearColumnFilter_caller<10, T, D, BrdColReflect101>,
linearColumnFilter_caller<11, T, D, BrdColReflect101>,
linearColumnFilter_caller<12, T, D, BrdColReflect101>,
linearColumnFilter_caller<13, T, D, BrdColReflect101>,
linearColumnFilter_caller<14, T, D, BrdColReflect101>,
linearColumnFilter_caller<15, T, D, BrdColReflect101>,
linearColumnFilter_caller<16, T, D, BrdColReflect101>,
linearColumnFilter_caller<17, T, D, BrdColReflect101>,
linearColumnFilter_caller<18, T, D, BrdColReflect101>,
linearColumnFilter_caller<19, T, D, BrdColReflect101>,
linearColumnFilter_caller<20, T, D, BrdColReflect101>,
linearColumnFilter_caller<21, T, D, BrdColReflect101>,
linearColumnFilter_caller<22, T, D, BrdColReflect101>,
linearColumnFilter_caller<23, T, D, BrdColReflect101>,
linearColumnFilter_caller<24, T, D, BrdColReflect101>,
linearColumnFilter_caller<25, T, D, BrdColReflect101>,
linearColumnFilter_caller<26, T, D, BrdColReflect101>,
linearColumnFilter_caller<27, T, D, BrdColReflect101>,
linearColumnFilter_caller<28, T, D, BrdColReflect101>,
linearColumnFilter_caller<29, T, D, BrdColReflect101>,
linearColumnFilter_caller<30, T, D, BrdColReflect101>,
linearColumnFilter_caller<31, T, D, BrdColReflect101>,
linearColumnFilter_caller<32, T, D, BrdColReflect101>
},
{
0,
linearColumnFilter_caller< 1, T, D, BrdColReplicate>,
linearColumnFilter_caller< 2, T, D, BrdColReplicate>,
linearColumnFilter_caller< 3, T, D, BrdColReplicate>,
linearColumnFilter_caller< 4, T, D, BrdColReplicate>,
linearColumnFilter_caller< 5, T, D, BrdColReplicate>,
linearColumnFilter_caller< 6, T, D, BrdColReplicate>,
linearColumnFilter_caller< 7, T, D, BrdColReplicate>,
linearColumnFilter_caller< 8, T, D, BrdColReplicate>,
linearColumnFilter_caller< 9, T, D, BrdColReplicate>,
linearColumnFilter_caller<10, T, D, BrdColReplicate>,
linearColumnFilter_caller<11, T, D, BrdColReplicate>,
linearColumnFilter_caller<12, T, D, BrdColReplicate>,
linearColumnFilter_caller<13, T, D, BrdColReplicate>,
linearColumnFilter_caller<14, T, D, BrdColReplicate>,
linearColumnFilter_caller<15, T, D, BrdColReplicate>,
linearColumnFilter_caller<16, T, D, BrdColReplicate>,
linearColumnFilter_caller<17, T, D, BrdColReplicate>,
linearColumnFilter_caller<18, T, D, BrdColReplicate>,
linearColumnFilter_caller<19, T, D, BrdColReplicate>,
linearColumnFilter_caller<20, T, D, BrdColReplicate>,
linearColumnFilter_caller<21, T, D, BrdColReplicate>,
linearColumnFilter_caller<22, T, D, BrdColReplicate>,
linearColumnFilter_caller<23, T, D, BrdColReplicate>,
linearColumnFilter_caller<24, T, D, BrdColReplicate>,
linearColumnFilter_caller<25, T, D, BrdColReplicate>,
linearColumnFilter_caller<26, T, D, BrdColReplicate>,
linearColumnFilter_caller<27, T, D, BrdColReplicate>,
linearColumnFilter_caller<28, T, D, BrdColReplicate>,
linearColumnFilter_caller<29, T, D, BrdColReplicate>,
linearColumnFilter_caller<30, T, D, BrdColReplicate>,
linearColumnFilter_caller<31, T, D, BrdColReplicate>,
linearColumnFilter_caller<32, T, D, BrdColReplicate>
},
{
0,
linearColumnFilter_caller< 1, T, D, BrdColConstant>,
linearColumnFilter_caller< 2, T, D, BrdColConstant>,
linearColumnFilter_caller< 3, T, D, BrdColConstant>,
linearColumnFilter_caller< 4, T, D, BrdColConstant>,
linearColumnFilter_caller< 5, T, D, BrdColConstant>,
linearColumnFilter_caller< 6, T, D, BrdColConstant>,
linearColumnFilter_caller< 7, T, D, BrdColConstant>,
linearColumnFilter_caller< 8, T, D, BrdColConstant>,
linearColumnFilter_caller< 9, T, D, BrdColConstant>,
linearColumnFilter_caller<10, T, D, BrdColConstant>,
linearColumnFilter_caller<11, T, D, BrdColConstant>,
linearColumnFilter_caller<12, T, D, BrdColConstant>,
linearColumnFilter_caller<13, T, D, BrdColConstant>,
linearColumnFilter_caller<14, T, D, BrdColConstant>,
linearColumnFilter_caller<15, T, D, BrdColConstant>,
linearColumnFilter_caller<16, T, D, BrdColConstant>,
linearColumnFilter_caller<17, T, D, BrdColConstant>,
linearColumnFilter_caller<18, T, D, BrdColConstant>,
linearColumnFilter_caller<19, T, D, BrdColConstant>,
linearColumnFilter_caller<20, T, D, BrdColConstant>,
linearColumnFilter_caller<21, T, D, BrdColConstant>,
linearColumnFilter_caller<22, T, D, BrdColConstant>,
linearColumnFilter_caller<23, T, D, BrdColConstant>,
linearColumnFilter_caller<24, T, D, BrdColConstant>,
linearColumnFilter_caller<25, T, D, BrdColConstant>,
linearColumnFilter_caller<26, T, D, BrdColConstant>,
linearColumnFilter_caller<27, T, D, BrdColConstant>,
linearColumnFilter_caller<28, T, D, BrdColConstant>,
linearColumnFilter_caller<29, T, D, BrdColConstant>,
linearColumnFilter_caller<30, T, D, BrdColConstant>,
linearColumnFilter_caller<31, T, D, BrdColConstant>,
linearColumnFilter_caller<32, T, D, BrdColConstant>
},
{
0,
linearColumnFilter_caller< 1, T, D, BrdColReflect>,
linearColumnFilter_caller< 2, T, D, BrdColReflect>,
linearColumnFilter_caller< 3, T, D, BrdColReflect>,
linearColumnFilter_caller< 4, T, D, BrdColReflect>,
linearColumnFilter_caller< 5, T, D, BrdColReflect>,
linearColumnFilter_caller< 6, T, D, BrdColReflect>,
linearColumnFilter_caller< 7, T, D, BrdColReflect>,
linearColumnFilter_caller< 8, T, D, BrdColReflect>,
linearColumnFilter_caller< 9, T, D, BrdColReflect>,
linearColumnFilter_caller<10, T, D, BrdColReflect>,
linearColumnFilter_caller<11, T, D, BrdColReflect>,
linearColumnFilter_caller<12, T, D, BrdColReflect>,
linearColumnFilter_caller<13, T, D, BrdColReflect>,
linearColumnFilter_caller<14, T, D, BrdColReflect>,
linearColumnFilter_caller<15, T, D, BrdColReflect>,
linearColumnFilter_caller<16, T, D, BrdColReflect>,
linearColumnFilter_caller<17, T, D, BrdColReflect>,
linearColumnFilter_caller<18, T, D, BrdColReflect>,
linearColumnFilter_caller<19, T, D, BrdColReflect>,
linearColumnFilter_caller<20, T, D, BrdColReflect>,
linearColumnFilter_caller<21, T, D, BrdColReflect>,
linearColumnFilter_caller<22, T, D, BrdColReflect>,
linearColumnFilter_caller<23, T, D, BrdColReflect>,
linearColumnFilter_caller<24, T, D, BrdColReflect>,
linearColumnFilter_caller<25, T, D, BrdColReflect>,
linearColumnFilter_caller<26, T, D, BrdColReflect>,
linearColumnFilter_caller<27, T, D, BrdColReflect>,
linearColumnFilter_caller<28, T, D, BrdColReflect>,
linearColumnFilter_caller<29, T, D, BrdColReflect>,
linearColumnFilter_caller<30, T, D, BrdColReflect>,
linearColumnFilter_caller<31, T, D, BrdColReflect>,
linearColumnFilter_caller<32, T, D, BrdColReflect>
},
{
0,
linearColumnFilter_caller< 1, T, D, BrdColWrap>,
linearColumnFilter_caller< 2, T, D, BrdColWrap>,
linearColumnFilter_caller< 3, T, D, BrdColWrap>,
linearColumnFilter_caller< 4, T, D, BrdColWrap>,
linearColumnFilter_caller< 5, T, D, BrdColWrap>,
linearColumnFilter_caller< 6, T, D, BrdColWrap>,
linearColumnFilter_caller< 7, T, D, BrdColWrap>,
linearColumnFilter_caller< 8, T, D, BrdColWrap>,
linearColumnFilter_caller< 9, T, D, BrdColWrap>,
linearColumnFilter_caller<10, T, D, BrdColWrap>,
linearColumnFilter_caller<11, T, D, BrdColWrap>,
linearColumnFilter_caller<12, T, D, BrdColWrap>,
linearColumnFilter_caller<13, T, D, BrdColWrap>,
