the first draft of transparent API and new UMat class. more files

modules/core/include/opencv2/core.hpp

@@ -158,6 +158,9 @@ enum { REDUCE_SUM = 0,
 //! swaps two matrices
 CV_EXPORTS void swap(Mat& a, Mat& b);
 
+//! swaps two umatrices
+CV_EXPORTS void swap( UMat& a, UMat& b );
+
 //! 1D interpolation function: returns coordinate of the "donor" pixel for the specified location p.
 CV_EXPORTS_W int borderInterpolate(int p, int len, int borderType);
 
@@ -439,7 +442,7 @@ CV_EXPORTS void calcCovarMatrix( const Mat* samples, int nsamples, Mat& covar, M
 
 //! computes covariation matrix of a set of samples
 CV_EXPORTS_W void calcCovarMatrix( InputArray samples, OutputArray covar,
-                                   OutputArray mean, int flags, int ctype = CV_64F);
+                                   InputOutputArray mean, int flags, int ctype = CV_64F);
 
 CV_EXPORTS_W void PCACompute(InputArray data, InputOutputArray mean,
                              OutputArray eigenvectors, int maxComponents = 0);

modules/core/include/opencv2/core/base.hpp

@@ -472,6 +472,9 @@ class CV_EXPORTS RNG;
 class CV_EXPORTS Mat;
 class CV_EXPORTS MatExpr;
 
+class CV_EXPORTS UMat;
+class CV_EXPORTS UMatExpr;
+
 class CV_EXPORTS SparseMat;
 typedef Mat MatND;
 

modules/core/src/convert.cpp

@@ -271,7 +271,7 @@ void cv::split(InputArray _m, OutputArrayOfArrays _mv)
         _mv.release();
         return;
     }
-    CV_Assert( !_mv.fixedType() || CV_MAT_TYPE(_mv.flags) == m.depth() );
+    CV_Assert( !_mv.fixedType() || _mv.type() == m.depth() );
     _mv.create(m.channels(), 1, m.depth());
     Mat* dst = &_mv.getMatRef(0);
     split(m, dst);

modules/core/src/matop.cpp

@@ -1610,7 +1610,7 @@ MatExpr Mat::mul(InputArray m, double scale) const
     MatExpr e;
     if(m.kind() == _InputArray::EXPR)
     {
-        const MatExpr& me = *(const MatExpr*)m.obj;
+        const MatExpr& me = *(const MatExpr*)m.getObj();
         me.op->multiply(MatExpr(*this), me, e, scale);
     }
     else

modules/core/src/precomp.hpp

@@ -50,6 +50,7 @@
 
 #include "opencv2/core/private.hpp"
 #include "opencv2/core/private.cuda.hpp"
+#include "opencv2/core/ocl.hpp"
 
 #include <assert.h>
 #include <ctype.h>
@@ -215,6 +216,19 @@ inline bool checkScalar(const Mat& sc, int atype, int sckind, int akind)
 
 void convertAndUnrollScalar( const Mat& sc, int buftype, uchar* scbuf, size_t blocksize );
 
+struct TLSData
+{
+    TLSData();
+    RNG rng;
+    int device;
+    ocl::Queue oclQueue;
+    int useOpenCL; // 1 - use, 0 - do not use, -1 - auto/not initialized
+
+    static TLSData* get();
+};
+
+namespace ocl { MatAllocator* getOpenCLAllocator(); }
+
 }
 
 #endif /*_CXCORE_INTERNAL_H_*/

modules/core/src/rand.cpp

@@ -727,85 +727,11 @@ void RNG::fill( InputOutputArray _mat, int disttype,
     }
 }
 
-#ifdef WIN32
-
-
-#ifdef HAVE_WINRT
-// using C++11 thread attribute for local thread data
-__declspec( thread ) RNG* rng = NULL;
-
- void deleteThreadRNGData()
- {
-    if (rng)
-        delete rng;
-}
-
-RNG& theRNG()
-{
-    if (!rng)
-    {
-        rng =  new RNG;
-    }
-    return *rng;
-}
-#else
-#ifdef WINCE
-#	define TLS_OUT_OF_INDEXES ((DWORD)0xFFFFFFFF)
-#endif
-static DWORD tlsRNGKey = TLS_OUT_OF_INDEXES;
-
- void deleteThreadRNGData()
- {
-     if( tlsRNGKey != TLS_OUT_OF_INDEXES )
-         delete (RNG*)TlsGetValue( tlsRNGKey );
 }
 
