refactored filter2D; eliminated restrictions sdepth == ddepth, delta == 0

This commit is contained in:
Ilya Lavrenov 2014-03-21 15:59:54 +04:00
parent b6833fdde7
commit e2c6ab0166
4 changed files with 97 additions and 255 deletions

View File

@ -3154,74 +3154,50 @@ static bool ocl_filter2D( InputArray _src, OutputArray _dst, int ddepth,
InputArray _kernel, Point anchor,
double delta, int borderType )
{
if (abs(delta) > FLT_MIN)
return false;
int type = _src.type(), cn = CV_MAT_CN(type);
int type = _src.type(), sdepth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type);
ddepth = ddepth < 0 ? sdepth : ddepth;
int dtype = CV_MAKE_TYPE(ddepth, cn), wdepth = std::max(std::max(sdepth, ddepth), CV_32F),
wtype = CV_MAKE_TYPE(wdepth, cn);
if (cn > 4)
return false;
int sdepth = CV_MAT_DEPTH(type);
Size ksize = _kernel.size();
if( anchor.x < 0 )
if (anchor.x < 0)
anchor.x = ksize.width / 2;
if( anchor.y < 0 )
if (anchor.y < 0)
anchor.y = ksize.height / 2;
if( ddepth < 0 )
ddepth = sdepth;
else if (ddepth != sdepth)
return false;
bool isIsolatedBorder = (borderType & BORDER_ISOLATED) != 0;
bool useDouble = (CV_64F == sdepth);
bool isolated = (borderType & BORDER_ISOLATED) != 0;
borderType &= ~BORDER_ISOLATED;
const cv::ocl::Device &device = cv::ocl::Device::getDefault();
int doubleFPConfig = device.doubleFPConfig();
if (useDouble && (0 == doubleFPConfig))
bool doubleSupport = device.doubleFPConfig() > 0;
if (wdepth == CV_64F && !doubleSupport)
return false;
const char* btype = NULL;
switch (borderType & ~BORDER_ISOLATED)
{
case BORDER_CONSTANT:
btype = "BORDER_CONSTANT";
break;
case BORDER_REPLICATE:
btype = "BORDER_REPLICATE";
break;
case BORDER_REFLECT:
btype = "BORDER_REFLECT";
break;
case BORDER_WRAP:
return false;
case BORDER_REFLECT101:
btype = "BORDER_REFLECT_101";
break;
}
const char * const borderMap[] = { "BORDER_CONSTANT", "BORDER_REPLICATE", "BORDER_REFLECT",
"BORDER_WRAP", "BORDER_REFLECT_101" };
cv::Mat kernelMat = _kernel.getMat();
std::vector<float> kernelMatDataFloat;
std::vector<double> kernelMatDataDouble;
int kernel_size_y2_aligned = useDouble ?
_prepareKernelFilter2D<double>(kernelMatDataDouble, kernelMat)
: _prepareKernelFilter2D<float>(kernelMatDataFloat, kernelMat);
int kernel_size_y2_aligned = _prepareKernelFilter2D<float>(kernelMatDataFloat, kernelMat);
cv::Size sz = _src.size(), wholeSize;
size_t globalsize[2] = { sz.width, sz.height }, localsize[2] = { 0, 1 };
cv::Size sz = _src.size();
size_t globalsize[2] = {sz.width, sz.height};
size_t localsize[2] = {0, 1};
ocl::Kernel kernel;
UMat src; Size wholeSize;
if (!isIsolatedBorder)
ocl::Kernel k;
UMat src = _src.getUMat();
if (!isolated)
{
src = _src.getUMat();
Point ofs;
src.locateROI(wholeSize, ofs);
}
size_t maxWorkItemSizes[32]; device.maxWorkItemSizes(maxWorkItemSizes);
size_t maxWorkItemSizes[32];
device.maxWorkItemSizes(maxWorkItemSizes);
size_t tryWorkItems = maxWorkItemSizes[0];
for (;;)
char cvt[2][40];
for ( ; ; )
{
size_t BLOCK_SIZE = tryWorkItems;