linearColumnFilter_caller<14, T, D, BrdColWrap>,
linearColumnFilter_caller<15, T, D, BrdColWrap>,
linearColumnFilter_caller<16, T, D, BrdColWrap>,
linearColumnFilter_caller<17, T, D, BrdColWrap>,
linearColumnFilter_caller<18, T, D, BrdColWrap>,
linearColumnFilter_caller<19, T, D, BrdColWrap>,
linearColumnFilter_caller<20, T, D, BrdColWrap>,
linearColumnFilter_caller<21, T, D, BrdColWrap>,
linearColumnFilter_caller<22, T, D, BrdColWrap>,
linearColumnFilter_caller<23, T, D, BrdColWrap>,
linearColumnFilter_caller<24, T, D, BrdColWrap>,
linearColumnFilter_caller<25, T, D, BrdColWrap>,
linearColumnFilter_caller<26, T, D, BrdColWrap>,
linearColumnFilter_caller<27, T, D, BrdColWrap>,
linearColumnFilter_caller<28, T, D, BrdColWrap>,
linearColumnFilter_caller<29, T, D, BrdColWrap>,
linearColumnFilter_caller<30, T, D, BrdColWrap>,
linearColumnFilter_caller<31, T, D, BrdColWrap>,
linearColumnFilter_caller<32, T, D, BrdColWrap>
}
};
loadKernel(kernel, ksize, stream);
callers[brd_type][ksize]((PtrStepSz<T>)src, (PtrStepSz<D>)dst, anchor, cc, stream);
}
template void linearColumnFilter_gpu<float , uchar >(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float3, uchar3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float4, uchar4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float3, short3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float , int >(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float , float >(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearColumnFilter_gpu<float4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
} // namespace column_filter
}}} // namespace cv { namespace gpu { namespace device
#endif /* CUDA_DISABLER */

373
modules/gpu/src/cuda/column_filter.h
Unescape View File

@ -0,0 +1,373 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "opencv2/gpu/device/common.hpp"
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/vec_math.hpp"
#include "opencv2/gpu/device/border_interpolate.hpp"
using namespace cv::gpu;
using namespace cv::gpu::device;
namespace column_filter
{
#define MAX_KERNEL_SIZE 32
__constant__ float c_kernel[MAX_KERNEL_SIZE];
template <int KSIZE, typename T, typename D, typename B>
__global__ void linearColumnFilter(const PtrStepSz<T> src, PtrStep<D> dst, const int anchor, const B brd)
{
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
const int BLOCK_DIM_X = 16;
const int BLOCK_DIM_Y = 16;
const int PATCH_PER_BLOCK = 4;
const int HALO_SIZE = KSIZE <= 16 ? 1 : 2;
#else
const int BLOCK_DIM_X = 16;
const int BLOCK_DIM_Y = 8;
const int PATCH_PER_BLOCK = 2;
const int HALO_SIZE = 2;
#endif
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type sum_t;
__shared__ sum_t smem[(PATCH_PER_BLOCK + 2 * HALO_SIZE) * BLOCK_DIM_Y][BLOCK_DIM_X];
const int x = blockIdx.x * BLOCK_DIM_X + threadIdx.x;
if (x >= src.cols)
return;
const T* src_col = src.ptr() + x;
const int yStart = blockIdx.y * (BLOCK_DIM_Y * PATCH_PER_BLOCK) + threadIdx.y;
if (blockIdx.y > 0)
{
//Upper halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(src(yStart - (HALO_SIZE - j) * BLOCK_DIM_Y, x));
}
else
{
//Upper halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(brd.at_low(yStart - (HALO_SIZE - j) * BLOCK_DIM_Y, src_col, src.step));
}
if (blockIdx.y + 2 < gridDim.y)
{
//Main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y + HALO_SIZE * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(src(yStart + j * BLOCK_DIM_Y, x));
//Lower halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(src(yStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_Y, x));
}
else
{
//Main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y + HALO_SIZE * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(brd.at_high(yStart + j * BLOCK_DIM_Y, src_col, src.step));
//Lower halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_Y + j * BLOCK_DIM_Y][threadIdx.x] = saturate_cast<sum_t>(brd.at_high(yStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_Y, src_col, src.step));
}
__syncthreads();
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
{
const int y = yStart + j * BLOCK_DIM_Y;
if (y < src.rows)
{
sum_t sum = VecTraits<sum_t>::all(0);
#pragma unroll
for (int k = 0; k < KSIZE; ++k)
sum = sum + smem[threadIdx.y + HALO_SIZE * BLOCK_DIM_Y + j * BLOCK_DIM_Y - anchor + k][threadIdx.x] * c_kernel[k];
dst(y, x) = saturate_cast<D>(sum);
}
}
}
template <int KSIZE, typename T, typename D, template<typename> class B>
void caller(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream)
{
int BLOCK_DIM_X;
int BLOCK_DIM_Y;
int PATCH_PER_BLOCK;
if (cc >= 20)
{
BLOCK_DIM_X = 16;
BLOCK_DIM_Y = 16;
PATCH_PER_BLOCK = 4;
}
else
{
BLOCK_DIM_X = 16;
BLOCK_DIM_Y = 8;
PATCH_PER_BLOCK = 2;
}
const dim3 block(BLOCK_DIM_X, BLOCK_DIM_Y);
const dim3 grid(divUp(src.cols, BLOCK_DIM_X), divUp(src.rows, BLOCK_DIM_Y * PATCH_PER_BLOCK));
B<T> brd(src.rows);
linearColumnFilter<KSIZE, T, D><<<grid, block, 0, stream>>>(src, dst, anchor, brd);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
namespace filter
{
template <typename T, typename D>
void linearColumn(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream)
{
typedef void (*caller_t)(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream);
static const caller_t callers[5][33] =
{
{
0,
column_filter::caller< 1, T, D, BrdColReflect101>,
column_filter::caller< 2, T, D, BrdColReflect101>,
column_filter::caller< 3, T, D, BrdColReflect101>,
column_filter::caller< 4, T, D, BrdColReflect101>,
column_filter::caller< 5, T, D, BrdColReflect101>,
column_filter::caller< 6, T, D, BrdColReflect101>,
column_filter::caller< 7, T, D, BrdColReflect101>,
column_filter::caller< 8, T, D, BrdColReflect101>,
column_filter::caller< 9, T, D, BrdColReflect101>,
column_filter::caller<10, T, D, BrdColReflect101>,
column_filter::caller<11, T, D, BrdColReflect101>,
column_filter::caller<12, T, D, BrdColReflect101>,
column_filter::caller<13, T, D, BrdColReflect101>,
column_filter::caller<14, T, D, BrdColReflect101>,
column_filter::caller<15, T, D, BrdColReflect101>,
column_filter::caller<16, T, D, BrdColReflect101>,
column_filter::caller<17, T, D, BrdColReflect101>,
column_filter::caller<18, T, D, BrdColReflect101>,
column_filter::caller<19, T, D, BrdColReflect101>,
column_filter::caller<20, T, D, BrdColReflect101>,
column_filter::caller<21, T, D, BrdColReflect101>,
column_filter::caller<22, T, D, BrdColReflect101>,
column_filter::caller<23, T, D, BrdColReflect101>,
column_filter::caller<24, T, D, BrdColReflect101>,
column_filter::caller<25, T, D, BrdColReflect101>,
column_filter::caller<26, T, D, BrdColReflect101>,
column_filter::caller<27, T, D, BrdColReflect101>,
column_filter::caller<28, T, D, BrdColReflect101>,
column_filter::caller<29, T, D, BrdColReflect101>,
column_filter::caller<30, T, D, BrdColReflect101>,
column_filter::caller<31, T, D, BrdColReflect101>,
column_filter::caller<32, T, D, BrdColReflect101>
},
{
0,
column_filter::caller< 1, T, D, BrdColReplicate>,
column_filter::caller< 2, T, D, BrdColReplicate>,
