thevar1able/ClickHouse
1
0
Fork 0
mirror of https://github.com/ClickHouse/ClickHouse.git synced 2024-12-03 13:02:00 +00:00
Code Issues Packages Projects Releases Wiki Activity
84a407f381
ClickHouse/src/Functions/FunctionBinaryArithmetic.h
feng lv 53fde56cb7 rewrite bitHammingDistance with FunctionBinaryArithmetic 
fix

fix typo
2022-01-19 09:07:05 +00:00

1928 lines
82 KiB
C++
Raw Blame History

#pragma once
// Include this first, because `#define _asan_poison_address` from
// llvm/Support/Compiler.h conflicts with its forward declaration in
// sanitizer/asan_interface.h
#include <memory>
#include <type_traits>
#include <base/wide_integer_to_string.h>
#include <Columns/ColumnAggregateFunction.h>
#include <Columns/ColumnConst.h>
#include <Columns/ColumnDecimal.h>
#include <Columns/ColumnFixedString.h>
#include <Columns/ColumnNullable.h>
#include <Columns/ColumnString.h>
#include <Columns/ColumnVector.h>
#include <Core/DecimalFunctions.h>
#include <DataTypes/DataTypeAggregateFunction.h>
#include <DataTypes/DataTypeDate.h>
#include <DataTypes/DataTypeDateTime.h>
#include <DataTypes/DataTypeDateTime64.h>
#include <DataTypes/DataTypeFactory.h>
#include <DataTypes/DataTypeFixedString.h>
#include <DataTypes/DataTypeInterval.h>
#include <DataTypes/DataTypeString.h>
#include <DataTypes/DataTypesDecimal.h>
#include <DataTypes/DataTypesNumber.h>
#include <DataTypes/Native.h>
#include <DataTypes/NumberTraits.h>
#include <Functions/DivisionUtils.h>
#include <Functions/FunctionFactory.h>
#include <Functions/FunctionHelpers.h>
#include <Functions/IFunction.h>
#include <Functions/IsOperation.h>
#include <Functions/castTypeToEither.h>
#include <Interpreters/castColumn.h>
#include <base/TypeList.h>
#include <base/map.h>
#include <Common/FieldVisitorsAccurateComparison.h>
#include <Common/assert_cast.h>
#include <Common/typeid_cast.h>
#if USE_EMBEDDED_COMPILER
# pragma GCC diagnostic push
# pragma GCC diagnostic ignored "-Wunused-parameter"
# include <llvm/IR/IRBuilder.h>
# pragma GCC diagnostic pop
#endif
#include <cassert>
namespace DB
{
namespace ErrorCodes
{
extern const int ILLEGAL_COLUMN;
extern const int ILLEGAL_TYPE_OF_ARGUMENT;
extern const int LOGICAL_ERROR;
extern const int DECIMAL_OVERFLOW;
extern const int CANNOT_ADD_DIFFERENT_AGGREGATE_STATES;
extern const int NUMBER_OF_ARGUMENTS_DOESNT_MATCH;
}
namespace traits_
{
struct InvalidType; /// Used to indicate undefined operation
template <bool V, typename T> struct Case : std::bool_constant<V> { using type = T; };
/// Switch<Case<C0, T0>, ...> -- select the first Ti for which Ci is true, InvalidType if none.
template <typename... Ts> using Switch = typename std::disjunction<Ts..., Case<true, InvalidType>>::type;
template <class T>
using DataTypeFromFieldType = std::conditional_t<std::is_same_v<T, NumberTraits::Error>,
InvalidType, DataTypeNumber<T>>;
template <typename DataType> constexpr bool IsIntegral = false;
template <> inline constexpr bool IsIntegral<DataTypeUInt8> = true;
template <> inline constexpr bool IsIntegral<DataTypeUInt16> = true;
template <> inline constexpr bool IsIntegral<DataTypeUInt32> = true;
template <> inline constexpr bool IsIntegral<DataTypeUInt64> = true;
template <> inline constexpr bool IsIntegral<DataTypeInt8> = true;
template <> inline constexpr bool IsIntegral<DataTypeInt16> = true;
template <> inline constexpr bool IsIntegral<DataTypeInt32> = true;
template <> inline constexpr bool IsIntegral<DataTypeInt64> = true;
template <typename DataType> constexpr bool IsExtended = false;
template <> inline constexpr bool IsExtended<DataTypeUInt128> = true;
template <> inline constexpr bool IsExtended<DataTypeUInt256> = true;
template <> inline constexpr bool IsExtended<DataTypeInt128> = true;
template <> inline constexpr bool IsExtended<DataTypeInt256> = true;
template <typename DataType> constexpr bool IsIntegralOrExtended = IsIntegral<DataType> || IsExtended<DataType>;
template <typename DataType> constexpr bool IsIntegralOrExtendedOrDecimal =
IsIntegralOrExtended<DataType> ||
IsDataTypeDecimal<DataType>;
template <typename DataType> constexpr bool IsFloatingPoint = false;
template <> inline constexpr bool IsFloatingPoint<DataTypeFloat32> = true;
template <> inline constexpr bool IsFloatingPoint<DataTypeFloat64> = true;
template <typename DataType> constexpr bool IsDateOrDateTime = false;
template <> inline constexpr bool IsDateOrDateTime<DataTypeDate> = true;
template <> inline constexpr bool IsDateOrDateTime<DataTypeDateTime> = true;
template <typename T0, typename T1> constexpr bool UseLeftDecimal = false;
template <> inline constexpr bool UseLeftDecimal<DataTypeDecimal<Decimal256>, DataTypeDecimal<Decimal128>> = true;
template <> inline constexpr bool UseLeftDecimal<DataTypeDecimal<Decimal256>, DataTypeDecimal<Decimal64>> = true;
template <> inline constexpr bool UseLeftDecimal<DataTypeDecimal<Decimal256>, DataTypeDecimal<Decimal32>> = true;
template <> inline constexpr bool UseLeftDecimal<DataTypeDecimal<Decimal128>, DataTypeDecimal<Decimal32>> = true;
template <> inline constexpr bool UseLeftDecimal<DataTypeDecimal<Decimal128>, DataTypeDecimal<Decimal64>> = true;
template <> inline constexpr bool UseLeftDecimal<DataTypeDecimal<Decimal64>, DataTypeDecimal<Decimal32>> = true;
template <template <typename, typename> class Operation, typename LeftDataType, typename RightDataType>
struct BinaryOperationTraits
{
using T0 = typename LeftDataType::FieldType;
using T1 = typename RightDataType::FieldType;
private: /// it's not correct for Decimal
using Op = Operation<T0, T1>;
public:
static constexpr bool allow_decimal = IsOperation<Operation>::allow_decimal;
/// Appropriate result type for binary operator on numeric types. "Date" can also mean
/// DateTime, but if both operands are Dates, their type must be the same (e.g. Date - DateTime is invalid).
using ResultDataType = Switch<
/// Decimal cases
Case<!allow_decimal && (IsDataTypeDecimal<LeftDataType> || IsDataTypeDecimal<RightDataType>), InvalidType>,
Case<IsDataTypeDecimal<LeftDataType> && IsDataTypeDecimal<RightDataType> && UseLeftDecimal<LeftDataType, RightDataType>, LeftDataType>,
Case<IsDataTypeDecimal<LeftDataType> && IsDataTypeDecimal<RightDataType>, RightDataType>,
Case<IsDataTypeDecimal<LeftDataType> && IsIntegralOrExtended<RightDataType>, LeftDataType>,
Case<IsDataTypeDecimal<RightDataType> && IsIntegralOrExtended<LeftDataType>, RightDataType>,
/// e.g Decimal +-*/ Float, least(Decimal, Float), greatest(Decimal, Float) = Float64
Case<IsOperation<Operation>::allow_decimal && IsDataTypeDecimal<LeftDataType> && IsFloatingPoint<RightDataType>,
DataTypeFloat64>,
Case<IsOperation<Operation>::allow_decimal && IsDataTypeDecimal<RightDataType> && IsFloatingPoint<LeftDataType>,
DataTypeFloat64>,
Case<IsOperation<Operation>::bit_hamming_distance && IsIntegral<LeftDataType> && IsIntegral<RightDataType>,
DataTypeUInt8>,
/// Decimal <op> Real is not supported (traditional DBs convert Decimal <op> Real to Real)
Case<IsDataTypeDecimal<LeftDataType> && !IsIntegralOrExtendedOrDecimal<RightDataType>, InvalidType>,
Case<IsDataTypeDecimal<RightDataType> && !IsIntegralOrExtendedOrDecimal<LeftDataType>, InvalidType>,
/// number <op> number -> see corresponding impl
Case<!IsDateOrDateTime<LeftDataType> && !IsDateOrDateTime<RightDataType>,
DataTypeFromFieldType<typename Op::ResultType>>,
/// Date + Integral -> Date
/// Integral + Date -> Date
Case<IsOperation<Operation>::plus, Switch<
Case<IsIntegral<RightDataType>, LeftDataType>,
Case<IsIntegral<LeftDataType>, RightDataType>>>,
/// Date - Date -> Int32
/// Date - Integral -> Date
Case<IsOperation<Operation>::minus, Switch<
Case<std::is_same_v<LeftDataType, RightDataType>, DataTypeInt32>,
Case<IsDateOrDateTime<LeftDataType> && IsIntegral<RightDataType>, LeftDataType>>>,
/// least(Date, Date) -> Date
/// greatest(Date, Date) -> Date
Case<std::is_same_v<LeftDataType, RightDataType> && (IsOperation<Operation>::least || IsOperation<Operation>::greatest),
LeftDataType>,
/// Date % Int32 -> Int32
/// Date % Float -> Float64
Case<IsOperation<Operation>::modulo, Switch<
Case<IsDateOrDateTime<LeftDataType> && IsIntegral<RightDataType>, RightDataType>,
Case<IsDateOrDateTime<LeftDataType> && IsFloatingPoint<RightDataType>, DataTypeFloat64>>>>;
};
}
namespace impl_
{
/** Arithmetic operations: +, -, *, /, %,
* intDiv (integer division)
* Bitwise operations: |, &, ^, ~.
