generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

checks

Browse Source
This commit is contained in:
Ilya Lavrenov
2015-01-12 10:59:31 +03:00
parent 1d3c860411
commit 6bce6ee34a
7 changed files with 841 additions and 632 deletions
Download Patch File Download Diff File Expand all files Collapse all files

89
modules/core/src/arithm.cpp
View File

@@ -64,7 +64,7 @@ FUNCTOR_TEMPLATE(VLoadStore128);
#if CV_SSE2
FUNCTOR_TEMPLATE(VLoadStore64);
FUNCTOR_TEMPLATE(VLoadStore128Aligned);
#if CV_AVX
#if CV_AVX2
FUNCTOR_TEMPLATE(VLoadStore256);
FUNCTOR_TEMPLATE(VLoadStore256Aligned);
#endif
@@ -2626,10 +2626,16 @@ struct Div_SIMD
template <>
struct Div_SIMD<uchar>
{
bool haveSIMD;
Div_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE4_1); }
int operator() (const uchar * src1, const uchar * src2, uchar * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2672,10 +2678,16 @@ struct Div_SIMD<uchar>
template <>
struct Div_SIMD<schar>
{
bool haveSIMD;
Div_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE2); }
int operator() (const schar * src1, const schar * src2, schar * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2718,10 +2730,16 @@ struct Div_SIMD<schar>
template <>
struct Div_SIMD<ushort>
{
bool haveSIMD;
Div_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE4_1); }
int operator() (const ushort * src1, const ushort * src2, ushort * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2763,10 +2781,16 @@ struct Div_SIMD<ushort>
template <>
struct Div_SIMD<short>
{
bool haveSIMD;
Div_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE2); }
int operator() (const short * src1, const short * src2, short * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2806,10 +2830,16 @@ struct Div_SIMD<short>
template <>
struct Div_SIMD<int>
{
bool haveSIMD;
Div_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE2); }
int operator() (const int * src1, const int * src2, int * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2902,10 +2932,16 @@ struct Recip_SIMD
template <>
struct Recip_SIMD<uchar>
{
bool haveSIMD;
Recip_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE4_1); }
int operator() (const uchar * src2, uchar * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2941,10 +2977,16 @@ struct Recip_SIMD<uchar>
template <>
struct Recip_SIMD<schar>
{
bool haveSIMD;
Recip_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE2); }
int operator() (const schar * src2, schar * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -2980,10 +3022,16 @@ struct Recip_SIMD<schar>
template <>
struct Recip_SIMD<ushort>
{
bool haveSIMD;
Recip_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE4_1); }
int operator() (const ushort * src2, ushort * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -3018,10 +3066,16 @@ struct Recip_SIMD<ushort>
template <>
struct Recip_SIMD<short>
{
bool haveSIMD;
Recip_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE2); }
