Skip to content

O
opencv
Commit 91d152e2 authored by Alexander Alekhin's avatar Alexander Alekhin
Browse files
Options

core: dispatch split.cpp

parent 1d8b30bf
Hide whitespace changes
Inline Side-by-side
Showing with 36 additions and 397 deletions
modules/core/CMakeLists.txt
View file @ 91d152e2
......@@ -8,6 +8,7 @@ ocv_add_dispatched_file(convert_scale SSE2 AVX2)
ocv_add_dispatched_file(count_non_zero SSE2 AVX2)
ocv_add_dispatched_file(matmul SSE2 AVX2)
ocv_add_dispatched_file(mean SSE2 AVX2)
ocv_add_dispatched_file(split SSE2 AVX2)
ocv_add_dispatched_file(sum SSE2 AVX2)
# dispatching for accuracy tests
......
modules/core/src/split.dispatch.cpp
View file @ 91d152e2
......@@ -6,213 +6,44 @@
#include "precomp.hpp"
#include "opencl_kernels_core.hpp"
namespace cv { namespace hal {
#if CV_SIMD
// see the comments for vecmerge_ in merge.cpp
template<typename T, typename VecT> static void
vecsplit_( const T* src, T** dst, int len, int cn )
{
const int VECSZ = VecT::nlanes;
int i, i0 = 0;
T* dst0 = dst[0];
T* dst1 = dst[1];
int r0 = (int)((size_t)(void*)dst0 % (VECSZ*sizeof(T)));
int r1 = (int)((size_t)(void*)dst1 % (VECSZ*sizeof(T)));
int r2 = cn > 2 ? (int)((size_t)(void*)dst[2] % (VECSZ*sizeof(T))) : r0;
int r3 = cn > 3 ? (int)((size_t)(void*)dst[3] % (VECSZ*sizeof(T))) : r0;
hal::StoreMode mode = hal::STORE_ALIGNED_NOCACHE;
if( (r0|r1|r2|r3) != 0 )
{
mode = hal::STORE_UNALIGNED;
if( r0 == r1 && r0 == r2 && r0 == r3 && r0 % sizeof(T) == 0 && len > VECSZ*2 )
i0 = VECSZ - (r0 / sizeof(T));
}
if( cn == 2 )
{
for( i = 0; i < len; i += VECSZ )
{
if( i > len - VECSZ )
{
i = len - VECSZ;
mode = hal::STORE_UNALIGNED;
}
VecT a, b;
v_load_deinterleave(src + i*cn, a, b);
v_store(dst0 + i, a, mode);
v_store(dst1 + i, b, mode);
if( i < i0 )
{
i = i0 - VECSZ;
mode = hal::STORE_ALIGNED_NOCACHE;
}
}
}
else if( cn == 3 )
{
T* dst2 = dst[2];
for( i = 0; i < len; i += VECSZ )
{
if( i > len - VECSZ )
{
i = len - VECSZ;
mode = hal::STORE_UNALIGNED;
}
VecT a, b, c;
v_load_deinterleave(src + i*cn, a, b, c);
v_store(dst0 + i, a, mode);
v_store(dst1 + i, b, mode);
v_store(dst2 + i, c, mode);
if( i < i0 )
{
i = i0 - VECSZ;
mode = hal::STORE_ALIGNED_NOCACHE;
}
}
}
else
{
CV_Assert( cn == 4 );
T* dst2 = dst[2];
T* dst3 = dst[3];
for( i = 0; i < len; i += VECSZ )
{
if( i > len - VECSZ )
{
i = len - VECSZ;
mode = hal::STORE_UNALIGNED;
}
VecT a, b, c, d;
v_load_deinterleave(src + i*cn, a, b, c, d);
v_store(dst0 + i, a, mode);
v_store(dst1 + i, b, mode);
v_store(dst2 + i, c, mode);
v_store(dst3 + i, d, mode);
if( i < i0 )
{
i = i0 - VECSZ;
mode = hal::STORE_ALIGNED_NOCACHE;
}
}
}
vx_cleanup();
}
#endif
template<typename T> static void
split_( const T* src, T** dst, int len, int cn )
{
int k = cn % 4 ? cn % 4 : 4;
int i, j;
if( k == 1 )
{
T* dst0 = dst[0];
if(cn == 1)
{
memcpy(dst0, src, len * sizeof(T));
}
else
{
for( i = 0, j = 0 ; i < len; i++, j += cn )
dst0[i] = src[j];
}
}
else if( k == 2 )
{
T *dst0 = dst[0], *dst1 = dst[1];
i = j = 0;
#include "split.simd.hpp"
#include "split.simd_declarations.hpp" // defines CV_CPU_DISPATCH_MODES_ALL=AVX2,...,BASELINE based on CMakeLists.txt content
for( ; i < len; i++, j += cn )
{
dst0[i] = src[j];
dst1[i] = src[j+1];
}
}
else if( k == 3 )
{
T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2];
i = j = 0;
for( ; i < len; i++, j += cn )
{
dst0[i] = src[j];
