database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-07-29 08:21:15 +03:00
Code Activity

Replace underlying type for avg and sum for int types from long double to wide decimal

Browse Source
This commit is contained in:
Leonid Fedorov
2021-04-13 21:15:33 +00:00
parent 123d602296
commit f81f743282
5 changed files with 159 additions and 142 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
datatypes/mcs_datatype.h
View File

@ -361,7 +361,6 @@ static inline bool isWideDecimalType(const datatypes::SystemCatalog::ColDataType
dt == SystemCatalog::UDECIMAL);
}
/** convenience function to determine if column type is a char
* type
*/
@ -480,27 +479,18 @@ inline bool isSignedInteger(const datatypes::SystemCatalog::ColDataType type)
}
}
/**
@brief Returns true if all arguments have a DECIMAL/UDECIMAL type
@brief The method netects whether type sum and avg aggregate will have
wide decimal underlying type
*/
static inline bool isDecimalOperands(const SystemCatalog::ColDataType resultDataType,
const SystemCatalog::ColDataType leftColDataType,
const SystemCatalog::ColDataType rightColDataType)
inline bool hasUnderlyingWideDecimalForSumAndAvg(datatypes::SystemCatalog::ColDataType type)
{
return ((resultDataType == SystemCatalog::DECIMAL ||
resultDataType == SystemCatalog::UDECIMAL) &&
(leftColDataType == SystemCatalog::DECIMAL ||
leftColDataType == SystemCatalog::UDECIMAL) &&
(rightColDataType == SystemCatalog::DECIMAL ||
rightColDataType == SystemCatalog::UDECIMAL));
return datatypes::isSignedInteger(type) || datatypes::isUnsigned(type);
}
} // end of namespace datatypes
namespace datatypes
{
@ -1957,7 +1947,7 @@ public:
MinMaxInfo widenMinMaxInfo(const SystemCatalog::TypeAttributesStd &attr,
const MinMaxInfo &a,
const MinMaxInfo &b) const override
{
{
return MinMaxInfo::widenSInt64(a, b);
}
MinMaxPartitionInfo getExtentPartitionInfo(const SystemCatalog::TypeAttributesStd &attr,
@ -2038,7 +2028,7 @@ public:
MinMaxInfo widenMinMaxInfo(const SystemCatalog::TypeAttributesStd &attr,
const MinMaxInfo &a,
const MinMaxInfo &b) const override
{
{
return MinMaxInfo::widenSInt64(a, b);
}
MinMaxPartitionInfo getExtentPartitionInfo(const SystemCatalog::TypeAttributesStd &attr,
@ -2123,7 +2113,7 @@ public:
MinMaxInfo widenMinMaxInfo(const SystemCatalog::TypeAttributesStd &attr,
const MinMaxInfo &a,
const MinMaxInfo &b) const override
{
{
return MinMaxInfo::widenSInt128(a, b);
}
MinMaxPartitionInfo getExtentPartitionInfo(const SystemCatalog::TypeAttributesStd &attr,
@ -2202,7 +2192,7 @@ public:
MinMaxInfo widenMinMaxInfo(const SystemCatalog::TypeAttributesStd &attr,
const MinMaxInfo &a,
const MinMaxInfo &b) const override
{
{
return MinMaxInfo::widenSInt128(a, b);
}
MinMaxPartitionInfo getExtentPartitionInfo(const SystemCatalog::TypeAttributesStd &attr,

2
dbcon/execplan/functioncolumn.h
View File

@ -285,7 +285,7 @@ public:
}
decimal.scale = fResultType.scale;
decimal.precision = fResultType.precision;
decimal.precision = std::max(fResultType.precision, static_cast<int32_t>(decimal.precision));
return decimal;
}
virtual bool getBoolVal(rowgroup::Row& row, bool& isNull)

57
dbcon/joblist/tupleaggregatestep.cpp
View File

@ -361,6 +361,13 @@ void wideDecimalOrLongDouble(const uint64_t colProj,
precisionAgg.push_back(precisionProj[colProj]);
widthAgg.push_back(width[colProj]);
}
else if (datatypes::hasUnderlyingWideDecimalForSumAndAvg(type))
{
typeAgg.push_back(CalpontSystemCatalog::DECIMAL);
scaleAgg.push_back(0);
precisionAgg.push_back(datatypes::INT128MAXPRECISION);
widthAgg.push_back(datatypes::MAXDECIMALWIDTH);
}
else
{
typeAgg.push_back(CalpontSystemCatalog::LONGDOUBLE);
@ -1683,6 +1690,14 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(
// for count column of average function
map<uint32_t, SP_ROWAGG_FUNC_t> avgFuncMap, avgDistFuncMap;