-RNG& theRNG()
+cv::RNG& cv::theRNG()
 {
-    if( tlsRNGKey == TLS_OUT_OF_INDEXES )
-    {
-       tlsRNGKey = TlsAlloc();
-       CV_Assert(tlsRNGKey != TLS_OUT_OF_INDEXES);
-    }
-    RNG* rng = (RNG*)TlsGetValue( tlsRNGKey );
-    if( !rng )
-    {
-       rng = new RNG;
-       TlsSetValue( tlsRNGKey, rng );
-    }
-    return *rng;
-}
-#endif //HAVE_WINRT
-#else
-
-static pthread_key_t tlsRNGKey = 0;
-static pthread_once_t tlsRNGKeyOnce = PTHREAD_ONCE_INIT;
-
-static void deleteRNG(void* data)
-{
-    delete (RNG*)data;
-}
-
-static void makeRNGKey()
-{
-    int errcode = pthread_key_create(&tlsRNGKey, deleteRNG);
-    CV_Assert(errcode == 0);
-}
-
-RNG& theRNG()
-{
-    pthread_once(&tlsRNGKeyOnce, makeRNGKey);
-    RNG* rng = (RNG*)pthread_getspecific(tlsRNGKey);
-    if( !rng )
-    {
-        rng = new RNG;
-        pthread_setspecific(tlsRNGKey, rng);
-    }
-    return *rng;
-}
-
-#endif
-
+    return TLSData::get()->rng;
 }
 
 void cv::randu(InputOutputArray dst, InputArray low, InputArray high)

modules/core/src/system.cpp

@@ -830,4 +830,92 @@ bool Mutex::trylock() { return impl->trylock(); }
 
 }
 
+//////////////////////////////// thread-local storage ////////////////////////////////
+
+namespace cv
+{
+
+TLSData::TLSData()
+{
+    device = 0;
+    useOpenCL = -1;
+}
+
+#ifdef WIN32
+
+#ifdef HAVE_WINRT
+    // using C++11 thread attribute for local thread data
+    static __declspec( thread ) TLSData* g_tlsdata = NULL;
+
+    static void deleteThreadRNGData()
+    {
+        if (g_tlsdata)
+            delete g_tlsdata;
+    }
+
+    TLSData* TLSData::get()
+    {
+        if (!g_tlsdata)
+        {
+            g_tlsdata = new TLSData;
+        }
+        return g_tlsdata;
+    }
+#else
+#ifdef WINCE
+#   define TLS_OUT_OF_INDEXES ((DWORD)0xFFFFFFFF)
+#endif
+    static DWORD tlsKey = TLS_OUT_OF_INDEXES;
+
+    void deleteThreadData()
+    {
+        if( tlsKey != TLS_OUT_OF_INDEXES )
+            delete (TLSData*)TlsGetValue( tlsKey );
+    }
+
+    TLSData* TLSData::get()
+    {
+        if( tlsKey == TLS_OUT_OF_INDEXES )
+        {
+            tlsRNGKey = TlsAlloc();
+            CV_Assert(tlsRNGKey != TLS_OUT_OF_INDEXES);
+        }
+        TLSData* d = (TLSData*)TlsGetValue( tlsKey );
+        if( !d )
+        {
+            d = new TLSData;
+            TlsSetValue( tlsRNGKey, d );
+        }
+        return d;
+    }
+#endif //HAVE_WINRT
+#else
+    static pthread_key_t tlsKey = 0;
+    static pthread_once_t tlsKeyOnce = PTHREAD_ONCE_INIT;
+
+    static void deleteTLSData(void* data)
+    {
+        delete (TLSData*)data;
+    }
+
+    static void makeKey()
+    {
+        int errcode = pthread_key_create(&tlsKey, deleteTLSData);
+        CV_Assert(errcode == 0);
+    }
+
+    TLSData* TLSData::get()
+    {
+        pthread_once(&tlsKeyOnce, makeKey);
+        TLSData* d = (TLSData*)pthread_getspecific(tlsKey);
+        if( !d )
+        {
+            d = new TLSData;
+            pthread_setspecific(tlsKey, d);
+        }
+        return d;
+    }
+#endif
+}
+
 /* End of file. */