dst1[i] = src[j+1];
dst2[i] = src[j+2];
}
}
else
{
T *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2], *dst3 = dst[3];
i = j = 0;
for( ; i < len; i++, j += cn )
{
dst0[i] = src[j]; dst1[i] = src[j+1];
dst2[i] = src[j+2]; dst3[i] = src[j+3];
}
}
for( ; k < cn; k += 4 )
{
T *dst0 = dst[k], *dst1 = dst[k+1], *dst2 = dst[k+2], *dst3 = dst[k+3];
for( i = 0, j = k; i < len; i++, j += cn )
{
dst0[i] = src[j]; dst1[i] = src[j+1];
dst2[i] = src[j+2]; dst3[i] = src[j+3];
}
}
}
namespace cv { namespace hal {
void split8u(const uchar* src, uchar** dst, int len, int cn )
{
CV_INSTRUMENT_REGION();
CALL_HAL(split8u, cv_hal_split8u, src,dst, len, cn)
#if CV_SIMD
if( len >= v_uint8::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<uchar, v_uint8>(src, dst, len, cn);
else
#endif
split_(src, dst, len, cn);
CV_CPU_DISPATCH(split8u, (src, dst, len, cn),
CV_CPU_DISPATCH_MODES_ALL);
}
void split16u(const ushort* src, ushort** dst, int len, int cn )
{
CV_INSTRUMENT_REGION();
CALL_HAL(split16u, cv_hal_split16u, src,dst, len, cn)
#if CV_SIMD
if( len >= v_uint16::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<ushort, v_uint16>(src, dst, len, cn);
else
#endif
split_(src, dst, len, cn);
CV_CPU_DISPATCH(split16u, (src, dst, len, cn),
CV_CPU_DISPATCH_MODES_ALL);
}
void split32s(const int* src, int** dst, int len, int cn )
{
CV_INSTRUMENT_REGION();
CALL_HAL(split32s, cv_hal_split32s, src,dst, len, cn)
#if CV_SIMD
if( len >= v_uint32::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<int, v_int32>(src, dst, len, cn);
else
#endif
split_(src, dst, len, cn);
CV_CPU_DISPATCH(split32s, (src, dst, len, cn),
CV_CPU_DISPATCH_MODES_ALL);
}
void split64s(const int64* src, int64** dst, int len, int cn )
{
CV_INSTRUMENT_REGION();
CALL_HAL(split64s, cv_hal_split64s, src,dst, len, cn)
#if CV_SIMD
if( len >= v_int64::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<int64, v_int64>(src, dst, len, cn);
else
#endif
split_(src, dst, len, cn);
CV_CPU_DISPATCH(split64s, (src, dst, len, cn),
CV_CPU_DISPATCH_MODES_ALL);
}
}} // cv::hal::
} // namespace cv::hal::
/****************************************************************************************\
* split & merge *
......@@ -233,7 +64,6 @@ static SplitFunc getSplitFunc(int depth)
#ifdef HAVE_IPP
namespace cv {
static bool ipp_split(const Mat& src, Mat* mv, int channels)
{
#ifdef HAVE_IPP_IW_LL
......@@ -282,10 +112,9 @@ static bool ipp_split(const Mat& src, Mat* mv, int channels)
return false;
#endif
}
}
#endif
void cv::split(const Mat& src, Mat* mv)
void split(const Mat& src, Mat* mv)
{
CV_INSTRUMENT_REGION();
......@@ -341,8 +170,6 @@ void cv::split(const Mat& src, Mat* mv)
#ifdef HAVE_OPENCL
namespace cv {
static bool ocl_split( InputArray _m, OutputArrayOfArrays _mv )
{
int type = _m.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
......@@ -381,11 +208,9 @@ static bool ocl_split( InputArray _m, OutputArrayOfArrays _mv )
return k.run(2, globalsize, NULL, false);
}
}
#endif
void cv::split(InputArray _m, OutputArrayOfArrays _mv)
void split(InputArray _m, OutputArrayOfArrays _mv)
{
CV_INSTRUMENT_REGION();
......@@ -411,3 +236,5 @@ void cv::split(InputArray _m, OutputArrayOfArrays _mv)
split(m, &dst[0]);
}
} // namespace
modules/core/src/split.simd.hpp
View file @ 91d152e2
......@@ -4,9 +4,16 @@
#include "precomp.hpp"
#include "opencl_kernels_core.hpp"
