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

update docs

Browse Source 
minor fixes and refactoring of GPU module
This commit is contained in:
Vladislav Vinogradov
2011-02-16 08:31:45 +00:00
parent 7d42dbdd71
commit 54fa600b9e
16 changed files with 77117 additions and 76809 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
modules/gpu/src/cuda/brute_force_matcher.cu
View File

@@ -582,10 +582,10 @@ namespace cv { namespace gpu { namespace bfmatcher
}
///////////////////////////////////////////////////////////////////////////////
// Match kernel chooser
// Match caller
template <typename Dist, typename T, typename Train, typename Mask>
void match_chooser(const DevMem2D_<T>& queryDescs, const Train& train,
void matchDispatcher(const DevMem2D_<T>& queryDescs, const Train& train,
const Mask& mask, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12)
{
@@ -616,11 +616,11 @@ namespace cv { namespace gpu { namespace bfmatcher
if (mask.data)
{
SingleMask m(mask);
match_chooser<L1Dist>((DevMem2D_<T>)queryDescs, train, m, trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, train, m, trainIdx, imgIdx, distance, cc_12);
}
else
{
match_chooser<L1Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
}
}
@@ -640,11 +640,11 @@ namespace cv { namespace gpu { namespace bfmatcher
if (mask.data)
{
SingleMask m(mask);
match_chooser<L2Dist>((DevMem2D_<T>)queryDescs, train, m, trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, train, m, trainIdx, imgIdx, distance, cc_12);
}
else
{
match_chooser<L2Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
}
}
@@ -664,11 +664,11 @@ namespace cv { namespace gpu { namespace bfmatcher
if (maskCollection.data)
{
MaskCollection mask(maskCollection.data);
match_chooser<L1Dist>((DevMem2D_<T>)queryDescs, train, mask, trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, train, mask, trainIdx, imgIdx, distance, cc_12);
}
else
{
match_chooser<L1Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
}
}
@@ -688,11 +688,11 @@ namespace cv { namespace gpu { namespace bfmatcher
if (maskCollection.data)
{
MaskCollection mask(maskCollection.data);
match_chooser<L2Dist>((DevMem2D_<T>)queryDescs, train, mask, trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, train, mask, trainIdx, imgIdx, distance, cc_12);
}
else
{
match_chooser<L2Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
matchDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, train, WithOutMask(), trainIdx, imgIdx, distance, cc_12);
}
}
@@ -942,22 +942,35 @@ namespace cv { namespace gpu { namespace bfmatcher
///////////////////////////////////////////////////////////////////////////////
// knn match caller
template <typename Dist, typename T, typename Mask>
void calcDistanceDispatcher(const DevMem2D_<T>& queryDescs, const DevMem2D_<T>& trainDescs,
const Mask& mask, const DevMem2Df& allDist)
{
calcDistance_caller<16, 16, Dist>(queryDescs, trainDescs, mask, allDist);
}
void findKnnMatchDispatcher(int knn, const DevMem2Di& trainIdx, const DevMem2Df& distance,
const DevMem2Df& allDist)
{
findKnnMatch_caller<256>(knn, trainIdx, distance, allDist);
}
template <typename T>
void knnMatchL1_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, int knn,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist)
{
if (mask.data)
{
calcDistance_caller<16, 16, L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
calcDistanceDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
SingleMask(mask), allDist);
}
else
{
calcDistance_caller<16, 16, L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
calcDistanceDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
WithOutMask(), allDist);
}
findKnnMatch_caller<256>(knn, trainIdx, distance, allDist);
findKnnMatchDispatcher(knn, trainIdx, distance, allDist);
}
template void knnMatchL1_gpu<uchar >(const DevMem2D& queryDescs, const DevMem2D& trainDescs, int knn, const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist);
@@ -973,16 +986,16 @@ namespace cv { namespace gpu { namespace bfmatcher
{
if (mask.data)
{
calcDistance_caller<16, 16, L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
calcDistanceDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
SingleMask(mask), allDist);
}
else
{
calcDistance_caller<16, 16, L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
calcDistanceDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
WithOutMask(), allDist);
}
findKnnMatch_caller<256>(knn, trainIdx, distance, allDist);
findKnnMatchDispatcher(knn, trainIdx, distance, allDist);
}
template void knnMatchL2_gpu<uchar >(const DevMem2D& queryDescs, const DevMem2D& trainDescs, int knn, const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist);
@@ -1061,7 +1074,16 @@ namespace cv { namespace gpu { namespace bfmatcher
}
///////////////////////////////////////////////////////////////////////////////
// Radius Match kernel chooser
// Radius Match caller
template <typename Dist, typename T, typename Mask>
void radiusMatchDispatcher(const DevMem2D_<T>& queryDescs, const DevMem2D_<T>& trainDescs,
float maxDistance, const Mask& mask, const DevMem2Di& trainIdx, unsigned int* nMatches,
const DevMem2Df& distance)
{
radiusMatch_caller<16, 16, Dist>(queryDescs, trainDescs, maxDistance, mask,
trainIdx, nMatches, distance);
}
template <typename T>
void radiusMatchL1_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, float maxDistance,
@@ -1069,12 +1091,12 @@ namespace cv { namespace gpu { namespace bfmatcher
{
if (mask.data)
{
radiusMatch_caller<16, 16, L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
radiusMatchDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
maxDistance, SingleMask(mask), trainIdx, nMatches, distance);
}
else
{
radiusMatch_caller<16, 16, L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
radiusMatchDispatcher<L1Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
maxDistance, WithOutMask(), trainIdx, nMatches, distance);
}
}
@@ -1092,12 +1114,12 @@ namespace cv { namespace gpu { namespace bfmatcher
{
if (mask.data)
{
radiusMatch_caller<16, 16, L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
radiusMatchDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
maxDistance, SingleMask(mask), trainIdx, nMatches, distance);
}
else
{
radiusMatch_caller<16, 16, L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
radiusMatchDispatcher<L2Dist>((DevMem2D_<T>)queryDescs, (DevMem2D_<T>)trainDescs,
maxDistance, WithOutMask(), trainIdx, nMatches, distance);
}
}

