SURF detect crashed with TBB (Bug #1349)

I updated and built OpenCV 2.32 SVN yesterday with the latest stable version of TBB using CMake 2.85 and Microsoft Visual Studio 2010 x64.

But SURF detect crashed in both cv cpp example code and my own program. It's my first time to build OpenCV with TBB, I followed the guide in the wiki. I've tried TBB3 in Intel C++ Composer XE, but it's the same problem.

Error Message:

Debug Assertion Failed!

Program: .......
Module: tbb_debug.dll
File: ../../src/tbb/mutex.cpp
Line: 43

Expression: false
mutex::scoped_lock: illegal mutex state

VS Call Stack(sorry for Chinese characters):

msvcr100d.dll!_CrtDbgBreak() 行 85 C

msvcr100d.dll!_VCrtDbgReportA(int nRptType=2, const char * szFile=0x000000005a0b8520, int nLine=1019, const char * szModule=0x0000000000000000, const char * szFormat=0x0000000000000000, char * arglist=0x00000000086bebb8) 行 287 C
msvcr100d.dll!_CrtDbgReportV(int nRptType=2, const char * szFile=0x000000005a0b8520, int nLine=1019, const char * szModule=0x0000000000000000, const char * szFormat=0x0000000000000000, char * arglist=0x00000000086bebb8) 行 242 C
msvcr100d.dll!_CrtDbgReport(int nRptType=2, const char * szFile=0x000000005a0b8520, int nLine=1019, const char * szModule=0x0000000000000000, const char * szFormat=0x0000000000000000, ...) 行 258 + 0x2c 字节 C
mfc100d.dll!AfxAssertFailedLine(const char * lpszFileName=0x000000005a0b8520, int nLine=1019) 行 25 + 0x27 字节 C++
mfc100d.dll!CWnd::DestroyWindow() 行 1019 + 0x3a 字节 C++
mfc100d.dll!CToolTipCtrl::DestroyToolTipCtrl() 行 76 C++
mfc100d.dll!AFX_MODULE_THREAD_STATE::~AFX_MODULE_THREAD_STATE() 行 372 C++
mfc100d.dll!AFX_MODULE_THREAD_STATE::@scalar deleting destructor'() + 0x17 字节 C++
mfc100d.dll!CThreadSlotData::DeleteValues(CThreadData * pData=0x0000000000272750, HINSTANCE+ * hInst=0x0000000000000000) 行 354 + 0x3d 字节 C++
mfc100d.dll!CThreadSlotData::DeleteValues(HINSTANCE+ * hInst=0x0000000000000000, int bAll=1) 行 396 C++
mfc100d.dll!AfxTermLocalData(HINSTANCE+ * hInst=0x0000000000000000, int bAll=1) 行 494 C++
mfc100d.dll!DllMain(HINSTANCE+ * hInstance=0x0000000059aa0000, unsigned long dwReason=0, void * formal=0x0000000000000001) 行 563 + 0xc 字节 C+
mfc100d.dll!+DllMainCRTStartup(void * hDllHandle=0x0000000059aa0000, unsigned long dwReason=0, void * lpreserved=0x0000000000000001) 行 512 + 0x13 字节 C
mfc100d.dll!_DllMainCRTStartup(void * hDllHandle=0x0000000059aa0000, unsigned long dwReason=0, void * lpreserved=0x0000000000000001) 行 477 C
ntdll.dll!0000000077a94371()
[下面的框架可能不正确和/或缺失，没有为 ntdll.dll 加载符号]
ntdll.dll!0000000077a94180()
msvcr100d.dll!+crtExitProcess(int status=3) 行 709 C
msvcr100d.dll!doexit(int code=3, int quick=1, int retcaller=0) 行 622 C
msvcr100d.dll!_exit(int code=3) 行 402 C
msvcr100d.dll!raise(int signum=22) 行 525 C
msvcr100d.dll!+crtMessageWindowA(int nRptType=2, const char * szFile=0x000007fef8808390, const char * szLine=0x00000000086c35e0, const char * szModule=0x000007fef88076c8, const char * szUserMessage=0x00000000086c45e0) 行 377 C
msvcr100d.dll!_VCrtDbgReportA(int nRptType=2, const char * szFile=0x000007fef8808390, int nLine=43, const char * szModule=0x000007fef88076c8, const char * szFormat=0x000007fef88076bc, char * arglist=0x00000000086c5718) 行 417 + 0x59 字节 C
msvcr100d.dll!_CrtDbgReportV(int nRptType=2, const char * szFile=0x000007fef8808390, int nLine=43, const char * szModule=0x000007fef88076c8, const char * szFormat=0x000007fef88076bc, char * arglist=0x00000000086c5718) 行 242 C
msvcr100d.dll!_CrtDbgReport(int nRptType=2, const char * szFile=0x000007fef8808390, int nLine=43, const char * szModule=0x000007fef88076c8, const char * szFormat=0x000007fef88076bc, ...) 行 258 + 0x2c 字节 C
tbb_debug.dll!tbb::assertion_failure(const char * filename=0x000007fef8808390, int line=43, const char * expression=0x000007fef8808388, const char * comment=0x000007fef8808360) 行 68 + 0x5c 字节 C++
tbb_debug.dll!tbb::mutex::scoped_lock::internal_acquire(tbb::mutex & m={...}) 行 50 C++
opencv_features2d232d.dll!tbb::mutex::scoped_lock::acquire(tbb::mutex & mutex={...}) 行 108 C++
opencv_features2d232d.dll!tbb::mutex::scoped_lock::scoped_lock(tbb::mutex & mutex={...}) 行 92 C++
opencv_features2d232d.dll!icvFindMaximaInLayer(const CvMat * sum=0x0000000001c5aae0, const CvMat * mask_sum=0x0000000000000000, const CvSURFParams * params=0x000000000332cdc0, CvMat * * dets=0x0000000003326a20, CvMat * * traces=0x0000000003327aa0, const int * sizes=0x0000000003328b20, int layer=4, int sampleStep=1, CvSeq * points=0x0000000001c7efd0) 行 360 + 0x14 字节 C++
opencv_features2d232d.dll!cv::SURFFindInvoker::operator()(const tbb::blocked_range<int> & range={...}) 行 425 C++
opencv_features2d232d.dll!tbb::internal::start_for<tbb::blocked_range<int>,cv::SURFFindInvoker,tbb::auto_partitioner>::execute() 行 100 C++
tbb_debug.dll!tbb::internal::custom_scheduler<tbb::internal::IntelSchedulerTraits>::local_wait_for_all(tbb::task & parent={...}, tbb::task * child=0x000007fffff4b340) 行 441 + 0x10 字节 C++
tbb_debug.dll!tbb::internal::arena::process(tbb::internal::generic_scheduler & s={...}) 行 93 C++
tbb_debug.dll!tbb::internal::market::process(rml::job & j={...}) 行 385 + 0xf 字节 C++
tbb_debug.dll!tbb::internal::rml::private_worker::run() 行 256 C++
tbb_debug.dll!tbb::internal::rml::private_worker::thread_routine(void * arg=0x000007fffff43e80) 行 224 C++
msvcr100d.dll!_callthreadstartex() 行 314 + 0x17 字节 C
msvcr100d.dll!_threadstartex(void * ptd=0x0000000003688dc0) 行 297 C
kernel32.dll!000000007786652d()
ntdll.dll!0000000077a9c521()