namespace cv { namespace hal {
CV_CPU_OPTIMIZATION_NAMESPACE_BEGIN
void split8u(const uchar* src, uchar** dst, int len, int cn);
void split16u(const ushort* src, ushort** dst, int len, int cn);
void split32s(const int* src, int** dst, int len, int cn);
void split64s(const int64* src, int64** dst, int len, int cn);
#ifndef CV_CPU_OPTIMIZATION_DECLARATIONS_ONLY
#if CV_SIMD
// see the comments for vecmerge_ in merge.cpp
......@@ -169,8 +176,7 @@ split_( const T* src, T** dst, int len, int cn )
void split8u(const uchar* src, uchar** dst, int len, int cn )
{
CALL_HAL(split8u, cv_hal_split8u, src,dst, len, cn)
CV_INSTRUMENT_REGION();
#if CV_SIMD
if( len >= v_uint8::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<uchar, v_uint8>(src, dst, len, cn);
......@@ -181,7 +187,7 @@ void split8u(const uchar* src, uchar** dst, int len, int cn )
void split16u(const ushort* src, ushort** dst, int len, int cn )
{
CALL_HAL(split16u, cv_hal_split16u, src,dst, len, cn)
CV_INSTRUMENT_REGION();
#if CV_SIMD
if( len >= v_uint16::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<ushort, v_uint16>(src, dst, len, cn);
......@@ -192,7 +198,7 @@ void split16u(const ushort* src, ushort** dst, int len, int cn )
void split32s(const int* src, int** dst, int len, int cn )
{
CALL_HAL(split32s, cv_hal_split32s, src,dst, len, cn)
CV_INSTRUMENT_REGION();
#if CV_SIMD
if( len >= v_uint32::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<int, v_int32>(src, dst, len, cn);
......@@ -203,7 +209,7 @@ void split32s(const int* src, int** dst, int len, int cn )
void split64s(const int64* src, int64** dst, int len, int cn )
{
CALL_HAL(split64s, cv_hal_split64s, src,dst, len, cn)
CV_INSTRUMENT_REGION();
#if CV_SIMD
if( len >= v_int64::nlanes && 2 <= cn && cn <= 4 )
vecsplit_<int64, v_int64>(src, dst, len, cn);
......@@ -212,202 +218,6 @@ void split64s(const int64* src, int64** dst, int len, int cn )
split_(src, dst, len, cn);
}
}} // cv::hal::
/****************************************************************************************\
* split & merge *
\****************************************************************************************/
typedef void (*SplitFunc)(const uchar* src, uchar** dst, int len, int cn);
static SplitFunc getSplitFunc(int depth)
{
static SplitFunc splitTab[] =
{
(SplitFunc)GET_OPTIMIZED(cv::hal::split8u), (SplitFunc)GET_OPTIMIZED(cv::hal::split8u), (SplitFunc)GET_OPTIMIZED(cv::hal::split16u), (SplitFunc)GET_OPTIMIZED(cv::hal::split16u),
(SplitFunc)GET_OPTIMIZED(cv::hal::split32s), (SplitFunc)GET_OPTIMIZED(cv::hal::split32s), (SplitFunc)GET_OPTIMIZED(cv::hal::split64s), 0
};
return splitTab[depth];
}
#ifdef HAVE_IPP
namespace cv {
static bool ipp_split(const Mat& src, Mat* mv, int channels)
{
#ifdef HAVE_IPP_IW_LL
CV_INSTRUMENT_REGION_IPP();
if(channels != 3 && channels != 4)
return false;
if(src.dims <= 2)
{
IppiSize size = ippiSize(src.size());
void *dstPtrs[4] = {NULL};
size_t dstStep = mv[0].step;
for(int i = 0; i < channels; i++)
{
dstPtrs[i] = mv[i].ptr();
if(dstStep != mv[i].step)
return false;
}
return CV_INSTRUMENT_FUN_IPP(llwiCopySplit, src.ptr(), (int)src.step, dstPtrs, (int)dstStep, size, (int)src.elemSize1(), channels, 0) >= 0;
}
else
{
const Mat *arrays[5] = {NULL};
uchar *ptrs[5] = {NULL};
arrays[0] = &src;
for(int i = 1; i < channels; i++)
{
arrays[i] = &mv[i-1];
}
