Merge 7a74f0223d into e0f8b8d351

src/Core/AccurateComparison.h

@@ -209,12 +209,14 @@ template <typename A, typename B> struct EqualsOp
     using SymmetricOp = EqualsOp<B, A>;
 
     static UInt8 apply(A a, B b) { return accurate::equalsOp(a, b); }
+    static constexpr bool compilable = true;
 };
 
 template <typename A, typename B> struct NotEqualsOp
 {
     using SymmetricOp = NotEqualsOp<B, A>;
     static UInt8 apply(A a, B b) { return accurate::notEqualsOp(a, b); }
+    static constexpr bool compilable = true;
 };
 
 template <typename A, typename B> struct GreaterOp;
@@ -223,12 +225,14 @@ template <typename A, typename B> struct LessOp
 {
     using SymmetricOp = GreaterOp<B, A>;
     static UInt8 apply(A a, B b) { return accurate::lessOp(a, b); }
+    static constexpr bool compilable = true;
 };
 
 template <typename A, typename B> struct GreaterOp
 {
     using SymmetricOp = LessOp<B, A>;
     static UInt8 apply(A a, B b) { return accurate::greaterOp(a, b); }
+    static constexpr bool compilable = true;
 };
 
 template <typename A, typename B> struct GreaterOrEqualsOp;
@@ -237,12 +241,14 @@ template <typename A, typename B> struct LessOrEqualsOp
 {
     using SymmetricOp = GreaterOrEqualsOp<B, A>;
     static UInt8 apply(A a, B b) { return accurate::lessOrEqualsOp(a, b); }
+    static constexpr bool compilable = true;
 };
 
 template <typename A, typename B> struct GreaterOrEqualsOp
 {
     using SymmetricOp = LessOrEqualsOp<B, A>;
     static UInt8 apply(A a, B b) { return accurate::greaterOrEqualsOp(a, b); }
+    static constexpr bool compilable = true;
 };
 
 }

src/DataTypes/IDataType.h

@@ -555,6 +555,7 @@ inline bool isNullableOrLowCardinalityNullable(const DataTypePtr & data_type)
 
 template <typename DataType> constexpr bool IsDataTypeDecimal = false;
 template <typename DataType> constexpr bool IsDataTypeNumber = false;
+template <typename DataType> constexpr bool IsDataTypeNativeNumber = false;
 template <typename DataType> constexpr bool IsDataTypeDateOrDateTime = false;
 template <typename DataType> constexpr bool IsDataTypeDate = false;
 template <typename DataType> constexpr bool IsDataTypeEnum = false;
@@ -581,6 +582,9 @@ template <is_decimal T> constexpr bool IsDataTypeDecimal<DataTypeDecimal<T>> = t
 template <> inline constexpr bool IsDataTypeDecimal<DataTypeDateTime64> = true;
 
 template <typename T> constexpr bool IsDataTypeNumber<DataTypeNumber<T>> = true;
+template <typename T>
+requires std::is_arithmetic_v<T>
+constexpr bool IsDataTypeNativeNumber<DataTypeNumber<T>> = true;
 
 template <> inline constexpr bool IsDataTypeDate<DataTypeDate> = true;
 template <> inline constexpr bool IsDataTypeDate<DataTypeDate32> = true;

src/DataTypes/NumberTraits.h

@@ -205,6 +205,39 @@ struct ResultOfIf
             ConstructedType, Error>>>;
 };
 
+/** Type casting for `modulo` function:
+    * UInt<x>,  UInt<y>   ->  UInt<max(x,y)>
+    * Int<x>,   Int<y>    ->  Int<max(x,y)>
+    * UInt<x>,  Int<y>    ->  Int<max(x*2, y)>
+    * UInt64,   Int<x>    ->  Error
+    * Float<x>, Float<y>  ->  Float64
+    * Float<x>, [U]Int<y> ->  Float64
+    */
+template <typename A, typename B>
+struct ResultOfModuloNativePromotion
+{
+    static_assert(is_arithmetic_v<A> && is_arithmetic_v<B>);
+
+    static constexpr bool has_float = std::is_floating_point_v<A> || std::is_floating_point_v<B>;
+    static constexpr bool has_integer = is_integer<A> || is_integer<B>;
+    static constexpr bool has_signed = is_signed_v<A> || is_signed_v<B>;
+    static constexpr bool has_unsigned = !is_signed_v<A> || !is_signed_v<B>;
+
+    static constexpr size_t max_size_of_unsigned_integer = max(is_signed_v<A> ? 0 : sizeof(A), is_signed_v<B> ? 0 : sizeof(B));
+    static constexpr size_t max_size_of_signed_integer = max(is_signed_v<A> ? sizeof(A) : 0, is_signed_v<B> ? sizeof(B) : 0);
+    static constexpr size_t max_size_of_integer = max(is_integer<A> ? sizeof(A) : 0, is_integer<B> ? sizeof(B) : 0);
+
+    using ConstructedType = typename Construct<
+        has_signed,
+        false,
+        (has_signed ^ has_unsigned) ? max(max_size_of_unsigned_integer * 2, max_size_of_signed_integer) : max(sizeof(A), sizeof(B))>::Type;
+
+    using Type = std::conditional_t<
+        std::is_same_v<A, B>,
+        A,
+        std::conditional_t<has_float, Float64, std::conditional_t<sizeof(ConstructedType) <= 8, ConstructedType, Error>>>;
+};
+
 /** Before applying operator `%` and bitwise operations, operands are casted to whole numbers. */
 template <typename A> struct ToInteger
 {

src/Functions/DivisionUtils.h

@@ -6,8 +6,14 @@
 #include <Common/NaNUtils.h>
 #include <DataTypes/NumberTraits.h>
 
+#include "DataTypes/Native.h"
 #include "config.h"
 
+#if USE_EMBEDDED_COMPILER
+#    include <Core/ValuesWithType.h>
+#    include <llvm/IR/IRBuilder.h>
+#endif
+
 
 namespace DB
 {
@@ -15,8 +21,42 @@ namespace DB
 namespace ErrorCodes
 {
     extern const int ILLEGAL_DIVISION;
+    extern const int LOGICAL_ERROR;
 }
 
+#if USE_EMBEDDED_COMPILER
+
+template <typename F>
+static llvm::Value * compileWithNullableValues(llvm::IRBuilder<> & b, llvm::Value * left, llvm::Value * right, bool is_signed, F && compile_func)
+{
+    auto * left_type = left->getType();
+    auto * right_type = right->getType();
+
+    if (!left_type->isStructTy() && !right_type->isStructTy())
+    {
+        // Both arguments are not nullable.
+        return compile_func(b, left, right, is_signed);
+    }
+
+    auto * denull_left = left_type->isStructTy() ? b.CreateExtractValue(left, {1}) : left;
+    auto * denull_right = right_type->isStructTy() ? b.CreateExtractValue(right, {1}) : right;
+    auto * denull_result = compile_func(b, denull_left, denull_right, is_signed);
+
+    auto * nullable_result_type = toNullableType(b, denull_result->getType());
+    llvm::Value * nullable_result = llvm::Constant::getNullValue(nullable_result_type);
+    nullable_result = b.CreateInsertValue(nullable_result, denull_result, {0});
+
+    auto * result_is_null = b.CreateExtractValue(nullable_result, {1});
+    if (left_type->isStructTy())
+        result_is_null = b.CreateOr(result_is_null, b.CreateExtractValue(left, {1}));
+    if (right_type->isStructTy())
+        result_is_null = b.CreateOr(result_is_null, b.CreateExtractValue(right, {1}));
+
+    return b.CreateInsertValue(nullable_result, result_is_null, {1});
+}
+
+#endif
+
 template <typename A, typename B>
 inline void throwIfDivisionLeadsToFPE(A a, B b)
 {
@@ -158,14 +198,39 @@ struct ModuloImpl
     }
 
