database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-08-08 14:22:09 +03:00
Code Activity

MCOL-4171

Browse Source
This commit is contained in:
David Hall
2020-07-30 17:28:11 -05:00
committed by Roman Nozdrin
parent 5287e6860b
commit 638202417f
40 changed files with 807 additions and 250 deletions
Download Patch File Download Diff File Expand all files Collapse all files

51
utils/regr/moda.cpp
View File

@@ -103,9 +103,12 @@ mcsv1_UDAF* moda::getImpl(mcsv1Context* context)
case 4:
data->modaImpl = &moda_impl_int32;
break;
default:
case 8:
data->modaImpl = &moda_impl_int64;
break;
case 16:
data->modaImpl = &moda_impl_int128;
break;
}
break;
case execplan::CalpontSystemCatalog::UTINYINT:
@@ -179,11 +182,18 @@ mcsv1_UDAF::ReturnCode moda::init(mcsv1Context* context,
{
context->setColWidth(4);
}
else
else if (colTypes[0].precision < 19)
{
context->setColWidth(8);
}
else
{
context->setColWidth(16);
}
context->setScale(colTypes[0].scale);
}
context->setPrecision(colTypes[0].precision);
mcsv1_UDAF* impl = getImpl(context);
@@ -203,10 +213,7 @@ template<class T>
mcsv1_UDAF::ReturnCode Moda_impl_T<T>::init(mcsv1Context* context,
ColumnDatum* colTypes)
{
context->setScale(context->getScale());
context->setPrecision(19);
return mcsv1_UDAF::SUCCESS;
}
template<class T>
@@ -224,7 +231,7 @@ mcsv1_UDAF::ReturnCode Moda_impl_T<T>::nextValue(mcsv1Context* context, ColumnDa
{
static_any::any& valIn = valsIn[0].columnData;
ModaData* data = static_cast<ModaData*>(context->getUserData());
std::unordered_map<T, uint32_t>* map = data->getMap<T>();
std::unordered_map<T, uint32_t, hasher<T> >* map = data->getMap<T>();
if (valIn.empty())
{
@@ -261,9 +268,9 @@ mcsv1_UDAF::ReturnCode Moda_impl_T<T>::subEvaluate(mcsv1Context* context, const
ModaData* outData = static_cast<ModaData*>(context->getUserData());
const ModaData* inData = static_cast<const ModaData*>(userDataIn);
std::unordered_map<T, uint32_t>* outMap = outData->getMap<T>();
std::unordered_map<T, uint32_t>* inMap = inData->getMap<T>();
typename std::unordered_map<T, uint32_t>::const_iterator iter;
std::unordered_map<T, uint32_t, hasher<T> >* outMap = outData->getMap<T>();
std::unordered_map<T, uint32_t, hasher<T> >* inMap = inData->getMap<T>();
typename std::unordered_map<T, uint32_t, hasher<T> >::const_iterator iter;
for (iter = inMap->begin(); iter != inMap->end(); ++iter)
{
@@ -283,7 +290,7 @@ mcsv1_UDAF::ReturnCode Moda_impl_T<T>::evaluate(mcsv1Context* context, static_an
long double avg = 0;
T val = 0;
ModaData* data = static_cast<ModaData*>(context->getUserData());
std::unordered_map<T, uint32_t>* map = data->getMap<T>();
std::unordered_map<T, uint32_t, hasher<T> >* map = data->getMap<T>();
if (map->size() == 0)
{
@@ -292,7 +299,7 @@ mcsv1_UDAF::ReturnCode Moda_impl_T<T>::evaluate(mcsv1Context* context, static_an
}
avg = data->fCount ? data->fSum / data->fCount : 0;
typename std::unordered_map<T, uint32_t>::iterator iter;
typename std::unordered_map<T, uint32_t, hasher<T> >::iterator iter;
for (iter = map->begin(); iter != map->end(); ++iter)
{
@@ -303,11 +310,13 @@ mcsv1_UDAF::ReturnCode Moda_impl_T<T>::evaluate(mcsv1Context* context, static_an
}
else if (iter->second == maxCnt)
{
T absval = val >= 0 ? val : -val;
T absfirst = iter->first >= 0 ? iter->first : -iter->first;
// Tie breaker: choose the closest to avg. If still tie, choose smallest
long double dist1 = val > avg ? (long double)val-avg : avg-(long double)val;
long double dist2 = iter->first > avg ? (long double)iter->first-avg : avg-(long double)iter->first;
if ((dist1 > dist2)
|| ((dist1 == dist2) && (std::fabs(val) > std::fabs(iter->first))))
|| ((dist1 == dist2) && (absval > absfirst)))
{
val = iter->first;
}
@@ -328,7 +337,7 @@ mcsv1_UDAF::ReturnCode Moda_impl_T<T>::dropValue(mcsv1Context* context, ColumnDa
{
static_any::any& valDropped = valsDropped[0].columnData;
ModaData* data = static_cast<ModaData*>(context->getUserData());
std::unordered_map<T, uint32_t>* map = data->getMap<T>();
std::unordered_map<T, uint32_t, hasher<T> >* map = data->getMap<T>();
if (valDropped.empty())
{
@@ -379,9 +388,12 @@ void ModaData::serialize(messageqcpp::ByteStream& bs) const
case 4:
serializeMap<int32_t>(bs);
break;
default:
case 8:
serializeMap<int64_t>(bs);
break;
case 16:
serializeMap<int128_t>(bs);
break;
}
break;
case execplan::CalpontSystemCatalog::UTINYINT:
@@ -447,9 +459,12 @@ void ModaData::unserialize(messageqcpp::ByteStream& bs)
case 4:
unserializeMap<int32_t>(bs);
break;
default:
case 8:
unserializeMap<int64_t>(bs);
break;
case 16:
unserializeMap<int128_t>(bs);
break;
}
break;
case execplan::CalpontSystemCatalog::UTINYINT:
@@ -519,10 +534,14 @@ void ModaData::cleanup()
clear<int32_t>();
deleteMap<int32_t>();
break;
default:
case 8:
clear<int64_t>();
deleteMap<int64_t>();
break;
case 16:
clear<int128_t>();
deleteMap<int128_t>();
break;
}
break;
case execplan::CalpontSystemCatalog::UTINYINT:

52
utils/regr/moda.h
View File

@@ -45,6 +45,7 @@
#include "mcsv1_udaf.h"
#include "calpontsystemcatalog.h"
#include "windowfunctioncolumn.h"
#include "hasher.h"
#if defined(_MSC_VER) && defined(xxxRGNODE_DLLEXPORT)
#define EXPORT __declspec(dllexport)
@@ -54,6 +55,38 @@
namespace mcsv1sdk
{
// A hasher that handles int128_t
template<class T>
struct hasher
{
inline size_t operator()(T val) const
{
return fHasher((char*) &val, sizeof(T));
}
private:
utils::Hasher fHasher;
};
template<>
struct hasher<long double>
{
inline size_t operator()(long double val) const
{
if (sizeof(long double) == 8) // Probably just MSC, but you never know.
{
return fHasher((char*) &val, sizeof(long double));
}
else
{
// For Linux x86_64, long double is stored in 128 bits, but only 80 are significant
return fHasher((char*) &val, 10);
}
}
private:
utils::Hasher fHasher;
};
// Override UserData for data storage
struct ModaData : public UserData
{
@@ -69,22 +102,22 @@ struct ModaData : public UserData
virtual void unserialize(messageqcpp::ByteStream& bs);
template<class T>
std::unordered_map<T, uint32_t>* getMap()
std::unordered_map<T, uint32_t, hasher<T> >* getMap()
{
if (!fMap)
{
// Just in time creation
fMap = new std::unordered_map<T, uint32_t>;
fMap = new std::unordered_map<T, uint32_t, hasher<T> >;
}
return (std::unordered_map<T, uint32_t>*) fMap;
return (std::unordered_map<T, uint32_t, hasher<T> >*) fMap;
}
// The const version is only called by serialize()
// It shouldn't (and can't) create a new map.
template<class T>
std::unordered_map<T, uint32_t>* getMap() const
std::unordered_map<T, uint32_t, hasher<T> >* getMap() const
{
return (std::unordered_map<T, uint32_t>*) fMap;
return (std::unordered_map<T, uint32_t, hasher<T> >*) fMap;
}
template<class T>
@@ -92,7 +125,7 @@ struct ModaData : public UserData
{
if (fMap)
{
delete (std::unordered_map<T, uint32_t>*) fMap;
delete (std::unordered_map<T, uint32_t, hasher<T> >*) fMap;
fMap = NULL;
}
}
@@ -123,10 +156,10 @@ private:
template<class T>
void serializeMap(messageqcpp::ByteStream& bs) const
{
std::unordered_map<T, uint32_t>* map = getMap<T>();
std::unordered_map<T, uint32_t, hasher<T> >* map = getMap<T>();
if (map)
{
typename std::unordered_map<T, uint32_t>::const_iterator iter;
typename std::unordered_map<T, uint32_t, hasher<T> >::const_iterator iter;
bs << (uint64_t)map->size();
for (iter = map->begin(); iter != map->end(); ++iter)
{
@@ -147,7 +180,7 @@ private:
T num;
uint64_t sz;
bs >> sz;
std::unordered_map<T, uint32_t>* map = getMap<T>();
std::unordered_map<T, uint32_t, hasher<T> >* map = getMap<T>();
map->clear();
for (uint64_t i = 0; i < sz; ++i)
{
@@ -234,6 +267,7 @@ protected:
Moda_impl_T<int16_t> moda_impl_int16;
Moda_impl_T<int32_t> moda_impl_int32;
Moda_impl_T<int64_t> moda_impl_int64;
Moda_impl_T<int128_t> moda_impl_int128;
Moda_impl_T<uint8_t> moda_impl_uint8;
Moda_impl_T<uint16_t> moda_impl_uint16;
Moda_impl_T<uint32_t> moda_impl_uint32;