// collect the projected column info, prepare for aggregation
vector<uint32_t> width;
for (uint64_t i = 0; i < keysProj.size(); i++)
{
width.push_back(projRG.getColumnWidth(i));
}
// associate the columns between projected RG and aggregate RG on UM
// populated the aggregate columns
// the groupby columns are put in front, even not a returned column
@ -1934,11 +1949,10 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(
oidsAgg.push_back(oidsProj[colProj]);
keysAgg.push_back(aggKey);
typeAgg.push_back(CalpontSystemCatalog::LONGDOUBLE);
csNumAgg.push_back(8);
precisionAgg.push_back(-1);
widthAgg.push_back(sizeof(long double));
scaleAgg.push_back(0);
wideDecimalOrLongDouble(colProj, typeProj[colProj],
precisionProj, scaleProj, width,
typeAgg, scaleAgg, precisionAgg, widthAgg);
colAgg++;
// has distinct step, put the count column for avg next to the sum
@ -2265,11 +2279,10 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(
oidsAggDist.push_back(oidsAgg[colAgg]);
keysAggDist.push_back(retKey);
typeAggDist.push_back(CalpontSystemCatalog::LONGDOUBLE);
wideDecimalOrLongDouble(colAgg, typeAgg[colAgg],
precisionAgg, scaleAgg, widthAgg,
typeAggDist, scaleAggDist, precisionAggDist, widthAggDist);
csNumAggDist.push_back(8);
precisionAggDist.push_back(-1);
widthAggDist.push_back(sizeof(long double));
scaleAggDist.push_back(0);
}
break;
@ -2343,11 +2356,10 @@ void TupleAggregateStep::prep1PhaseDistinctAggregate(
{
oidsAggDist.push_back(oidsAgg[colAgg]);
keysAggDist.push_back(retKey);
scaleAggDist.push_back(0);
typeAggDist.push_back(CalpontSystemCatalog::LONGDOUBLE);
csNumAggDist.push_back(8);
precisionAggDist.push_back(-1);
widthAggDist.push_back(sizeof(long double));
wideDecimalOrLongDouble(colAgg, typeAgg[colAgg],
precisionAgg, scaleAgg, widthAgg,
typeAggDist, scaleAggDist, precisionAggDist, widthAggDist);
}
else
{
@ -4166,11 +4178,10 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(
oidsAggPm.push_back(oidsProj[colProj]);
keysAggPm.push_back(aggKey);
typeAggPm.push_back(CalpontSystemCatalog::LONGDOUBLE);
csNumAggPm.push_back(8);
precisionAggPm.push_back(-1);
widthAggPm.push_back(sizeof(long double));
scaleAggPm.push_back(0);
wideDecimalOrLongDouble(colProj, typeProj[colProj],
precisionProj, scaleProj, width,
typeAggPm, scaleAggPm, precisionAggPm, widthAggPm);
colAggPm++;
}
@ -4546,11 +4557,10 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(
oidsAggDist.push_back(oidsAggUm[colUm]);
keysAggDist.push_back(retKey);
typeAggDist.push_back(CalpontSystemCatalog::LONGDOUBLE);
csNumAggDist.push_back(8);
precisionAggDist.push_back(-1);
widthAggDist.push_back(sizeof(long double));
scaleAggDist.push_back(0);
wideDecimalOrLongDouble(colUm, typeAggPm[colUm],
precisionAggPm, scaleAggPm, widthAggPm,
typeAggDist, scaleAggDist, precisionAggDist, widthAggDist);
}
// PM: put the count column for avg next to the sum
// let fall through to add a count column for average function
@ -4614,11 +4624,10 @@ void TupleAggregateStep::prep2PhasesDistinctAggregate(
{
oidsAggDist.push_back(oidsAggUm[colUm]);
keysAggDist.push_back(retKey);
scaleAggDist.push_back(0);
typeAggDist.push_back(CalpontSystemCatalog::LONGDOUBLE);
csNumAggDist.push_back(8);
precisionAggDist.push_back(-1);
widthAggDist.push_back(sizeof(long double));
wideDecimalOrLongDouble(colUm, typeAggUm[colUm],
precisionAggUm, scaleAggUm, widthAggUm,
typeAggDist, scaleAggDist, precisionAggDist, widthAggDist);
}
else
{

97
dbcon/mysql/ha_mcs_execplan.cpp
View File

@ -342,7 +342,7 @@ bool nonConstFunc(Item_func* ifp)
***********************************************************/
void getColNameFromItem(std::ostringstream& ostream, Item* item)
{
// Item_func doesn't have proper db.table.field values
// Item_func doesn't have proper db.table.field values
// inherited from Item_ident. TBD what is the valid output.