column_filter::caller< 3, T, D, BrdColReplicate>,
column_filter::caller< 4, T, D, BrdColReplicate>,
column_filter::caller< 5, T, D, BrdColReplicate>,
column_filter::caller< 6, T, D, BrdColReplicate>,
column_filter::caller< 7, T, D, BrdColReplicate>,
column_filter::caller< 8, T, D, BrdColReplicate>,
column_filter::caller< 9, T, D, BrdColReplicate>,
column_filter::caller<10, T, D, BrdColReplicate>,
column_filter::caller<11, T, D, BrdColReplicate>,
column_filter::caller<12, T, D, BrdColReplicate>,
column_filter::caller<13, T, D, BrdColReplicate>,
column_filter::caller<14, T, D, BrdColReplicate>,
column_filter::caller<15, T, D, BrdColReplicate>,
column_filter::caller<16, T, D, BrdColReplicate>,
column_filter::caller<17, T, D, BrdColReplicate>,
column_filter::caller<18, T, D, BrdColReplicate>,
column_filter::caller<19, T, D, BrdColReplicate>,
column_filter::caller<20, T, D, BrdColReplicate>,
column_filter::caller<21, T, D, BrdColReplicate>,
column_filter::caller<22, T, D, BrdColReplicate>,
column_filter::caller<23, T, D, BrdColReplicate>,
column_filter::caller<24, T, D, BrdColReplicate>,
column_filter::caller<25, T, D, BrdColReplicate>,
column_filter::caller<26, T, D, BrdColReplicate>,
column_filter::caller<27, T, D, BrdColReplicate>,
column_filter::caller<28, T, D, BrdColReplicate>,
column_filter::caller<29, T, D, BrdColReplicate>,
column_filter::caller<30, T, D, BrdColReplicate>,
column_filter::caller<31, T, D, BrdColReplicate>,
column_filter::caller<32, T, D, BrdColReplicate>
},
{
0,
column_filter::caller< 1, T, D, BrdColConstant>,
column_filter::caller< 2, T, D, BrdColConstant>,
column_filter::caller< 3, T, D, BrdColConstant>,
column_filter::caller< 4, T, D, BrdColConstant>,
column_filter::caller< 5, T, D, BrdColConstant>,
column_filter::caller< 6, T, D, BrdColConstant>,
column_filter::caller< 7, T, D, BrdColConstant>,
column_filter::caller< 8, T, D, BrdColConstant>,
column_filter::caller< 9, T, D, BrdColConstant>,
column_filter::caller<10, T, D, BrdColConstant>,
column_filter::caller<11, T, D, BrdColConstant>,
column_filter::caller<12, T, D, BrdColConstant>,
column_filter::caller<13, T, D, BrdColConstant>,
column_filter::caller<14, T, D, BrdColConstant>,
column_filter::caller<15, T, D, BrdColConstant>,
column_filter::caller<16, T, D, BrdColConstant>,
column_filter::caller<17, T, D, BrdColConstant>,
column_filter::caller<18, T, D, BrdColConstant>,
column_filter::caller<19, T, D, BrdColConstant>,
column_filter::caller<20, T, D, BrdColConstant>,
column_filter::caller<21, T, D, BrdColConstant>,
column_filter::caller<22, T, D, BrdColConstant>,
column_filter::caller<23, T, D, BrdColConstant>,
column_filter::caller<24, T, D, BrdColConstant>,
column_filter::caller<25, T, D, BrdColConstant>,
column_filter::caller<26, T, D, BrdColConstant>,
column_filter::caller<27, T, D, BrdColConstant>,
column_filter::caller<28, T, D, BrdColConstant>,
column_filter::caller<29, T, D, BrdColConstant>,
column_filter::caller<30, T, D, BrdColConstant>,
column_filter::caller<31, T, D, BrdColConstant>,
column_filter::caller<32, T, D, BrdColConstant>
},
{
0,
column_filter::caller< 1, T, D, BrdColReflect>,
column_filter::caller< 2, T, D, BrdColReflect>,
column_filter::caller< 3, T, D, BrdColReflect>,
column_filter::caller< 4, T, D, BrdColReflect>,
column_filter::caller< 5, T, D, BrdColReflect>,
column_filter::caller< 6, T, D, BrdColReflect>,
column_filter::caller< 7, T, D, BrdColReflect>,
column_filter::caller< 8, T, D, BrdColReflect>,
column_filter::caller< 9, T, D, BrdColReflect>,
column_filter::caller<10, T, D, BrdColReflect>,
column_filter::caller<11, T, D, BrdColReflect>,
column_filter::caller<12, T, D, BrdColReflect>,
column_filter::caller<13, T, D, BrdColReflect>,
column_filter::caller<14, T, D, BrdColReflect>,
column_filter::caller<15, T, D, BrdColReflect>,
column_filter::caller<16, T, D, BrdColReflect>,
column_filter::caller<17, T, D, BrdColReflect>,
column_filter::caller<18, T, D, BrdColReflect>,
column_filter::caller<19, T, D, BrdColReflect>,
column_filter::caller<20, T, D, BrdColReflect>,
column_filter::caller<21, T, D, BrdColReflect>,
column_filter::caller<22, T, D, BrdColReflect>,
column_filter::caller<23, T, D, BrdColReflect>,
column_filter::caller<24, T, D, BrdColReflect>,
column_filter::caller<25, T, D, BrdColReflect>,
column_filter::caller<26, T, D, BrdColReflect>,
column_filter::caller<27, T, D, BrdColReflect>,
column_filter::caller<28, T, D, BrdColReflect>,
column_filter::caller<29, T, D, BrdColReflect>,
column_filter::caller<30, T, D, BrdColReflect>,
column_filter::caller<31, T, D, BrdColReflect>,
column_filter::caller<32, T, D, BrdColReflect>
},
{
0,
column_filter::caller< 1, T, D, BrdColWrap>,
column_filter::caller< 2, T, D, BrdColWrap>,
column_filter::caller< 3, T, D, BrdColWrap>,
column_filter::caller< 4, T, D, BrdColWrap>,
column_filter::caller< 5, T, D, BrdColWrap>,
column_filter::caller< 6, T, D, BrdColWrap>,
column_filter::caller< 7, T, D, BrdColWrap>,
column_filter::caller< 8, T, D, BrdColWrap>,
column_filter::caller< 9, T, D, BrdColWrap>,
column_filter::caller<10, T, D, BrdColWrap>,
column_filter::caller<11, T, D, BrdColWrap>,
column_filter::caller<12, T, D, BrdColWrap>,
column_filter::caller<13, T, D, BrdColWrap>,
column_filter::caller<14, T, D, BrdColWrap>,
column_filter::caller<15, T, D, BrdColWrap>,
column_filter::caller<16, T, D, BrdColWrap>,
column_filter::caller<17, T, D, BrdColWrap>,
column_filter::caller<18, T, D, BrdColWrap>,
column_filter::caller<19, T, D, BrdColWrap>,
column_filter::caller<20, T, D, BrdColWrap>,
column_filter::caller<21, T, D, BrdColWrap>,
column_filter::caller<22, T, D, BrdColWrap>,
column_filter::caller<23, T, D, BrdColWrap>,
column_filter::caller<24, T, D, BrdColWrap>,
column_filter::caller<25, T, D, BrdColWrap>,
column_filter::caller<26, T, D, BrdColWrap>,
column_filter::caller<27, T, D, BrdColWrap>,
column_filter::caller<28, T, D, BrdColWrap>,
column_filter::caller<29, T, D, BrdColWrap>,
column_filter::caller<30, T, D, BrdColWrap>,
column_filter::caller<31, T, D, BrdColWrap>,
column_filter::caller<32, T, D, BrdColWrap>
}
};
if (stream == 0)
cudaSafeCall( cudaMemcpyToSymbol(column_filter::c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice) );
else
cudaSafeCall( cudaMemcpyToSymbolAsync(column_filter::c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice, stream) );
callers[brd_type][ksize]((PtrStepSz<T>)src, (PtrStepSz<D>)dst, anchor, cc, stream);
}
}

53
