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

Implementation detector and selector for IPP and OpenCL;

Browse Source 
IPP can be switched on and off on runtime;

Optional implementation collector was added (switched off by default in CMake). Gathers data of implementation used in functions and report this info through performance TS;

TS modifications for implementations control;
This commit is contained in:
Pavel Vlasov
2014-10-03 15:17:28 +04:00
parent 83ef276697
commit 45958eaabc
49 changed files with 4279 additions and 2799 deletions
Download Patch File Download Diff File Expand all files Collapse all files

610
modules/core/src/arithm.cpp
View File

@@ -562,10 +562,16 @@ static void add8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_8u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_8u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpAdd<uchar>, IF_SIMD(VAdd<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -582,10 +588,16 @@ static void add16u( const ushort* src1, size_t step1,
ushort* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_16u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_16u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<ushort, OpAdd<ushort>, IF_SIMD(VAdd<ushort>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -595,10 +607,16 @@ static void add16s( const short* src1, size_t step1,
short* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_16s_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_16s_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<short, OpAdd<short>, IF_SIMD(VAdd<short>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -615,10 +633,16 @@ static void add32f( const float* src1, size_t step1,
float* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAdd_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp32<float, OpAdd<float>, IF_SIMD(VAdd<float>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -635,10 +659,16 @@ static void sub8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_8u_C1RSfs(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz), 0))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_8u_C1RSfs(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz), 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpSub<uchar>, IF_SIMD(VSub<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -655,10 +685,16 @@ static void sub16u( const ushort* src1, size_t step1,
ushort* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_16u_C1RSfs(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz), 0))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_16u_C1RSfs(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz), 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<ushort, OpSub<ushort>, IF_SIMD(VSub<ushort>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -668,10 +704,16 @@ static void sub16s( const short* src1, size_t step1,
short* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_16s_C1RSfs(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz), 0))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_16s_C1RSfs(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz), 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<short, OpSub<short>, IF_SIMD(VSub<short>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -688,10 +730,16 @@ static void sub32f( const float* src1, size_t step1,
float* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_32f_C1R(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiSub_32f_C1R(src2, (int)step2, src1, (int)step1, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp32<float, OpSub<float>, IF_SIMD(VSub<float>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -711,22 +759,28 @@ static void max8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
uchar* s1 = (uchar*)src1;
uchar* s2 = (uchar*)src2;
uchar* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMaxEvery_8u(s1, s2, d, sz.width))
break;
s1 += step1;
s2 += step2;
d += step;
uchar* s1 = (uchar*)src1;
uchar* s2 = (uchar*)src2;
uchar* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMaxEvery_8u(s1, s2, d, sz.width))
break;
s1 += step1;
s2 += step2;
d += step;
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp<uchar, OpMax<uchar>, IF_SIMD(VMax<uchar>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -743,22 +797,28 @@ static void max16u( const ushort* src1, size_t step1,
ushort* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
ushort* s1 = (ushort*)src1;
ushort* s2 = (ushort*)src2;
ushort* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMaxEvery_16u(s1, s2, d, sz.width))
break;
s1 = (ushort*)((uchar*)s1 + step1);
s2 = (ushort*)((uchar*)s2 + step2);
d = (ushort*)((uchar*)d + step);
ushort* s1 = (ushort*)src1;
ushort* s2 = (ushort*)src2;
ushort* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMaxEvery_16u(s1, s2, d, sz.width))
break;
s1 = (ushort*)((uchar*)s1 + step1);
s2 = (ushort*)((uchar*)s2 + step2);
d = (ushort*)((uchar*)d + step);
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp<ushort, OpMax<ushort>, IF_SIMD(VMax<ushort>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -782,22 +842,28 @@ static void max32f( const float* src1, size_t step1,
float* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
float* s1 = (float*)src1;
float* s2 = (float*)src2;
float* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMaxEvery_32f(s1, s2, d, sz.width))
break;
s1 = (float*)((uchar*)s1 + step1);
s2 = (float*)((uchar*)s2 + step2);
d = (float*)((uchar*)d + step);
float* s1 = (float*)src1;
float* s2 = (float*)src2;
float* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMaxEvery_32f(s1, s2, d, sz.width))
break;
s1 = (float*)((uchar*)s1 + step1);
s2 = (float*)((uchar*)s2 + step2);
d = (float*)((uchar*)d + step);
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp32<float, OpMax<float>, IF_SIMD(VMax<float>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -807,22 +873,28 @@ static void max64f( const double* src1, size_t step1,
double* dst, size_t step, Size sz, void* )
{
#if ARITHM_USE_IPP == 1
double* s1 = (double*)src1;
double* s2 = (double*)src2;
double* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMaxEvery_64f(s1, s2, d, sz.width))
break;
s1 = (double*)((uchar*)s1 + step1);
s2 = (double*)((uchar*)s2 + step2);
d = (double*)((uchar*)d + step);
double* s1 = (double*)src1;
double* s2 = (double*)src2;
double* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMaxEvery_64f(s1, s2, d, sz.width))
break;
s1 = (double*)((uchar*)s1 + step1);
s2 = (double*)((uchar*)s2 + step2);
d = (double*)((uchar*)d + step);
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp64<double, OpMax<double>, IF_SIMD(VMax<double>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -832,22 +904,28 @@ static void min8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
uchar* s1 = (uchar*)src1;
uchar* s2 = (uchar*)src2;
uchar* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMinEvery_8u(s1, s2, d, sz.width))
break;
s1 += step1;
s2 += step2;
d += step;
uchar* s1 = (uchar*)src1;
uchar* s2 = (uchar*)src2;
uchar* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMinEvery_8u(s1, s2, d, sz.width))
break;
s1 += step1;
s2 += step2;
d += step;
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp<uchar, OpMin<uchar>, IF_SIMD(VMin<uchar>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -864,22 +942,28 @@ static void min16u( const ushort* src1, size_t step1,
ushort* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
ushort* s1 = (ushort*)src1;
ushort* s2 = (ushort*)src2;
ushort* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMinEvery_16u(s1, s2, d, sz.width))
break;
s1 = (ushort*)((uchar*)s1 + step1);
s2 = (ushort*)((uchar*)s2 + step2);
d = (ushort*)((uchar*)d + step);
ushort* s1 = (ushort*)src1;
ushort* s2 = (ushort*)src2;
ushort* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMinEvery_16u(s1, s2, d, sz.width))
break;
s1 = (ushort*)((uchar*)s1 + step1);
s2 = (ushort*)((uchar*)s2 + step2);
d = (ushort*)((uchar*)d + step);
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp<ushort, OpMin<ushort>, IF_SIMD(VMin<ushort>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -903,22 +987,28 @@ static void min32f( const float* src1, size_t step1,
float* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
float* s1 = (float*)src1;
float* s2 = (float*)src2;
float* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMinEvery_32f(s1, s2, d, sz.width))
break;
s1 = (float*)((uchar*)s1 + step1);
s2 = (float*)((uchar*)s2 + step2);
d = (float*)((uchar*)d + step);
float* s1 = (float*)src1;
float* s2 = (float*)src2;
float* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMinEvery_32f(s1, s2, d, sz.width))
break;
s1 = (float*)((uchar*)s1 + step1);
s2 = (float*)((uchar*)s2 + step2);
d = (float*)((uchar*)d + step);
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp32<float, OpMin<float>, IF_SIMD(VMin<float>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -928,22 +1018,28 @@ static void min64f( const double* src1, size_t step1,
double* dst, size_t step, Size sz, void* )
{
#if ARITHM_USE_IPP == 1
double* s1 = (double*)src1;
double* s2 = (double*)src2;
double* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
CV_IPP_CHECK()
{
if (0 > ippsMinEvery_64f(s1, s2, d, sz.width))
break;
s1 = (double*)((uchar*)s1 + step1);
s2 = (double*)((uchar*)s2 + step2);
d = (double*)((uchar*)d + step);
double* s1 = (double*)src1;
double* s2 = (double*)src2;
double* d = dst;
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
int i = 0;
for(; i < sz.height; i++)
{
if (0 > ippsMinEvery_64f(s1, s2, d, sz.width))
break;
s1 = (double*)((uchar*)s1 + step1);
s2 = (double*)((uchar*)s2 + step2);
d = (double*)((uchar*)d + step);
}
if (i == sz.height)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (i == sz.height)
return;
setIppErrorStatus();
#endif
vBinOp64<double, OpMin<double>, IF_SIMD(VMin<double>)>(src1, step1, src2, step2, dst, step, sz);
}
@@ -953,10 +1049,16 @@ static void absdiff8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAbsDiff_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAbsDiff_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpAbsDiff<uchar>, IF_SIMD(VAbsDiff<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -973,10 +1075,16 @@ static void absdiff16u( const ushort* src1, size_t step1,
ushort* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAbsDiff_16u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAbsDiff_16u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<ushort, OpAbsDiff<ushort>, IF_SIMD(VAbsDiff<ushort>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -1000,10 +1108,16 @@ static void absdiff32f( const float* src1, size_t step1,
float* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAbsDiff_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAbsDiff_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp32<float, OpAbsDiff<float>, IF_SIMD(VAbsDiff<float>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -1021,10 +1135,16 @@ static void and8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAnd_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiAnd_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpAnd<uchar>, IF_SIMD(VAnd<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -1034,10 +1154,16 @@ static void or8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiOr_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiOr_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpOr<uchar>, IF_SIMD(VOr<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -1047,10 +1173,16 @@ static void xor8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiXor_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiXor_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpXor<uchar>, IF_SIMD(VXor<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -1060,10 +1192,16 @@ static void not8u( const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz, void* )
{
#if (ARITHM_USE_IPP == 1)
fixSteps(sz, sizeof(dst[0]), step1, step2, step); (void)src2;
if (0 <= ippiNot_8u_C1R(src1, (int)step1, dst, (int)step, ippiSize(sz)))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
fixSteps(sz, sizeof(dst[0]), step1, step2, step); (void)src2;
if (0 <= ippiNot_8u_C1R(src1, (int)step1, dst, (int)step, ippiSize(sz)))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
(vBinOp<uchar, OpNot<uchar>, IF_SIMD(VNot<uchar>)>(src1, step1, src2, step2, dst, step, sz));
}
@@ -2376,11 +2514,17 @@ static void mul8u( const uchar* src1, size_t step1, const uchar* src2, size_t st
{
float fscale = (float)*(const double*)scale;
#if defined HAVE_IPP
if (std::fabs(fscale - 1) <= FLT_EPSILON)
CV_IPP_CHECK()
{
if (ippiMul_8u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0) >= 0)
return;
setIppErrorStatus();
if (std::fabs(fscale - 1) <= FLT_EPSILON)
{
if (ippiMul_8u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
mul_(src1, step1, src2, step2, dst, step, sz, fscale);
@@ -2397,11 +2541,17 @@ static void mul16u( const ushort* src1, size_t step1, const ushort* src2, size_t
{
float fscale = (float)*(const double*)scale;
#if defined HAVE_IPP
if (std::fabs(fscale - 1) <= FLT_EPSILON)
CV_IPP_CHECK()
{
if (ippiMul_16u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0) >= 0)
return;
setIppErrorStatus();
if (std::fabs(fscale - 1) <= FLT_EPSILON)
{
if (ippiMul_16u_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
mul_(src1, step1, src2, step2, dst, step, sz, fscale);
@@ -2412,11 +2562,17 @@ static void mul16s( const short* src1, size_t step1, const short* src2, size_t s
{
float fscale = (float)*(const double*)scale;
#if defined HAVE_IPP
if (std::fabs(fscale - 1) <= FLT_EPSILON)
CV_IPP_CHECK()
{
if (ippiMul_16s_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0) >= 0)
return;
setIppErrorStatus();
if (std::fabs(fscale - 1) <= FLT_EPSILON)
{
if (ippiMul_16s_C1RSfs(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz), 0) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
mul_(src1, step1, src2, step2, dst, step, sz, fscale);
@@ -2433,11 +2589,17 @@ static void mul32f( const float* src1, size_t step1, const float* src2, size_t s
{
float fscale = (float)*(const double*)scale;
#if defined HAVE_IPP
if (std::fabs(fscale - 1) <= FLT_EPSILON)
CV_IPP_CHECK()
{
if (ippiMul_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)) >= 0)
return;
setIppErrorStatus();
if (std::fabs(fscale - 1) <= FLT_EPSILON)
{
if (ippiMul_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(sz)) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
mul_(src1, step1, src2, step2, dst, step, sz, fscale);
@@ -3185,13 +3347,19 @@ static void cmp8u(const uchar* src1, size_t step1, const uchar* src2, size_t ste
uchar* dst, size_t step, Size size, void* _cmpop)
{
#if ARITHM_USE_IPP
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op >= 0 )
CV_IPP_CHECK()
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
return;
setIppErrorStatus();
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op >= 0 )
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_8u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
//vz optimized cmp_(src1, step1, src2, step2, dst, step, size, *(int*)_cmpop);
@@ -3284,13 +3452,19 @@ static void cmp16u(const ushort* src1, size_t step1, const ushort* src2, size_t
uchar* dst, size_t step, Size size, void* _cmpop)
{
#if ARITHM_USE_IPP
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op >= 0 )
CV_IPP_CHECK()
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_16u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
return;
setIppErrorStatus();
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op >= 0 )
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_16u_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
cmp_(src1, step1, src2, step2, dst, step, size, *(int*)_cmpop);
@@ -3300,13 +3474,19 @@ static void cmp16s(const short* src1, size_t step1, const short* src2, size_t st
uchar* dst, size_t step, Size size, void* _cmpop)
{
#if ARITHM_USE_IPP
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op > 0 )
CV_IPP_CHECK()
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_16s_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
return;
setIppErrorStatus();
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op > 0 )
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_16s_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
//vz optimized cmp_(src1, step1, src2, step2, dst, step, size, *(int*)_cmpop);
@@ -3438,13 +3618,19 @@ static void cmp32f(const float* src1, size_t step1, const float* src2, size_t st
uchar* dst, size_t step, Size size, void* _cmpop)
{
#if ARITHM_USE_IPP
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op >= 0 )
CV_IPP_CHECK()
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
return;
setIppErrorStatus();
IppCmpOp op = convert_cmp(*(int *)_cmpop);
if( op >= 0 )
{
fixSteps(size, sizeof(dst[0]), step1, step2, step);
if (0 <= ippiCompare_32f_C1R(src1, (int)step1, src2, (int)step2, dst, (int)step, ippiSize(size), op))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
cmp_(src1, step1, src2, step2, dst, step, size, *(int*)_cmpop);