int operator() (const short * src2, short * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -3054,10 +3108,16 @@ struct Recip_SIMD<short>
template <>
struct Recip_SIMD<int>
{
bool haveSIMD;
Recip_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE2); }
int operator() (const int * src2, int * dst, int width, double scale) const
{
int x = 0;
if (!haveSIMD)
return x;
__m128d v_scale = _mm_set1_pd(scale);
__m128i v_zero = _mm_setzero_si128();
@@ -4126,7 +4186,8 @@ static void cmp8u(const uchar* src1, size_t step1, const uchar* src2, size_t ste
{
int x =0;
#if CV_SSE2
if( USE_SSE2 ){
if( USE_SSE2 )
{
__m128i m128 = code == CMP_GT ? _mm_setzero_si128() : _mm_set1_epi8 (-1);
__m128i c128 = _mm_set1_epi8 (-128);
for( ; x <= size.width - 16; x += 16 )
@@ -4142,7 +4203,7 @@ static void cmp8u(const uchar* src1, size_t step1, const uchar* src2, size_t ste
}
}
#elif CV_NEON
#elif CV_NEON
uint8x16_t mask = code == CMP_GT ? vdupq_n_u8(0) : vdupq_n_u8(255);
for( ; x <= size.width - 16; x += 16 )
@@ -4164,7 +4225,8 @@ static void cmp8u(const uchar* src1, size_t step1, const uchar* src2, size_t ste
{
int x = 0;
#if CV_SSE2
if( USE_SSE2 ){
if( USE_SSE2 )
{
__m128i m128 = code == CMP_EQ ? _mm_setzero_si128() : _mm_set1_epi8 (-1);
for( ; x <= size.width - 16; x += 16 )
{
@@ -4174,7 +4236,7 @@ static void cmp8u(const uchar* src1, size_t step1, const uchar* src2, size_t ste
_mm_storeu_si128((__m128i*)(dst + x), r00);
}
}
#elif CV_NEON
#elif CV_NEON
uint8x16_t mask = code == CMP_EQ ? vdupq_n_u8(0) : vdupq_n_u8(255);
for( ; x <= size.width - 16; x += 16 )
@@ -4254,7 +4316,8 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
{
int x =0;
#if CV_SSE2
if( USE_SSE2){//
if( USE_SSE2)
{
__m128i m128 = code == CMP_GT ? _mm_setzero_si128() : _mm_set1_epi16 (-1);
for( ; x <= size.width - 16; x += 16 )
{
@@ -4278,7 +4341,7 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
x += 8;
}
}
#elif CV_NEON
#elif CV_NEON
uint8x16_t mask = code == CMP_GT ? vdupq_n_u8(0) : vdupq_n_u8(255);
for( ; x <= size.width - 16; x += 16 )
@@ -4293,8 +4356,7 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
vst1q_u8(dst+x, veorq_u8(vcombine_u8(t1, t2), mask));
}
#endif
#endif
for( ; x < size.width; x++ ){
dst[x] = (uchar)(-(src1[x] > src2[x]) ^ m);
@@ -4308,7 +4370,8 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
{
int x = 0;
#if CV_SSE2
if( USE_SSE2 ){
if( USE_SSE2 )
{
__m128i m128 = code == CMP_EQ ? _mm_setzero_si128() : _mm_set1_epi16 (-1);
for( ; x <= size.width - 16; x += 16 )
{
@@ -4332,7 +4395,7 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
x += 8;
}
}
#elif CV_NEON
#elif CV_NEON
uint8x16_t mask = code == CMP_EQ ? vdupq_n_u8(0) : vdupq_n_u8(255);
for( ; x <= size.width - 16; x += 16 )
@@ -4347,8 +4410,8 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
vst1q_u8(dst+x, veorq_u8(vcombine_u8(t1, t2), mask));
}
#endif
for( ; x < size.width; x++ )
#endif
for( ; x < size.width; x++ )
dst[x] = (uchar)(-(src1[x] == src2[x]) ^ m);
}
}

