Skip to content

O
opencv
Commit a202dc9a authored by Vitaly Tuzov's avatar Vitaly Tuzov
Browse files
Options

threshold() reworked to use wide universal intrinsics

parent 0e9c90a0
Show whitespace changes
Inline Side-by-side
Showing with 286 additions and 166 deletions
modules/imgproc/src/thresh.cpp
View file @ a202dc9a
......@@ -195,22 +195,19 @@ thresh_8u( const Mat& _src, Mat& _dst, uchar thresh, uchar maxval, int type )
int j = 0;
const uchar* src = _src.ptr();
uchar* dst = _dst.ptr();
#if CV_SIMD128
bool useSIMD = checkHardwareSupport( CV_CPU_SSE2 ) || checkHardwareSupport( CV_CPU_NEON );
if( useSIMD )
{
v_uint8x16 thresh_u = v_setall_u8( thresh );
v_uint8x16 maxval16 = v_setall_u8( maxval );
#if CV_SIMD
v_uint8 thresh_u = vx_setall_u8( thresh );
v_uint8 maxval16 = vx_setall_u8( maxval );
switch( type )
{
case THRESH_BINARY:
for( int i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
for( j = 0; j <= roi.width - 16; j += 16 )
for( j = 0; j <= roi.width - v_uint8::nlanes; j += v_uint8::nlanes)
{
v_uint8x16 v0;
v0 = v_load( src + j );
v_uint8 v0;
v0 = vx_load( src + j );
v0 = thresh_u < v0;
v0 = v0 & maxval16;
v_store( dst + j, v0 );
......@@ -221,10 +218,10 @@ thresh_8u( const Mat& _src, Mat& _dst, uchar thresh, uchar maxval, int type )
case THRESH_BINARY_INV:
for( int i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
for( j = 0; j <= roi.width - 16; j += 16 )
for( j = 0; j <= roi.width - v_uint8::nlanes; j += v_uint8::nlanes)
{
v_uint8x16 v0;
v0 = v_load( src + j );
v_uint8 v0;
v0 = vx_load( src + j );
v0 = v0 <= thresh_u;
v0 = v0 & maxval16;
v_store( dst + j, v0 );
......@@ -235,10 +232,10 @@ thresh_8u( const Mat& _src, Mat& _dst, uchar thresh, uchar maxval, int type )
case THRESH_TRUNC:
for( int i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
for( j = 0; j <= roi.width - 16; j += 16 )
for( j = 0; j <= roi.width - v_uint8::nlanes; j += v_uint8::nlanes)
{
v_uint8x16 v0;
v0 = v_load( src + j );
v_uint8 v0;
v0 = vx_load( src + j );
v0 = v0 - ( v0 - thresh_u );
v_store( dst + j, v0 );
}
......@@ -248,10 +245,10 @@ thresh_8u( const Mat& _src, Mat& _dst, uchar thresh, uchar maxval, int type )
case THRESH_TOZERO:
for( int i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
for( j = 0; j <= roi.width - 16; j += 16 )
for( j = 0; j <= roi.width - v_uint8::nlanes; j += v_uint8::nlanes)
{
v_uint8x16 v0;
v0 = v_load( src + j );
v_uint8 v0;
v0 = vx_load( src + j );
v0 = ( thresh_u < v0 ) & v0;
v_store( dst + j, v0 );
}
......@@ -261,17 +258,16 @@ thresh_8u( const Mat& _src, Mat& _dst, uchar thresh, uchar maxval, int type )
case THRESH_TOZERO_INV:
for( int i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
for( j = 0; j <= roi.width - 16; j += 16 )
for( j = 0; j <= roi.width - v_uint8::nlanes; j += v_uint8::nlanes)
{
v_uint8x16 v0;
v0 = v_load( src + j );
v_uint8 v0;
v0 = vx_load( src + j );
v0 = ( v0 <= thresh_u ) & v0;
v_store( dst + j, v0 );
}
}
break;
}
}
#endif
int j_scalar = j;
......@@ -362,30 +358,35 @@ thresh_16u(const Mat& _src, Mat& _dst, ushort thresh, ushort maxval, int type)
const ushort* src = _src.ptr<ushort>();
ushort* dst = _dst.ptr<ushort>();
#if CV_SIMD128
bool useSIMD = checkHardwareSupport(CV_CPU_SSE2) || checkHardwareSupport(CV_CPU_NEON);
if (useSIMD)
{
#if CV_SIMD
int i, j;
v_uint16x8 thresh_u = v_setall_u16(thresh);
