diff --git a/cmake/linux/default_libs.cmake b/cmake/linux/default_libs.cmake
index 51620bc9f33..79875e1ed6b 100644
--- a/cmake/linux/default_libs.cmake
+++ b/cmake/linux/default_libs.cmake
@@ -3,8 +3,7 @@
 
 set (DEFAULT_LIBS "-nodefaultlibs")
 
-# We need builtins from Clang's RT even without libcxx - for ubsan+int128.
-# See https://bugs.llvm.org/show_bug.cgi?id=16404
+# We need builtins from Clang
 execute_process (COMMAND
     ${CMAKE_CXX_COMPILER} --target=${CMAKE_CXX_COMPILER_TARGET} --print-libgcc-file-name --rtlib=compiler-rt
     OUTPUT_VARIABLE BUILTINS_LIBRARY
diff --git a/src/AggregateFunctions/ReservoirSampler.h b/src/AggregateFunctions/ReservoirSampler.h
index 2668e0dc890..870cb429fb7 100644
--- a/src/AggregateFunctions/ReservoirSampler.h
+++ b/src/AggregateFunctions/ReservoirSampler.h
@@ -276,6 +276,6 @@ private:
     {
         if (OnEmpty == ReservoirSamplerOnEmpty::THROW)
             throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Quantile of empty ReservoirSampler");
-        return NanLikeValueConstructor<ResultType, std::is_floating_point_v<ResultType>>::getValue();
+        return NanLikeValueConstructor<ResultType, is_floating_point<ResultType>>::getValue();
     }
 };
diff --git a/src/Columns/ColumnUnique.cpp b/src/Columns/ColumnUnique.cpp
index 54f45204c00..773edbfd590 100644
--- a/src/Columns/ColumnUnique.cpp
+++ b/src/Columns/ColumnUnique.cpp
@@ -16,6 +16,7 @@ template class ColumnUnique<ColumnInt128>;
 template class ColumnUnique<ColumnUInt128>;
 template class ColumnUnique<ColumnInt256>;
 template class ColumnUnique<ColumnUInt256>;
+template class ColumnUnique<ColumnBFloat16>;
 template class ColumnUnique<ColumnFloat32>;
 template class ColumnUnique<ColumnFloat64>;
 template class ColumnUnique<ColumnString>;
diff --git a/src/Columns/ColumnUnique.h b/src/Columns/ColumnUnique.h
index ffa7c311e9e..ce7bbf0766f 100644
--- a/src/Columns/ColumnUnique.h
+++ b/src/Columns/ColumnUnique.h
@@ -760,6 +760,7 @@ extern template class ColumnUnique<ColumnInt128>;
 extern template class ColumnUnique<ColumnUInt128>;
 extern template class ColumnUnique<ColumnInt256>;
 extern template class ColumnUnique<ColumnUInt256>;
+extern template class ColumnUnique<ColumnBFloat16>;
 extern template class ColumnUnique<ColumnFloat32>;
 extern template class ColumnUnique<ColumnFloat64>;
 extern template class ColumnUnique<ColumnString>;
diff --git a/src/Columns/IColumn.cpp b/src/Columns/IColumn.cpp
index c9a0514af4e..4a3886dddb6 100644
--- a/src/Columns/IColumn.cpp
+++ b/src/Columns/IColumn.cpp
@@ -443,6 +443,7 @@ template class IColumnHelper<ColumnVector<Int32>, ColumnFixedSizeHelper>;
 template class IColumnHelper<ColumnVector<Int64>, ColumnFixedSizeHelper>;
 template class IColumnHelper<ColumnVector<Int128>, ColumnFixedSizeHelper>;
 template class IColumnHelper<ColumnVector<Int256>, ColumnFixedSizeHelper>;
+template class IColumnHelper<ColumnVector<BFloat16>, ColumnFixedSizeHelper>;
 template class IColumnHelper<ColumnVector<Float32>, ColumnFixedSizeHelper>;
 template class IColumnHelper<ColumnVector<Float64>, ColumnFixedSizeHelper>;
 template class IColumnHelper<ColumnVector<UUID>, ColumnFixedSizeHelper>;
diff --git a/src/Common/FieldVisitorConvertToNumber.cpp b/src/Common/FieldVisitorConvertToNumber.cpp
index 75b3fbfe02a..a5963e3d028 100644
--- a/src/Common/FieldVisitorConvertToNumber.cpp
+++ b/src/Common/FieldVisitorConvertToNumber.cpp
@@ -1,5 +1,4 @@
 #include <Common/FieldVisitorConvertToNumber.h>
-#include "base/Decimal.h"
 
