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

other kernels now use row scheme

Browse Source
This commit is contained in:
Ilya Lavrenov
2014-05-14 18:56:25 +04:00
parent 15738bf7ef
commit bd5e4c6c49
9 changed files with 141 additions and 92 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
modules/core/src/opencl/inrange.cl
View File

@@ -52,37 +52,47 @@
__kernel void inrange(__global const uchar * src1ptr, int src1_step, int src1_offset,
__global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols,
#ifdef HAVE_SCALAR
__global const T * src2, __global const T * src3
__global const T * src2, __global const T * src3,
#else
__global const uchar * src2ptr, int src2_step, int src2_offset,
__global const uchar * src3ptr, int src3_step, int src3_offset
__global const uchar * src3ptr, int src3_step, int src3_offset,
#endif
)
int rowsPerWI)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y0 = get_global_id(1) * rowsPerWI;
if (x < dst_cols && y < dst_rows)
if (x < dst_cols)
{
int src1_index = mad24(y, src1_step, mad24(x, (int)sizeof(T) * cn, src1_offset));
int dst_index = mad24(y, dst_step, x + dst_offset);
__global const T * src1 = (__global const T *)(src1ptr + src1_index);
__global uchar * dst = dstptr + dst_index;
int src1_index = mad24(y0, src1_step, mad24(x, (int)sizeof(T) * cn, src1_offset));
int dst_index = mad24(y0, dst_step, x + dst_offset);
#ifndef HAVE_SCALAR
int src2_index = mad24(y, src2_step, mad24(x, (int)sizeof(T) * cn, src2_offset));
int src3_index = mad24(y, src3_step, mad24(x, (int)sizeof(T) * cn, src3_offset));
__global const T * src2 = (__global const T *)(src2ptr + src2_index);
__global const T * src3 = (__global const T *)(src3ptr + src3_index);
int src2_index = mad24(y0, src2_step, mad24(x, (int)sizeof(T) * cn, src2_offset));
int src3_index = mad24(y0, src3_step, mad24(x, (int)sizeof(T) * cn, src3_offset));
#endif
dst[0] = 255;
for (int y = y0, y1 = min(dst_rows, y0 + rowsPerWI); y < y1; ++y, src1_index += src1_step, dst_index += dst_step)
{
__global const T * src1 = (__global const T *)(src1ptr + src1_index);
__global uchar * dst = dstptr + dst_index;
#ifndef HAVE_SCALAR
__global const T * src2 = (__global const T *)(src2ptr + src2_index);
__global const T * src3 = (__global const T *)(src3ptr + src3_index);
#endif
for (int c = 0; c < cn; ++c)
if (src2[c] > src1[c] || src3[c] < src1[c])
{
dst[0] = 0;
break;
}
dst[0] = 255;
for (int c = 0; c < cn; ++c)
if (src2[c] > src1[c] || src3[c] < src1[c])
{
dst[0] = 0;
break;
}
#ifndef HAVE_SCALAR
src2_index += src2_step;
src3_index += src3_step;
#endif
}
}
}

