mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-29 05:29:54 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

minor refactoring:

Browse Source 
moved lbp.hpp to src/cuda folder
added missing cv::gpu::device namespace
deleted whitespaces
This commit is contained in:
Vladislav Vinogradov 2012-10-05 17:43:22 +04:00
parent a703df5456
commit 3ebec7448d
8 changed files with 222 additions and 217 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
modules/gpu/src/cuda/lbp.cu
View File

@ -42,9 +42,9 @@
#if !defined CUDA_DISABLER
#include <opencv2/gpu/device/lbp.hpp>
#include <opencv2/gpu/device/vec_traits.hpp>
#include <opencv2/gpu/device/saturate_cast.hpp>
#include "lbp.hpp"
#include "opencv2/gpu/device/vec_traits.hpp"
#include "opencv2/gpu/device/saturate_cast.hpp"
namespace cv { namespace gpu { namespace device
{
@ -299,4 +299,4 @@ namespace cv { namespace gpu { namespace device
}
}}}
#endif /* CUDA_DISABLER */
#endif /* CUDA_DISABLER */

0
modules/gpu/src/opencv2/gpu/device/lbp.hpp → modules/gpu/src/cuda/lbp.hpp
View File

2
modules/gpu/src/opencv2/gpu/device/detail/color_detail.hpp
View File

@ -1535,6 +1535,8 @@ namespace cv { namespace gpu { namespace device
return functor_type(); \
} \
};
#undef CV_DESCALE
}}} // namespace cv { namespace gpu { namespace device
#endif // __OPENCV_GPU_COLOR_DETAIL_HPP__

184
modules/gpu/src/opencv2/gpu/device/detail/utility_detail.hpp → modules/gpu/src/opencv2/gpu/device/detail/reduction_detail.hpp
View File