v_uint16x8 maxval16 = v_setall_u16(maxval);
v_uint16 thresh_u = vx_setall_u16(thresh);
v_uint16 maxval16 = vx_setall_u16(maxval);
switch (type)
{
case THRESH_BINARY:
for (i = 0; i < roi.height; i++, src += src_step, dst += dst_step)
{
for (j = 0; j <= roi.width - 16; j += 16)
for (j = 0; j <= roi.width - 2*v_uint16::nlanes; j += 2*v_uint16::nlanes)
{
v_uint16x8 v0, v1;
v0 = v_load(src + j);
v1 = v_load(src + j + 8);
v_uint16 v0, v1;
v0 = vx_load(src + j);
v1 = vx_load(src + j + v_uint16::nlanes);
v0 = thresh_u < v0;
v1 = thresh_u < v1;
v0 = v0 & maxval16;
v1 = v1 & maxval16;
v_store(dst + j, v0);
v_store(dst + j + 8, v1);
v_store(dst + j + v_uint16::nlanes, v1);
}
if (j <= roi.width - v_uint16::nlanes)
{
v_uint16 v0 = vx_load(src + j);
v0 = thresh_u < v0;
v0 = v0 & maxval16;
v_store(dst + j, v0);
j += v_uint16::nlanes;
}
for (; j < roi.width; j++)
......@@ -397,17 +398,25 @@ thresh_16u(const Mat& _src, Mat& _dst, ushort thresh, ushort maxval, int type)
for (i = 0; i < roi.height; i++, src += src_step, dst += dst_step)
{
j = 0;
for (; j <= roi.width - 16; j += 16)
for (; j <= roi.width - 2*v_uint16::nlanes; j += 2*v_uint16::nlanes)
{
v_uint16x8 v0, v1;
v0 = v_load(src + j);
v1 = v_load(src + j + 8);
v_uint16 v0, v1;
v0 = vx_load(src + j);
v1 = vx_load(src + j + v_uint16::nlanes);
v0 = v0 <= thresh_u;
v1 = v1 <= thresh_u;
v0 = v0 & maxval16;
v1 = v1 & maxval16;
v_store(dst + j, v0);
v_store(dst + j + 8, v1);
v_store(dst + j + v_uint16::nlanes, v1);
}
if (j <= roi.width - v_uint16::nlanes)
{
v_uint16 v0 = vx_load(src + j);
v0 = v0 <= thresh_u;
v0 = v0 & maxval16;
v_store(dst + j, v0);
j += v_uint16::nlanes;
}
for (; j < roi.width; j++)
......@@ -419,15 +428,22 @@ thresh_16u(const Mat& _src, Mat& _dst, ushort thresh, ushort maxval, int type)
for (i = 0; i < roi.height; i++, src += src_step, dst += dst_step)
{
j = 0;
for (; j <= roi.width - 16; j += 16)
for (; j <= roi.width - 2*v_uint16::nlanes; j += 2*v_uint16::nlanes)
{
v_uint16x8 v0, v1;
v0 = v_load(src + j);
v1 = v_load(src + j + 8);
v_uint16 v0, v1;
v0 = vx_load(src + j);
v1 = vx_load(src + j + v_uint16::nlanes);
v0 = v_min(v0, thresh_u);
v1 = v_min(v1, thresh_u);
v_store(dst + j, v0);
v_store(dst + j + 8, v1);
v_store(dst + j + v_uint16::nlanes, v1);
}
if (j <= roi.width - v_uint16::nlanes)
{
v_uint16 v0 = vx_load(src + j);
v0 = v_min(v0, thresh_u);
v_store(dst + j, v0);
j += v_uint16::nlanes;
}
for (; j < roi.width; j++)
......@@ -439,15 +455,22 @@ thresh_16u(const Mat& _src, Mat& _dst, ushort thresh, ushort maxval, int type)
for (i = 0; i < roi.height; i++, src += src_step, dst += dst_step)
{
j = 0;
for (; j <= roi.width - 16; j += 16)
for (; j <= roi.width - 2*v_uint16::nlanes; j += 2*v_uint16::nlanes)
{
v_uint16x8 v0, v1;
v0 = v_load(src + j);
v1 = v_load(src + j + 8);
v_uint16 v0, v1;
v0 = vx_load(src + j);
v1 = vx_load(src + j + v_uint16::nlanes);
v0 = (thresh_u < v0) & v0;
v1 = (thresh_u < v1) & v1;
v_store(dst + j, v0);
v_store(dst + j + 8, v1);
v_store(dst + j + v_uint16::nlanes, v1);
}
if (j <= roi.width - v_uint16::nlanes)
{
v_uint16 v0 = vx_load(src + j);
v0 = (thresh_u < v0) & v0;
v_store(dst + j, v0);
j += v_uint16::nlanes;
}
for (; j < roi.width; j++)
......@@ -459,15 +482,22 @@ thresh_16u(const Mat& _src, Mat& _dst, ushort thresh, ushort maxval, int type)