// !!!dynamic_cast fails compilation
ostream << "'";
@ -506,7 +506,7 @@ bool sortItemIsInGrouping(Item* sort_item, ORDER* groupcol)
{
Item* group_item = *(groupcol->item);
found = (group_item->eq(sort_item, false)) ? true : false;
// Detect aggregation functions first then traverse
// Detect aggregation functions first then traverse
// if sort field is a Func and group field
// is either Field or Func
// Consider nonConstFunc() check here
@ -524,9 +524,9 @@ bool sortItemIsInGrouping(Item* sort_item, ORDER* groupcol)
/*@brief buildAggFrmTempField- build aggr func from extSELECT list item*/
/***********************************************************
* DESCRIPTION:
* Server adds additional aggregation items to extended SELECT list and
* references them in projection and HAVING. This f() finds
* corresponding item in extSelAggColsItems and builds
* Server adds additional aggregation items to extended SELECT list and
* references them in projection and HAVING. This f() finds
* corresponding item in extSelAggColsItems and builds
* ReturnedColumn using the item.
* PARAMETERS:
* item Item* used to build aggregation
@ -585,7 +585,7 @@ ReturnedColumn* buildAggFrmTempField(Item* item, gp_walk_info& gwi)
* searches for semantically equal SF in the list of already
* applied equi predicates.
* TODO
* We must move find_select_handler to either future or
* We must move find_select_handler to either future or
* later execution phase.
* PARAMETERS:
* gwi main structure
@ -2289,7 +2289,7 @@ bool buildPredicateItem(Item_func* ifp, gp_walk_info* gwip)
ConstantColumn* nlhs1 = new ConstantColumn("", ConstantColumn::NULLDATA);
nlhs1->timeZone(gwip->thd->variables.time_zone->get_name()->ptr());
sop.reset(new PredicateOperator("isnull"));
sop->setOpType(lhs->resultType(), rhs->resultType());
sop->setOpType(lhs->resultType(), rhs->resultType());
sfn1 = new SimpleFilter(sop, rhs, nlhs1);
sfn1->timeZone(gwip->thd->variables.time_zone->get_name()->ptr());
ParseTree* ptpl = new ParseTree(sfn1);
@ -2307,7 +2307,7 @@ bool buildPredicateItem(Item_func* ifp, gp_walk_info* gwip)
ptpn->right(ptpr);
// a = b
sop.reset(new PredicateOperator("="));
sop->setOpType(lhs->resultType(), rhs->resultType());
sop->setOpType(lhs->resultType(), rhs->resultType());
sfo = new SimpleFilter(sop, lhs->clone(), rhs->clone());
sfo->timeZone(gwip->thd->variables.time_zone->get_name()->ptr());
// OR with the NULL comparison tree
@ -3158,7 +3158,7 @@ CalpontSystemCatalog::ColType colType_MysqlToIDB (const Item* item)
ct.colDataType = CalpontSystemCatalog::VARCHAR;
// MCOL-4758 the longest TEXT we deal with is 16777215 so
// limit to that.