 namespace DB
 {
@@ -17,6 +16,7 @@ template class FieldVisitorConvertToNumber<Int128>;
 template class FieldVisitorConvertToNumber<UInt128>;
 template class FieldVisitorConvertToNumber<Int256>;
 template class FieldVisitorConvertToNumber<UInt256>;
+//template class FieldVisitorConvertToNumber<BFloat16>;
 template class FieldVisitorConvertToNumber<Float32>;
 template class FieldVisitorConvertToNumber<Float64>;
 
diff --git a/src/Common/FieldVisitorConvertToNumber.h b/src/Common/FieldVisitorConvertToNumber.h
index 638b8805b6a..38d5dc473c4 100644
--- a/src/Common/FieldVisitorConvertToNumber.h
+++ b/src/Common/FieldVisitorConvertToNumber.h
@@ -129,6 +129,7 @@ extern template class FieldVisitorConvertToNumber<Int128>;
 extern template class FieldVisitorConvertToNumber<UInt128>;
 extern template class FieldVisitorConvertToNumber<Int256>;
 extern template class FieldVisitorConvertToNumber<UInt256>;
+//extern template class FieldVisitorConvertToNumber<BFloat16>;
 extern template class FieldVisitorConvertToNumber<Float32>;
 extern template class FieldVisitorConvertToNumber<Float64>;
 
diff --git a/src/DataTypes/DataTypesBinaryEncoding.cpp b/src/DataTypes/DataTypesBinaryEncoding.cpp
index dc0f2f3f5aa..c3190b462c3 100644
--- a/src/DataTypes/DataTypesBinaryEncoding.cpp
+++ b/src/DataTypes/DataTypesBinaryEncoding.cpp
@@ -96,6 +96,7 @@ enum class BinaryTypeIndex : uint8_t
     SimpleAggregateFunction = 0x2E,
     Nested = 0x2F,
     JSON = 0x30,
+    BFloat16 = 0x31,
 };
 
 /// In future we can introduce more arguments in the JSON data type definition.
@@ -151,6 +152,8 @@ BinaryTypeIndex getBinaryTypeIndex(const DataTypePtr & type)
             return BinaryTypeIndex::Int128;
         case TypeIndex::Int256:
             return BinaryTypeIndex::Int256;
+        case TypeIndex::BFloat16:
+            return BinaryTypeIndex::BFloat16;
         case TypeIndex::Float32:
             return BinaryTypeIndex::Float32;
         case TypeIndex::Float64:
@@ -565,6 +568,8 @@ DataTypePtr decodeDataType(ReadBuffer & buf)
             return std::make_shared<DataTypeInt128>();
         case BinaryTypeIndex::Int256:
             return std::make_shared<DataTypeInt256>();
+        case BinaryTypeIndex::BFloat16:
+            return std::make_shared<DataTypeBFloat16>();
         case BinaryTypeIndex::Float32:
             return std::make_shared<DataTypeFloat32>();
         case BinaryTypeIndex::Float64:
diff --git a/src/DataTypes/DataTypesNumber.cpp b/src/DataTypes/DataTypesNumber.cpp
index 5972cebbca1..4c8918521fe 100644
--- a/src/DataTypes/DataTypesNumber.cpp
+++ b/src/DataTypes/DataTypesNumber.cpp
@@ -112,6 +112,7 @@ template class DataTypeNumber<Int8>;
 template class DataTypeNumber<Int16>;
 template class DataTypeNumber<Int32>;
 template class DataTypeNumber<Int64>;
+template class DataTypeNumber<BFloat16>;
 template class DataTypeNumber<Float32>;
 template class DataTypeNumber<Float64>;
 