 #if USE_EMBEDDED_COMPILER
-    static constexpr bool compilable = false; /// don't know how to throw from LLVM IR
+    static constexpr bool compilable = true; /// Ignore exceptions in LLVM IR
-#endif
+
+    static llvm::Value * compile(llvm::IRBuilder<> & b, llvm::Value * left, llvm::Value * right, bool is_signed)
+    {
+        return compileWithNullableValues(
+            b,
+            left,
+            right,
+            is_signed,
+            [](auto & b_, auto * left_, auto * right_, auto is_signed_) { return compileImpl(b_, left_, right_, is_signed_); });
+    }
+
+    static llvm::Value * compileImpl(llvm::IRBuilder<> & b, llvm::Value * left, llvm::Value * right, bool is_signed)
+    {
+        if (left->getType()->isFloatingPointTy())
+            return b.CreateFRem(left, right);
+        else if (left->getType()->isIntegerTy())
+            return is_signed ? b.CreateSRem(left, right) : b.CreateURem(left, right);
+        else
+            throw Exception(ErrorCodes::LOGICAL_ERROR, "ModuloImpl compilation expected native integer or floating point type");
+    }
+
+    #endif
 };
 
 template <typename A, typename B>
 struct ModuloLegacyImpl : ModuloImpl<A, B>
 {
     using ResultType = typename NumberTraits::ResultOfModuloLegacy<A, B>::Type;
+
+#if USE_EMBEDDED_COMPILER
+    static constexpr bool compilable = false; /// moduloLegacy is only used in partition key expression
+#endif
 };
 
 template <typename A, typename B>
@@ -194,6 +259,36 @@ struct PositiveModuloImpl : ModuloImpl<A, B>
         }
         return static_cast<ResultType>(res);
     }
+
+#if USE_EMBEDDED_COMPILER
+    static constexpr bool compilable = true; /// Ignore exceptions in LLVM IR
+
+    static llvm::Value * compile(llvm::IRBuilder<> & b, llvm::Value * left, llvm::Value * right, bool is_signed)
+    {
+        return compileWithNullableValues(
+            b,
+            left,
+            right,
+            is_signed,
+            [](auto & b_, auto * left_, auto * right_, auto is_signed_) { return compileImpl(b_, left_, right_, is_signed_); });
+    }
+
+    static llvm::Value * compileImpl(llvm::IRBuilder<> & b, llvm::Value * left, llvm::Value * right, bool is_signed)
+    {
+        auto * result = ModuloImpl<A, B>::compileImpl(b, left, right, is_signed);
+        if (is_signed)
+        {
+            /// If result is negative, result += abs(right).
+            auto * zero = llvm::Constant::getNullValue(result->getType());
+            auto * is_negative = b.CreateICmpSLT(result, zero);
+            auto * abs_right = b.CreateSelect(b.CreateICmpSLT(right, zero), b.CreateNeg(right), right);
+            return b.CreateSelect(is_negative, b.CreateAdd(result, abs_right), result);
+        }
+        else
+            return result;
+    }
+#endif
+
 };
 
 }

src/Functions/FunctionBinaryArithmetic.h

@@ -810,6 +810,7 @@ class FunctionBinaryArithmetic : public IFunction
     static constexpr bool is_division = IsOperation<Op>::division;
     static constexpr bool is_bit_hamming_distance = IsOperation<Op>::bit_hamming_distance;
     static constexpr bool is_modulo = IsOperation<Op>::modulo;
+    static constexpr bool is_positive_modulo = IsOperation<Op>::positive_modulo;
     static constexpr bool is_int_div = IsOperation<Op>::int_div;
     static constexpr bool is_int_div_or_zero = IsOperation<Op>::int_div_or_zero;
 
@@ -2387,59 +2388,105 @@ ColumnPtr executeStringInteger(const ColumnsWithTypeAndName & arguments, const A
         if (!canBeNativeType(*arguments[0]) || !canBeNativeType(*arguments[1]) || !canBeNativeType(*result_type))
             return false;
 
-        WhichDataType data_type_lhs(arguments[0]);
+        auto denull_left_type = removeNullable(arguments[0]);
-        WhichDataType data_type_rhs(arguments[1]);
+        auto denull_right_type = removeNullable(arguments[1]);
+        WhichDataType data_type_lhs(denull_left_type);
+        WhichDataType data_type_rhs(denull_right_type);
         if ((data_type_lhs.isDateOrDate32() || data_type_lhs.isDateTime()) ||
             (data_type_rhs.isDateOrDate32() || data_type_rhs.isDateTime()))
             return false;
 
-        return castBothTypes(arguments[0].get(), arguments[1].get(), [&](const auto & left, const auto & right)
+        return castBothTypes(
-        {
+            denull_left_type.get(),
-            using LeftDataType = std::decay_t<decltype(left)>;
+            denull_right_type.get(),
-            using RightDataType = std::decay_t<decltype(right)>;
+            [&](const auto & left, const auto & 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 LeftDataType = std::decay_t<decltype(left)>;
-                using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
+                using RightDataType = std::decay_t<decltype(right)>;
-                if constexpr (!std::is_same_v<ResultDataType, InvalidType> && !IsDataTypeDecimal<ResultDataType> && OpSpec::compilable)
+                if constexpr (
-                    return true;
+                    !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;
+                {
-        });
+                    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>
+                        && !IsDataTypeDecimal<LeftDataType> && !IsDataTypeDecimal<RightDataType> && OpSpec::compilable)
+                    {
+                        if constexpr (is_modulo || is_positive_modulo)
+                        {
+                            using LeftType = typename LeftDataType::FieldType;
+                            using RightType = typename RightDataType::FieldType;
+                            using PromotedType = typename NumberTraits::ResultOfModuloNativePromotion<LeftType, RightType>::Type;
+                            if constexpr (std::is_arithmetic_v<PromotedType>)
+                            {
+                                return true;
+                            }
+                        }
+                        else
+                            return true;
+                    }
+                }
+                return false;
+            });
     }
 
     llvm::Value * compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & arguments, const DataTypePtr & result_type) const override
     {
         assert(2 == arguments.size());
 
