Administrator/opencv-MIRROR
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-09-11 04:43:22 -05:00
Code Issues Packages Projects Releases Wiki Activity

imgproc: vectorize symmetric CV_32S, CV_16S column filter

Browse Source
This commit is contained in:
Pratham Kumar
2026-08-27 15:51:44 +05:30
parent 1c59b23c9f
commit 8714d8afec
1 changed files with 118 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

139
modules/imgproc/src/filter.simd.hpp
View File

@@ -796,13 +796,14 @@ struct RowVec_8u32f
struct SymmRowSmallVec_8u32s
{
SymmRowSmallVec_8u32s() { smallValues = false; symmetryType = 0; }
SymmRowSmallVec_8u32s() { smallValues = false; int16Sums = false; symmetryType = 0; }
SymmRowSmallVec_8u32s( const Mat& _kernel, int _symmetryType )
{
kernel = _kernel;
symmetryType = _symmetryType;
smallValues = true;
int k, ksize = kernel.rows + kernel.cols - 1;
int absSum = 0;
for( k = 0; k < ksize; k++ )
{
int v = kernel.ptr<int>()[k];
@@ -811,7 +812,9 @@ struct SymmRowSmallVec_8u32s
smallValues = false;
break;
}
absSum += std::abs(v);
}
int16Sums = smallValues && absSum*UCHAR_MAX <= SHRT_MAX;
}
int operator()(const uchar* src, uchar* _dst, int width, int cn) const
@@ -944,34 +947,46 @@ struct SymmRowSmallVec_8u32s
}
else if( _ksize == 5 )
{
if( kx[0] == -2 && kx[1] == 0 && kx[2] == 1 )
if( int16Sums )
{
v_int16 k0 = vx_setall_s16((short)kx[0]);
v_int16 k1 = vx_setall_s16((short)kx[1]);
v_int16 k2 = vx_setall_s16((short)kx[2]);
for( ; i <= width - VTraits<v_uint8>::vlanes(); i += VTraits<v_uint8>::vlanes(), src += VTraits<v_uint8>::vlanes() )
{
v_uint16 x0l, x0h, x1l, x1h, x2l, x2h;
v_expand(vx_load(src - 2*cn), x0l, x0h);
v_expand(vx_load(src), x1l, x1h);
v_expand(vx_load(src + 2*cn), x2l, x2h);
x1l = v_sub_wrap(v_add_wrap(x0l, x2l), v_add_wrap(x1l, x1l));
x1h = v_sub_wrap(v_add_wrap(x0h, x2h), v_add_wrap(x1h, x1h));
v_store(dst + i, v_expand_low(v_reinterpret_as_s16(x1l)));
v_store(dst + i + VTraits<v_int32>::vlanes(), v_expand_high(v_reinterpret_as_s16(x1l)));
v_store(dst + i + 2*VTraits<v_int32>::vlanes(), v_expand_low(v_reinterpret_as_s16(x1h)));
v_store(dst + i + 3*VTraits<v_int32>::vlanes(), v_expand_high(v_reinterpret_as_s16(x1h)));
v_uint16 c0, c1, a0, a1, b0, b1, d0, d1, e0, e1;
v_expand(vx_load(src), c0, c1);
v_expand(vx_load(src - cn), a0, a1);
v_expand(vx_load(src + cn), b0, b1);
v_expand(vx_load(src - 2*cn), d0, d1);
v_expand(vx_load(src + 2*cn), e0, e1);
v_int16 sl = v_add_wrap(v_add_wrap(v_mul_wrap(v_reinterpret_as_s16(c0), k0),
v_mul_wrap(v_reinterpret_as_s16(v_add_wrap(a0, b0)), k1)),
v_mul_wrap(v_reinterpret_as_s16(v_add_wrap(d0, e0)), k2));
v_int16 sh = v_add_wrap(v_add_wrap(v_mul_wrap(v_reinterpret_as_s16(c1), k0),