* Etc.
*/
enum class OpCase { Vector, LeftConstant, RightConstant };
constexpr const auto & undec(const auto & x) { return x; }
constexpr const auto & undec(const is_decimal auto & x) { return x.value; }
template <typename A, typename B, typename Op, typename OpResultType = typename Op::ResultType>
struct BinaryOperation
{
using ResultType = OpResultType;
static const constexpr bool allow_fixed_string = false;
static const constexpr bool allow_string_integer = false;
template <OpCase op_case>
static void NO_INLINE process(const A * __restrict a, const B * __restrict b, ResultType * __restrict c, size_t size, const NullMap * right_nullmap = nullptr)
{
if constexpr (op_case == OpCase::RightConstant)
{
if (right_nullmap && (*right_nullmap)[0])
return;
for (size_t i = 0; i < size; ++i)
c[i] = Op::template apply<ResultType>(a[i], *b);
}
else
{
if (right_nullmap)
{
for (size_t i = 0; i < size; ++i)
if ((*right_nullmap)[i])
c[i] = ResultType();
else
apply<op_case>(a, b, c, i);
}
else
for (size_t i = 0; i < size; ++i)
apply<op_case>(a, b, c, i);
}
}
static ResultType process(A a, B b) { return Op::template apply<ResultType>(a, b); }
private:
template <OpCase op_case>
static inline void apply(const A * __restrict a, const B * __restrict b, ResultType * __restrict c, size_t i)
{
if constexpr (op_case == OpCase::Vector)
c[i] = Op::template apply<ResultType>(a[i], b[i]);
else
c[i] = Op::template apply<ResultType>(*a, b[i]);
}
};
template <typename B, typename Op>
struct StringIntegerOperationImpl
{
static const constexpr bool allow_fixed_string = false;
static const constexpr bool allow_string_integer = true;
template <OpCase op_case>
static void NO_INLINE processFixedString(const UInt8 * __restrict in_vec, const UInt64 n, const B * __restrict b, ColumnFixedString::Chars & out_vec, size_t size)
{
size_t prev_offset = 0;
out_vec.reserve(n * size);
for (size_t i = 0; i < size; ++i)
{
if constexpr (op_case == OpCase::LeftConstant)
{
Op::apply(&in_vec[0], &in_vec[n], b[i], out_vec);
}
else
{
size_t new_offset = prev_offset + n;
if constexpr (op_case == OpCase::Vector)
{
Op::apply(&in_vec[prev_offset], &in_vec[new_offset], b[i], out_vec);
}
else
{
Op::apply(&in_vec[prev_offset], &in_vec[new_offset], b[0], out_vec);
}
prev_offset = new_offset;
}
}
}
template <OpCase op_case>
static void NO_INLINE processString(const UInt8 * __restrict in_vec, const UInt64 * __restrict in_offsets, const B * __restrict b, ColumnString::Chars & out_vec, ColumnString::Offsets & out_offsets, size_t size)
{
size_t prev_offset = 0;
for (size_t i = 0; i < size; ++i)
{
if constexpr (op_case == OpCase::LeftConstant)
{
Op::apply(&in_vec[0], &in_vec[in_offsets[0] - 1], b[i], out_vec, out_offsets);
}
else
{
size_t new_offset = in_offsets[i];
if constexpr (op_case == OpCase::Vector)
{
Op::apply(&in_vec[prev_offset], &in_vec[new_offset - 1], b[i], out_vec, out_offsets);
}
else
{
Op::apply(&in_vec[prev_offset], &in_vec[new_offset - 1], b[0], out_vec, out_offsets);
}
prev_offset = new_offset;
}
}
}
};
template <typename Op>
struct FixedStringOperationImpl
{
template <OpCase op_case>
static void NO_INLINE process(
const UInt8 * __restrict a, const UInt8 * __restrict b, UInt8 * __restrict result,
size_t size, [[maybe_unused]] size_t N)
{
if constexpr (op_case == OpCase::Vector)
for (size_t i = 0; i < size; ++i)
result[i] = Op::template apply<UInt8>(a[i], b[i]);
else if constexpr (op_case == OpCase::LeftConstant)
withConst<true>(b, a, result, size, N);
else
withConst<false>(a, b, result, size, N);
}
private:
template <bool inverted>
static void NO_INLINE withConst(const UInt8 * __restrict a, const UInt8 * __restrict b, UInt8 * __restrict c, size_t size, size_t N)
{
/// These complications are needed to avoid integer division in inner loop.
/// Create a pattern of repeated values of b with at least 16 bytes,
/// so we can read 16 bytes of this repeated pattern starting from any offset inside b.
///
/// Example:
///
/// N = 6
/// ------
/// [abcdefabcdefabcdefabc]
/// ^^^^^^^^^^^^^^^^
/// 16 bytes starting from the last offset inside b.
const size_t b_repeated_size = N + 15;
UInt8 b_repeated[b_repeated_size];
for (size_t i = 0; i < b_repeated_size; ++i)
b_repeated[i] = b[i % N];
size_t b_offset = 0;
const size_t b_increment = 16 % N;
/// Example:
///
/// At first iteration we copy 16 bytes at offset 0 from b_repeated:
/// [abcdefabcdefabcdefabc]
/// ^^^^^^^^^^^^^^^^
/// At second iteration we copy 16 bytes at offset 4 = 16 % 6 from b_repeated:
/// [abcdefabcdefabcdefabc]
/// ^^^^^^^^^^^^^^^^
/// At third iteration we copy 16 bytes at offset 2 = (16 * 2) % 6 from b_repeated:
/// [abcdefabcdefabcdefabc]
/// ^^^^^^^^^^^^^^^^
/// PaddedPODArray allows overflow for 15 bytes.
for (size_t i = 0; i < size; i += 16)
{
/// This loop is formed in a way to be vectorized into two SIMD mov.
for (size_t j = 0; j < 16; ++j)
c[i + j] = inverted
? Op::template apply<UInt8>(a[i + j], b_repeated[b_offset + j])
: Op::template apply<UInt8>(b_repeated[b_offset + j], a[i + j]);
b_offset += b_increment;
if (b_offset >= N) /// This condition is easily predictable.
b_offset -= N;
}
}
};
template <typename A, typename B, typename Op, typename ResultType = typename Op::ResultType>
struct BinaryOperationImpl : BinaryOperation<A, B, Op, ResultType> { };
/**
* Binary operations with Decimals (either Decimal OP Decimal or Decimal Op Float) need to scale the args correctly.
* - + (plus), - (minus), * (multiply), least and greatest operations scale one of the args (which scale factor is not 1).
* The resulting scale is either left or the right scale.
* - / (divide) operation scales the first argument.
* The resulting scale is the first one's.