+        auto denull_left_type = removeNullable(arguments[0].type);
+        auto denull_right_type = removeNullable(arguments[1].type);
         llvm::Value * result = nullptr;
-        castBothTypes(arguments[0].type.get(), arguments[1].type.get(), [&](const auto & left, const auto & right)
+
-        {
+        castBothTypes(
-            using LeftDataType = std::decay_t<decltype(left)>;
+            denull_left_type.get(),
-            using RightDataType = std::decay_t<decltype(right)>;
+            denull_right_type.get(),
-            if constexpr (!std::is_same_v<DataTypeFixedString, LeftDataType> &&
+            [&](const auto & left, const auto & right)
-                !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 LeftDataType = std::decay_t<decltype(left)>;
-                using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
+                using RightDataType = std::decay_t<decltype(right)>;
-                if constexpr (!std::is_same_v<ResultDataType, InvalidType> && !IsDataTypeDecimal<ResultDataType> && OpSpec::compilable)
+                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>)
                 {
-                    auto & b = static_cast<llvm::IRBuilder<> &>(builder);
+                    using ResultDataType = typename BinaryOperationTraits<Op, LeftDataType, RightDataType>::ResultDataType;
-                    auto * lval = nativeCast(b, arguments[0], result_type);
+                    using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
-                    auto * rval = nativeCast(b, arguments[1], result_type);
+                    if constexpr (
-                    result = OpSpec::compile(b, lval, rval, std::is_signed_v<typename ResultDataType::FieldType>);
+                        !std::is_same_v<ResultDataType, InvalidType> && !IsDataTypeDecimal<ResultDataType>
+                        && !IsDataTypeDecimal<LeftDataType> && !IsDataTypeDecimal<RightDataType> && OpSpec::compilable)
+                    {
+                        auto & b = static_cast<llvm::IRBuilder<> &>(builder);
+                        if constexpr (is_modulo || is_positive_modulo)
+                        {
+                            using LeftType = typename LeftDataType::FieldType;
+                            using RightType = typename RightDataType::FieldType;
+                            using PromotedType = typename NumberTraits::ResultOfModuloNativePromotion<LeftType, RightType>::Type;
+                            if constexpr (std::is_arithmetic_v<PromotedType>)
+                            {
+                                DataTypePtr promoted_type = std::make_shared<DataTypeNumber<PromotedType>>();
+                                if (result_type->isNullable())
+                                    promoted_type = std::make_shared<DataTypeNullable>(promoted_type);
 
-                    return true;
+                                auto * lval = nativeCast(b, arguments[0], promoted_type);
+                                auto * rval = nativeCast(b, arguments[1], promoted_type);
+                                result = nativeCast(
+                                    b, promoted_type, OpSpec::compile(b, lval, rval, std::is_signed_v<PromotedType>), result_type);
+                                return true;
+                            }
+                        }
+                        else
+                        {
+                            auto * lval = nativeCast(b, arguments[0], result_type);
+                            auto * rval = nativeCast(b, arguments[1], result_type);
+                            result = OpSpec::compile(b, lval, rval, std::is_signed_v<typename ResultDataType::FieldType>);
+                            return true;
+                        }
+                    }
                 }
-            }
+                return false;
-
+            });
-            return false;
-        });
 
         return result;
     }

src/Functions/FunctionUnaryArithmetic.h

@@ -489,9 +489,7 @@ public:
         {
             using DataType = std::decay_t<decltype(type)>;
             if constexpr (std::is_same_v<DataTypeFixedString, DataType> || std::is_same_v<DataTypeString, DataType>)
-            {
                 return false;
-            }
             else
             {
                 using T0 = typename DataType::FieldType;
@@ -513,9 +511,7 @@ public:
         {
             using DataType = std::decay_t<decltype(type)>;
             if constexpr (std::is_same_v<DataTypeFixedString, DataType> || std::is_same_v<DataTypeString, DataType>)
-            {
                 return false;
-            }
             else
             {
                 using T0 = typename DataType::FieldType;
@@ -523,8 +519,16 @@ public:
                 if constexpr (!std::is_same_v<T1, InvalidType> && !IsDataTypeDecimal<DataType> && Op<T0>::compilable)
                 {
                     auto & b = static_cast<llvm::IRBuilder<> &>(builder);
-                    auto * v = nativeCast(b, arguments[0], result_type);
+                    if constexpr (std::is_same_v<Op<T0>, AbsImpl<T0>> || std::is_same_v<Op<T0>, BitCountImpl<T0>>)
-                    result = Op<T0>::compile(b, v, is_signed_v<T1>);
+                    {
+                        /// We don't need to cast the argument to the result type if it's abs/bitcount function.
+                        result = Op<T0>::compile(b, arguments[0].value, is_signed_v<T0>);
+                    }
+                    else
+                    {
+                        auto * v = nativeCast(b, arguments[0], result_type);
+                        result = Op<T0>::compile(b, v, is_signed_v<T1>);
+                    }
 
                     return true;
                 }

src/Functions/FunctionsComparison.h

@@ -1,17 +1,21 @@
 #pragma once
 
-#include <Common/memcmpSmall.h>
+// Include this first, because `#define _asan_poison_address` from
-#include <Common/assert_cast.h>
+// llvm/Support/Compiler.h conflicts with its forward declaration in
-#include <Common/TargetSpecific.h>
+// sanitizer/asan_interface.h
+#include <memory>
+#include <limits>
+#include <type_traits>
 
-#include <Columns/ColumnsNumber.h>
+#include <Columns/ColumnArray.h>
 #include <Columns/ColumnConst.h>
 #include <Columns/ColumnDecimal.h>
-#include <Columns/ColumnString.h>
 #include <Columns/ColumnFixedString.h>
+#include <Columns/ColumnString.h>
 #include <Columns/ColumnTuple.h>
-#include <Columns/ColumnArray.h>
+#include <Columns/ColumnsNumber.h>
-
+#include <Core/AccurateComparison.h>
+#include <Core/DecimalComparison.h>
 #include <DataTypes/DataTypeDate.h>
 #include <DataTypes/DataTypeDateTime.h>
 #include <DataTypes/DataTypeDateTime64.h>
@@ -24,22 +28,23 @@
 #include <DataTypes/DataTypeUUID.h>
 #include <DataTypes/DataTypesNumber.h>
 #include <DataTypes/getLeastSupertype.h>
-
-#include <Interpreters/convertFieldToType.h>
-#include <Interpreters/castColumn.h>
-
-#include <Functions/IFunctionAdaptors.h>
 #include <Functions/FunctionHelpers.h>
+#include <Functions/IFunctionAdaptors.h>
 #include <Functions/IsOperation.h>
-
-#include <Core/AccurateComparison.h>
-#include <Core/DecimalComparison.h>
-
 #include <IO/ReadBufferFromMemory.h>
 #include <IO/ReadHelpers.h>
+#include <Interpreters/castColumn.h>
+#include <Interpreters/convertFieldToType.h>
+#include <Common/TargetSpecific.h>
+#include <Common/assert_cast.h>
+#include <Common/memcmpSmall.h>
+#include "DataTypes/NumberTraits.h"
 
