Merge remote-tracking branch 'upstream/3.4' into merge-3.4

modules/core/include/opencv2/core/private.hpp

@@ -141,13 +141,20 @@ namespace cv
 {
 CV_EXPORTS void scalarToRawData(const cv::Scalar& s, void* buf, int type, int unroll_to = 0);
 
-//! Allocate all memory buffers which will not be freed, ease filtering memcheck issues
+//! Allocate memory buffers which will not be freed, ease filtering memcheck issues. Uses fastMalloc() call.
 CV_EXPORTS void* allocSingletonBuffer(size_t size);
 
-//! Allocate all memory buffers which will not be freed, ease filtering memcheck issues
+//! Allocate memory buffers which will not be freed, ease filtering memcheck issues. Uses fastMalloc() call
 template <typename T> static inline
 T* allocSingleton(size_t count = 1) { return static_cast<T*>(allocSingletonBuffer(sizeof(T) * count)); }
 
+//! Allocate memory buffers which will not be freed, ease filtering memcheck issues. Uses generic malloc() call.
+CV_EXPORTS void* allocSingletonNewBuffer(size_t size);
+
+//! Allocate memory buffers which will not be freed, ease filtering memcheck issues.  Uses generic malloc() call.
+template <typename T> static inline
+T* allocSingletonNew() { return new(allocSingletonNewBuffer(sizeof(T))) T(); }
+
 } // namespace
 
 #if 1 // TODO: Remove in OpenCV 4.x

modules/core/include/opencv2/core/utils/allocator_stats.hpp

@@ -0,0 +1,29 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#ifndef OPENCV_CORE_ALLOCATOR_STATS_HPP
+#define OPENCV_CORE_ALLOCATOR_STATS_HPP
+
+#include "../cvdef.h"
+
+namespace cv { namespace utils {
+
+class AllocatorStatisticsInterface
+{
+protected:
+    AllocatorStatisticsInterface() {}
+    virtual ~AllocatorStatisticsInterface() {}
+public:
+    virtual uint64_t getCurrentUsage() const = 0;
+    virtual uint64_t getTotalUsage() const = 0;
+    virtual uint64_t getNumberOfAllocations() const = 0;
+    virtual uint64_t getPeakUsage() const = 0;
+
+    /** set peak usage = current usage */
+    virtual void resetPeakUsage() = 0;
+};
+
+}} // namespace
+
+#endif // OPENCV_CORE_ALLOCATOR_STATS_HPP

