Skip to content

O
opencv
Commit 5726e80f authored by Andrey Pavlenko's avatar Andrey Pavlenko Committed by OpenCV Buildbot
Browse files
Options

Merge pull request #2475 from ilya-lavrenov:ocl_2.4_fix

parents 836635d2 61c347fb
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 72 additions and 628 deletions
modules/nonfree/perf/perf_main.cpp
View file @ 5726e80f
......@@ -5,7 +5,7 @@ static const char * impls[] = {
#ifdef HAVE_CUDA
"cuda",
#endif
#ifdef HAVE_OPENCL
#ifdef HAVE_OPENCV_OCL
"ocl",
#endif
"plain"
......
modules/nonfree/perf/perf_surf_ocl.cpp
View file @ 5726e80f
......@@ -59,7 +59,7 @@ typedef perf::TestBaseWithParam<std::string> OCL_SURF;
#define OCL_TEST_CYCLE() for( ; startTimer(), next(); cv::ocl::finish(), stopTimer())
PERF_TEST_P(OCL_SURF, with_data_transfer, testing::Values(SURF_IMAGES))
PERF_TEST_P(OCL_SURF, DISABLED_with_data_transfer, testing::Values(SURF_IMAGES))
{
string filename = getDataPath(GetParam());
Mat src = imread(filename, IMREAD_GRAYSCALE);
......@@ -94,7 +94,7 @@ PERF_TEST_P(OCL_SURF, with_data_transfer, testing::Values(SURF_IMAGES))
SANITY_CHECK_NOTHING();
}
PERF_TEST_P(OCL_SURF, without_data_transfer, testing::Values(SURF_IMAGES))
PERF_TEST_P(OCL_SURF, DISABLED_without_data_transfer, testing::Values(SURF_IMAGES))
{
string filename = getDataPath(GetParam());
Mat src = imread(filename, IMREAD_GRAYSCALE);
......
modules/ocl/src/arithm.cpp
View file @ 5726e80f
This diff is collapsed.
modules/ocl/src/gftt.cpp
View file @ 5726e80f
......@@ -146,34 +146,33 @@ static void minMaxEig_caller(const oclMat &src, oclMat &dst, oclMat & tozero)
CV_Assert(groupnum != 0);
int dbsize = groupnum * 2 * src.elemSize();
ensureSizeIsEnough(1, dbsize, CV_8UC1, dst);
cl_mem dst_data = reinterpret_cast<cl_mem>(dst.data);
int all_cols = src.step / src.elemSize();
int pre_cols = (src.offset % src.step) / src.elemSize();
int sec_cols = all_cols - (src.offset % src.step + src.cols * src.elemSize() - 1) / src.elemSize() - 1;
int invalid_cols = pre_cols + sec_cols;
int cols = all_cols - invalid_cols , elemnum = cols * src.rows;
int offset = src.offset / src.elemSize();
int vElemSize = src.elemSize1();
int src_step = src.step / vElemSize, src_offset = src.offset / vElemSize;
int total = src.size().area();
{// first parallel pass
{
// first parallel pass
vector<pair<size_t , const void *> > args;
args.push_back( make_pair( sizeof(cl_mem) , (void *)&src.data));
args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&cols ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&invalid_cols ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&offset));
args.push_back( make_pair( sizeof(cl_int) , (void *)&elemnum));
args.push_back( make_pair( sizeof(cl_int) , (void *)&src_step));
args.push_back( make_pair( sizeof(cl_int) , (void *)&src_offset));
args.push_back( make_pair( sizeof(cl_int) , (void *)&src.rows ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&src.cols ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&total));
args.push_back( make_pair( sizeof(cl_int) , (void *)&groupnum));
args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data ));
size_t globalThreads[3] = {groupnum * 256, 1, 1};
size_t localThreads[3] = {256, 1, 1};
openCLExecuteKernel(src.clCxt, &arithm_minMax, "arithm_op_minMax", globalThreads, localThreads,
args, -1, -1, "-D T=float -D DEPTH_5");
args, -1, -1, "-D T=float -D DEPTH_5 -D vlen=1");
}
{// run final "serial" kernel to find accumulate results from threads and reset corner counter
{
// run final "serial" kernel to find accumulate results from threads and reset corner counter
vector<pair<size_t , const void *> > args;
args.push_back( make_pair( sizeof(cl_mem) , (void *)&dst_data ));
args.push_back( make_pair( sizeof(cl_int) , (void *)&groupnum ));
......