@ -40,27 +40,27 @@
//
//M*/
#ifndef __OPENCV_GPU_UTILITY_DETAIL_HPP__
#define __OPENCV_GPU_UTILITY_DETAIL_HPP__
#ifndef __OPENCV_GPU_REDUCTION_DETAIL_HPP__
#define __OPENCV_GPU_REDUCTION_DETAIL_HPP__
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace utility_detail
{
///////////////////////////////////////////////////////////////////////////////
// Reduction
// Reductor
template <int n> struct WarpReductor
{
template <typename T, typename Op> static __device__ __forceinline__ void reduce(volatile T* data, T& partial_reduction, int tid, const Op& op)
{
if (tid < n)
data[tid] = partial_reduction;
data[tid] = partial_reduction;
if (n > 32) __syncthreads();
if (n > 32)
{
if (tid < n - 32)
if (tid < n - 32)
data[tid] = partial_reduction = op(partial_reduction, data[tid + 32]);
if (tid < 16)
{
@ -73,7 +73,7 @@ namespace cv { namespace gpu { namespace device
}
else if (n > 16)
{
if (tid < n - 16)
if (tid < n - 16)
data[tid] = partial_reduction = op(partial_reduction, data[tid + 16]);
if (tid < 8)
{
@ -85,7 +85,7 @@ namespace cv { namespace gpu { namespace device
}
else if (n > 8)
{
if (tid < n - 8)
if (tid < n - 8)
data[tid] = partial_reduction = op(partial_reduction, data[tid + 8]);
if (tid < 4)
{
@ -96,23 +96,23 @@ namespace cv { namespace gpu { namespace device
}
else if (n > 4)
{
if (tid < n - 4)
if (tid < n - 4)
data[tid] = partial_reduction = op(partial_reduction, data[tid + 4]);
if (tid < 2)
{
data[tid] = partial_reduction = op(partial_reduction, data[tid + 2]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1]);
}
}
}
else if (n > 2)
{
if (tid < n - 2)
if (tid < n - 2)
data[tid] = partial_reduction = op(partial_reduction, data[tid + 2]);
if (tid < 2)
{
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1]);
}
}
}
}
};
template <> struct WarpReductor<64>
@ -121,15 +121,15 @@ namespace cv { namespace gpu { namespace device
{
data[tid] = partial_reduction;
__syncthreads();
if (tid < 32)
if (tid < 32)
{
data[tid] = partial_reduction = op(partial_reduction, data[tid + 32]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 16]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 8 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 4 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 2 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
}
}
};
@ -138,14 +138,14 @@ namespace cv { namespace gpu { namespace device
template <typename T, typename Op> static __device__ void reduce(volatile T* data, T& partial_reduction, int tid, const Op& op)
{
data[tid] = partial_reduction;
if (tid < 16)
if (tid < 16)
{
data[tid] = partial_reduction = op(partial_reduction, data[tid + 16]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 8 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 4 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 2 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
}
}
};
@ -154,13 +154,13 @@ namespace cv { namespace gpu { namespace device
template <typename T, typename Op> static __device__ void reduce(volatile T* data, T& partial_reduction, int tid, const Op& op)
{
data[tid] = partial_reduction;
if (tid < 8)
if (tid < 8)
{
data[tid] = partial_reduction = op(partial_reduction, data[tid + 8 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 4 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 2 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