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

@@ -2989,11 +2989,17 @@ dtype* dst, size_t dstep, Size size, double* scale) \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
dtype* dst, size_t dstep, Size size, double*) \
{ \
if (src && dst)\
CV_IPP_CHECK()\
{\
if (ippiConvert_##ippFavor(src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height)) >= 0) \
return; \
setIppErrorStatus(); \
if (src && dst)\
{\
if (ippiConvert_##ippFavor(src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height)) >= 0) \
{\
CV_IMPL_ADD(CV_IMPL_IPP)\
return; \
}\
setIppErrorStatus(); \
}\
}\
cvt_(src, sstep, dst, dstep, size); \
}
@@ -3002,11 +3008,17 @@ static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
static void cvt##suffix( const stype* src, size_t sstep, const uchar*, size_t, \
dtype* dst, size_t dstep, Size size, double*) \
{ \
if (src && dst)\
CV_IPP_CHECK()\
{\
if (ippiConvert_##ippFavor(src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height), ippRndFinancial, 0) >= 0) \
return; \
setIppErrorStatus(); \
if (src && dst)\
{\
if (ippiConvert_##ippFavor(src, (int)sstep, dst, (int)dstep, ippiSize(size.width, size.height), ippRndFinancial, 0) >= 0) \
{\
CV_IMPL_ADD(CV_IMPL_IPP)\
return; \
}\
setIppErrorStatus(); \
}\
}\
cvt_(src, sstep, dst, dstep, size); \
}
@@ -3524,6 +3536,7 @@ public:
setIppErrorStatus();
*ok = false;
}
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
}
private:
IppLUTParallelBody_LUTC1(const IppLUTParallelBody_LUTC1&);
@@ -3567,6 +3580,7 @@ public:
setIppErrorStatus();
return;
}
CV_IMPL_ADD(CV_IMPL_IPP);
}
else if (lutcn == 4)
{
@@ -3576,6 +3590,7 @@ public:
setIppErrorStatus();
return;
}
CV_IMPL_ADD(CV_IMPL_IPP);
}
*ok = true;
@@ -3605,14 +3620,20 @@ public:
if (ippiLUTPalette_8u_C3R(
src.ptr(), (int)src.step[0], dst.ptr(), (int)dst.step[0],
ippiSize(dst.size()), lutTable, 8) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
return;
}
}
else if (lutcn == 4)
{
if (ippiLUTPalette_8u_C4R(
src.ptr(), (int)src.step[0], dst.ptr(), (int)dst.step[0],
ippiSize(dst.size()), lutTable, 8) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
return;
}
}
setIppErrorStatus();
*ok = false;
@@ -3690,19 +3711,22 @@ void cv::LUT( InputArray _src, InputArray _lut, OutputArray _dst )
bool ok = false;
Ptr<ParallelLoopBody> body;
#if defined(HAVE_IPP)
size_t elemSize1 = CV_ELEM_SIZE1(dst.depth());
CV_IPP_CHECK()
{
size_t elemSize1 = CV_ELEM_SIZE1(dst.depth());
#if 0 // there are no performance benefits (PR #2653)
if (lutcn == 1)
{
ParallelLoopBody* p = new ipp::IppLUTParallelBody_LUTC1(src, lut, dst, &ok);
body.reset(p);
}
else
if (lutcn == 1)
{
ParallelLoopBody* p = new ipp::IppLUTParallelBody_LUTC1(src, lut, dst, &ok);
body.reset(p);
}
else
#endif
if ((lutcn == 3 || lutcn == 4) && elemSize1 == 1)
{
ParallelLoopBody* p = new ipp::IppLUTParallelBody_LUTCN(src, lut, dst, &ok);
body.reset(p);
if ((lutcn == 3 || lutcn == 4) && elemSize1 == 1)
{
ParallelLoopBody* p = new ipp::IppLUTParallelBody_LUTCN(src, lut, dst, &ok);
body.reset(p);
}
}
#endif
if (body == NULL || ok == false)