// limit to that.
if (item->max_length < 16777215)
ct.colWidth = item->max_length;
else
@ -3444,7 +3444,7 @@ ReturnedColumn* buildReturnedColumn(
gwi.fatalParseError = true;
gwi.parseErrorText = IDBErrorInfo::instance()->errorMsg(ERR_NON_SUPPORT_SELECT_SUB);
break;
default:
gwi.fatalParseError = true;
gwi.parseErrorText = "Unknown REF item";
@ -3541,7 +3541,7 @@ ReturnedColumn* buildReturnedColumn(
if (rc)
rc->charsetNumber(item->collation.collation->number);
return rc;
}
@ -3725,9 +3725,7 @@ ArithmeticColumn* buildArithmeticColumn(
const CalpontSystemCatalog::ColType& leftColType = pt->left()->data()->resultType();
const CalpontSystemCatalog::ColType& rightColType = pt->right()->data()->resultType();
// Only tinker with the type if all columns involved are decimal
if (datatypes::isDecimalOperands(mysqlType.colDataType,
leftColType.colDataType, rightColType.colDataType))
if (datatypes::isDecimal(leftColType.colDataType) || datatypes::isDecimal(rightColType.colDataType))
{
int32_t leftColWidth = leftColType.colWidth;
int32_t rightColWidth = rightColType.colWidth;
@ -3742,6 +3740,8 @@ ArithmeticColumn* buildArithmeticColumn(
int32_t scale1 = leftColType.scale;
int32_t scale2 = rightColType.scale;
mysqlType.precision = datatypes::INT128MAXPRECISION;
if (funcName == "/" &&
(mysqlType.scale - (scale1 - scale2)) > datatypes::INT128MAXPRECISION)
{
@ -4276,7 +4276,7 @@ ReturnedColumn* buildFunctionColumn(
fc->resultType(ct);
}
fc->expressionId(ci->expressionId++);
// A few functions use a different collation than that found in
// A few functions use a different collation than that found in
// the base ifp class
if (funcName == "locate" ||
funcName == "find_in_set" ||
@ -4421,7 +4421,7 @@ FunctionColumn* buildCaseFunction(Item_func* item, gp_walk_info& gwi, bool& nonS
if ((item->functype() == Item_func::CASE_SEARCHED_FUNC) &&
(i < arg_offset))
{
// MCOL-1472 Nested CASE with an ISNULL predicate. We don't want the predicate
// MCOL-1472 Nested CASE with an ISNULL predicate. We don't want the predicate
// to pull off of rcWorkStack, so we set this inCaseStmt flag to tell it
// not to.
gwi.inCaseStmt = true;
@ -4539,7 +4539,7 @@ ConstantColumn* buildDecimalColumn(Item* item, gp_walk_info& gwi)
bool specialPrecision = false;
// handle the case when the constant value is 0.12345678901234567890123456789012345678
// In this case idp->decimal_precision() = 39, but we can
// In this case idp->decimal_precision() = 39, but we can
if (((i + 1) < str->length()) &&
str->ptr()[i] == '0' && str->ptr()[i + 1] == '.')
specialPrecision = true;
@ -4735,7 +4735,7 @@ ReturnedColumn* buildAggregateColumn(Item* item, gp_walk_info& gwi)
// Argument_count() is the # of formal parms to the agg fcn. Columnstore
// only supports 1 argument except UDAnF, COUNT(DISTINC) and GROUP_CONCAT
if (isp->argument_count() != 1
if (isp->argument_count() != 1
&& isp->sum_func() != Item_sum::COUNT_DISTINCT_FUNC
&& isp->sum_func() != Item_sum::GROUP_CONCAT_FUNC
&& isp->sum_func() != Item_sum::UDF_SUM_FUNC)
@ -4813,7 +4813,7 @@ ReturnedColumn* buildAggregateColumn(Item* item, gp_walk_info& gwi)
{
Item* ord_col = *(*order_item)->item;
if (ord_col->type() == Item::CONST_ITEM
if (ord_col->type() == Item::CONST_ITEM
&& ord_col->cmp_type() == INT_RESULT)
{
Item_int* id = (Item_int*)ord_col;
@ -5058,6 +5058,19 @@ ReturnedColumn* buildAggregateColumn(Item* item, gp_walk_info& gwi)