while (BLOCK_SIZE > 32 && BLOCK_SIZE >= (size_t)ksize.width * 2 && BLOCK_SIZE > (size_t)sz.width * 2)
@ -3241,32 +3217,36 @@ static bool ocl_filter2D( InputArray _src, OutputArray _dst, int ddepth,
int requiredLeft = (int)BLOCK_SIZE; // not this: anchor.x;
int requiredBottom = ksize.height - 1 - anchor.y;
int requiredRight = (int)BLOCK_SIZE; // not this: ksize.width - 1 - anchor.x;
int h = isIsolatedBorder ? sz.height : wholeSize.height;
int w = isIsolatedBorder ? sz.width : wholeSize.width;
int h = isolated ? sz.height : wholeSize.height;
int w = isolated ? sz.width : wholeSize.width;
bool extra_extrapolation = h < requiredTop || h < requiredBottom || w < requiredLeft || w < requiredRight;
if ((w < ksize.width) || (h < ksize.height))
return false;
char build_options[1024];
sprintf(build_options, "-D LOCAL_SIZE=%d -D BLOCK_SIZE_Y=%d -D DATA_DEPTH=%d -D DATA_CHAN=%d -D USE_DOUBLE=%d "
"-D ANCHOR_X=%d -D ANCHOR_Y=%d -D KERNEL_SIZE_X=%d -D KERNEL_SIZE_Y=%d -D KERNEL_SIZE_Y2_ALIGNED=%d "
"-D %s -D %s -D %s",
(int)BLOCK_SIZE, (int)BLOCK_SIZE_Y,
sdepth, cn, useDouble ? 1 : 0,
anchor.x, anchor.y, ksize.width, ksize.height, kernel_size_y2_aligned,
btype,
extra_extrapolation ? "EXTRA_EXTRAPOLATION" : "NO_EXTRA_EXTRAPOLATION",
isIsolatedBorder ? "BORDER_ISOLATED" : "NO_BORDER_ISOLATED");
String opts = format("-D LOCAL_SIZE=%d -D BLOCK_SIZE_Y=%d -D cn=%d "
"-D ANCHOR_X=%d -D ANCHOR_Y=%d -D KERNEL_SIZE_X=%d -D KERNEL_SIZE_Y=%d "
"-D KERNEL_SIZE_Y2_ALIGNED=%d -D %s -D %s -D %s%s "
"-D srcT=%s -D srcT1=%s -D dstT=%s -D dstT1=%s -D WT=%s -D WT1=%s "
"-D convertToWT=%s -D convertToDstT=%s",
(int)BLOCK_SIZE, (int)BLOCK_SIZE_Y, cn, anchor.x, anchor.y,
ksize.width, ksize.height, kernel_size_y2_aligned, borderMap[borderType],
extra_extrapolation ? "EXTRA_EXTRAPOLATION" : "NO_EXTRA_EXTRAPOLATION",
isolated ? "BORDER_ISOLATED" : "NO_BORDER_ISOLATED",
doubleSupport ? " -D DOUBLE_SUPPORT" : "",
ocl::typeToStr(type), ocl::typeToStr(sdepth), ocl::typeToStr(dtype),
ocl::typeToStr(ddepth), ocl::typeToStr(wtype), ocl::typeToStr(wdepth),
ocl::convertTypeStr(sdepth, wdepth, cn, cvt[0]),
ocl::convertTypeStr(wdepth, ddepth, cn, cvt[1]));
localsize[0] = BLOCK_SIZE;
globalsize[0] = DIVUP(sz.width, BLOCK_SIZE - (ksize.width - 1)) * BLOCK_SIZE;
globalsize[1] = DIVUP(sz.height, BLOCK_SIZE_Y);
cv::String errmsg;
if (!kernel.create("filter2D", cv::ocl::imgproc::filter2D_oclsrc, build_options))
if (!k.create("filter2D", cv::ocl::imgproc::filter2D_oclsrc, opts))
return false;
size_t kernelWorkGroupSize = kernel.workGroupSize();
size_t kernelWorkGroupSize = k.workGroupSize();
if (localsize[0] <= kernelWorkGroupSize)
break;
if (BLOCK_SIZE < kernelWorkGroupSize)
@ -3274,46 +3254,19 @@ static bool ocl_filter2D( InputArray _src, OutputArray _dst, int ddepth,
tryWorkItems = kernelWorkGroupSize;
}
_dst.create(sz, CV_MAKETYPE(ddepth, cn));
UMat dst = _dst.getUMat();
if (src.empty())
src = _src.getUMat();
int idxArg = 0;
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(src));
