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

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

@@ -390,6 +390,8 @@ CV_EXPORTS CV_NORETURN void error(int _code, const String& _err, const char* _fu
 #define CV_Error(...) do { abort(); } while (0)
 #define CV_Error_( code, args ) do { cv::format args; abort(); } while (0)
 #define CV_Assert( expr ) do { if (!(expr)) abort(); } while (0)
+#define CV_ErrorNoReturn CV_Error
+#define CV_ErrorNoReturn_ CV_Error_
 
 #else // CV_STATIC_ANALYSIS
 

modules/core/include/opencv2/core/hal/intrin.hpp

@@ -368,6 +368,9 @@ inline unsigned int trailingZeros32(unsigned int value) {
     unsigned long index = 0;
     _BitScanForward(&index, value);
     return (unsigned int)index;
+#elif defined(__clang__)
+    // clang-cl doesn't export _tzcnt_u32 for non BMI systems
+    return value ? __builtin_ctz(value) : 32;
 #else
     return _tzcnt_u32(value);
 #endif

modules/core/include/opencv2/core/ocl.hpp

@@ -352,7 +352,8 @@ public:
     KernelArg(int _flags, UMat* _m, int wscale=1, int iwscale=1, const void* _obj=0, size_t _sz=0);
     KernelArg();
 
-    static KernelArg Local() { return KernelArg(LOCAL, 0); }
+    static KernelArg Local(size_t localMemSize)
+    { return KernelArg(LOCAL, 0, 1, 1, 0, localMemSize); }
     static KernelArg PtrWriteOnly(const UMat& m)
     { return KernelArg(PTR_ONLY+WRITE_ONLY, (UMat*)&m); }
     static KernelArg PtrReadOnly(const UMat& m)

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

@@ -142,9 +142,13 @@ 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
-template <typename T>
-T* allocSingleton(size_t count) { return static_cast<T*>(fastMalloc(sizeof(T) * count)); }
-}
+CV_EXPORTS void* allocSingletonBuffer(size_t size);
+
+//! Allocate all memory buffers which will not be freed, ease filtering memcheck issues
+template <typename T> static inline
+T* allocSingleton(size_t count = 1) { return static_cast<T*>(allocSingletonBuffer(sizeof(T) * count)); }
+
+} // namespace
 
 #if 1 // TODO: Remove in OpenCV 4.x