modules/gpu/src/cuda/row_filter.0.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<uchar, float>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.1.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<uchar3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.10.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<unsigned short, float>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.11.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<ushort3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.12.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<ushort4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.13.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<int3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.14.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<int4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.2.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<uchar4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.3.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<short3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.4.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<int, float>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.5.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<float, float>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.6.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<float3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.7.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<float4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.8.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<short, float>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

53
modules/gpu/src/cuda/row_filter.9.cu
Unescape View File

@ -0,0 +1,53 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "row_filter.h"
namespace filter
{
template void linearRow<short4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
#endif /* CUDA_DISABLER */

390
modules/gpu/src/cuda/row_filter.cu
Unescape View File

@ -1,390 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#if !defined CUDA_DISABLER
#include "internal_shared.hpp"
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/vec_math.hpp"
#include "opencv2/gpu/device/limits.hpp"
#include "opencv2/gpu/device/border_interpolate.hpp"
#include "opencv2/gpu/device/static_check.hpp"
namespace cv { namespace gpu { namespace device
{
namespace row_filter
{
#define MAX_KERNEL_SIZE 32
__constant__ float c_kernel[MAX_KERNEL_SIZE];
void loadKernel(const float* kernel, int ksize, cudaStream_t stream)
{
if (stream == 0)
cudaSafeCall( cudaMemcpyToSymbol(c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice) );
else
cudaSafeCall( cudaMemcpyToSymbolAsync(c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice, stream) );
}
template <int KSIZE, typename T, typename D, typename B>
__global__ void linearRowFilter(const PtrStepSz<T> src, PtrStep<D> dst, const int anchor, const B brd)
{
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
const int BLOCK_DIM_X = 32;
const int BLOCK_DIM_Y = 8;
const int PATCH_PER_BLOCK = 4;
const int HALO_SIZE = 1;
#else
const int BLOCK_DIM_X = 32;
const int BLOCK_DIM_Y = 4;
const int PATCH_PER_BLOCK = 4;
const int HALO_SIZE = 1;
#endif
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type sum_t;
__shared__ sum_t smem[BLOCK_DIM_Y][(PATCH_PER_BLOCK + 2 * HALO_SIZE) * BLOCK_DIM_X];
const int y = blockIdx.y * BLOCK_DIM_Y + threadIdx.y;
if (y >= src.rows)
return;
const T* src_row = src.ptr(y);
const int xStart = blockIdx.x * (PATCH_PER_BLOCK * BLOCK_DIM_X) + threadIdx.x;
if (blockIdx.x > 0)
{
//Load left halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + j * BLOCK_DIM_X] = saturate_cast<sum_t>(src_row[xStart - (HALO_SIZE - j) * BLOCK_DIM_X]);
}
else
{
//Load left halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + j * BLOCK_DIM_X] = saturate_cast<sum_t>(brd.at_low(xStart - (HALO_SIZE - j) * BLOCK_DIM_X, src_row));
}
if (blockIdx.x + 2 < gridDim.x)
{
//Load main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y][threadIdx.x + HALO_SIZE * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(src_row[xStart + j * BLOCK_DIM_X]);
//Load right halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(src_row[xStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_X]);
}
else
{
//Load main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y][threadIdx.x + HALO_SIZE * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(brd.at_high(xStart + j * BLOCK_DIM_X, src_row));
//Load right halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(brd.at_high(xStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_X, src_row));
}
__syncthreads();
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
{
const int x = xStart + j * BLOCK_DIM_X;
if (x < src.cols)
{
sum_t sum = VecTraits<sum_t>::all(0);
#pragma unroll
for (int k = 0; k < KSIZE; ++k)
sum = sum + smem[threadIdx.y][threadIdx.x + HALO_SIZE * BLOCK_DIM_X + j * BLOCK_DIM_X - anchor + k] * c_kernel[k];
dst(y, x) = saturate_cast<D>(sum);
}
}
}
template <int KSIZE, typename T, typename D, template<typename> class B>
void linearRowFilter_caller(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream)
{
int BLOCK_DIM_X;
int BLOCK_DIM_Y;
int PATCH_PER_BLOCK;
if (cc >= 20)
{
BLOCK_DIM_X = 32;
BLOCK_DIM_Y = 8;
PATCH_PER_BLOCK = 4;
}
else
{
BLOCK_DIM_X = 32;
BLOCK_DIM_Y = 4;
PATCH_PER_BLOCK = 4;
}
const dim3 block(BLOCK_DIM_X, BLOCK_DIM_Y);
const dim3 grid(divUp(src.cols, BLOCK_DIM_X * PATCH_PER_BLOCK), divUp(src.rows, BLOCK_DIM_Y));
B<T> brd(src.cols);
linearRowFilter<KSIZE, T, D><<<grid, block, 0, stream>>>(src, dst, anchor, brd);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
template <typename T, typename D>
void linearRowFilter_gpu(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream)
{
typedef void (*caller_t)(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream);
static const caller_t callers[5][33] =
{
{
0,
linearRowFilter_caller< 1, T, D, BrdRowReflect101>,
linearRowFilter_caller< 2, T, D, BrdRowReflect101>,
linearRowFilter_caller< 3, T, D, BrdRowReflect101>,
linearRowFilter_caller< 4, T, D, BrdRowReflect101>,
linearRowFilter_caller< 5, T, D, BrdRowReflect101>,
linearRowFilter_caller< 6, T, D, BrdRowReflect101>,
linearRowFilter_caller< 7, T, D, BrdRowReflect101>,
linearRowFilter_caller< 8, T, D, BrdRowReflect101>,
linearRowFilter_caller< 9, T, D, BrdRowReflect101>,
linearRowFilter_caller<10, T, D, BrdRowReflect101>,
linearRowFilter_caller<11, T, D, BrdRowReflect101>,
linearRowFilter_caller<12, T, D, BrdRowReflect101>,
linearRowFilter_caller<13, T, D, BrdRowReflect101>,
linearRowFilter_caller<14, T, D, BrdRowReflect101>,
linearRowFilter_caller<15, T, D, BrdRowReflect101>,
linearRowFilter_caller<16, T, D, BrdRowReflect101>,
linearRowFilter_caller<17, T, D, BrdRowReflect101>,
linearRowFilter_caller<18, T, D, BrdRowReflect101>,
linearRowFilter_caller<19, T, D, BrdRowReflect101>,
linearRowFilter_caller<20, T, D, BrdRowReflect101>,
linearRowFilter_caller<21, T, D, BrdRowReflect101>,
linearRowFilter_caller<22, T, D, BrdRowReflect101>,
linearRowFilter_caller<23, T, D, BrdRowReflect101>,
linearRowFilter_caller<24, T, D, BrdRowReflect101>,
linearRowFilter_caller<25, T, D, BrdRowReflect101>,
linearRowFilter_caller<26, T, D, BrdRowReflect101>,
linearRowFilter_caller<27, T, D, BrdRowReflect101>,
linearRowFilter_caller<28, T, D, BrdRowReflect101>,
linearRowFilter_caller<29, T, D, BrdRowReflect101>,
linearRowFilter_caller<30, T, D, BrdRowReflect101>,
linearRowFilter_caller<31, T, D, BrdRowReflect101>,
