diff --git a/programs/obfuscator/Obfuscator.cpp b/programs/obfuscator/Obfuscator.cpp
index 056ed85c670..6891d2113a5 100644
--- a/programs/obfuscator/Obfuscator.cpp
+++ b/programs/obfuscator/Obfuscator.cpp
@@ -46,6 +46,7 @@
 #include <boost/algorithm/string.hpp>
 #include <boost/container/flat_map.hpp>
 #include <Common/TerminalSize.h>
+#include <bit>
 
 
 static const char * documentation = R"(
@@ -186,7 +187,7 @@ static UInt64 transform(UInt64 x, UInt64 seed)
     if (x == 2 || x == 3)
         return x ^ (seed & 1);
 
-    size_t num_leading_zeros = __builtin_clzll(x);
+    size_t num_leading_zeros = std::countl_zero(x);
 
     return feistelNetwork(x, 64 - num_leading_zeros - 1, seed);
 }
diff --git a/src/Columns/ColumnDecimal.cpp b/src/Columns/ColumnDecimal.cpp
index 4290a7a4cb1..63b76dbb230 100644
--- a/src/Columns/ColumnDecimal.cpp
+++ b/src/Columns/ColumnDecimal.cpp
@@ -298,7 +298,7 @@ ColumnPtr ColumnDecimal<T>::filter(const IColumn::Filter & filt, ssize_t result_
         {
             while (mask)
             {
-                size_t index = __builtin_ctzll(mask);
+                size_t index = std::countr_zero(mask);
                 res_data.push_back(data_pos[index]);
             #ifdef __BMI__
                 mask = _blsr_u64(mask);
diff --git a/src/Columns/ColumnFixedString.cpp b/src/Columns/ColumnFixedString.cpp
index b71323f2ec7..158d154f3f8 100644
--- a/src/Columns/ColumnFixedString.cpp
+++ b/src/Columns/ColumnFixedString.cpp
@@ -240,7 +240,7 @@ ColumnPtr ColumnFixedString::filter(const IColumn::Filter & filt, ssize_t result
             size_t res_chars_size = res->chars.size();
             while (mask)
             {
-                size_t index = __builtin_ctzll(mask);
+                size_t index = std::countr_zero(mask);
                 res->chars.resize(res_chars_size + n);
                 memcpySmallAllowReadWriteOverflow15(&res->chars[res_chars_size], data_pos + index * n, n);
                 res_chars_size += n;
diff --git a/src/Columns/ColumnVector.cpp b/src/Columns/ColumnVector.cpp
index 60423e2b0fe..9084e47c2f5 100644
--- a/src/Columns/ColumnVector.cpp
+++ b/src/Columns/ColumnVector.cpp
@@ -508,7 +508,7 @@ ColumnPtr ColumnVector<T>::filter(const IColumn::Filter & filt, ssize_t result_s
         {
             while (mask)
             {
-                size_t index = __builtin_ctzll(mask);
+                size_t index = std::countr_zero(mask);
                 res_data.push_back(data_pos[index]);
             #ifdef __BMI__
                 mask = _blsr_u64(mask);
diff --git a/src/Columns/ColumnsCommon.cpp b/src/Columns/ColumnsCommon.cpp
index cbc62140bef..17f521c242c 100644
--- a/src/Columns/ColumnsCommon.cpp
+++ b/src/Columns/ColumnsCommon.cpp
@@ -2,13 +2,14 @@
 #include <Columns/ColumnVector.h>
 #include <Common/typeid_cast.h>
 #include <Common/HashTable/HashSet.h>
+#include <bit>
 #include "ColumnsCommon.h"
 
 
 namespace DB
 {
 
-#if defined(__SSE2__) && defined(__POPCNT__)
+#if defined(__SSE2__)
 /// Transform 64-byte mask to 64-bit mask.
 static UInt64 toBits64(const Int8 * bytes64)
 {
@@ -41,11 +42,11 @@ size_t countBytesInFilter(const UInt8 * filt, size_t start, size_t end)
 
     const Int8 * end_pos = pos + (end - start);
 
-#if defined(__SSE2__) && defined(__POPCNT__)
+#if defined(__SSE2__)
     const Int8 * end_pos64 = pos + (end - start) / 64 * 64;
 
     for (; pos < end_pos64; pos += 64)
-        count += __builtin_popcountll(toBits64(pos));
+        count += std::popcount(toBits64(pos));
 