modules/core/include/opencv2/core/utils/allocator_stats.impl.hpp

@@ -0,0 +1,117 @@
+// This file is part of OpenCV project.
+// It is subject to the license terms in the LICENSE file found in the top-level directory
+// of this distribution and at http://opencv.org/license.html.
+
+#ifndef OPENCV_CORE_ALLOCATOR_STATS_IMPL_HPP
+#define OPENCV_CORE_ALLOCATOR_STATS_IMPL_HPP
+
+#include "./allocator_stats.hpp"
+
+#ifdef CV_CXX11
+#include <atomic>
+#endif
+
+namespace cv { namespace utils {
+
+#ifdef CV__ALLOCATOR_STATS_LOG
+namespace {
+#endif
+
+class AllocatorStatistics : public AllocatorStatisticsInterface
+{
+protected:
+#ifdef CV_CXX11
+    std::atomic<long long> curr, total, total_allocs, peak;
+#else
+    volatile long long curr, total, total_allocs, peak;  // overflow is possible, CV_XADD operates with 'int' only
+#endif
+
+public:
+    AllocatorStatistics()
+#ifndef CV_CXX11
+        : curr(0), total(0), total_allocs(0), peak(0)
+#endif
+    {}
+    ~AllocatorStatistics() CV_OVERRIDE {}
+
+    // AllocatorStatisticsInterface
+
+#ifdef CV_CXX11
+    uint64_t getCurrentUsage() const CV_OVERRIDE { return (uint64_t)curr.load(); }
+    uint64_t getTotalUsage() const CV_OVERRIDE { return (uint64_t)total.load(); }
+    uint64_t getNumberOfAllocations() const CV_OVERRIDE { return (uint64_t)total_allocs.load(); }
+    uint64_t getPeakUsage() const CV_OVERRIDE { return (uint64_t)peak.load(); }
+
+    /** set peak usage = current usage */
+    void resetPeakUsage() CV_OVERRIDE { peak.store(curr.load()); }
+
+    // Controller interface
+    void onAllocate(size_t sz)
+    {
+#ifdef CV__ALLOCATOR_STATS_LOG
+        CV__ALLOCATOR_STATS_LOG(cv::format("allocate: %lld (curr=%lld)", (long long int)sz, (long long int)curr.load()));
+#endif
+
+        long long new_curr = curr.fetch_add((long long)sz) + (long long)sz;
+
+        // peak = std::max((uint64_t)peak, new_curr);
+        auto prev_peak = peak.load();
+        while (prev_peak < new_curr)
+        {
+            if (peak.compare_exchange_weak(prev_peak, new_curr))
+                break;
+        }
+        // end of peak = max(...)
+
+        total += (long long)sz;
+        total_allocs++;
+    }
+    void onFree(size_t sz)
+    {
+#ifdef CV__ALLOCATOR_STATS_LOG
+        CV__ALLOCATOR_STATS_LOG(cv::format("free: %lld (curr=%lld)", (long long int)sz, (long long int)curr.load()));
+#endif
+        curr -= (long long)sz;
+    }
+
+#else
+    uint64_t getCurrentUsage() const CV_OVERRIDE { return (uint64_t)curr; }
+    uint64_t getTotalUsage() const CV_OVERRIDE { return (uint64_t)total; }
+    uint64_t getNumberOfAllocations() const CV_OVERRIDE { return (uint64_t)total_allocs; }
+    uint64_t getPeakUsage() const CV_OVERRIDE { return (uint64_t)peak; }
+
+    void resetPeakUsage() CV_OVERRIDE { peak = curr; }
+
+    // Controller interface
+    void onAllocate(size_t sz)
+    {
+#ifdef CV__ALLOCATOR_STATS_LOG
+        CV__ALLOCATOR_STATS_LOG(cv::format("allocate: %lld (curr=%lld)", (long long int)sz, (long long int)curr));
+#endif
+
+        uint64_t new_curr = (uint64_t)CV_XADD(&curr, (uint64_t)sz) + sz;
+
+        peak = std::max((uint64_t)peak, new_curr);  // non-thread safe
+
+        //CV_XADD(&total, (uint64_t)sz);  // overflow with int, non-reliable...
+        total += sz;
+
+        CV_XADD(&total_allocs, (uint64_t)1);
+    }
+    void onFree(size_t sz)
+    {
+#ifdef CV__ALLOCATOR_STATS_LOG
+        CV__ALLOCATOR_STATS_LOG(cv::format("free: %lld (curr=%lld)", (long long int)sz, (long long int)curr));
+#endif
+        CV_XADD(&curr, (uint64_t)-sz);
+    }
+#endif
+};
+
+#ifdef CV__ALLOCATOR_STATS_LOG
+} // namespace
+#endif
+
+}} // namespace
+
+#endif // OPENCV_CORE_ALLOCATOR_STATS_IMPL_HPP

modules/core/src/alloc.cpp

@@ -42,12 +42,29 @@
 
 #include "precomp.hpp"
 
+#include <opencv2/core/utils/logger.defines.hpp>
+#undef CV_LOG_STRIP_LEVEL
+#define CV_LOG_STRIP_LEVEL CV_LOG_LEVEL_VERBOSE + 1
+#include <opencv2/core/utils/logger.hpp>
+
+#define CV__ALLOCATOR_STATS_LOG(...) CV_LOG_VERBOSE(NULL, 0, "alloc.cpp: " << __VA_ARGS__)
+#include "opencv2/core/utils/allocator_stats.impl.hpp"
+#undef CV__ALLOCATOR_STATS_LOG
+
+//#define OPENCV_ALLOC_ENABLE_STATISTICS
+#define OPENCV_ALLOC_STATISTICS_LIMIT 4096  // don't track buffers less than N bytes
+
+
 #ifdef HAVE_POSIX_MEMALIGN
 #include <stdlib.h>
 #elif defined HAVE_MALLOC_H
 #include <malloc.h>
 #endif
 