}
}
};
@ -169,12 +169,12 @@ namespace cv { namespace gpu { namespace device
template <typename T, typename Op> static __device__ void reduce(volatile T* data, T& partial_reduction, int tid, const Op& op)
{
data[tid] = partial_reduction;
if (tid < 4)
if (tid < 4)
{
data[tid] = partial_reduction = op(partial_reduction, data[tid + 4 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 2 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
data[tid] = partial_reduction = op(partial_reduction, data[tid + 1 ]);
}
}
};
@ -214,11 +214,11 @@ namespace cv { namespace gpu { namespace device
///////////////////////////////////////////////////////////////////////////////
// PredValWarpReductor
template <int n> struct PredValWarpReductor;
template <> struct PredValWarpReductor<64>
{
template <typename T, typename V, typename Pred>
template <typename T, typename V, typename Pred>
static __device__ void reduce(T& myData, V& myVal, volatile T* sdata, V* sval, int tid, const Pred& pred)
{
if (tid < 32)
@ -253,14 +253,14 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -272,7 +272,7 @@ namespace cv { namespace gpu { namespace device
};
template <> struct PredValWarpReductor<32>
{
template <typename T, typename V, typename Pred>
template <typename T, typename V, typename Pred>
static __device__ void reduce(T& myData, V& myVal, volatile T* sdata, V* sval, int tid, const Pred& pred)
{
if (tid < 16)
@ -300,14 +300,14 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -320,7 +320,7 @@ namespace cv { namespace gpu { namespace device
template <> struct PredValWarpReductor<16>
{
template <typename T, typename V, typename Pred>
template <typename T, typename V, typename Pred>
static __device__ void reduce(T& myData, V& myVal, volatile T* sdata, V* sval, int tid, const Pred& pred)
{
if (tid < 8)
@ -341,14 +341,14 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -360,7 +360,7 @@ namespace cv { namespace gpu { namespace device
};
template <> struct PredValWarpReductor<8>
{
template <typename T, typename V, typename Pred>
template <typename T, typename V, typename Pred>
static __device__ void reduce(T& myData, V& myVal, volatile T* sdata, V* sval, int tid, const Pred& pred)
{
if (tid < 4)
@ -374,14 +374,14 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -407,7 +407,7 @@ namespace cv { namespace gpu { namespace device
myData = sdata[tid];
myVal = sval[tid];
if (n >= 512 && tid < 256)
if (n >= 512 && tid < 256)
{
T reg = sdata[tid + 256];
@ -416,9 +416,9 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 256];
}
__syncthreads();
__syncthreads();
}
if (n >= 256 && tid < 128)
if (n >= 256 && tid < 128)
{
T reg = sdata[tid + 128];
@ -427,9 +427,9 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 128];
}
__syncthreads();
__syncthreads();
}
if (n >= 128 && tid < 64)
if (n >= 128 && tid < 64)
{
T reg = sdata[tid + 64];
@ -438,13 +438,13 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 64];
}
__syncthreads();
}
__syncthreads();
}
if (tid < 32)
{
if (n >= 64)
{
if (n >= 64)
{
T reg = sdata[tid + 32];
if (pred(reg, myData))
@ -453,8 +453,8 @@ namespace cv { namespace gpu { namespace device
sval[tid] = myVal = sval[tid + 32];
}
}
if (n >= 32)
{
if (n >= 32)
{
T reg = sdata[tid + 16];
if (pred(reg, myData))