     /// TODO Add duff device for tail?
 #endif
@@ -74,11 +75,11 @@ size_t countBytesInFilterWithNull(const IColumn::Filter & filt, const UInt8 * nu
     const Int8 * pos2 = reinterpret_cast<const Int8 *>(null_map) + start;
     const Int8 * end_pos = pos + (end - start);
 
-#if defined(__SSE2__) && defined(__POPCNT__)
+#if defined(__SSE2__)
     const Int8 * end_pos64 = pos + (end - start) / 64 * 64;
 
     for (; pos < end_pos64; pos += 64, pos2 += 64)
-        count += __builtin_popcountll(toBits64(pos) & ~toBits64(pos2));
+        count += std::popcount(toBits64(pos) & ~toBits64(pos2));
 
         /// TODO Add duff device for tail?
 #endif
@@ -259,7 +260,7 @@ namespace
             {
                 while (mask)
                 {
-                    size_t index = __builtin_ctzll(mask);
+                    size_t index = std::countr_zero(mask);
                     copy_array(offsets_pos + index);
                 #ifdef __BMI__
                     mask = _blsr_u64(mask);
diff --git a/src/Columns/ColumnsCommon.h b/src/Columns/ColumnsCommon.h
index 1e5849e2b88..607cff75a98 100644
--- a/src/Columns/ColumnsCommon.h
+++ b/src/Columns/ColumnsCommon.h
@@ -36,7 +36,7 @@ inline UInt64 bytes64MaskToBits64Mask(const UInt8 * bytes64)
         _mm256_loadu_si256(reinterpret_cast<const __m256i *>(bytes64)), zero32))) & 0xffffffff)
         | (static_cast<UInt64>(_mm256_movemask_epi8(_mm256_cmpeq_epi8(
         _mm256_loadu_si256(reinterpret_cast<const __m256i *>(bytes64+32)), zero32))) << 32);
-#elif defined(__SSE2__) && defined(__POPCNT__)
+#elif defined(__SSE2__)
     static const __m128i zero16 = _mm_setzero_si128();
     UInt64 res =
         (static_cast<UInt64>(_mm_movemask_epi8(_mm_cmpeq_epi8(
diff --git a/src/Common/HyperLogLogCounter.h b/src/Common/HyperLogLogCounter.h
index 4644d18ecf3..cdd4f246e53 100644
--- a/src/Common/HyperLogLogCounter.h
+++ b/src/Common/HyperLogLogCounter.h
@@ -11,6 +11,7 @@
 #include <IO/WriteHelpers.h>
 #include <Core/Defines.h>
 
+#include <bit>
 #include <cmath>
 #include <cstring>
 
@@ -205,7 +206,7 @@ struct TrailingZerosCounter<UInt32>
 {
     static int apply(UInt32 val)
     {
-        return __builtin_ctz(val);
+        return std::countr_zero(val);
     }
 };
 
@@ -214,7 +215,7 @@ struct TrailingZerosCounter<UInt64>
 {
     static int apply(UInt64 val)
     {
-        return __builtin_ctzll(val);
+        return std::countr_zero(val);
     }
 };
 
diff --git a/src/Common/IPv6ToBinary.cpp b/src/Common/IPv6ToBinary.cpp
index a8363a46de7..8d335d89353 100644
--- a/src/Common/IPv6ToBinary.cpp
+++ b/src/Common/IPv6ToBinary.cpp
@@ -5,6 +5,7 @@
 #include <Common/formatIPv6.h>
 
 #include <cstring>
+#include <bit>
 
 
 namespace DB
@@ -89,7 +90,7 @@ bool matchIPv6Subnet(const uint8_t * addr, const uint8_t * cidr_addr, UInt8 pref
 
     if (mask)
     {
-        auto offset = __builtin_ctz(mask);
+        auto offset = std::countr_zero(mask);
 
