Skip to content

O
opencv
Commit 3ebec744 authored by Vladislav Vinogradov's avatar Vladislav Vinogradov
Browse files
Options

minor refactoring: 

moved lbp.hpp to src/cuda folder
added missing cv::gpu::device namespace
deleted whitespaces
parent a703df54
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 132 additions and 130 deletions
modules/gpu/src/cuda/lbp.cu
View file @ 3ebec744
......@@ -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 */
\ No newline at end of file
#endif /* CUDA_DISABLER */
modules/gpu/src/opencv2/gpu/device/detail/color_detail.hpp
View file @ 3ebec744
......@@ -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__
modules/gpu/src/opencv2/gpu/device/detail/transform_detail.hpp
View file @ 3ebec744
......@@ -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
......
modules/gpu/src/opencv2/gpu/device/scan.hpp
View file @ 3ebec744
......@@ -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
{
__device__ __forceinline__ WarpScan() {}
__device__ __forceinline__ WarpScan(const WarpScan& other) { (void)other; }
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
const unsigned int lane = idx & 31;
F op;
namespace cv { namespace gpu { namespace device
{
enum ScanKind { EXCLUSIVE = 0, INCLUSIVE = 1 };
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]);
template <ScanKind Kind, typename T, typename F> struct WarpScan
{
__device__ __forceinline__ WarpScan() {}
__device__ __forceinline__ WarpScan(const WarpScan& other) { (void)other; }
if( Kind == INCLUSIVE )
return ptr [idx];
else
return (lane > 0) ? ptr [idx - 1] : 0;
}
__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( 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__ unsigned int index(const unsigned int tid)
{
return tid;
}
__device__ __forceinline__ void init(volatile T *ptr){}
__device__ __forceinline__ void init(volatile T *ptr){}
static const int warp_offset = 0;
static const int warp_offset = 0;
typedef WarpScan<INCLUSIVE, T, F> merge;
};
typedef WarpScan<INCLUSIVE, T, F> merge;
};
template <ScanKind Kind , typename T, typename F> struct WarpScanNoComp
{
__device__ __forceinline__ WarpScanNoComp() {}
__device__ __forceinline__ WarpScanNoComp(const WarpScanNoComp& other) { (void)other; }
template <ScanKind Kind , typename T, typename F> struct WarpScanNoComp
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
const unsigned int lane = threadIdx.x & 31;
F op;
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]);
if( Kind == INCLUSIVE )
return ptr [idx];
else
return (lane > 0) ? ptr [idx - 1] : 0;
}
__device__ __forceinline__ unsigned int index(const unsigned int tid)
{
__device__ __forceinline__ WarpScanNoComp() {}
__device__ __forceinline__ WarpScanNoComp(const WarpScanNoComp& other) { (void)other; }
__device__ __forceinline__ T operator()( volatile T *ptr , const unsigned int idx)
{
const unsigned int lane = threadIdx.x & 31;
F op;
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]);
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 >> warp_log) * warp_smem_stride + 16 + (tid & warp_mask);
}
__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
return (tid >> warp_log) * warp_smem_stride + 16 + (tid & warp_mask);
}
__device__ __forceinline__ void init(volatile T *ptr)
{
__device__ __forceinline__ BlockScan() {}
__device__ __forceinline__ BlockScan(const BlockScan& other) { (void)other; }
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;
};
__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;
template <ScanKind Kind , typename T, typename Sc, typename F> struct BlockScan
{
__device__ __forceinline__ BlockScan() {}
__device__ __forceinline__ BlockScan(const BlockScan& other) { (void)other; }
__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;
Sc scan;
typename Sc::merge merge_scan;
const unsigned int idx = scan.index(tid);
Sc scan;
typename Sc::merge merge_scan;
const unsigned int idx = scan.index(tid);
T val = scan(ptr, idx);
__syncthreads ();
T val = scan(ptr, idx);
__syncthreads ();
if( warp == 0)
scan.init(ptr);
__syncthreads ();
if( warp == 0)
scan.init(ptr);
__syncthreads ();
if( lane == 31 )
ptr [scan.warp_offset + warp ] = (Kind == INCLUSIVE) ? val : ptr [idx];
__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 )
merge_scan(ptr, idx);
__syncthreads();
if ( warp > 0)
val = ptr [scan.warp_offset + warp - 1] + val;
__syncthreads ();
if ( warp > 0)
val = ptr [scan.warp_offset + warp - 1] + val;
__syncthreads ();
ptr[idx] = val;
__syncthreads ();
ptr[idx] = val;
__syncthreads ();
return val ;
}
return val ;
}
static const int warp_log = 5;
static const int warp_mask = 31;
};
static const int warp_log = 5;
static const int warp_mask = 31;
};
}}}
#endif
\ No newline at end of file
#endif // __OPENCV_GPU_SCAN_HPP__
modules/gpu/src/opencv2/gpu/device/static_check.hpp
View file @ 3ebec744
......@@ -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__ */
\ No newline at end of file
#endif /* __OPENCV_GPU_GPU_DEVICE_STATIC_CHECK_HPP__ */
modules/gpu/src/opencv2/gpu/device/utility.hpp
View file @ 3ebec744
......@@ -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
{
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment