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

optimizations:

Browse Source 
- new reduce implementation (with kepler optimizations)
- saturate_cast via asm command
- video SIMD instructions in element operations
- float arithmetics instead of double
- new deviceSupports function
This commit is contained in:
Vladislav Vinogradov
2013-01-23 14:43:36 +04:00
parent ae6266e101
commit 281d036fcf
45 changed files with 9379 additions and 8846 deletions
Download Patch File Download Diff File Expand all files Collapse all files

66
modules/gpu/src/warp.cpp
View File

@@ -61,13 +61,13 @@ namespace cv { namespace gpu { namespace device
template <typename T>
void warpAffine_gpu(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, float coeffs[2 * 3], PtrStepSzb dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
int borderMode, const float* borderValue, cudaStream_t stream, bool cc20);
void buildWarpPerspectiveMaps_gpu(float coeffs[3 * 3], PtrStepSzf xmap, PtrStepSzf ymap, cudaStream_t stream);
template <typename T>
void warpPerspective_gpu(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, float coeffs[3 * 3], PtrStepSzb dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
int borderMode, const float* borderValue, cudaStream_t stream, bool cc20);
}
}}}
@@ -143,33 +143,31 @@ namespace
{
typedef typename NppWarpFunc<DEPTH>::npp_t npp_t;
static void call(const cv::gpu::GpuMat& src, cv::Size wholeSize, cv::Point ofs, cv::gpu::GpuMat& dst,
double coeffs[][3], cv::Size dsize, int interpolation, cudaStream_t stream)
static void call(const cv::gpu::GpuMat& src, cv::gpu::GpuMat& dst, double coeffs[][3], int interpolation, cudaStream_t stream)
{
static const int npp_inter[] = {NPPI_INTER_NN, NPPI_INTER_LINEAR, NPPI_INTER_CUBIC};
dst.create(dsize, src.type());
dst.setTo(cv::Scalar::all(0));
NppiSize srcsz;
srcsz.height = wholeSize.height;
srcsz.width = wholeSize.width;
srcsz.height = src.rows;
srcsz.width = src.cols;
NppiRect srcroi;
srcroi.x = ofs.x;
srcroi.y = ofs.y;
srcroi.x = 0;
srcroi.y = 0;
srcroi.height = src.rows;
srcroi.width = src.cols;
NppiRect dstroi;
dstroi.x = dstroi.y = 0;
dstroi.x = 0;
dstroi.y = 0;
dstroi.height = dst.rows;
dstroi.width = dst.cols;
cv::gpu::NppStreamHandler h(stream);
nppSafeCall( func((npp_t*)src.datastart, srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<npp_t>(), static_cast<int>(dst.step), dstroi, coeffs, npp_inter[interpolation]) );
nppSafeCall( func(src.ptr<npp_t>(), srcsz, static_cast<int>(src.step), srcroi,
dst.ptr<npp_t>(), static_cast<int>(dst.step), dstroi,
coeffs, npp_inter[interpolation]) );
if (stream == 0)
cudaSafeCall( cudaDeviceSynchronize() );
@@ -187,6 +185,8 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
dst.create(dsize, src.type());
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
@@ -231,8 +231,7 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
}
};
bool useNpp = borderMode == BORDER_CONSTANT;
useNpp = useNpp && useNppTab[src.depth()][src.channels() - 1][interpolation];
bool useNpp = borderMode == BORDER_CONSTANT && ofs.x == 0 && ofs.y == 0 && useNppTab[src.depth()][src.channels() - 1][interpolation];
#ifdef linux
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
@@ -240,7 +239,7 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
if (useNpp)
{
typedef void (*func_t)(const cv::gpu::GpuMat& src, cv::Size wholeSize, cv::Point ofs, cv::gpu::GpuMat& dst, double coeffs[][3], cv::Size dsize, int flags, cudaStream_t stream);
typedef void (*func_t)(const cv::gpu::GpuMat& src, cv::gpu::GpuMat& dst, double coeffs[][3], int flags, cudaStream_t stream);
static const func_t funcs[2][6][4] =
{
@@ -262,6 +261,8 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
}
};
dst.setTo(borderValue);
double coeffs[2][3];
Mat coeffsMat(2, 3, CV_64F, (void*)coeffs);
M.convertTo(coeffsMat, coeffsMat.type());
@@ -269,14 +270,14 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
const func_t func = funcs[(flags & WARP_INVERSE_MAP) != 0][src.depth()][src.channels() - 1];
CV_Assert(func != 0);