linearRowFilter_caller<32, T, D, BrdRowReflect101>
},
{
0,
linearRowFilter_caller< 1, T, D, BrdRowReplicate>,
linearRowFilter_caller< 2, T, D, BrdRowReplicate>,
linearRowFilter_caller< 3, T, D, BrdRowReplicate>,
linearRowFilter_caller< 4, T, D, BrdRowReplicate>,
linearRowFilter_caller< 5, T, D, BrdRowReplicate>,
linearRowFilter_caller< 6, T, D, BrdRowReplicate>,
linearRowFilter_caller< 7, T, D, BrdRowReplicate>,
linearRowFilter_caller< 8, T, D, BrdRowReplicate>,
linearRowFilter_caller< 9, T, D, BrdRowReplicate>,
linearRowFilter_caller<10, T, D, BrdRowReplicate>,
linearRowFilter_caller<11, T, D, BrdRowReplicate>,
linearRowFilter_caller<12, T, D, BrdRowReplicate>,
linearRowFilter_caller<13, T, D, BrdRowReplicate>,
linearRowFilter_caller<14, T, D, BrdRowReplicate>,
linearRowFilter_caller<15, T, D, BrdRowReplicate>,
linearRowFilter_caller<16, T, D, BrdRowReplicate>,
linearRowFilter_caller<17, T, D, BrdRowReplicate>,
linearRowFilter_caller<18, T, D, BrdRowReplicate>,
linearRowFilter_caller<19, T, D, BrdRowReplicate>,
linearRowFilter_caller<20, T, D, BrdRowReplicate>,
linearRowFilter_caller<21, T, D, BrdRowReplicate>,
linearRowFilter_caller<22, T, D, BrdRowReplicate>,
linearRowFilter_caller<23, T, D, BrdRowReplicate>,
linearRowFilter_caller<24, T, D, BrdRowReplicate>,
linearRowFilter_caller<25, T, D, BrdRowReplicate>,
linearRowFilter_caller<26, T, D, BrdRowReplicate>,
linearRowFilter_caller<27, T, D, BrdRowReplicate>,
linearRowFilter_caller<28, T, D, BrdRowReplicate>,
linearRowFilter_caller<29, T, D, BrdRowReplicate>,
linearRowFilter_caller<30, T, D, BrdRowReplicate>,
linearRowFilter_caller<31, T, D, BrdRowReplicate>,
linearRowFilter_caller<32, T, D, BrdRowReplicate>
},
{
0,
linearRowFilter_caller< 1, T, D, BrdRowConstant>,
linearRowFilter_caller< 2, T, D, BrdRowConstant>,
linearRowFilter_caller< 3, T, D, BrdRowConstant>,
linearRowFilter_caller< 4, T, D, BrdRowConstant>,
linearRowFilter_caller< 5, T, D, BrdRowConstant>,
linearRowFilter_caller< 6, T, D, BrdRowConstant>,
linearRowFilter_caller< 7, T, D, BrdRowConstant>,
linearRowFilter_caller< 8, T, D, BrdRowConstant>,
linearRowFilter_caller< 9, T, D, BrdRowConstant>,
linearRowFilter_caller<10, T, D, BrdRowConstant>,
linearRowFilter_caller<11, T, D, BrdRowConstant>,
linearRowFilter_caller<12, T, D, BrdRowConstant>,
linearRowFilter_caller<13, T, D, BrdRowConstant>,
linearRowFilter_caller<14, T, D, BrdRowConstant>,
linearRowFilter_caller<15, T, D, BrdRowConstant>,
linearRowFilter_caller<16, T, D, BrdRowConstant>,
linearRowFilter_caller<17, T, D, BrdRowConstant>,
linearRowFilter_caller<18, T, D, BrdRowConstant>,
linearRowFilter_caller<19, T, D, BrdRowConstant>,
linearRowFilter_caller<20, T, D, BrdRowConstant>,
linearRowFilter_caller<21, T, D, BrdRowConstant>,
linearRowFilter_caller<22, T, D, BrdRowConstant>,
linearRowFilter_caller<23, T, D, BrdRowConstant>,
linearRowFilter_caller<24, T, D, BrdRowConstant>,
linearRowFilter_caller<25, T, D, BrdRowConstant>,
linearRowFilter_caller<26, T, D, BrdRowConstant>,
linearRowFilter_caller<27, T, D, BrdRowConstant>,
linearRowFilter_caller<28, T, D, BrdRowConstant>,
linearRowFilter_caller<29, T, D, BrdRowConstant>,
linearRowFilter_caller<30, T, D, BrdRowConstant>,
linearRowFilter_caller<31, T, D, BrdRowConstant>,
linearRowFilter_caller<32, T, D, BrdRowConstant>
},
{
0,
linearRowFilter_caller< 1, T, D, BrdRowReflect>,
linearRowFilter_caller< 2, T, D, BrdRowReflect>,
linearRowFilter_caller< 3, T, D, BrdRowReflect>,
linearRowFilter_caller< 4, T, D, BrdRowReflect>,
linearRowFilter_caller< 5, T, D, BrdRowReflect>,
linearRowFilter_caller< 6, T, D, BrdRowReflect>,
linearRowFilter_caller< 7, T, D, BrdRowReflect>,
linearRowFilter_caller< 8, T, D, BrdRowReflect>,
linearRowFilter_caller< 9, T, D, BrdRowReflect>,
linearRowFilter_caller<10, T, D, BrdRowReflect>,
linearRowFilter_caller<11, T, D, BrdRowReflect>,
linearRowFilter_caller<12, T, D, BrdRowReflect>,
linearRowFilter_caller<13, T, D, BrdRowReflect>,
linearRowFilter_caller<14, T, D, BrdRowReflect>,
linearRowFilter_caller<15, T, D, BrdRowReflect>,
linearRowFilter_caller<16, T, D, BrdRowReflect>,
linearRowFilter_caller<17, T, D, BrdRowReflect>,
linearRowFilter_caller<18, T, D, BrdRowReflect>,
linearRowFilter_caller<19, T, D, BrdRowReflect>,
linearRowFilter_caller<20, T, D, BrdRowReflect>,
linearRowFilter_caller<21, T, D, BrdRowReflect>,
linearRowFilter_caller<22, T, D, BrdRowReflect>,
linearRowFilter_caller<23, T, D, BrdRowReflect>,
linearRowFilter_caller<24, T, D, BrdRowReflect>,
linearRowFilter_caller<25, T, D, BrdRowReflect>,
linearRowFilter_caller<26, T, D, BrdRowReflect>,
linearRowFilter_caller<27, T, D, BrdRowReflect>,
linearRowFilter_caller<28, T, D, BrdRowReflect>,
linearRowFilter_caller<29, T, D, BrdRowReflect>,
linearRowFilter_caller<30, T, D, BrdRowReflect>,
linearRowFilter_caller<31, T, D, BrdRowReflect>,
linearRowFilter_caller<32, T, D, BrdRowReflect>
},
{
0,
linearRowFilter_caller< 1, T, D, BrdRowWrap>,
linearRowFilter_caller< 2, T, D, BrdRowWrap>,
linearRowFilter_caller< 3, T, D, BrdRowWrap>,
linearRowFilter_caller< 4, T, D, BrdRowWrap>,
linearRowFilter_caller< 5, T, D, BrdRowWrap>,
linearRowFilter_caller< 6, T, D, BrdRowWrap>,
linearRowFilter_caller< 7, T, D, BrdRowWrap>,
linearRowFilter_caller< 8, T, D, BrdRowWrap>,
linearRowFilter_caller< 9, T, D, BrdRowWrap>,
linearRowFilter_caller<10, T, D, BrdRowWrap>,
linearRowFilter_caller<11, T, D, BrdRowWrap>,
linearRowFilter_caller<12, T, D, BrdRowWrap>,
linearRowFilter_caller<13, T, D, BrdRowWrap>,
linearRowFilter_caller<14, T, D, BrdRowWrap>,
linearRowFilter_caller<15, T, D, BrdRowWrap>,
linearRowFilter_caller<16, T, D, BrdRowWrap>,
linearRowFilter_caller<17, T, D, BrdRowWrap>,
linearRowFilter_caller<18, T, D, BrdRowWrap>,
linearRowFilter_caller<19, T, D, BrdRowWrap>,
linearRowFilter_caller<20, T, D, BrdRowWrap>,
linearRowFilter_caller<21, T, D, BrdRowWrap>,
linearRowFilter_caller<22, T, D, BrdRowWrap>,
linearRowFilter_caller<23, T, D, BrdRowWrap>,
linearRowFilter_caller<24, T, D, BrdRowWrap>,
linearRowFilter_caller<25, T, D, BrdRowWrap>,
linearRowFilter_caller<26, T, D, BrdRowWrap>,
linearRowFilter_caller<27, T, D, BrdRowWrap>,
linearRowFilter_caller<28, T, D, BrdRowWrap>,
linearRowFilter_caller<29, T, D, BrdRowWrap>,
linearRowFilter_caller<30, T, D, BrdRowWrap>,
linearRowFilter_caller<31, T, D, BrdRowWrap>,
linearRowFilter_caller<32, T, D, BrdRowWrap>
}
};
loadKernel(kernel, ksize, stream);
callers[brd_type][ksize]((PtrStepSz<T>)src, (PtrStepSz<D>)dst, anchor, cc, stream);
}
template void linearRowFilter_gpu<uchar , float >(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<uchar3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<uchar4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<short3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<int , float >(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<float , float >(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<float3, float3>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template void linearRowFilter_gpu<float4, float4>(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