ct.precision = precision;
ct.scale = scale;
}
else if (datatypes::hasUnderlyingWideDecimalForSumAndAvg(ct.colDataType))
{
uint32_t precision = datatypes::INT128MAXPRECISION;
uint32_t scale = ct.scale;
ct.colDataType = CalpontSystemCatalog::DECIMAL;
ct.colWidth = datatypes::MAXDECIMALWIDTH;
if (isAvg)
{
datatypes::Decimal::setScalePrecision4Avg(precision, scale);
}
ct.scale = scale;
ct.precision = precision;
}
else
{
ct.colDataType = CalpontSystemCatalog::LONGDOUBLE;
@ -5164,7 +5177,7 @@ because it has multiple arguments.";
{
for (uint32_t i = 0; i < gwi.returnedCols.size(); i++)
{
if (*ac == gwi.returnedCols[i].get()
if (*ac == gwi.returnedCols[i].get()
&& ac->alias() == gwi.returnedCols[i].get()->alias())
ac->expressionId(gwi.returnedCols[i]->expressionId());
}
@ -5922,7 +5935,7 @@ void gp_walk(const Item* item, void* arg)
{
boost::shared_ptr<SimpleColumn> scsp(sc->clone());
gwip->scsp = scsp;
if (col->type() == Item::FIELD_ITEM)
{
const auto &field_name = string(((Item_field*)item)->field_name.str);
@ -6155,8 +6168,8 @@ void gp_walk(const Item* item, void* arg)
* functions or arithmetic expressions for vtable post process.
*/
void parse_item (Item* item, vector<Item_field*>& field_vec,
bool& hasNonSupportItem,
uint16_t& parseInfo,
bool& hasNonSupportItem,
uint16_t& parseInfo,
gp_walk_info* gwi)
{
Item::Type itype = item->type();
@ -6226,7 +6239,7 @@ void parse_item (Item* item, vector<Item_field*>& field_vec,
// special handling for count(*). This should not be treated as constant.
if (isp->argument_count() == 1 &&
( sfitempp[0]->type() == Item::CONST_ITEM &&
( sfitempp[0]->type() == Item::CONST_ITEM &&
(sfitempp[0]->cmp_type() == INT_RESULT ||
sfitempp[0]->cmp_type() == STRING_RESULT ||
sfitempp[0]->cmp_type() == REAL_RESULT ||
@ -6250,7 +6263,7 @@ void parse_item (Item* item, vector<Item_field*>& field_vec,
ReturnedColumn* rc = NULL;
if (gwi)
rc = buildAggFrmTempField(ref, *gwi);
if (!rc)
{
Item_field* ifp = reinterpret_cast<Item_field*>(*(ref->ref));
@ -6476,7 +6489,7 @@ int processFrom(bool &isUnion,
setError(gwi.thd, ER_CHECK_NOT_IMPLEMENTED, gwi.parseErrorText, gwi);
return ER_CHECK_NOT_IMPLEMENTED;
}
string viewName = getViewName(table_ptr);
if (lower_case_table_names)
{
@ -7173,7 +7186,7 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
}
setExecutionParams(gwi, csep);
gwi.subSelectType = csep->subType();
uint32_t sessionID = csep->sessionID();
gwi.sessionid = sessionID;
@ -7478,7 +7491,7 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
{
Message::Args args;
args.add(ifp->func_name());
gwi.parseErrorText =
gwi.parseErrorText =
IDBErrorInfo::instance()->errorMsg(ERR_NON_SUPPORTED_FUNCTION, args);
setError(gwi.thd, ER_CHECK_NOT_IMPLEMENTED, gwi.parseErrorText, gwi);
return ER_CHECK_NOT_IMPLEMENTED;
@ -7521,7 +7534,7 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
}
// MCOL-2178 This switch doesn't handl
// ROW_
default:
default:
{
IDEBUG(cerr << "Warning unsupported cmp_type() in projection" << endl);
//noop
@ -7651,7 +7664,7 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
}
break;
}
default:
{
@ -8141,7 +8154,7 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
if ((*(ordercol->item))->type() == Item::WINDOW_FUNC_ITEM)
gwi.hasWindowFunc = true;
// MCOL-2166 Looking for this sorting item in GROUP_BY items list.