idxArg = kernel.set(idxArg, (int)src.step);
_dst.create(sz, dtype);
UMat dst = _dst.getUMat(), kernalDataUMat(kernelMatDataFloat, true);
int srcOffsetX = (int)((src.offset % src.step) / src.elemSize());
int srcOffsetY = (int)(src.offset / src.step);
int srcEndX = (isIsolatedBorder ? (srcOffsetX + sz.width) : wholeSize.width);
int srcEndY = (isIsolatedBorder ? (srcOffsetY + sz.height) : wholeSize.height);
idxArg = kernel.set(idxArg, srcOffsetX);
idxArg = kernel.set(idxArg, srcOffsetY);
idxArg = kernel.set(idxArg, srcEndX);
idxArg = kernel.set(idxArg, srcEndY);
int srcEndX = (isolated ? (srcOffsetX + sz.width) : wholeSize.width);
int srcEndY = (isolated ? (srcOffsetY + sz.height) : wholeSize.height);
idxArg = kernel.set(idxArg, ocl::KernelArg::WriteOnly(dst));
float borderValue[4] = {0, 0, 0, 0};
double borderValueDouble[4] = {0, 0, 0, 0};
if ((borderType & ~BORDER_ISOLATED) == BORDER_CONSTANT)
{
int cnocl = 3 == cn ? 4 : cn;
if (useDouble)
idxArg = kernel.set(idxArg, (void *)&borderValueDouble[0], sizeof(double) * cnocl);
else
idxArg = kernel.set(idxArg, (void *)&borderValue[0], sizeof(float) * cnocl);
}
if (useDouble)
{
UMat kernalDataUMat(kernelMatDataDouble, true);
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(kernalDataUMat));
}
else
{
UMat kernalDataUMat(kernelMatDataFloat, true);
idxArg = kernel.set(idxArg, ocl::KernelArg::PtrReadOnly(kernalDataUMat));
}
return kernel.run(2, globalsize, localsize, true);
k.args(ocl::KernelArg::PtrReadOnly(src), (int)src.step, srcOffsetX, srcOffsetY,
srcEndX, srcEndY, ocl::KernelArg::WriteOnly(dst),
ocl::KernelArg::PtrReadOnly(kernalDataUMat), (float)delta);
return k.run(2, globalsize, localsize, false);
}
static bool ocl_sepRowFilter2D(const UMat & src, UMat & buf, const Mat & kernelX, int anchor,

View File

@ -122,7 +122,7 @@
}
#ifdef BORDER_REFLECT
#define EXTRAPOLATE(x, y, minX, minY, maxX, maxY) EXTRAPOLATE_(x, y, minX, minY, maxX, maxY, 0)
#elif defined(BORDER_REFLECT_101)
#elif defined(BORDER_REFLECT_101) || defined(BORDER_REFLECT101)
#define EXTRAPOLATE(x, y, minX, minY, maxX, maxY) EXTRAPOLATE_(x, y, minX, minY, maxX, maxY, 1)
#endif
#else
@ -142,127 +142,49 @@
}
#endif
#if USE_DOUBLE
#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
#define FPTYPE double
#define CONVERT_TO_FPTYPE CAT(convert_double, VEC_SIZE)
#else
#define FPTYPE float
#define CONVERT_TO_FPTYPE CAT(convert_float, VEC_SIZE)
#endif
#if DATA_DEPTH == 0
#define BASE_TYPE uchar
#elif DATA_DEPTH == 1
#define BASE_TYPE char
#elif DATA_DEPTH == 2
#define BASE_TYPE ushort
#elif DATA_DEPTH == 3
#define BASE_TYPE short
#elif DATA_DEPTH == 4
#define BASE_TYPE int
#elif DATA_DEPTH == 5
#define BASE_TYPE float
#elif DATA_DEPTH == 6
#define BASE_TYPE double
#if cn != 3
#define loadpix(addr) *(__global const srcT *)(addr)
#define storepix(val, addr) *(__global dstT *)(addr) = val
#define SRCSIZE (int)sizeof(srcT)
#define DSTSIZE (int)sizeof(dstT)
#else
#error data_depth
#define loadpix(addr) vload3(0, (__global const srcT1 *)(addr))