modules/ocl/src/opencl/arithm_bitwise_binary.cl → modules/ocl/src/opencl/arithm_bitwise.cl
View file @ 5726e80f
......@@ -48,35 +48,46 @@
/////////////////////////////////////////// bitwise_binary //////////////////////////////////////////
/////////////////////////////////////////////////////////////////////////////////////////////////////
__kernel void arithm_bitwise_binary(__global uchar * src1, int src1_step, int src1_offset,
__global uchar * src2, int src2_step, int src2_offset,
__global uchar * dst, int dst_step, int dst_offset,
int cols, int rows)
__kernel void arithm_bitwise(__global uchar * src1ptr, int src1_step, int src1_offset,
#ifdef OP_BINARY
__global uchar * src2ptr, int src2_step, int src2_offset,
#elif defined HAVE_SCALAR
T scalar,
#endif
#ifdef HAVE_MASK
__global uchar * mask, int mask_step, int mask_offset,
#endif
__global uchar * dstptr, int dst_step, int dst_offset, int dst_rows, int dst_cols)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
if (x < dst_cols && y < dst_rows)
{
#if elemSize > 1
x *= elemSize;
#ifdef HAVE_MASK
mask += mad24(y, mask_step, x + mask_offset);
if (mask[0])
#endif
int src1_index = mad24(y, src1_step, x + src1_offset);
int src2_index = mad24(y, src2_step, x + src2_offset);
int dst_index = mad24(y, dst_step, x + dst_offset);
#if elemSize > 1
#pragma unroll
for (int i = 0; i < elemSize; i += vlen)
{
ucharv t0 = vloadn(0, src1 + src1_index + i);
ucharv t1 = vloadn(0, src2 + src2_index + i);
ucharv t2 = t0 Operation t1;
int src1_index = mad24(y, src1_step, mad24(x, (int)sizeof(T), src1_offset));
#ifdef OP_BINARY
int src2_index = mad24(y, src2_step, mad24(x, (int)sizeof(T), src2_offset));
#endif
int dst_index = mad24(y, dst_step, mad24(x, (int)sizeof(T), dst_offset));
vstoren(t2, 0, dst + dst_index + i);
}
__global const T * src1 = (__global const T *)(src1ptr + src1_index);
#ifdef OP_BINARY
__global const T * src2 = (__global const T *)(src2ptr + src2_index);
#endif
__global T * dst = (__global T *)(dstptr + dst_index);
#ifdef OP_BINARY
dst[0] = src1[0] Operation src2[0];
#elif defined HAVE_SCALAR
dst[0] = src1[0] Operation scalar;
#else
dst[dst_index] = src1[src1_index] Operation src2[src2_index];
dst[0] = Operation src1[0];
#endif
}
}
}
modules/ocl/src/opencl/arithm_bitwise_binary_mask.cl deleted 100644 → 0
View file @ 836635d2
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jiang Liyuan, jlyuan001.good@163.com
// Peng Xiao, pengxiao@outlook.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
//////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////bitwise_binary////////////////////////////////////////////
//////////////////////////////////////////////////////////////////////////////////////////////////////
__kernel void arithm_bitwise_binary_mask(__global uchar * src1, int src1_step, int src1_offset,
__global uchar * src2, int src2_step, int src2_offset,
__global uchar * mask, int mask_step, int mask_offset,
__global uchar * dst, int dst_step, int dst_offset,
int cols1, int rows)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols1 && y < rows)
{
int mask_index = mad24(y, mask_step, mask_offset + x);
if (mask[mask_index])
{
#if elemSize > 1
x *= elemSize;
#endif
int src1_index = mad24(y, src1_step, x + src1_offset);
int src2_index = mad24(y, src2_step, x + src2_offset);
int dst_index = mad24(y, dst_step, x + dst_offset);
#if elemSize > 1
#pragma unroll
for (int i = 0; i < elemSize; i += vlen)
{
ucharv t0 = vloadn(0, src1 + src1_index + i);
ucharv t1 = vloadn(0, src2 + src2_index + i);
ucharv t2 = t0 Operation t1;