+#ifdef OPENCV_ALLOC_ENABLE_STATISTICS
+#include <map>
+#endif
+
 namespace cv {
 
 static void* OutOfMemoryError(size_t size)
@@ -55,8 +72,21 @@ static void* OutOfMemoryError(size_t size)
     CV_Error_(CV_StsNoMem, ("Failed to allocate %llu bytes", (unsigned long long)size));
 }
 
+CV_EXPORTS cv::utils::AllocatorStatisticsInterface& getAllocatorStatistics();
 
-void* fastMalloc( size_t size )
+static cv::utils::AllocatorStatistics allocator_stats;
+
+cv::utils::AllocatorStatisticsInterface& getAllocatorStatistics()
+{
+    return allocator_stats;
+}
+
+#ifdef OPENCV_ALLOC_ENABLE_STATISTICS
+static inline
+void* fastMalloc_(size_t size)
+#else
+void* fastMalloc(size_t size)
+#endif
 {
 #ifdef HAVE_POSIX_MEMALIGN
     void* ptr = NULL;
@@ -80,7 +110,12 @@ void* fastMalloc( size_t size )
 #endif
 }
 
+#ifdef OPENCV_ALLOC_ENABLE_STATISTICS
+static inline
+void fastFree_(void* ptr)
+#else
 void fastFree(void* ptr)
+#endif
 {
 #if defined HAVE_POSIX_MEMALIGN || defined HAVE_MEMALIGN
     free(ptr);
@@ -95,6 +130,47 @@ void fastFree(void* ptr)
 #endif
 }
 
+#ifdef OPENCV_ALLOC_ENABLE_STATISTICS
+
+static
+Mutex& getAllocationStatisticsMutex()
+{
+    static Mutex* p_alloc_mutex = allocSingletonNew<Mutex>();
+    CV_Assert(p_alloc_mutex);
+    return *p_alloc_mutex;
+}
+
+static std::map<void*, size_t> allocated_buffers;  // guarded by getAllocationStatisticsMutex()
+
+void* fastMalloc(size_t size)
+{
+    void* res = fastMalloc_(size);
+    if (res && size >= OPENCV_ALLOC_STATISTICS_LIMIT)
+    {
+        cv::AutoLock lock(getAllocationStatisticsMutex());
+        allocated_buffers.insert(std::make_pair(res, size));
+        allocator_stats.onAllocate(size);
+    }
+    return res;
+}
+
+void fastFree(void* ptr)
+{
+    {
+        cv::AutoLock lock(getAllocationStatisticsMutex());
+        std::map<void*, size_t>::iterator i = allocated_buffers.find(ptr);
+        if (i != allocated_buffers.end())
+        {
+            size_t size = i->second;
+            allocator_stats.onFree(size);
+            allocated_buffers.erase(i);
+        }
+    }
+    fastFree_(ptr);
+}
+
+#endif // OPENCV_ALLOC_ENABLE_STATISTICS
+
 } // namespace
 
 CV_IMPL void* cvAlloc( size_t size )

modules/core/src/ocl.cpp

@@ -54,6 +54,9 @@
 
 #include <opencv2/core/utils/configuration.private.hpp>
 
+#include <opencv2/core/utils/logger.defines.hpp>
+#undef CV_LOG_STRIP_LEVEL
+#define CV_LOG_STRIP_LEVEL CV_LOG_LEVEL_DEBUG + 1
 #include <opencv2/core/utils/logger.hpp>
 
 #include "opencv2/core/ocl_genbase.hpp"
@@ -63,6 +66,10 @@
 #include "opencv2/core/utils/filesystem.hpp"
 #include "opencv2/core/utils/filesystem.private.hpp"
 
+#define CV__ALLOCATOR_STATS_LOG(...) CV_LOG_VERBOSE(NULL, 0, "OpenCL allocator: " << __VA_ARGS__)
+#include "opencv2/core/utils/allocator_stats.impl.hpp"
+#undef CV__ALLOCATOR_STATS_LOG
+
 #define CV_OPENCL_ALWAYS_SHOW_BUILD_LOG          0
 
