Poor performance from stdext::hash_map

• Previous message: tom_usenet: "Re: VC 6.0 to VC 7.0 (version 1.3) porting"
• Next in thread: P.J. Plauger: "Re: Poor performance from stdext::hash_map"
• Reply: P.J. Plauger: "Re: Poor performance from stdext::hash_map"
• Messages sorted by: [ date ] [ thread ]

Date: Fri, 11 Jun 2004 12:38:02 +0100

Hi,

I've just noticed that MS are now shipping hash_map and hash_set with
DevStudio. However, I'm finding that I'm getting worse performance than the
standard map container and a lot worse than the MFC CMap. So I'm assuming
I'm doing something wrong!

Here's the code snipet - any help would be greatly appreciated:

void DoMFCSTLCompare()
{
     typedef CMap<CString, const TCHAR*, long, long> MFCMap;
     typedef std::map<std::string, long> STLMap;
     typedef stdext::hash_map<std::string, long> STLExtMap;

     const unsigned long nNumValues = 10000;

     long l;
     CString strMFC;
     std::string strSTL;

    // Init some strings:
     CString strMFC_Init[10000];
     std::string strSTL_Init[10000];
     for (unsigned long i = 0; i != nNumValues; ++i)
     {
          strMFC.Format(_T("Hello world %d"), i);
          strMFC_Init[i] = strMFC;
          strSTL_Init[i] = strMFC;
     }

     // STL version:
     const unsigned long nSTLStartTic = ::GetTickCount();
     STLMap stlMap;

     for (unsigned long i = 0; i != nNumValues; ++i)
          stlMap.insert(std::make_pair(strSTL_Init[i], i));

     const unsigned long nSTLInsertEndTic = ::GetTickCount();

     const STLMap::const_iterator itEnd = stlMap.end();
     for (STLMap::const_iterator it = stlMap.begin(); it != itEnd; ++it)
          ;

     const unsigned long nSTLExtractEndTic = ::GetTickCount();

     for (unsigned long i = 0; i != nNumValues; ++i)
          assert(stlMap.find(strSTL_Init[i]) != stlMap.end());

     const unsigned long nSTLSearchEndTic = ::GetTickCount();

     // STLExt version:
     const unsigned long nSTLExtStartTic = ::GetTickCount();
     STLExtMap stlextMap;

     for (unsigned long i = 0; i != nNumValues; ++i)
          stlextMap.insert(std::make_pair(strSTL_Init[i], i));

     const unsigned long nSTLExtInsertEndTic = ::GetTickCount();

     const STLExtMap::const_iterator itEnd2 = stlextMap.end();
     for (STLExtMap::const_iterator it2 = stlextMap.begin(); it2 != itEnd2;
++it2)
          ;

     const unsigned long nSTLExtExtractEndTic = ::GetTickCount();

     for (unsigned long i = 0; i != nNumValues; ++i)
          assert(stlextMap.find(strSTL_Init[i]) != stlextMap.end());

     const unsigned long nSTLExtSearchEndTic = ::GetTickCount();

     // MFC version:
     const unsigned long nMFCStartTic = ::GetTickCount();
     MFCMap mfcMap;
     mfcMap.InitHashTable(12007);

     for (unsigned long i = 0; i != nNumValues; ++i)
          mfcMap.SetAt(strMFC_Init[i], i);

     const unsigned long nMFCInsertEndTic = ::GetTickCount();

     POSITION pos = mfcMap.GetStartPosition();
     while (pos != 0)
          mfcMap.GetNextAssoc(pos, strMFC, l);

     const unsigned long nMFCExtractEndTic = ::GetTickCount();

     for (unsigned long i = 0; i != nNumValues; ++i)
          assert(mfcMap.Lookup(strMFC_Init[i], l) == TRUE);

     const unsigned long nMFCSearchEndTic = ::GetTickCount();
}

If I run this on .Net 2003 I get the following times (Debug build):

 nMFCInsertEndTic - nMFCStartTic 94 (MFC
Insert)
 nMFCExtractEndTic - nMFCInsertEndTic 16 (MFC Iterate)
 nMFCSearchEndTic - nMFCExtractEndTic 41 (MFC Search)
 nSTLInsertEndTic - nSTLStartTic 734 (STL
Insert)
 nSTLExtractEndTic - nSTLInsertEndTic 16 (STL
Iterate)
 nSTLSearchEndTic - nSTLExtractEndTic 235 (STL Search)
 nSTLExtInsertEndTic - nSTLExtStartTic 2531 (STLExt
Insert)
 nSTLExtExtractEndTic - nSTLExtInsertEndTic 0 (STLExt
Iterate)
 nSTLExtSearchEndTic - nSTLExtExtractEndTic 11687 (STLExt Search)

If I remove the InitHashMap call for the MFC version I get a slightly higher
time, but still much quicker than the std::map (as expected as CMap is a
hash container). What does puzzel me is the hash_map times.

I'm knocking up a coding standards document and I'm a big fan of the STL, so
I'd like to be able to recommend the STL containers over their MFC
conterparts; but as it stands I won't be able to do that.

Thanks for any help,

Andy

• Previous message: tom_usenet: "Re: VC 6.0 to VC 7.0 (version 1.3) porting"
• Next in thread: P.J. Plauger: "Re: Poor performance from stdext::hash_map"
• Reply: P.J. Plauger: "Re: Poor performance from stdext::hash_map"
• Messages sorted by: [ date ] [ thread ]