for (i = 0; i < roi.height; i++, src += src_step, dst += dst_step)
{
j = 0;
for (; j <= roi.width - 16; j += 16)
for (; j <= roi.width - 2*v_uint16::nlanes; j += 2*v_uint16::nlanes)
{
v_uint16x8 v0, v1;
v0 = v_load(src + j);
v1 = v_load(src + j + 8);
v_uint16 v0, v1;
v0 = vx_load(src + j);
v1 = vx_load(src + j + v_uint16::nlanes);
v0 = (v0 <= thresh_u) & v0;
v1 = (v1 <= thresh_u) & v1;
v_store(dst + j, v0);
v_store(dst + j + 8, v1);
v_store(dst + j + v_uint16::nlanes, v1);
}
if (j <= roi.width - v_uint16::nlanes)
{
v_uint16 v0 = vx_load(src + j);
v0 = (v0 <= thresh_u) & v0;
v_store(dst + j, v0);
j += v_uint16::nlanes;
}
for (; j < roi.width; j++)
......@@ -475,12 +505,9 @@ thresh_16u(const Mat& _src, Mat& _dst, ushort thresh, ushort maxval, int type)
}
break;
}
}
else
#endif
{
#else
threshGeneric<ushort>(roi, src, src_step, dst, dst_step, thresh, maxval, type);
}
#endif
}
static void
......@@ -556,13 +583,10 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
}
#endif
#if CV_SIMD128
bool useSIMD = checkHardwareSupport( CV_CPU_SSE2 ) || checkHardwareSupport( CV_CPU_NEON );
if( useSIMD )
{
#if CV_SIMD
int i, j;
v_int16x8 thresh8 = v_setall_s16( thresh );
v_int16x8 maxval8 = v_setall_s16( maxval );
v_int16 thresh8 = vx_setall_s16( thresh );
v_int16 maxval8 = vx_setall_s16( maxval );
switch( type )
{
......@@ -570,17 +594,25 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 16; j += 16 )
for( ; j <= roi.width - 2*v_int16::nlanes; j += 2*v_int16::nlanes )
{
v_int16x8 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 8 );
v_int16 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_int16::nlanes );
v0 = thresh8 < v0;
v1 = thresh8 < v1;
v0 = v0 & maxval8;
v1 = v1 & maxval8;
v_store( dst + j, v0 );
v_store( dst + j + 8, v1 );
v_store( dst + j + v_int16::nlanes, v1 );
}
if( j <= roi.width - v_int16::nlanes )
{
v_int16 v0 = vx_load( src + j );
v0 = thresh8 < v0;
v0 = v0 & maxval8;
v_store( dst + j, v0 );
j += v_int16::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -592,17 +624,25 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 16; j += 16 )
for( ; j <= roi.width - 2*v_int16::nlanes; j += 2*v_int16::nlanes )
{
v_int16x8 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 8 );
v_int16 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_int16::nlanes );
v0 = v0 <= thresh8;
v1 = v1 <= thresh8;
v0 = v0 & maxval8;
v1 = v1 & maxval8;
v_store( dst + j, v0 );
v_store( dst + j + 8, v1 );
v_store( dst + j + v_int16::nlanes, v1 );
}
if( j <= roi.width - v_int16::nlanes )
{
v_int16 v0 = vx_load( src + j );
v0 = v0 <= thresh8;
v0 = v0 & maxval8;
v_store( dst + j, v0 );
j += v_int16::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -614,15 +654,22 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 16; j += 16 )
for( ; j <= roi.width - 2*v_int16::nlanes; j += 2*v_int16::nlanes )
{
v_int16x8 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 8 );
v_int16 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_int16::nlanes );
v0 = v_min( v0, thresh8 );
v1 = v_min( v1, thresh8 );
v_store( dst + j, v0 );
v_store( dst + j + 8, v1 );
v_store( dst + j + v_int16::nlanes, v1 );
}
if( j <= roi.width - v_int16::nlanes )
{
v_int16 v0 = vx_load( src + j );
v0 = v_min( v0, thresh8 );
v_store( dst + j, v0 );