62
modules/core/src/convert.cpp
View File

@@ -158,7 +158,7 @@ struct VSplit2<data_type>
\
VSplit2() \
{ \
support = true; \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, \
@@ -191,7 +191,7 @@ struct VSplit3<data_type>
\
VSplit3() \
{ \
support = true; \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, \
@@ -229,7 +229,7 @@ struct VSplit4<data_type>
\
VSplit4() \
{ \
support = true; \
support = checkHardwareSupport(CV_CPU_SSE2); \
} \
\
void operator()(const data_type * src, data_type * dst0, data_type * dst1, \
@@ -502,7 +502,7 @@ struct VMerge4
bool support;
};
#define MERGE2_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor) \
#define MERGE2_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor, se) \
template <> \
struct VMerge2<data_type> \
{ \
@@ -513,7 +513,7 @@ struct VMerge2<data_type>
\
VMerge2() \
{ \
support = true; \
support = checkHardwareSupport(se); \
} \
\
void operator()(const data_type * src0, const data_type * src1, \
@@ -535,7 +535,7 @@ struct VMerge2<data_type>
bool support; \
}
#define MERGE3_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor) \
#define MERGE3_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor, se) \
template <> \
struct VMerge3<data_type> \
{ \
@@ -546,7 +546,7 @@ struct VMerge3<data_type>
\
VMerge3() \
{ \
support = true; \
support = checkHardwareSupport(se); \
} \
\
void operator()(const data_type * src0, const data_type * src1, const data_type * src2,\
@@ -573,7 +573,7 @@ struct VMerge3<data_type>
bool support; \
}
#define MERGE4_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor) \
#define MERGE4_KERNEL_TEMPLATE(data_type, reg_type, cast_type, _mm_interleave, flavor, se) \
template <> \
struct VMerge4<data_type> \
{ \
@@ -584,7 +584,7 @@ struct VMerge4<data_type>
\
VMerge4() \
{ \
support = true; \
support = checkHardwareSupport(se); \
} \
\
void operator()(const data_type * src0, const data_type * src1, \
@@ -616,19 +616,19 @@ struct VMerge4<data_type>
bool support; \
}
MERGE2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128);
MERGE3_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128);
MERGE4_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128);
MERGE2_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128, CV_CPU_SSE2);
MERGE3_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128, CV_CPU_SSE2);
MERGE4_KERNEL_TEMPLATE( uchar, __m128i, __m128i, _mm_interleave_epi8, si128, CV_CPU_SSE2);
#if CV_SSE4_1
MERGE2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128);
MERGE3_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128);
MERGE4_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128);
MERGE2_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128, CV_CPU_SSE4_1);
MERGE3_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128, CV_CPU_SSE4_1);
MERGE4_KERNEL_TEMPLATE(ushort, __m128i, __m128i, _mm_interleave_epi16, si128, CV_CPU_SSE4_1);
#endif
MERGE2_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps);
MERGE3_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps);
MERGE4_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps);
MERGE2_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps, CV_CPU_SSE2);
MERGE3_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps, CV_CPU_SSE2);
MERGE4_KERNEL_TEMPLATE( int, __m128, float, _mm_interleave_ps, ps, CV_CPU_SSE2);
#endif
@@ -4404,6 +4404,9 @@ struct Cvt_SIMD<double, uchar>
{
int x = 0;
if (!USE_SSE2)
return x;
for ( ; x <= width - 8; x += 8)
{
__m128 v_src0 = _mm_cvtpd_ps(_mm_loadu_pd(src + x));
@@ -4430,6 +4433,9 @@ struct Cvt_SIMD<double, schar>
{
int x = 0;
if (!USE_SSE2)
return x;
for ( ; x <= width - 8; x += 8)
{
__m128 v_src0 = _mm_cvtpd_ps(_mm_loadu_pd(src + x));
@@ -4454,10 +4460,16 @@ struct Cvt_SIMD<double, schar>
template <>
struct Cvt_SIMD<double, ushort>
{
bool haveSIMD;
Cvt_SIMD() { haveSIMD = checkHardwareSupport(CV_CPU_SSE4_1); }
int operator() (const double * src, ushort * dst, int width) const
{
int x = 0;
if (!haveSIMD)
return x;
for ( ; x <= width - 8; x += 8)
{
__m128 v_src0 = _mm_cvtpd_ps(_mm_loadu_pd(src + x));
@@ -4486,6 +4498,9 @@ struct Cvt_SIMD<double, short>
{
int x = 0;
if (!USE_SSE2)
return x;
for ( ; x <= width - 8; x += 8)
{
__m128 v_src0 = _mm_cvtpd_ps(_mm_loadu_pd(src + x));
@@ -4512,6 +4527,9 @@ struct Cvt_SIMD<double, int>
{
int x = 0;
if (!USE_SSE2)
return x;
for ( ; x <= width - 4; x += 4)
{
__m128 v_src0 = _mm_cvtpd_ps(_mm_loadu_pd(src + x));
@@ -4532,6 +4550,9 @@ struct Cvt_SIMD<double, float>
{
int x = 0;
if (!USE_SSE2)
return x;
for ( ; x <= width - 4; x += 4)
{
__m128 v_src0 = _mm_cvtpd_ps(_mm_loadu_pd(src + x));
@@ -5114,8 +5135,9 @@ cvt_<float, short>( const float* src, size_t sstep,
{
int x = 0;
#if CV_SSE2
if(USE_SSE2){
for( ; x <= size.width - 8; x += 8 )
if(USE_SSE2)
{
for( ; x <= size.width - 8; x += 8 )
{
__m128 src128 = _mm_loadu_ps (src + x);
__m128i src_int128 = _mm_cvtps_epi32 (src128);

