Chiebot-Mirror
/
opencv
8
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

fix gpu module compilation under linux

Browse Source
pull/13383/head
Vladislav Vinogradov 15 years ago
parent 4cdcf37139
commit 863d61e9eb
4 changed files with 425 additions and 252 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 143
      modules/gpu/src/arithm.cpp
  2. 6
      modules/gpu/src/graphcuts.cpp
  3. 26
      modules/gpu/src/imgproc_gpu.cpp
  4. 30
      modules/gpu/src/matrix_operations.cpp

143
modules/gpu/src/arithm.cpp
Unescape View File

@ -84,6 +84,12 @@ void cv::gpu::polarToCart(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool,
#else /* !defined (HAVE_CUDA) */
#define NPP_VERSION (10 * NPP_VERSION_MAJOR + NPP_VERSION_MINOR)
#if (defined(_WIN32) || defined(_WIN64)) && (NPP_VERSION >= 32)
# define NPP_HAVE_COMPLEX_TYPE
#endif
////////////////////////////////////////////////////////////////////////
// add subtract multiply divide
@ -102,7 +108,11 @@ namespace
{
CV_DbgAssert(src1.size() == src2.size() && src1.type() == src2.type());
#if NPP_VERSION >= 32
CV_Assert(src1.type() == CV_8UC1 || src1.type() == CV_8UC4 || src1.type() == CV_32SC1 || src1.type() == CV_32FC1);
#else
CV_Assert(src1.type() == CV_8UC1 || src1.type() == CV_8UC4 || src1.type() == CV_32FC1);
#endif
dst.create( src1.size(), src1.type() );
@ -122,11 +132,13 @@ namespace
src2.ptr<Npp8u>(), src2.step,
dst.ptr<Npp8u>(), dst.step, sz, 0) );
break;
#if NPP_VERSION >= 32
case CV_32SC1:
nppSafeCall( npp_func_32sc1(src1.ptr<Npp32s>(), src1.step,
src2.ptr<Npp32s>(), src2.step,
dst.ptr<Npp32s>(), dst.step, sz) );
break;
#endif
case CV_32FC1:
nppSafeCall( npp_func_32fc1(src1.ptr<Npp32f>(), src1.step,
src2.ptr<Npp32f>(), src2.step,
@ -143,11 +155,13 @@ namespace
typedef NppStatus (*func_ptr)(const Npp32f *pSrc, int nSrcStep, Npp32f nValue, Npp32f *pDst,
int nDstStep, NppiSize oSizeROI);
};
#ifdef NPP_HAVE_COMPLEX_TYPE
template<> struct NppArithmScalarFunc<2>
{
typedef NppStatus (*func_ptr)(const Npp32fc *pSrc, int nSrcStep, Npp32fc nValue, Npp32fc *pDst,
int nDstStep, NppiSize oSizeROI);
};
#endif
template<int SCN, typename NppArithmScalarFunc<SCN>::func_ptr func> struct NppArithmScalar;
template<typename NppArithmScalarFunc<1>::func_ptr func> struct NppArithmScalar<1, func>
@ -163,6 +177,7 @@ namespace
nppSafeCall( func(src.ptr<Npp32f>(), src.step, (Npp32f)sc[0], dst.ptr<Npp32f>(), dst.step, sz) );
}
};
#ifdef NPP_HAVE_COMPLEX_TYPE
template<typename NppArithmScalarFunc<2>::func_ptr func> struct NppArithmScalar<2, func>
{
static void calc(const GpuMat& src, const Scalar& sc, GpuMat& dst)
@ -180,66 +195,119 @@ namespace
nppSafeCall( func(src.ptr<Npp32fc>(), src.step, nValue, dst.ptr<Npp32fc>(), dst.step, sz) );
}
};
#endif
}
void cv::gpu::add(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
#if NPP_VERSION >= 32
nppArithmCaller(src1, src2, dst, nppiAdd_8u_C1RSfs, nppiAdd_8u_C4RSfs, nppiAdd_32s_C1R, nppiAdd_32f_C1R);
#else
nppArithmCaller(src1, src2, dst, nppiAdd_8u_C1RSfs, nppiAdd_8u_C4RSfs, 0, nppiAdd_32f_C1R);
#endif
}
void cv::gpu::subtract(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