j += v_int16::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -634,15 +681,22 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 16; j += 16 )
for( ; j <= roi.width - 2*v_int16::nlanes; j += 2*v_int16::nlanes )
{
v_int16x8 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 8 );
v_int16 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_int16::nlanes );
v0 = ( thresh8 < v0 ) & v0;
v1 = ( thresh8 < v1 ) & v1;
v_store( dst + j, v0 );
v_store( dst + j + 8, v1 );
v_store( dst + j + v_int16::nlanes, v1 );
}
if( j <= roi.width - v_int16::nlanes )
{
v_int16 v0 = vx_load( src + j );
v0 = ( thresh8 < v0 ) & v0;
v_store( dst + j, v0 );
j += v_int16::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -654,15 +708,22 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 16; j += 16 )
for( ; j <= roi.width - 2*v_int16::nlanes; j += 2*v_int16::nlanes )
{
v_int16x8 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 8 );
v_int16 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_int16::nlanes );
v0 = ( v0 <= thresh8 ) & v0;
v1 = ( v1 <= thresh8 ) & v1;
v_store( dst + j, v0 );
v_store( dst + j + 8, v1 );
v_store( dst + j + v_int16::nlanes, v1 );
}
if( j <= roi.width - v_int16::nlanes )
{
v_int16 v0 = vx_load( src + j );
v0 = ( v0 <= thresh8 ) & v0;
v_store( dst + j, v0 );
j += v_int16::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -672,12 +733,9 @@ thresh_16s( const Mat& _src, Mat& _dst, short thresh, short maxval, int type )
default:
CV_Error( CV_StsBadArg, "" ); return;
}
}
else
#endif
{
#else
threshGeneric<short>(roi, src, src_step, dst, dst_step, thresh, maxval, type);
}
#endif
}
......@@ -736,13 +794,10 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
}
#endif
#if CV_SIMD128
bool useSIMD = checkHardwareSupport( CV_CPU_SSE2 ) || checkHardwareSupport( CV_CPU_NEON );
if( useSIMD )
{
#if CV_SIMD
int i, j;
v_float32x4 thresh4 = v_setall_f32( thresh );
v_float32x4 maxval4 = v_setall_f32( maxval );
v_float32 thresh4 = vx_setall_f32( thresh );
v_float32 maxval4 = vx_setall_f32( maxval );
switch( type )
{
......@@ -750,17 +805,25 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 8; j += 8 )
for( ; j <= roi.width - 2*v_float32::nlanes; j += 2*v_float32::nlanes )
{
v_float32x4 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 4 );
v_float32 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float32::nlanes );
v0 = thresh4 < v0;
v1 = thresh4 < v1;
v0 = v0 & maxval4;
v1 = v1 & maxval4;
v_store( dst + j, v0 );
v_store( dst + j + 4, v1 );
v_store( dst + j + v_float32::nlanes, v1 );
}
if( j <= roi.width - v_float32::nlanes )
{
v_float32 v0 = vx_load( src + j );
v0 = thresh4 < v0;
v0 = v0 & maxval4;
v_store( dst + j, v0 );
j += v_float32::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -772,17 +835,25 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 8; j += 8 )
for( ; j <= roi.width - 2*v_float32::nlanes; j += 2*v_float32::nlanes )
{
v_float32x4 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 4 );
v_float32 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float32::nlanes );
v0 = v0 <= thresh4;
v1 = v1 <= thresh4;
v0 = v0 & maxval4;
v1 = v1 & maxval4;
v_store( dst + j, v0 );
v_store( dst + j + 4, v1 );
v_store( dst + j + v_float32::nlanes, v1 );
}
if( j <= roi.width - v_float32::nlanes )
{
v_float32 v0 = vx_load( src + j );
v0 = v0 <= thresh4;
v0 = v0 & maxval4;