diff --git a/src/DataTypes/DataTypesNumber.h b/src/DataTypes/DataTypesNumber.h
index 29899847c4b..a9e77e01b13 100644
--- a/src/DataTypes/DataTypesNumber.h
+++ b/src/DataTypes/DataTypesNumber.h
@@ -63,6 +63,7 @@ extern template class DataTypeNumber<Int8>;
 extern template class DataTypeNumber<Int16>;
 extern template class DataTypeNumber<Int32>;
 extern template class DataTypeNumber<Int64>;
+extern template class DataTypeNumber<BFloat16>;
 extern template class DataTypeNumber<Float32>;
 extern template class DataTypeNumber<Float64>;
 
diff --git a/src/Formats/JSONExtractTree.cpp b/src/Formats/JSONExtractTree.cpp
index ae6051823b7..62905a2e630 100644
--- a/src/Formats/JSONExtractTree.cpp
+++ b/src/Formats/JSONExtractTree.cpp
@@ -131,7 +131,7 @@ bool tryGetNumericValueFromJSONElement(
     switch (element.type())
     {
         case ElementType::DOUBLE:
-            if constexpr (std::is_floating_point_v<NumberType>)
+            if constexpr (is_floating_point<NumberType>)
             {
                 /// We permit inaccurate conversion of double to float.
                 /// Example: double 0.1 from JSON is not representable in float.
@@ -175,7 +175,7 @@ bool tryGetNumericValueFromJSONElement(
                 return false;
 
             auto rb = ReadBufferFromMemory{element.getString()};
-            if constexpr (std::is_floating_point_v<NumberType>)
+            if constexpr (is_floating_point<NumberType>)
             {
                 if (!tryReadFloatText(value, rb) || !rb.eof())
                 {
diff --git a/src/Functions/FunctionBinaryArithmetic.h b/src/Functions/FunctionBinaryArithmetic.h
index df239b820af..854b40df441 100644
--- a/src/Functions/FunctionBinaryArithmetic.h
+++ b/src/Functions/FunctionBinaryArithmetic.h
@@ -110,6 +110,7 @@ template <typename DataType> constexpr bool IsIntegralOrExtendedOrDecimal =
     IsDataTypeDecimal<DataType>;
 
 template <typename DataType> constexpr bool IsFloatingPoint = false;
+template <> inline constexpr bool IsFloatingPoint<DataTypeBFloat16> = true;
 template <> inline constexpr bool IsFloatingPoint<DataTypeFloat32> = true;
 template <> inline constexpr bool IsFloatingPoint<DataTypeFloat64> = true;
 
diff --git a/src/Functions/FunctionsConversion.cpp b/src/Functions/FunctionsConversion.cpp
index 70ec390b576..1c662dd1d9a 100644
--- a/src/Functions/FunctionsConversion.cpp
+++ b/src/Functions/FunctionsConversion.cpp
@@ -2930,6 +2930,7 @@ template <> struct FunctionTo<DataTypeInt32> { using Type = FunctionToInt32; };
 template <> struct FunctionTo<DataTypeInt64> { using Type = FunctionToInt64; };
 template <> struct FunctionTo<DataTypeInt128> { using Type = FunctionToInt128; };
 template <> struct FunctionTo<DataTypeInt256> { using Type = FunctionToInt256; };
+//template <> struct FunctionTo<DataTypeBFloat16> { using Type = FunctionToBFloat16; };
 template <> struct FunctionTo<DataTypeFloat32> { using Type = FunctionToFloat32; };
 template <> struct FunctionTo<DataTypeFloat64> { using Type = FunctionToFloat64; };
 