*/
template <template <typename, typename> typename Operation, class OpResultType, bool check_overflow = true>
struct DecimalBinaryOperation
{
private:
using ResultType = OpResultType; // e.g. Decimal32
using NativeResultType = NativeType<ResultType>; // e.g. UInt32 for Decimal32
using ResultContainerType = typename ColumnVectorOrDecimal<ResultType>::Container;
public:
template <OpCase op_case, bool is_decimal_a, bool is_decimal_b>
static void NO_INLINE process(const auto & a, const auto & b, ResultContainerType & c,
NativeResultType scale_a, NativeResultType scale_b, const NullMap * right_nullmap = nullptr)
{
if constexpr (op_case == OpCase::LeftConstant) static_assert(!is_decimal<decltype(a)>);
if constexpr (op_case == OpCase::RightConstant) static_assert(!is_decimal<decltype(b)>);
size_t size;
if constexpr (op_case == OpCase::LeftConstant)
size = b.size();
else
size = a.size();
if constexpr (is_plus_minus_compare)
{
if (scale_a != 1)
{
for (size_t i = 0; i < size; ++i)
c[i] = applyScaled<true>(
unwrap<op_case, OpCase::LeftConstant>(a, i),
unwrap<op_case, OpCase::RightConstant>(b, i),
scale_a);
return;
}
else if (scale_b != 1)
{
for (size_t i = 0; i < size; ++i)
c[i] = applyScaled<false>(
unwrap<op_case, OpCase::LeftConstant>(a, i),
unwrap<op_case, OpCase::RightConstant>(b, i),
scale_b);
return;
}
}
else if constexpr (is_multiply)
{
if (scale_a != 1)
{
for (size_t i = 0; i < size; ++i)
c[i] = applyScaled<true, false>(
unwrap<op_case, OpCase::LeftConstant>(a, i),
unwrap<op_case, OpCase::RightConstant>(b, i),
scale_a);
return;
}
else if (scale_b != 1)
{
for (size_t i = 0; i < size; ++i)
c[i] = applyScaled<false, false>(
unwrap<op_case, OpCase::LeftConstant>(a, i),
unwrap<op_case, OpCase::RightConstant>(b, i),
scale_b);
return;
}
}
else if constexpr (is_division && is_decimal_b)
{
processWithRightNullmapImpl<op_case>(a, b, c, size, right_nullmap, [&scale_a](const auto & left, const auto & right)
{
return applyScaledDiv<is_decimal_a>(left, right, scale_a);
});
return;
}
processWithRightNullmapImpl<op_case>(a, b, c, size, right_nullmap, [](const auto & left, const auto & right){ return apply(left, right); });
}
template <bool is_decimal_a, bool is_decimal_b, class A, class B>
static ResultType process(A a, B b, NativeResultType scale_a, NativeResultType scale_b)
requires(!is_decimal<A> && !is_decimal<B>)
{
if constexpr (is_division && is_decimal_b)
return applyScaledDiv<is_decimal_a>(a, b, scale_a);
else if constexpr (is_plus_minus_compare)
{
if (scale_a != 1)
return applyScaled<true>(a, b, scale_a);
if (scale_b != 1)
return applyScaled<false>(a, b, scale_b);
}
return apply(a, b);
}
private:
template <OpCase op_case, typename ApplyFunc>
static inline void processWithRightNullmapImpl(const auto & a, const auto & b, ResultContainerType & c, size_t size, const NullMap * right_nullmap, ApplyFunc apply_func)
{
if (right_nullmap)
{
if constexpr (op_case == OpCase::RightConstant)
{
if ((*right_nullmap)[0])
return;
for (size_t i = 0; i < size; ++i)
c[i] = apply_func(undec(a[i]), undec(b));
}
else
{
for (size_t i = 0; i < size; ++i)
{
if ((*right_nullmap)[i])
c[i] = ResultType();
else
c[i] = apply_func(unwrap<op_case, OpCase::LeftConstant>(a, i), undec(b[i]));
}
}
}
else
for (size_t i = 0; i < size; ++i)
c[i] = apply_func(unwrap<op_case, OpCase::LeftConstant>(a, i), unwrap<op_case, OpCase::RightConstant>(b, i));
}
static constexpr bool is_plus_minus = IsOperation<Operation>::plus ||
IsOperation<Operation>::minus;
static constexpr bool is_multiply = IsOperation<Operation>::multiply;
static constexpr bool is_float_division = IsOperation<Operation>::div_floating;
static constexpr bool is_int_division = IsOperation<Operation>::div_int ||
IsOperation<Operation>::div_int_or_zero;
static constexpr bool is_division = is_float_division || is_int_division;
static constexpr bool is_compare = IsOperation<Operation>::least ||
IsOperation<Operation>::greatest;
static constexpr bool is_plus_minus_compare = is_plus_minus || is_compare;
static constexpr bool can_overflow = is_plus_minus || is_multiply;
using Op = std::conditional_t<is_float_division,
DivideIntegralImpl<NativeResultType, NativeResultType>, /// substitute divide by intDiv (throw on division by zero)
Operation<NativeResultType, NativeResultType>>;
template <OpCase op_case, OpCase target, class E>
static auto unwrap(const E& elem, size_t i)
{
if constexpr (op_case == target)
return undec(elem);
else
return undec(elem[i]);
}
/// there's implicit type conversion here
static NativeResultType apply(NativeResultType a, NativeResultType b)
{
if constexpr (can_overflow && check_overflow)
{
NativeResultType res;
if (Op::template apply<NativeResultType>(a, b, res))
throw Exception("Decimal math overflow", ErrorCodes::DECIMAL_OVERFLOW);
return res;
}
else
return Op::template apply<NativeResultType>(a, b);
}
template <bool scale_left, bool may_check_overflow = true>
static NO_SANITIZE_UNDEFINED NativeResultType applyScaled(NativeResultType a, NativeResultType b, NativeResultType scale)
{
static_assert(is_plus_minus_compare || is_multiply);
NativeResultType res;
if constexpr (check_overflow && may_check_overflow)
{
bool overflow = false;
if constexpr (scale_left)
overflow |= common::mulOverflow(a, scale, a);
else
overflow |= common::mulOverflow(b, scale, b);
if constexpr (can_overflow)
overflow |= Op::template apply<NativeResultType>(a, b, res);
else
res = Op::template apply<NativeResultType>(a, b);
if (overflow)
throw Exception("Decimal math overflow", ErrorCodes::DECIMAL_OVERFLOW);
}
else
{
if constexpr (scale_left)
a *= scale;
else
b *= scale;
res = Op::template apply<NativeResultType>(a, b);
}
return res;
}
template <bool is_decimal_a>
static NO_SANITIZE_UNDEFINED NativeResultType applyScaledDiv(NativeResultType a, NativeResultType b, NativeResultType scale)
{
if constexpr (is_division)
{
if constexpr (check_overflow)
{
bool overflow = false;
if constexpr (!is_decimal_a)
overflow |= common::mulOverflow(scale, scale, scale);
overflow |= common::mulOverflow(a, scale, a);
if (overflow)
throw Exception("Decimal math overflow", ErrorCodes::DECIMAL_OVERFLOW);
}
else
{
if constexpr (!is_decimal_a)
scale *= scale;
a *= scale;
}
return Op::template apply<NativeResultType>(a, b);
}
}
};
}
using namespace traits_;
using namespace impl_;
template <template <typename, typename> class Op, typename Name, bool valid_on_default_arguments = true, bool valid_on_float_arguments = true, bool division_by_nullable = false>
class FunctionBinaryArithmetic : public IFunction
{
static constexpr const bool is_plus = IsOperation<Op>::plus;
static constexpr const bool is_minus = IsOperation<Op>::minus;
static constexpr const bool is_multiply = IsOperation<Op>::multiply;
static constexpr const bool is_division = IsOperation<Op>::division;
ContextPtr context;
bool check_decimal_overflow = true;
static bool castType(const IDataType * type, auto && f)
{
using Types = TypeList<
DataTypeUInt8, DataTypeUInt16, DataTypeUInt32, DataTypeUInt64, DataTypeUInt128, DataTypeUInt256,
DataTypeInt8, DataTypeInt16, DataTypeInt32, DataTypeInt64, DataTypeInt128, DataTypeInt256,
DataTypeDecimal32, DataTypeDecimal64, DataTypeDecimal128, DataTypeDecimal256,
DataTypeDate, DataTypeDateTime,
DataTypeFixedString, DataTypeString>;
using Floats = TypeList<DataTypeFloat32, DataTypeFloat64>;
using ValidTypes = std::conditional_t<valid_on_float_arguments,
TypeListConcat<Types, Floats>,
Types>;
return castTypeToEither(ValidTypes{}, type, std::forward<decltype(f)>(f));
}
template <typename F>
static bool castBothTypes(const IDataType * left, const IDataType * right, F && f)
{
return castType(left, [&](const auto & left_)
{
return castType(right, [&](const auto & right_)
{
return f(left_, right_);
});
});
}
static FunctionOverloadResolverPtr
getFunctionForIntervalArithmetic(const DataTypePtr & type0, const DataTypePtr & type1, ContextPtr context)
{
bool first_is_date_or_datetime = isDate(type0) || isDateTime(type0) || isDateTime64(type0);
bool second_is_date_or_datetime = isDate(type1) || isDateTime(type1) || isDateTime64(type1);
/// Exactly one argument must be Date or DateTime
if (first_is_date_or_datetime == second_is_date_or_datetime)
return {};
/// Special case when the function is plus or minus, one of arguments is Date/DateTime and another is Interval.
/// We construct another function (example: addMonths) and call it.
if constexpr (!is_plus && !is_minus)
return {};
const DataTypePtr & type_time = first_is_date_or_datetime ? type0 : type1;
const DataTypePtr & type_interval = first_is_date_or_datetime ? type1 : type0;
bool interval_is_number = isNumber(type_interval);
const DataTypeInterval * interval_data_type = nullptr;
if (!interval_is_number)
{
interval_data_type = checkAndGetDataType<DataTypeInterval>(type_interval.get());
if (!interval_data_type)
return {};
}
if (second_is_date_or_datetime && is_minus)
throw Exception("Wrong order of arguments for function " + String(name) + ": argument of type Interval cannot be first",
ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
std::string function_name;
if (interval_data_type)
{
function_name = fmt::format("{}{}s",
is_plus ? "add" : "subtract",
interval_data_type->getKind().toString());
}
else
{
if (isDate(type_time))
function_name = is_plus ? "addDays" : "subtractDays";
else
function_name = is_plus ? "addSeconds" : "subtractSeconds";
}
return FunctionFactory::instance().get(function_name, context);
}
static FunctionOverloadResolverPtr
getFunctionForTupleArithmetic(const DataTypePtr & type0, const DataTypePtr & type1, ContextPtr context)
{
if (!isTuple(type0) || !isTuple(type1))
return {};
/// Special case when the function is plus, minus or multiply, both arguments are tuples.
