Beispiel #1
0
//---------------------------------------------------------------------------
//	@function:
//		CBitSetTest::EresUnittest_SetOps
//
//	@doc:
//		Test for set operations
//
//---------------------------------------------------------------------------
GPOS_RESULT
CBitSetTest::EresUnittest_SetOps()
{
	// create memory pool
	CAutoMemoryPool amp;
	IMemoryPool *pmp = amp.Pmp();

	ULONG cSizeBits = 32;
	ULONG cInserts = 10;

	CBitSet *pbs1 = GPOS_NEW(pmp) CBitSet(pmp, cSizeBits);
	for (ULONG i = 0; i < cInserts; i += 2)
	{
		pbs1->FExchangeSet(i * cSizeBits);
	}

	CBitSet *pbs2 = GPOS_NEW(pmp) CBitSet(pmp, cSizeBits);
	for (ULONG i = 1; i < cInserts; i += 2)
	{
		pbs2->FExchangeSet(i * cSizeBits);
	}
	CBitSet *pbs = GPOS_NEW(pmp) CBitSet(pmp, cSizeBits);

	pbs->Union(pbs1);
	GPOS_ASSERT(pbs->FEqual(pbs1));

	pbs->Intersection(pbs1);
	GPOS_ASSERT(pbs->FEqual(pbs1));
	GPOS_ASSERT(pbs->FEqual(pbs));
	GPOS_ASSERT(pbs1->FEqual(pbs1));

	pbs->Union(pbs2);
	GPOS_ASSERT(!pbs->FEqual(pbs1) && !pbs->FEqual(pbs2));
	GPOS_ASSERT(pbs->FSubset(pbs1) && pbs->FSubset(pbs2));
	
	pbs->Difference(pbs2);
	GPOS_ASSERT(pbs->FEqual(pbs1));

	pbs1->Release();

	pbs->Union(pbs2);
	pbs->Intersection(pbs2);
	GPOS_ASSERT(pbs->FEqual(pbs2));
	GPOS_ASSERT(pbs->FSubset(pbs2));

	GPOS_ASSERT(pbs->CElements() == pbs2->CElements());

	pbs2->Release();

	pbs->Release();

	return GPOS_OK;
}
Beispiel #2
0
//---------------------------------------------------------------------------
//	@function:
//		CJoinOrderDP::PexprBestJoinOrder
//
//	@doc:
//		find best join order for a given set of elements;
//
//---------------------------------------------------------------------------
CExpression *
CJoinOrderDP::PexprBestJoinOrder
	(
	CBitSet *pbs
	)
{
	GPOS_CHECK_STACK_SIZE;
	GPOS_CHECK_ABORT;

	GPOS_ASSERT(NULL != pbs);

	// start by looking-up cost in the DP map
	CExpression *pexpr = PexprLookup(pbs);

	if (pexpr == m_pexprDummy)
	{
		// no join order could be created
		return NULL;
	}

	if (NULL != pexpr)
	{
		// join order is found by looking up map
		return pexpr;
	}

	// find maximal covered subset
	CBitSet *pbsCovered = PbsCovered(pbs);
	if (0 == pbsCovered->Size())
	{
		// set is not covered, return a cross product
		pbsCovered->Release();

		return PexprCross(pbs);
	}

	if (!pbsCovered->Equals(pbs))
	{
		// create a cross product for uncovered subset
		CBitSet *pbsUncovered = GPOS_NEW(m_mp) CBitSet(m_mp, *pbs);
		pbsUncovered->Difference(pbsCovered);
		CExpression *pexprResult =
			PexprJoinCoveredSubsetWithUncoveredSubset(pbs, pbsCovered, pbsUncovered);
		pbsCovered->Release();
		pbsUncovered->Release();

		return pexprResult;
	}
	pbsCovered->Release();

	// if set has size 2, there is only one possible solution
	if (2 == pbs->Size())
	{
		return PexprJoin(pbs);
	}

	// otherwise, compute best join order using dynamic programming
	CExpression *pexprBestJoinOrder = PexprBestJoinOrderDP(pbs);
	if (pexprBestJoinOrder == m_pexprDummy)
	{
		// no join order could be created
		return NULL;
	}

	return pexprBestJoinOrder;
}
Beispiel #3
0
//---------------------------------------------------------------------------
//	@function:
//		CJoinOrderDP::PexprBestJoinOrderDP
//
//	@doc:
//		Find the best join order of a given set of elements using dynamic
//		programming;
//		given a set of elements (e.g., {A, B, C}), we find all possible splits
//		of the set (e.g., {A}, {B, C}) where at least one edge connects the
//		two subsets resulting from the split,
//		for each split, we find the best join orders of left and right subsets
//		recursively,
//		the function finds the split with the least cost, and stores the join
//		of its two subsets as the best join order of the given set
//
//
//---------------------------------------------------------------------------
CExpression *
CJoinOrderDP::PexprBestJoinOrderDP
	(
	CBitSet *pbs // set of elements to be joined
	)
{
	CDouble dMinCost(0.0);
	CExpression *pexprResult = NULL;

	CBitSetArray *pdrgpbsSubsets = PdrgpbsSubsets(m_mp, pbs);
	const ULONG ulSubsets = pdrgpbsSubsets->Size();
	for (ULONG ul = 0; ul < ulSubsets; ul++)
	{
		CBitSet *pbsCurrent = (*pdrgpbsSubsets)[ul];
		CBitSet *pbsRemaining = GPOS_NEW(m_mp) CBitSet(m_mp, *pbs);
		pbsRemaining->Difference(pbsCurrent);

		// check if subsets are connected with one or more edges
		CExpression *pexprPred = PexprPred(pbsCurrent, pbsRemaining);
		if (NULL != pexprPred)
		{
			// compute solutions of left and right subsets recursively
			CExpression *pexprLeft = PexprBestJoinOrder(pbsCurrent);
			CExpression *pexprRight = PexprBestJoinOrder(pbsRemaining);

			if (NULL != pexprLeft && NULL != pexprRight)
			{
				// we found solutions of left and right subsets, we check if
				// this gives a better solution for the input set
				CExpression *pexprJoin = PexprJoin(pbsCurrent, pbsRemaining);
				CDouble dCost = DCost(pexprJoin);

				if (NULL == pexprResult || dCost < dMinCost)
				{
					// this is the first solution, or we found a better solution
					dMinCost = dCost;
					CRefCount::SafeRelease(pexprResult);
					pexprJoin->AddRef();
					pexprResult = pexprJoin;
				}

				if (m_ulComps == pbs->Size())
				{
					AddJoinOrder(pexprJoin, dCost);
				}

				pexprJoin->Release();
			}
		}
		pbsRemaining->Release();
	}
	pdrgpbsSubsets->Release();

	// store solution in DP table
	if (NULL == pexprResult)
	{
		m_pexprDummy->AddRef();
		pexprResult = m_pexprDummy;
	}

	DeriveStats(pexprResult);
	pbs->AddRef();
#ifdef GPOS_DEBUG
	BOOL fInserted =
#endif // GPOS_DEBUG
		m_phmbsexpr->Insert(pbs, pexprResult);
	GPOS_ASSERT(fInserted);

	// add expression cost to cost map
	InsertExpressionCost(pexprResult, dMinCost, false /*fValidateInsert*/);

	return pexprResult;
}