vstoren(t2, 0, dst + dst_index + i);
}
#else
dst[dst_index] = src1[src1_index] Operation src2[src2_index];
#endif
}
}
}
modules/ocl/src/opencl/arithm_bitwise_binary_scalar.cl deleted 100644 → 0
View file @ 836635d2
////////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jiang Liyuan, jlyuan001.good@163.com
// Peng Xiao, pengxiao@outlook.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//
///////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////bitwise_binary/////////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////////////////////////
__kernel void arithm_bitwise_binary_scalar(
__global uchar *src1, int src1_step, int src1_offset,
__global uchar *src2,
__global uchar *dst, int dst_step, int dst_offset,
int cols, int rows)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
#if elemSize > 1
x *= elemSize;
#endif
int src1_index = mad24(y, src1_step, src1_offset + x);
int dst_index = mad24(y, dst_step, dst_offset + x);
#if elemSize > 1
#pragma unroll
for (int i = 0; i < elemSize; i += vlen)
{
ucharv t0 = vloadn(0, src1 + src1_index + i);
ucharv t1 = vloadn(0, src2 + i);
ucharv t2 = t0 Operation t1;
vstoren(t2, 0, dst + dst_index + i);
}
#else
dst[dst_index] = src1[src1_index] Operation src2[0];
#endif
}
}
modules/ocl/src/opencl/arithm_bitwise_binary_scalar_mask.cl deleted 100644 → 0
View file @ 836635d2
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jiang Liyuan, jlyuan001.good@163.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
//////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////bitwise_binary////////////////////////////////////////////
//////////////////////////////////////////////////////////////////////////////////////////////////////
__kernel void arithm_bitwise_binary_scalar_mask(__global uchar *src1, int src1_step, int src1_offset,
__global uchar *src2,
__global uchar *mask, int mask_step, int mask_offset,
__global uchar *dst, int dst_step, int dst_offset,
int cols, int rows)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int mask_index = mad24(y, mask_step, x + mask_offset);
if (mask[mask_index])
{
#if elemSize > 1
x *= elemSize;
#endif
int src1_index = mad24(y, src1_step, x + src1_offset);
int dst_index = mad24(y, dst_step, x + dst_offset);
#if elemSize > 1
#pragma unroll
for (int i = 0; i < elemSize; i += vlen)
{
ucharv t0 = vloadn(0, src1 + src1_index + i);
ucharv t1 = vloadn(0, src2 + i);
ucharv t2 = t0 Operation t1;
vstoren(t2, 0, dst + dst_index + i);
}
#else
dst[dst_index] = src1[src1_index] Operation src2[0];
#endif
}
}
}
modules/ocl/src/opencl/arithm_bitwise_not.cl deleted 100644 → 0
View file @ 836635d2
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jiang Liyuan, jlyuan001.good@163.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifdef DOUBLE_SUPPORT
#ifdef cl_amd_fp64
#pragma OPENCL EXTENSION cl_amd_fp64:enable
#elif defined (cl_khr_fp64)
#pragma OPENCL EXTENSION cl_khr_fp64:enable
#endif
#endif
///////////////////////////////////////////////////////////////////////////////////////////////////////
////////////////////////////////////////////BITWISE_NOT////////////////////////////////////////////////
///////////////////////////////////////////////////////////////////////////////////////////////////////
__kernel void arithm_bitwise_not_D0 (__global uchar *src1, int src1_step, int src1_offset,
__global uchar *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 2;
int src1_index = mad24(y, src1_step, x + src1_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + x);
uchar4 src1_data = vload4(0, src1 + src1_index);