func(src, wholeSize, ofs, dst, coeffs, dsize, interpolation, StreamAccessor::getStream(s));
func(src, dst, coeffs, interpolation, StreamAccessor::getStream(s));
}
else
{
using namespace cv::gpu::device::imgproc;
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, float coeffs[2 * 3], PtrStepSzb dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
int borderMode, const float* borderValue, cudaStream_t stream, bool cc20);
static const func_t funcs[6][4] =
{
@@ -294,8 +295,6 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(dsize, src.type());
float coeffs[2 * 3];
Mat coeffsMat(2, 3, CV_32F, (void*)coeffs);
@@ -311,11 +310,8 @@ void cv::gpu::warpAffine(const GpuMat& src, GpuMat& dst, const Mat& M, Size dsiz
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
func(src, PtrStepSzb(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, coeffs,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(s), cc);
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(s), deviceSupports(FEATURE_SET_COMPUTE_20));
}
}
@@ -329,6 +325,8 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
CV_Assert(interpolation == INTER_NEAREST || interpolation == INTER_LINEAR || interpolation == INTER_CUBIC);
CV_Assert(borderMode == BORDER_REFLECT101 || borderMode == BORDER_REPLICATE || borderMode == BORDER_CONSTANT || borderMode == BORDER_REFLECT || borderMode == BORDER_WRAP);
dst.create(dsize, src.type());
Size wholeSize;
Point ofs;
src.locateROI(wholeSize, ofs);
@@ -373,8 +371,7 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
}
};
bool useNpp = borderMode == BORDER_CONSTANT;
useNpp = useNpp && useNppTab[src.depth()][src.channels() - 1][interpolation];
bool useNpp = borderMode == BORDER_CONSTANT && ofs.x == 0 && ofs.y == 0 && useNppTab[src.depth()][src.channels() - 1][interpolation];
#ifdef linux
// NPP bug on float data
useNpp = useNpp && src.depth() != CV_32F;
@@ -382,7 +379,7 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
if (useNpp)
{
typedef void (*func_t)(const cv::gpu::GpuMat& src, cv::Size wholeSize, cv::Point ofs, cv::gpu::GpuMat& dst, double coeffs[][3], cv::Size dsize, int flags, cudaStream_t stream);
typedef void (*func_t)(const cv::gpu::GpuMat& src, cv::gpu::GpuMat& dst, double coeffs[][3], int flags, cudaStream_t stream);
static const func_t funcs[2][6][4] =
{
@@ -404,6 +401,8 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
}
};
dst.setTo(borderValue);
double coeffs[3][3];
Mat coeffsMat(3, 3, CV_64F, (void*)coeffs);
M.convertTo(coeffsMat, coeffsMat.type());
@@ -411,14 +410,14 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
const func_t func = funcs[(flags & WARP_INVERSE_MAP) != 0][src.depth()][src.channels() - 1];
CV_Assert(func != 0);
func(src, wholeSize, ofs, dst, coeffs, dsize, interpolation, StreamAccessor::getStream(s));
func(src, dst, coeffs, interpolation, StreamAccessor::getStream(s));
}
else
{
using namespace cv::gpu::device::imgproc;
typedef void (*func_t)(PtrStepSzb src, PtrStepSzb srcWhole, int xoff, int yoff, float coeffs[2 * 3], PtrStepSzb dst, int interpolation,
int borderMode, const float* borderValue, cudaStream_t stream, int cc);
int borderMode, const float* borderValue, cudaStream_t stream, bool cc20);
static const func_t funcs[6][4] =
{
@@ -436,8 +435,6 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
int gpuBorderType;
CV_Assert(tryConvertToGpuBorderType(borderMode, gpuBorderType));
dst.create(dsize, src.type());
float coeffs[3 * 3];
Mat coeffsMat(3, 3, CV_32F, (void*)coeffs);
@@ -453,11 +450,8 @@ void cv::gpu::warpPerspective(const GpuMat& src, GpuMat& dst, const Mat& M, Size
Scalar_<float> borderValueFloat;
borderValueFloat = borderValue;
DeviceInfo info;
int cc = info.majorVersion() * 10 + info.minorVersion();
func(src, PtrStepSzb(wholeSize.height, wholeSize.width, src.datastart, src.step), ofs.x, ofs.y, coeffs,
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(s), cc);
dst, interpolation, gpuBorderType, borderValueFloat.val, StreamAccessor::getStream(s), deviceSupports(FEATURE_SET_COMPUTE_20));
}
}