modules/imgproc/perf/perf_histogram.cpp

@@ -9,7 +9,7 @@ using std::tr1::get;
 
 typedef tr1::tuple<Size, MatType> Size_Source_t;
 typedef TestBaseWithParam<Size_Source_t> Size_Source;
-typedef TestBaseWithParam<Size> MatSize;
+typedef TestBaseWithParam<Size> TestMatSize;
 
 static const float rangeHight = 256.0f;
 static const float rangeLow = 0.0f;
@@ -99,7 +99,7 @@ PERF_TEST_P(Size_Source, calcHist3d,
     SANITY_CHECK(hist);
 }
 
-PERF_TEST_P(MatSize, equalizeHist,
+PERF_TEST_P(TestMatSize, equalizeHist,
             testing::Values(TYPICAL_MAT_SIZES)
             )
 {

modules/legacy/src/em.cpp

@@ -102,7 +102,8 @@ float
 CvEM::predict( const CvMat* _sample, CvMat* _probs ) const
 {
     Mat prbs0 = cvarrToMat(_probs), prbs = prbs0, sample = cvarrToMat(_sample);
-    int cls = static_cast<int>(emObj.predict(sample, _probs ? _OutputArray(prbs) : cv::noArray())[1]);
+    int cls = static_cast<int>(emObj.predict(sample, _probs ? _OutputArray(prbs) :
+                                             (OutputArray)cv::noArray())[1]);
     if(_probs)
     {
         if( prbs.data != prbs0.data )
@@ -208,13 +209,16 @@ bool CvEM::train( const Mat& _samples, const Mat& _sample_idx,
     bool isOk = false;
     if( _params.start_step == EM::START_AUTO_STEP )
         isOk = emObj.train(_samples,
-                           logLikelihoods, _labels ? _OutputArray(*_labels) : cv::noArray(), probs);
+                           logLikelihoods, _labels ? _OutputArray(*_labels) :
+                           (OutputArray)cv::noArray(), probs);
     else if( _params.start_step == EM::START_E_STEP )
         isOk = emObj.trainE(_samples, means, covshdrs, weights,
-                            logLikelihoods, _labels ? _OutputArray(*_labels) : cv::noArray(), probs);
+                            logLikelihoods, _labels ? _OutputArray(*_labels) :
+                            (OutputArray)cv::noArray(), probs);
     else if( _params.start_step == EM::START_M_STEP )
         isOk = emObj.trainM(_samples, prbs,
-                            logLikelihoods, _labels ? _OutputArray(*_labels) : cv::noArray(), probs);
+                            logLikelihoods, _labels ? _OutputArray(*_labels) :
+                            (OutputArray)cv::noArray(), probs);
     else
         CV_Error(CV_StsBadArg, "Bad start type of EM algorithm");
 
@@ -230,7 +234,9 @@ bool CvEM::train( const Mat& _samples, const Mat& _sample_idx,
 float
 CvEM::predict( const Mat& _sample, Mat* _probs ) const
 {
-    return static_cast<float>(emObj.predict(_sample, _probs ? _OutputArray(*_probs) : cv::noArray())[1]);
+    return static_cast<float>(emObj.predict(_sample, _probs ?
+                                            _OutputArray(*_probs) :
+                                            (OutputArray)cv::noArray())[1]);
 }
 
 int CvEM::getNClusters() const

modules/legacy/src/features2d.cpp

@@ -82,7 +82,7 @@ cvExtractSURF( const CvArr* _img, const CvArr* _mask,
     surf->set("upright", params.upright != 0);
     surf->set("extended", params.extended != 0);
 
