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

modified according to CUDA 4.0 API updates

Browse Source
This commit is contained in:
Vladislav Vinogradov
2011-05-31 08:31:10 +00:00
parent 98d663e7e0
commit 926a6bba00
40 changed files with 1134 additions and 1818 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
modules/gpu/src/brute_force_matcher.cpp
View File

@@ -54,18 +54,18 @@ const vector<GpuMat>& cv::gpu::BruteForceMatcher_GPU_base::getTrainDescriptors()
void cv::gpu::BruteForceMatcher_GPU_base::clear() { throw_nogpu(); }
bool cv::gpu::BruteForceMatcher_GPU_base::empty() const { throw_nogpu(); return true; }
bool cv::gpu::BruteForceMatcher_GPU_base::isMaskSupported() const { throw_nogpu(); return true; }
void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, const GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::matchDownload(const GpuMat&, const GpuMat&, vector<DMatch>&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::match(const GpuMat&, const GpuMat&, vector<DMatch>&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::makeGpuCollection(GpuMat&, GpuMat&, const vector<GpuMat>&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, GpuMat&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, GpuMat&, const GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::matchDownload(const GpuMat&, const GpuMat&, const GpuMat&, std::vector<DMatch>&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::match(const GpuMat&, std::vector<DMatch>&, const std::vector<GpuMat>&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, GpuMat&, int, const GpuMat&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, GpuMat&, int, const GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::knnMatchDownload(const GpuMat&, const GpuMat&, std::vector< std::vector<DMatch> >&, bool) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat&, const GpuMat&, std::vector< std::vector<DMatch> >&, int, const GpuMat&, bool) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat&, std::vector< std::vector<DMatch> >&, int, const std::vector<GpuMat>&, bool) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, GpuMat&, float, const GpuMat&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, GpuMat&, float, const GpuMat&, Stream&) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchDownload(const GpuMat&, const GpuMat&, const GpuMat&, std::vector< std::vector<DMatch> >&, bool) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat&, const GpuMat&, std::vector< std::vector<DMatch> >&, float, const GpuMat&, bool) { throw_nogpu(); }
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat&, std::vector< std::vector<DMatch> >&, float, const std::vector<GpuMat>&, bool) { throw_nogpu(); }
@@ -77,47 +77,47 @@ namespace cv { namespace gpu { namespace bfmatcher
template <typename T>
void matchSingleL1_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
template <typename T>
void matchSingleL2_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
template <typename T>
void matchSingleHamming_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
template <typename T>
void matchCollectionL1_gpu(const DevMem2D& queryDescs, const DevMem2D& trainCollection,
const DevMem2D_<PtrStep>& maskCollection, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
template <typename T>
void matchCollectionL2_gpu(const DevMem2D& queryDescs, const DevMem2D& trainCollection,
const DevMem2D_<PtrStep>& maskCollection, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
template <typename T>
void matchCollectionHamming_gpu(const DevMem2D& queryDescs, const DevMem2D& trainCollection,
const DevMem2D_<PtrStep>& maskCollection, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
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);
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist, cudaStream_t stream);
template <typename T>
void knnMatchL2_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, int knn,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist);
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist, cudaStream_t stream);
template <typename T>
void knnMatchHamming_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, int knn,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist);
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist, cudaStream_t stream);
template <typename T>
void radiusMatchL1_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, float maxDistance,
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance);
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance, cudaStream_t stream);
template <typename T>
void radiusMatchL2_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, float maxDistance,
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance);
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance, cudaStream_t stream);
template <typename T>
void radiusMatchHamming_gpu(const DevMem2D& queryDescs, const DevMem2D& trainDescs, float maxDistance,
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance);
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance, cudaStream_t stream);
}}}
namespace
@@ -168,7 +168,7 @@ bool cv::gpu::BruteForceMatcher_GPU_base::isMaskSupported() const
// Match
void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat& queryDescs, const GpuMat& trainDescs,
GpuMat& trainIdx, GpuMat& distance, const GpuMat& mask)
GpuMat& trainIdx, GpuMat& distance, const GpuMat& mask, Stream& stream)
{
if (queryDescs.empty() || trainDescs.empty())
return;
@@ -177,7 +177,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat& queryDescs,
typedef void (*match_caller_t)(const DevMem2D& queryDescs, const DevMem2D& trainDescs,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx, const DevMem2Df& distance,
bool cc_12);