-#include <limits>
+#if USE_EMBEDDED_COMPILER
-#include <type_traits>
+#    include <DataTypes/Native.h>
+#    include <Functions/castTypeToEither.h>
+#    include <llvm/IR/IRBuilder.h>
+#endif
 
 namespace DB
 {
@@ -1356,6 +1361,109 @@ public:
 
         return executeGeneric(col_with_type_and_name_left, col_with_type_and_name_right);
     }
+
+#if USE_EMBEDDED_COMPILER
+    template <typename F>
+    static bool castType(const IDataType * type, F && f)
+    {
+        return castTypeToEither<
+            DataTypeUInt8,
+            DataTypeUInt16,
+            DataTypeUInt32,
+            DataTypeUInt64,
+            DataTypeInt8,
+            DataTypeInt16,
+            DataTypeInt32,
+            DataTypeInt64,
+            DataTypeFloat32,
+            DataTypeFloat64>(type, std::forward<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_);
+            });
+        });
+    }
+
+    // bool isCompilableImpl(const DataTypes &, const DataTypePtr &) const override { return false; }
+    bool isCompilableImpl(const DataTypes & arguments, const DataTypePtr & result_type) const override
+    {
+        if (2 != arguments.size())
+            return false;
+
+        if (!canBeNativeType(*arguments[0]) || !canBeNativeType(*arguments[1]) || !canBeNativeType(*result_type))
+            return false;
+
+        WhichDataType data_type_lhs(arguments[0]);
+        WhichDataType data_type_rhs(arguments[1]);
+        /// TODO support date/date32
+        if ((data_type_lhs.isDateOrDate32() || data_type_lhs.isDateTime()) ||
+            (data_type_rhs.isDateOrDate32() || data_type_rhs.isDateTime()))
+            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)>;
+            using LeftType = typename LeftDataType::FieldType;
+            using RightType = typename RightDataType::FieldType;
+            using PromotedType = typename NumberTraits::ResultOfIf<LeftType, RightType>::Type;
+            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>
+                && (std::is_integral_v<PromotedType> || std::is_floating_point_v<PromotedType>))
+            {
+                using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
+                return OpSpec::compilable;
+            }
+            return false;
+        });
+        return false;
+    }
+
+    llvm::Value * compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & arguments, const DataTypePtr &) const override
+    {
+        assert(2 == arguments.size());
+
+        llvm::Value * result = nullptr;
+        castBothTypes(arguments[0].type.get(), arguments[1].type.get(), [&](const auto & left, const auto & right)
+        {
+            using LeftDataType = std::decay_t<decltype(left)>;
+            using RightDataType = std::decay_t<decltype(right)>;
+            using LeftType = typename LeftDataType::FieldType;
+            using RightType = typename RightDataType::FieldType;
+            using PromotedType = typename NumberTraits::ResultOfIf<LeftType, RightType>::Type;
+
+            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>
+                && (std::is_integral_v<PromotedType> || std::is_floating_point_v<PromotedType>))
+            {
+                using OpSpec = Op<typename LeftDataType::FieldType, typename RightDataType::FieldType>;
+                if constexpr (OpSpec::compilable)
+                {
+                    auto promoted_type = std::make_shared<DataTypeNumber<PromotedType>>();
+                    auto & b = static_cast<llvm::IRBuilder<> &>(builder);
+                    auto * left_value = nativeCast(b, arguments[0], promoted_type);
+                    auto * right_value = nativeCast(b, arguments[1], promoted_type);
+                    result = b.CreateSelect(
+                        CompileOp<Op>::compile(b, left_value, right_value, std::is_signed_v<PromotedType>), b.getInt8(1), b.getInt8(0));
+                    return true;
+                }
+            }
+            return false;
+        });
+
+        return result;
+    }
+#endif
+
 };
 
 }

src/Functions/FunctionsConversion.cpp

@@ -71,6 +71,9 @@
 #include <Common/assert_cast.h>
 #include <Common/quoteString.h>
 
+#if USE_EMBEDDED_COMPILER
+#include <DataTypes/Native.h>
+#endif
 
 namespace DB
 {
@@ -115,6 +118,43 @@ namespace ErrorCodes
 namespace
 {
 
+#if USE_EMBEDDED_COMPILER
+bool castType(const IDataType * type, auto && f)
+{
+    using Types = TypeList<
+        DataTypeUInt8,
+        DataTypeUInt16,
+        DataTypeUInt32,
+        DataTypeUInt64,
+        DataTypeUInt128,
+        DataTypeUInt256,
+        DataTypeInt8,
+        DataTypeInt16,
+        DataTypeInt32,
+        DataTypeInt64,
+        DataTypeInt128,
+        DataTypeInt256,
+        DataTypeFloat32,
+        DataTypeFloat64,
+        DataTypeDecimal32,
+        DataTypeDecimal64,
+        DataTypeDecimal128,
+        DataTypeDecimal256,
+        DataTypeDate,
+        DataTypeDateTime,
+        DataTypeFixedString,
+        DataTypeString,
+        DataTypeInterval>;
+    return castTypeToEither(Types{}, type, std::forward<decltype(f)>(f));
+}
+
+template <typename F>
+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_); }); });
+}
+#endif
+
 /** Type conversion functions.
   * toType - conversion in "natural way";
   */
@@ -2238,6 +2278,52 @@ public:
         return Monotonic::get(type, left, right);
     }
 
+#if USE_EMBEDDED_COMPILER
+    bool isCompilableImpl(const DataTypes & types, const DataTypePtr & result_type) const override
+    {
+        if (types.size() != 1)
+            return false;
+
+        if (!canBeNativeType(types[0]) || !canBeNativeType(result_type))
+            return false;
+
+        return castBothTypes(types[0].get(), result_type.get(), [](const auto & left, const auto & right)
+        {
+            using LeftDataType = std::decay_t<decltype(left)>;
+            using RightDataType = std::decay_t<decltype(right)>;
+
+            if constexpr (IsDataTypeNativeNumber<LeftDataType> && IsDataTypeNativeNumber<RightDataType>)
+                return true;
+
+            return false;
+        });
+    }
+
+    llvm::Value *
+    compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & arguments, const DataTypePtr & result_type) const override
+    {
+        llvm::Value * result = nullptr;
+        castBothTypes(
+            arguments[0].type.get(),
+            result_type.get(),
+            [&](const auto & left, const auto & right)
+            {
+                using LeftDataType = std::decay_t<decltype(left)>;
+                using RightDataType = std::decay_t<decltype(right)>;
+
+                if constexpr (IsDataTypeNativeNumber<LeftDataType> && IsDataTypeNativeNumber<RightDataType>)
+                {
+                    result = nativeCast(builder, arguments[0], result_type);
+                    return true;
+                }
+
+                return false;
+            });
+
+        return result;
+    }
+#endif
+
 private:
     ContextPtr context;
     mutable bool checked_return_type = false;
@@ -3228,6 +3314,61 @@ public:
         return monotonicity_for_range(type, left, right);
     }
 
+#if USE_EMBEDDED_COMPILER
+    bool isCompilable() const override
+    {
+        if (getName() != "CAST" || argument_types.size() != 2)
+            return false;
+
+        const auto & from_type = argument_types[0];
+        const auto & to_type = return_type;
+        auto denull_from_type = removeNullable(from_type);
+        auto denull_to_type = removeNullable(to_type);
+        if (!canBeNativeType(denull_from_type) || !canBeNativeType(denull_to_type))
+            return false;
+
+        return castBothTypes(denull_from_type.get(), denull_to_type.get(), [](const auto & left, const auto & right)
+        {
+            using LeftDataType = std::decay_t<decltype(left)>;
+            using RightDataType = std::decay_t<decltype(right)>;
+
+            if constexpr (IsDataTypeNativeNumber<LeftDataType> && IsDataTypeNativeNumber<RightDataType>)
+                return true;
+
+            return false;
+        });
+    }
+
+    llvm::Value * compile(llvm::IRBuilderBase & builder, const ValuesWithType & arguments) const override
+    {
+        llvm::Value * result = nullptr;
+
+        const auto & from_type = arguments[0].type;
+        const auto & to_type = return_type;
+        auto denull_from_type = removeNullable(from_type);
+        auto denull_to_type = removeNullable(to_type);
+        castBothTypes(
+            denull_from_type.get(),
+            denull_to_type.get(),
+            [&](const auto & left, const auto & right)
+            {
+                using LeftDataType = std::decay_t<decltype(left)>;
+                using RightDataType = std::decay_t<decltype(right)>;
+
+                if constexpr (IsDataTypeNativeNumber<LeftDataType> && IsDataTypeNativeNumber<RightDataType>)
+                {
+                    result = nativeCast(builder, arguments[0], return_type);
+                    return true;
+                }
+
+                return false;
+            });
+
+        return result;
+    }
+
+#endif
+
 private:
     const char * cast_name;
     MonotonicityForRange monotonicity_for_range;