/// We construct another function (example: tuplePlus) and call it.
if constexpr (!is_plus && !is_minus && !is_multiply)
return {};
std::string function_name;
if (is_plus)
{
function_name = "tuplePlus";
}
else if (is_minus)
{
function_name = "tupleMinus";
}
else
{
function_name = "dotProduct";
}
return FunctionFactory::instance().get(function_name, context);
}
static FunctionOverloadResolverPtr
getFunctionForTupleAndNumberArithmetic(const DataTypePtr & type0, const DataTypePtr & type1, ContextPtr context)
{
if (!(isTuple(type0) && isNumber(type1)) && !(isTuple(type1) && isNumber(type0)))
return {};
/// Special case when the function is multiply or divide, one of arguments is Tuple and another is Number.
/// We construct another function (example: tupleMultiplyByNumber) and call it.
if constexpr (!is_multiply && !is_division)
return {};
if (isNumber(type0) && is_division)
throw Exception("Wrong order of arguments for function " + String(name) + ": argument of numeric type cannot be first",
ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
std::string function_name;
if (is_multiply)
{
function_name = "tupleMultiplyByNumber";
}
else
{
function_name = "tupleDivideByNumber";
}
return FunctionFactory::instance().get(function_name, context);
}
static bool isAggregateMultiply(const DataTypePtr & type0, const DataTypePtr & type1)
{
if constexpr (!is_multiply)
return false;
WhichDataType which0(type0);
WhichDataType which1(type1);
return (which0.isAggregateFunction() && which1.isNativeUInt())
|| (which0.isNativeUInt() && which1.isAggregateFunction());
}
static bool isAggregateAddition(const DataTypePtr & type0, const DataTypePtr & type1)
{
if constexpr (!is_plus)
return false;
WhichDataType which0(type0);
WhichDataType which1(type1);
return which0.isAggregateFunction() && which1.isAggregateFunction();
}
/// Multiply aggregation state by integer constant: by merging it with itself specified number of times.
ColumnPtr executeAggregateMultiply(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const
{
ColumnsWithTypeAndName new_arguments = arguments;
if (WhichDataType(new_arguments[1].type).isAggregateFunction())
std::swap(new_arguments[0], new_arguments[1]);
if (!isColumnConst(*new_arguments[1].column))
throw Exception{"Illegal column " + new_arguments[1].column->getName()
+ " of argument of aggregation state multiply. Should be integer constant", ErrorCodes::ILLEGAL_COLUMN};
const IColumn & agg_state_column = *new_arguments[0].column;
bool agg_state_is_const = isColumnConst(agg_state_column);
const ColumnAggregateFunction & column = typeid_cast<const ColumnAggregateFunction &>(
agg_state_is_const ? assert_cast<const ColumnConst &>(agg_state_column).getDataColumn() : agg_state_column);
AggregateFunctionPtr function = column.getAggregateFunction();
size_t size = agg_state_is_const ? 1 : input_rows_count;
auto column_to = ColumnAggregateFunction::create(function);
column_to->reserve(size);
auto column_from = ColumnAggregateFunction::create(function);
column_from->reserve(size);
for (size_t i = 0; i < size; ++i)
{
column_to->insertDefault();
column_from->insertFrom(column.getData()[i]);
}
auto & vec_to = column_to->getData();
auto & vec_from = column_from->getData();
UInt64 m = typeid_cast<const ColumnConst *>(new_arguments[1].column.get())->getValue<UInt64>();
// Since we merge the function states by ourselves, we have to have an
// Arena for this. Pass it to the resulting column so that the arena
// has a proper lifetime.
auto arena = std::make_shared<Arena>();
column_to->addArena(arena);
/// We use exponentiation by squaring algorithm to perform multiplying aggregate states by N in O(log(N)) operations
/// https://en.wikipedia.org/wiki/Exponentiation_by_squaring
while (m)
{
if (m % 2)
{
for (size_t i = 0; i < size; ++i)
function->merge(vec_to[i], vec_from[i], arena.get());
--m;
}
else
{
for (size_t i = 0; i < size; ++i)
function->merge(vec_from[i], vec_from[i], arena.get());
m /= 2;
}
}
if (agg_state_is_const)
return ColumnConst::create(std::move(column_to), input_rows_count);
else
return column_to;
}
/// Merge two aggregation states together.
ColumnPtr executeAggregateAddition(const ColumnsWithTypeAndName & arguments, const DataTypePtr &, size_t input_rows_count) const
{
const IColumn & lhs_column = *arguments[0].column;
const IColumn & rhs_column = *arguments[1].column;
bool lhs_is_const = isColumnConst(lhs_column);
bool rhs_is_const = isColumnConst(rhs_column);
const ColumnAggregateFunction & lhs = typeid_cast<const ColumnAggregateFunction &>(
lhs_is_const ? assert_cast<const ColumnConst &>(lhs_column).getDataColumn() : lhs_column);
const ColumnAggregateFunction & rhs = typeid_cast<const ColumnAggregateFunction &>(
rhs_is_const ? assert_cast<const ColumnConst &>(rhs_column).getDataColumn() : rhs_column);
AggregateFunctionPtr function = lhs.getAggregateFunction();
size_t size = (lhs_is_const && rhs_is_const) ? 1 : input_rows_count;
auto column_to = ColumnAggregateFunction::create(function);
column_to->reserve(size);
for (size_t i = 0; i < size; ++i)
{
column_to->insertFrom(lhs.getData()[lhs_is_const ? 0 : i]);
column_to->insertMergeFrom(rhs.getData()[rhs_is_const ? 0 : i]);
}
if (lhs_is_const && rhs_is_const)
return ColumnConst::create(std::move(column_to), input_rows_count);
else
return column_to;
}
ColumnPtr executeDateTimeIntervalPlusMinus(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type,
size_t input_rows_count, const FunctionOverloadResolverPtr & function_builder) const
{
ColumnsWithTypeAndName new_arguments = arguments;
/// Interval argument must be second.
if (isDate(arguments[1].type) || isDateTime(arguments[1].type) || isDateTime64(arguments[1].type))
std::swap(new_arguments[0], new_arguments[1]);
/// Change interval argument type to its representation
new_arguments[1].type = std::make_shared<DataTypeNumber<DataTypeInterval::FieldType>>();
auto function = function_builder->build(new_arguments);
return function->execute(new_arguments, result_type, input_rows_count);
}
ColumnPtr executeTupleNumberOperator(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type,
size_t input_rows_count, const FunctionOverloadResolverPtr & function_builder) const
{
ColumnsWithTypeAndName new_arguments = arguments;
/// Number argument must be second.
if (isNumber(arguments[0].type))
std::swap(new_arguments[0], new_arguments[1]);
auto function = function_builder->build(new_arguments);
return function->execute(new_arguments, result_type, input_rows_count);
}
template <typename T, typename ResultDataType>
static auto helperGetOrConvert(const auto & col_const, const auto & col)
{
using ResultType = typename ResultDataType::FieldType;
using NativeResultType = NativeType<ResultType>;
if constexpr (IsFloatingPoint<ResultDataType> && is_decimal<T>)
return DecimalUtils::convertTo<NativeResultType>(col_const->template getValue<T>(), col.getScale());
else if constexpr (is_decimal<T>)
return col_const->template getValue<T>().value;
else
return col_const->template getValue<T>();
}
template <OpCase op_case, bool left_decimal, bool right_decimal, typename OpImpl, typename OpImplCheck>
void helperInvokeEither(const auto& left, const auto& right, auto& vec_res, auto scale_a, auto scale_b, const NullMap * right_nullmap) const
{
if (check_decimal_overflow)
OpImplCheck::template process<op_case, left_decimal, right_decimal>(left, right, vec_res, scale_a, scale_b, right_nullmap);
else
OpImpl::template process<op_case, left_decimal, right_decimal>(left, right, vec_res, scale_a, scale_b, right_nullmap);
}
template <class LeftDataType, class RightDataType, class ResultDataType>
ColumnPtr executeNumericWithDecimal(
const auto & left, const auto & right,
const ColumnConst * const col_left_const, const ColumnConst * const col_right_const,
const auto * const col_left, const auto * const col_right,
size_t col_left_size, const NullMap * right_nullmap) const
{
using T0 = typename LeftDataType::FieldType;
using T1 = typename RightDataType::FieldType;
using ResultType = typename ResultDataType::FieldType;
using NativeResultType = NativeType<ResultType>;
using OpImpl = DecimalBinaryOperation<Op, ResultType, false>;
using OpImplCheck = DecimalBinaryOperation<Op, ResultType, true>;
using ColVecResult = ColumnVectorOrDecimal<ResultType>;
static constexpr const bool left_is_decimal = is_decimal<T0>;
static constexpr const bool right_is_decimal = is_decimal<T1>;
typename ColVecResult::MutablePtr col_res = nullptr;
const ResultDataType type = decimalResultType<is_multiply, is_division>(left, right);
const ResultType scale_a = [&]
{
if constexpr (IsDataTypeDecimal<RightDataType> && is_division)
return right.getScaleMultiplier(); // the division impl uses only the scale_a
else
{
if constexpr (is_multiply)
// the decimal impl uses scales, but if the result is decimal, both of the arguments are decimal,
// so they would multiply correctly, so we need to scale the result to the neutral element (1).