22
modules/core/src/opencl/mixchannels.cl
View File

@@ -45,20 +45,28 @@
__global const uchar * src##i##ptr, int src##i##_step, int src##i##_offset,
#define DECLARE_OUTPUT_MAT(i) \
__global uchar * dst##i##ptr, int dst##i##_step, int dst##i##_offset,
#define DECLARE_INDEX(i) \
int src##i##_index = mad24(src##i##_step, y0, mad24(x, (int)sizeof(T) * scn##i, src##i##_offset)); \
int dst##i##_index = mad24(dst##i##_step, y0, mad24(x, (int)sizeof(T) * dcn##i, dst##i##_offset));
#define PROCESS_ELEM(i) \
int src##i##_index = mad24(src##i##_step, y, mad24(x, (int)sizeof(T) * scn##i, src##i##_offset)); \
__global const T * src##i = (__global const T *)(src##i##ptr + src##i##_index); \
int dst##i##_index = mad24(dst##i##_step, y, mad24(x, (int)sizeof(T) * dcn##i, dst##i##_offset)); \
__global T * dst##i = (__global T *)(dst##i##ptr + dst##i##_index); \
dst##i[0] = src##i[0];
dst##i[0] = src##i[0]; \
src##i##_index += src##i##_step; \
dst##i##_index += dst##i##_step;
__kernel void mixChannels(DECLARE_INPUT_MATS DECLARE_OUTPUT_MATS int rows, int cols)
__kernel void mixChannels(DECLARE_INPUT_MAT_N DECLARE_OUTPUT_MAT_N int rows, int cols, int rowsPerWI)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y0 = get_global_id(1) * rowsPerWI;
if (x < cols && y < rows)
if (x < cols)
{
PROCESS_ELEMS
DECLARE_INDEX_N
for (int y = y0, y1 = min(y0 + rowsPerWI, rows); y < y1; ++y)
{
PROCESS_ELEM_N
}
}
}

28
modules/core/src/opencl/mulspectrums.cl
View File

@@ -56,26 +56,30 @@ inline float2 conjf(float2 a)
__kernel void mulAndScaleSpectrums(__global const uchar * src1ptr, int src1_step, int src1_offset,
__global const uchar * src2ptr, int src2_step, int src2_offset,
__global uchar * dstptr, int dst_step, int dst_offset,
int dst_rows, int dst_cols)
int dst_rows, int dst_cols, int rowsPerWI)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y0 = get_global_id(1) * rowsPerWI;
if (x < dst_cols && y < dst_rows)
if (x < dst_cols)
{
int src1_index = mad24(y, src1_step, mad24(x, (int)sizeof(float2), src1_offset));
int src2_index = mad24(y, src2_step, mad24(x, (int)sizeof(float2), src2_offset));
int dst_index = mad24(y, dst_step, mad24(x, (int)sizeof(float2), dst_offset));
int src1_index = mad24(y0, src1_step, mad24(x, (int)sizeof(float2), src1_offset));
int src2_index = mad24(y0, src2_step, mad24(x, (int)sizeof(float2), src2_offset));
int dst_index = mad24(y0, dst_step, mad24(x, (int)sizeof(float2), dst_offset));
float2 src0 = *(__global const float2 *)(src1ptr + src1_index);
float2 src1 = *(__global const float2 *)(src2ptr + src2_index);
__global float2 * dst = (__global float2 *)(dstptr + dst_index);
for (int y = y0, y1 = min(dst_rows, y0 + rowsPerWI); y < y1; ++y,
src1_index += src1_step, src2_index += src2_step, dst_index += dst_step)
{
float2 src0 = *(__global const float2 *)(src1ptr + src1_index);
float2 src1 = *(__global const float2 *)(src2ptr + src2_index);
__global float2 * dst = (__global float2 *)(dstptr + dst_index);
#ifdef CONJ
float2 v = cmulf(src0, conjf(src1));
float2 v = cmulf(src0, conjf(src1));
#else
float2 v = cmulf(src0, src1);
float2 v = cmulf(src0, src1);
#endif
dst[0] = v;
dst[0] = v;
}
}
}

11
modules/core/src/opencl/set_identity.cl
View File

@@ -56,15 +56,16 @@
#endif
__kernel void setIdentity(__global uchar * srcptr, int src_step, int src_offset, int rows, int cols,
ST scalar_)
ST scalar_, int rowsPerWI)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y0 = get_global_id(1) * rowsPerWI;
if (x < cols && y < rows)
if (x < cols)
{
int src_index = mad24(y, src_step, mad24(x, TSIZE, src_offset));
int src_index = mad24(y0, src_step, mad24(x, TSIZE, src_offset));
storepix(x == y ? scalar : (T)(0), srcptr + src_index);
for (int y = y0, y1 = min(rows, y0 + rowsPerWI); y < y1; ++y, src_index += src_step)
storepix(x == y ? scalar : (T)(0), srcptr + src_index);
}
}

