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

AUX column scan(MCOL-5021) effectively disables vectorized scanning on

Browse Source 
ARM platforms. This patch resolves this issue and unifies AUX column
processing at x86 and ARM using tempate class SimdProcessor.
The patch also replaces uint16_t mask previously used in column.cpp and
SimProcessor code with a native masks that platform uses, e.g. __m128i
or __m128 on x86 and variety of masks on ARM.
To unify the processing I introduced a new filtering Compare Operator - COMPARE_NULLEQ.
with a 'c1 IS NULL semantics'.
This commit is contained in:
NTH19
2022-08-26 19:37:40 +08:00
committed by Roman Nozdrin
parent 85a6121f76
commit 7d76dc4534
7 changed files with 1531 additions and 1407 deletions
Download Patch File Download Diff File Expand all files Collapse all files

127
primitives/linux-port/primitiveprocessor.h
View File

@ -55,7 +55,7 @@
class PrimTest;
// XXX: turn off dictionary range setting during scan.
#define XXX_PRIMITIVES_TOKEN_RANGES_XXX
#define XXX_PRIMITIVES_TOKEN_RANGES_XXX
namespace primitives
{
@ -472,6 +472,103 @@ class PrimitiveProcessor
friend class ::PrimTest;
};
// Bit pattern representing NULL value for given column type/width
// TBD Use TypeHandler
template <typename T, typename std::enable_if<sizeof(T) == sizeof(int128_t), T>::type* = nullptr>
T getNullValue(uint8_t type)
{
return datatypes::Decimal128Null;
}
template <typename T, typename std::enable_if<sizeof(T) == sizeof(int64_t), T>::type* = nullptr>
T getNullValue(uint8_t type)
{
switch (type)
{
case execplan::CalpontSystemCatalog::DOUBLE:
case execplan::CalpontSystemCatalog::UDOUBLE: return joblist::DOUBLENULL;
case execplan::CalpontSystemCatalog::CHAR:
case execplan::CalpontSystemCatalog::VARCHAR:
case execplan::CalpontSystemCatalog::DATE:
case execplan::CalpontSystemCatalog::DATETIME:
case execplan::CalpontSystemCatalog::TIMESTAMP:
case execplan::CalpontSystemCatalog::TIME:
case execplan::CalpontSystemCatalog::VARBINARY:
case execplan::CalpontSystemCatalog::BLOB:
case execplan::CalpontSystemCatalog::TEXT: return joblist::CHAR8NULL;
case execplan::CalpontSystemCatalog::UBIGINT: return joblist::UBIGINTNULL;
default: return joblist::BIGINTNULL;
}
}
template <typename T, typename std::enable_if<sizeof(T) == sizeof(int32_t), T>::type* = nullptr>
T getNullValue(uint8_t type)
{
switch (type)
{
case execplan::CalpontSystemCatalog::FLOAT:
case execplan::CalpontSystemCatalog::UFLOAT: return joblist::FLOATNULL;
case execplan::CalpontSystemCatalog::CHAR:
case execplan::CalpontSystemCatalog::VARCHAR:
case execplan::CalpontSystemCatalog::BLOB:
case execplan::CalpontSystemCatalog::TEXT: return joblist::CHAR4NULL;
case execplan::CalpontSystemCatalog::DATE:
case execplan::CalpontSystemCatalog::DATETIME:
case execplan::CalpontSystemCatalog::TIMESTAMP:
case execplan::CalpontSystemCatalog::TIME: return joblist::DATENULL;
case execplan::CalpontSystemCatalog::UINT:
case execplan::CalpontSystemCatalog::UMEDINT: return joblist::UINTNULL;
default: return joblist::INTNULL;
}
}
template <typename T, typename std::enable_if<sizeof(T) == sizeof(int16_t), T>::type* = nullptr>
T getNullValue(uint8_t type)
{
switch (type)
{
case execplan::CalpontSystemCatalog::CHAR:
case execplan::CalpontSystemCatalog::VARCHAR:
case execplan::CalpontSystemCatalog::BLOB:
case execplan::CalpontSystemCatalog::TEXT:
case execplan::CalpontSystemCatalog::DATE:
case execplan::CalpontSystemCatalog::DATETIME:
case execplan::CalpontSystemCatalog::TIMESTAMP:
case execplan::CalpontSystemCatalog::TIME: return joblist::CHAR2NULL;
case execplan::CalpontSystemCatalog::USMALLINT: return joblist::USMALLINTNULL;
default: return joblist::SMALLINTNULL;
}
}
template <typename T, typename std::enable_if<sizeof(T) == sizeof(int8_t), T>::type* = nullptr>
T getNullValue(uint8_t type)
{
switch (type)
{
case execplan::CalpontSystemCatalog::CHAR:
case execplan::CalpontSystemCatalog::VARCHAR:
case execplan::CalpontSystemCatalog::BLOB:
case execplan::CalpontSystemCatalog::TEXT:
case execplan::CalpontSystemCatalog::DATE:
case execplan::CalpontSystemCatalog::DATETIME:
case execplan::CalpontSystemCatalog::TIMESTAMP:
case execplan::CalpontSystemCatalog::TIME: return joblist::CHAR1NULL;
case execplan::CalpontSystemCatalog::UTINYINT: return joblist::UTINYINTNULL;
default: return joblist::TINYINTNULL;
}
}
//
// COMPILE A COLUMN FILTER
//
@ -518,13 +615,32 @@ boost::shared_ptr<ParsedColumnFilter> _parseColumnFilter(
// Pointer to ColArgs structure representing argIndex'th element in the BLOB
auto args = reinterpret_cast<const ColArgs*>(filterString + (argIndex * filterSize));
ret->prestored_cops[argIndex] = args->COP;
ret->prestored_rfs[argIndex] = args->rf;
if (datatypes::isUnsigned((execplan::CalpontSystemCatalog::ColDataType)colType))
ret->storeFilterArg(argIndex, reinterpret_cast<const UT*>(args->val));
auto colDataType = (execplan::CalpontSystemCatalog::ColDataType)colType;
bool isNullEqCmp = false;
if (datatypes::isUnsigned(colDataType))
{
const auto nullValue = getNullValue<UT>(colDataType);
const UT* filterValue = reinterpret_cast<const UT*>(args->val);
isNullEqCmp =
(args->COP == COMPARE_EQ && memcmp(filterValue, &nullValue, sizeof(nullValue)) == 0) ? true : false;
ret->storeFilterArg(argIndex, filterValue);
}
else
ret->storeFilterArg(argIndex, reinterpret_cast<const T*>(args->val));
{
const auto nullValue = getNullValue<T>(colDataType);
const T* filterValue = reinterpret_cast<const T*>(args->val);
isNullEqCmp =
(args->COP == COMPARE_EQ && memcmp(filterValue, &nullValue, sizeof(nullValue)) == 0) ? true : false;
ret->storeFilterArg(argIndex, filterValue);
}
// IS NULL filtering expression is translated into COMPARE_EQ + NULL magic in the filter.
// This if replaces an operation id once to avoid additional branching in the main loop
// of vectorizedFiltering_ in column.cpp.
// It would be cleaner to place in into EM though.
ret->prestored_cops[argIndex] = (isNullEqCmp) ? COMPARE_NULLEQ : args->COP;
}
/* Decide which structure to use. I think the only cases where we can use the set
@ -575,4 +691,3 @@ boost::shared_ptr<ParsedColumnFilter> _parseColumnFilter(
}
} // namespace primitives