diff --git a/src/Functions/FunctionsRound.h b/src/Functions/FunctionsRound.h
index 809905c692e..255eca5b406 100644
--- a/src/Functions/FunctionsRound.h
+++ b/src/Functions/FunctionsRound.h
@@ -694,7 +694,7 @@ public:
                 if (arguments.size() > 1)
                 {
                     const ColumnWithTypeAndName & scale_column = arguments[1];
-                    res = Dispatcher<DataType, rounding_mode, tie_breaking_mode>::template apply<ScaleType>(value_arg.column.get(), scale_column.column.get());
+                    res = Dispatcher<DataType, rounding_mode, tie_breaking_mode>::template apply<int>(value_arg.column.get(), scale_column.column.get());
                     return true;
                 }
                 res = Dispatcher<DataType, rounding_mode, tie_breaking_mode>::template apply<int>(value_arg.column.get());
diff --git a/src/IO/readFloatText.cpp b/src/IO/readFloatText.cpp
index 17ccc1b25b7..fb3c86fd7b6 100644
--- a/src/IO/readFloatText.cpp
+++ b/src/IO/readFloatText.cpp
@@ -47,26 +47,35 @@ void assertNaN(ReadBuffer & buf)
 }
 
 
+template void readFloatTextPrecise<BFloat16>(BFloat16 &, ReadBuffer &);
 template void readFloatTextPrecise<Float32>(Float32 &, ReadBuffer &);
 template void readFloatTextPrecise<Float64>(Float64 &, ReadBuffer &);
+template bool tryReadFloatTextPrecise<BFloat16>(BFloat16 &, ReadBuffer &);
 template bool tryReadFloatTextPrecise<Float32>(Float32 &, ReadBuffer &);
 template bool tryReadFloatTextPrecise<Float64>(Float64 &, ReadBuffer &);
 
+template void readFloatTextFast<BFloat16>(BFloat16 &, ReadBuffer &);
 template void readFloatTextFast<Float32>(Float32 &, ReadBuffer &);
 template void readFloatTextFast<Float64>(Float64 &, ReadBuffer &);
+template bool tryReadFloatTextFast<BFloat16>(BFloat16 &, ReadBuffer &);
 template bool tryReadFloatTextFast<Float32>(Float32 &, ReadBuffer &);
 template bool tryReadFloatTextFast<Float64>(Float64 &, ReadBuffer &);
 
+template void readFloatTextSimple<BFloat16>(BFloat16 &, ReadBuffer &);
 template void readFloatTextSimple<Float32>(Float32 &, ReadBuffer &);
 template void readFloatTextSimple<Float64>(Float64 &, ReadBuffer &);
+template bool tryReadFloatTextSimple<BFloat16>(BFloat16 &, ReadBuffer &);
 template bool tryReadFloatTextSimple<Float32>(Float32 &, ReadBuffer &);
 template bool tryReadFloatTextSimple<Float64>(Float64 &, ReadBuffer &);
 
+template void readFloatText<BFloat16>(BFloat16 &, ReadBuffer &);
 template void readFloatText<Float32>(Float32 &, ReadBuffer &);
 template void readFloatText<Float64>(Float64 &, ReadBuffer &);
+template bool tryReadFloatText<BFloat16>(BFloat16 &, ReadBuffer &);
 template bool tryReadFloatText<Float32>(Float32 &, ReadBuffer &);
 template bool tryReadFloatText<Float64>(Float64 &, ReadBuffer &);
 
+template bool tryReadFloatTextNoExponent<BFloat16>(BFloat16 &, ReadBuffer &);
 template bool tryReadFloatTextNoExponent<Float32>(Float32 &, ReadBuffer &);
 template bool tryReadFloatTextNoExponent<Float64>(Float64 &, ReadBuffer &);
 
diff --git a/src/IO/readFloatText.h b/src/IO/readFloatText.h
index c2fec9d4b0b..a7fd6058dd9 100644
--- a/src/IO/readFloatText.h
+++ b/src/IO/readFloatText.h
@@ -222,7 +222,6 @@ ReturnType readFloatTextPreciseImpl(T & x, ReadBuffer & buf)
             break;
         }
 