// The explicit type is needed as the int (in contrast with float) can't be implicitly converted
// to decimal.
return ResultType{1};
else
return type.scaleFactorFor(left, false);
}
}();
const ResultType scale_b = [&]
{
if constexpr (is_multiply)
return ResultType{1};
else
return type.scaleFactorFor(right, is_division);
}();
/// non-vector result
if (col_left_const && col_right_const)
{
const NativeResultType const_a = helperGetOrConvert<T0, ResultDataType>(col_left_const, left);
const NativeResultType const_b = helperGetOrConvert<T1, ResultDataType>(col_right_const, right);
ResultType res = {};
if (!right_nullmap || !(*right_nullmap)[0])
res = check_decimal_overflow
? OpImplCheck::template process<left_is_decimal, right_is_decimal>(const_a, const_b, scale_a, scale_b)
: OpImpl::template process<left_is_decimal, right_is_decimal>(const_a, const_b, scale_a, scale_b);
return ResultDataType(type.getPrecision(), type.getScale())
.createColumnConst(col_left_const->size(), toField(res, type.getScale()));
}
col_res = ColVecResult::create(0, type.getScale());
auto & vec_res = col_res->getData();
vec_res.resize(col_left_size);
if (col_left && col_right)
{
helperInvokeEither<OpCase::Vector, left_is_decimal, right_is_decimal, OpImpl, OpImplCheck>(
col_left->getData(), col_right->getData(), vec_res, scale_a, scale_b, right_nullmap);
}
else if (col_left_const && col_right)
{
const NativeResultType const_a = helperGetOrConvert<T0, ResultDataType>(col_left_const, left);
helperInvokeEither<OpCase::LeftConstant, left_is_decimal, right_is_decimal, OpImpl, OpImplCheck>(
const_a, col_right->getData(), vec_res, scale_a, scale_b, right_nullmap);
}
else if (col_left && col_right_const)
{
const NativeResultType const_b = helperGetOrConvert<T1, ResultDataType>(col_right_const, right);
helperInvokeEither<OpCase::RightConstant, left_is_decimal, right_is_decimal, OpImpl, OpImplCheck>(
col_left->getData(), const_b, vec_res, scale_a, scale_b, right_nullmap);
}
else
return nullptr;
return col_res;
}
public:
static constexpr auto name = Name::name;
static FunctionPtr create(ContextPtr context) { return std::make_shared<FunctionBinaryArithmetic>(context); }
explicit FunctionBinaryArithmetic(ContextPtr context_)
: context(context_),
check_decimal_overflow(decimalCheckArithmeticOverflow(context))
{}
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 2; }
bool useDefaultImplementationForNulls() const override
{
/// We shouldn't use default implementation for nulls for the case when operation is divide,
/// intDiv or modulo and denominator is Nullable(Something), because it may cause division
/// by zero error (when value is Null we store default value 0 in nested column).
return !division_by_nullable;
}
bool isSuitableForShortCircuitArgumentsExecution(const DataTypesWithConstInfo & arguments) const override
{
return ((IsOperation<Op>::div_int || IsOperation<Op>::modulo) && !arguments[1].is_const)
|| (IsOperation<Op>::div_floating && (isDecimalOrNullableDecimal(arguments[0].type) || isDecimalOrNullableDecimal(arguments[1].type)));
}
DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
{
return getReturnTypeImplStatic(arguments, context);
}
static DataTypePtr getReturnTypeImplStatic(const DataTypes & arguments, ContextPtr context)
{
/// Special case when multiply aggregate function state
if (isAggregateMultiply(arguments[0], arguments[1]))
{
if (WhichDataType(arguments[0]).isAggregateFunction())
return arguments[0];
return arguments[1];
}
/// Special case - addition of two aggregate functions states
if (isAggregateAddition(arguments[0], arguments[1]))
{
if (!arguments[0]->equals(*arguments[1]))
throw Exception("Cannot add aggregate states of different functions: "
+ arguments[0]->getName() + " and " + arguments[1]->getName(), ErrorCodes::CANNOT_ADD_DIFFERENT_AGGREGATE_STATES);
return arguments[0];
}
/// Special case when the function is plus or minus, one of arguments is Date/DateTime and another is Interval.
if (auto function_builder = getFunctionForIntervalArithmetic(arguments[0], arguments[1], context))
{
ColumnsWithTypeAndName new_arguments(2);
for (size_t i = 0; i < 2; ++i)
new_arguments[i].type = arguments[i];
/// Interval argument must be second.
if (isDate(new_arguments[1].type) || isDateTime(new_arguments[1].type) || isDateTime64(new_arguments[1].type))
std::swap(new_arguments[0], new_arguments[1]);
/// Change interval argument to its representation
new_arguments[1].type = std::make_shared<DataTypeNumber<DataTypeInterval::FieldType>>();
auto function = function_builder->build(new_arguments);
return function->getResultType();
}
/// Special case when the function is plus, minus or multiply, both arguments are tuples.
if (auto function_builder = getFunctionForTupleArithmetic(arguments[0], arguments[1], context))
{
ColumnsWithTypeAndName new_arguments(2);
for (size_t i = 0; i < 2; ++i)
new_arguments[i].type = arguments[i];
auto function = function_builder->build(new_arguments);
return function->getResultType();
}
/// Special case when the function is multiply or divide, one of arguments is Tuple and another is Number.
if (auto function_builder = getFunctionForTupleAndNumberArithmetic(arguments[0], arguments[1], context))
{
ColumnsWithTypeAndName new_arguments(2);
for (size_t i = 0; i < 2; ++i)
new_arguments[i].type = arguments[i];
/// Number argument must be second.
if (isNumber(new_arguments[0].type))
std::swap(new_arguments[0], new_arguments[1]);
auto function = function_builder->build(new_arguments);
return function->getResultType();
}
DataTypePtr type_res;
const bool valid = castBothTypes(arguments[0].get(), arguments[1].get(), [&](const auto & left, const auto & right)
{
using LeftDataType = std::decay_t<decltype(left)>;
using RightDataType = std::decay_t<decltype(right)>;
if constexpr ((std::is_same_v<DataTypeFixedString, LeftDataType> || std::is_same_v<DataTypeString, LeftDataType>) ||
(std::is_same_v<DataTypeFixedString, RightDataType> || std::is_same_v<DataTypeString, RightDataType>))
{
if constexpr (std::is_same_v<DataTypeFixedString, LeftDataType> &&
std::is_same_v<DataTypeFixedString, RightDataType>)
{
if constexpr (!Op<DataTypeFixedString, DataTypeFixedString>::allow_fixed_string)
return false;
else
{
if (left.getN() == right.getN())
{
type_res = std::make_shared<LeftDataType>(left.getN());
return true;
}
}
}
if constexpr (!Op<LeftDataType, RightDataType>::allow_string_integer)
return false;
else if constexpr (!IsIntegral<RightDataType>)
return false;
else if constexpr (std::is_same_v<DataTypeFixedString, LeftDataType>)
type_res = std::make_shared<LeftDataType>(left.getN());
else
type_res = std::make_shared<DataTypeString>();
return true;
}
else
{
using ResultDataType = typename BinaryOperationTraits<Op, LeftDataType, RightDataType>::ResultDataType;
if constexpr (!std::is_same_v<ResultDataType, InvalidType>)
{
if constexpr (IsDataTypeDecimal<LeftDataType> && IsDataTypeDecimal<RightDataType>)
{
ResultDataType result_type = decimalResultType<is_multiply, is_division>(left, right);
type_res = std::make_shared<ResultDataType>(result_type.getPrecision(), result_type.getScale());
}
else if constexpr ((IsDataTypeDecimal<LeftDataType> && IsFloatingPoint<RightDataType>) ||
(IsDataTypeDecimal<RightDataType> && IsFloatingPoint<LeftDataType>))
type_res = std::make_shared<DataTypeFloat64>();
else if constexpr (IsDataTypeDecimal<LeftDataType>)
type_res = std::make_shared<LeftDataType>(left.getPrecision(), left.getScale());
else if constexpr (IsDataTypeDecimal<RightDataType>)
type_res = std::make_shared<RightDataType>(right.getPrecision(), right.getScale());
else if constexpr (std::is_same_v<ResultDataType, DataTypeDateTime>)
{
// Special case for DateTime: binary OPS should reuse timezone
// of DateTime argument as timezeone of result type.