uchar4 dst_data = vload4(0, dst + dst_index);
uchar4 tmp_data = ~src1_data;
dst_data.x = dst_index + 0 < dst_end ? tmp_data.x : dst_data.x;
dst_data.y = dst_index + 1 < dst_end ? tmp_data.y : dst_data.y;
dst_data.z = dst_index + 2 < dst_end ? tmp_data.z : dst_data.z;
dst_data.w = dst_index + 3 < dst_end ? tmp_data.w : dst_data.w;
vstore4(dst_data, 0, dst + dst_index);
}
}
__kernel void arithm_bitwise_not_D1 (__global char *src1, int src1_step, int src1_offset,
__global char *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 2;
int src1_index = mad24(y, src1_step, x + src1_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + x);
char4 src1_data = vload4(0, src1 + src1_index);
char4 dst_data = vload4(0, dst + dst_index);
char4 tmp_data = ~src1_data;
dst_data.x = dst_index + 0 < dst_end ? tmp_data.x : dst_data.x;
dst_data.y = dst_index + 1 < dst_end ? tmp_data.y : dst_data.y;
dst_data.z = dst_index + 2 < dst_end ? tmp_data.z : dst_data.z;
dst_data.w = dst_index + 3 < dst_end ? tmp_data.w : dst_data.w;
vstore4(dst_data, 0, dst + dst_index);
}
}
__kernel void arithm_bitwise_not_D2 (__global ushort *src1, int src1_step, int src1_offset,
__global ushort *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 2;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align ((dst_offset >> 1) & 3)
int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffff8);
ushort4 src1_data = vload4(0, (__global ushort *)((__global char *)src1 + src1_index));
ushort4 dst_data = *((__global ushort4 *)((__global char *)dst + dst_index));
ushort4 tmp_data = ~ src1_data;
dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
dst_data.z = ((dst_index + 4 >= dst_start) && (dst_index + 4 < dst_end)) ? tmp_data.z : dst_data.z;
dst_data.w = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end)) ? tmp_data.w : dst_data.w;
*((__global ushort4 *)((__global char *)dst + dst_index)) = dst_data;
}
}
__kernel void arithm_bitwise_not_D3 (__global short *src1, int src1_step, int src1_offset,
__global short *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
x = x << 2;
#ifdef dst_align
#undef dst_align
#endif
#define dst_align ((dst_offset >> 1) & 3)
int src1_index = mad24(y, src1_step, (x << 1) + src1_offset - (dst_align << 1));
int dst_start = mad24(y, dst_step, dst_offset);
int dst_end = mad24(y, dst_step, dst_offset + dst_step1);
int dst_index = mad24(y, dst_step, dst_offset + (x << 1) & (int)0xfffffff8);
short4 src1_data = vload4(0, (__global short *)((__global char *)src1 + src1_index));
short4 dst_data = *((__global short4 *)((__global char *)dst + dst_index));
short4 tmp_data = ~ src1_data;
dst_data.x = ((dst_index + 0 >= dst_start) && (dst_index + 0 < dst_end)) ? tmp_data.x : dst_data.x;
dst_data.y = ((dst_index + 2 >= dst_start) && (dst_index + 2 < dst_end)) ? tmp_data.y : dst_data.y;
dst_data.z = ((dst_index + 4 >= dst_start) && (dst_index + 4 < dst_end)) ? tmp_data.z : dst_data.z;
dst_data.w = ((dst_index + 6 >= dst_start) && (dst_index + 6 < dst_end)) ? tmp_data.w : dst_data.w;
*((__global short4 *)((__global char *)dst + dst_index)) = dst_data;
}
}
__kernel void arithm_bitwise_not_D4 (__global int *src1, int src1_step, int src1_offset,
__global int *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src1_index = mad24(y, src1_step, (x << 2) + src1_offset);
int dst_index = mad24(y, dst_step, (x << 2) + dst_offset);
int data1 = *((__global int *)((__global char *)src1 + src1_index));
int tmp = ~ data1;
*((__global int *)((__global char *)dst + dst_index)) = tmp;
}
}