-    surf->operator()(img, mask, kpt, _descriptors ? _OutputArray(descr) : noArray(),
+    surf->operator()(img, mask, kpt, _descriptors ? _OutputArray(descr) : (OutputArray)noArray(),
                      useProvidedKeyPts != 0);
 
     if( _keypoints )

modules/ocl/src/matrix_operations.cpp

@@ -154,30 +154,24 @@ void cv::ocl::oclMat::upload(const Mat &m)
 
 cv::ocl::oclMat::operator cv::_InputArray()
 {
-    _InputArray newInputArray;
-    newInputArray.flags = cv::_InputArray::OCL_MAT;
-    newInputArray.obj   = reinterpret_cast<void *>(this);
-    return newInputArray;
+    return _InputArray(cv::_InputArray::OCL_MAT, this);
 }
 
 cv::ocl::oclMat::operator cv::_OutputArray()
 {
-    _OutputArray newOutputArray;
-    newOutputArray.flags = cv::_InputArray::OCL_MAT;
-    newOutputArray.obj   = reinterpret_cast<void *>(this);
-    return newOutputArray;
+    return _OutputArray(cv::_InputArray::OCL_MAT, this);
 }
 
 cv::ocl::oclMat& cv::ocl::getOclMatRef(InputArray src)
 {
-    CV_Assert(src.flags & cv::_InputArray::OCL_MAT);
-    return *reinterpret_cast<oclMat*>(src.obj);
+    CV_Assert(src.kind() == cv::_InputArray::OCL_MAT);
+    return *(oclMat*)src.getObj();
 }
 
 cv::ocl::oclMat& cv::ocl::getOclMatRef(OutputArray src)
 {
-    CV_Assert(src.flags & cv::_InputArray::OCL_MAT);
-    return *reinterpret_cast<oclMat*>(src.obj);
+    CV_Assert(src.kind() == cv::_InputArray::OCL_MAT);
+    return *(oclMat*)src.getObj();
 }
 
 void cv::ocl::oclMat::download(cv::Mat &m) const

modules/superres/src/optical_flow.cpp

@@ -163,7 +163,9 @@ namespace
 
     void Farneback::impl(const Mat& input0, const Mat& input1, OutputArray dst)
     {
-        calcOpticalFlowFarneback(input0, input1, dst, pyrScale_, numLevels_, winSize_, numIters_, polyN_, polySigma_, flags_);
+        calcOpticalFlowFarneback(input0, input1, (InputOutputArray)dst, pyrScale_,
+                                 numLevels_, winSize_, numIters_,
+                                 polyN_, polySigma_, flags_);
     }
 }
 
@@ -325,7 +327,7 @@ namespace
         alg_->set("iterations", iterations_);
         alg_->set("useInitialFlow", useInitialFlow_);
 
-        alg_->calc(input0, input1, dst);
+        alg_->calc(input0, input1, (InputOutputArray)dst);
     }
 
     void DualTVL1::collectGarbage()

modules/video/src/compat_video.cpp

@@ -352,7 +352,7 @@ cvCalcOpticalFlowPyrLK( const void* arrA, const void* arrB,
     if( error )
         err = cv::Mat(count, 1, CV_32F, (void*)error);
     cv::calcOpticalFlowPyrLK( A, B, ptA, ptB, st,
-                              error ? cv::_OutputArray(err) : cv::noArray(),
+                              error ? cv::_OutputArray(err) : (cv::_OutputArray)cv::noArray(),
                               winSize, level, criteria, flags);
 }
 

modules/video/src/optflowgf.cpp

@@ -564,7 +564,7 @@ FarnebackUpdateFlow_GaussianBlur( const Mat& _R0, const Mat& _R1,
 }
 
 void cv::calcOpticalFlowFarneback( InputArray _prev0, InputArray _next0,
-                               OutputArray _flow0, double pyr_scale, int levels, int winsize,
+                               InputOutputArray _flow0, double pyr_scale, int levels, int winsize,
                                int iterations, int poly_n, double poly_sigma, int flags )
 {
     Mat prev0 = _prev0.getMat(), next0 = _next0.getMat();