Merge pull request opencv#8404 from khnaba:stream-with-custom-allocator

vpisarev · vpisarev · commit 8abd16346459 · 2017-03-21T20:06:56.000Z
diff --git a/modules/core/include/opencv2/core/cuda.hpp b/modules/core/include/opencv2/core/cuda.hpp
@@ -507,6 +507,9 @@ class CV_EXPORTS Stream
     //! 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;
diff --git a/modules/core/src/cuda_stream.cpp b/modules/core/src/cuda_stream.cpp
@@ -282,9 +282,10 @@ class cv::cuda::Stream::Impl
     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

Original file line number	Diff line number	Diff line change
`@@ -282,9 +282,10 @@ class cv::cuda::Stream::Impl`
`282`	`282`	`cudaStream_t stream;`
`283`	`283`	`bool ownStream;`
`284`	`284`
`285`		`- Ptr<StackAllocator> stackAllocator;`
	`285`	`+ Ptr<GpuMat::Allocator> allocator;`
`286`	`286`
`287`	`287`	`Impl();`
	`288`	`+ Impl(const Ptr<GpuMat::Allocator>& allocator);`
`288`	`289`	`explicit Impl(cudaStream_t stream);`
`289`	`290`
`290`	`291`	`~Impl();`
`@@ -295,17 +296,23 @@ cv::cuda::Stream::Impl::Impl() : stream(0), ownStream(false)`
`295`	`296`	`cudaSafeCall( cudaStreamCreate(&stream) );`
`296`	`297`	`ownStream = true;`
`297`	`298`
`298`		`- stackAllocator = makePtr<StackAllocator>(stream);`
	`299`	`+ allocator = makePtr<StackAllocator>(stream);`
	`300`	`+}`
	`301`	`+`
	`302`	`+cv::cuda::Stream::Impl::Impl(const Ptr<GpuMat::Allocator>& allocator) : stream(0), ownStream(false), allocator(allocator)`
	`303`	`+{`
	`304`	`+ cudaSafeCall( cudaStreamCreate(&stream) );`
	`305`	`+ ownStream = true;`
`299`	`306`	`}`
`300`	`307`
`301`	`308`	`cv::cuda::Stream::Impl::Impl(cudaStream_t stream_) : stream(stream_), ownStream(false)`
`302`	`309`	`{`
`303`		`- stackAllocator = makePtr<StackAllocator>(stream);`
	`310`	`+ allocator = makePtr<StackAllocator>(stream);`
`304`	`311`	`}`
`305`	`312`
`306`	`313`	`cv::cuda::Stream::Impl::~Impl()`
`307`	`314`	`{`
`308`		`- stackAllocator.release();`
	`315`	`+ allocator.release();`
`309`	`316`
`310`	`317`	`if (stream && ownStream)`
`311`	`318`	`{`
`@@ -417,6 +424,16 @@ cv::cuda::Stream::Stream()`
`417`	`424`	`#endif`
`418`	`425`	`}`
`419`	`426`
	`427`	`+cv::cuda::Stream::Stream(const Ptr<GpuMat::Allocator>& allocator)`
	`428`	`+{`
	`429`	`+#ifndef HAVE_CUDA`
	`430`	`+ (void) allocator;`
	`431`	`+ throw_no_cuda();`
	`432`	`+#else`
	`433`	`+ impl_ = makePtr<Impl>(allocator);`
	`434`	`+#endif`
	`435`	`+}`
	`436`	`+`
`420`	`437`	`bool cv::cuda::Stream::queryIfComplete() const`
`421`	`438`	`{`
`422`	`439`	`#ifndef HAVE_CUDA`
`@@ -675,7 +692,7 @@ cv::cuda::BufferPool::BufferPool(Stream& stream)`
`675`	`692`	`throw_no_cuda();`
`676`	`693`	`}`
`677`	`694`	`#else`
`678`		`-cv::cuda::BufferPool::BufferPool(Stream& stream) : allocator_(stream.impl_->stackAllocator)`
	`695`	`+cv::cuda::BufferPool::BufferPool(Stream& stream) : allocator_(stream.impl_->allocator)`
`679`	`696`	`{`
`680`	`697`	`}`
`681`	`698`	`#endif`