v_store( dst + j, v0 );
j += v_float32::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -794,15 +865,22 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 8; j += 8 )
for( ; j <= roi.width - 2*v_float32::nlanes; j += 2*v_float32::nlanes )
{
v_float32x4 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 4 );
v_float32 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float32::nlanes );
v0 = v_min( v0, thresh4 );
v1 = v_min( v1, thresh4 );
v_store( dst + j, v0 );
v_store( dst + j + 4, v1 );
v_store( dst + j + v_float32::nlanes, v1 );
}
if( j <= roi.width - v_float32::nlanes )
{
v_float32 v0 = vx_load( src + j );
v0 = v_min( v0, thresh4 );
v_store( dst + j, v0 );
j += v_float32::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -814,15 +892,22 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 8; j += 8 )
for( ; j <= roi.width - 2*v_float32::nlanes; j += 2*v_float32::nlanes )
{
v_float32x4 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 4 );
v_float32 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float32::nlanes );
v0 = ( thresh4 < v0 ) & v0;
v1 = ( thresh4 < v1 ) & v1;
v_store( dst + j, v0 );
v_store( dst + j + 4, v1 );
v_store( dst + j + v_float32::nlanes, v1 );
}
if( j <= roi.width - v_float32::nlanes )
{
v_float32 v0 = vx_load( src + j );
v0 = ( thresh4 < v0 ) & v0;
v_store( dst + j, v0 );
j += v_float32::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -834,15 +919,22 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 8; j += 8 )
for( ; j <= roi.width - 2*v_float32::nlanes; j += 2*v_float32::nlanes )
{
v_float32x4 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 4 );
v_float32 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float32::nlanes );
v0 = ( v0 <= thresh4 ) & v0;
v1 = ( v1 <= thresh4 ) & v1;
v_store( dst + j, v0 );
v_store( dst + j + 4, v1 );
v_store( dst + j + v_float32::nlanes, v1 );
}
if( j <= roi.width - v_float32::nlanes )
{
v_float32 v0 = vx_load( src + j );
v0 = ( v0 <= thresh4 ) & v0;
v_store( dst + j, v0 );
j += v_float32::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -852,12 +944,9 @@ thresh_32f( const Mat& _src, Mat& _dst, float thresh, float maxval, int type )
default:
CV_Error( CV_StsBadArg, "" ); return;
}
}
else
#endif
{
#else
threshGeneric<float>(roi, src, src_step, dst, dst_step, thresh, maxval, type);
}
#endif
}
static void
......@@ -876,13 +965,10 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
roi.height = 1;
}
#if CV_SIMD128_64F
bool useSIMD = checkHardwareSupport( CV_CPU_SSE2 ) || checkHardwareSupport( CV_CPU_NEON );
if( useSIMD )
{
#if CV_SIMD_64F
int i, j;
v_float64x2 thresh2 = v_setall_f64( thresh );
v_float64x2 maxval2 = v_setall_f64( maxval );
v_float64 thresh2 = vx_setall_f64( thresh );
v_float64 maxval2 = vx_setall_f64( maxval );
switch( type )
{
......@@ -890,17 +976,25 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 4; j += 4 )
for( ; j <= roi.width - 2*v_float64::nlanes; j += 2*v_float64::nlanes )
{
v_float64x2 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 2 );
v_float64 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float64::nlanes );
v0 = thresh2 < v0;
v1 = thresh2 < v1;
v0 = v0 & maxval2;
v1 = v1 & maxval2;
v_store( dst + j, v0 );
v_store( dst + j + 2, v1 );
v_store( dst + j + v_float64::nlanes, v1 );
}
if( j <= roi.width - v_float64::nlanes )