// NOTE: binary plus/minus are not allowed on DateTime64, and we are not handling it here.
const TimezoneMixin * tz = nullptr;
if constexpr (std::is_same_v<RightDataType, DataTypeDateTime>)
tz = &right;
if constexpr (std::is_same_v<LeftDataType, DataTypeDateTime>)
tz = &left;
type_res = std::make_shared<ResultDataType>(*tz);
}
else
type_res = std::make_shared<ResultDataType>();
return true;
}
}
return false;
});
if (valid)
return type_res;
throw Exception(
"Illegal types " + arguments[0]->getName() +
" and " + arguments[1]->getName() +
" of arguments of function " + String(name),
ErrorCodes::ILLEGAL_TYPE_OF_ARGUMENT);
}
ColumnPtr executeFixedString(const ColumnsWithTypeAndName & arguments) const
{
using OpImpl = FixedStringOperationImpl<Op<UInt8, UInt8>>;
const auto * const col_left_raw = arguments[0].column.get();
const auto * const col_right_raw = arguments[1].column.get();
if (const auto * col_left_const = checkAndGetColumnConst<ColumnFixedString>(col_left_raw))
{
if (const auto * col_right_const = checkAndGetColumnConst<ColumnFixedString>(col_right_raw))
{
const auto * col_left = checkAndGetColumn<ColumnFixedString>(col_left_const->getDataColumn());
const auto * col_right = checkAndGetColumn<ColumnFixedString>(col_right_const->getDataColumn());
if (col_left->getN() != col_right->getN())
return nullptr;
auto col_res = ColumnFixedString::create(col_left->getN());
auto & out_chars = col_res->getChars();
out_chars.resize(col_left->getN());
OpImpl::template process<OpCase::Vector>(
col_left->getChars().data(),
col_right->getChars().data(),
out_chars.data(),
out_chars.size(), {});
return ColumnConst::create(std::move(col_res), col_left_raw->size());
}
}
const bool is_left_column_const = checkAndGetColumnConst<ColumnFixedString>(col_left_raw) != nullptr;
const bool is_right_column_const = checkAndGetColumnConst<ColumnFixedString>(col_right_raw) != nullptr;
const auto * col_left = is_left_column_const
? checkAndGetColumn<ColumnFixedString>(
checkAndGetColumnConst<ColumnFixedString>(col_left_raw)->getDataColumn())
: checkAndGetColumn<ColumnFixedString>(col_left_raw);
const auto * col_right = is_right_column_const
? checkAndGetColumn<ColumnFixedString>(
checkAndGetColumnConst<ColumnFixedString>(col_right_raw)->getDataColumn())
: checkAndGetColumn<ColumnFixedString>(col_right_raw);
if (col_left && col_right)
{
if (col_left->getN() != col_right->getN())
return nullptr;
auto col_res = ColumnFixedString::create(col_left->getN());
auto & out_chars = col_res->getChars();
out_chars.resize((is_right_column_const ? col_left->size() : col_right->size()) * col_left->getN());
if (!is_left_column_const && !is_right_column_const)
{
OpImpl::template process<OpCase::Vector>(
col_left->getChars().data(),
col_right->getChars().data(),
out_chars.data(),
out_chars.size(), {});
}
else if (is_left_column_const)
{
OpImpl::template process<OpCase::LeftConstant>(
col_left->getChars().data(),
col_right->getChars().data(),
out_chars.data(),
out_chars.size(),
col_left->getN());
}
else
{
OpImpl::template process<OpCase::RightConstant>(
col_left->getChars().data(),
col_right->getChars().data(),
out_chars.data(),
out_chars.size(),
col_left->getN());
}
return col_res;
}
return nullptr;
}
template <typename LeftColumnType, typename A, typename B>
ColumnPtr executeStringInteger(const ColumnsWithTypeAndName & arguments, const A & left, const B & right) const
{
using LeftDataType = std::decay_t<decltype(left)>;
using RightDataType = std::decay_t<decltype(right)>;
const auto * const col_left_raw = arguments[0].column.get();
const auto * const col_right_raw = arguments[1].column.get();
using T1 = typename RightDataType::FieldType;
using ColVecT1 = ColumnVector<T1>;
const ColVecT1 * const col_right = checkAndGetColumn<ColVecT1>(col_right_raw);
const ColumnConst * const col_right_const = checkAndGetColumnConst<ColVecT1>(col_right_raw);
using OpImpl = StringIntegerOperationImpl<T1, Op<LeftDataType, T1>>;
const ColumnConst * const col_left_const = checkAndGetColumnConst<LeftColumnType>(col_left_raw);
const auto * col_left = col_left_const ? checkAndGetColumn<LeftColumnType>(col_left_const->getDataColumn())
: checkAndGetColumn<LeftColumnType>(col_left_raw);
if (!col_left)
return nullptr;
const typename LeftColumnType::Chars & in_vec = col_left->getChars();
typename LeftColumnType::MutablePtr col_res;
if constexpr (std::is_same_v<LeftDataType, DataTypeFixedString>)
col_res = LeftColumnType::create(col_left->getN());
else
col_res = LeftColumnType::create();
typename LeftColumnType::Chars & out_vec = col_res->getChars();
if (col_left_const && col_right_const)
{
const T1 value = col_right_const->template getValue<T1>();
if constexpr (std::is_same_v<LeftDataType, DataTypeFixedString>)
{
OpImpl::template processFixedString<OpCase::Vector>(in_vec.data(), col_left->getN(), &value, out_vec, 1);
}
else
{
ColumnString::Offsets & out_offsets = col_res->getOffsets();
OpImpl::template processString<OpCase::Vector>(in_vec.data(), col_left->getOffsets().data(), &value, out_vec, out_offsets, 1);
}
return ColumnConst::create(std::move(col_res), col_left->size());
}
else if (!col_left_const && !col_right_const && col_right)
{
if constexpr (std::is_same_v<LeftDataType, DataTypeFixedString>)
{
OpImpl::template processFixedString<OpCase::Vector>(in_vec.data(), col_left->getN(), col_right->getData().data(), out_vec, col_left->size());
}
else
{
ColumnString::Offsets & out_offsets = col_res->getOffsets();
out_offsets.reserve(col_left->size());
OpImpl::template processString<OpCase::Vector>(
in_vec.data(), col_left->getOffsets().data(), col_right->getData().data(), out_vec, out_offsets, col_left->size());
}
}
else if (col_left_const && col_right)
{
if constexpr (std::is_same_v<LeftDataType, DataTypeFixedString>)
{
OpImpl::template processFixedString<OpCase::LeftConstant>(
in_vec.data(), col_left->getN(), col_right->getData().data(), out_vec, col_right->size());
}
else
{
ColumnString::Offsets & out_offsets = col_res->getOffsets();
out_offsets.reserve(col_right->size());
OpImpl::template processString<OpCase::LeftConstant>(
in_vec.data(), col_left->getOffsets().data(), col_right->getData().data(), out_vec, out_offsets, col_right->size());
}
}
else if (col_right_const)
{
const T1 value = col_right_const->template getValue<T1>();
if constexpr (std::is_same_v<LeftDataType, DataTypeFixedString>)
{
OpImpl::template processFixedString<OpCase::RightConstant>(in_vec.data(), col_left->getN(), &value, out_vec, col_left->size());
}
else
{
ColumnString::Offsets & out_offsets = col_res->getOffsets();
out_offsets.reserve(col_left->size());
OpImpl::template processString<OpCase::RightConstant>(
in_vec.data(), col_left->getOffsets().data(), &value, out_vec, out_offsets, col_left->size());
}
}
else
return nullptr;
return col_res;
}
template <typename A, typename B>
ColumnPtr executeNumeric(const ColumnsWithTypeAndName & arguments, const A & left, const B & right, const NullMap * right_nullmap) const
{
using LeftDataType = std::decay_t<decltype(left)>;
using RightDataType = std::decay_t<decltype(right)>;
using ResultDataType = typename BinaryOperationTraits<Op, LeftDataType, RightDataType>::ResultDataType;
if constexpr (std::is_same_v<ResultDataType, InvalidType>)
return nullptr;
else // we can't avoid the else because otherwise the compiler may assume the ResultDataType may be Invalid
// and that would produce the compile error.
{
constexpr bool decimal_with_float = (IsDataTypeDecimal<LeftDataType> && IsFloatingPoint<RightDataType>)
|| (IsFloatingPoint<LeftDataType> && IsDataTypeDecimal<RightDataType>);
using T0 = std::conditional_t<decimal_with_float, Float64, typename LeftDataType::FieldType>;
using T1 = std::conditional_t<decimal_with_float, Float64, typename RightDataType::FieldType>;
using ResultType = typename ResultDataType::FieldType;
using ColVecT0 = ColumnVectorOrDecimal<T0>;
using ColVecT1 = ColumnVectorOrDecimal<T1>;
using ColVecResult = ColumnVectorOrDecimal<ResultType>;
ColumnPtr left_col = nullptr;
ColumnPtr right_col = nullptr;
/// When Decimal op Float32/64, convert both of them into Float64
if constexpr (decimal_with_float)
{
const auto converted_type = std::make_shared<DataTypeFloat64>();
left_col = castColumn(arguments[0], converted_type);
right_col = castColumn(arguments[1], converted_type);
}
else
{
left_col = arguments[0].column;
right_col = arguments[1].column;
}
const auto * const col_left_raw = left_col.get();
const auto * const col_right_raw = right_col.get();
const size_t col_left_size = col_left_raw->size();
const ColumnConst * const col_left_const = checkAndGetColumnConst<ColVecT0>(col_left_raw);
const ColumnConst * const col_right_const = checkAndGetColumnConst<ColVecT1>(col_right_raw);
const ColVecT0 * const col_left = checkAndGetColumn<ColVecT0>(col_left_raw);
const ColVecT1 * const col_right = checkAndGetColumn<ColVecT1>(col_right_raw);
if constexpr (IsDataTypeDecimal<ResultDataType>)
{
return executeNumericWithDecimal<LeftDataType, RightDataType, ResultDataType>(
left, right,
col_left_const, col_right_const,
col_left, col_right,
col_left_size,
right_nullmap);
}
else // can't avoid else and another indentation level, otherwise the compiler would try to instantiate
// ColVecResult for Decimals which would lead to a compile error.