// Shouldn't look into this if query doesn't have GROUP BY or
// Shouldn't look into this if query doesn't have GROUP BY or
// aggregations
if(select_lex.agg_func_used() && select_lex.group_list.first
&& !sortItemIsInGrouping(*ordercol->item,
@ -8180,9 +8193,9 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
Item* ord_item = *(ordercol->item);
// ignore not_used column on order by.
if ((ord_item->type() == Item::CONST_ITEM
&& ord_item->cmp_type() == INT_RESULT)
&& ord_item->full_name()
if ((ord_item->type() == Item::CONST_ITEM
&& ord_item->cmp_type() == INT_RESULT)
&& ord_item->full_name()
&& !strcmp(ord_item->full_name(), "Not_used"))
{
continue;
@ -8325,7 +8338,7 @@ int getSelectPlan(gp_walk_info& gwi, SELECT_LEX& select_lex,
for (; ordercol; ordercol = ordercol->next)
{
Item* ord_item = *(ordercol->item);
if (ord_item->name.length)
{
// for union order by 1 case. For unknown reason, it doesn't show in_field_list
@ -8681,7 +8694,7 @@ ConstantColumn* buildConstColFromFilter(SimpleColumn* originalSC,
continue;
op = simpFilter->op();
execplan::ReturnedColumn* rc = dynamic_cast<execplan::ReturnedColumn*>(simpleCol);
execplan::ReturnedColumn* rc = dynamic_cast<execplan::ReturnedColumn*>(simpleCol);
// The filter could have any kind of op
if ( originalSC->sameColumn(rc) )
@ -9183,7 +9196,7 @@ int getGroupPlan(gp_walk_info& gwi, SELECT_LEX& select_lex, SCSEP& csep, cal_gro
boost::shared_ptr<ReturnedColumn> spac(ac);
gwi.returnedCols.push_back(spac);
// This item could be used in projection or HAVING later.
gwi.extSelAggColsItems.push_back(item);
gwi.extSelAggColsItems.push_back(item);
break;
}
@ -9973,9 +9986,9 @@ int getGroupPlan(gp_walk_info& gwi, SELECT_LEX& select_lex, SCSEP& csep, cal_gro
bool nonAggField = true;
// ignore not_used column on order by.
if ((ord_item->type() == Item::CONST_ITEM
&& ord_item->cmp_type() == INT_RESULT)
&& ord_item->full_name()
if ((ord_item->type() == Item::CONST_ITEM
&& ord_item->cmp_type() == INT_RESULT)
&& ord_item->full_name()
&& !strcmp(ord_item->full_name(), "Not_used"))
{
continue;
@ -10360,13 +10373,13 @@ int getGroupPlan(gp_walk_info& gwi, SELECT_LEX& select_lex, SCSEP& csep, cal_gro
// LIMIT and OFFSET are extracted from TABLE_LIST elements.
// All of JOIN-ed tables contain relevant limit and offset.
uint64_t limit = (uint64_t)-1;
if (gi.groupByTables->select_lex->limit_params.select_limit &&
if (gi.groupByTables->select_lex->limit_params.select_limit &&
( limit = static_cast<Item_int*>(gi.groupByTables->select_lex->limit_params.select_limit)->val_int() ) &&
limit != (uint64_t)-1 )
{
csep->limitNum(limit);
}
else if (csep->hasOrderBy())
else if (csep->hasOrderBy())
{
// We use LimitedOrderBy so set the limit to
// go through the check in addOrderByAndLimit

119
utils/rowgroup/rowaggregation.cpp
View File

@ -1278,7 +1278,7 @@ void RowAggregation::doMinMax(const Row& rowIn, int64_t colIn, int64_t colOut, i
//------------------------------------------------------------------------------
void RowAggregation::doSum(const Row& rowIn, int64_t colIn, int64_t colOut, int funcType)
{
int colDataType = rowIn.getColType(colIn);
datatypes::SystemCatalog::ColDataType colDataType = rowIn.getColType(colIn);
long double valIn = 0;
bool isWideDataType = false;
void *wideValInPtr = nullptr;
@ -1328,7 +1328,7 @@ void RowAggregation::doSum(const Row& rowIn, int64_t colIn, int64_t colOut, int
idbassert(0);
throw std::logic_error("RowAggregation::doSum(): DECIMAL bad length.");
}
break;
}
@ -1379,9 +1379,26 @@ void RowAggregation::doSum(const Row& rowIn, int64_t colIn, int64_t colOut, int
break;
}