         if (prefix / 8 != offset)
             return prefix / 8 < offset;
diff --git a/src/Common/SpaceSaving.h b/src/Common/SpaceSaving.h
index 0f577349722..20d6812b91b 100644
--- a/src/Common/SpaceSaving.h
+++ b/src/Common/SpaceSaving.h
@@ -78,7 +78,7 @@ private:
     constexpr uint64_t nextAlphaSize(uint64_t x)
     {
         constexpr uint64_t alpha_map_elements_per_counter = 6;
-        return 1ULL << (sizeof(uint64_t) * 8 - __builtin_clzll(x * alpha_map_elements_per_counter));
+        return 1ULL << (sizeof(uint64_t) * 8 - std::countl_zero(x * alpha_map_elements_per_counter));
     }
 
 public:
diff --git a/src/Common/UTF8Helpers.cpp b/src/Common/UTF8Helpers.cpp
index b1d38c4f31a..0af31726f40 100644
--- a/src/Common/UTF8Helpers.cpp
+++ b/src/Common/UTF8Helpers.cpp
@@ -2,6 +2,7 @@
 #include <Common/StringUtils/StringUtils.h>
 
 #include <widechar_width.h>
+#include <bit>
 
 
 namespace DB
@@ -124,7 +125,7 @@ size_t computeWidthImpl(const UInt8 * data, size_t size, size_t prefix, size_t l
 
             if (non_regular_width_mask)
             {
-                auto num_regular_chars = __builtin_ctz(non_regular_width_mask);
+                auto num_regular_chars = std::countr_zero(non_regular_width_mask);
                 width += num_regular_chars;
                 i += num_regular_chars;
                 break;
diff --git a/src/Common/UTF8Helpers.h b/src/Common/UTF8Helpers.h
index 72bdb965789..ce90af3d5ce 100644
--- a/src/Common/UTF8Helpers.h
+++ b/src/Common/UTF8Helpers.h
@@ -83,7 +83,7 @@ inline size_t countCodePoints(const UInt8 * data, size_t size)
     const auto threshold = vdupq_n_s8(0xBF);
 
     for (; data < src_end_sse; data += bytes_sse)
-        res += __builtin_popcountll(get_nibble_mask(vcgtq_s8(vld1q_s8(reinterpret_cast<const int8_t *>(data)), threshold)));
+        res += std::popcount(get_nibble_mask(vcgtq_s8(vld1q_s8(reinterpret_cast<const int8_t *>(data)), threshold)));
     res >>= 2;
 #endif
 
diff --git a/src/Common/examples/average.cpp b/src/Common/examples/average.cpp
index 5a42d8c5020..d2802717fc8 100644
--- a/src/Common/examples/average.cpp
+++ b/src/Common/examples/average.cpp
@@ -1,5 +1,6 @@
 #include <iostream>
 #include <string>
+#include <bit>
 
 #include <fmt/format.h>
 
@@ -561,7 +562,7 @@ int main(int argc, char ** argv)
     /// Fill source data
     for (size_t i = 0; i < size; ++i)
     {
-        keys[i] = __builtin_ctz(i + 1); /// Make keys to have just slightly more realistic distribution.
+        keys[i] = std::countr_zero(i + 1); /// Make keys to have just slightly more realistic distribution.
         values[i] = 1234.5; /// The distribution of values does not affect execution speed.
     }
 
diff --git a/src/Common/memcmpSmall.h b/src/Common/memcmpSmall.h
index 7b977a4a23c..e95a21b836d 100644
--- a/src/Common/memcmpSmall.h
+++ b/src/Common/memcmpSmall.h
@@ -1,6 +1,7 @@
 #pragma once
 
 #include <algorithm>
+#include <bit>
 #include <cstdint>
 
 #include <Core/Defines.h>
@@ -50,7 +51,7 @@ inline int memcmpSmallAllowOverflow15(const Char * a, size_t a_size, const Char
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= min_size)
                 break;
@@ -82,7 +83,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= min_size)
                 break;
@@ -123,7 +124,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= max_size)
                 return 0;
@@ -150,7 +151,7 @@ inline int memcmpSmallAllowOverflow15(const Char * a, const Char * b, size_t siz
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= size)
                 return 0;
@@ -180,7 +181,7 @@ inline bool memequalSmallAllowOverflow15(const Char * a, size_t a_size, const Ch
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
             return offset >= a_size;
         }
     }
@@ -203,7 +204,7 @@ inline int memcmpSmallMultipleOf16(const Char * a, const Char * b, size_t size)
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
             return detail::cmp(a[offset], b[offset]);
         }
     }
@@ -222,7 +223,7 @@ inline int memcmp16(const Char * a, const Char * b)
 
     if (mask)
     {
-        auto offset = __builtin_ctz(mask);
+        auto offset = std::countr_zero(mask);
         return detail::cmp(a[offset], b[offset]);
     }
 