{
using OpImpl = BinaryOperationImpl<T0, T1, Op<T0, T1>, ResultType>;
/// non-vector result
if (col_left_const && col_right_const)
{
const auto res = right_nullmap && (*right_nullmap)[0] ? ResultType() : OpImpl::process(
col_left_const->template getValue<T0>(),
col_right_const->template getValue<T1>());
return ResultDataType().createColumnConst(col_left_const->size(), toField(res));
}
typename ColVecResult::MutablePtr col_res = ColVecResult::create();
auto & vec_res = col_res->getData();
vec_res.resize(col_left_size);
if (col_left && col_right)
{
OpImpl::template process<OpCase::Vector>(
col_left->getData().data(),
col_right->getData().data(),
vec_res.data(),
vec_res.size(),
right_nullmap);
}
else if (col_left_const && col_right)
{
const T0 value = col_left_const->template getValue<T0>();
OpImpl::template process<OpCase::LeftConstant>(
&value,
col_right->getData().data(),
vec_res.data(),
vec_res.size(),
right_nullmap);
}
else if (col_left && col_right_const)
{
const T1 value = col_right_const->template getValue<T1>();
OpImpl::template process<OpCase::RightConstant>(
col_left->getData().data(), &value, vec_res.data(), vec_res.size(), right_nullmap);
}
else
return nullptr;
return col_res;
}
}
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) const override
{
/// Special case when multiply aggregate function state
if (isAggregateMultiply(arguments[0].type, arguments[1].type))
{
return executeAggregateMultiply(arguments, result_type, input_rows_count);
}
/// Special case - addition of two aggregate functions states
if (isAggregateAddition(arguments[0].type, arguments[1].type))
{
return executeAggregateAddition(arguments, result_type, input_rows_count);
}
/// Special case when the function is plus or minus, one of arguments is Date/DateTime and another is Interval.
if (auto function_builder = getFunctionForIntervalArithmetic(arguments[0].type, arguments[1].type, context))
{
return executeDateTimeIntervalPlusMinus(arguments, result_type, input_rows_count, function_builder);
}
/// Special case when the function is plus, minus or multiply, both arguments are tuples.
if (auto function_builder = getFunctionForTupleArithmetic(arguments[0].type, arguments[1].type, context))
{
return function_builder->build(arguments)->execute(arguments, result_type, input_rows_count);
}
/// Special case when the function is multiply or divide, one of arguments is Tuple and another is Number.
if (auto function_builder = getFunctionForTupleAndNumberArithmetic(arguments[0].type, arguments[1].type, context))
{
return executeTupleNumberOperator(arguments, result_type, input_rows_count, function_builder);
}
return executeImpl2(arguments, result_type, input_rows_count);
}
ColumnPtr executeImpl2(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count, const NullMap * right_nullmap = nullptr) const
{
const auto & left_argument = arguments[0];
const auto & right_argument = arguments[1];
/// Process special case when operation is divide, intDiv or modulo and denominator
/// is Nullable(Something) to prevent division by zero error.
if (division_by_nullable && !right_nullmap)
{
assert(right_argument.type->isNullable());
bool is_const = checkColumnConst<ColumnNullable>(right_argument.column.get());
const ColumnNullable * nullable_column = is_const ? checkAndGetColumnConstData<ColumnNullable>(right_argument.column.get())
: checkAndGetColumn<ColumnNullable>(*right_argument.column);
const auto & null_bytemap = nullable_column->getNullMapData();
auto res = executeImpl2(createBlockWithNestedColumns(arguments), removeNullable(result_type), input_rows_count, &null_bytemap);
return wrapInNullable(res, arguments, result_type, input_rows_count);
}
const auto * const left_generic = left_argument.type.get();
const auto * const right_generic = right_argument.type.get();
ColumnPtr res;
const bool valid = castBothTypes(left_generic, right_generic, [&](const auto & left, const auto & right)
{
using LeftDataType = std::decay_t<decltype(left)>;
using RightDataType = std::decay_t<decltype(right)>;
if constexpr ((std::is_same_v<DataTypeFixedString, LeftDataType> || std::is_same_v<DataTypeString, LeftDataType>) ||
(std::is_same_v<DataTypeFixedString, RightDataType> || std::is_same_v<DataTypeString, RightDataType>))
{
if constexpr (std::is_same_v<DataTypeFixedString, LeftDataType> &&
std::is_same_v<DataTypeFixedString, RightDataType>)
{
if constexpr (!Op<DataTypeFixedString, DataTypeFixedString>::allow_fixed_string)
return false;
else
return (res = executeFixedString(arguments)) != nullptr;
}
if constexpr (!Op<LeftDataType, RightDataType>::allow_string_integer)
return false;
else if constexpr (!IsIntegral<RightDataType>)
return false;
else if constexpr (std::is_same_v<DataTypeFixedString, LeftDataType>)
{
return (res = executeStringInteger<ColumnFixedString>(arguments, left, right)) != nullptr;
}
else if constexpr (std::is_same_v<DataTypeString, LeftDataType>)
return (res = executeStringInteger<ColumnString>(arguments, left, right)) != nullptr;
}
else
return (res = executeNumeric(arguments, left, right, right_nullmap)) != nullptr;
});
if (!valid)
{
// This is a logical error, because the types should have been checked
// by getReturnTypeImpl().
throw Exception(ErrorCodes::LOGICAL_ERROR,
"Arguments of '{}' have incorrect data types: '{}' of type '{}',"
" '{}' of type '{}'", getName(),
left_argument.name, left_argument.type->getName(),
right_argument.name, right_argument.type->getName());
}
return res;
}
#if USE_EMBEDDED_COMPILER
bool isCompilableImpl(const DataTypes & arguments) const override
{
if (2 != arguments.size())
return false;
return castBothTypes(arguments[0].get(), arguments[1].get(), [&](const auto & left, const auto & right)
{
using LeftDataType = std::decay_t<decltype(left)>;
using RightDataType = std::decay_t<decltype(right)>;
if constexpr (std::is_same_v<DataTypeFixedString, LeftDataType> || std::is_same_v<DataTypeFixedString, RightDataType> || std::is_same_v<DataTypeString, LeftDataType> || std::is_same_v<DataTypeString, RightDataType>)
return false;
else
{
using ResultDataType = typename BinaryOperationTraits<Op, LeftDataType, RightDataType>::ResultDataType;
using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
return !std::is_same_v<ResultDataType, InvalidType> && !IsDataTypeDecimal<ResultDataType> && OpSpec::compilable;
}
});
}
llvm::Value * compileImpl(llvm::IRBuilderBase & builder, const DataTypes & types, Values values) const override
{
assert(2 == types.size() && 2 == values.size());
llvm::Value * result = nullptr;
castBothTypes(types[0].get(), types[1].get(), [&](const auto & left, const auto & right)
{
using LeftDataType = std::decay_t<decltype(left)>;
using RightDataType = std::decay_t<decltype(right)>;
if constexpr (!std::is_same_v<DataTypeFixedString, LeftDataType> && !std::is_same_v<DataTypeFixedString, RightDataType> && !std::is_same_v<DataTypeString, LeftDataType> && !std::is_same_v<DataTypeString, RightDataType>)
{
using ResultDataType = typename BinaryOperationTraits<Op, LeftDataType, RightDataType>::ResultDataType;
using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
if constexpr (!std::is_same_v<ResultDataType, InvalidType> && !IsDataTypeDecimal<ResultDataType> && OpSpec::compilable)
{
auto & b = static_cast<llvm::IRBuilder<> &>(builder);
auto type = std::make_shared<ResultDataType>();
auto * lval = nativeCast(b, types[0], values[0], type);
auto * rval = nativeCast(b, types[1], values[1], type);
result = OpSpec::compile(b, lval, rval, std::is_signed_v<typename ResultDataType::FieldType>);
return true;
}
}
return false;
});
return result;
}
#endif
bool canBeExecutedOnDefaultArguments() const override { return valid_on_default_arguments; }
};
template <template <typename, typename> class Op, typename Name, bool valid_on_default_arguments = true, bool valid_on_float_arguments = true, bool division_by_nullable = false>
class FunctionBinaryArithmeticWithConstants : public FunctionBinaryArithmetic<Op, Name, valid_on_default_arguments, valid_on_float_arguments, division_by_nullable>
{
public:
using Base = FunctionBinaryArithmetic<Op, Name, valid_on_default_arguments, valid_on_float_arguments, division_by_nullable>;
using Monotonicity = typename Base::Monotonicity;
static FunctionPtr create(
const ColumnWithTypeAndName & left_,
const ColumnWithTypeAndName & right_,
const DataTypePtr & return_type_,
ContextPtr context)
{
return std::make_shared<FunctionBinaryArithmeticWithConstants>(left_, right_, return_type_, context);
}
FunctionBinaryArithmeticWithConstants(
const ColumnWithTypeAndName & left_,
const ColumnWithTypeAndName & right_,
const DataTypePtr & return_type_,
ContextPtr context_)
: Base(context_), left(left_), right(right_), return_type(return_type_)
{
}
ColumnPtr executeImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type, size_t input_rows_count) const override
{
if (left.column && isColumnConst(*left.column) && arguments.size() == 1)
{
ColumnsWithTypeAndName columns_with_constant
= {{left.column->cloneResized(input_rows_count), left.type, left.name},
arguments[0]};
return Base::executeImpl(columns_with_constant, result_type, input_rows_count);
}
else if (right.column && isColumnConst(*right.column) && arguments.size() == 1)
{
ColumnsWithTypeAndName columns_with_constant
= {arguments[0],
{right.column->cloneResized(input_rows_count), right.type, right.name}};
return Base::executeImpl(columns_with_constant, result_type, input_rows_count);
}
else
return Base::executeImpl(arguments, result_type, input_rows_count);
}
bool hasInformationAboutMonotonicity() const override
{
const std::string_view name_view = Name::name;
return (name_view == "minus" || name_view == "plus" || name_view == "divide" || name_view == "intDiv");
}
Monotonicity getMonotonicityForRange(const IDataType &, const Field & left_point, const Field & right_point) const override
{
const std::string_view name_view = Name::name;
// For simplicity, we treat null values as monotonicity breakers, except for variable / non-zero constant.