-
         char tmp_buf[MAX_LENGTH];
         int num_copied_chars = 0;
 
@@ -597,22 +596,85 @@ ReturnType readFloatTextSimpleImpl(T & x, ReadBuffer & buf)
     return ReturnType(true);
 }
 
-template <typename T> void readFloatTextPrecise(T & x, ReadBuffer & in) { readFloatTextPreciseImpl<T, void>(x, in); }
-template <typename T> bool tryReadFloatTextPrecise(T & x, ReadBuffer & in) { return readFloatTextPreciseImpl<T, bool>(x, in); }
+template <typename T> void readFloatTextPrecise(T & x, ReadBuffer & in)
+{
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        readFloatTextPreciseImpl<Float32, void>(tmp, in);
+        x = BFloat16(tmp);
+    }
+    else
+        readFloatTextPreciseImpl<T, void>(x, in);
+}
+
+template <typename T> bool tryReadFloatTextPrecise(T & x, ReadBuffer & in)
+{
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        bool res = readFloatTextPreciseImpl<Float32, bool>(tmp, in);
+        if (res)
+            x = BFloat16(tmp);
+        return res;
+    }
+    else
+        return readFloatTextPreciseImpl<T, bool>(x, in);
+}
 
 template <typename T> void readFloatTextFast(T & x, ReadBuffer & in)
 {
     bool has_fractional;
-    readFloatTextFastImpl<T, void>(x, in, has_fractional);
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        readFloatTextFastImpl<Float32, void>(tmp, in, has_fractional);
+        x = BFloat16(tmp);
+    }
+    else
+        readFloatTextFastImpl<T, void>(x, in, has_fractional);
 }
+
 template <typename T> bool tryReadFloatTextFast(T & x, ReadBuffer & in)
 {
     bool has_fractional;
-    return readFloatTextFastImpl<T, bool>(x, in, has_fractional);
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        bool res = readFloatTextFastImpl<Float32, bool>(tmp, in, has_fractional);
+        if (res)
+            x = BFloat16(tmp);
+        return res;
+    }
+    else
+        return readFloatTextFastImpl<T, bool>(x, in, has_fractional);
 }
 
-template <typename T> void readFloatTextSimple(T & x, ReadBuffer & in) { readFloatTextSimpleImpl<T, void>(x, in); }
-template <typename T> bool tryReadFloatTextSimple(T & x, ReadBuffer & in) { return readFloatTextSimpleImpl<T, bool>(x, in); }
+template <typename T> void readFloatTextSimple(T & x, ReadBuffer & in)
+{
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        readFloatTextSimpleImpl<Float32, void>(tmp, in);
+        x = BFloat16(tmp);
+    }
+    else
+        readFloatTextSimpleImpl<T, void>(x, in);
+}
+
+template <typename T> bool tryReadFloatTextSimple(T & x, ReadBuffer & in)
+{
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        bool res = readFloatTextSimpleImpl<Float32, bool>(tmp, in);
+        if (res)
+            x = BFloat16(tmp);
+        return res;
+    }
+    else
+        return readFloatTextSimpleImpl<T, bool>(x, in);
+}
 
 
 /// Implementation that is selected as default.
@@ -624,18 +686,47 @@ template <typename T> bool tryReadFloatText(T & x, ReadBuffer & in) { return try
 template <typename T> bool tryReadFloatTextNoExponent(T & x, ReadBuffer & in)
 {
     bool has_fractional;
-    return readFloatTextFastImpl<T, bool, false>(x, in, has_fractional);
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        bool res = readFloatTextFastImpl<Float32, bool, false>(tmp, in, has_fractional);
+        if (res)
+            x = BFloat16(tmp);
+        return res;
+
+    }
+    else
+        return readFloatTextFastImpl<T, bool, false>(x, in, has_fractional);
 }
 