#if NPP_VERSION >= 32
nppArithmCaller(src2, src1, dst, nppiSub_8u_C1RSfs, nppiSub_8u_C4RSfs, nppiSub_32s_C1R, nppiSub_32f_C1R);
#else
nppArithmCaller(src2, src1, dst, nppiSub_8u_C1RSfs, nppiSub_8u_C4RSfs, 0, nppiSub_32f_C1R);
#endif
}
void cv::gpu::multiply(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
#if NPP_VERSION >= 32
nppArithmCaller(src1, src2, dst, nppiMul_8u_C1RSfs, nppiMul_8u_C4RSfs, nppiMul_32s_C1R, nppiMul_32f_C1R);
#else
nppArithmCaller(src1, src2, dst, nppiMul_8u_C1RSfs, nppiMul_8u_C4RSfs, 0, nppiMul_32f_C1R);
#endif
}
void cv::gpu::divide(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
#if NPP_VERSION >= 32
nppArithmCaller(src2, src1, dst, nppiDiv_8u_C1RSfs, nppiDiv_8u_C4RSfs, nppiDiv_32s_C1R, nppiDiv_32f_C1R);
#else
nppArithmCaller(src2, src1, dst, nppiDiv_8u_C1RSfs, nppiDiv_8u_C4RSfs, 0, nppiDiv_32f_C1R);
#endif
}
void cv::gpu::add(const GpuMat& src, const Scalar& sc, GpuMat& dst)
{
#ifdef NPP_HAVE_COMPLEX_TYPE
typedef void (*caller_t)(const GpuMat& src, const Scalar& sc, GpuMat& dst);
static const caller_t callers[] = {NppArithmScalar<1, nppiAddC_32f_C1R>::calc, NppArithmScalar<2, nppiAddC_32fc_C1R>::calc};
static const caller_t callers[] = {0, NppArithmScalar<1, nppiAddC_32f_C1R>::calc, NppArithmScalar<2, nppiAddC_32fc_C1R>::calc};
CV_Assert(src.type() == CV_32FC1 || src.type() == CV_32FC2);
callers[src.channels()](src, sc, dst);
#else
# if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
NppArithmScalar<1, nppiAddC_32f_C1R>::calc(src, sc, dst);
# else
CV_Assert(!"This function doesn't supported");
# endif
#endif
}
void cv::gpu::subtract(const GpuMat& src, const Scalar& sc, GpuMat& dst)
{
#ifdef NPP_HAVE_COMPLEX_TYPE
typedef void (*caller_t)(const GpuMat& src, const Scalar& sc, GpuMat& dst);
static const caller_t callers[] = {NppArithmScalar<1, nppiSubC_32f_C1R>::calc, NppArithmScalar<2, nppiSubC_32fc_C1R>::calc};
static const caller_t callers[] = {0, NppArithmScalar<1, nppiSubC_32f_C1R>::calc, NppArithmScalar<2, nppiSubC_32fc_C1R>::calc};
CV_Assert(src.type() == CV_32FC1 || src.type() == CV_32FC2);
callers[src.channels()](src, sc, dst);
#else
# if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
NppArithmScalar<1, nppiSubC_32f_C1R>::calc(src, sc, dst);
# else
CV_Assert(!"This function doesn't supported");
# endif
#endif
}
void cv::gpu::multiply(const GpuMat& src, const Scalar& sc, GpuMat& dst)
{
#ifdef NPP_HAVE_COMPLEX_TYPE
typedef void (*caller_t)(const GpuMat& src, const Scalar& sc, GpuMat& dst);
static const caller_t callers[] = {NppArithmScalar<1, nppiMulC_32f_C1R>::calc, NppArithmScalar<2, nppiMulC_32fc_C1R>::calc};
static const caller_t callers[] = {0, NppArithmScalar<1, nppiMulC_32f_C1R>::calc, NppArithmScalar<2, nppiMulC_32fc_C1R>::calc};
CV_Assert(src.type() == CV_32FC1 || src.type() == CV_32FC2);
callers[src.channels()](src, sc, dst);
#else
# if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
NppArithmScalar<1, nppiMulC_32f_C1R>::calc(src, sc, dst);
# else
CV_Assert(!"This function doesn't supported");
# endif
#endif
}
void cv::gpu::divide(const GpuMat& src, const Scalar& sc, GpuMat& dst)
{
#ifdef NPP_HAVE_COMPLEX_TYPE
typedef void (*caller_t)(const GpuMat& src, const Scalar& sc, GpuMat& dst);
static const caller_t callers[] = {NppArithmScalar<1, nppiDivC_32f_C1R>::calc, NppArithmScalar<2, nppiDivC_32fc_C1R>::calc};
static const caller_t callers[] = {0, NppArithmScalar<1, nppiDivC_32f_C1R>::calc, NppArithmScalar<2, nppiDivC_32fc_C1R>::calc};
CV_Assert(src.type() == CV_32FC1 || src.type() == CV_32FC2);
callers[src.channels()](src, sc, dst);
#else
# if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
NppArithmScalar<1, nppiDivC_32f_C1R>::calc(src, sc, dst);
# else
CV_Assert(!"This function doesn't supported");
# endif
#endif
}
////////////////////////////////////////////////////////////////////////
@ -265,7 +333,11 @@ void cv::gpu::absdiff(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
CV_DbgAssert(src1.size() == src2.size() && src1.type() == src2.type());
#if NPP_VERSION >= 32
CV_Assert(src1.type() == CV_8UC1 || src1.type() == CV_8UC4 || src1.type() == CV_32SC1 || src1.type() == CV_32FC1);
#else
CV_Assert(src1.type() == CV_8UC1 || src1.type() == CV_8UC4 || src1.type() == CV_32FC1);
#endif
dst.create( src1.size(), src1.type() );
@ -285,11 +357,13 @@ void cv::gpu::absdiff(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
src2.ptr<Npp8u>(), src2.step,
dst.ptr<Npp8u>(), dst.step, sz) );
break;
#if NPP_VERSION >= 32
case CV_32SC1:
nppSafeCall( nppiAbsDiff_32s_C1R(src1.ptr<Npp32s>(), src1.step,
src2.ptr<Npp32s>(), src2.step,
dst.ptr<Npp32s>(), dst.step, sz) );
break;
#endif
case CV_32FC1:
nppSafeCall( nppiAbsDiff_32f_C1R(src1.ptr<Npp32f>(), src1.step,
src2.ptr<Npp32f>(), src2.step,
@ -302,6 +376,7 @@ void cv::gpu::absdiff(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
void cv::gpu::absdiff(const GpuMat& src, const Scalar& s, GpuMat& dst)
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
dst.create( src.size(), src.type() );
@ -311,6 +386,9 @@ void cv::gpu::absdiff(const GpuMat& src, const Scalar& s, GpuMat& dst)
sz.height = src.rows;
nppSafeCall( nppiAbsDiffC_32f_C1R(src.ptr<Npp32f>(), src.step, dst.ptr<Npp32f>(), dst.step, sz, (Npp32f)s[0]) );
#else
CV_Assert(!"This function doesn't supported");
#endif
}
////////////////////////////////////////////////////////////////////////
@ -444,13 +522,16 @@ void cv::gpu::flip(const GpuMat& src, GpuMat& dst, int flipCode)
Scalar cv::gpu::sum(const GpuMat& src)
{
CV_Assert(!"disabled until fix crash");
CV_Assert(src.type() == CV_8UC1 || src.type() == CV_8UC4);
NppiSize sz;
sz.width = src.cols;
sz.height = src.rows;
Scalar res;
#if NPP_VERSION >= 32
CV_Assert(!"disabled until fix crash");
int bufsz;
if (src.type() == CV_8UC1)
@ -458,19 +539,23 @@ Scalar cv::gpu::sum(const GpuMat& src)
nppiReductionGetBufferHostSize_8u_C1R(sz, &bufsz);
GpuMat buf(1, bufsz, CV_32S);
Scalar res;
nppSafeCall( nppiSum_8u_C1R(src.ptr<Npp8u>(), src.step, sz, buf.ptr<Npp32s>(), res.val) );
return res;
}
else
{
nppiReductionGetBufferHostSize_8u_C4R(sz, &bufsz);
GpuMat buf(1, bufsz, CV_32S);
Scalar res;
