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

cv::transpose

Browse Source
This commit is contained in:
Ilya Lavrenov
2014-06-27 12:44:32 +04:00
parent cbd0511a56
commit 9c8b9fc733
2 changed files with 5 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
modules/core/src/matrix.cpp
View File

@@ -3002,8 +3002,8 @@ static bool ocl_transpose( InputArray _src, OutputArray _dst )
k.args(ocl::KernelArg::ReadOnly(src), k.args(ocl::KernelArg::ReadOnly(src),
ocl::KernelArg::WriteOnlyNoSize(dst)); ocl::KernelArg::WriteOnlyNoSize(dst));
size_t localsize[3] = { TILE_DIM, BLOCK_ROWS, 1 }; size_t localsize[2] = { TILE_DIM, BLOCK_ROWS };
size_t globalsize[3] = { src.cols, inplace ? src.rows : divUp(src.rows, TILE_DIM) * BLOCK_ROWS, 1 }; size_t globalsize[2] = { src.cols, inplace ? src.rows : divUp(src.rows, TILE_DIM) * BLOCK_ROWS };
return k.run(2, globalsize, localsize, false); return k.run(2, globalsize, localsize, false);
} }

4
modules/core/src/opencl/transpose.cl
View File

@@ -53,7 +53,7 @@
#define TSIZE ((int)sizeof(T1)*3) #define TSIZE ((int)sizeof(T1)*3)
#endif #endif
#define LDS_STEP TILE_DIM #define LDS_STEP (TILE_DIM + 1)
__kernel void transpose(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols, __kernel void transpose(__global const uchar * srcptr, int src_step, int src_offset, int src_rows, int src_cols,
__global uchar * dstptr, int dst_step, int dst_offset) __global uchar * dstptr, int dst_step, int dst_offset)
@@ -90,6 +90,7 @@ __kernel void transpose(__global const uchar * srcptr, int src_step, int src_off
{ {
int index_src = mad24(y, src_step, mad24(x, TSIZE, src_offset)); int index_src = mad24(y, src_step, mad24(x, TSIZE, src_offset));
#pragma unroll
for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS) for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS)
if (y + i < src_rows) if (y + i < src_rows)
{ {
@@ -103,6 +104,7 @@ __kernel void transpose(__global const uchar * srcptr, int src_step, int src_off
{ {
int index_dst = mad24(y_index, dst_step, mad24(x_index, TSIZE, dst_offset)); int index_dst = mad24(y_index, dst_step, mad24(x_index, TSIZE, dst_offset));
#pragma unroll
for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS) for (int i = 0; i < TILE_DIM; i += BLOCK_ROWS)
if ((y_index + i) < src_cols) if ((y_index + i) < src_cols)
{ {