bool cc_12, cudaStream_t stream);
static const match_caller_t match_callers[3][8] =
{
@@ -213,7 +213,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchSingle(const GpuMat& queryDescs,
// For single train there is no need to save imgIdx, so we just save imgIdx to trainIdx.
// trainIdx store after imgIdx, so we doesn't lose it value.
func(queryDescs, trainDescs, mask, trainIdx, trainIdx, distance, cc_12);
func(queryDescs, trainDescs, mask, trainIdx, trainIdx, distance, cc_12, StreamAccessor::getStream(stream));
}
void cv::gpu::BruteForceMatcher_GPU_base::matchDownload(const GpuMat& trainIdx, const GpuMat& distance,
@@ -301,7 +301,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::makeGpuCollection(GpuMat& trainCollect
}
void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat& queryDescs, const GpuMat& trainCollection,
GpuMat& trainIdx, GpuMat& imgIdx, GpuMat& distance, const GpuMat& maskCollection)
GpuMat& trainIdx, GpuMat& imgIdx, GpuMat& distance, const GpuMat& maskCollection, Stream& stream)
{
if (queryDescs.empty() || trainCollection.empty())
return;
@@ -310,7 +310,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat& queryDes
typedef void (*match_caller_t)(const DevMem2D& queryDescs, const DevMem2D& trainCollection,
const DevMem2D_<PtrStep>& maskCollection, const DevMem2Di& trainIdx, const DevMem2Di& imgIdx,
const DevMem2Df& distance, bool cc_12);
const DevMem2Df& distance, bool cc_12, cudaStream_t stream);
static const match_caller_t match_callers[3][8] =
{
@@ -344,7 +344,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::matchCollection(const GpuMat& queryDes
bool cc_12 = TargetArchs::builtWith(FEATURE_SET_COMPUTE_12) && DeviceInfo().supports(FEATURE_SET_COMPUTE_12);
func(queryDescs, trainCollection, maskCollection, trainIdx, imgIdx, distance, cc_12);
func(queryDescs, trainCollection, maskCollection, trainIdx, imgIdx, distance, cc_12, StreamAccessor::getStream(stream));
}
void cv::gpu::BruteForceMatcher_GPU_base::matchDownload(const GpuMat& trainIdx, const GpuMat& imgIdx,
@@ -403,7 +403,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::match(const GpuMat& queryDescs, vector
// KnnMatch
void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat& queryDescs, const GpuMat& trainDescs,
GpuMat& trainIdx, GpuMat& distance, GpuMat& allDist, int k, const GpuMat& mask)
GpuMat& trainIdx, GpuMat& distance, GpuMat& allDist, int k, const GpuMat& mask, Stream& stream)
{
if (queryDescs.empty() || trainDescs.empty())
return;
@@ -411,7 +411,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat& queryDescs, con
using namespace cv::gpu::bfmatcher;
typedef void (*match_caller_t)(const DevMem2D& queryDescs, const DevMem2D& trainDescs, int knn,
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist);
const DevMem2D& mask, const DevMem2Di& trainIdx, const DevMem2Df& distance, const DevMem2Df& allDist, cudaStream_t stream);
static const match_caller_t match_callers[3][8] =
{
@@ -436,16 +436,24 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat& queryDescs, con
const int nTrain = trainDescs.rows;
trainIdx.create(nQuery, k, CV_32S);
trainIdx.setTo(Scalar::all(-1));
distance.create(nQuery, k, CV_32F);
ensureSizeIsEnough(nQuery, nTrain, CV_32FC1, allDist);
allDist.setTo(Scalar::all(numeric_limits<float>::max()));
if (stream)
{
stream.enqueueMemSet(trainIdx, Scalar::all(-1));
stream.enqueueMemSet(allDist, Scalar::all(numeric_limits<float>::max()));
}
else
{
trainIdx.setTo(Scalar::all(-1));
allDist.setTo(Scalar::all(numeric_limits<float>::max()));
}
match_caller_t func = match_callers[distType][queryDescs.depth()];
CV_Assert(func != 0);
func(queryDescs, trainDescs, k, mask, trainIdx, distance, allDist);
func(queryDescs, trainDescs, k, mask, trainIdx, distance, allDist, StreamAccessor::getStream(stream));
}
void cv::gpu::BruteForceMatcher_GPU_base::knnMatchDownload(const GpuMat& trainIdx, const GpuMat& distance,
@@ -547,7 +555,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::knnMatch(const GpuMat& queryDescs,
// RadiusMatch
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat& queryDescs, const GpuMat& trainDescs,
GpuMat& trainIdx, GpuMat& nMatches, GpuMat& distance, float maxDistance, const GpuMat& mask)
GpuMat& trainIdx, GpuMat& nMatches, GpuMat& distance, float maxDistance, const GpuMat& mask, Stream& stream)
{
if (queryDescs.empty() || trainDescs.empty())
return;
@@ -555,7 +563,7 @@ void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat& queryDescs,
using namespace cv::gpu::bfmatcher;
typedef void (*radiusMatch_caller_t)(const DevMem2D& queryDescs, const DevMem2D& trainDescs, float maxDistance,
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance);
const DevMem2D& mask, const DevMem2Di& trainIdx, unsigned int* nMatches, const DevMem2Df& distance, cudaStream_t stream);
static const radiusMatch_caller_t radiusMatch_callers[3][8] =
{
@@ -583,17 +591,21 @@ void cv::gpu::BruteForceMatcher_GPU_base::radiusMatch(const GpuMat& queryDescs,
CV_Assert(trainIdx.empty() || (trainIdx.rows == nQuery && trainIdx.size() == distance.size()));
ensureSizeIsEnough(1, nQuery, CV_32SC1, nMatches);
nMatches.setTo(Scalar::all(0));
if (trainIdx.empty())
{
trainIdx.create(nQuery, nTrain, CV_32SC1);
distance.create(nQuery, nTrain, CV_32FC1);
}
if (stream)
stream.enqueueMemSet(nMatches, Scalar::all(0));
else
nMatches.setTo(Scalar::all(0));
radiusMatch_caller_t func = radiusMatch_callers[distType][queryDescs.depth()];
CV_Assert(func != 0);
func(queryDescs, trainDescs, maxDistance, mask, trainIdx, nMatches.ptr<unsigned int>(), distance);
func(queryDescs, trainDescs, maxDistance, mask, trainIdx, nMatches.ptr<unsigned int>(), distance, StreamAccessor::getStream(stream));
}
void cv::gpu::BruteForceMatcher_GPU_base::radiusMatchDownload(const GpuMat& trainIdx, const GpuMat& nMatches,