534
modules/core/src/copy.cpp
View File

@@ -82,9 +82,15 @@ template<> void
copyMask_<uchar>(const uchar* _src, size_t sstep, const uchar* mask, size_t mstep, uchar* _dst, size_t dstep, Size size)
{
#if defined HAVE_IPP
if (ippiCopy_8u_C1MR(_src, (int)sstep, _dst, (int)dstep, ippiSize(size), mask, (int)mstep) >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (ippiCopy_8u_C1MR(_src, (int)sstep, _dst, (int)dstep, ippiSize(size), mask, (int)mstep) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
for( ; size.height--; mask += mstep, _src += sstep, _dst += dstep )
@@ -126,9 +132,15 @@ template<> void
copyMask_<ushort>(const uchar* _src, size_t sstep, const uchar* mask, size_t mstep, uchar* _dst, size_t dstep, Size size)
{
#if defined HAVE_IPP
if (ippiCopy_16u_C1MR((const Ipp16u *)_src, (int)sstep, (Ipp16u *)_dst, (int)dstep, ippiSize(size), mask, (int)mstep) >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (ippiCopy_16u_C1MR((const Ipp16u *)_src, (int)sstep, (Ipp16u *)_dst, (int)dstep, ippiSize(size), mask, (int)mstep) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
for( ; size.height--; mask += mstep, _src += sstep, _dst += dstep )
@@ -201,9 +213,15 @@ static void copyMask##suffix(const uchar* src, size_t sstep, const uchar* mask,
static void copyMask##suffix(const uchar* src, size_t sstep, const uchar* mask, size_t mstep, \
uchar* dst, size_t dstep, Size size, void*) \
{ \
if (ippiCopy_##ippfavor((const ipptype *)src, (int)sstep, (ipptype *)dst, (int)dstep, ippiSize(size), (const Ipp8u *)mask, (int)mstep) >= 0) \
return; \
setIppErrorStatus(); \
CV_IPP_CHECK()\
{\
if (ippiCopy_##ippfavor((const ipptype *)src, (int)sstep, (ipptype *)dst, (int)dstep, ippiSize(size), (const Ipp8u *)mask, (int)mstep) >= 0) \
{\
CV_IMPL_ADD(CV_IMPL_IPP);\
return;\
}\
setIppErrorStatus(); \
}\
copyMask_<type>(src, sstep, mask, mstep, dst, dstep, size); \
}
#else
@@ -301,9 +319,15 @@ void Mat::copyTo( OutputArray _dst ) const
size_t len = sz.width*elemSize();
#if defined HAVE_IPP
if (ippiCopy_8u_C1R(sptr, (int)step, dptr, (int)dst.step, ippiSize((int)len, sz.height)) >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (ippiCopy_8u_C1R(sptr, (int)step, dptr, (int)dst.step, ippiSize((int)len, sz.height)) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP)
return;
}
setIppErrorStatus();
}
#endif
for( ; sz.height--; sptr += step, dptr += dst.step )
@@ -380,23 +404,32 @@ Mat& Mat::operator = (const Scalar& s)
if( is[0] == 0 && is[1] == 0 && is[2] == 0 && is[3] == 0 )
{
#if defined HAVE_IPP && !defined HAVE_IPP_ICV_ONLY && 0
if (dims <= 2 || isContinuous())
CV_IPP_CHECK()
{
IppiSize roisize = { cols, rows };
if (isContinuous())
if (dims <= 2 || isContinuous())
{
roisize.width = (int)total();
roisize.height = 1;
IppiSize roisize = { cols, rows };
if (isContinuous())
{
roisize.width = (int)total();
roisize.height = 1;
if (ippsZero_8u(data, static_cast<int>(roisize.width * elemSize())) >= 0)
if (ippsZero_8u(data, static_cast<int>(roisize.width * elemSize())) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP)
return *this;
}
setIppErrorStatus();
}
roisize.width *= (int)elemSize();
if (ippiSet_8u_C1R(0, data, (int)step, roisize) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP)
return *this;
}
setIppErrorStatus();
}
roisize.width *= (int)elemSize();
if (ippiSet_8u_C1R(0, data, (int)step, roisize) >= 0)
return *this;
setIppErrorStatus();
}
#endif
@@ -439,77 +472,83 @@ Mat& Mat::setTo(InputArray _value, InputArray _mask)
CV_Assert( mask.empty() || (mask.type() == CV_8U && size == mask.size) );
#if defined HAVE_IPP
int cn = channels(), depth0 = depth();
if (!mask.empty() && (dims <= 2 || (isContinuous() && mask.isContinuous())) &&
(/*depth0 == CV_8U ||*/ depth0 == CV_16U || depth0 == CV_16S || depth0 == CV_32S || depth0 == CV_32F) &&
(cn == 1 || cn == 3 || cn == 4))
CV_IPP_CHECK()
{
uchar _buf[32];
void * buf = _buf;
convertAndUnrollScalar( value, type(), _buf, 1 );
int cn = channels(), depth0 = depth();
IppStatus status = (IppStatus)-1;
IppiSize roisize = { cols, rows };
int mstep = (int)mask.step[0], dstep = (int)step[0];
if (!mask.empty() && (dims <= 2 || (isContinuous() && mask.isContinuous())) &&
(/*depth0 == CV_8U ||*/ depth0 == CV_16U || depth0 == CV_16S || depth0 == CV_32S || depth0 == CV_32F) &&
(cn == 1 || cn == 3 || cn == 4))
{
uchar _buf[32];
void * buf = _buf;
convertAndUnrollScalar( value, type(), _buf, 1 );
if (isContinuous() && mask.isContinuous())
{
roisize.width = (int)total();
roisize.height = 1;
}
IppStatus status = (IppStatus)-1;
IppiSize roisize = { cols, rows };
int mstep = (int)mask.step[0], dstep = (int)step[0];
if (cn == 1)
{
/*if (depth0 == CV_8U)
status = ippiSet_8u_C1MR(*(Ipp8u *)buf, (Ipp8u *)data, dstep, roisize, mask.data, mstep);
else*/ if (depth0 == CV_16U)
status = ippiSet_16u_C1MR(*(Ipp16u *)buf, (Ipp16u *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_16S)
status = ippiSet_16s_C1MR(*(Ipp16s *)buf, (Ipp16s *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_32S)
status = ippiSet_32s_C1MR(*(Ipp32s *)buf, (Ipp32s *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_32F)
status = ippiSet_32f_C1MR(*(Ipp32f *)buf, (Ipp32f *)data, dstep, roisize, mask.data, mstep);
}
else if (cn == 3 || cn == 4)
{
if (isContinuous() && mask.isContinuous())
{
roisize.width = (int)total();
roisize.height = 1;
}
if (cn == 1)
{
/*if (depth0 == CV_8U)
status = ippiSet_8u_C1MR(*(Ipp8u *)buf, (Ipp8u *)data, dstep, roisize, mask.data, mstep);
else*/ if (depth0 == CV_16U)
status = ippiSet_16u_C1MR(*(Ipp16u *)buf, (Ipp16u *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_16S)
status = ippiSet_16s_C1MR(*(Ipp16s *)buf, (Ipp16s *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_32S)
status = ippiSet_32s_C1MR(*(Ipp32s *)buf, (Ipp32s *)data, dstep, roisize, mask.data, mstep);
else if (depth0 == CV_32F)
status = ippiSet_32f_C1MR(*(Ipp32f *)buf, (Ipp32f *)data, dstep, roisize, mask.data, mstep);
}
else if (cn == 3 || cn == 4)
{
#define IPP_SET(ippfavor, ippcn) \
do \
{ \
typedef Ipp##ippfavor ipptype; \
ipptype ippvalue[4] = { ((ipptype *)buf)[0], ((ipptype *)buf)[1], ((ipptype *)buf)[2], ((ipptype *)buf)[3] }; \
status = ippiSet_##ippfavor##_C##ippcn##MR(ippvalue, (ipptype *)data, dstep, roisize, mask.data, mstep); \
} while ((void)0, 0)
do \
{ \
typedef Ipp##ippfavor ipptype; \
ipptype ippvalue[4] = { ((ipptype *)buf)[0], ((ipptype *)buf)[1], ((ipptype *)buf)[2], ((ipptype *)buf)[3] }; \
status = ippiSet_##ippfavor##_C##ippcn##MR(ippvalue, (ipptype *)data, dstep, roisize, mask.data, mstep); \
} while ((void)0, 0)
#define IPP_SET_CN(ippcn) \
do \
{ \
if (cn == ippcn) \
do \
{ \
/*if (depth0 == CV_8U) \
IPP_SET(8u, ippcn); \
else*/ if (depth0 == CV_16U) \
IPP_SET(16u, ippcn); \
else if (depth0 == CV_16S) \
IPP_SET(16s, ippcn); \
else if (depth0 == CV_32S) \
IPP_SET(32s, ippcn); \
else if (depth0 == CV_32F) \
IPP_SET(32f, ippcn); \
} \
} while ((void)0, 0)
if (cn == ippcn) \
{ \
/*if (depth0 == CV_8U) \
IPP_SET(8u, ippcn); \
else*/ if (depth0 == CV_16U) \
IPP_SET(16u, ippcn); \
else if (depth0 == CV_16S) \
IPP_SET(16s, ippcn); \
else if (depth0 == CV_32S) \
IPP_SET(32s, ippcn); \
else if (depth0 == CV_32F) \
IPP_SET(32f, ippcn); \
} \
} while ((void)0, 0)
IPP_SET_CN(3);
IPP_SET_CN(4);
IPP_SET_CN(3);
IPP_SET_CN(4);
#undef IPP_SET_CN
#undef IPP_SET
}
}
if (status >= 0)
return *this;
setIppErrorStatus();
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return *this;
}
setIppErrorStatus();
}
}
#endif
@@ -714,66 +753,75 @@ void flip( InputArray _src, OutputArray _dst, int flip_mode )
size_t esz = CV_ELEM_SIZE(type);
#if defined HAVE_IPP
typedef IppStatus (CV_STDCALL * ippiMirror)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize, IppiAxis flip);
typedef IppStatus (CV_STDCALL * ippiMirrorI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize, IppiAxis flip);
ippiMirror ippFunc = 0;
ippiMirrorI ippFuncI = 0;
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippiMirror)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize, IppiAxis flip);
typedef IppStatus (CV_STDCALL * ippiMirrorI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize, IppiAxis flip);
ippiMirror ippFunc = 0;
ippiMirrorI ippFuncI = 0;
if (src.data == dst.data)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_8UC1 ? (ippiMirrorI)ippiMirror_8u_C1IR :
type == CV_8UC3 ? (ippiMirrorI)ippiMirror_8u_C3IR :
type == CV_8UC4 ? (ippiMirrorI)ippiMirror_8u_C4IR :
type == CV_16UC1 ? (ippiMirrorI)ippiMirror_16u_C1IR :
type == CV_16UC3 ? (ippiMirrorI)ippiMirror_16u_C3IR :
type == CV_16UC4 ? (ippiMirrorI)ippiMirror_16u_C4IR :
type == CV_16SC1 ? (ippiMirrorI)ippiMirror_16s_C1IR :
type == CV_16SC3 ? (ippiMirrorI)ippiMirror_16s_C3IR :
type == CV_16SC4 ? (ippiMirrorI)ippiMirror_16s_C4IR :
type == CV_32SC1 ? (ippiMirrorI)ippiMirror_32s_C1IR :
type == CV_32SC3 ? (ippiMirrorI)ippiMirror_32s_C3IR :
type == CV_32SC4 ? (ippiMirrorI)ippiMirror_32s_C4IR :
type == CV_32FC1 ? (ippiMirrorI)ippiMirror_32f_C1IR :
type == CV_32FC3 ? (ippiMirrorI)ippiMirror_32f_C3IR :
type == CV_32FC4 ? (ippiMirrorI)ippiMirror_32f_C4IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_8UC1 ? (ippiMirror)ippiMirror_8u_C1R :
type == CV_8UC3 ? (ippiMirror)ippiMirror_8u_C3R :
type == CV_8UC4 ? (ippiMirror)ippiMirror_8u_C4R :
type == CV_16UC1 ? (ippiMirror)ippiMirror_16u_C1R :
type == CV_16UC3 ? (ippiMirror)ippiMirror_16u_C3R :
type == CV_16UC4 ? (ippiMirror)ippiMirror_16u_C4R :
type == CV_16SC1 ? (ippiMirror)ippiMirror_16s_C1R :
type == CV_16SC3 ? (ippiMirror)ippiMirror_16s_C3R :
type == CV_16SC4 ? (ippiMirror)ippiMirror_16s_C4R :
type == CV_32SC1 ? (ippiMirror)ippiMirror_32s_C1R :
type == CV_32SC3 ? (ippiMirror)ippiMirror_32s_C3R :
type == CV_32SC4 ? (ippiMirror)ippiMirror_32s_C4R :
type == CV_32FC1 ? (ippiMirror)ippiMirror_32f_C1R :
type == CV_32FC3 ? (ippiMirror)ippiMirror_32f_C3R :
type == CV_32FC4 ? (ippiMirror)ippiMirror_32f_C4R : 0;
}
IppiAxis axis = flip_mode == 0 ? ippAxsHorizontal :
flip_mode > 0 ? ippAxsVertical : ippAxsBoth;
IppiSize roisize = { dst.cols, dst.rows };
if (src.data == dst.data)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_8UC1 ? (ippiMirrorI)ippiMirror_8u_C1IR :
type == CV_8UC3 ? (ippiMirrorI)ippiMirror_8u_C3IR :
type == CV_8UC4 ? (ippiMirrorI)ippiMirror_8u_C4IR :
type == CV_16UC1 ? (ippiMirrorI)ippiMirror_16u_C1IR :
type == CV_16UC3 ? (ippiMirrorI)ippiMirror_16u_C3IR :
type == CV_16UC4 ? (ippiMirrorI)ippiMirror_16u_C4IR :
type == CV_16SC1 ? (ippiMirrorI)ippiMirror_16s_C1IR :
type == CV_16SC3 ? (ippiMirrorI)ippiMirror_16s_C3IR :
type == CV_16SC4 ? (ippiMirrorI)ippiMirror_16s_C4IR :
type == CV_32SC1 ? (ippiMirrorI)ippiMirror_32s_C1IR :
type == CV_32SC3 ? (ippiMirrorI)ippiMirror_32s_C3IR :
type == CV_32SC4 ? (ippiMirrorI)ippiMirror_32s_C4IR :
type == CV_32FC1 ? (ippiMirrorI)ippiMirror_32f_C1IR :
type == CV_32FC3 ? (ippiMirrorI)ippiMirror_32f_C3IR :
type == CV_32FC4 ? (ippiMirrorI)ippiMirror_32f_C4IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_8UC1 ? (ippiMirror)ippiMirror_8u_C1R :
type == CV_8UC3 ? (ippiMirror)ippiMirror_8u_C3R :
type == CV_8UC4 ? (ippiMirror)ippiMirror_8u_C4R :
type == CV_16UC1 ? (ippiMirror)ippiMirror_16u_C1R :
type == CV_16UC3 ? (ippiMirror)ippiMirror_16u_C3R :
type == CV_16UC4 ? (ippiMirror)ippiMirror_16u_C4R :
type == CV_16SC1 ? (ippiMirror)ippiMirror_16s_C1R :
type == CV_16SC3 ? (ippiMirror)ippiMirror_16s_C3R :
type == CV_16SC4 ? (ippiMirror)ippiMirror_16s_C4R :
type == CV_32SC1 ? (ippiMirror)ippiMirror_32s_C1R :
type == CV_32SC3 ? (ippiMirror)ippiMirror_32s_C3R :
type == CV_32SC4 ? (ippiMirror)ippiMirror_32s_C4R :
type == CV_32FC1 ? (ippiMirror)ippiMirror_32f_C1R :
type == CV_32FC3 ? (ippiMirror)ippiMirror_32f_C3R :
type == CV_32FC4 ? (ippiMirror)ippiMirror_32f_C4R : 0;
}
IppiAxis axis = flip_mode == 0 ? ippAxsHorizontal :
flip_mode > 0 ? ippAxsVertical : ippAxsBoth;