#define storepix(val, addr) vstore3(val, 0, (__global dstT1 *)(addr))
#define SRCSIZE (int)sizeof(srcT1) * cn
#define DSTSIZE (int)sizeof(dstT1) * cn
#endif
#define __CAT(x, y) x##y
#define CAT(x, y) __CAT(x, y)
#define uchar1 uchar
#define char1 char
#define ushort1 ushort
#define short1 short
#define int1 int
#define float1 float
#define double1 double
#define convert_uchar1_sat_rte convert_uchar_sat_rte
#define convert_char1_sat_rte convert_char_sat_rte
#define convert_ushort1_sat_rte convert_ushort_sat_rte
#define convert_short1_sat_rte convert_short_sat_rte
#define convert_int1_sat_rte convert_int_sat_rte
#define convert_float1
#define convert_double1
#if DATA_DEPTH == 5 || DATA_DEPTH == 6
#define CONVERT_TO_TYPE CAT(CAT(convert_, BASE_TYPE), VEC_SIZE)
#else
#define CONVERT_TO_TYPE CAT(CAT(CAT(convert_, BASE_TYPE), VEC_SIZE), _sat_rte)
#endif
#define VEC_SIZE DATA_CHAN
#define VEC_TYPE CAT(BASE_TYPE, VEC_SIZE)
#define TYPE VEC_TYPE
#if VEC_SIZE == 3
#define SCALAR_TYPE CAT(FPTYPE, 4)
#else
#define SCALAR_TYPE CAT(FPTYPE, VEC_SIZE)
#endif
#define INTERMEDIATE_TYPE CAT(FPTYPE, VEC_SIZE)
#if DATA_CHAN != 3
#define loadpix(addr) *(__global const TYPE *)(addr)
#define storepix(val, addr) *(__global TYPE *)(addr) = val
#define TSIZE (int)sizeof(TYPE)
#else
#define loadpix(addr) vload3(0, (__global const BASE_TYPE *)(addr))
#define storepix(val, addr) vstore3(val, 0, (__global BASE_TYPE *)(addr))
#define TSIZE (int)sizeof(BASE_TYPE)*DATA_CHAN
#endif
#define noconvert
struct RectCoords
{
int x1, y1, x2, y2;
};
//#define DEBUG
#ifdef DEBUG
#define DEBUG_ONLY(x) x
#define ASSERT(condition) do { if (!(condition)) { printf("BUG in boxFilter kernel (global=%d,%d): " #condition "\n", get_global_id(0), get_global_id(1)); } } while (0)
#else
#define DEBUG_ONLY(x) (void)0
#define ASSERT(condition) (void)0
#endif
inline INTERMEDIATE_TYPE readSrcPixel(int2 pos, __global const uchar* srcptr, int srcstep, const struct RectCoords srcCoords
#ifdef BORDER_CONSTANT
, SCALAR_TYPE borderValue
#endif
)
inline WT readSrcPixel(int2 pos, __global const uchar * srcptr, int src_step, const struct RectCoords srcCoords)
{
#ifdef BORDER_ISOLATED
if(pos.x >= srcCoords.x1 && pos.y >= srcCoords.y1 && pos.x < srcCoords.x2 && pos.y < srcCoords.y2)
if (pos.x >= srcCoords.x1 && pos.y >= srcCoords.y1 && pos.x < srcCoords.x2 && pos.y < srcCoords.y2)
#else
if(pos.x >= 0 && pos.y >= 0 && pos.x < srcCoords.x2 && pos.y < srcCoords.y2)
if (pos.x >= 0 && pos.y >= 0 && pos.x < srcCoords.x2 && pos.y < srcCoords.y2)
#endif
{
//__global TYPE* ptr = (__global TYPE*)((__global char*)src + pos.x * sizeof(TYPE) + pos.y * srcStepBytes);
__global TYPE* ptr = (__global TYPE*)(srcptr + pos.y * srcstep + pos.x * TSIZE);
return CONVERT_TO_FPTYPE(loadpix(ptr));
return convertToWT(loadpix(srcptr + mad24(pos.y, src_step, pos.x * SRCSIZE)));
}
else
{
#ifdef BORDER_CONSTANT
#if VEC_SIZE == 3
return (INTERMEDIATE_TYPE)(borderValue.x, borderValue.y, borderValue.z);