@ -463,8 +463,8 @@ namespace cv { namespace gpu { namespace device
sval[tid] = myVal = sval[tid + 16];
}
}
if (n >= 16)
{
if (n >= 16)
{
T reg = sdata[tid + 8];
if (pred(reg, myData))
@ -473,8 +473,8 @@ namespace cv { namespace gpu { namespace device
sval[tid] = myVal = sval[tid + 8];
}
}
if (n >= 8)
{
if (n >= 8)
{
T reg = sdata[tid + 4];
if (pred(reg, myData))
@ -483,18 +483,18 @@ namespace cv { namespace gpu { namespace device
sval[tid] = myVal = sval[tid + 4];
}
}
if (n >= 4)
{
if (n >= 4)
{
T reg = sdata[tid + 2];
if (pred(reg, myData))
{
sdata[tid] = myData = reg;
sval[tid] = myVal = sval[tid + 2];
}
}
}
if (n >= 2)
{
if (n >= 2)
{
T reg = sdata[tid + 1];
if (pred(reg, myData))
@ -513,7 +513,7 @@ namespace cv { namespace gpu { namespace device
template <int n> struct PredVal2WarpReductor;
template <> struct PredVal2WarpReductor<64>
{
template <typename T, typename V1, typename V2, typename Pred>
template <typename T, typename V1, typename V2, typename Pred>
static __device__ void reduce(T& myData, V1& myVal1, V2& myVal2, volatile T* sdata, V1* sval1, V2* sval2, int tid, const Pred& pred)
{
if (tid < 32)
@ -553,7 +553,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 4];
sval2[tid] = myVal2 = sval2[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
@ -561,7 +561,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 2];
sval2[tid] = myVal2 = sval2[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -574,7 +574,7 @@ namespace cv { namespace gpu { namespace device
};
template <> struct PredVal2WarpReductor<32>
{
template <typename T, typename V1, typename V2, typename Pred>
template <typename T, typename V1, typename V2, typename Pred>
static __device__ void reduce(T& myData, V1& myVal1, V2& myVal2, volatile T* sdata, V1* sval1, V2* sval2, int tid, const Pred& pred)
{
if (tid < 16)
@ -606,7 +606,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 4];
sval2[tid] = myVal2 = sval2[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
@ -614,7 +614,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 2];
sval2[tid] = myVal2 = sval2[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -628,7 +628,7 @@ namespace cv { namespace gpu { namespace device
template <> struct PredVal2WarpReductor<16>
{
template <typename T, typename V1, typename V2, typename Pred>
template <typename T, typename V1, typename V2, typename Pred>
static __device__ void reduce(T& myData, V1& myVal1, V2& myVal2, volatile T* sdata, V1* sval1, V2* sval2, int tid, const Pred& pred)
{
if (tid < 8)
@ -652,7 +652,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 4];
sval2[tid] = myVal2 = sval2[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
@ -660,7 +660,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 2];
sval2[tid] = myVal2 = sval2[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -673,7 +673,7 @@ namespace cv { namespace gpu { namespace device
};
template <> struct PredVal2WarpReductor<8>
{
template <typename T, typename V1, typename V2, typename Pred>
template <typename T, typename V1, typename V2, typename Pred>
static __device__ void reduce(T& myData, V1& myVal1, V2& myVal2, volatile T* sdata, V1* sval1, V2* sval2, int tid, const Pred& pred)
{
if (tid < 4)
@ -689,7 +689,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 4];
sval2[tid] = myVal2 = sval2[tid + 4];