IppiSize roisize = { dst.cols, dst.rows };
if (ippFunc != 0)
{
if (ippFunc(src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, ippiSize(src.cols, src.rows), axis) >= 0)
return;
setIppErrorStatus();
}
else if (ippFuncI != 0)
{
if (ippFuncI(dst.ptr(), (int)dst.step, roisize, axis) >= 0)
return;
setIppErrorStatus();
if (ippFunc != 0)
{
if (ippFunc(src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, ippiSize(src.cols, src.rows), axis) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
else if (ippFuncI != 0)
{
if (ippFuncI(dst.ptr(), (int)dst.step, roisize, axis) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
@@ -1135,111 +1183,117 @@ void cv::copyMakeBorder( InputArray _src, OutputArray _dst, int top, int bottom,
borderType &= ~BORDER_ISOLATED;
#if defined HAVE_IPP && 0
typedef IppStatus (CV_STDCALL * ippiCopyMakeBorder)(const void * pSrc, int srcStep, IppiSize srcRoiSize, void * pDst,
int dstStep, IppiSize dstRoiSize, int topBorderHeight, int leftBorderWidth);
typedef IppStatus (CV_STDCALL * ippiCopyMakeBorderI)(const void * pSrc, int srcDstStep, IppiSize srcRoiSize, IppiSize dstRoiSize,
int topBorderHeight, int leftborderwidth);
typedef IppStatus (CV_STDCALL * ippiCopyConstBorder)(const void * pSrc, int srcStep, IppiSize srcRoiSize, void * pDst, int dstStep,
IppiSize dstRoiSize, int topBorderHeight, int leftBorderWidth, void * value);
IppiSize srcRoiSize = { src.cols, src.rows }, dstRoiSize = { dst.cols, dst.rows };
ippiCopyMakeBorder ippFunc = 0;
ippiCopyMakeBorderI ippFuncI = 0;
ippiCopyConstBorder ippFuncConst = 0;
bool inplace = dst.datastart == src.datastart;
if (borderType == BORDER_CONSTANT)
CV_IPP_CHECK()
{
ippFuncConst =
// type == CV_8UC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_8u_C1R : bug in IPP 8.1
type == CV_16UC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_16u_C1R :
// type == CV_16SC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_16s_C1R : bug in IPP 8.1
// type == CV_32SC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_32s_C1R : bug in IPP 8.1
// type == CV_32FC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_32f_C1R : bug in IPP 8.1
type == CV_8UC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_8u_C3R :
type == CV_16UC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_16u_C3R :
type == CV_16SC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_16s_C3R :
type == CV_32SC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_32s_C3R :
type == CV_32FC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_32f_C3R :
type == CV_8UC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_8u_C4R :
type == CV_16UC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_16u_C4R :
type == CV_16SC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_16s_C4R :
type == CV_32SC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_32s_C4R :
type == CV_32FC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_32f_C4R : 0;
}
else if (borderType == BORDER_WRAP)
{
if (inplace)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_32SC1 ? (ippiCopyMakeBorderI)ippiCopyWrapBorder_32s_C1IR :
type == CV_32FC1 ? (ippiCopyMakeBorderI)ippiCopyWrapBorder_32s_C1IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_32SC1 ? (ippiCopyMakeBorder)ippiCopyWrapBorder_32s_C1R :
type == CV_32FC1 ? (ippiCopyMakeBorder)ippiCopyWrapBorder_32s_C1R : 0;
}
}
else if (borderType == BORDER_REPLICATE)
{
if (inplace)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_8UC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_8u_C1IR :
type == CV_16UC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16u_C1IR :
type == CV_16SC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16s_C1IR :
type == CV_32SC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32s_C1IR :
type == CV_32FC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32f_C1IR :
type == CV_8UC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_8u_C3IR :
type == CV_16UC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16u_C3IR :
type == CV_16SC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16s_C3IR :
type == CV_32SC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32s_C3IR :
type == CV_32FC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32f_C3IR :
type == CV_8UC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_8u_C4IR :
type == CV_16UC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16u_C4IR :
type == CV_16SC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16s_C4IR :
type == CV_32SC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32s_C4IR :
type == CV_32FC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32f_C4IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_8UC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_8u_C1R :
type == CV_16UC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16u_C1R :
type == CV_16SC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16s_C1R :
type == CV_32SC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32s_C1R :
type == CV_32FC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32f_C1R :
type == CV_8UC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_8u_C3R :
type == CV_16UC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16u_C3R :
type == CV_16SC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16s_C3R :
type == CV_32SC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32s_C3R :
type == CV_32FC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32f_C3R :
type == CV_8UC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_8u_C4R :
type == CV_16UC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16u_C4R :
type == CV_16SC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16s_C4R :
type == CV_32SC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32s_C4R :
type == CV_32FC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32f_C4R : 0;
}
}
typedef IppStatus (CV_STDCALL * ippiCopyMakeBorder)(const void * pSrc, int srcStep, IppiSize srcRoiSize, void * pDst,
int dstStep, IppiSize dstRoiSize, int topBorderHeight, int leftBorderWidth);
typedef IppStatus (CV_STDCALL * ippiCopyMakeBorderI)(const void * pSrc, int srcDstStep, IppiSize srcRoiSize, IppiSize dstRoiSize,
int topBorderHeight, int leftborderwidth);
typedef IppStatus (CV_STDCALL * ippiCopyConstBorder)(const void * pSrc, int srcStep, IppiSize srcRoiSize, void * pDst, int dstStep,
IppiSize dstRoiSize, int topBorderHeight, int leftBorderWidth, void * value);
if (ippFunc || ippFuncI || ippFuncConst)
{
uchar scbuf[32];
scalarToRawData(value, scbuf, type);
IppiSize srcRoiSize = { src.cols, src.rows }, dstRoiSize = { dst.cols, dst.rows };
ippiCopyMakeBorder ippFunc = 0;
ippiCopyMakeBorderI ippFuncI = 0;
ippiCopyConstBorder ippFuncConst = 0;
bool inplace = dst.datastart == src.datastart;
if ( (ippFunc && ippFunc(src.data, (int)src.step, srcRoiSize, dst.data, (int)dst.step, dstRoiSize, top, left) >= 0) ||
(ippFuncI && ippFuncI(src.data, (int)src.step, srcRoiSize, dstRoiSize, top, left) >= 0) ||
(ippFuncConst && ippFuncConst(src.data, (int)src.step, srcRoiSize, dst.data, (int)dst.step,
dstRoiSize, top, left, scbuf) >= 0))
return;
if (borderType == BORDER_CONSTANT)
{
ippFuncConst =
// type == CV_8UC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_8u_C1R : bug in IPP 8.1
type == CV_16UC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_16u_C1R :
// type == CV_16SC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_16s_C1R : bug in IPP 8.1
// type == CV_32SC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_32s_C1R : bug in IPP 8.1
// type == CV_32FC1 ? (ippiCopyConstBorder)ippiCopyConstBorder_32f_C1R : bug in IPP 8.1
type == CV_8UC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_8u_C3R :
type == CV_16UC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_16u_C3R :
type == CV_16SC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_16s_C3R :
type == CV_32SC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_32s_C3R :
type == CV_32FC3 ? (ippiCopyConstBorder)ippiCopyConstBorder_32f_C3R :
type == CV_8UC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_8u_C4R :
type == CV_16UC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_16u_C4R :
type == CV_16SC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_16s_C4R :
type == CV_32SC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_32s_C4R :
type == CV_32FC4 ? (ippiCopyConstBorder)ippiCopyConstBorder_32f_C4R : 0;
}
else if (borderType == BORDER_WRAP)
{
if (inplace)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_32SC1 ? (ippiCopyMakeBorderI)ippiCopyWrapBorder_32s_C1IR :
type == CV_32FC1 ? (ippiCopyMakeBorderI)ippiCopyWrapBorder_32s_C1IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_32SC1 ? (ippiCopyMakeBorder)ippiCopyWrapBorder_32s_C1R :
type == CV_32FC1 ? (ippiCopyMakeBorder)ippiCopyWrapBorder_32s_C1R : 0;
}
}
else if (borderType == BORDER_REPLICATE)
{
if (inplace)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_8UC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_8u_C1IR :
type == CV_16UC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16u_C1IR :
type == CV_16SC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16s_C1IR :
type == CV_32SC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32s_C1IR :
type == CV_32FC1 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32f_C1IR :
type == CV_8UC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_8u_C3IR :
type == CV_16UC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16u_C3IR :
type == CV_16SC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16s_C3IR :
type == CV_32SC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32s_C3IR :
type == CV_32FC3 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32f_C3IR :
type == CV_8UC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_8u_C4IR :
type == CV_16UC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16u_C4IR :
type == CV_16SC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_16s_C4IR :
type == CV_32SC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32s_C4IR :
type == CV_32FC4 ? (ippiCopyMakeBorderI)ippiCopyReplicateBorder_32f_C4IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_8UC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_8u_C1R :
type == CV_16UC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16u_C1R :
type == CV_16SC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16s_C1R :
type == CV_32SC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32s_C1R :
type == CV_32FC1 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32f_C1R :
type == CV_8UC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_8u_C3R :
type == CV_16UC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16u_C3R :
type == CV_16SC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16s_C3R :
type == CV_32SC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32s_C3R :
type == CV_32FC3 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32f_C3R :
type == CV_8UC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_8u_C4R :
type == CV_16UC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16u_C4R :
type == CV_16SC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_16s_C4R :
type == CV_32SC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32s_C4R :
type == CV_32FC4 ? (ippiCopyMakeBorder)ippiCopyReplicateBorder_32f_C4R : 0;
}
}
setIppErrorStatus();
if (ippFunc || ippFuncI || ippFuncConst)
{
uchar scbuf[32];
scalarToRawData(value, scbuf, type);
if ( (ippFunc && ippFunc(src.data, (int)src.step, srcRoiSize, dst.data, (int)dst.step, dstRoiSize, top, left) >= 0) ||
(ippFuncI && ippFuncI(src.data, (int)src.step, srcRoiSize, dstRoiSize, top, left) >= 0) ||
(ippFuncConst && ippFuncConst(src.data, (int)src.step, srcRoiSize, dst.data, (int)dst.step,
dstRoiSize, top, left, scbuf) >= 0))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif

122
modules/core/src/dxt.cpp
View File

@@ -554,12 +554,18 @@ DFT( const Complex<T>* src, Complex<T>* dst, int n,
if( !inv )
{
if (ippsDFTFwd_CToC( src, dst, spec, (uchar*)buf ) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
}
else
{
if (ippsDFTInv_CToC( src, dst, spec, (uchar*)buf ) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
}
setIppErrorStatus();
}
@@ -997,6 +1003,7 @@ RealDFT( const T* src, T* dst, int n, int nf, int* factors, const int* itab,
if( (n & 1) == 0 )
dst[n] = 0;
}
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
@@ -1124,7 +1131,12 @@ CCSIDFT( const T* src, T* dst, int n, int nf, int* factors, const int* itab,
if( spec )
{
if (ippsDFTInv_PackToR( src, dst, spec, (uchar*)buf ) >=0)
goto finalize;
{
if( complex_input )
((T*)src)[0] = (T)save_s1;
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
@@ -1249,10 +1261,6 @@ CCSIDFT( const T* src, T* dst, int n, int nf, int* factors, const int* itab,
dst[j+1] = t1;
}
}
#ifdef USE_IPP_DFT
finalize:
#endif
if( complex_input )
((T*)src)[0] = (T)save_s1;
}
@@ -1556,6 +1564,7 @@ public:
ippFree( pBuffer );
ippFree( pDFTSpec );
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
}
private:
@@ -1629,6 +1638,7 @@ public:
ippFree( pBuffer );
ippFree( pDFTSpec );
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
}
private:
@@ -1727,8 +1737,13 @@ static bool ippi_DFT_C_32F(const Mat& src, Mat& dst, bool inv, int norm_flag)
ippFree( pDFTSpec );
return status >= 0;
if(status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return true;
}
return false;
}
static bool ippi_DFT_R_32F(const Mat& src, Mat& dst, bool inv, int norm_flag)
{
@@ -1776,7 +1791,12 @@ static bool ippi_DFT_R_32F(const Mat& src, Mat& dst, bool inv, int norm_flag)
ippFree( pDFTSpec );
return status >= 0;
if(status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return true;
}
return false;
}
#endif
@@ -2455,39 +2475,53 @@ void cv::dft( InputArray _src0, OutputArray _dst, int flags, int nonzero_rows )
Mat dst = _dst.getMat();
#if defined USE_IPP_DFT
if ((src.depth() == CV_32F) && (src.total()>(int)(1<<6)) && nonzero_rows == 0)
CV_IPP_CHECK()
{
if ((flags & DFT_ROWS) == 0)
if ((src.depth() == CV_32F) && (src.total()>(int)(1<<6)) && nonzero_rows == 0)
{
if (src.channels() == 2 && !(inv && (flags & DFT_REAL_OUTPUT)))
if ((flags & DFT_ROWS) == 0)
{
if (ippi_DFT_C_32F(src, dst, inv, ipp_norm_flag))
return;
setIppErrorStatus();
if (src.channels() == 2 && !(inv && (flags & DFT_REAL_OUTPUT)))
{
if (ippi_DFT_C_32F(src, dst, inv, ipp_norm_flag))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
if (src.channels() == 1 && (inv || !(flags & DFT_COMPLEX_OUTPUT)))
{
if (ippi_DFT_R_32F(src, dst, inv, ipp_norm_flag))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
if (src.channels() == 1 && (inv || !(flags & DFT_COMPLEX_OUTPUT)))
else
{
if (ippi_DFT_R_32F(src, dst, inv, ipp_norm_flag))
return;
setIppErrorStatus();
}
}
else
{
if (src.channels() == 2 && !(inv && (flags & DFT_REAL_OUTPUT)))
{
ippiDFT_C_Func ippiFunc = inv ? (ippiDFT_C_Func)ippiDFTInv_CToC_32fc_C1R : (ippiDFT_C_Func)ippiDFTFwd_CToC_32fc_C1R;
if (Dft_C_IPPLoop(src, dst, IPPDFT_C_Functor(ippiFunc),ipp_norm_flag))
return;
setIppErrorStatus();
}
if (src.channels() == 1 && (inv || !(flags & DFT_COMPLEX_OUTPUT)))
{
ippiDFT_R_Func ippiFunc = inv ? (ippiDFT_R_Func)ippiDFTInv_PackToR_32f_C1R : (ippiDFT_R_Func)ippiDFTFwd_RToPack_32f_C1R;
if (Dft_R_IPPLoop(src, dst, IPPDFT_R_Functor(ippiFunc),ipp_norm_flag))
return;
setIppErrorStatus();
if (src.channels() == 2 && !(inv && (flags & DFT_REAL_OUTPUT)))
{
ippiDFT_C_Func ippiFunc = inv ? (ippiDFT_C_Func)ippiDFTInv_CToC_32fc_C1R : (ippiDFT_C_Func)ippiDFTFwd_CToC_32fc_C1R;
if (Dft_C_IPPLoop(src, dst, IPPDFT_C_Functor(ippiFunc),ipp_norm_flag))
{
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
return;
}
setIppErrorStatus();
}
if (src.channels() == 1 && (inv || !(flags & DFT_COMPLEX_OUTPUT)))
{
ippiDFT_R_Func ippiFunc = inv ? (ippiDFT_R_Func)ippiDFTInv_PackToR_32f_C1R : (ippiDFT_R_Func)ippiDFTFwd_RToPack_32f_C1R;
if (Dft_R_IPPLoop(src, dst, IPPDFT_R_Functor(ippiFunc),ipp_norm_flag))
{
CV_IMPL_ADD(CV_IMPL_IPP|CV_IMPL_MT);
return;
}
setIppErrorStatus();
}
}
}
}
@@ -2538,7 +2572,7 @@ void cv::dft( InputArray _src0, OutputArray _dst, int flags, int nonzero_rows )
spec = 0;
#ifdef USE_IPP_DFT
if( len*count >= 64 ) // use IPP DFT if available
if( CV_IPP_CHECK_COND && (len*count >= 64) ) // use IPP DFT if available
{
int specsize=0, initsize=0, worksize=0;
IppDFTGetSizeFunc getSizeFunc = 0;
@@ -3432,12 +3466,18 @@ void cv::dct( InputArray _src0, OutputArray _dst, int flags )
Mat dst = _dst.getMat();
#if defined (HAVE_IPP) && (IPP_VERSION_MAJOR >= 7)
bool row = (flags & DCT_ROWS) != 0;
if (src.type() == CV_32F)
CV_IPP_CHECK()
{
if(ippi_DCT_32f(src,dst,inv, row))
return;
setIppErrorStatus();
bool row = (flags & DCT_ROWS) != 0;
if (src.type() == CV_32F)
{
if(ippi_DCT_32f(src,dst,inv, row))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif

204
modules/core/src/mathfuncs.cpp
View File

@@ -240,10 +240,16 @@ float cubeRoot( float value )
static void Magnitude_32f(const float* x, const float* y, float* mag, int len)
{
#if defined HAVE_IPP && 0
IppStatus status = ippsMagnitude_32f(x, y, mag, len);
if (status >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
IppStatus status = ippsMagnitude_32f(x, y, mag, len);
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
int i = 0;
@@ -286,10 +292,16 @@ static void Magnitude_32f(const float* x, const float* y, float* mag, int len)
static void Magnitude_64f(const double* x, const double* y, double* mag, int len)
{
#if defined(HAVE_IPP)
IppStatus status = ippsMagnitude_64f(x, y, mag, len);
if (status >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
IppStatus status = ippsMagnitude_64f(x, y, mag, len);
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
int i = 0;
@@ -320,9 +332,15 @@ static void Magnitude_64f(const double* x, const double* y, double* mag, int len
static void InvSqrt_32f(const float* src, float* dst, int len)
{
#if defined(HAVE_IPP)
if (ippsInvSqrt_32f_A21(src, dst, len) >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (ippsInvSqrt_32f_A21(src, dst, len) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
int i = 0;
@@ -380,9 +398,15 @@ static void InvSqrt_64f(const double* src, double* dst, int len)
static void Sqrt_32f(const float* src, float* dst, int len)
{
#if defined(HAVE_IPP)
if (ippsSqrt_32f_A21(src, dst, len) >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (ippsSqrt_32f_A21(src, dst, len) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
int i = 0;
@@ -414,9 +438,15 @@ static void Sqrt_32f(const float* src, float* dst, int len)
static void Sqrt_64f(const double* src, double* dst, int len)
{
#if defined(HAVE_IPP)
if (ippsSqrt_64f_A50(src, dst, len) >= 0)
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (ippsSqrt_64f_A50(src, dst, len) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
#endif
int i = 0;
@@ -790,19 +820,25 @@ void polarToCart( InputArray src1, InputArray src2,
Mat X = dst1.getMat(), Y = dst2.getMat();
#if defined(HAVE_IPP)
if (Mag.isContinuous() && Angle.isContinuous() && X.isContinuous() && Y.isContinuous() && !angleInDegrees)
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippsPolarToCart)(const void * pSrcMagn, const void * pSrcPhase,
void * pDstRe, void * pDstIm, int len);
ippsPolarToCart ippFunc =
depth == CV_32F ? (ippsPolarToCart)ippsPolarToCart_32f :
depth == CV_64F ? (ippsPolarToCart)ippsPolarToCart_64f : 0;
CV_Assert(ippFunc != 0);
if (Mag.isContinuous() && Angle.isContinuous() && X.isContinuous() && Y.isContinuous() && !angleInDegrees)
{
typedef IppStatus (CV_STDCALL * ippsPolarToCart)(const void * pSrcMagn, const void * pSrcPhase,
void * pDstRe, void * pDstIm, int len);
ippsPolarToCart ippFunc =
depth == CV_32F ? (ippsPolarToCart)ippsPolarToCart_32f :
depth == CV_64F ? (ippsPolarToCart)ippsPolarToCart_64f : 0;
CV_Assert(ippFunc != 0);
IppStatus status = ippFunc(Mag.ptr(), Angle.ptr(), X.ptr(), Y.ptr(), static_cast<int>(cn * X.total()));
if (status >= 0)
return;
setIppErrorStatus();
IppStatus status = ippFunc(Mag.ptr(), Angle.ptr(), X.ptr(), Y.ptr(), static_cast<int>(cn * X.total()));
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
@@ -1353,17 +1389,29 @@ static void Exp_64f( const double *_x, double *y, int n )
#ifdef HAVE_IPP
static void Exp_32f_ipp(const float *x, float *y, int n)
{
if (0 <= ippsExp_32f_A21(x, y, n))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (0 <= ippsExp_32f_A21(x, y, n))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
Exp_32f(x, y, n);
}
static void Exp_64f_ipp(const double *x, double *y, int n)
{
if (0 <= ippsExp_64f_A50(x, y, n))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (0 <= ippsExp_64f_A50(x, y, n))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
Exp_64f(x, y, n);
}
@@ -2013,17 +2061,29 @@ static void Log_64f( const double *x, double *y, int n )
#ifdef HAVE_IPP
static void Log_32f_ipp(const float *x, float *y, int n)
{
if (0 <= ippsLn_32f_A21(x, y, n))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (0 <= ippsLn_32f_A21(x, y, n))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
Log_32f(x, y, n);
}
static void Log_64f_ipp(const double *x, double *y, int n)
{
if (0 <= ippsLn_64f_A50(x, y, n))
return;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (0 <= ippsLn_64f_A50(x, y, n))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
Log_64f(x, y, n);
}
@@ -2216,28 +2276,34 @@ void pow( InputArray _src, double power, OutputArray _dst )
return;
case 2:
#if defined(HAVE_IPP)
if (depth == CV_32F && !same && ( (_src.dims() <= 2 && !ocl::useOpenCL()) ||
(_src.dims() > 2 && _src.isContinuous() && _dst.isContinuous()) ))
CV_IPP_CHECK()
{
Mat src = _src.getMat();
_dst.create( src.dims, src.size, type );
Mat dst = _dst.getMat();
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, esz = CV_ELEM_SIZE(type);
if (src.isContinuous() && dst.isContinuous())
if (depth == CV_32F && !same && ( (_src.dims() <= 2 && !ocl::useOpenCL()) ||
(_src.dims() > 2 && _src.isContinuous() && _dst.isContinuous()) ))
{
size.width = (int)src.total();
size.height = 1;
srcstep = dststep = (int)src.total() * esz;
Mat src = _src.getMat();
_dst.create( src.dims, src.size, type );
Mat dst = _dst.getMat();
Size size = src.size();
int srcstep = (int)src.step, dststep = (int)dst.step, esz = CV_ELEM_SIZE(type);
if (src.isContinuous() && dst.isContinuous())
{
size.width = (int)src.total();
size.height = 1;
srcstep = dststep = (int)src.total() * esz;
}
size.width *= cn;
IppStatus status = ippiSqr_32f_C1R(src.ptr<Ipp32f>(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
size.width *= cn;
IppStatus status = ippiSqr_32f_C1R(src.ptr<Ipp32f>(), srcstep, dst.ptr<Ipp32f>(), dststep, ippiSize(size.width, size.height));
if (status >= 0)
return;
setIppErrorStatus();
}
#endif
if (same)
@@ -2288,15 +2354,21 @@ void pow( InputArray _src, double power, OutputArray _dst )
else
{
#if defined(HAVE_IPP)
if (src.isContinuous() && dst.isContinuous())
CV_IPP_CHECK()
{
IppStatus status = depth == CV_32F ?
ippsPowx_32f_A21(src.ptr<Ipp32f>(), (Ipp32f)power, dst.ptr<Ipp32f>(), (Ipp32s)(src.total() * cn)) :
ippsPowx_64f_A50(src.ptr<Ipp64f>(), power, dst.ptr<Ipp64f>(), (Ipp32s)(src.total() * cn));
if (src.isContinuous() && dst.isContinuous())
{
IppStatus status = depth == CV_32F ?
ippsPowx_32f_A21(src.ptr<Ipp32f>(), (Ipp32f)power, dst.ptr<Ipp32f>(), (Ipp32s)(src.total() * cn)) :
ippsPowx_64f_A50(src.ptr<Ipp64f>(), power, dst.ptr<Ipp64f>(), (Ipp32s)(src.total() * cn));
if (status >= 0)
return;
setIppErrorStatus();
if (status >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif

84
modules/core/src/matmul.cpp
View File

@@ -2821,11 +2821,17 @@ static double dotProd_8u(const uchar* src1, const uchar* src2, int len)
{
double r = 0;
#if ARITHM_USE_IPP && 0
if (0 <= ippiDotProd_8u64f_C1R(src1, (int)(len*sizeof(src1[0])),
src2, (int)(len*sizeof(src2[0])),
ippiSize(len, 1), &r))
return r;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (0 <= ippiDotProd_8u64f_C1R(src1, (int)(len*sizeof(src1[0])),
src2, (int)(len*sizeof(src2[0])),
ippiSize(len, 1), &r))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return r;
}
setIppErrorStatus();
}
#endif
int i = 0;
@@ -2968,10 +2974,16 @@ static double dotProd_8s(const schar* src1, const schar* src2, int len)
static double dotProd_16u(const ushort* src1, const ushort* src2, int len)
{
#if (ARITHM_USE_IPP == 1)
double r = 0;
if (0 <= ippiDotProd_16u64f_C1R(src1, (int)(len*sizeof(src1[0])), src2, (int)(len*sizeof(src2[0])), ippiSize(len, 1), &r))
return r;
setIppErrorStatus();
CV_IPP_CHECK()
{
double r = 0;
if (0 <= ippiDotProd_16u64f_C1R(src1, (int)(len*sizeof(src1[0])), src2, (int)(len*sizeof(src2[0])), ippiSize(len, 1), &r))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return r;
}
setIppErrorStatus();
}
#endif
return dotProd_(src1, src2, len);
}
@@ -2979,10 +2991,16 @@ static double dotProd_16u(const ushort* src1, const ushort* src2, int len)
static double dotProd_16s(const short* src1, const short* src2, int len)
{
#if (ARITHM_USE_IPP == 1)
double r = 0;
if (0 <= ippiDotProd_16s64f_C1R(src1, (int)(len*sizeof(src1[0])), src2, (int)(len*sizeof(src2[0])), ippiSize(len, 1), &r))
return r;
setIppErrorStatus();
CV_IPP_CHECK()
{
double r = 0;
if (0 <= ippiDotProd_16s64f_C1R(src1, (int)(len*sizeof(src1[0])), src2, (int)(len*sizeof(src2[0])), ippiSize(len, 1), &r))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return r;
}
setIppErrorStatus();
}
#endif
return dotProd_(src1, src2, len);
}
@@ -2990,10 +3008,16 @@ static double dotProd_16s(const short* src1, const short* src2, int len)
static double dotProd_32s(const int* src1, const int* src2, int len)
{
#if (ARITHM_USE_IPP == 1)
double r = 0;
if (0 <= ippiDotProd_32s64f_C1R(src1, (int)(len*sizeof(src1[0])), src2, (int)(len*sizeof(src2[0])), ippiSize(len, 1), &r))
return r;
setIppErrorStatus();
CV_IPP_CHECK()
{
double r = 0;
if (0 <= ippiDotProd_32s64f_C1R(src1, (int)(len*sizeof(src1[0])), src2, (int)(len*sizeof(src2[0])), ippiSize(len, 1), &r))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return r;
}
setIppErrorStatus();
}
#endif
return dotProd_(src1, src2, len);
}
@@ -3004,9 +3028,15 @@ static double dotProd_32f(const float* src1, const float* src2, int len)
int i = 0;
#if (ARITHM_USE_IPP == 1)
if (0 <= ippsDotProd_32f64f(src1, src2, len, &r))
return r;
setIppErrorStatus();
CV_IPP_CHECK()
{
if (0 <= ippsDotProd_32f64f(src1, src2, len, &r))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return r;
}
setIppErrorStatus();
}
#elif CV_NEON
int len0 = len & -4, blockSize0 = (1 << 13), blockSize;
float32x4_t v_zero = vdupq_n_f32(0.0f);
@@ -3035,10 +3065,16 @@ static double dotProd_32f(const float* src1, const float* src2, int len)
static double dotProd_64f(const double* src1, const double* src2, int len)
{
#if (ARITHM_USE_IPP == 1)
double r = 0;
if (0 <= ippsDotProd_64f(src1, src2, len, &r))
return r;
setIppErrorStatus();
CV_IPP_CHECK()
{
double r = 0;
if (0 <= ippsDotProd_64f(src1, src2, len, &r))
{
CV_IMPL_ADD(CV_IMPL_IPP);
return r;
}
setIppErrorStatus();
}
#endif
return dotProd_(src1, src2, len);
}