nppSafeCall( nppiSum_8u_C4R(src.ptr<Npp8u>(), src.step, sz, buf.ptr<Npp32s>(), res.val) );
return res;
}
#else
if (src.type() == CV_8UC1)
nppSafeCall( nppiSum_8u_C1R(src.ptr<Npp8u>(), src.step, sz, res.val) );
else
nppSafeCall( nppiSum_8u_C4R(src.ptr<Npp8u>(), src.step, sz, res.val) );
#endif
return res;
}
////////////////////////////////////////////////////////////////////////
@ -501,18 +586,26 @@ namespace
sz.width = src.cols;
sz.height = src.rows;
Npp8u* cuMin = nppsMalloc_8u(4);
Npp8u* cuMax = nppsMalloc_8u(4);
Npp8u* cuMem;
#if NPP_VERSION >= 32
cuMem = nppsMalloc_8u(8);
#else
cudaSafeCall( cudaMalloc((void**)&cuMem, 8 * sizeof(Npp8u)) );
#endif
nppSafeCall( nppiMinMax_8u_C4R(src.ptr<Npp8u>(), src.step, sz, cuMin, cuMax) );
nppSafeCall( nppiMinMax_8u_C4R(src.ptr<Npp8u>(), src.step, sz, cuMem, cuMem + 4) );
if (minVal)
cudaMemcpy(minVal, cuMin, 4 * sizeof(Npp8u), cudaMemcpyDeviceToHost);
cudaMemcpy(minVal, cuMem, 4 * sizeof(Npp8u), cudaMemcpyDeviceToHost);
if (maxVal)
cudaMemcpy(maxVal, cuMax, 4 * sizeof(Npp8u), cudaMemcpyDeviceToHost);
cudaMemcpy(maxVal, cuMem + 4, 4 * sizeof(Npp8u), cudaMemcpyDeviceToHost);
nppsFree(cuMin);
nppsFree(cuMax);
#if NPP_VERSION >= 32
nppsFree(cuMem);
#else
cudaSafeCall( cudaFree(cuMem) );
#endif
}
}
@ -591,6 +684,7 @@ void cv::gpu::LUT(const GpuMat& src, const Mat& lut, GpuMat& dst)
void cv::gpu::exp(const GpuMat& src, GpuMat& dst)
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
dst.create(src.size(), src.type());
@ -600,6 +694,9 @@ void cv::gpu::exp(const GpuMat& src, GpuMat& dst)
sz.height = src.rows;
nppSafeCall( nppiExp_32f_C1R(src.ptr<Npp32f>(), src.step, dst.ptr<Npp32f>(), dst.step, sz) );
#else
CV_Assert(!"This function doesn't supported");
#endif
}
////////////////////////////////////////////////////////////////////////
@ -607,6 +704,7 @@ void cv::gpu::exp(const GpuMat& src, GpuMat& dst)
void cv::gpu::log(const GpuMat& src, GpuMat& dst)
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_32FC1);
dst.create(src.size(), src.type());
@ -616,11 +714,15 @@ void cv::gpu::log(const GpuMat& src, GpuMat& dst)
sz.height = src.rows;
nppSafeCall( nppiLn_32f_C1R(src.ptr<Npp32f>(), src.step, dst.ptr<Npp32f>(), dst.step, sz) );
#else
CV_Assert(!"This function doesn't supported");
#endif
}
////////////////////////////////////////////////////////////////////////
// NPP magnitide
#ifdef NPP_HAVE_COMPLEX_TYPE
namespace
{
typedef NppStatus (*nppMagnitude_t)(const Npp32fc* pSrc, int nSrcStep, Npp32f* pDst, int nDstStep, NppiSize oSizeROI);
@ -638,15 +740,24 @@ namespace
nppSafeCall( func(src.ptr<Npp32fc>(), src.step, dst.ptr<Npp32f>(), dst.step, sz) );
}
}
#endif
void cv::gpu::magnitude(const GpuMat& src, GpuMat& dst)
{
#ifdef NPP_HAVE_COMPLEX_TYPE
::npp_magnitude(src, dst, nppiMagnitude_32fc32f_C1R);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
void cv::gpu::magnitudeSqr(const GpuMat& src, GpuMat& dst)
{
#ifdef NPP_HAVE_COMPLEX_TYPE
::npp_magnitude(src, dst, nppiMagnitudeSqr_32fc32f_C1R);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
////////////////////////////////////////////////////////////////////////