} // namespace row_filter
}}} // namespace cv { namespace gpu { namespace device
#endif /* CUDA_DISABLER */

372
modules/gpu/src/cuda/row_filter.h
Unescape View File

@ -0,0 +1,372 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
// Copyright (C) 1993-2011, NVIDIA Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "opencv2/gpu/device/common.hpp"
#include "opencv2/gpu/device/saturate_cast.hpp"
#include "opencv2/gpu/device/vec_math.hpp"
#include "opencv2/gpu/device/border_interpolate.hpp"
using namespace cv::gpu;
using namespace cv::gpu::device;
namespace row_filter
{
#define MAX_KERNEL_SIZE 32
__constant__ float c_kernel[MAX_KERNEL_SIZE];
template <int KSIZE, typename T, typename D, typename B>
__global__ void linearRowFilter(const PtrStepSz<T> src, PtrStep<D> dst, const int anchor, const B brd)
{
#if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
const int BLOCK_DIM_X = 32;
const int BLOCK_DIM_Y = 8;
const int PATCH_PER_BLOCK = 4;
const int HALO_SIZE = 1;
#else
const int BLOCK_DIM_X = 32;
const int BLOCK_DIM_Y = 4;
const int PATCH_PER_BLOCK = 4;
const int HALO_SIZE = 1;
#endif
typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type sum_t;
__shared__ sum_t smem[BLOCK_DIM_Y][(PATCH_PER_BLOCK + 2 * HALO_SIZE) * BLOCK_DIM_X];
const int y = blockIdx.y * BLOCK_DIM_Y + threadIdx.y;
if (y >= src.rows)
return;
const T* src_row = src.ptr(y);
const int xStart = blockIdx.x * (PATCH_PER_BLOCK * BLOCK_DIM_X) + threadIdx.x;
if (blockIdx.x > 0)
{
//Load left halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + j * BLOCK_DIM_X] = saturate_cast<sum_t>(src_row[xStart - (HALO_SIZE - j) * BLOCK_DIM_X]);
}
else
{
//Load left halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + j * BLOCK_DIM_X] = saturate_cast<sum_t>(brd.at_low(xStart - (HALO_SIZE - j) * BLOCK_DIM_X, src_row));
}
if (blockIdx.x + 2 < gridDim.x)
{
//Load main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y][threadIdx.x + HALO_SIZE * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(src_row[xStart + j * BLOCK_DIM_X]);
//Load right halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(src_row[xStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_X]);
}
else
{
//Load main data
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
smem[threadIdx.y][threadIdx.x + HALO_SIZE * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(brd.at_high(xStart + j * BLOCK_DIM_X, src_row));
//Load right halo
#pragma unroll
for (int j = 0; j < HALO_SIZE; ++j)
smem[threadIdx.y][threadIdx.x + (PATCH_PER_BLOCK + HALO_SIZE) * BLOCK_DIM_X + j * BLOCK_DIM_X] = saturate_cast<sum_t>(brd.at_high(xStart + (PATCH_PER_BLOCK + j) * BLOCK_DIM_X, src_row));
}
__syncthreads();
#pragma unroll
for (int j = 0; j < PATCH_PER_BLOCK; ++j)
{
const int x = xStart + j * BLOCK_DIM_X;
if (x < src.cols)
{
sum_t sum = VecTraits<sum_t>::all(0);
#pragma unroll
for (int k = 0; k < KSIZE; ++k)
sum = sum + smem[threadIdx.y][threadIdx.x + HALO_SIZE * BLOCK_DIM_X + j * BLOCK_DIM_X - anchor + k] * c_kernel[k];
dst(y, x) = saturate_cast<D>(sum);
}
}
}
template <int KSIZE, typename T, typename D, template<typename> class B>
void caller(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream)
{
int BLOCK_DIM_X;
int BLOCK_DIM_Y;
int PATCH_PER_BLOCK;
if (cc >= 20)
{
BLOCK_DIM_X = 32;
BLOCK_DIM_Y = 8;
PATCH_PER_BLOCK = 4;
}
else
{
BLOCK_DIM_X = 32;
BLOCK_DIM_Y = 4;
PATCH_PER_BLOCK = 4;
}
const dim3 block(BLOCK_DIM_X, BLOCK_DIM_Y);
const dim3 grid(divUp(src.cols, BLOCK_DIM_X * PATCH_PER_BLOCK), divUp(src.rows, BLOCK_DIM_Y));
B<T> brd(src.cols);
linearRowFilter<KSIZE, T, D><<<grid, block, 0, stream>>>(src, dst, anchor, brd);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
}
}
namespace filter
{
template <typename T, typename D>
void linearRow(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream)
{
typedef void (*caller_t)(PtrStepSz<T> src, PtrStepSz<D> dst, int anchor, int cc, cudaStream_t stream);
static const caller_t callers[5][33] =
{
{
0,
row_filter::caller< 1, T, D, BrdRowReflect101>,
row_filter::caller< 2, T, D, BrdRowReflect101>,
row_filter::caller< 3, T, D, BrdRowReflect101>,
row_filter::caller< 4, T, D, BrdRowReflect101>,
row_filter::caller< 5, T, D, BrdRowReflect101>,
row_filter::caller< 6, T, D, BrdRowReflect101>,
row_filter::caller< 7, T, D, BrdRowReflect101>,
row_filter::caller< 8, T, D, BrdRowReflect101>,
row_filter::caller< 9, T, D, BrdRowReflect101>,
row_filter::caller<10, T, D, BrdRowReflect101>,
row_filter::caller<11, T, D, BrdRowReflect101>,
row_filter::caller<12, T, D, BrdRowReflect101>,
row_filter::caller<13, T, D, BrdRowReflect101>,
row_filter::caller<14, T, D, BrdRowReflect101>,
row_filter::caller<15, T, D, BrdRowReflect101>,
row_filter::caller<16, T, D, BrdRowReflect101>,
row_filter::caller<17, T, D, BrdRowReflect101>,
row_filter::caller<18, T, D, BrdRowReflect101>,
row_filter::caller<19, T, D, BrdRowReflect101>,
row_filter::caller<20, T, D, BrdRowReflect101>,
row_filter::caller<21, T, D, BrdRowReflect101>,
row_filter::caller<22, T, D, BrdRowReflect101>,
row_filter::caller<23, T, D, BrdRowReflect101>,
row_filter::caller<24, T, D, BrdRowReflect101>,
row_filter::caller<25, T, D, BrdRowReflect101>,
row_filter::caller<26, T, D, BrdRowReflect101>,
row_filter::caller<27, T, D, BrdRowReflect101>,
row_filter::caller<28, T, D, BrdRowReflect101>,
row_filter::caller<29, T, D, BrdRowReflect101>,
row_filter::caller<30, T, D, BrdRowReflect101>,
row_filter::caller<31, T, D, BrdRowReflect101>,
row_filter::caller<32, T, D, BrdRowReflect101>
},
{
0,
row_filter::caller< 1, T, D, BrdRowReplicate>,
row_filter::caller< 2, T, D, BrdRowReplicate>,
row_filter::caller< 3, T, D, BrdRowReplicate>,
row_filter::caller< 4, T, D, BrdRowReplicate>,
row_filter::caller< 5, T, D, BrdRowReplicate>,
row_filter::caller< 6, T, D, BrdRowReplicate>,
row_filter::caller< 7, T, D, BrdRowReplicate>,
row_filter::caller< 8, T, D, BrdRowReplicate>,
row_filter::caller< 9, T, D, BrdRowReplicate>,
row_filter::caller<10, T, D, BrdRowReplicate>,
row_filter::caller<11, T, D, BrdRowReplicate>,
row_filter::caller<12, T, D, BrdRowReplicate>,
row_filter::caller<13, T, D, BrdRowReplicate>,
row_filter::caller<14, T, D, BrdRowReplicate>,
row_filter::caller<15, T, D, BrdRowReplicate>,
row_filter::caller<16, T, D, BrdRowReplicate>,
row_filter::caller<17, T, D, BrdRowReplicate>,
row_filter::caller<18, T, D, BrdRowReplicate>,
row_filter::caller<19, T, D, BrdRowReplicate>,
row_filter::caller<20, T, D, BrdRowReplicate>,
row_filter::caller<21, T, D, BrdRowReplicate>,
row_filter::caller<22, T, D, BrdRowReplicate>,
row_filter::caller<23, T, D, BrdRowReplicate>,
row_filter::caller<24, T, D, BrdRowReplicate>,
row_filter::caller<25, T, D, BrdRowReplicate>,
row_filter::caller<26, T, D, BrdRowReplicate>,
row_filter::caller<27, T, D, BrdRowReplicate>,
row_filter::caller<28, T, D, BrdRowReplicate>,
row_filter::caller<29, T, D, BrdRowReplicate>,
row_filter::caller<30, T, D, BrdRowReplicate>,