NAryMatIterator it(arrays, ptrs);
IppiSize size = { (int)it.size, 1 };
for( size_t i = 0; i < it.nplanes; i++, ++it )
{
if(CV_INSTRUMENT_FUN_IPP(llwiCopySplit, ptrs[0], 0, (void**)&ptrs[1], 0, size, (int)src.elemSize1(), channels, 0) < 0)
return false;
}
return true;
}
#else
CV_UNUSED(src); CV_UNUSED(mv); CV_UNUSED(channels);
return false;
#endif
}
}
#endif
void cv::split(const Mat& src, Mat* mv)
{
CV_INSTRUMENT_REGION();
int k, depth = src.depth(), cn = src.channels();
if( cn == 1 )
{
src.copyTo(mv[0]);
return;
}
for( k = 0; k < cn; k++ )
{
mv[k].create(src.dims, src.size, depth);
}
CV_IPP_RUN_FAST(ipp_split(src, mv, cn));
SplitFunc func = getSplitFunc(depth);
CV_Assert( func != 0 );
size_t esz = src.elemSize(), esz1 = src.elemSize1();
size_t blocksize0 = (BLOCK_SIZE + esz-1)/esz;
AutoBuffer<uchar> _buf((cn+1)*(sizeof(Mat*) + sizeof(uchar*)) + 16);
const Mat** arrays = (const Mat**)_buf.data();
uchar** ptrs = (uchar**)alignPtr(arrays + cn + 1, 16);
arrays[0] = &src;
for( k = 0; k < cn; k++ )
{
arrays[k+1] = &mv[k];
}
NAryMatIterator it(arrays, ptrs, cn+1);
size_t total = it.size;
size_t blocksize = std::min((size_t)CV_SPLIT_MERGE_MAX_BLOCK_SIZE(cn), cn <= 4 ? total : std::min(total, blocksize0));
for( size_t i = 0; i < it.nplanes; i++, ++it )
{
for( size_t j = 0; j < total; j += blocksize )
{
size_t bsz = std::min(total - j, blocksize);
func( ptrs[0], &ptrs[1], (int)bsz, cn );
if( j + blocksize < total )
{
ptrs[0] += bsz*esz;
for( k = 0; k < cn; k++ )
ptrs[k+1] += bsz*esz1;
}
}
}
}
#ifdef HAVE_OPENCL
namespace cv {
static bool ocl_split( InputArray _m, OutputArrayOfArrays _mv )
{
int type = _m.type(), depth = CV_MAT_DEPTH(type), cn = CV_MAT_CN(type),
rowsPerWI = ocl::Device::getDefault().isIntel() ? 4 : 1;
String dstargs, processelem, indexdecl;
for (int i = 0; i < cn; ++i)
{
dstargs += format("DECLARE_DST_PARAM(%d)", i);
indexdecl += format("DECLARE_INDEX(%d)", i);
processelem += format("PROCESS_ELEM(%d)", i);
}
ocl::Kernel k("split", ocl::core::split_merge_oclsrc,
format("-D T=%s -D OP_SPLIT -D cn=%d -D DECLARE_DST_PARAMS=%s"
" -D PROCESS_ELEMS_N=%s -D DECLARE_INDEX_N=%s",
ocl::memopTypeToStr(depth), cn, dstargs.c_str(),
processelem.c_str(), indexdecl.c_str()));
if (k.empty())
return false;
Size size = _m.size();
_mv.create(cn, 1, depth);
for (int i = 0; i < cn; ++i)
_mv.create(size, depth, i);
std::vector<UMat> dst;
_mv.getUMatVector(dst);
int argidx = k.set(0, ocl::KernelArg::ReadOnly(_m.getUMat()));
for (int i = 0; i < cn; ++i)
argidx = k.set(argidx, ocl::KernelArg::WriteOnlyNoSize(dst[i]));
k.set(argidx, rowsPerWI);
size_t globalsize[2] = { (size_t)size.width, ((size_t)size.height + rowsPerWI - 1) / rowsPerWI };
return k.run(2, globalsize, NULL, false);
}
}
#endif
void cv::split(InputArray _m, OutputArrayOfArrays _mv)
{
CV_INSTRUMENT_REGION();
CV_OCL_RUN(_m.dims() <= 2 && _mv.isUMatVector(),
ocl_split(_m, _mv))
Mat m = _m.getMat();
if( m.empty() )
{
_mv.release();
return;
}
CV_Assert( !_mv.fixedType() || _mv.empty() || _mv.type() == m.depth() );
int depth = m.depth(), cn = m.channels();
_mv.create(cn, 1, depth);
for (int i = 0; i < cn; ++i)
_mv.create(m.dims, m.size.p, depth, i);
std::vector<Mat> dst;
_mv.getMatVector(dst);
split(m, &dst[0]);
}
CV_CPU_OPTIMIZATION_NAMESPACE_END
}} // namespace
\ No newline at end of file
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