}
reg = sdata[tid + 2];
if (pred(reg, myData))
{
@ -697,7 +697,7 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 2];
sval2[tid] = myVal2 = sval2[tid + 2];
}
reg = sdata[tid + 1];
if (pred(reg, myData))
{
@ -712,7 +712,7 @@ namespace cv { namespace gpu { namespace device
template <bool warp> struct PredVal2ReductionDispatcher;
template <> struct PredVal2ReductionDispatcher<true>
{
template <int n, typename T, typename V1, typename V2, typename Pred>
template <int n, typename T, typename V1, typename V2, typename Pred>
static __device__ void reduce(T& myData, V1& myVal1, V2& myVal2, volatile T* sdata, V1* sval1, V2* sval2, int tid, const Pred& pred)
{
PredVal2WarpReductor<n>::reduce(myData, myVal1, myVal2, sdata, sval1, sval2, tid, pred);
@ -720,14 +720,14 @@ namespace cv { namespace gpu { namespace device
};
template <> struct PredVal2ReductionDispatcher<false>
{
template <int n, typename T, typename V1, typename V2, typename Pred>
template <int n, typename T, typename V1, typename V2, typename Pred>
static __device__ void reduce(T& myData, V1& myVal1, V2& myVal2, volatile T* sdata, V1* sval1, V2* sval2, int tid, const Pred& pred)
{
myData = sdata[tid];
myVal1 = sval1[tid];
myVal2 = sval2[tid];
if (n >= 512 && tid < 256)
if (n >= 512 && tid < 256)
{
T reg = sdata[tid + 256];
@ -737,9 +737,9 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 256];
sval2[tid] = myVal2 = sval2[tid + 256];
}
__syncthreads();
__syncthreads();
}
if (n >= 256 && tid < 128)
if (n >= 256 && tid < 128)
{
T reg = sdata[tid + 128];
@ -749,9 +749,9 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 128];
sval2[tid] = myVal2 = sval2[tid + 128];
}
__syncthreads();
__syncthreads();
}
if (n >= 128 && tid < 64)
if (n >= 128 && tid < 64)
{
T reg = sdata[tid + 64];
@ -761,13 +761,13 @@ namespace cv { namespace gpu { namespace device
sval1[tid] = myVal1 = sval1[tid + 64];
sval2[tid] = myVal2 = sval2[tid + 64];
}
__syncthreads();
}
__syncthreads();
}
if (tid < 32)
{
if (n >= 64)
{
if (n >= 64)
{
T reg = sdata[tid + 32];
if (pred(reg, myData))
@ -777,8 +777,8 @@ namespace cv { namespace gpu { namespace device
sval2[tid] = myVal2 = sval2[tid + 32];
}
}
if (n >= 32)
{
if (n >= 32)
{
T reg = sdata[tid + 16];
if (pred(reg, myData))
@ -788,8 +788,8 @@ namespace cv { namespace gpu { namespace device
sval2[tid] = myVal2 = sval2[tid + 16];
}
}
if (n >= 16)
{
if (n >= 16)
{
T reg = sdata[tid + 8];
if (pred(reg, myData))
@ -799,8 +799,8 @@ namespace cv { namespace gpu { namespace device
sval2[tid] = myVal2 = sval2[tid + 8];
}
}
if (n >= 8)
{
if (n >= 8)
{
T reg = sdata[tid + 4];
if (pred(reg, myData))
@ -810,8 +810,8 @@ namespace cv { namespace gpu { namespace device
sval2[tid] = myVal2 = sval2[tid + 4];
}
}
if (n >= 4)
{
if (n >= 4)
{
T reg = sdata[tid + 2];
if (pred(reg, myData))
@ -819,10 +819,10 @@ namespace cv { namespace gpu { namespace device
sdata[tid] = myData = reg;
sval1[tid] = myVal1 = sval1[tid + 2];
sval2[tid] = myVal2 = sval2[tid + 2];
}
}
}
if (n >= 2)
{
if (n >= 2)
{
T reg = sdata[tid + 1];
if (pred(reg, myData))
@ -838,4 +838,4 @@ namespace cv { namespace gpu { namespace device
} // namespace utility_detail
}}} // namespace cv { namespace gpu { namespace device
#endif // __OPENCV_GPU_UTILITY_DETAIL_HPP__
#endif // __OPENCV_GPU_REDUCTION_DETAIL_HPP__