@@ -252,7 +253,7 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size)
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
             return offset >= size;
         }
     }
@@ -285,7 +286,7 @@ inline int memcmpSmallAllowOverflow15(const Char * a, size_t a_size, const Char
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= min_size)
                 break;
@@ -317,7 +318,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= min_size)
                 break;
@@ -359,7 +360,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= max_size)
                 return 0;
@@ -386,7 +387,7 @@ inline int memcmpSmallAllowOverflow15(const Char * a, const Char * b, size_t siz
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
 
             if (offset >= size)
                 return 0;
@@ -416,7 +417,7 @@ inline bool memequalSmallAllowOverflow15(const Char * a, size_t a_size, const Ch
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
             return offset >= a_size;
         }
     }
@@ -439,7 +440,7 @@ inline int memcmpSmallMultipleOf16(const Char * a, const Char * b, size_t size)
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
             return detail::cmp(a[offset], b[offset]);
         }
     }
@@ -459,7 +460,7 @@ inline int memcmp16(const Char * a, const Char * b)
 
     if (mask)
     {
-        auto offset = __builtin_ctz(mask);
+        auto offset = std::countr_zero(mask);
         return detail::cmp(a[offset], b[offset]);
     }
 
@@ -490,7 +491,7 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size)
 
         if (mask)
         {
-            offset += __builtin_ctz(mask);
+            offset += std::countr_zero(mask);
             return offset >= size;
         }
     }
@@ -523,7 +524,7 @@ inline int memcmpSmallAllowOverflow15(const Char * a, size_t a_size, const Char
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
 
             if (offset >= min_size)
                 break;
@@ -548,7 +549,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
 
             if (offset >= min_size)
                 break;
@@ -589,7 +590,7 @@ inline int memcmpSmallLikeZeroPaddedAllowOverflow15(const Char * a, size_t a_siz
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
 
             if (offset >= max_size)
                 return 0;
@@ -611,7 +612,7 @@ inline int memcmpSmallAllowOverflow15(const Char * a, const Char * b, size_t siz
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
 
             if (offset >= size)
                 return 0;
@@ -637,7 +638,7 @@ inline bool memequalSmallAllowOverflow15(const Char * a, size_t a_size, const Ch
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
             return offset >= a_size;
         }
     }
@@ -656,7 +657,7 @@ inline int memcmpSmallMultipleOf16(const Char * a, const Char * b, size_t size)
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
             return detail::cmp(a[offset], b[offset]);
         }
     }
@@ -672,7 +673,7 @@ inline int memcmp16(const Char * a, const Char * b)
     mask = ~mask;
     if (mask)
     {
-        auto offset = __builtin_ctzll(mask) >> 2;
+        auto offset = std::countr_zero(mask) >> 2;
         return detail::cmp(a[offset], b[offset]);
     }
     return 0;
@@ -694,7 +695,7 @@ inline bool memoryIsZeroSmallAllowOverflow15(const void * data, size_t size)
 
         if (mask)
         {
-            offset += __builtin_ctzll(mask) >> 2;
+            offset += std::countr_zero(mask) >> 2;
             return offset >= size;
         }
     }
diff --git a/src/Compression/CompressedReadBufferBase.cpp b/src/Compression/CompressedReadBufferBase.cpp
index 2c85dc6d9a9..9101caf568e 100644
--- a/src/Compression/CompressedReadBufferBase.cpp
+++ b/src/Compression/CompressedReadBufferBase.cpp
@@ -1,5 +1,6 @@
 #include "CompressedReadBufferBase.h"
 
+#include <bit>
 #include <cstring>
 #include <cassert>
 #include <city.h>
@@ -93,8 +94,8 @@ static void validateChecksum(char * data, size_t size, const Checksum expected_c
     }
 