return (WT)(0);
#else
return borderValue;
#endif
#else
int selected_col = pos.x;
int selected_row = pos.y;
int selected_col = pos.x, selected_row = pos.y;
EXTRAPOLATE(selected_col, selected_row,
#ifdef BORDER_ISOLATED
@ -273,68 +195,40 @@ inline INTERMEDIATE_TYPE readSrcPixel(int2 pos, __global const uchar* srcptr, in
srcCoords.x2, srcCoords.y2
);
// debug border mapping
//printf("pos=%d,%d --> %d, %d\n", pos.x, pos.y, selected_col, selected_row);
pos = (int2)(selected_col, selected_row);
if(pos.x >= 0 && pos.y >= 0 && pos.x < srcCoords.x2 && pos.y < srcCoords.y2)
{
//__global TYPE* ptr = (__global TYPE*)((__global char*)src + pos.x * sizeof(TYPE) + pos.y * srcStepBytes);
__global TYPE* ptr = (__global TYPE*)(srcptr + pos.y * srcstep + pos.x * TSIZE);
return CONVERT_TO_FPTYPE(loadpix(ptr));
}
else
{
// for debug only
DEBUG_ONLY(printf("BUG in boxFilter kernel\n"));
return (FPTYPE)(0.0f);
}
return convertToWT(loadpix(srcptr + mad24(selected_row, src_step, selected_col * SRCSIZE)));
#endif
}
}
// INPUT PARAMETER: BLOCK_SIZE_Y (via defines)
__kernel
__attribute__((reqd_work_group_size(LOCAL_SIZE, 1, 1)))
void filter2D(__global const uchar* srcptr, int srcstep, int srcOffsetX, int srcOffsetY, int srcEndX, int srcEndY,
__global uchar* dstptr, int dststep, int dstoffset,
int rows, int cols,
#ifdef BORDER_CONSTANT
SCALAR_TYPE borderValue,
#endif
__constant FPTYPE* kernelData // transposed: [KERNEL_SIZE_X][KERNEL_SIZE_Y2_ALIGNED]
)
__kernel void filter2D(__global const uchar * srcptr, int src_step, int srcOffsetX, int srcOffsetY, int srcEndX, int srcEndY,
__global uchar * dstptr, int dst_step, int dst_offset, int rows, int cols, __constant WT1 * kernelData, float delta)
{
const struct RectCoords srcCoords = {srcOffsetX, srcOffsetY, srcEndX, srcEndY}; // for non-isolated border: offsetX, offsetY, wholeX, wholeY
const struct RectCoords srcCoords = { srcOffsetX, srcOffsetY, srcEndX, srcEndY }; // for non-isolated border: offsetX, offsetY, wholeX, wholeY
const int local_id = get_local_id(0);
const int x = local_id + (LOCAL_SIZE - (KERNEL_SIZE_X - 1)) * get_group_id(0) - ANCHOR_X;
const int y = get_global_id(1) * BLOCK_SIZE_Y;
int local_id = get_local_id(0);
int x = local_id + (LOCAL_SIZE - (KERNEL_SIZE_X - 1)) * get_group_id(0) - ANCHOR_X;
int y = get_global_id(1) * BLOCK_SIZE_Y;
INTERMEDIATE_TYPE data[KERNEL_SIZE_Y];
__local INTERMEDIATE_TYPE sumOfCols[LOCAL_SIZE];
WT data[KERNEL_SIZE_Y];
__local WT sumOfCols[LOCAL_SIZE];
int2 srcPos = (int2)(srcCoords.x1 + x, srcCoords.y1 + y - ANCHOR_Y);
int2 pos = (int2)(x, y);
__global TYPE* dstPtr = (__global TYPE*)((__global char*)dstptr + pos.y * dststep + dstoffset + pos.x * TSIZE); // Pointer can be out of bounds!
bool writeResult = ((local_id >= ANCHOR_X) && (local_id < LOCAL_SIZE - (KERNEL_SIZE_X - 1 - ANCHOR_X)) &&
(pos.x >= 0) && (pos.x < cols));
__global dstT * dst = (__global dstT *)(dstptr + mad24(pos.y, dst_step, mad24(pos.x, DSTSIZE, dst_offset))); // Pointer can be out of bounds!