__kernel void arithm_bitwise_not_D5 (__global char *src, int src_step, int src_offset,
__global char *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src_index = mad24(y, src_step, (x << 2) + src_offset);
int dst_index = mad24(y, dst_step, (x << 2) + dst_offset);
char4 data;
data = *((__global char4 *)((__global char *)src + src_index));
data = ~ data;
*((__global char4 *)((__global char *)dst + dst_index)) = data;
}
}
#if defined (DOUBLE_SUPPORT)
__kernel void arithm_bitwise_not_D6 (__global char *src, int src_step, int src_offset,
__global char *dst, int dst_step, int dst_offset,
int rows, int cols, int dst_step1)
{
int x = get_global_id(0);
int y = get_global_id(1);
if (x < cols && y < rows)
{
int src_index = mad24(y, src_step, (x << 3) + src_offset);
int dst_index = mad24(y, dst_step, (x << 3) + dst_offset);
char8 data;
data = *((__global char8 *)((__global char *)src + src_index));
data = ~ data;
*((__global char8 *)((__global char *)dst + dst_index)) = data;
}
}
#endif
modules/ocl/src/opencl/arithm_minMax.cl
View file @ 5726e80f
......@@ -63,81 +63,31 @@
/**************************************Array minMax**************************************/
__kernel void arithm_op_minMax(__global const T * src, __global T * dst,
int cols, int invalid_cols, int offset, int elemnum, int groupnum)
{
int lid = get_local_id(0);
int gid = get_group_id(0);
int id = get_global_id(0);
int idx = offset + id + (id / cols) * invalid_cols;
__local T localmem_max[128], localmem_min[128];
T minval = (T)(MAX_VAL), maxval = (T)(MIN_VAL), temp;
for (int grainSize = groupnum << 8; id < elemnum; id += grainSize)
{
idx = offset + id + (id / cols) * invalid_cols;
temp = src[idx];
minval = min(minval, temp);
maxval = max(maxval, temp);
}
if (lid > 127)
{
localmem_min[lid - 128] = minval;
localmem_max[lid - 128] = maxval;
}
barrier(CLK_LOCAL_MEM_FENCE);
if (lid < 128)
{
localmem_min[lid] = min(minval, localmem_min[lid]);
localmem_max[lid] = max(maxval, localmem_max[lid]);
}
barrier(CLK_LOCAL_MEM_FENCE);
for (int lsize = 64; lsize > 0; lsize >>= 1)
{
if (lid < lsize)
{
int lid2 = lsize + lid;
localmem_min[lid] = min(localmem_min[lid], localmem_min[lid2]);
localmem_max[lid] = max(localmem_max[lid], localmem_max[lid2]);
}
barrier(CLK_LOCAL_MEM_FENCE);
}
if (lid == 0)
{
dst[gid] = localmem_min[0];
dst[gid + groupnum] = localmem_max[0];
}
}
__kernel void arithm_op_minMax_mask(__global const T * src, __global T * dst,
int cols, int invalid_cols, int offset,
int elemnum, int groupnum,
const __global uchar * mask, int minvalid_cols, int moffset)
__kernel void arithm_op_minMax(__global const T * src, int src_step, int src_offset, int src_rows, int src_cols,
int total, int groupnum, __global T * dst
#ifdef WITH_MASK
, __global const uchar * mask, int mask_step, int mask_offset
#endif
)
{
int lid = get_local_id(0);
int gid = get_group_id(0);
int id = get_global_id(0);
int idx = offset + id + (id / cols) * invalid_cols;
int midx = moffset + id + (id / cols) * minvalid_cols;
__local T localmem_max[128], localmem_min[128];
T minval = (T)(MAX_VAL), maxval = (T)(MIN_VAL), temp;
int y, x;
for (int grainSize = groupnum << 8; id < elemnum; id += grainSize)
for (int grainSize = groupnum << 8; id < total; id += grainSize)
{
idx = offset + id + (id / cols) * invalid_cols;
midx = moffset + id + (id / cols) * minvalid_cols;
y = id / src_cols;
x = id % src_cols;
if (mask[midx])
#ifdef WITH_MASK
if (mask[mad24(y, mask_step, x + mask_offset)])
#endif
{
temp = src[idx];
temp = src[mad24(y, src_step, x + src_offset)];
minval = min(minval, temp);
maxval = max(maxval, temp);
}
......
