potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 15:23:05 +04:00
Code Packages Releases Wiki Activity

imgproc: optimize tiled parallel filter with TLS buffers and custom border fill

Browse Source
This commit is contained in:
Ismail
2026-04-24 09:22:35 +02:00
parent a5295015b2
commit d16f2bfef2
9 changed files with 472 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/imgproc/src/morph.simd.hpp
+11 -5
View File
@@ -43,6 +43,7 @@
#include "precomp.hpp"
#include <limits.h>
#include "opencv2/core/hal/intrin.hpp"
#include "filter.hpp"
/****************************************************************************************\
Basic Morphological Operations: Erosion & Dilation
@@ -495,6 +496,8 @@ template<class Op, class VecOp> struct MorphRowFilter : public BaseRowFilter
anchor = _anchor;
}
bool isStateless() const CV_OVERRIDE { return true; }
void operator()(const uchar* src, uchar* dst, int width, int cn) CV_OVERRIDE
{
CV_INSTRUMENT_REGION();
@@ -551,6 +554,8 @@ template<class Op, class VecOp> struct MorphColumnFilter : public BaseColumnFilt
anchor = _anchor;
}
bool isStateless() const CV_OVERRIDE { return true; }
void operator()(const uchar** _src, uchar* dst, int dststep, int count, int width) CV_OVERRIDE
{
CV_INSTRUMENT_REGION();
@@ -638,7 +643,7 @@ template<class Op, class VecOp> struct MorphColumnFilter : public BaseColumnFilt
};
template<class Op, class VecOp> struct MorphFilter : BaseFilter
template<class Op, class VecOp> struct MorphFilter : public BaseFilter
{
typedef typename Op::rtype T;
@@ -651,16 +656,18 @@ template<class Op, class VecOp> struct MorphFilter : BaseFilter
std::vector<uchar> coeffs; // we do not really the values of non-zero
// kernel elements, just their locations
preprocess2DKernel( _kernel, coords, coeffs );
ptrs.resize( coords.size() );
}
bool isStateless() const CV_OVERRIDE { return true; }
void operator()(const uchar** src, uchar* dst, int dststep, int count, int width, int cn) CV_OVERRIDE
{
CV_INSTRUMENT_REGION();
const Point* pt = &coords[0];
const T** kp = (const T**)&ptrs[0];
int i, k, nz = (int)coords.size();
AutoBuffer<const T*> _kp(nz);
const T** kp = _kp.data();
Op op;
width *= cn;
@@ -671,7 +678,7 @@ template<class Op, class VecOp> struct MorphFilter : BaseFilter
for( k = 0; k < nz; k++ )
kp[k] = (const T*)src[pt[k].y] + pt[k].x*cn;
i = vecOp(&ptrs[0], nz, dst, width);
i = vecOp((uchar**)kp, nz, dst, width);
#if CV_ENABLE_UNROLLED
for( ; i <= width - 4; i += 4 )
{
@@ -700,7 +707,6 @@ template<class Op, class VecOp> struct MorphFilter : BaseFilter
}
std::vector<Point> coords;
std::vector<uchar*> ptrs;
VecOp vecOp;
};