| [ Web Proxy ] |
| Viewing: https://raw.githubusercontent.com/pythonmjs/opencv/master/modules/core/src/split.cpp | [Back] [Original] |
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html
#include "precomp.hpp"
#include "opencl_kernels_core.hpp"
namespace cv { namespace hal {
#if CV_NEON
template struct VSplit2;
template struct VSplit3;
template struct VSplit4;
#define SPLIT2_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template \
struct name \
{ \
void operator()(const data_type* src, data_type* dst0, \
data_type* dst1) const \
{ \
reg_type r = load_func(src); \
store_func(dst0, r.val[0]); \
store_func(dst1, r.val[1]); \
} \
}
#define SPLIT3_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template \
struct name \
{ \
void operator()(const data_type* src, data_type* dst0, data_type* dst1, \
data_type* dst2) const \
{ \
reg_type r = load_func(src); \
store_func(dst0, r.val[0]); \
store_func(dst1, r.val[1]); \
store_func(dst2, r.val[2]); \
} \
}
#define SPLIT4_KERNEL_TEMPLATE(name, data_type, reg_type, load_func, store_func) \
template \
struct name \
{ \
void operator()(const data_type* src, data_type* dst0, data_type* dst1, \
data_type* dst2, data_type* dst3) const \
{ \
reg_type r = load_func(src); \
store_func(dst0, r.val[0]); \
store_func(dst1, r.val[1]); \
store_func(dst2, r.val[2]); \
store_func(dst3, r.val[3]); \
} \
}
SPLIT2_KERNEL_TEMPLATE(VSplit2, uchar , uint8x16x2_t, vld2q_u8 , vst1q_u8 );
SPLIT2_KERNEL_TEMPLATE(VSplit2, ushort, uint16x8x2_t, vld2q_u16, vst1q_u16);
SPLIT2_KERNEL_TEMPLATE(VSplit2, int , int32x4x2_t, vld2q_s32, vst1q_s32);
SPLIT2_KERNEL_TEMPLATE(VSplit2, int64 , int64x1x2_t, vld2_s64 , vst1_s64 );
SPLIT3_KERNEL_TEMPLATE(VSplit3, uchar , uint8x16x3_t, vld3q_u8 , vst1q_u8 );
SPLIT3_KERNEL_TEMPLATE(VSplit3, ushort, uint16x8x3_t, vld3q_u16, vst1q_u16);
SPLIT3_KERNEL_TEMPLATE(VSplit3, int , int32x4x3_t, vld3q_s32, vst1q_s32);
SPLIT3_KERNEL_TEMPLATE(VSplit3, int64 , int64x1x3_t, vld3_s64 , vst1_s64 );
SPLIT4_KERNEL_TEMPLATE(VSplit4, uchar , uint8x16x4_t, vld4q_u8 , vst1q_u8 );
SPLIT4_KERNEL_TEMPLATE(VSplit4, ushort, uint16x8x4_t, vld4q_u16, vst1q_u16);
SPLIT4_KERNEL_TEMPLATE(VSplit4, int , int32x4x4_t, vld4q_s32, vst1q_s32);
SPLIT4_KERNEL_TEMPLATE(VSplit4, int64 , int64x1x4_t, vld4_s64 , vst1_s64 );
#elif CV_SSE2
template
struct VSplit2
{
VSplit2() : support(false) { }
void operator()(const T *, T *, T *) const { }
bool support;
};
template
struct VSplit3
{
VSplit3() : support(false) { }
void operator()(const T *, T *, T *, T *) const { }
bool support;
};
template
struct VSplit4
{
VSplit4() : support(false) { }
void operator()(const T *, T *, T *, T *, T *) const { }
bool support;
};
#define SPLIT2_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_deinterleave, flavor) \
template \
struct VSplit2 \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VSplit2() \
{ \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, \
data_type * dst0, data_type * dst1) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((cast_type const *)(src)); \
reg_type v_src1 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 2)); \
reg_type v_src3 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 3)); \
\
_mm_deinterleave(v_src0, v_src1, v_src2, v_src3); \
\
_mm_storeu_##flavor((cast_type *)(dst0), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst0 + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst1), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst1 + ELEMS_IN_VEC), v_src3); \
} \
\
bool support; \
}
#define SPLIT3_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_deinterleave, flavor) \
template \