v_mul_wrap(v_reinterpret_as_s16(v_add_wrap(a1, b1)), k1)),
v_mul_wrap(v_reinterpret_as_s16(v_add_wrap(d1, e1)), k2));
v_store(dst + i, v_expand_low(sl));
v_store(dst + i + VTraits<v_int32>::vlanes(), v_expand_high(sl));
v_store(dst + i + 2*VTraits<v_int32>::vlanes(), v_expand_low(sh));
v_store(dst + i + 3*VTraits<v_int32>::vlanes(), v_expand_high(sh));
}
if( i <= width - VTraits<v_uint16>::vlanes() )
{
v_uint16 x = vx_load_expand(src);
x = v_sub_wrap(v_add_wrap(vx_load_expand(src - 2*cn), vx_load_expand(src + 2*cn)), v_add_wrap(x, x));
v_store(dst + i, v_expand_low(v_reinterpret_as_s16(x)));
v_store(dst + i + VTraits<v_int32>::vlanes(), v_expand_high(v_reinterpret_as_s16(x)));
v_int16 s = v_add_wrap(v_add_wrap(v_mul_wrap(v_reinterpret_as_s16(vx_load_expand(src)), k0),
v_mul_wrap(v_reinterpret_as_s16(v_add_wrap(vx_load_expand(src - cn), vx_load_expand(src + cn))), k1)),
v_mul_wrap(v_reinterpret_as_s16(v_add_wrap(vx_load_expand(src - 2*cn), vx_load_expand(src + 2*cn))), k2));
v_store(dst + i, v_expand_low(s));
v_store(dst + i + VTraits<v_int32>::vlanes(), v_expand_high(s));
i += VTraits<v_uint16>::vlanes(); src += VTraits<v_uint16>::vlanes();
}
if( i <= width - VTraits<v_uint32>::vlanes() )
{
v_int32 x = v_reinterpret_as_s32(vx_load_expand_q(src));
x = v_sub(v_reinterpret_as_s32(v_add(vx_load_expand_q(src - 2 * cn), vx_load_expand_q(src + 2 * cn))), v_add(x, x));
v_store(dst + i, x);
v_store(dst + i, v_muladd(v_reinterpret_as_s32(vx_load_expand_q(src)), vx_setall_s32(kx[0]),
v_muladd(v_reinterpret_as_s32(v_add(vx_load_expand_q(src - cn), vx_load_expand_q(src + cn))), vx_setall_s32(kx[1]),
v_mul(v_reinterpret_as_s32(v_add(vx_load_expand_q(src - 2 * cn), vx_load_expand_q(src + 2 * cn))), vx_setall_s32(kx[2])))));
i += VTraits<v_uint32>::vlanes();
}
}
@@ -1289,6 +1304,7 @@ struct SymmRowSmallVec_8u32s
Mat kernel;
int symmetryType;
bool smallValues;
bool int16Sums;
};
@@ -1491,6 +1507,85 @@ struct SymmColumnVec_32f8u
Mat kernel;
};
struct SymmColumnVec_32s16s
{
SymmColumnVec_32s16s() { symmetryType = 0; delta = 0; }
SymmColumnVec_32s16s(const Mat& _kernel, int _symmetryType, int, double _delta)
{
symmetryType = _symmetryType;
kernel = _kernel;
delta = saturate_cast<int>(_delta);
CV_Assert( (symmetryType & (KERNEL_SYMMETRICAL | KERNEL_ASYMMETRICAL)) != 0 );
}
int operator()(const uchar** _src, uchar* _dst, int width) const
{
CV_INSTRUMENT_REGION();
int _ksize = kernel.rows + kernel.cols - 1;
if( _ksize == 1 )
return 0;
int ksize2 = _ksize/2;
const int* ky = kernel.ptr<int>() + ksize2;
int i = 0, k;
bool symmetrical = (symmetryType & KERNEL_SYMMETRICAL) != 0;
const int** src = (const int**)_src;
short* dst = (short*)_dst;
const int step = VTraits<v_int32>::vlanes();