src/Functions/FunctionsLogical.cpp

@@ -39,6 +39,7 @@ namespace ErrorCodes
     extern const int ILLEGAL_TYPE_OF_ARGUMENT;
     extern const int TOO_FEW_ARGUMENTS_FOR_FUNCTION;
     extern const int ILLEGAL_COLUMN;
+    extern const int NOT_IMPLEMENTED;
 }
 
 namespace
@@ -511,6 +512,44 @@ ColumnPtr basicExecuteImpl(ColumnRawPtrs arguments, size_t input_rows_count)
 
 }
 
+namespace FunctionsLogicalDetail
+{
+
+#if USE_EMBEDDED_COMPILER
+
+/// Cast LLVM value with type to ternary
+llvm::Value * nativeTernaryCast(llvm::IRBuilderBase & b, const DataTypePtr & from_type, llvm::Value * value)
+{
+    auto * result_type = llvm::Type::getInt8Ty(b.getContext());
+
+    if (from_type->isNullable())
+    {
+        auto * ternary_null = llvm::ConstantInt::get(result_type, 1);
+        auto * inner = nativeTernaryCast(b, removeNullable(from_type), b.CreateExtractValue(value, {0}));
+        auto * is_null = b.CreateExtractValue(value, {1});
+        return b.CreateSelect(is_null, ternary_null, inner);
+    }
+
+    auto * zero = llvm::Constant::getNullValue(value->getType());
+    auto * ternary_true = llvm::ConstantInt::get(result_type, 2);
+    auto * ternary_false = llvm::ConstantInt::get(result_type, 0);
+    if (value->getType()->isIntegerTy())
+        return b.CreateSelect(b.CreateICmpNE(value, zero), ternary_true, ternary_false);
+    else if (value->getType()->isFloatingPointTy())
+        return b.CreateSelect(b.CreateFCmpONE(value, zero), ternary_true, ternary_false);
+    else
+        throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Cannot cast non-number {} to ternary", from_type->getName());
+}
+
+/// Cast LLVM value with type to ternary
+llvm::Value * nativeTernaryCast(llvm::IRBuilderBase & b, const ValueWithType & value_with_type)
+{
+    return nativeTernaryCast(b, value_with_type.type, value_with_type.value);
+}
+
+#endif
+}
+
 template <typename Impl, typename Name>
 DataTypePtr FunctionAnyArityLogical<Impl, Name>::getReturnTypeImpl(const DataTypes & arguments) const
 {

src/Functions/FunctionsLogical.h

@@ -6,7 +6,6 @@
 #include <DataTypes/DataTypesNumber.h>
 #include <Functions/IFunction.h>
 #include <IO/WriteHelpers.h>
-#include <type_traits>
 #include <Interpreters/Context_fwd.h>
 
 
@@ -79,6 +78,15 @@ namespace Ternary
     }
 }
 
+#if USE_EMBEDDED_COMPILER
+
+/// Cast LLVM value with type to Ternary
+llvm::Value * nativeTernaryCast(llvm::IRBuilderBase & b, const DataTypePtr & from_type, llvm::Value * value);
+
+/// Cast LLVM value with type to Ternary
+llvm::Value * nativeTernaryCast(llvm::IRBuilderBase & b, const ValueWithType & value_with_type);
+
+#endif
 
 struct AndImpl
 {
@@ -98,6 +106,18 @@ struct AndImpl
 
     /// Will use three-valued logic for NULLs (see above) or default implementation (any operation with NULL returns NULL).
     static constexpr bool specialImplementationForNulls() { return true; }
+
+#if USE_EMBEDDED_COMPILER
+    static llvm::Value * apply(llvm::IRBuilder<> & builder, llvm::Value * a, llvm::Value * b)
+    {
+        return builder.CreateAnd(a, b);
+    }
+
+    static llvm::Value * ternaryApply(llvm::IRBuilder<> & builder, llvm::Value * a, llvm::Value * b)
+    {
+        return builder.CreateSelect(builder.CreateICmpUGT(a, b), b, a);
+    }
+#endif
 };
 
 struct OrImpl
@@ -110,6 +130,19 @@ struct OrImpl
     static constexpr ResultType apply(UInt8 a, UInt8 b) { return a | b; }
     static constexpr ResultType ternaryApply(UInt8 a, UInt8 b) { return std::max(a, b); }
     static constexpr bool specialImplementationForNulls() { return true; }
+
+#if USE_EMBEDDED_COMPILER
+    static llvm::Value * apply(llvm::IRBuilder<> & builder, llvm::Value * a, llvm::Value * b)
+    {
+        return builder.CreateOr(a, b);
+    }
+
+    static llvm::Value * ternaryApply(llvm::IRBuilder<> & builder, llvm::Value * a, llvm::Value * b)
+    {
+        return builder.CreateSelect(builder.CreateICmpUGT(a, b), a, b);
+    }
+#endif
+
 };
 
 struct XorImpl
@@ -128,6 +161,12 @@ struct XorImpl
     {
         return builder.CreateXor(a, b);
     }
+
+    static llvm::Value * ternaryApply(llvm::IRBuilder<> & builder, llvm::Value * a, llvm::Value * b)
+    {
+        llvm::Value * xor_result = builder.CreateXor(a, b);
+        return builder.CreateSelect(xor_result, builder.getInt8(Ternary::True), builder.getInt8(Ternary::False));
+    }
 #endif
 };
 
@@ -184,47 +223,51 @@ public:
     ColumnPtr getConstantResultForNonConstArguments(const ColumnsWithTypeAndName & arguments, const DataTypePtr & result_type) const override;
 
 #if USE_EMBEDDED_COMPILER
-    bool isCompilableImpl(const DataTypes &, const DataTypePtr &) const override { return useDefaultImplementationForNulls(); }
+    bool isCompilableImpl(const DataTypes & arguments, const DataTypePtr &) const override
+    {
+        for (const auto & arg : arguments)
+        {
+            if (!canBeNativeType(arg))
+                return false;
+        }
+        return true;
+    }
 