row_filter::caller<31, T, D, BrdRowReplicate>,
row_filter::caller<32, T, D, BrdRowReplicate>
},
{
0,
row_filter::caller< 1, T, D, BrdRowConstant>,
row_filter::caller< 2, T, D, BrdRowConstant>,
row_filter::caller< 3, T, D, BrdRowConstant>,
row_filter::caller< 4, T, D, BrdRowConstant>,
row_filter::caller< 5, T, D, BrdRowConstant>,
row_filter::caller< 6, T, D, BrdRowConstant>,
row_filter::caller< 7, T, D, BrdRowConstant>,
row_filter::caller< 8, T, D, BrdRowConstant>,
row_filter::caller< 9, T, D, BrdRowConstant>,
row_filter::caller<10, T, D, BrdRowConstant>,
row_filter::caller<11, T, D, BrdRowConstant>,
row_filter::caller<12, T, D, BrdRowConstant>,
row_filter::caller<13, T, D, BrdRowConstant>,
row_filter::caller<14, T, D, BrdRowConstant>,
row_filter::caller<15, T, D, BrdRowConstant>,
row_filter::caller<16, T, D, BrdRowConstant>,
row_filter::caller<17, T, D, BrdRowConstant>,
row_filter::caller<18, T, D, BrdRowConstant>,
row_filter::caller<19, T, D, BrdRowConstant>,
row_filter::caller<20, T, D, BrdRowConstant>,
row_filter::caller<21, T, D, BrdRowConstant>,
row_filter::caller<22, T, D, BrdRowConstant>,
row_filter::caller<23, T, D, BrdRowConstant>,
row_filter::caller<24, T, D, BrdRowConstant>,
row_filter::caller<25, T, D, BrdRowConstant>,
row_filter::caller<26, T, D, BrdRowConstant>,
row_filter::caller<27, T, D, BrdRowConstant>,
row_filter::caller<28, T, D, BrdRowConstant>,
row_filter::caller<29, T, D, BrdRowConstant>,
row_filter::caller<30, T, D, BrdRowConstant>,
row_filter::caller<31, T, D, BrdRowConstant>,
row_filter::caller<32, T, D, BrdRowConstant>
},
{
0,
row_filter::caller< 1, T, D, BrdRowReflect>,
row_filter::caller< 2, T, D, BrdRowReflect>,
row_filter::caller< 3, T, D, BrdRowReflect>,
row_filter::caller< 4, T, D, BrdRowReflect>,
row_filter::caller< 5, T, D, BrdRowReflect>,
row_filter::caller< 6, T, D, BrdRowReflect>,
row_filter::caller< 7, T, D, BrdRowReflect>,
row_filter::caller< 8, T, D, BrdRowReflect>,
row_filter::caller< 9, T, D, BrdRowReflect>,
row_filter::caller<10, T, D, BrdRowReflect>,
row_filter::caller<11, T, D, BrdRowReflect>,
row_filter::caller<12, T, D, BrdRowReflect>,
row_filter::caller<13, T, D, BrdRowReflect>,
row_filter::caller<14, T, D, BrdRowReflect>,
row_filter::caller<15, T, D, BrdRowReflect>,
row_filter::caller<16, T, D, BrdRowReflect>,
row_filter::caller<17, T, D, BrdRowReflect>,
row_filter::caller<18, T, D, BrdRowReflect>,
row_filter::caller<19, T, D, BrdRowReflect>,
row_filter::caller<20, T, D, BrdRowReflect>,
row_filter::caller<21, T, D, BrdRowReflect>,
row_filter::caller<22, T, D, BrdRowReflect>,
row_filter::caller<23, T, D, BrdRowReflect>,
row_filter::caller<24, T, D, BrdRowReflect>,
row_filter::caller<25, T, D, BrdRowReflect>,
row_filter::caller<26, T, D, BrdRowReflect>,
row_filter::caller<27, T, D, BrdRowReflect>,
row_filter::caller<28, T, D, BrdRowReflect>,
row_filter::caller<29, T, D, BrdRowReflect>,
row_filter::caller<30, T, D, BrdRowReflect>,
row_filter::caller<31, T, D, BrdRowReflect>,
row_filter::caller<32, T, D, BrdRowReflect>
},
{
0,
row_filter::caller< 1, T, D, BrdRowWrap>,
row_filter::caller< 2, T, D, BrdRowWrap>,
row_filter::caller< 3, T, D, BrdRowWrap>,
row_filter::caller< 4, T, D, BrdRowWrap>,
row_filter::caller< 5, T, D, BrdRowWrap>,
row_filter::caller< 6, T, D, BrdRowWrap>,
row_filter::caller< 7, T, D, BrdRowWrap>,
row_filter::caller< 8, T, D, BrdRowWrap>,
row_filter::caller< 9, T, D, BrdRowWrap>,
row_filter::caller<10, T, D, BrdRowWrap>,
row_filter::caller<11, T, D, BrdRowWrap>,
row_filter::caller<12, T, D, BrdRowWrap>,
row_filter::caller<13, T, D, BrdRowWrap>,
row_filter::caller<14, T, D, BrdRowWrap>,
row_filter::caller<15, T, D, BrdRowWrap>,
row_filter::caller<16, T, D, BrdRowWrap>,
row_filter::caller<17, T, D, BrdRowWrap>,
row_filter::caller<18, T, D, BrdRowWrap>,
row_filter::caller<19, T, D, BrdRowWrap>,
row_filter::caller<20, T, D, BrdRowWrap>,
row_filter::caller<21, T, D, BrdRowWrap>,
row_filter::caller<22, T, D, BrdRowWrap>,
row_filter::caller<23, T, D, BrdRowWrap>,
row_filter::caller<24, T, D, BrdRowWrap>,
row_filter::caller<25, T, D, BrdRowWrap>,
row_filter::caller<26, T, D, BrdRowWrap>,
row_filter::caller<27, T, D, BrdRowWrap>,
row_filter::caller<28, T, D, BrdRowWrap>,
row_filter::caller<29, T, D, BrdRowWrap>,
row_filter::caller<30, T, D, BrdRowWrap>,
row_filter::caller<31, T, D, BrdRowWrap>,
row_filter::caller<32, T, D, BrdRowWrap>
}
};
if (stream == 0)
cudaSafeCall( cudaMemcpyToSymbol(row_filter::c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice) );
else
cudaSafeCall( cudaMemcpyToSymbolAsync(row_filter::c_kernel, kernel, ksize * sizeof(float), 0, cudaMemcpyDeviceToDevice, stream) );
callers[brd_type][ksize]((PtrStepSz<T>)src, (PtrStepSz<D>)dst, anchor, cc, stream);
}
}

160
modules/gpu/src/filtering.cpp
Unescape View File

@ -830,20 +830,14 @@ void cv::gpu::filter2D(const GpuMat& src, GpuMat& dst, int ddepth, const Mat& ke
//////////////////////////////////////////////////////////////////////////////////////////////////// ////////////////////////////////////////////////////////////////////////////////////////////////////
// Separable Linear Filter // Separable Linear Filter
namespace cv { namespace gpu { namespace device namespace filter
{ {
namespace row_filter template <typename T, typename D>
{ void linearRow(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template <typename T, typename D>
void linearRowFilter_gpu(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
namespace column_filter template <typename T, typename D>
{ void linearColumn(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
template <typename T, typename D> }
void linearColumnFilter_gpu(PtrStepSzb src, PtrStepSzb dst, const float* kernel, int ksize, int anchor, int brd_type, int cc, cudaStream_t stream);
}
}}}
namespace namespace
{ {
@ -899,72 +893,56 @@ namespace
Ptr<BaseRowFilter_GPU> cv::gpu::getLinearRowFilter_GPU(int srcType, int bufType, const Mat& rowKernel, int anchor, int borderType) Ptr<BaseRowFilter_GPU> cv::gpu::getLinearRowFilter_GPU(int srcType, int bufType, const Mat& rowKernel, int anchor, int borderType)
{ {
using namespace ::cv::gpu::device::row_filter; static const gpuFilter1D_t funcs[7][4] =
{
static const nppFilter1D_t nppFilter1D_callers[] = {0, nppiFilterRow_8u_C1R, 0, 0, nppiFilterRow_8u_C4R}; {filter::linearRow<uchar, float>, 0, filter::linearRow<uchar3, float3>, filter::linearRow<uchar4, float4>},
{0, 0, 0, 0},
{filter::linearRow<ushort, float>, 0, filter::linearRow<ushort3, float3>, filter::linearRow<ushort4, float4>},
{filter::linearRow<short, float>, 0, filter::linearRow<short3, float3>, filter::linearRow<short4, float4>},
{filter::linearRow<int, float>, 0, filter::linearRow<int3, float3>, filter::linearRow<int4, float4>},
{filter::linearRow<float, float>, 0, filter::linearRow<float3, float3>, filter::linearRow<float4, float4>},
{0, 0, 0, 0}
};
static const nppFilter1D_t npp_funcs[] =
{
0, nppiFilterRow_8u_C1R, 0, 0, nppiFilterRow_8u_C4R
};