1526
modules/gpu/src/cuda/color.cu
View File

File diff suppressed because it is too large Load Diff

11
modules/gpu/src/cudastream.cpp
View File

@@ -190,6 +190,9 @@ void cv::gpu::Stream::enqueueCopy(const GpuMat& src, GpuMat& dst) { devcopy(src,
void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar val)
{
CV_Assert((src.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*set_caller_t)(GpuMat& src, const Scalar& s, cudaStream_t stream);
static const set_caller_t set_callers[] =
{
@@ -201,6 +204,11 @@ void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar val)
void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar val, const GpuMat& mask)
{
CV_Assert((src.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
CV_Assert(mask.type() == CV_8UC1);
typedef void (*set_caller_t)(GpuMat& src, const Scalar& s, const GpuMat& mask, cudaStream_t stream);
static const set_caller_t set_callers[] =
{
@@ -212,6 +220,9 @@ void cv::gpu::Stream::enqueueMemSet(GpuMat& src, Scalar val, const GpuMat& mask)
void cv::gpu::Stream::enqueueConvert(const GpuMat& src, GpuMat& dst, int rtype, double alpha, double beta)
{
CV_Assert((src.depth() != CV_64F && CV_MAT_DEPTH(rtype) != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
bool noScale = fabs(alpha-1) < std::numeric_limits<double>::epsilon() && fabs(beta) < std::numeric_limits<double>::epsilon();
if( rtype < 0 )

33
modules/gpu/src/element_operations.cpp
View File

@@ -625,7 +625,11 @@ namespace
}
void cv::gpu::min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -637,6 +641,10 @@ void cv::gpu::min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
void cv::gpu::min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Stream& stream)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -648,6 +656,9 @@ void cv::gpu::min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Str
void cv::gpu::min(const GpuMat& src1, double src2, GpuMat& dst)
{
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, double src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -659,6 +670,9 @@ void cv::gpu::min(const GpuMat& src1, double src2, GpuMat& dst)
void cv::gpu::min(const GpuMat& src1, double src2, GpuMat& dst, const Stream& stream)
{
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, double src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -670,6 +684,10 @@ void cv::gpu::min(const GpuMat& src1, double src2, GpuMat& dst, const Stream& st
void cv::gpu::max(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -681,6 +699,10 @@ void cv::gpu::max(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
void cv::gpu::max(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Stream& stream)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -692,6 +714,9 @@ void cv::gpu::max(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Str
void cv::gpu::max(const GpuMat& src1, double src2, GpuMat& dst)
{
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, double src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -703,6 +728,9 @@ void cv::gpu::max(const GpuMat& src1, double src2, GpuMat& dst)
void cv::gpu::max(const GpuMat& src1, double src2, GpuMat& dst, const Stream& stream)
{
CV_Assert((src1.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*func_t)(const GpuMat& src1, double src2, GpuMat& dst, cudaStream_t stream);
static const func_t funcs[] =
{
@@ -749,6 +777,9 @@ double cv::gpu::threshold(const GpuMat& src, GpuMat& dst, double thresh, double
}
else
{
CV_Assert((src.depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
typedef void (*caller_t)(const GpuMat& src, GpuMat& dst, double thresh, double maxVal, int type,
cudaStream_t stream);