 /// With a @has_fractional flag
 /// Used for input_format_try_infer_integers
 template <typename T> bool tryReadFloatTextExt(T & x, ReadBuffer & in, bool & has_fractional)
 {
-    return readFloatTextFastImpl<T, bool>(x, in, has_fractional);
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        bool res = readFloatTextFastImpl<Float32, bool>(tmp, in, has_fractional);
+        if (res)
+            x = BFloat16(tmp);
+        return res;
+    }
+    else
+        return readFloatTextFastImpl<T, bool>(x, in, has_fractional);
 }
+
 template <typename T> bool tryReadFloatTextExtNoExponent(T & x, ReadBuffer & in, bool & has_fractional)
 {
-    return readFloatTextFastImpl<T, bool, false>(x, in, has_fractional);
+    if constexpr (std::is_same_v<T, BFloat16>)
+    {
+        Float32 tmp;
+        bool res = readFloatTextFastImpl<Float32, bool, false>(tmp, in, has_fractional);
+        if (res)
+            x = BFloat16(tmp);
+        return res;
+    }
+    else
+        return readFloatTextFastImpl<T, bool, false>(x, in, has_fractional);
 }
 
 }
diff --git a/src/Processors/Formats/Impl/Parquet/ParquetDataValuesReader.cpp b/src/Processors/Formats/Impl/Parquet/ParquetDataValuesReader.cpp
index b471989076b..4b79be98810 100644
--- a/src/Processors/Formats/Impl/Parquet/ParquetDataValuesReader.cpp
+++ b/src/Processors/Formats/Impl/Parquet/ParquetDataValuesReader.cpp
@@ -580,6 +580,7 @@ template class ParquetPlainValuesReader<ColumnInt32>;
 template class ParquetPlainValuesReader<ColumnUInt32>;
 template class ParquetPlainValuesReader<ColumnInt64>;
 template class ParquetPlainValuesReader<ColumnUInt64>;
+template class ParquetPlainValuesReader<ColumnBFloat16>;
 template class ParquetPlainValuesReader<ColumnFloat32>;
 template class ParquetPlainValuesReader<ColumnFloat64>;
 template class ParquetPlainValuesReader<ColumnDecimal<Decimal32>>;
@@ -602,6 +603,7 @@ template class ParquetRleDictReader<ColumnInt32>;
 template class ParquetRleDictReader<ColumnUInt32>;
 template class ParquetRleDictReader<ColumnInt64>;
 template class ParquetRleDictReader<ColumnUInt64>;
+template class ParquetRleDictReader<ColumnBFloat16>;
 template class ParquetRleDictReader<ColumnFloat32>;
 template class ParquetRleDictReader<ColumnFloat64>;
 template class ParquetRleDictReader<ColumnDecimal<Decimal32>>;
diff --git a/src/Processors/Formats/Impl/Parquet/ParquetLeafColReader.cpp b/src/Processors/Formats/Impl/Parquet/ParquetLeafColReader.cpp
index c3c7db510ed..328dd37107e 100644
--- a/src/Processors/Formats/Impl/Parquet/ParquetLeafColReader.cpp
+++ b/src/Processors/Formats/Impl/Parquet/ParquetLeafColReader.cpp
@@ -644,6 +644,7 @@ template class ParquetLeafColReader<ColumnInt32>;
 template class ParquetLeafColReader<ColumnUInt32>;
 template class ParquetLeafColReader<ColumnInt64>;
 template class ParquetLeafColReader<ColumnUInt64>;
+template class ParquetLeafColReader<ColumnBFloat16>;
 template class ParquetLeafColReader<ColumnFloat32>;
 template class ParquetLeafColReader<ColumnFloat64>;
 template class ParquetLeafColReader<ColumnString>;