6
modules/gpu/src/graphcuts.cpp
Unescape View File

@ -48,8 +48,11 @@ void cv::gpu::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Gpu
#else /* !defined (HAVE_CUDA) */
#define NPP_VERSION (10 * NPP_VERSION_MAJOR + NPP_VERSION_MINOR)
void cv::gpu::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& bottom, GpuMat& labels, GpuMat& buf)
{
#if NPP_VERSION >= 32
CV_Assert(leftTransp.type() == CV_32S && rightTransp.type() == CV_32S);
CV_Assert(terminals.type() == CV_32S && bottom.type() == CV_32S && top.type() == CV_32S);
CV_Assert(terminals.size() == leftTransp.size());
@ -71,6 +74,9 @@ void cv::gpu::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTrans
nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
terminals.step, leftTransp.step, sznpp, labels.ptr<Npp8u>(), labels.step, buf.ptr<Npp8u>()) );
#else
CV_Assert(!"This function doesn't supported");
#endif
}

26
modules/gpu/src/imgproc_gpu.cpp
Unescape View File

@ -71,6 +71,8 @@ void cv::gpu::histRange(const GpuMat&, GpuMat*, const GpuMat*) { throw_nogpu();
#else /* !defined (HAVE_CUDA) */
#define NPP_VERSION (10 * NPP_VERSION_MAJOR + NPP_VERSION_MINOR)
namespace cv { namespace gpu { namespace imgproc
{
void remap_gpu_1c(const DevMem2D& src, const DevMem2Df& xmap, const DevMem2Df& ymap, DevMem2D dst);
@ -577,6 +579,7 @@ void cv::gpu::rectStdDev(const GpuMat& src, const GpuMat& sqr, GpuMat& dst, cons
void cv::gpu::Canny(const GpuMat& image, GpuMat& edges, double threshold1, double threshold2, int apertureSize)
{
#if NPP_VERSION >= 32
CV_Assert(!"disabled until fix crash");
CV_Assert(image.type() == CV_8UC1);
@ -600,6 +603,9 @@ void cv::gpu::Canny(const GpuMat& image, GpuMat& edges, double threshold1, doubl
nppSafeCall( nppiCanny_32f8u_C1R(srcDx.ptr<Npp32f>(), srcDx.step, srcDy.ptr<Npp32f>(), srcDy.step,
edges.ptr<Npp8u>(), edges.step, sz, (Npp32f)threshold1, (Npp32f)threshold2, buf.ptr<Npp8u>()) );
#else
CV_Assert(!"This function doesn't supported");
#endif
}
////////////////////////////////////////////////////////////////////////
@ -783,13 +789,18 @@ namespace
void cv::gpu::evenLevels(GpuMat& levels, int nLevels, int lowerLevel, int upperLevel)
{
#if NPP_VERSION >= 32
Mat host_levels(1, nLevels, CV_32SC1);
nppSafeCall( nppiEvenLevelsHost_32s(host_levels.ptr<Npp32s>(), nLevels, lowerLevel, upperLevel) );
levels.upload(host_levels);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
void cv::gpu::histEven(const GpuMat& src, GpuMat& hist, int histSize, int lowerLevel, int upperLevel)
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_8UC1 || src.type() == CV_16UC1 || src.type() == CV_16SC1 );
typedef void (*hist_t)(const GpuMat& src, GpuMat& hist, int levels, int lowerLevel, int upperLevel);
@ -802,10 +813,14 @@ void cv::gpu::histEven(const GpuMat& src, GpuMat& hist, int histSize, int lowerL
};
hist_callers[src.depth()](src, hist, histSize, lowerLevel, upperLevel);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
void cv::gpu::histEven(const GpuMat& src, GpuMat hist[4], int histSize[4], int lowerLevel[4], int upperLevel[4])
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_8UC4 || src.type() == CV_16UC4 || src.type() == CV_16SC4 );
typedef void (*hist_t)(const GpuMat& src, GpuMat hist[4], int levels[4], int lowerLevel[4], int upperLevel[4]);
@ -818,10 +833,14 @@ void cv::gpu::histEven(const GpuMat& src, GpuMat hist[4], int histSize[4], int l
};
hist_callers[src.depth()](src, hist, histSize, lowerLevel, upperLevel);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
void cv::gpu::histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels)
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_8UC1 || src.type() == CV_16UC1 || src.type() == CV_16SC1 || src.type() == CV_32FC1);
typedef void (*hist_t)(const GpuMat& src, GpuMat& hist, const GpuMat& levels);
@ -836,10 +855,14 @@ void cv::gpu::histRange(const GpuMat& src, GpuMat& hist, const GpuMat& levels)
};
hist_callers[src.depth()](src, hist, levels);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
void cv::gpu::histRange(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4])
{
#if NPP_VERSION >= 32
CV_Assert(src.type() == CV_8UC4 || src.type() == CV_16UC4 || src.type() == CV_16SC4 || src.type() == CV_32FC4);
typedef void (*hist_t)(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4]);
@ -854,6 +877,9 @@ void cv::gpu::histRange(const GpuMat& src, GpuMat hist[4], const GpuMat levels[4
};
hist_callers[src.depth()](src, hist, levels);
#else
CV_Assert(!"This function doesn't supported");
#endif
}
#endif /* !defined (HAVE_CUDA) */