v_int32 d4 = vx_setall_s32(delta);
if( symmetrical )
{
v_int32 f0 = vx_setall_s32(ky[0]);
for( ; i <= width - 2*VTraits<v_int16>::vlanes(); i += 2*VTraits<v_int16>::vlanes() )
{
const int* S = src[0] + i;
v_int32 s0 = v_muladd(vx_load(S), f0, d4);
v_int32 s1 = v_muladd(vx_load(S + step), f0, d4);
v_int32 s2 = v_muladd(vx_load(S + 2*step), f0, d4);
v_int32 s3 = v_muladd(vx_load(S + 3*step), f0, d4);
for( k = 1; k <= ksize2; k++ )
{
v_int32 f = vx_setall_s32(ky[k]);
const int* S0 = src[k] + i;
const int* S1 = src[-k] + i;
s0 = v_muladd(v_add(vx_load(S0), vx_load(S1)), f, s0);
s1 = v_muladd(v_add(vx_load(S0 + step), vx_load(S1 + step)), f, s1);
s2 = v_muladd(v_add(vx_load(S0 + 2*step), vx_load(S1 + 2*step)), f, s2);
s3 = v_muladd(v_add(vx_load(S0 + 3*step), vx_load(S1 + 3*step)), f, s3);
}
v_store(dst + i, v_pack(s0, s1));
v_store(dst + i + VTraits<v_int16>::vlanes(), v_pack(s2, s3));
}
}
else
{
for( ; i <= width - 2*VTraits<v_int16>::vlanes(); i += 2*VTraits<v_int16>::vlanes() )
{
v_int32 s0 = d4, s1 = d4, s2 = d4, s3 = d4;
for( k = 1; k <= ksize2; k++ )
{
v_int32 f = vx_setall_s32(ky[k]);
const int* S0 = src[k] + i;
const int* S1 = src[-k] + i;
s0 = v_muladd(v_sub(vx_load(S0), vx_load(S1)), f, s0);
s1 = v_muladd(v_sub(vx_load(S0 + step), vx_load(S1 + step)), f, s1);
s2 = v_muladd(v_sub(vx_load(S0 + 2*step), vx_load(S1 + 2*step)), f, s2);
s3 = v_muladd(v_sub(vx_load(S0 + 3*step), vx_load(S1 + 3*step)), f, s3);
}
v_store(dst + i, v_pack(s0, s1));
v_store(dst + i + VTraits<v_int16>::vlanes(), v_pack(s2, s3));
}
}
return i;
}
int symmetryType;
int delta;
Mat kernel;
};
struct SymmColumnSmallVec_32s16s
{
@@ -2656,6 +2751,7 @@ typedef SymmRowSmallNoVec SymmRowSmallVec_8u32s;
typedef SymmRowSmallNoVec SymmRowSmallVec_32f;
typedef ColumnNoVec SymmColumnVec_32s8u;
typedef ColumnNoVec SymmColumnVec_32f8u;
typedef ColumnNoVec SymmColumnVec_32s16s;
typedef ColumnNoVec SymmColumnVec_32f16s;
typedef ColumnNoVec SymmColumnVec_32f;
typedef SymmColumnSmallNoVec SymmColumnSmallVec_32s16s;
@@ -3363,8 +3459,9 @@ Ptr<BaseColumnFilter> getLinearColumnFilter(
return makePtr<SymmColumnFilter<Cast<double, ushort>, ColumnNoVec> >
(kernel, anchor, delta, symmetryType);
if( ddepth == CV_16S && sdepth == CV_32S )
return makePtr<SymmColumnFilter<Cast<int, short>, ColumnNoVec> >
(kernel, anchor, delta, symmetryType);
return makePtr<SymmColumnFilter<Cast<int, short>, SymmColumnVec_32s16s> >
(kernel, anchor, delta, symmetryType, Cast<int, short>(),
SymmColumnVec_32s16s(kernel, symmetryType, bits, delta));
if( ddepth == CV_16S && sdepth == CV_32F )
return makePtr<SymmColumnFilter<Cast<float, short>, SymmColumnVec_32f16s> >
(kernel, anchor, delta, symmetryType, Cast<float, short>(),