102
modules/core/src/mathfuncs.cpp
View File

@@ -597,15 +597,18 @@ void phase( InputArray src1, InputArray src2, OutputArray dst, bool angleInDegre
k = 0;
#if CV_SSE2
for ( ; k <= len - 4; k += 4)
if (USE_SSE2)
{
__m128 v_dst0 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(x + k)),
_mm_cvtpd_ps(_mm_loadu_pd(x + k + 2)));
__m128 v_dst1 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(y + k)),
_mm_cvtpd_ps(_mm_loadu_pd(y + k + 2)));
for ( ; k <= len - 4; k += 4)
{
__m128 v_dst0 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(x + k)),
_mm_cvtpd_ps(_mm_loadu_pd(x + k + 2)));
__m128 v_dst1 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(y + k)),
_mm_cvtpd_ps(_mm_loadu_pd(y + k + 2)));
_mm_storeu_ps(buf[0] + k, v_dst0);
_mm_storeu_ps(buf[1] + k, v_dst1);
_mm_storeu_ps(buf[0] + k, v_dst0);
_mm_storeu_ps(buf[1] + k, v_dst1);
}
}
#endif
@@ -619,11 +622,14 @@ void phase( InputArray src1, InputArray src2, OutputArray dst, bool angleInDegre
k = 0;
#if CV_SSE2
for ( ; k <= len - 4; k += 4)
if (USE_SSE2)
{
__m128 v_src = _mm_loadu_ps(buf[0] + k);
_mm_storeu_pd(angle + k, _mm_cvtps_pd(v_src));
_mm_storeu_pd(angle + k + 2, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
for ( ; k <= len - 4; k += 4)
{
__m128 v_src = _mm_loadu_ps(buf[0] + k);
_mm_storeu_pd(angle + k, _mm_cvtps_pd(v_src));
_mm_storeu_pd(angle + k + 2, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
}
}
#endif
@@ -728,15 +734,18 @@ void cartToPolar( InputArray src1, InputArray src2,
k = 0;
#if CV_SSE2
for ( ; k <= len - 4; k += 4)
if (USE_SSE2)
{
__m128 v_dst0 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(x + k)),
_mm_cvtpd_ps(_mm_loadu_pd(x + k + 2)));
__m128 v_dst1 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(y + k)),
_mm_cvtpd_ps(_mm_loadu_pd(y + k + 2)));
for ( ; k <= len - 4; k += 4)
{
__m128 v_dst0 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(x + k)),
_mm_cvtpd_ps(_mm_loadu_pd(x + k + 2)));
__m128 v_dst1 = _mm_movelh_ps(_mm_cvtpd_ps(_mm_loadu_pd(y + k)),
_mm_cvtpd_ps(_mm_loadu_pd(y + k + 2)));
_mm_storeu_ps(buf[0] + k, v_dst0);
_mm_storeu_ps(buf[1] + k, v_dst1);
_mm_storeu_ps(buf[0] + k, v_dst0);
_mm_storeu_ps(buf[1] + k, v_dst1);
}
}
#endif
@@ -750,11 +759,14 @@ void cartToPolar( InputArray src1, InputArray src2,
k = 0;
#if CV_SSE2
for ( ; k <= len - 4; k += 4)
if (USE_SSE2)
{
__m128 v_src = _mm_loadu_ps(buf[0] + k);
_mm_storeu_pd(angle + k, _mm_cvtps_pd(v_src));
_mm_storeu_pd(angle + k + 2, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
for ( ; k <= len - 4; k += 4)
{
__m128 v_src = _mm_loadu_ps(buf[0] + k);
_mm_storeu_pd(angle + k, _mm_cvtps_pd(v_src));
_mm_storeu_pd(angle + k + 2, _mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_src), 8))));
}
}
#endif
@@ -832,17 +844,16 @@ static void SinCos_32f( const float *angle, float *sinval, float* cosval,
k1 = N/360.;
#if CV_AVX2
__m128d v_i = _mm_set_pd(1, 0);
__m128d v_k1 = _mm_set1_pd(k1);
__m128d v_1 = _mm_set1_pd(1);
__m128i v_N1 = _mm_set1_epi32(N - 1);
__m128i v_N4 = _mm_set1_epi32(N >> 2);
__m128d v_sin_a0 = _mm_set1_pd(sin_a0);
__m128d v_sin_a2 = _mm_set1_pd(sin_a2);
__m128d v_cos_a0 = _mm_set1_pd(cos_a0);
if (USE_AVX2)
{