38
modules/gpu/src/opencv2/gpu/device/detail/transform_detail.hpp
View File

@ -47,7 +47,7 @@
#include "../vec_traits.hpp"
#include "../functional.hpp"
namespace cv { namespace gpu { namespace device
namespace cv { namespace gpu { namespace device
{
namespace transform_detail
{
@ -203,7 +203,7 @@ namespace cv { namespace gpu { namespace device
};
template <typename T, typename D, typename UnOp, typename Mask>
__global__ static void transformSmart(const PtrStepSz<T> src_, PtrStep<D> dst_, const Mask mask, const UnOp op)
static __global__ void transformSmart(const PtrStepSz<T> src_, PtrStep<D> dst_, const Mask mask, const UnOp op)
{
typedef TransformFunctorTraits<UnOp> ft;
typedef typename UnaryReadWriteTraits<T, D, ft::smart_shift>::read_type read_type;
@ -239,10 +239,10 @@ namespace cv { namespace gpu { namespace device
}
template <typename T, typename D, typename UnOp, typename Mask>
static __global__ void transformSimple(const PtrStepSz<T> src, PtrStep<D> dst, const Mask mask, const UnOp op)
__global__ static void transformSimple(const PtrStepSz<T> src, PtrStep<D> dst, const Mask mask, const UnOp op)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < src.cols && y < src.rows && mask(y, x))
{
@ -251,7 +251,7 @@ namespace cv { namespace gpu { namespace device
}
template <typename T1, typename T2, typename D, typename BinOp, typename Mask>
__global__ static void transformSmart(const PtrStepSz<T1> src1_, const PtrStep<T2> src2_, PtrStep<D> dst_,
static __global__ void transformSmart(const PtrStepSz<T1> src1_, const PtrStep<T2> src2_, PtrStep<D> dst_,
const Mask mask, const BinOp op)
{
typedef TransformFunctorTraits<BinOp> ft;
@ -274,7 +274,7 @@ namespace cv { namespace gpu { namespace device
const read_type1 src1_n_el = ((const read_type1*)src1)[x];
const read_type2 src2_n_el = ((const read_type2*)src2)[x];
write_type dst_n_el = ((const write_type*)dst)[x];
OpUnroller<ft::smart_shift>::unroll(src1_n_el, src2_n_el, dst_n_el, mask, op, x_shifted, y);
((write_type*)dst)[x] = dst_n_el;
@ -291,11 +291,11 @@ namespace cv { namespace gpu { namespace device
}
template <typename T1, typename T2, typename D, typename BinOp, typename Mask>
static __global__ void transformSimple(const PtrStepSz<T1> src1, const PtrStep<T2> src2, PtrStep<D> dst,
static __global__ void transformSimple(const PtrStepSz<T1> src1, const PtrStep<T2> src2, PtrStep<D> dst,
const Mask mask, const BinOp op)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < src1.cols && y < src1.rows && mask(y, x))
{
@ -314,13 +314,13 @@ namespace cv { namespace gpu { namespace device
typedef TransformFunctorTraits<UnOp> ft;
const dim3 threads(ft::simple_block_dim_x, ft::simple_block_dim_y, 1);
const dim3 grid(divUp(src.cols, threads.x), divUp(src.rows, threads.y), 1);
const dim3 grid(divUp(src.cols, threads.x), divUp(src.rows, threads.y), 1);
transformSimple<T, D><<<grid, threads, 0, stream>>>(src, dst, mask, op);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaDeviceSynchronize() );
}
template <typename T1, typename T2, typename D, typename BinOp, typename Mask>
@ -329,13 +329,13 @@ namespace cv { namespace gpu { namespace device
typedef TransformFunctorTraits<BinOp> ft;
const dim3 threads(ft::simple_block_dim_x, ft::simple_block_dim_y, 1);
const dim3 grid(divUp(src1.cols, threads.x), divUp(src1.rows, threads.y), 1);
const dim3 grid(divUp(src1.cols, threads.x), divUp(src1.rows, threads.y), 1);
transformSimple<T1, T2, D><<<grid, threads, 0, stream>>>(src1, src2, dst, mask, op);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
template<> struct TransformDispatcher<true>
@ -347,7 +347,7 @@ namespace cv { namespace gpu { namespace device
StaticAssert<ft::smart_shift != 1>::check();
if (!isAligned(src.data, ft::smart_shift * sizeof(T)) || !isAligned(src.step, ft::smart_shift * sizeof(T)) ||
if (!isAligned(src.data, ft::smart_shift * sizeof(T)) || !isAligned(src.step, ft::smart_shift * sizeof(T)) ||
!isAligned(dst.data, ft::smart_shift * sizeof(D)) || !isAligned(dst.step, ft::smart_shift * sizeof(D)))
{
TransformDispatcher<false>::call(src, dst, op, mask, stream);
@ -355,7 +355,7 @@ namespace cv { namespace gpu { namespace device
}
const dim3 threads(ft::smart_block_dim_x, ft::smart_block_dim_y, 1);
const dim3 grid(divUp(src.cols, threads.x * ft::smart_shift), divUp(src.rows, threads.y), 1);
const dim3 grid(divUp(src.cols, threads.x * ft::smart_shift), divUp(src.rows, threads.y), 1);
transformSmart<T, D><<<grid, threads, 0, stream>>>(src, dst, mask, op);
cudaSafeCall( cudaGetLastError() );
@ -380,15 +380,15 @@ namespace cv { namespace gpu { namespace device
}
const dim3 threads(ft::smart_block_dim_x, ft::smart_block_dim_y, 1);
const dim3 grid(divUp(src1.cols, threads.x * ft::smart_shift), divUp(src1.rows, threads.y), 1);
const dim3 grid(divUp(src1.cols, threads.x * ft::smart_shift), divUp(src1.rows, threads.y), 1);
transformSmart<T1, T2, D><<<grid, threads, 0, stream>>>(src1, src2, dst, mask, op);
cudaSafeCall( cudaGetLastError() );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
cudaSafeCall( cudaDeviceSynchronize() );
}
};
};
} // namespace transform_detail
}}} // namespace cv { namespace gpu { namespace device