if ((bufType == srcType) && (srcType == CV_8UC1 || srcType == CV_8UC4)) if ((bufType == srcType) && (srcType == CV_8UC1 || srcType == CV_8UC4))
{ {
CV_Assert(borderType == BORDER_CONSTANT); CV_Assert( borderType == BORDER_CONSTANT );
GpuMat gpu_row_krnl; GpuMat gpu_row_krnl;
int nDivisor; int nDivisor;
normalizeKernel(rowKernel, gpu_row_krnl, CV_32S, &nDivisor, true); normalizeKernel(rowKernel, gpu_row_krnl, CV_32S, &nDivisor, true);
int ksize = gpu_row_krnl.cols; const int ksize = gpu_row_krnl.cols;
normalizeAnchor(anchor, ksize); normalizeAnchor(anchor, ksize);
return Ptr<BaseRowFilter_GPU>(new NppLinearRowFilter(ksize, anchor, gpu_row_krnl, nDivisor, return Ptr<BaseRowFilter_GPU>(new NppLinearRowFilter(ksize, anchor, gpu_row_krnl, nDivisor, npp_funcs[CV_MAT_CN(srcType)]));
nppFilter1D_callers[CV_MAT_CN(srcType)]));
} }
CV_Assert(borderType == BORDER_REFLECT101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP); CV_Assert( borderType == BORDER_REFLECT101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP );
int gpuBorderType; int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderType, gpuBorderType)); CV_Assert( tryConvertToGpuBorderType(borderType, gpuBorderType) );
CV_Assert(srcType == CV_8UC1 || srcType == CV_8UC3 || srcType == CV_8UC4 || srcType == CV_16SC3 || srcType == CV_32SC1 || srcType == CV_32FC1 || srcType == CV_32FC3 || srcType == CV_32FC4); const int sdepth = CV_MAT_DEPTH(srcType);
const int cn = CV_MAT_CN(srcType);
CV_Assert( sdepth <= CV_64F && cn <= 4 );
CV_Assert( CV_MAT_DEPTH(bufType) == CV_32F && CV_MAT_CN(bufType) == cn );
CV_Assert(CV_MAT_DEPTH(bufType) == CV_32F && CV_MAT_CN(srcType) == CV_MAT_CN(bufType)); const gpuFilter1D_t func = funcs[sdepth][cn - 1];
CV_Assert( func != 0 );
GpuMat gpu_row_krnl; GpuMat gpu_row_krnl;
normalizeKernel(rowKernel, gpu_row_krnl, CV_32F); normalizeKernel(rowKernel, gpu_row_krnl, CV_32F);
int ksize = gpu_row_krnl.cols; const int ksize = gpu_row_krnl.cols;
CV_Assert( ksize > 0 && ksize <= 32 );
CV_Assert(ksize > 0 && ksize <= 32);
normalizeAnchor(anchor, ksize); normalizeAnchor(anchor, ksize);
gpuFilter1D_t func = 0;
switch (srcType)
{
case CV_8UC1:
func = linearRowFilter_gpu<uchar, float>;
break;
case CV_8UC3:
func = linearRowFilter_gpu<uchar3, float3>;
break;
case CV_8UC4:
func = linearRowFilter_gpu<uchar4, float4>;
break;
case CV_16SC3:
func = linearRowFilter_gpu<short3, float3>;
break;
case CV_32SC1:
func = linearRowFilter_gpu<int, float>;
break;
case CV_32FC1:
func = linearRowFilter_gpu<float, float>;
break;
case CV_32FC3:
func = linearRowFilter_gpu<float3, float3>;
break;
case CV_32FC4:
func = linearRowFilter_gpu<float4, float4>;
break;
}
return Ptr<BaseRowFilter_GPU>(new GpuLinearRowFilter(ksize, anchor, gpu_row_krnl, func, gpuBorderType)); return Ptr<BaseRowFilter_GPU>(new GpuLinearRowFilter(ksize, anchor, gpu_row_krnl, func, gpuBorderType));
} }
@ -1020,72 +998,56 @@ namespace
Ptr<BaseColumnFilter_GPU> cv::gpu::getLinearColumnFilter_GPU(int bufType, int dstType, const Mat& columnKernel, int anchor, int borderType) Ptr<BaseColumnFilter_GPU> cv::gpu::getLinearColumnFilter_GPU(int bufType, int dstType, const Mat& columnKernel, int anchor, int borderType)
{ {
using namespace ::cv::gpu::device::column_filter; static const gpuFilter1D_t funcs[7][4] =
{
static const nppFilter1D_t nppFilter1D_callers[] = {0, nppiFilterColumn_8u_C1R, 0, 0, nppiFilterColumn_8u_C4R}; {filter::linearColumn<float, uchar>, 0, filter::linearColumn<float3, uchar3>, filter::linearColumn<float4, uchar4>},
{0, 0, 0, 0},
{filter::linearColumn<float, ushort>, 0, filter::linearColumn<float3, ushort3>, filter::linearColumn<float4, ushort4>},
{filter::linearColumn<float, short>, 0, filter::linearColumn<float3, short3>, filter::linearColumn<float4, short4>},
{filter::linearColumn<float, int>, 0, filter::linearColumn<float3, int3>, filter::linearColumn<float4, int4>},
{filter::linearColumn<float, float>, 0, filter::linearColumn<float3, float3>, filter::linearColumn<float4, float4>},
{0, 0, 0, 0}
};
static const nppFilter1D_t npp_funcs[] =
{
0, nppiFilterColumn_8u_C1R, 0, 0, nppiFilterColumn_8u_C4R
};
if ((bufType == dstType) && (bufType == CV_8UC1 || bufType == CV_8UC4)) if ((bufType == dstType) && (bufType == CV_8UC1 || bufType == CV_8UC4))
{ {
CV_Assert(borderType == BORDER_CONSTANT); CV_Assert( borderType == BORDER_CONSTANT );
GpuMat gpu_col_krnl; GpuMat gpu_col_krnl;
int nDivisor; int nDivisor;
normalizeKernel(columnKernel, gpu_col_krnl, CV_32S, &nDivisor, true); normalizeKernel(columnKernel, gpu_col_krnl, CV_32S, &nDivisor, true);
int ksize = gpu_col_krnl.cols; const int ksize = gpu_col_krnl.cols;
normalizeAnchor(anchor, ksize); normalizeAnchor(anchor, ksize);
return Ptr<BaseColumnFilter_GPU>(new NppLinearColumnFilter(ksize, anchor, gpu_col_krnl, nDivisor, return Ptr<BaseColumnFilter_GPU>(new NppLinearColumnFilter(ksize, anchor, gpu_col_krnl, nDivisor, npp_funcs[CV_MAT_CN(bufType)]));
nppFilter1D_callers[CV_MAT_CN(bufType)]));
} }
CV_Assert(borderType == BORDER_REFLECT101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP); CV_Assert( borderType == BORDER_REFLECT101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP );
int gpuBorderType; int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderType, gpuBorderType)); CV_Assert( tryConvertToGpuBorderType(borderType, gpuBorderType) );
CV_Assert(dstType == CV_8UC1 || dstType == CV_8UC3 || dstType == CV_8UC4 || dstType == CV_16SC3 || dstType == CV_32SC1 || dstType == CV_32FC1 || dstType == CV_32FC3 || dstType == CV_32FC4); const int ddepth = CV_MAT_DEPTH(dstType);
const int cn = CV_MAT_CN(dstType);
CV_Assert( ddepth <= CV_64F && cn <= 4 );
CV_Assert( CV_MAT_DEPTH(bufType) == CV_32F && CV_MAT_CN(bufType) == cn );
CV_Assert(CV_MAT_DEPTH(bufType) == CV_32F && CV_MAT_CN(dstType) == CV_MAT_CN(bufType)); gpuFilter1D_t func = funcs[ddepth][cn - 1];
CV_Assert( func != 0 );
GpuMat gpu_col_krnl; GpuMat gpu_col_krnl;
normalizeKernel(columnKernel, gpu_col_krnl, CV_32F); normalizeKernel(columnKernel, gpu_col_krnl, CV_32F);
int ksize = gpu_col_krnl.cols; const int ksize = gpu_col_krnl.cols;
CV_Assert(ksize > 0 && ksize <= 32); CV_Assert(ksize > 0 && ksize <= 32);
normalizeAnchor(anchor, ksize); normalizeAnchor(anchor, ksize);
gpuFilter1D_t func = 0;
switch (dstType)
{
case CV_8UC1:
func = linearColumnFilter_gpu<float, uchar>;
break;
case CV_8UC3:
func = linearColumnFilter_gpu<float3, uchar3>;
break;
case CV_8UC4:
func = linearColumnFilter_gpu<float4, uchar4>;
break;
case CV_16SC3:
func = linearColumnFilter_gpu<float3, short3>;
break;
case CV_32SC1:
func = linearColumnFilter_gpu<float, int>;
break;
case CV_32FC1:
func = linearColumnFilter_gpu<float, float>;
break;
case CV_32FC3:
func = linearColumnFilter_gpu<float3, float3>;
break;
case CV_32FC4:
func = linearColumnFilter_gpu<float4, float4>;
break;
}
return Ptr<BaseColumnFilter_GPU>(new GpuLinearColumnFilter(ksize, anchor, gpu_col_krnl, func, gpuBorderType)); return Ptr<BaseColumnFilter_GPU>(new GpuLinearColumnFilter(ksize, anchor, gpu_col_krnl, func, gpuBorderType));
} }

Write Preview
Loading…
Cancel
Save