__m128d v_k1 = _mm_set1_pd(k1);
__m128d v_1 = _mm_set1_pd(1);
__m128i v_N1 = _mm_set1_epi32(N - 1);
__m128i v_N4 = _mm_set1_epi32(N >> 2);
__m128d v_sin_a0 = _mm_set1_pd(sin_a0);
__m128d v_sin_a2 = _mm_set1_pd(sin_a2);
__m128d v_cos_a0 = _mm_set1_pd(cos_a0);
for ( ; i <= len - 4; i += 4)
{
__m128 v_angle = _mm_loadu_ps(angle + i);
@@ -859,8 +870,8 @@ static void SinCos_32f( const float *angle, float *sinval, float* cosval,
__m128d v_sin_b = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(v_sin_a0, v_t2), v_sin_a2), v_t);
__m128d v_cos_b = _mm_add_pd(_mm_mul_pd(v_cos_a0, v_t2), v_1);
__m128d v_sin_a = _mm_i32gather_pd(sin_table, v_sin_idx, 1);
__m128d v_cos_a = _mm_i32gather_pd(sin_table, v_cos_idx, 1);
__m128d v_sin_a = _mm_i32gather_pd(sin_table, v_sin_idx, 8);
__m128d v_cos_a = _mm_i32gather_pd(sin_table, v_cos_idx, 8);
__m128d v_sin_val_0 = _mm_add_pd(_mm_mul_pd(v_sin_a, v_cos_b),
_mm_mul_pd(v_cos_a, v_sin_b));
@@ -868,7 +879,7 @@ static void SinCos_32f( const float *angle, float *sinval, float* cosval,
_mm_mul_pd(v_sin_a, v_sin_b));
// 2-3
v_t = _mm_mul_pd(_mm_cvtps_pd(_mm_castsi128_ps(_mm_slli_si128(_mm_castps_si128(v_angle), 8))), v_k1);
v_t = _mm_mul_pd(_mm_cvtps_pd(_mm_castsi128_ps(_mm_srli_si128(_mm_castps_si128(v_angle), 8))), v_k1);
v_it = _mm_cvtpd_epi32(v_t);
v_t = _mm_sub_pd(v_t, _mm_cvtepi32_pd(v_it));
@@ -879,8 +890,8 @@ static void SinCos_32f( const float *angle, float *sinval, float* cosval,
v_sin_b = _mm_mul_pd(_mm_add_pd(_mm_mul_pd(v_sin_a0, v_t2), v_sin_a2), v_t);
v_cos_b = _mm_add_pd(_mm_mul_pd(v_cos_a0, v_t2), v_1);
v_sin_a = _mm_i32gather_pd(sin_table, v_sin_idx, 1);
v_cos_a = _mm_i32gather_pd(sin_table, v_cos_idx, 1);
v_sin_a = _mm_i32gather_pd(sin_table, v_sin_idx, 8);
v_cos_a = _mm_i32gather_pd(sin_table, v_cos_idx, 8);
__m128d v_sin_val_1 = _mm_add_pd(_mm_mul_pd(v_sin_a, v_cos_b),
_mm_mul_pd(v_cos_a, v_sin_b));
@@ -1032,11 +1043,14 @@ void polarToCart( InputArray src1, InputArray src2,
vst1q_f32(y + k, vmulq_f32(vld1q_f32(y + k), v_m));
}
#elif CV_SSE2
for( ; k <= len - 4; k += 4 )
if (USE_SSE2)
{
__m128 v_m = _mm_loadu_ps(mag + k);
_mm_storeu_ps(x + k, _mm_mul_ps(_mm_loadu_ps(x + k), v_m));
_mm_storeu_ps(y + k, _mm_mul_ps(_mm_loadu_ps(y + k), v_m));
for( ; k <= len - 4; k += 4 )
{
__m128 v_m = _mm_loadu_ps(mag + k);
_mm_storeu_ps(x + k, _mm_mul_ps(_mm_loadu_ps(x + k), v_m));
_mm_storeu_ps(y + k, _mm_mul_ps(_mm_loadu_ps(y + k), v_m));
}
}
#endif
@@ -1063,10 +1077,10 @@ void polarToCart( InputArray src1, InputArray src2,
x[k] = buf[0][k]*m; y[k] = buf[1][k]*m;
}
else
for( k = 0; k < len; k++ )
{
x[k] = buf[0][k]; y[k] = buf[1][k];
}
{
std::memcpy(x, buf[0], sizeof(float) * len);
std::memcpy(y, buf[1], sizeof(float) * len);
}
}
if( ptrs[0] )

6
modules/core/src/stat.cpp
View File

@@ -397,6 +397,8 @@ static int countNonZero_(const T* src, int len )
return nz;
}
#if CV_SSE2
static const uchar * initPopcountTable()
{
static uchar tab[256];
@@ -425,6 +427,8 @@ static const uchar * initPopcountTable()
return tab;
}
#endif
static int countNonZero8u( const uchar* src, int len )
{
int i=0, nz = 0;
@@ -645,7 +649,7 @@ static int countNonZero32f( const float* src, int len )
}
static int countNonZero64f( const double* src, int len )
{
{
int i = 0, nz = 0;
#if CV_SSE2
if (USE_SSE2)