struct VSplit3 \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VSplit3() \
{ \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, \
data_type * dst0, data_type * dst1, data_type * dst2) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((cast_type const *)(src)); \
reg_type v_src1 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 2)); \
reg_type v_src3 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 3)); \
reg_type v_src4 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 4)); \
reg_type v_src5 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 5)); \
\
_mm_deinterleave(v_src0, v_src1, v_src2, \
v_src3, v_src4, v_src5); \
\
_mm_storeu_##flavor((cast_type *)(dst0), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst0 + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst1), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst1 + ELEMS_IN_VEC), v_src3); \
_mm_storeu_##flavor((cast_type *)(dst2), v_src4); \
_mm_storeu_##flavor((cast_type *)(dst2 + ELEMS_IN_VEC), v_src5); \
} \
\
bool support; \
}
#define SPLIT4_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_deinterleave, flavor) \
template \
struct VSplit4 \
{ \
enum \
{ \
ELEMS_IN_VEC = 16 / sizeof(data_type) \
}; \
\
VSplit4() \
{ \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, data_type * dst0, data_type * dst1, \
data_type * dst2, data_type * dst3) const \
{ \
reg_type v_src0 = _mm_loadu_##flavor((cast_type const *)(src)); \
reg_type v_src1 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC)); \
reg_type v_src2 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 2)); \
reg_type v_src3 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 3)); \
reg_type v_src4 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 4)); \
reg_type v_src5 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 5)); \
reg_type v_src6 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 6)); \
reg_type v_src7 = _mm_loadu_##flavor((cast_type const *)(src + ELEMS_IN_VEC * 7)); \
\
_mm_deinterleave(v_src0, v_src1, v_src2, v_src3, \
v_src4, v_src5, v_src6, v_src7); \
\
_mm_storeu_##flavor((cast_type *)(dst0), v_src0); \
_mm_storeu_##flavor((cast_type *)(dst0 + ELEMS_IN_VEC), v_src1); \
_mm_storeu_##flavor((cast_type *)(dst1), v_src2); \
_mm_storeu_##flavor((cast_type *)(dst1 + ELEMS_IN_VEC), v_src3); \
_mm_storeu_##flavor((cast_type *)(dst2), v_src4); \
_mm_storeu_##flavor((cast_type *)(dst2 + ELEMS_IN_VEC), v_src5); \
_mm_storeu_##flavor((cast_type *)(dst3), v_src6); \
_mm_storeu_##flavor((cast_type *)(dst3 + ELEMS_IN_VEC), v_src7); \
} \
\
bool support; \
}
SPLIT2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
SPLIT2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
SPLIT2_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
SPLIT3_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
SPLIT3_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
SPLIT3_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
SPLIT4_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_deinterleave_epi8, si128);
SPLIT4_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_deinterleave_epi16, si128);
SPLIT4_KERNEL_TEMPLATE( int, __m128, float, _mm_deinterleave_ps, ps);
#endif
template 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;
#if CV_NEON
if(cn == 2)
{
int inc_i = (sizeof(T) == 8)? 1: 16/sizeof(T);
int inc_j = 2 * inc_i;
VSplit2 vsplit;
for( ; i < len - inc_i; i += inc_i, j += inc_j)
vsplit(src + j, dst0 + i, dst1 + i);
}
#elif CV_SSE2
if (cn == 2)
{
int inc_i = 32/sizeof(T);
int inc_j = 2 * inc_i;
VSplit2 vsplit;
if (vsplit.support)
{
for( ; i
| Web Proxy Viewer | New URL | Original Page |