52
modules/core/src/opencl/split_merge.cl
View File

@@ -44,42 +44,58 @@
#ifdef OP_MERGE
#define DECLARE_SRC_PARAM(index) __global const uchar * src##index##ptr, int src##index##_step, int src##index##_offset,
#define DECLARE_DATA(index) __global const T * src##index = \
(__global T *)(src##index##ptr + mad24(src##index##_step, y, mad24(x, (int)sizeof(T) * scn##index, src##index##_offset)));
#define PROCESS_ELEM(index) dst[index] = src##index[0];
#define DECLARE_INDEX(index) int src##index##_index = mad24(src##index##_step, y0, mad24(x, (int)sizeof(T) * scn##index, src##index##_offset));
#define PROCESS_ELEM(index) \
__global const T * src##index = (__global const T *)(src##index##ptr + src##index##_index); \
dst[index] = src##index[0]; \
src##index##_index += src##index##_step;
__kernel void merge(DECLARE_SRC_PARAMS_N
__global uchar * dstptr, int dst_step, int dst_offset,
int rows, int cols)
int rows, int cols, int rowsPerWI)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y0 = get_global_id(1) * rowsPerWI;
if (x < cols && y < rows)
if (x < cols)
{
DECLARE_DATA_N
__global T * dst = (__global T *)(dstptr + mad24(dst_step, y, mad24(x, (int)sizeof(T) * cn, dst_offset)));
PROCESS_ELEMS_N
DECLARE_INDEX_N
int dst_index = mad24(dst_step, y0, mad24(x, (int)sizeof(T) * cn, dst_offset));
for (int y = y0, y1 = min(rows, y0 + rowsPerWI); y < y1; ++y, dst_index += dst_step)
{
__global T * dst = (__global T *)(dstptr + dst_index);
PROCESS_ELEMS_N
}
}
}
#elif defined OP_SPLIT
#define DECLARE_DST_PARAM(index) , __global uchar * dst##index##ptr, int dst##index##_step, int dst##index##_offset
#define DECLARE_DATA(index) __global T * dst##index = \
(__global T *)(dst##index##ptr + mad24(y, dst##index##_step, mad24(x, (int)sizeof(T), dst##index##_offset)));
#define PROCESS_ELEM(index) dst##index[0] = src[index];
#define DECLARE_INDEX(index) int dst##index##_index = mad24(y0, dst##index##_step, mad24(x, (int)sizeof(T), dst##index##_offset));
#define PROCESS_ELEM(index) \
__global T * dst##index = (__global T *)(dst##index##ptr + dst##index##_index); \
dst##index[0] = src[index]; \
dst##index##_index += dst##index##_step;
__kernel void split(__global uchar* srcptr, int src_step, int src_offset, int rows, int cols DECLARE_DST_PARAMS)
__kernel void split(__global uchar* srcptr, int src_step, int src_offset, int rows, int cols DECLARE_DST_PARAMS, int rowsPerWI)
{
int x = get_global_id(0);
int y = get_global_id(1);
int y0 = get_global_id(1) * rowsPerWI;
if (x < cols && y < rows)
if (x < cols)
{
DECLARE_DATA_N
__global const T * src = (__global const T *)(srcptr + mad24(y, src_step, mad24(x, cn * (int)sizeof(T), src_offset)));
PROCESS_ELEMS_N
DECLARE_INDEX_N
int src_index = mad24(y0, src_step, mad24(x, cn * (int)sizeof(T), src_offset));
for (int y = y0, y1 = min(rows, y0 + rowsPerWI); y < y1; ++y, src_index += src_step)
{
__global const T * src = (__global const T *)(srcptr + src_index);
PROCESS_ELEMS_N
}
}
}