201
modules/gpu/src/opencv2/gpu/device/scan.hpp
View File

@ -43,124 +43,129 @@
#ifndef __OPENCV_GPU_SCAN_HPP__
#define __OPENCV_GPU_SCAN_HPP__
enum ScanKind { EXCLUSIVE = 0, INCLUSIVE = 1 };
#include "common.hpp"
template <ScanKind Kind, typename T, typename F> struct WarpScan
namespace cv { namespace gpu { namespace device
{
enum ScanKind { EXCLUSIVE = 0, INCLUSIVE = 1 };
template <ScanKind Kind, typename T, typename F> struct WarpScan
{
__device__ __forceinline__ WarpScan() {}
__device__ __forceinline__ WarpScan(const WarpScan& other) { (void)other; }
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
__device__ __forceinline__ WarpScan() {}
__device__ __forceinline__ WarpScan(const WarpScan& other) { (void)other; }
const unsigned int lane = idx & 31;
F op;
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
const unsigned int lane = idx & 31;
F op;
if ( lane >= 1) ptr [idx ] = op(ptr [idx - 1], ptr [idx]);
if ( lane >= 2) ptr [idx ] = op(ptr [idx - 2], ptr [idx]);
if ( lane >= 4) ptr [idx ] = op(ptr [idx - 4], ptr [idx]);
if ( lane >= 8) ptr [idx ] = op(ptr [idx - 8], ptr [idx]);
if ( lane >= 16) ptr [idx ] = op(ptr [idx - 16], ptr [idx]);
if ( lane >= 1) ptr [idx ] = op(ptr [idx - 1], ptr [idx]);
if ( lane >= 2) ptr [idx ] = op(ptr [idx - 2], ptr [idx]);
if ( lane >= 4) ptr [idx ] = op(ptr [idx - 4], ptr [idx]);
if ( lane >= 8) ptr [idx ] = op(ptr [idx - 8], ptr [idx]);
if ( lane >= 16) ptr [idx ] = op(ptr [idx - 16], ptr [idx]);
if( Kind == INCLUSIVE )
return ptr [idx];
else
return (lane > 0) ? ptr [idx - 1] : 0;
}
if( Kind == INCLUSIVE )
return ptr [idx];
else
return (lane > 0) ? ptr [idx - 1] : 0;
}
__device__ __forceinline__ unsigned int index(const unsigned int tid)
{
return tid;
}
__device__ __forceinline__ void init(volatile T *ptr){}
static const int warp_offset = 0;
typedef WarpScan<INCLUSIVE, T, F> merge;
};
template <ScanKind Kind , typename T, typename F> struct WarpScanNoComp
__device__ __forceinline__ unsigned int index(const unsigned int tid)
{
__device__ __forceinline__ WarpScanNoComp() {}
__device__ __forceinline__ WarpScanNoComp(const WarpScanNoComp& other) { (void)other; }
return tid;
}
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
const unsigned int lane = threadIdx.x & 31;
F op;
__device__ __forceinline__ void init(volatile T *ptr){}
ptr [idx ] = op(ptr [idx - 1], ptr [idx]);
ptr [idx ] = op(ptr [idx - 2], ptr [idx]);
ptr [idx ] = op(ptr [idx - 4], ptr [idx]);
ptr [idx ] = op(ptr [idx - 8], ptr [idx]);
ptr [idx ] = op(ptr [idx - 16], ptr [idx]);
static const int warp_offset = 0;
if( Kind == INCLUSIVE )
return ptr [idx];
else
return (lane > 0) ? ptr [idx - 1] : 0;
}
typedef WarpScan<INCLUSIVE, T, F> merge;
};
__device__ __forceinline__ unsigned int index(const unsigned int tid)
{
return (tid >> warp_log) * warp_smem_stride + 16 + (tid & warp_mask);
}
template <ScanKind Kind , typename T, typename F> struct WarpScanNoComp
{
__device__ __forceinline__ WarpScanNoComp() {}
__device__ __forceinline__ WarpScanNoComp(const WarpScanNoComp& other) { (void)other; }
__device__ __forceinline__ void init(volatile T *ptr)
{
ptr[threadIdx.x] = 0;
}
static const int warp_smem_stride = 32 + 16 + 1;
static const int warp_offset = 16;
static const int warp_log = 5;
static const int warp_mask = 31;
typedef WarpScanNoComp<INCLUSIVE, T, F> merge;
