diff --git a/modules/core/include/opencv2/core/mat.hpp b/modules/core/include/opencv2/core/mat.hpp
index 2cf1e35dc5..f864510de6 100644
--- a/modules/core/include/opencv2/core/mat.hpp
+++ b/modules/core/include/opencv2/core/mat.hpp
@@ -283,7 +283,6 @@ public:
                                void* data, size_t* step, int flags) const = 0;
     virtual bool allocate(UMatData* data, int accessflags) const = 0;
     virtual void deallocate(UMatData* data) const = 0;
-    virtual void sync(UMatData* u) const;
     virtual void map(UMatData* data, int accessflags) const;
     virtual void unmap(UMatData* data) const;
     virtual void download(UMatData* data, void* dst, int dims, const size_t sz[],
diff --git a/modules/core/src/matrix.cpp b/modules/core/src/matrix.cpp
index 07da4ad724..919452dc98 100644
--- a/modules/core/src/matrix.cpp
+++ b/modules/core/src/matrix.cpp
@@ -48,10 +48,6 @@
 
 namespace cv {
 
-void MatAllocator::sync(UMatData*) const
-{
-}
-
 void MatAllocator::map(UMatData*, int) const
 {
 }
diff --git a/modules/core/src/ocl.cpp b/modules/core/src/ocl.cpp
index 1616267bee..64460efb0d 100644
--- a/modules/core/src/ocl.cpp
+++ b/modules/core/src/ocl.cpp
@@ -2268,8 +2268,8 @@ bool Kernel::run(int dims, size_t _globalsize[], size_t _localsize[],
     CV_Assert(_globalsize != 0);
     for (int i = 0; i < dims; i++)
     {
-        size_t val = _localsize ? _localsize[i] : 1;
-            //dims == 1 ? 64 : dims == 2 ? (16>>i) : dims == 3 ? (8>>(int)(i>0)) : 1;
+        size_t val = _localsize ? _localsize[i] :
+            dims == 1 ? 64 : dims == 2 ? (16>>i) : dims == 3 ? (8>>(int)(i>0)) : 1;
         CV_Assert( val > 0 );
         total *= _globalsize[i];
         globalsize[i] = ((_globalsize[i] + val - 1)/val)*val;
@@ -2721,7 +2721,7 @@ public:
         return true;
     }
 
-    void sync(UMatData* u) const
+    /*void sync(UMatData* u) const
     {
         cl_command_queue q = (cl_command_queue)Queue::getDefault().ptr();
         UMatDataAutoLock lock(u);
@@ -2749,19 +2749,19 @@ public:
             clEnqueueWriteBuffer(q, (cl_mem)u->handle, CL_TRUE, 0,
                                  u->size, u->data, 0, 0, 0);
         }
-    }
+    }*/
 
     void deallocate(UMatData* u) const
     {
         if(!u)
             return;
-        UMatDataAutoLock lock(u);
 
         // TODO: !!! when we add Shared Virtual Memory Support,
         // this function (as well as the others) should be corrected
         CV_Assert(u->handle != 0 && u->urefcount == 0);
         if(u->tempUMat())
         {
+            UMatDataAutoLock lock(u);
             if( u->hostCopyObsolete() && u->refcount > 0 )
             {
                 cl_command_queue q = (cl_command_queue)Queue::getDefault().ptr();
@@ -2792,6 +2792,7 @@ public:
         }
         else
         {
+            CV_Assert(u->refcount == 0);
             if(u->data && u->copyOnMap() && !(u->flags & UMatData::USER_ALLOCATED))
             {
                 fastFree(u->data);
diff --git a/modules/core/src/umatrix.cpp b/modules/core/src/umatrix.cpp
index de8cd07be0..e8144cd646 100644
--- a/modules/core/src/umatrix.cpp
+++ b/modules/core/src/umatrix.cpp
@@ -286,8 +286,6 @@ void UMat::copySize(const UMat& m)
 
 UMat::~UMat()
 {
-    //if( u && u->refcount > 0 )
-    //    u->currAllocator->sync(u);
     release();
     if( step.p != step.buf )
         fastFree(step.p);
diff --git a/modules/ocl/test/test_api.cpp b/modules/ocl/test/test_api.cpp
index 68888a7c08..a54ebb7cdf 100644
--- a/modules/ocl/test/test_api.cpp
+++ b/modules/ocl/test/test_api.cpp
@@ -83,12 +83,8 @@ TEST(TestAPI, openCLExecuteKernelInterop)
 TEST(OCL_TestTAPI, performance)
 {
     cv::RNG rng;
-#if 1
     cv::Mat src(1280,768,CV_8UC4), dst;
     rng.fill(src, RNG::UNIFORM, 0, 255);
-#else
-    cv::Mat src = cv::imread("/Users/vp/work/opencv/samples/c/lena.jpg", 1), dst;
-#endif
 
     cv::UMat usrc, udst;
     src.copyTo(usrc);
@@ -130,9 +126,4 @@ TEST(OCL_TestTAPI, performance)
     }
     t = (double)cv::getTickCount() - t;
     printf("cpu exec time = %gms per iter\n", t*1000./niters/cv::getTickFrequency());
-    /*cv::imshow("result0", dst);
-    cv::imshow("result1", udst);
-    cv::waitKey();
-    cv::destroyWindow("result0");
-    cv::destroyWindow("result1");*/
 }