30
modules/gpu/src/matrix_operations.cpp
Unescape View File

@ -77,6 +77,8 @@ namespace cv
#else /* !defined (HAVE_CUDA) */
#define NPP_VERSION (10 * NPP_VERSION_MAJOR + NPP_VERSION_MINOR)
namespace cv
{
namespace gpu
@ -232,7 +234,11 @@ void cv::gpu::GpuMat::convertTo( GpuMat& dst, int rtype, double alpha, double be
{NppCvt<CV_8U, CV_16U, nppiConvert_8u16u_C1R>::cvt,convertToKernelCaller,convertToKernelCaller,NppCvt<CV_8U, CV_16U, nppiConvert_8u16u_C4R>::cvt},
{NppCvt<CV_8U, CV_16S, nppiConvert_8u16s_C1R>::cvt,convertToKernelCaller,convertToKernelCaller,NppCvt<CV_8U, CV_16S, nppiConvert_8u16s_C4R>::cvt},
{convertToKernelCaller,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
#if NPP_VERSION >= 32
{NppCvt<CV_8U, CV_32F, nppiConvert_8u32f_C1R>::cvt,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
#else
{convertToKernelCaller,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
#endif
{convertToKernelCaller,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
{0,0,0,0}
},
@ -277,7 +283,11 @@ void cv::gpu::GpuMat::convertTo( GpuMat& dst, int rtype, double alpha, double be
{0,0,0,0}
},
{
#if NPP_VERSION >= 32
{NppCvt<CV_32F, CV_8U, nppiConvert_32f8u_C1R>::cvt,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
#else
{convertToKernelCaller,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
#endif
{convertToKernelCaller,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
{NppCvt<CV_32F, CV_16U, nppiConvert_32f16u_C1R>::cvt,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
{NppCvt<CV_32F, CV_16S, nppiConvert_32f16s_C1R>::cvt,convertToKernelCaller,convertToKernelCaller,convertToKernelCaller},
@ -421,10 +431,26 @@ GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask)
{
{NppSet<CV_8U, 1, nppiSet_8u_C1R>::set,kernelSet,kernelSet,NppSet<CV_8U, 4, nppiSet_8u_C4R>::set},
{kernelSet,kernelSet,kernelSet,kernelSet},
#if NPP_VERSION >= 32
{NppSet<CV_16U, 1, nppiSet_16u_C1R>::set,kernelSet,kernelSet,NppSet<CV_16U, 4, nppiSet_16u_C4R>::set},
#else
{kernelSet,kernelSet,kernelSet,kernelSet},
#endif
#if NPP_VERSION >= 32
{NppSet<CV_16S, 1, nppiSet_16s_C1R>::set,kernelSet,kernelSet,NppSet<CV_16S, 4, nppiSet_16s_C4R>::set},
#else
{kernelSet,kernelSet,kernelSet,kernelSet},
#endif
#if NPP_VERSION >= 32
{NppSet<CV_32S, 1, nppiSet_32s_C1R>::set,kernelSet,kernelSet,NppSet<CV_32S, 4, nppiSet_32s_C4R>::set},
#else
{NppSet<CV_32S, 1, nppiSet_32s_C1R>::set,kernelSet,kernelSet,kernelSet},
#endif
#if NPP_VERSION >= 32
{NppSet<CV_32F, 1, nppiSet_32f_C1R>::set,kernelSet,kernelSet,NppSet<CV_32F, 4, nppiSet_32f_C4R>::set},
#else
{NppSet<CV_32F, 1, nppiSet_32f_C1R>::set,kernelSet,kernelSet,kernelSet},
#endif
{kernelSet,kernelSet,kernelSet,kernelSet},
{0,0,0,0}
};
@ -432,6 +458,7 @@ GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask)
}
else
{
#if NPP_VERSION >= 32
typedef void (*set_caller_t)(GpuMat& src, const Scalar& s, const GpuMat& mask);
static const set_caller_t set_callers[8][4] =
{
@ -445,6 +472,9 @@ GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask)
{0,0,0,0}
};
set_callers[depth()][channels()-1](*this, s, mask);
#else
kernelSetMask(*this, s, mask);
#endif
}
return *this;

Write Preview
Loading…
Cancel
Save