214
modules/core/src/matrix.cpp
View File

@@ -3066,64 +3066,73 @@ void cv::transpose( InputArray _src, OutputArray _dst )
}
#if defined HAVE_IPP
typedef IppStatus (CV_STDCALL * ippiTranspose)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiTransposeI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize);
ippiTranspose ippFunc = 0;
ippiTransposeI ippFuncI = 0;
CV_IPP_CHECK()
{
typedef IppStatus (CV_STDCALL * ippiTranspose)(const void * pSrc, int srcStep, void * pDst, int dstStep, IppiSize roiSize);
typedef IppStatus (CV_STDCALL * ippiTransposeI)(const void * pSrcDst, int srcDstStep, IppiSize roiSize);
ippiTranspose ippFunc = 0;
ippiTransposeI ippFuncI = 0;
if (dst.data == src.data && dst.cols == dst.rows)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_8UC1 ? (ippiTransposeI)ippiTranspose_8u_C1IR :
type == CV_8UC3 ? (ippiTransposeI)ippiTranspose_8u_C3IR :
type == CV_8UC4 ? (ippiTransposeI)ippiTranspose_8u_C4IR :
type == CV_16UC1 ? (ippiTransposeI)ippiTranspose_16u_C1IR :
type == CV_16UC3 ? (ippiTransposeI)ippiTranspose_16u_C3IR :
type == CV_16UC4 ? (ippiTransposeI)ippiTranspose_16u_C4IR :
type == CV_16SC1 ? (ippiTransposeI)ippiTranspose_16s_C1IR :
type == CV_16SC3 ? (ippiTransposeI)ippiTranspose_16s_C3IR :
type == CV_16SC4 ? (ippiTransposeI)ippiTranspose_16s_C4IR :
type == CV_32SC1 ? (ippiTransposeI)ippiTranspose_32s_C1IR :
type == CV_32SC3 ? (ippiTransposeI)ippiTranspose_32s_C3IR :
type == CV_32SC4 ? (ippiTransposeI)ippiTranspose_32s_C4IR :
type == CV_32FC1 ? (ippiTransposeI)ippiTranspose_32f_C1IR :
type == CV_32FC3 ? (ippiTransposeI)ippiTranspose_32f_C3IR :
type == CV_32FC4 ? (ippiTransposeI)ippiTranspose_32f_C4IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_8UC1 ? (ippiTranspose)ippiTranspose_8u_C1R :
type == CV_8UC3 ? (ippiTranspose)ippiTranspose_8u_C3R :
type == CV_8UC4 ? (ippiTranspose)ippiTranspose_8u_C4R :
type == CV_16UC1 ? (ippiTranspose)ippiTranspose_16u_C1R :
type == CV_16UC3 ? (ippiTranspose)ippiTranspose_16u_C3R :
type == CV_16UC4 ? (ippiTranspose)ippiTranspose_16u_C4R :
type == CV_16SC1 ? (ippiTranspose)ippiTranspose_16s_C1R :
type == CV_16SC3 ? (ippiTranspose)ippiTranspose_16s_C3R :
type == CV_16SC4 ? (ippiTranspose)ippiTranspose_16s_C4R :
type == CV_32SC1 ? (ippiTranspose)ippiTranspose_32s_C1R :
type == CV_32SC3 ? (ippiTranspose)ippiTranspose_32s_C3R :
type == CV_32SC4 ? (ippiTranspose)ippiTranspose_32s_C4R :
type == CV_32FC1 ? (ippiTranspose)ippiTranspose_32f_C1R :
type == CV_32FC3 ? (ippiTranspose)ippiTranspose_32f_C3R :
type == CV_32FC4 ? (ippiTranspose)ippiTranspose_32f_C4R : 0;
}
if (dst.data == src.data && dst.cols == dst.rows)
{
CV_SUPPRESS_DEPRECATED_START
ippFuncI =
type == CV_8UC1 ? (ippiTransposeI)ippiTranspose_8u_C1IR :
type == CV_8UC3 ? (ippiTransposeI)ippiTranspose_8u_C3IR :
type == CV_8UC4 ? (ippiTransposeI)ippiTranspose_8u_C4IR :
type == CV_16UC1 ? (ippiTransposeI)ippiTranspose_16u_C1IR :
type == CV_16UC3 ? (ippiTransposeI)ippiTranspose_16u_C3IR :
type == CV_16UC4 ? (ippiTransposeI)ippiTranspose_16u_C4IR :
type == CV_16SC1 ? (ippiTransposeI)ippiTranspose_16s_C1IR :
type == CV_16SC3 ? (ippiTransposeI)ippiTranspose_16s_C3IR :
type == CV_16SC4 ? (ippiTransposeI)ippiTranspose_16s_C4IR :
type == CV_32SC1 ? (ippiTransposeI)ippiTranspose_32s_C1IR :
type == CV_32SC3 ? (ippiTransposeI)ippiTranspose_32s_C3IR :
type == CV_32SC4 ? (ippiTransposeI)ippiTranspose_32s_C4IR :
type == CV_32FC1 ? (ippiTransposeI)ippiTranspose_32f_C1IR :
type == CV_32FC3 ? (ippiTransposeI)ippiTranspose_32f_C3IR :
type == CV_32FC4 ? (ippiTransposeI)ippiTranspose_32f_C4IR : 0;
CV_SUPPRESS_DEPRECATED_END
}
else
{
ippFunc =
type == CV_8UC1 ? (ippiTranspose)ippiTranspose_8u_C1R :
type == CV_8UC3 ? (ippiTranspose)ippiTranspose_8u_C3R :
type == CV_8UC4 ? (ippiTranspose)ippiTranspose_8u_C4R :
type == CV_16UC1 ? (ippiTranspose)ippiTranspose_16u_C1R :
type == CV_16UC3 ? (ippiTranspose)ippiTranspose_16u_C3R :
type == CV_16UC4 ? (ippiTranspose)ippiTranspose_16u_C4R :
type == CV_16SC1 ? (ippiTranspose)ippiTranspose_16s_C1R :
type == CV_16SC3 ? (ippiTranspose)ippiTranspose_16s_C3R :
type == CV_16SC4 ? (ippiTranspose)ippiTranspose_16s_C4R :
type == CV_32SC1 ? (ippiTranspose)ippiTranspose_32s_C1R :
type == CV_32SC3 ? (ippiTranspose)ippiTranspose_32s_C3R :
type == CV_32SC4 ? (ippiTranspose)ippiTranspose_32s_C4R :
type == CV_32FC1 ? (ippiTranspose)ippiTranspose_32f_C1R :
type == CV_32FC3 ? (ippiTranspose)ippiTranspose_32f_C3R :
type == CV_32FC4 ? (ippiTranspose)ippiTranspose_32f_C4R : 0;
}
IppiSize roiSize = { src.cols, src.rows };
if (ippFunc != 0)
{
if (ippFunc(src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, roiSize) >= 0)
return;
setIppErrorStatus();
}
else if (ippFuncI != 0)
{
if (ippFuncI(dst.ptr(), (int)dst.step, roiSize) >= 0)
return;
setIppErrorStatus();
IppiSize roiSize = { src.cols, src.rows };
if (ippFunc != 0)
{
if (ippFunc(src.ptr(), (int)src.step, dst.ptr(), (int)dst.step, roiSize) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
else if (ippFuncI != 0)
{
if (ippFuncI(dst.ptr(), (int)dst.step, roiSize) >= 0)
{
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
setIppErrorStatus();
}
}
#endif
@@ -3353,27 +3362,32 @@ static inline void reduceSumC_8u16u16s32f_64f(const cv::Mat& srcmat, cv::Mat& ds
}
CV_Assert(!(ippFunc && ippFuncHint) && func);
if (ippFunc)
CV_IPP_CHECK()
{
for (int y = 0; y < size.height; ++y)
if (ippFunc(srcmat.ptr(y), sstep, roisize, dstmat.ptr<Ipp64f>(y)) < 0)
{
setIppErrorStatus();
cv::Mat dstroi = dstmat.rowRange(y, y + 1);
func(srcmat.rowRange(y, y + 1), dstroi);
}
return;
}
else if (ippFuncHint)
{
for (int y = 0; y < size.height; ++y)
if (ippFuncHint(srcmat.ptr(y), sstep, roisize, dstmat.ptr<Ipp64f>(y), ippAlgHintAccurate) < 0)
{
setIppErrorStatus();
cv::Mat dstroi = dstmat.rowRange(y, y + 1);
func(srcmat.rowRange(y, y + 1), dstroi);
}
return;
if (ippFunc)
{
for (int y = 0; y < size.height; ++y)
if (ippFunc(srcmat.ptr(y), sstep, roisize, dstmat.ptr<Ipp64f>(y)) < 0)
{
setIppErrorStatus();
cv::Mat dstroi = dstmat.rowRange(y, y + 1);
func(srcmat.rowRange(y, y + 1), dstroi);
}
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
else if (ippFuncHint)
{
for (int y = 0; y < size.height; ++y)
if (ippFuncHint(srcmat.ptr(y), sstep, roisize, dstmat.ptr<Ipp64f>(y), ippAlgHintAccurate) < 0)
{
setIppErrorStatus();
cv::Mat dstroi = dstmat.rowRange(y, y + 1);
func(srcmat.rowRange(y, y + 1), dstroi);
}
CV_IMPL_ADD(CV_IMPL_IPP);
return;
}
}
func(srcmat, dstmat);
@@ -3409,7 +3423,7 @@ static inline void reduce##optype##C##favor(const cv::Mat& srcmat, cv::Mat& dstm
IppiSize roisize = ippiSize(size.width, 1);\
int sstep = (int)srcmat.step; \
\
if (srcmat.channels() == 1) \
if (CV_IPP_CHECK_COND && (srcmat.channels() == 1)) \
{ \
for (int y = 0; y < size.height; ++y) \
if (ippi##optype##_##favor##_C1R(srcmat.ptr<IppType>(y), sstep, roisize, dstmat.ptr<IppType>(y)) < 0) \
@@ -3418,6 +3432,10 @@ static inline void reduce##optype##C##favor(const cv::Mat& srcmat, cv::Mat& dstm
cv::Mat dstroi = dstmat.rowRange(y, y + 1); \
cv::reduceC_ < type1, type2, cv::Op##optype < type2 > >(srcmat.rowRange(y, y + 1), dstroi); \
} \
else \
{ \
CV_IMPL_ADD(CV_IMPL_IPP);\
} \
return; \
} \
cv::reduceC_ < type1, type2, cv::Op##optype < type2 > >(srcmat, dstmat); \
@@ -3768,8 +3786,13 @@ template<typename T> static void sort_( const Mat& src, Mat& dst, int flags )
#ifdef USE_IPP_SORT
int depth = src.depth();
IppSortFunc ippSortFunc = getSortFunc(depth, sortDescending);
IppFlipFunc ippFlipFunc = getFlipFunc(depth);
IppSortFunc ippSortFunc = 0;
IppFlipFunc ippFlipFunc = 0;
CV_IPP_CHECK()
{
ippSortFunc = getSortFunc(depth, sortDescending);
ippFlipFunc = getFlipFunc(depth);
}
#endif
for( i = 0; i < n; i++ )
@@ -3812,8 +3835,20 @@ template<typename T> static void sort_( const Mat& src, Mat& dst, int flags )
for( j = 0; j < len/2; j++ )
std::swap(ptr[j], ptr[len-1-j]);
}
#ifdef USE_IPP_SORT
else
{
CV_IMPL_ADD(CV_IMPL_IPP);
}
#endif
}
}
#ifdef USE_IPP_SORT
else
{
CV_IMPL_ADD(CV_IMPL_IPP);
}
#endif
if( !sortRows )
for( j = 0; j < len; j++ )
@@ -3878,8 +3913,13 @@ template<typename T> static void sortIdx_( const Mat& src, Mat& dst, int flags )
#if defined USE_IPP_SORT && 0
int depth = src.depth();
IppSortIndexFunc ippFunc = getSortIndexFunc(depth, sortDescending);
IppFlipFunc ippFlipFunc = getFlipFunc(depth);
IppSortIndexFunc ippFunc = 0;
IppFlipFunc ippFlipFunc = 0;
CV_IPP_CHECK()
{
ippFunc = getSortIndexFunc(depth, sortDescending);
ippFlipFunc = getFlipFunc(depth);
}
#endif
for( i = 0; i < n; i++ )
@@ -3920,8 +3960,20 @@ template<typename T> static void sortIdx_( const Mat& src, Mat& dst, int flags )
for( j = 0; j < len/2; j++ )
std::swap(iptr[j], iptr[len-1-j]);
}
#if defined USE_IPP_SORT && 0
else
{
CV_IMPL_ADD(CV_IMPL_IPP);
}
#endif
}
}
#if defined USE_IPP_SORT && 0
else
{
CV_IMPL_ADD(CV_IMPL_IPP);
}
#endif
if( !sortRows )
for( j = 0; j < len; j++ )