bool writeResult = local_id >= ANCHOR_X && local_id < LOCAL_SIZE - (KERNEL_SIZE_X - 1 - ANCHOR_X) &&
pos.x >= 0 && pos.x < cols;
#if BLOCK_SIZE_Y > 1
bool readAllpixels = true;
int sy_index = 0; // current index in data[] array
dstRowsMax = min(rows, pos.y + BLOCK_SIZE_Y);
for (;
pos.y < dstRowsMax;
pos.y++,
dstPtr = (__global TYPE*)((__global char*)dstptr + dststep))
for ( ;
pos.y < dstRowsMax;
pos.y++, dst = (__global dstT *)((__global uchar *)dst + dst_step))
#endif
{
ASSERT(pos.y < dstRowsMax);
for (
#if BLOCK_SIZE_Y > 1
int sy = readAllpixels ? 0 : -1; sy < (readAllpixels ? KERNEL_SIZE_Y : 0);
@ -343,27 +237,21 @@ void filter2D(__global const uchar* srcptr, int srcstep, int srcOffsetX, int src
#endif
sy++, srcPos.y++)
{
data[sy + sy_index] = readSrcPixel(srcPos, srcptr, srcstep, srcCoords
#ifdef BORDER_CONSTANT
, borderValue
#endif
);
data[sy + sy_index] = readSrcPixel(srcPos, srcptr, src_step, srcCoords);
}
INTERMEDIATE_TYPE total_sum = 0;
WT total_sum = 0;
for (int sx = 0; sx < KERNEL_SIZE_X; sx++)
{
{
__constant FPTYPE* k = &kernelData[KERNEL_SIZE_Y2_ALIGNED * sx
__constant WT1 * k = &kernelData[KERNEL_SIZE_Y2_ALIGNED * sx
#if BLOCK_SIZE_Y > 1
+ KERNEL_SIZE_Y - sy_index
#endif
];
INTERMEDIATE_TYPE tmp_sum = 0;
WT tmp_sum = 0;
for (int sy = 0; sy < KERNEL_SIZE_Y; sy++)
{
tmp_sum += data[sy] * k[sy];
}
sumOfCols[local_id] = tmp_sum;
barrier(CLK_LOCAL_MEM_FENCE);
@ -377,14 +265,12 @@ void filter2D(__global const uchar* srcptr, int srcstep, int srcOffsetX, int src
}
if (writeResult)
{
storepix(CONVERT_TO_TYPE(total_sum), dstPtr);
}
storepix(convertToDstT(total_sum + (WT)(delta)), dst);
#if BLOCK_SIZE_Y > 1
readAllpixels = false;
#if BLOCK_SIZE_Y > KERNEL_SIZE_Y
sy_index = (sy_index + 1 <= KERNEL_SIZE_Y) ? sy_index + 1 : 1;
sy_index = sy_index + 1 <= KERNEL_SIZE_Y ? sy_index + 1 : 1;
#else
sy_index++;
#endif

View File

@ -62,6 +62,7 @@ PARAM_TEST_CASE(Filter2D, MatDepth, Channels, BorderType, bool, bool)
int borderType;
bool useRoi;
Mat kernel;
double delta;
TEST_DECLARE_INPUT_PARAMETER(src);
TEST_DECLARE_OUTPUT_PARAMETER(dst);
@ -91,6 +92,8 @@ PARAM_TEST_CASE(Filter2D, MatDepth, Channels, BorderType, bool, bool)
anchor.x = randomInt(-1, ksize.width);
anchor.y = randomInt(-1, ksize.height);
delta = randomDouble(-100, 100);
UMAT_UPLOAD_INPUT_PARAMETER(src);
UMAT_UPLOAD_OUTPUT_PARAMETER(dst);
}
@ -108,8 +111,8 @@ OCL_TEST_P(Filter2D, Mat)
{
random_roi();
OCL_OFF(cv::filter2D(src_roi, dst_roi, -1, kernel, anchor, 0.0, borderType));
OCL_ON(cv::filter2D(usrc_roi, udst_roi, -1, kernel, anchor, 0.0, borderType));
OCL_OFF(cv::filter2D(src_roi, dst_roi, -1, kernel, anchor, delta, borderType));
OCL_ON(cv::filter2D(usrc_roi, udst_roi, -1, kernel, anchor, delta, borderType));
Near(1.0);
}

View File

@ -152,8 +152,8 @@ OCL_TEST_P(LaplacianTest, Accuracy)
{
random_roi();
OCL_OFF(cv::Laplacian(src_roi, dst_roi, -1, ksize, scale, 0, borderType));
OCL_ON(cv::Laplacian(usrc_roi, udst_roi, -1, ksize, scale, 0, borderType));
OCL_OFF(cv::Laplacian(src_roi, dst_roi, -1, ksize, scale, 10, borderType));
OCL_ON(cv::Laplacian(usrc_roi, udst_roi, -1, ksize, scale, 10, borderType));
Near();
}