generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

optimized gpumat::copyTo()

Browse Source
This commit is contained in:
Andrey Morozov 2010-07-27 09:25:46 +00:00
parent 290c967b8f
commit 5a88e8cf61
1 changed files with 23 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

45
modules/gpu/src/cuda/matrix_operations.cu
View File

@ -54,28 +54,6 @@ namespace mat_operators
{ {
__constant__ double scalar_d[4]; __constant__ double scalar_d[4];
///////////////////////////////////////////////////////////////////////////
////////////////////////////////// CopyTo /////////////////////////////////
///////////////////////////////////////////////////////////////////////////
template<typename T>
__global__ void kernel_copy_to_with_mask(T * mat_src, T * mat_dst, const unsigned char * mask, int cols, int rows, int step_mat, int step_mask, int channels)
{
size_t x = blockIdx.x * blockDim.x + threadIdx.x;
size_t y = blockIdx.y * blockDim.y + threadIdx.y;
if ((x < cols * channels ) && (y < rows))
if (mask[y * step_mask + x / channels] != 0)
{
size_t idx = y * (step_mat / sizeof(T)) + x;
mat_dst[idx] = mat_src[idx];
}
}
///////////////////////////////////////////////////////////////////////////
////////////////////////////////// SetTo //////////////////////////////////
///////////////////////////////////////////////////////////////////////////
template <typename T> template <typename T>
class shift_and_sizeof; class shift_and_sizeof;
@ -129,6 +107,29 @@ namespace mat_operators
enum { shift = 3 }; enum { shift = 3 };
}; };
///////////////////////////////////////////////////////////////////////////
////////////////////////////////// CopyTo /////////////////////////////////
///////////////////////////////////////////////////////////////////////////
template<typename T>
__global__ void kernel_copy_to_with_mask(T * mat_src, T * mat_dst, const unsigned char * mask, int cols, int rows, int step_mat, int step_mask, int channels)
{
size_t x = blockIdx.x * blockDim.x + threadIdx.x;
size_t y = blockIdx.y * blockDim.y + threadIdx.y;
if ((x < cols * channels ) && (y < rows))
if (mask[y * step_mask + x / channels] != 0)
{
size_t idx = y * ( step_mat >> shift_and_sizeof<T>::shift ) + x;
mat_dst[idx] = mat_src[idx];
}
}
///////////////////////////////////////////////////////////////////////////
////////////////////////////////// SetTo //////////////////////////////////
///////////////////////////////////////////////////////////////////////////
template<typename T> template<typename T>
__global__ void kernel_set_to_without_mask(T * mat, int cols, int rows, int step, int channels) __global__ void kernel_set_to_without_mask(T * mat, int cols, int rows, int step, int channels)
{ {