-    llvm::Value * compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & values, const DataTypePtr &) const override
+    llvm::Value * compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & values, const DataTypePtr & result_type) const override
     {
         assert(!values.empty());
 
         auto & b = static_cast<llvm::IRBuilder<> &>(builder);
-        if constexpr (!Impl::isSaturable())
+        if (useDefaultImplementationForNulls() || !result_type->isNullable())
         {
-            auto * result = nativeBoolCast(b, values[0]);
+            llvm::Value * result = nativeBoolCast(b, values[0]);
             for (size_t i = 1; i < values.size(); ++i)
-                result = Impl::apply(b, result, nativeBoolCast(b, values[i]));
+            {
+                llvm::Value * casted_value = nativeBoolCast(b, values[i]);
+                result = Impl::apply(b, result, casted_value);
+            }
             return b.CreateSelect(result, b.getInt8(1), b.getInt8(0));
         }
-
+        else
-        constexpr bool break_on_true = Impl::isSaturatedValue(true);
-        auto * next = b.GetInsertBlock();
-        auto * stop = llvm::BasicBlock::Create(next->getContext(), "", next->getParent());
-        b.SetInsertPoint(stop);
-
-        auto * phi = b.CreatePHI(b.getInt8Ty(), static_cast<unsigned>(values.size()));
-
-        for (size_t i = 0; i < values.size(); ++i)
         {
-            b.SetInsertPoint(next);
+            /// First we need to cast all values to ternary logic
-            auto * value = values[i].value;
+            llvm::Value * ternary_result = nativeTernaryCast(b, values[0]);
-            auto * truth = nativeBoolCast(b, values[i]);
+            for (size_t i = 1; i < values.size(); ++i)
-            if (!values[i].type->equals(DataTypeUInt8{}))
-                value = b.CreateSelect(truth, b.getInt8(1), b.getInt8(0));
-            phi->addIncoming(value, b.GetInsertBlock());
-            if (i + 1 < values.size())
             {
-                next = llvm::BasicBlock::Create(next->getContext(), "", next->getParent());
+                llvm::Value * casted_value = nativeTernaryCast(b, values[i]);
-                b.CreateCondBr(truth, break_on_true ? stop : next, break_on_true ? next : stop);
+                ternary_result = Impl::ternaryApply(b, ternary_result, casted_value);
             }
+
+            /// Then transform ternary logic to struct which represents nullable result
+            llvm::Value * is_null = b.CreateICmpEQ(ternary_result, b.getInt8(Ternary::Null));
+            llvm::Value * is_true = b.CreateICmpEQ(ternary_result, b.getInt8(Ternary::True));
+
+            auto * nullable_result_type = toNativeType(b, result_type);
+            auto * nullable_result = llvm::Constant::getNullValue(nullable_result_type);
+            auto * nullable_result_with_value
+                = b.CreateInsertValue(nullable_result, b.CreateSelect(is_true, b.getInt8(1), b.getInt8(0)), {0});
+            return b.CreateInsertValue(nullable_result_with_value, is_null, {1});
         }
-
-        b.CreateBr(stop);
-        b.SetInsertPoint(stop);
-
-        return phi;
     }
 #endif
 };

src/Functions/abs.cpp

@@ -6,6 +6,11 @@
 namespace DB
 {
 
+namespace ErrorCodes
+{
+extern const int LOGICAL_ERROR;
+}
+
 template <typename A>
 struct AbsImpl
 {
@@ -27,25 +32,65 @@ struct AbsImpl
     }
 
 #if USE_EMBEDDED_COMPILER
-    static constexpr bool compilable = false; /// special type handling, some other time
+    static constexpr bool compilable = true;
+
+    static llvm::Value * compile(llvm::IRBuilder<> & b, llvm::Value * arg, bool sign)
+    {
+        const auto & type = arg->getType();
+        if (type->isIntegerTy())
+        {
+            if (sign)
+            {
+                auto & context = b.getContext();
+                auto * signed_type = arg->getType();
+                auto * unsigned_type = llvm::IntegerType::get(context, signed_type->getIntegerBitWidth());
+
+                auto * is_negative = b.CreateICmpSLT(arg, llvm::ConstantInt::get(signed_type, 0));
+                auto * neg_value = b.CreateNeg(arg);
+                auto * abs_value = b.CreateSelect(is_negative, neg_value, arg);
+                return b.CreateZExt(abs_value, unsigned_type);
+            }
+            else
+            {
+                return arg;
+            }
+        }
+        else if (type->isDoubleTy() || type->isFloatTy())
+        {
+            auto * func_fabs = llvm::Intrinsic::getDeclaration(b.GetInsertBlock()->getModule(), llvm::Intrinsic::fabs, {type});
+            return b.CreateCall(func_fabs, {arg});
+        }
+        else
+            throw Exception(ErrorCodes::LOGICAL_ERROR, "AbsImpl compilation expected native integer or floating point type");
+    }
 #endif
 };
 
-struct NameAbs { static constexpr auto name = "abs"; };
+struct NameAbs
+{
+    static constexpr auto name = "abs";
+};
 using FunctionAbs = FunctionUnaryArithmetic<AbsImpl, NameAbs, false>;
 
-template <> struct FunctionUnaryArithmeticMonotonicity<NameAbs>
+template <>
+struct FunctionUnaryArithmeticMonotonicity<NameAbs>
 {
     static bool has() { return true; }
     static IFunction::Monotonicity get(const Field & left, const Field & right)
     {
-        Float64 left_float = left.isNull() ? -std::numeric_limits<Float64>::infinity() : applyVisitor(FieldVisitorConvertToNumber<Float64>(), left);
+        Float64 left_float
-        Float64 right_float = right.isNull() ? std::numeric_limits<Float64>::infinity() : applyVisitor(FieldVisitorConvertToNumber<Float64>(), right);
+            = left.isNull() ? -std::numeric_limits<Float64>::infinity() : applyVisitor(FieldVisitorConvertToNumber<Float64>(), left);
+        Float64 right_float
+            = right.isNull() ? std::numeric_limits<Float64>::infinity() : applyVisitor(FieldVisitorConvertToNumber<Float64>(), right);
 
         if ((left_float < 0 && right_float > 0) || (left_float > 0 && right_float < 0))
             return {};
 
-        return { .is_monotonic = true, .is_positive = std::min(left_float, right_float) >= 0, .is_strict = true, };
+        return {
+            .is_monotonic = true,
+            .is_positive = std::min(left_float, right_float) >= 0,
+            .is_strict = true,
+        };
     }
 };
 

src/Functions/assumeNotNull.cpp

@@ -4,6 +4,11 @@
 #include <Core/ColumnNumbers.h>
 #include <Columns/ColumnNullable.h>
 
+#if USE_EMBEDDED_COMPILER
+#    include <DataTypes/Native.h>
+#    include <llvm/IR/IRBuilder.h>
+#endif
+
 
 namespace DB
 {
@@ -58,6 +63,22 @@ public:
             return nullable_col->getNestedColumnPtr();
         return col;
     }
+
+#if USE_EMBEDDED_COMPILER
+    bool isCompilableImpl(const DataTypes & arguments, const DataTypePtr &) const override { return canBeNativeType(arguments[0]); }
+
+    llvm::Value *
+    compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & arguments, const DataTypePtr & /*result_type*/) const override
+    {
+        auto & b = static_cast<llvm::IRBuilder<> &>(builder);
+        if (arguments[0].type->isNullable())
+            return b.CreateExtractValue(arguments[0].value, {0});
+        else
+            return arguments[0].value;
+    }
+#endif
+
+
 };
 
 }

src/Functions/bitCount.cpp

@@ -7,6 +7,10 @@
 namespace DB
 {
 
+namespace ErrorCodes
+{
+extern const int LOGICAL_ERROR;
+}
 template <typename A>
 struct BitCountImpl
 {
@@ -38,7 +42,26 @@ struct BitCountImpl
     }
 
