Support for creating streams with custom allocator

modules/core/include/opencv2/core/cuda.hpp

@@ -507,6 +507,9 @@ public:
     //! creates a new asynchronous stream
     Stream();
 
+    //! creates a new asynchronous stream with custom allocator
+    Stream(const Ptr<GpuMat::Allocator>& allocator);
+
     /** @brief Returns true if the current stream queue is finished. Otherwise, it returns false.
     */
     bool queryIfComplete() const;

modules/core/src/cuda_stream.cpp

@@ -282,9 +282,10 @@ public:
     cudaStream_t stream;
     bool ownStream;
 
-    Ptr<StackAllocator> stackAllocator;
+    Ptr<GpuMat::Allocator> allocator;
 
     Impl();
+    Impl(const Ptr<GpuMat::Allocator>& allocator);
     explicit Impl(cudaStream_t stream);
 
     ~Impl();
@@ -295,17 +296,23 @@ cv::cuda::Stream::Impl::Impl() : stream(0), ownStream(false)
     cudaSafeCall( cudaStreamCreate(&stream) );
     ownStream = true;
 
-    stackAllocator = makePtr<StackAllocator>(stream);
+    allocator = makePtr<StackAllocator>(stream);
+}
+
+cv::cuda::Stream::Impl::Impl(const Ptr<GpuMat::Allocator>& allocator) : stream(0), ownStream(false), allocator(allocator)
+{
+    cudaSafeCall( cudaStreamCreate(&stream) );
+    ownStream = true;
 }
 
 cv::cuda::Stream::Impl::Impl(cudaStream_t stream_) : stream(stream_), ownStream(false)
 {
-    stackAllocator = makePtr<StackAllocator>(stream);
+    allocator = makePtr<StackAllocator>(stream);
 }
 
 cv::cuda::Stream::Impl::~Impl()
 {
-    stackAllocator.release();
+    allocator.release();
 
     if (stream && ownStream)
     {
@@ -417,6 +424,16 @@ cv::cuda::Stream::Stream()
 #endif
 }
 
+cv::cuda::Stream::Stream(const Ptr<GpuMat::Allocator>& allocator)
+{
+#ifndef HAVE_CUDA
+    (void) allocator;
+    throw_no_cuda();
+#else
+    impl_ = makePtr<Impl>(allocator);
+#endif
+}
+
 bool cv::cuda::Stream::queryIfComplete() const
 {
 #ifndef HAVE_CUDA
@@ -675,7 +692,7 @@ cv::cuda::BufferPool::BufferPool(Stream& stream)
     throw_no_cuda();
 }
 #else
-cv::cuda::BufferPool::BufferPool(Stream& stream) : allocator_(stream.impl_->stackAllocator)
+cv::cuda::BufferPool::BufferPool(Stream& stream) : allocator_(stream.impl_->allocator)
 {
 }
 #endif