}
if (LIKELY(!isWideDataType))
bool notFirstValue = !isNull(fRowGroupOut, fRow, colOut);
if (datatypes::hasUnderlyingWideDecimalForSumAndAvg(colDataType) || isWideDataType)
{
if (LIKELY(!isNull(fRowGroupOut, fRow, colOut)))
if (LIKELY(notFirstValue))
{
int128_t *valOutPtr = fRow.getBinaryField<int128_t>(colOut);
int128_t sum = (isWideDataType) ? *valOutPtr + *reinterpret_cast<int128_t*>(wideValInPtr) : *valOutPtr + valIn;
fRow.setBinaryField(&sum, colOut);
}
else
{
int128_t sum = (isWideDataType) ? *reinterpret_cast<int128_t*>(wideValInPtr) : valIn;
fRow.setBinaryField(&sum, colOut);
}
}
else
{
if (notFirstValue)
{
long double valOut = fRow.getLongDoubleField(colOut);
fRow.setLongDoubleField(valIn+valOut, colOut);
@ -1390,22 +1407,7 @@ void RowAggregation::doSum(const Row& rowIn, int64_t colIn, int64_t colOut, int
{
fRow.setLongDoubleField(valIn, colOut);
}
}
else
{
uint32_t offset = fRow.getOffset(colOut);
int128_t* dec = reinterpret_cast<int128_t*>(wideValInPtr);
if (LIKELY(!isNull(fRowGroupOut, fRow, colOut)))
{
int128_t *valOutPtr = fRow.getBinaryField<int128_t>(colOut);
int128_t sum = *valOutPtr + *dec;
fRow.setBinaryField_offset(&sum, sizeof(sum), offset);
}
else
{
fRow.setBinaryField_offset(dec, sizeof(*dec), offset);
}
} // end-of isWideDataType block
} // end-of long double processing block
}
//------------------------------------------------------------------------------
@ -1823,9 +1825,8 @@ void RowAggregation::doAvg(const Row& rowIn, int64_t colIn, int64_t colOut, int6
if (rowIn.isNullValue(colIn))
return;
int colDataType = rowIn.getColType(colIn);
datatypes::SystemCatalog::ColDataType colDataType = rowIn.getColType(colIn);
long double valIn = 0;
long double valOut = fRow.getLongDoubleField(colOut);
bool isWideDataType = false;
void *wideValInPtr = nullptr;
@ -1907,8 +1908,6 @@ void RowAggregation::doAvg(const Row& rowIn, int64_t colIn, int64_t colOut, int6
// min(count) = 0
uint64_t count = fRow.getUintField(colAux);
bool notFirstValue = count > 0;
// Set count column
if (merge)
{
@ -1919,29 +1918,34 @@ void RowAggregation::doAvg(const Row& rowIn, int64_t colIn, int64_t colOut, int6
fRow.setUintField<8>(count + 1, colAux);
}
// Set sum column
if (LIKELY(!isWideDataType))
bool notFirstValue = count > 0;
if (datatypes::hasUnderlyingWideDecimalForSumAndAvg(colDataType) || isWideDataType)
{
if (LIKELY(notFirstValue))
fRow.setLongDoubleField(valIn + valOut, colOut);
else // This is the first value
fRow.setLongDoubleField(valIn, colOut);
}
else
{
uint32_t offset = fRow.getOffset(colOut);
int128_t* dec = reinterpret_cast<int128_t*>(wideValInPtr);
if (LIKELY(notFirstValue))
{
int128_t *valOutPtr = fRow.getBinaryField<int128_t>(colOut);
int128_t sum = *valOutPtr + *dec;
fRow.setBinaryField_offset(&sum, sizeof(sum), offset);
int128_t sum = (isWideDataType) ? *valOutPtr + *reinterpret_cast<int128_t*>(wideValInPtr) : *valOutPtr + valIn;
fRow.setBinaryField(&sum, colOut);
}
else
{
fRow.setBinaryField_offset(dec, sizeof(*dec), offset);
int128_t sum = (isWideDataType) ? *reinterpret_cast<int128_t*>(wideValInPtr) : valIn;
fRow.setBinaryField(&sum, colOut);
}
}
else
{
if (LIKELY(notFirstValue))
{
long double valOut = fRow.getLongDoubleField(colOut);
fRow.setLongDoubleField(valIn+valOut, colOut);
}
else
{
fRow.setLongDoubleField(valIn, colOut);
}
} // end-of long double processing block
}
@ -3614,7 +3618,7 @@ void RowAggregationUM::doNotNullConstantAggregate(const ConstantAggData& aggData
{
fRow.setIntField(strtol(aggData.fConstValue.c_str(), nullptr, 10), colOut);
}
break;
break;
// AVG should not be uint32_t result type.