 #if USE_EMBEDDED_COMPILER
-    static constexpr bool compilable = false;
+    static constexpr bool compilable = true;
+
+    static llvm::Value * compile(llvm::IRBuilder<> & b, llvm::Value * arg, bool)
+    {
+        const auto & type = arg->getType();
+        llvm::Value * int_value = nullptr;
+
+        if (type->isIntegerTy())
+            int_value = arg;
+        else if (type->isFloatTy())
+            int_value = b.CreateBitCast(arg, llvm::Type::getInt32Ty(b.getContext()));
+        else if (type->isDoubleTy())
+            int_value = b.CreateBitCast(arg, llvm::Type::getInt64Ty(b.getContext()));
+        else
+            throw Exception(ErrorCodes::LOGICAL_ERROR, "BitCountImpl compilation expected native integer or floating-point type");
+
+        auto * func_ctpop = llvm::Intrinsic::getDeclaration(b.GetInsertBlock()->getModule(), llvm::Intrinsic::ctpop, {int_value->getType()});
+        llvm::Value * ctpop_value = b.CreateCall(func_ctpop, {int_value});
+        return b.CreateZExtOrTrunc(ctpop_value, llvm::Type::getInt8Ty(b.getContext()));
+    }
 #endif
 };
 

src/Functions/identity.h

@@ -2,6 +2,10 @@
 #include <Functions/IFunction.h>
 #include <Interpreters/Context_fwd.h>
 
+#if USE_EMBEDDED_COMPILER
+#    include <DataTypes/Native.h>
+#    include <llvm/IR/IRBuilder.h>
+#endif
 
 namespace DB
 {
@@ -11,6 +15,11 @@ namespace ErrorCodes
     extern const int BAD_ARGUMENTS;
 }
 
+struct IdentityName
+{
+    static constexpr auto name = "identity";
+};
+
 template<typename Name>
 class FunctionIdentityBase : public IFunction
 {
@@ -32,12 +41,21 @@ public:
     {
         return arguments.front().column;
     }
+
+#if USE_EMBEDDED_COMPILER
+    bool isCompilableImpl(const DataTypes & /*types*/, const DataTypePtr & result_type) const override
+    {
+        return Name::name == IdentityName::name && canBeNativeType(result_type);
+    }
+
+    llvm::Value *
+    compileImpl(llvm::IRBuilderBase & /*builder*/, const ValuesWithType & arguments, const DataTypePtr & /*result_type*/) const override
+    {
+        return arguments[0].value;
+    }
+#endif
 };
 
-struct IdentityName
-{
-    static constexpr auto name = "identity";
-};
 
 struct ScalarSubqueryResultName
 {

src/Functions/isNotNull.cpp

@@ -11,6 +11,11 @@
 #include <Interpreters/Context.h>
 #include <Common/assert_cast.h>
 
+#if USE_EMBEDDED_COMPILER
+#    include <DataTypes/Native.h>
+#    include <llvm/IR/IRBuilder.h>
+#endif
+
 namespace DB
 {
 namespace Setting
@@ -110,6 +115,23 @@ public:
         return DataTypeUInt8().createColumnConst(elem.column->size(), 1u);
     }
 
+#if USE_EMBEDDED_COMPILER
+    bool isCompilableImpl(const DataTypes & arguments, const DataTypePtr &) const override { return canBeNativeType(arguments[0]); }
+
+    llvm::Value *
+    compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & arguments, const DataTypePtr & /*result_type*/) const override
+    {
+        auto & b = static_cast<llvm::IRBuilder<> &>(builder);
+        if (arguments[0].type->isNullable())
+        {
+            auto * is_null = b.CreateExtractValue(arguments[0].value, {1});
+            return b.CreateNot(is_null);
+        }
+        else
+            return b.getInt8(1);
+    }
+#endif
+
 private:
     MULTITARGET_FUNCTION_AVX2_SSE42(
     MULTITARGET_FUNCTION_HEADER(static void NO_INLINE), vectorImpl, MULTITARGET_FUNCTION_BODY((const PaddedPODArray<UInt8> & null_map, PaddedPODArray<UInt8> & res) /// NOLINT

src/Functions/isNull.cpp

@@ -10,6 +10,11 @@
 #include <Core/Settings.h>
 #include <Interpreters/Context.h>
 
+#if USE_EMBEDDED_COMPILER
+#    include <DataTypes/Native.h>
+#    include <llvm/IR/IRBuilder.h>
+#endif
+
 
 namespace DB
 {
@@ -107,6 +112,21 @@ public:
         return DataTypeUInt8().createColumnConst(elem.column->size(), 0u);
     }
 
+#if USE_EMBEDDED_COMPILER
+    bool isCompilableImpl(const DataTypes & arguments, const DataTypePtr &) const override { return canBeNativeType(arguments[0]); }
+
+    llvm::Value *
+    compileImpl(llvm::IRBuilderBase & builder, const ValuesWithType & arguments, const DataTypePtr & /*result_type*/) const override
+    {
+        auto & b = static_cast<llvm::IRBuilder<> &>(builder);
+        if (arguments[0].type->isNullable())
+            return b.CreateExtractValue(arguments[0].value, {1});
+        else
+            return b.getInt8(0);
+    }
+#endif
+
+
 private:
     bool use_analyzer;
 };

src/Functions/sign.cpp

@@ -5,6 +5,12 @@
 
 namespace DB
 {
+
+namespace ErrorCodes
+{
+extern const int LOGICAL_ERROR;
+}
+
 template <typename A>
 struct SignImpl
 {
@@ -22,7 +28,44 @@ struct SignImpl
     }
 
 #if USE_EMBEDDED_COMPILER
-    static constexpr bool compilable = false;
+    static constexpr bool compilable = true;
+
+    static llvm::Value * compile(llvm::IRBuilder<> & b, llvm::Value * arg, bool sign)
+    {
+        auto * result_type = b.getInt8Ty();
+        auto * res_zero = llvm::ConstantInt::getSigned(result_type, 0);
+        auto * res_one = llvm::ConstantInt::getSigned(result_type, 1);
+        auto * res_minus_one = llvm::ConstantInt::getSigned(result_type, -1);
+
+        const auto & type = arg->getType();
+        if (type->isIntegerTy())
+        {
+            auto * zero = llvm::ConstantInt::get(type, 0, sign);
+            auto * is_zero = b.CreateICmpEQ(arg, zero);
+
+            if (sign)
+            {
+                auto * is_negative = b.CreateICmpSLT(arg, res_zero);
+                auto * select_zero = b.CreateSelect(is_zero, res_zero, res_one);
+                return b.CreateSelect(is_negative, res_minus_one, select_zero);
+            }
+            else
+                return b.CreateSelect(is_zero, res_zero, res_one);
+        }
+        else if (type->isDoubleTy() || type->isFloatTy())
+        {
+            auto * zero = llvm::ConstantFP::get(type, 0.0);
+            auto * is_zero = b.CreateFCmpOEQ(arg, zero);
+            auto * is_negative = b.CreateFCmpOLT(arg, zero);
+
+            auto * select_zero = b.CreateSelect(is_zero, res_zero, res_one);
+            return b.CreateSelect(is_negative, res_minus_one, select_zero);
+        }
+        else
+            throw Exception(ErrorCodes::LOGICAL_ERROR, "SignImpl compilation expected native integer or floating point type");
+    }
+
+
 #endif
 };
 