modules/ocl/src/opencl/arithm_nonzero.cl
View file @ 5726e80f
......@@ -52,23 +52,18 @@
/**************************************Count NonZero**************************************/
__kernel void arithm_op_nonzero(int cols, int invalid_cols, int offset, int elemnum, int groupnum,
__global srcT *src, __global dstT *dst)
__kernel void arithm_op_nonzero(__global srcT * src, int src_step, int src_offset, int src_cols,
int total, int groupnum, __global dstT * dst)
{
int lid = get_local_id(0);
int gid = get_group_id(0);
int id = get_global_id(0);
int idx = offset + id + (id / cols) * invalid_cols;
__local dstT localmem_nonzero[128];
dstT nonzero = (dstT)(0);
srcT zero = (srcT)(0), one = (srcT)(1);
for (int grain = groupnum << 8; id < elemnum; id += grain)
{
idx = offset + id + (id / cols) * invalid_cols;
nonzero += src[idx] == zero ? zero : one;
}
for (int grain = groupnum << 8; id < total; id += grain)
nonzero += convertToDstT(src[mad24(id / src_cols, src_step, id % src_cols + src_offset)] == (srcT)(0)) ? (dstT)(0) : (dstT)(1);
if (lid > 127)
localmem_nonzero[lid - 128] = nonzero;
......
modules/ocl/src/opencl/arithm_sum.cl
View file @ 5726e80f
......@@ -63,21 +63,19 @@
/**************************************Array buffer SUM**************************************/
__kernel void arithm_op_sum(int cols,int invalid_cols,int offset,int elemnum,int groupnum,
__global srcT *src, __global dstT *dst)
__kernel void arithm_op_sum(__global srcT * src, int src_step, int src_offset, int src_cols,
int total, int groupnum, __global dstT * dst)
{
int lid = get_local_id(0);
int gid = get_group_id(0);
int id = get_global_id(0);
int idx = offset + id + (id / cols) * invalid_cols;
__local dstT localmem_sum[128];
dstT sum = (dstT)(0), temp;
for (int grainSize = groupnum << 8; id < elemnum; id += grainSize)
for (int grainSize = groupnum << 8; id < total; id += grainSize)
{
idx = offset + id + (id / cols) * invalid_cols;
temp = convertToDstT(src[idx]);
temp = convertToDstT(src[mad24(id / src_cols, src_step, id % src_cols + src_offset)]);
FUNC(temp, sum);
}
......
modules/ocl/test/test_arithm.cpp
View file @ 5726e80f
......@@ -198,7 +198,7 @@ PARAM_TEST_CASE(ArithmTestBase, MatDepth, Channels, bool)
Size roiSize = randomSize(1, MAX_VALUE);
Border src1Border = randomBorder(0, use_roi ? MAX_VALUE : 0);
randomSubMat(src1, src1_roi, roiSize, src1Border, type, 2, 11);
randomSubMat(src1, src1_roi, roiSize, src1Border, type, -11, 11);
Border src2Border = randomBorder(0, use_roi ? MAX_VALUE : 0);
randomSubMat(src2, src2_roi, roiSize, src2Border, type, -1540, 1740);
......@@ -1163,7 +1163,7 @@ OCL_TEST_P(CountNonZero, MAT)
int cpures = cv::countNonZero(src1_roi);
int gpures = cv::ocl::countNonZero(gsrc1_roi);
EXPECT_DOUBLE_EQ((double)cpures, (double)gpures);
EXPECT_EQ(cpures, gpures);
}
}
......
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