case execplan::CalpontSystemCatalog::UTINYINT:
@ -3777,7 +3781,7 @@ void RowAggregationUM::doNotNullConstantAggregate(const ConstantAggData& aggData
// TODO: isn't precision loss possible below?
dbl *= datatypes::scaleDivisor<double>(fRowGroupOut->getScale()[i]);
dbl *= rowCnt;
if ((dbl > 0 && dbl > (double) numeric_limits<int64_t>::max()) ||
(dbl < 0 && dbl < (double) numeric_limits<int64_t>::min()))
throw logging::QueryDataExcept(overflowMsg, logging::aggregateDataErr);
@ -4280,9 +4284,8 @@ void RowAggregationUMP2::doAvg(const Row& rowIn, int64_t colIn, int64_t colOut,
if (rowIn.isNullValue(colIn))
return;
int colDataType = rowIn.getColType(colIn);
datatypes::SystemCatalog::ColDataType colDataType = rowIn.getColType(colIn);
long double valIn = 0;
long double valOut = fRow.getLongDoubleField(colOut);
bool isWideDataType = false;
void *wideValInPtr = nullptr;
@ -4362,37 +4365,39 @@ void RowAggregationUMP2::doAvg(const Row& rowIn, int64_t colIn, int64_t colOut,
}
}
uint64_t cnt = fRow.getUintField(colAux);
if (LIKELY(!isWideDataType))
uint64_t count = fRow.getUintField(colAux);
bool notFirstValue = count > 0;
if (datatypes::hasUnderlyingWideDecimalForSumAndAvg(colDataType) || isWideDataType)
{
if (LIKELY(cnt > 0))
if (LIKELY(notFirstValue))
{
fRow.setLongDoubleField(valIn + valOut, colOut);
fRow.setUintField(rowIn.getUintField(colIn + 1) + cnt, colAux);
int128_t *valOutPtr = fRow.getBinaryField<int128_t>(colOut);
int128_t sum = (isWideDataType) ? *valOutPtr + *reinterpret_cast<int128_t*>(wideValInPtr) : *valOutPtr + valIn;
fRow.setUintField(rowIn.getUintField(colIn + 1) + count, colAux);
fRow.setBinaryField(&sum, colOut);
}
else
{
fRow.setLongDoubleField(valIn, colOut);
int128_t sum = (isWideDataType) ? *reinterpret_cast<int128_t*>(wideValInPtr) : valIn;
fRow.setUintField(rowIn.getUintField(colIn + 1), colAux);
fRow.setBinaryField(&sum, colOut);
}
}
else
{
uint32_t offset = fRow.getOffset(colOut);
int128_t* dec = reinterpret_cast<int128_t*>(wideValInPtr);
if (LIKELY(cnt > 0))
if (LIKELY(notFirstValue))
{
int128_t *valOutPtr = fRow.getBinaryField<int128_t>(colOut);
int128_t sum = *valOutPtr + *dec;
fRow.setBinaryField_offset(&sum, sizeof(sum), offset);
fRow.setUintField(rowIn.getUintField(colIn + 1) + cnt, colAux);
long double valOut = fRow.getLongDoubleField(colOut);
fRow.setUintField(rowIn.getUintField(colIn + 1) + count, colAux);
fRow.setLongDoubleField(valIn+valOut, colOut);
}
else
{
fRow.setBinaryField_offset(dec, sizeof(*dec), offset);
fRow.setUintField(rowIn.getUintField(colIn + 1), colAux);
fRow.setLongDoubleField(valIn, colOut);
}
}
} // end-of long double processing block
}
//------------------------------------------------------------------------------