if (left_point.isNull() || right_point.isNull())
{
if (name_view == "divide" || name_view == "intDiv")
{
// variable / constant
if (right.column && isColumnConst(*right.column))
{
auto constant = (*right.column)[0];
if (applyVisitor(FieldVisitorAccurateEquals(), constant, Field(0)))
return {false, true, false}; // variable / 0 is undefined, let's treat it as non-monotonic
bool is_constant_positive = applyVisitor(FieldVisitorAccurateLess(), Field(0), constant);
// division is saturated to `inf`, thus it doesn't have overflow issues.
return {true, is_constant_positive, true};
}
}
return {false, true, false};
}
// For simplicity, we treat every single value interval as positive monotonic.
if (applyVisitor(FieldVisitorAccurateEquals(), left_point, right_point))
return {true, true, false};
if (name_view == "minus" || name_view == "plus")
{
// const +|- variable
if (left.column && isColumnConst(*left.column))
{
auto transform = [&](const Field & point)
{
ColumnsWithTypeAndName columns_with_constant
= {{left.column->cloneResized(1), left.type, left.name},
{right.type->createColumnConst(1, point), right.type, right.name}};
auto col = Base::executeImpl(columns_with_constant, return_type, 1);
Field point_transformed;
col->get(0, point_transformed);
return point_transformed;
};
transform(left_point);
transform(right_point);
if (name_view == "plus")
{
// Check if there is an overflow
if (applyVisitor(FieldVisitorAccurateLess(), left_point, right_point)
== applyVisitor(FieldVisitorAccurateLess(), transform(left_point), transform(right_point)))
return {true, true, false};
else
return {false, true, false};
}
else
{
// Check if there is an overflow
if (applyVisitor(FieldVisitorAccurateLess(), left_point, right_point)
!= applyVisitor(FieldVisitorAccurateLess(), transform(left_point), transform(right_point)))
return {true, false, false};
else
return {false, false, false};
}
}
// variable +|- constant
else if (right.column && isColumnConst(*right.column))
{
auto transform = [&](const Field & point)
{
ColumnsWithTypeAndName columns_with_constant
= {{left.type->createColumnConst(1, point), left.type, left.name},
{right.column->cloneResized(1), right.type, right.name}};
auto col = Base::executeImpl(columns_with_constant, return_type, 1);
Field point_transformed;
col->get(0, point_transformed);
return point_transformed;
};
// Check if there is an overflow
if (applyVisitor(FieldVisitorAccurateLess(), left_point, right_point)
== applyVisitor(FieldVisitorAccurateLess(), transform(left_point), transform(right_point)))
return {true, true, false};
else
return {false, true, false};
}
}
if (name_view == "divide" || name_view == "intDiv")
{
// const / variable
if (left.column && isColumnConst(*left.column))
{
auto constant = (*left.column)[0];
if (applyVisitor(FieldVisitorAccurateEquals(), constant, Field(0)))
return {true, true, false}; // 0 / 0 is undefined, thus it's not always monotonic
bool is_constant_positive = applyVisitor(FieldVisitorAccurateLess(), Field(0), constant);
if (applyVisitor(FieldVisitorAccurateLess(), left_point, Field(0))
&& applyVisitor(FieldVisitorAccurateLess(), right_point, Field(0)))
{
return {true, is_constant_positive, false};
}
else if (
applyVisitor(FieldVisitorAccurateLess(), Field(0), left_point)
&& applyVisitor(FieldVisitorAccurateLess(), Field(0), right_point))
{
return {true, !is_constant_positive, false};
}
}
// variable / constant
else if (right.column && isColumnConst(*right.column))
{
auto constant = (*right.column)[0];
if (applyVisitor(FieldVisitorAccurateEquals(), constant, Field(0)))
return {false, true, false}; // variable / 0 is undefined, let's treat it as non-monotonic
bool is_constant_positive = applyVisitor(FieldVisitorAccurateLess(), Field(0), constant);
// division is saturated to `inf`, thus it doesn't have overflow issues.
return {true, is_constant_positive, true};
}
}
return {false, true, false};
}
private:
ColumnWithTypeAndName left;
ColumnWithTypeAndName right;
DataTypePtr return_type;
};
template <template <typename, typename> class Op, typename Name, bool valid_on_default_arguments = true, bool valid_on_float_arguments = true>
class BinaryArithmeticOverloadResolver : public IFunctionOverloadResolver
{
public:
static constexpr auto name = Name::name;
static FunctionOverloadResolverPtr create(ContextPtr context)
{
return std::make_unique<BinaryArithmeticOverloadResolver>(context);
}
explicit BinaryArithmeticOverloadResolver(ContextPtr context_) : context(context_) {}
String getName() const override { return name; }
size_t getNumberOfArguments() const override { return 2; }
bool isVariadic() const override { return false; }
FunctionBasePtr buildImpl(const ColumnsWithTypeAndName & arguments, const DataTypePtr & return_type) const override
{
/// Check the case when operation is divide, intDiv or modulo and denominator is Nullable(Something).
/// For divide operation we should check only Nullable(Decimal), because only this case can throw division by zero error.
bool division_by_nullable = !arguments[0].type->onlyNull() && !arguments[1].type->onlyNull() && arguments[1].type->isNullable()
&& (IsOperation<Op>::div_int || IsOperation<Op>::modulo
|| (IsOperation<Op>::div_floating
&& (isDecimalOrNullableDecimal(arguments[0].type) || isDecimalOrNullableDecimal(arguments[1].type))));
/// More efficient specialization for two numeric arguments.
if (arguments.size() == 2
&& ((arguments[0].column && isColumnConst(*arguments[0].column))
|| (arguments[1].column && isColumnConst(*arguments[1].column))))
{
auto function = division_by_nullable ? FunctionBinaryArithmeticWithConstants<Op, Name, valid_on_default_arguments, valid_on_float_arguments, true>::create(
arguments[0], arguments[1], return_type, context)
: FunctionBinaryArithmeticWithConstants<Op, Name, valid_on_default_arguments, valid_on_float_arguments, false>::create(
arguments[0], arguments[1], return_type, context);
return std::make_unique<FunctionToFunctionBaseAdaptor>(
function,
collections::map<DataTypes>(arguments, [](const auto & elem) { return elem.type; }),
return_type);
}
auto function = division_by_nullable
? FunctionBinaryArithmetic<Op, Name, valid_on_default_arguments, valid_on_float_arguments, true>::create(context)
: FunctionBinaryArithmetic<Op, Name, valid_on_default_arguments, valid_on_float_arguments, false>::create(context);
return std::make_unique<FunctionToFunctionBaseAdaptor>(
function,
collections::map<DataTypes>(arguments, [](const auto & elem) { return elem.type; }),
return_type);
}
DataTypePtr getReturnTypeImpl(const DataTypes & arguments) const override
{
if (arguments.size() != 2)
throw Exception(
"Number of arguments for function " + getName() + " doesn't match: passed " + toString(arguments.size()) + ", should be 2",
ErrorCodes::NUMBER_OF_ARGUMENTS_DOESNT_MATCH);
return FunctionBinaryArithmetic<Op, Name, valid_on_default_arguments, valid_on_float_arguments>::getReturnTypeImplStatic(arguments, context);
}
private:
ContextPtr context;
};
}
Reference in New Issue View Git Blame Copy Permalink