{
v_float64 v0 = vx_load( src + j );
v0 = thresh2 < v0;
v0 = v0 & maxval2;
v_store( dst + j, v0 );
j += v_float64::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -912,17 +1006,25 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 4; j += 4 )
for( ; j <= roi.width - 2*v_float64::nlanes; j += 2*v_float64::nlanes )
{
v_float64x2 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 2 );
v_float64 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float64::nlanes );
v0 = v0 <= thresh2;
v1 = v1 <= thresh2;
v0 = v0 & maxval2;
v1 = v1 & maxval2;
v_store( dst + j, v0 );
v_store( dst + j + 2, v1 );
v_store( dst + j + v_float64::nlanes, v1 );
}
if( j <= roi.width - v_float64::nlanes )
{
v_float64 v0 = vx_load( src + j );
v0 = v0 <= thresh2;
v0 = v0 & maxval2;
v_store( dst + j, v0 );
j += v_float64::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -934,15 +1036,22 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 4; j += 4 )
for( ; j <= roi.width - 2*v_float64::nlanes; j += 2*v_float64::nlanes )
{
v_float64x2 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 2 );
v_float64 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float64::nlanes );
v0 = v_min( v0, thresh2 );
v1 = v_min( v1, thresh2 );
v_store( dst + j, v0 );
v_store( dst + j + 2, v1 );
v_store( dst + j + v_float64::nlanes, v1 );
}
if( j <= roi.width - v_float64::nlanes )
{
v_float64 v0 = vx_load( src + j );
v0 = v_min( v0, thresh2 );
v_store( dst + j, v0 );
j += v_float64::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -954,15 +1063,22 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 4; j += 4 )
for( ; j <= roi.width - 2*v_float64::nlanes; j += 2*v_float64::nlanes )
{
v_float64x2 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 2 );
v_float64 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float64::nlanes );
v0 = ( thresh2 < v0 ) & v0;
v1 = ( thresh2 < v1 ) & v1;
v_store( dst + j, v0 );
v_store( dst + j + 2, v1 );
v_store( dst + j + v_float64::nlanes, v1 );
}
if( j <= roi.width - v_float64::nlanes )
{
v_float64 v0 = vx_load( src + j );
v0 = ( thresh2 < v0 ) & v0;
v_store( dst + j, v0 );
j += v_float64::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -974,15 +1090,22 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
for( i = 0; i < roi.height; i++, src += src_step, dst += dst_step )
{
j = 0;
for( ; j <= roi.width - 4; j += 4 )
for( ; j <= roi.width - 2*v_float64::nlanes; j += 2*v_float64::nlanes )
{
v_float64x2 v0, v1;
v0 = v_load( src + j );
v1 = v_load( src + j + 2 );
v_float64 v0, v1;
v0 = vx_load( src + j );
v1 = vx_load( src + j + v_float64::nlanes );
v0 = ( v0 <= thresh2 ) & v0;
v1 = ( v1 <= thresh2 ) & v1;
v_store( dst + j, v0 );
v_store( dst + j + 2, v1 );
v_store( dst + j + v_float64::nlanes, v1 );
}
if( j <= roi.width - v_float64::nlanes )
{
v_float64 v0 = vx_load( src + j );
v0 = ( v0 <= thresh2 ) & v0;
v_store( dst + j, v0 );
j += v_float64::nlanes;
}
for( ; j < roi.width; j++ )
......@@ -992,12 +1115,9 @@ thresh_64f(const Mat& _src, Mat& _dst, double thresh, double maxval, int type)
default:
CV_Error(CV_StsBadArg, ""); return;
}
}
else
#endif
{
#else
threshGeneric<double>(roi, src, src_step, dst, dst_step, thresh, maxval, type);
}
#endif
}
#ifdef HAVE_IPP
......
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