18
modules/core/src/ocl.cpp
View File

@@ -2893,6 +2893,9 @@ bool Kernel::create(const char* kname, const Program& prog)
p->release();
p = 0;
}
#ifdef CV_OPENCL_RUN_ASSERT // check kernel compilation fails
CV_Assert(p);
#endif
return p != 0;
}
@@ -3523,6 +3526,10 @@ protected:
entry.clBuffer_ = clCreateBuffer((cl_context)ctx.ptr(), CL_MEM_READ_WRITE, entry.capacity_, 0, &retval);
CV_Assert(retval == CL_SUCCESS);
CV_Assert(entry.clBuffer_ != NULL);
if(retval == CL_SUCCESS)
{
CV_IMPL_ADD(CV_IMPL_OCL);
}
LOG_BUFFER_POOL("OpenCL allocate %lld (0x%llx) bytes: %p\n",
(long long)entry.capacity_, (long long)entry.capacity_, entry.clBuffer_);
}
@@ -3747,6 +3754,7 @@ public:
CL_MEM_READ_WRITE|createFlags, total, 0, &retval);
if( !handle || retval != CL_SUCCESS )
return defaultAllocate(dims, sizes, type, data, step, flags, usageFlags);
CV_IMPL_ADD(CV_IMPL_OCL)
}
UMatData* u = new UMatData(this);
u->data = 0;
@@ -4187,20 +4195,24 @@ public:
CV_Assert(dst->refcount == 0);
cl_command_queue q = (cl_command_queue)Queue::getDefault().ptr();
cl_int retval;
if( iscontinuous )
{
CV_Assert( clEnqueueCopyBuffer(q, (cl_mem)src->handle, (cl_mem)dst->handle,
srcrawofs, dstrawofs, total, 0, 0, 0) == CL_SUCCESS );
CV_Assert( (retval = clEnqueueCopyBuffer(q, (cl_mem)src->handle, (cl_mem)dst->handle,
srcrawofs, dstrawofs, total, 0, 0, 0)) == CL_SUCCESS );
}
else
{
cl_int retval;
CV_Assert( (retval = clEnqueueCopyBufferRect(q, (cl_mem)src->handle, (cl_mem)dst->handle,
new_srcofs, new_dstofs, new_sz,
new_srcstep[0], new_srcstep[1],
new_dststep[0], new_dststep[1],
0, 0, 0)) == CL_SUCCESS );
}
if(retval == CL_SUCCESS)
{
CV_IMPL_ADD(CV_IMPL_OCL)
}
dst->markHostCopyObsolete(true);
dst->markDeviceCopyObsolete(false);

16
modules/core/src/precomp.hpp
View File

@@ -233,13 +233,25 @@ void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t bl
struct CoreTLSData
{
CoreTLSData() : device(0), useOpenCL(-1)
{}
CoreTLSData() : device(0), useOpenCL(-1), useIPP(-1), useCollection(false)
{
#ifdef CV_COLLECT_IMPL_DATA
implFlags = 0;
#endif
}
RNG rng;
int device;
ocl::Queue oclQueue;
int useOpenCL; // 1 - use, 0 - do not use, -1 - auto/not initialized
int useIPP; // 1 - use, 0 - do not use, -1 - auto/not initialized
bool useCollection; // enable/disable impl data collection
#ifdef CV_COLLECT_IMPL_DATA
int implFlags;
std::vector<int> implCode;
std::vector<String> implFun;
#endif
};
extern TLSData<CoreTLSData> coreTlsData;

1350
modules/core/src/stat.cpp
View File

File diff suppressed because it is too large Load Diff

74
modules/core/src/system.cpp
View File

@@ -1059,6 +1059,51 @@ TLSStorage::~TLSStorage()
TLSData<CoreTLSData> coreTlsData;
#ifdef CV_COLLECT_IMPL_DATA
void setImpl(int flags)
{
CoreTLSData* data = coreTlsData.get();
data->implFlags = flags;
data->implCode.clear();
data->implFun.clear();
}
void addImpl(int flag, const char* func)
{
CoreTLSData* data = coreTlsData.get();
data->implFlags |= flag;
if(func) // use lazy collection if name was not specified
{
size_t index = data->implCode.size();
if(!index || (data->implCode[index-1] != flag || data->implFun[index-1].compare(func))) // avoid duplicates
{
data->implCode.push_back(flag);
data->implFun.push_back(func);
}
}
}
int getImpl(std::vector<int> &impl, std::vector<String> &funName)
{
CoreTLSData* data = coreTlsData.get();
impl = data->implCode;
funName = data->implFun;
return data->implFlags; // return actual flags for lazy collection
}
bool useCollection()
{
CoreTLSData* data = coreTlsData.get();
return data->useCollection;
}
void setUseCollection(bool flag)
{
CoreTLSData* data = coreTlsData.get();
data->useCollection = flag;
}
#endif
namespace ipp
{
@@ -1084,6 +1129,35 @@ String getIppErrorLocation()
return format("%s:%d %s", filename ? filename : "", linen, funcname ? funcname : "");
}
bool useIPP()
{
#ifdef HAVE_IPP
CoreTLSData* data = coreTlsData.get();
if(data->useIPP < 0)
{
const char* pIppEnv = getenv("OPENCV_IPP");
if(pIppEnv && (cv::String(pIppEnv) == "disabled"))
data->useIPP = false;
else
data->useIPP = true;
}
return (data->useIPP > 0);
#else
return false;
#endif
}
void setUseIPP(bool flag)
{
CoreTLSData* data = coreTlsData.get();
#ifdef HAVE_IPP
data->useIPP = flag;
#else
(void)flag;
data->useIPP = false;
#endif
}
} // namespace ipp
} // namespace cv

9
modules/core/src/umatrix.cpp
View File

@@ -697,7 +697,10 @@ void UMat::copyTo(OutputArray _dst, InputArray _mask) const
size_t globalsize[2] = { cols, rows };
if (k.run(2, globalsize, NULL, false))
{
CV_IMPL_ADD(CV_IMPL_OCL);
return;
}
}
}
#endif
@@ -753,7 +756,10 @@ void UMat::convertTo(OutputArray _dst, int _type, double alpha, double beta) con
size_t globalsize[2] = { dst.cols * cn, (dst.rows + rowsPerWI - 1) / rowsPerWI };
if (k.run(2, globalsize, NULL, false))
{
CV_IMPL_ADD(CV_IMPL_OCL);
return;
}
}
}
#endif
@@ -802,7 +808,10 @@ UMat& UMat::setTo(InputArray _value, InputArray _mask)
size_t globalsize[] = { cols, (rows + rowsPerWI - 1) / rowsPerWI };
if( setK.run(2, globalsize, NULL, false) )
{
CV_IMPL_ADD(CV_IMPL_OCL);
return *this;
}
}
}
#endif