 #define CV_OPENCL_SHOW_RUN_KERNELS               0
@@ -132,6 +139,14 @@ namespace cv { namespace ocl {
     void release() { if( CV_XADD(&refcount, -1) == 1 && !cv::__termination) delete this; } \
     int refcount
 
+static cv::utils::AllocatorStatistics opencl_allocator_stats;
+
+CV_EXPORTS cv::utils::AllocatorStatisticsInterface& getOpenCLAllocatorStatistics();
+cv::utils::AllocatorStatisticsInterface& getOpenCLAllocatorStatistics()
+{
+    return opencl_allocator_stats;
+}
+
 #ifndef HAVE_OPENCL
 #define CV_OPENCL_NO_SUPPORT() CV_Error(cv::Error::OpenCLApiCallError, "OpenCV build without OpenCL support")
 namespace {
@@ -4534,15 +4549,17 @@ class OpenCLAllocator CV_FINAL : public MatAllocator
     mutable OpenCLSVMBufferPoolImpl bufferPoolSVM;
 #endif
 
+public:
     enum AllocatorFlags
     {
         ALLOCATOR_FLAGS_BUFFER_POOL_USED = 1 << 0,
-        ALLOCATOR_FLAGS_BUFFER_POOL_HOST_PTR_USED = 1 << 1
+        ALLOCATOR_FLAGS_BUFFER_POOL_HOST_PTR_USED = 1 << 1,
 #ifdef HAVE_OPENCL_SVM
-        ,ALLOCATOR_FLAGS_BUFFER_POOL_SVM_USED = 1 << 2
+        ALLOCATOR_FLAGS_BUFFER_POOL_SVM_USED = 1 << 2,
 #endif
+        ALLOCATOR_FLAGS_EXTERNAL_BUFFER = 1 << 3  // convertFromBuffer()
     };
-public:
+
     OpenCLAllocator()
         : bufferPool(0),
           bufferPoolHostPtr(CL_MEM_ALLOC_HOST_PTR)
@@ -4648,6 +4665,7 @@ public:
         u->allocatorFlags_ = allocatorFlags;
         CV_DbgAssert(!u->tempUMat()); // for bufferPool.release() consistency in deallocate()
         u->markHostCopyObsolete(true);
+        opencl_allocator_stats.onAllocate(u->size);
         return u;
     }
 
@@ -4757,6 +4775,7 @@ public:
         }
         if (!!(accessFlags & ACCESS_WRITE))
             u->markHostCopyObsolete(true);
+        opencl_allocator_stats.onAllocate(u->size);
         return true;
     }
 
@@ -4809,6 +4828,13 @@ public:
 
     void deallocate_(UMatData* u) const
     {
+        CV_Assert(u);
+        CV_Assert(u->handle);
+        if ((u->allocatorFlags_ & ALLOCATOR_FLAGS_EXTERNAL_BUFFER) == 0)
+        {
+            opencl_allocator_stats.onFree(u->size);
+        }
+
 #ifdef _WIN32
         if (cv::__termination)  // process is not in consistent state (after ExitProcess call) and terminating
             return;             // avoid any OpenCL calls
@@ -5790,7 +5816,7 @@ void convertFromBuffer(void* cl_mem_buffer, size_t step, int rows, int cols, int
     // attach clBuffer to UMatData
     dst.u = new UMatData(getOpenCLAllocator());
     dst.u->data            = 0;
-    dst.u->allocatorFlags_ = 0; // not allocated from any OpenCV buffer pool
+    dst.u->allocatorFlags_ = OpenCLAllocator::ALLOCATOR_FLAGS_EXTERNAL_BUFFER;  // not allocated from any OpenCV buffer pool
     dst.u->flags           = static_cast<UMatData::MemoryFlag>(0);
     dst.u->handle          = cl_mem_buffer;
     dst.u->origdata        = 0;

modules/core/src/system.cpp

@@ -71,6 +71,8 @@ static bool param_dumpErrors = utils::getConfigurationParameterBool("OPENCV_DUMP
 );
 
 void* allocSingletonBuffer(size_t size) { return fastMalloc(size); }
+void* allocSingletonNewBuffer(size_t size) { return malloc(size); }
+
 
 } // namespace cv