};
template <ScanKind Kind , typename T, typename Sc, typename F> struct BlockScan
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
__device__ __forceinline__ BlockScan() {}
__device__ __forceinline__ BlockScan(const BlockScan& other) { (void)other; }
const unsigned int lane = threadIdx.x & 31;
F op;
__device__ __forceinline__ T operator()(volatile T *ptr)
{
const unsigned int tid = threadIdx.x;
const unsigned int lane = tid & warp_mask;
const unsigned int warp = tid >> warp_log;
ptr [idx ] = op(ptr [idx - 1], ptr [idx]);
ptr [idx ] = op(ptr [idx - 2], ptr [idx]);
ptr [idx ] = op(ptr [idx - 4], ptr [idx]);
ptr [idx ] = op(ptr [idx - 8], ptr [idx]);
ptr [idx ] = op(ptr [idx - 16], ptr [idx]);
Sc scan;
typename Sc::merge merge_scan;
const unsigned int idx = scan.index(tid);
if( Kind == INCLUSIVE )
return ptr [idx];
else
return (lane > 0) ? ptr [idx - 1] : 0;
}
T val = scan(ptr, idx);
__syncthreads ();
__device__ __forceinline__ unsigned int index(const unsigned int tid)
{
return (tid >> warp_log) * warp_smem_stride + 16 + (tid & warp_mask);
}
if( warp == 0)
scan.init(ptr);
__syncthreads ();
__device__ __forceinline__ void init(volatile T *ptr)
{
ptr[threadIdx.x] = 0;
}
if( lane == 31 )
ptr [scan.warp_offset + warp ] = (Kind == INCLUSIVE) ? val : ptr [idx];
__syncthreads ();
static const int warp_smem_stride = 32 + 16 + 1;
static const int warp_offset = 16;
static const int warp_log = 5;
static const int warp_mask = 31;
if( warp == 0 )
merge_scan(ptr, idx);
__syncthreads();
typedef WarpScanNoComp<INCLUSIVE, T, F> merge;
};
if ( warp > 0)
val = ptr [scan.warp_offset + warp - 1] + val;
__syncthreads ();
template <ScanKind Kind , typename T, typename Sc, typename F> struct BlockScan
{
__device__ __forceinline__ BlockScan() {}
__device__ __forceinline__ BlockScan(const BlockScan& other) { (void)other; }
ptr[idx] = val;
__syncthreads ();
__device__ __forceinline__ T operator()(volatile T *ptr)
{
const unsigned int tid = threadIdx.x;
const unsigned int lane = tid & warp_mask;
const unsigned int warp = tid >> warp_log;
return val ;
}
Sc scan;
typename Sc::merge merge_scan;
const unsigned int idx = scan.index(tid);
static const int warp_log = 5;
static const int warp_mask = 31;
};
T val = scan(ptr, idx);
__syncthreads ();
#endif
if( warp == 0)
scan.init(ptr);
__syncthreads ();
if( lane == 31 )
ptr [scan.warp_offset + warp ] = (Kind == INCLUSIVE) ? val : ptr [idx];
__syncthreads ();
if( warp == 0 )
merge_scan(ptr, idx);
__syncthreads();
if ( warp > 0)
val = ptr [scan.warp_offset + warp - 1] + val;
__syncthreads ();
ptr[idx] = val;
__syncthreads ();
return val ;
}
static const int warp_log = 5;
static const int warp_mask = 31;
};
}}}
#endif // __OPENCV_GPU_SCAN_HPP__

4
modules/gpu/src/opencv2/gpu/device/static_check.hpp
View File

@ -60,10 +60,8 @@ namespace cv { namespace gpu
__OPENCV_GPU_HOST_DEVICE__ static void check() {};
};
}
using ::cv::gpu::device::Static;
}}
#undef __OPENCV_GPU_HOST_DEVICE__
#endif /* __OPENCV_GPU_GPU_DEVICE_STATIC_CHECK_HPP__ */
#endif /* __OPENCV_GPU_GPU_DEVICE_STATIC_CHECK_HPP__ */

2
modules/gpu/src/opencv2/gpu/device/utility.hpp
View File

@ -45,7 +45,7 @@
#include "saturate_cast.hpp"
#include "datamov_utils.hpp"
#include "detail/utility_detail.hpp"
#include "detail/reduction_detail.hpp"
namespace cv { namespace gpu { namespace device
{