src/Interpreters/ActionsDAG.cpp

@@ -1460,7 +1460,8 @@ ActionsDAG ActionsDAG::makeConvertingActions(
     MatchColumnsMode mode,
     bool ignore_constant_values,
     bool add_casted_columns,
-    NameToNameMap * new_names)
+    NameToNameMap * new_names,
+    NameSet * columns_contain_compiled_function)
 {
     size_t num_input_columns = source.size();
     size_t num_result_columns = result.size();
@@ -1533,6 +1534,15 @@ ActionsDAG ActionsDAG::makeConvertingActions(
                         "Cannot convert column `{}` because it is constant but values of constants are different in source and result",
                         res_elem.name);
             }
+            else if (columns_contain_compiled_function && columns_contain_compiled_function->contains(res_elem.name))
+            {
+                /// It may happen when JIT compilation is enabled that source column is constant and destination column is not constant.
+                /// e.g. expression "and(equals(materialize(null::Nullable(UInt64)), null::Nullable(UInt64)), equals(null::Nullable(UInt64), null::Nullable(UInt64)))"
+                /// compiled expression is "and(equals(input: Nullable(UInt64), null), null). Partial evaluation of the compiled expression isn't able to infer that the result column is constant.
+                /// It causes inconsistency between pipeline header(source column is not constant) and output header of ExpressionStep(destination column is constant).
+                /// So we need to convert non-constant column to constant column under this condition.
+                dst_node = &actions_dag.addColumn(res_elem);
+            }
             else
                 throw Exception(
                     ErrorCodes::ILLEGAL_COLUMN,

src/Interpreters/ActionsDAG.h

@@ -306,7 +306,8 @@ public:
         MatchColumnsMode mode,
         bool ignore_constant_values = false,
         bool add_casted_columns = false,
-        NameToNameMap * new_names = nullptr);
+        NameToNameMap * new_names = nullptr,
+        NameSet * columns_contain_compiled_function = nullptr);
 
     /// Create expression which add const column and then materialize it.
     static ActionsDAG makeAddingColumnActions(ColumnWithTypeAndName column);

src/Interpreters/ExpressionActions.cpp

@@ -1,4 +1,5 @@
 #include <Interpreters/Set.h>
+#include <Common/logger_useful.h>
 #include <Common/ProfileEvents.h>
 #include <Interpreters/ArrayJoinAction.h>
 #include <Interpreters/ExpressionActions.h>
@@ -59,7 +60,15 @@ ExpressionActions::ExpressionActions(ActionsDAG actions_dag_, const ExpressionAc
 
 #if USE_EMBEDDED_COMPILER
     if (settings.can_compile_expressions && settings.compile_expressions == CompileExpressions::yes)
+    {
+        LOG_TEST(
+            getLogger("ExpressionActions"),
+            "Actions before compilation: {} with {} lazy_executed_nodes",
+            actions_dag.dumpDAG(),
+            lazy_executed_nodes.size());
         actions_dag.compileExpressions(settings.min_count_to_compile_expression, lazy_executed_nodes);
+        LOG_TEST(getLogger("ExpressionActions"), "Actions after compilation: {}", actions_dag.dumpDAG());
+    }
 #endif
 
     linearizeActions(lazy_executed_nodes);

src/Interpreters/JIT/CHJIT.cpp

@@ -3,6 +3,7 @@
 #if USE_EMBEDDED_COMPILER
 
 #include <sys/mman.h>
+#include <cmath>
 
 #include <boost/noncopyable.hpp>
 
@@ -370,6 +371,9 @@ CHJIT::CHJIT()
     symbol_resolver->registerSymbol("memset", reinterpret_cast<void *>(&memset));
     symbol_resolver->registerSymbol("memcpy", reinterpret_cast<void *>(&memcpy));
     symbol_resolver->registerSymbol("memcmp", reinterpret_cast<void *>(&memcmp));
+
+    double (*fmod_ptr)(double, double) = &fmod;
+    symbol_resolver->registerSymbol("fmod", reinterpret_cast<void *>(fmod_ptr));
 }
 
 CHJIT::~CHJIT() = default;

src/Interpreters/JIT/CompileDAG.cpp

@@ -46,14 +46,18 @@ ValueWithType CompileDAG::compile(llvm::IRBuilderBase & builder, const ValuesWit
             {
                 ValuesWithType temporary_values;
                 temporary_values.reserve(node.arguments.size());
-
                 for (auto argument_index : node.arguments)
                 {
                     assert(compiled_values[argument_index].value != nullptr);
                     temporary_values.emplace_back(compiled_values[argument_index]);
                 }
 
-                compiled_values[compiled_values_index] = {node.function->compile(builder, temporary_values), node.result_type};
+                ValueWithType compiled_value{node.function->compile(builder, temporary_values), node.function->getResultType()};
+                if (!node.result_type->equals(*node.function->getResultType()))
+                    compiled_values[compiled_values_index] = {nativeCast(b, compiled_value, node.result_type), node.result_type};
+                else
+                    compiled_values[compiled_values_index] = std::move(compiled_value);
+
                 break;
             }
             case CompileType::INPUT:

src/Processors/QueryPlan/ExpressionStep.cpp

@@ -25,6 +25,34 @@ static ITransformingStep::Traits getTraits(const ActionsDAG & actions)
     };
 }
 
+static bool containsCompiledFunction(const ActionsDAG::Node * node)
+{
+    if (node->type == ActionsDAG::ActionType::FUNCTION && node->is_function_compiled)
+        return true;
+
+    const auto & children = node->children;
+    if (children.empty())
+        return false;
+
+    bool result = false;
+    for (const auto & child : children)
+        result |= containsCompiledFunction(child);
+    return result;
+}
+
+static NameSet getColumnsContainCompiledFunction(const ActionsDAG & actions_dag)
+{
+    NameSet result;
+    for (const auto * node : actions_dag.getOutputs())
+    {
+        if (containsCompiledFunction(node))
+        {
+            result.insert(node->result_name);
+        }
+    }
+    return result;
+}
+
 ExpressionStep::ExpressionStep(const Header & input_header_, ActionsDAG actions_dag_)
     : ITransformingStep(
         input_header_,
@@ -45,10 +73,15 @@ void ExpressionStep::transformPipeline(QueryPipelineBuilder & pipeline, const Bu
 
     if (!blocksHaveEqualStructure(pipeline.getHeader(), *output_header))
     {
+        auto columns_contain_compiled_function = getColumnsContainCompiledFunction(expression->getActionsDAG());
         auto convert_actions_dag = ActionsDAG::makeConvertingActions(
-                pipeline.getHeader().getColumnsWithTypeAndName(),
+            pipeline.getHeader().getColumnsWithTypeAndName(),
-                output_header->getColumnsWithTypeAndName(),
+            output_header->getColumnsWithTypeAndName(),
-                ActionsDAG::MatchColumnsMode::Name);
+            ActionsDAG::MatchColumnsMode::Name,
+            false,
+            false,
+            nullptr,
+            &columns_contain_compiled_function);
         auto convert_actions = std::make_shared<ExpressionActions>(std::move(convert_actions_dag), settings.getActionsSettings());
 
         pipeline.addSimpleTransform([&](const Block & header)