6
modules/gpu/src/matrix_operations.cpp
View File

@@ -205,6 +205,9 @@ namespace
void cv::gpu::GpuMat::convertTo( GpuMat& dst, int rtype, double alpha, double beta ) const
{
CV_Assert((depth() != CV_64F && CV_MAT_DEPTH(rtype) != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
bool noScale = fabs(alpha-1) < std::numeric_limits<double>::epsilon() && fabs(beta) < std::numeric_limits<double>::epsilon();
if( rtype < 0 )
@@ -428,6 +431,9 @@ GpuMat& GpuMat::setTo(const Scalar& s, const GpuMat& mask)
{
CV_Assert(mask.type() == CV_8UC1);
CV_Assert((depth() != CV_64F) ||
(TargetArchs::builtWith(NATIVE_DOUBLE) && DeviceInfo().supports(NATIVE_DOUBLE)));
CV_DbgAssert(!this->empty());
NppiSize sz;

30
modules/gpu/src/opencv2/gpu/device/transform.hpp
View File

@@ -393,11 +393,37 @@ namespace cv
}
};
template <typename T, typename D, int scn, int dcn> struct UseSmartUn_
{
static const bool value = false;
};
template <typename T, typename D> struct UseSmartUn_<T, D, 1, 1>
{
static const bool value = device::UnReadWriteTraits<T, D>::shift != 1;
};
template <typename T, typename D> struct UseSmartUn
{
static const bool value = UseSmartUn_<T, D, device::VecTraits<T>::cn, device::VecTraits<D>::cn>::value;
};
template <typename T1, typename T2, typename D, int src1cn, int src2cn, int dstcn> struct UseSmartBin_
{
static const bool value = false;
};
template <typename T1, typename T2, typename D> struct UseSmartBin_<T1, T2, D, 1, 1, 1>
{
static const bool value = device::BinReadWriteTraits<T1, T2, D>::shift != 1;
};
template <typename T1, typename T2, typename D> struct UseSmartBin
{
static const bool value = UseSmartBin_<T1, T2, D, device::VecTraits<T1>::cn, device::VecTraits<T2>::cn, device::VecTraits<D>::cn>::value;
};
template <typename T, typename D, typename UnOp, typename Mask>
static void transform_caller(const DevMem2D_<T>& src, const DevMem2D_<D>& dst, UnOp op, const Mask& mask,
cudaStream_t stream = 0)
{
TransformDispatcher<device::VecTraits<T>::cn == 1 && device::VecTraits<D>::cn == 1 && device::UnReadWriteTraits<T, D>::shift != 1>::call(src, dst, op, mask, stream);
TransformDispatcher< UseSmartUn<T, D>::value >::call(src, dst, op, mask, stream);
}
template <typename T, typename D, typename UnOp>
@@ -416,7 +442,7 @@ namespace cv
static void transform_caller(const DevMem2D_<T1>& src1, const DevMem2D_<T2>& src2, const DevMem2D_<D>& dst,
BinOp op, const Mask& mask, cudaStream_t stream = 0)
{
TransformDispatcher<device::VecTraits<T1>::cn == 1 && device::VecTraits<T2>::cn == 1 && device::VecTraits<D>::cn == 1 && device::BinReadWriteTraits<T1, T2, D>::shift != 1>::call(src1, src2, dst, op, mask, stream);
TransformDispatcher< UseSmartBin<T1, T2, D>::value >::call(src1, src2, dst, op, mask, stream);
}
template <typename T1, typename T2, typename D, typename BinOp>