     /// Check if the difference caused by single bit flip in stored checksum.
-    size_t difference = __builtin_popcountll(expected_checksum.first ^ calculated_checksum.first)
-        + __builtin_popcountll(expected_checksum.second ^ calculated_checksum.second);
+    size_t difference = std::popcount(expected_checksum.first ^ calculated_checksum.first)
+        + std::popcount(expected_checksum.second ^ calculated_checksum.second);
 
     if (difference == 1)
     {
diff --git a/src/Compression/CompressionCodecT64.cpp b/src/Compression/CompressionCodecT64.cpp
index 0efa98fbd82..9ed37c2d676 100644
--- a/src/Compression/CompressionCodecT64.cpp
+++ b/src/Compression/CompressionCodecT64.cpp
@@ -8,6 +8,7 @@
 #include <Parsers/ASTFunction.h>
 #include <IO/WriteHelpers.h>
 #include <Core/Types.h>
+#include <bit>
 
 
 namespace DB
@@ -413,7 +414,7 @@ UInt32 getValuableBitsNumber(UInt64 min, UInt64 max)
 {
     UInt64 diff_bits = min ^ max;
     if (diff_bits)
-        return 64 - __builtin_clzll(diff_bits);
+        return 64 - std::countl_zero(diff_bits);
     return 0;
 }
 
diff --git a/src/Functions/FunctionsBitToArray.cpp b/src/Functions/FunctionsBitToArray.cpp
index 22a56ba35e6..f154884a0fd 100644
--- a/src/Functions/FunctionsBitToArray.cpp
+++ b/src/Functions/FunctionsBitToArray.cpp
@@ -8,6 +8,7 @@
 #include <Functions/IFunction.h>
 #include <IO/WriteBufferFromVector.h>
 #include <IO/WriteHelpers.h>
+#include <bit>
 
 
 namespace DB
@@ -285,7 +286,7 @@ public:
             {
                 while (x)
                 {
-                    result_array_values_data.push_back(getTrailingZeroBitsUnsafe(x));
+                    result_array_values_data.push_back(std::countr_zero(x));
                     x &= (x - 1);
                 }
             }
diff --git a/src/Functions/bitHammingDistance.cpp b/src/Functions/bitHammingDistance.cpp
index 2090d17432c..75928c2a8af 100644
--- a/src/Functions/bitHammingDistance.cpp
+++ b/src/Functions/bitHammingDistance.cpp
@@ -1,5 +1,6 @@
 #include <Functions/FunctionBinaryArithmetic.h>
 #include <Functions/FunctionFactory.h>
+#include <bit>
 
 namespace DB
 {
@@ -14,7 +15,7 @@ struct BitHammingDistanceImpl
     static inline NO_SANITIZE_UNDEFINED Result apply(A a, B b)
     {
         UInt64 res = static_cast<UInt64>(a) ^ static_cast<UInt64>(b);
-        return __builtin_popcountll(res);
+        return std::popcount(res);
     }
 
 #if USE_EMBEDDED_COMPILER
diff --git a/src/IO/ReadHelpers.cpp b/src/IO/ReadHelpers.cpp
index c2b0a0f65d7..fb5d0b9aea4 100644
--- a/src/IO/ReadHelpers.cpp
+++ b/src/IO/ReadHelpers.cpp
@@ -10,6 +10,7 @@
 #include <IO/Operators.h>
 #include <base/find_symbols.h>
 #include <cstdlib>
+#include <bit>
 
 #ifdef __SSE2__
     #include <emmintrin.h>
@@ -698,7 +699,7 @@ void readCSVStringInto(Vector & s, ReadBuffer & buf, const FormatSettings::CSV &
                     uint16_t bit_mask = _mm_movemask_epi8(eq);
                     if (bit_mask)
                     {
-                        next_pos += __builtin_ctz(bit_mask);
+                        next_pos += std::countr_zero(bit_mask);
                         return;
                     }
                 }
@@ -716,7 +717,7 @@ void readCSVStringInto(Vector & s, ReadBuffer & buf, const FormatSettings::CSV &
                     uint64_t bit_mask = get_nibble_mask(eq);
                     if (bit_mask)
                     {
-                        next_pos += __builtin_ctzll(bit_mask) >> 2;
+                        next_pos += std::countr_zero(bit_mask) >> 2;
                         return;
                     }
                 }
diff --git a/src/Interpreters/ConcurrentHashJoin.cpp b/src/Interpreters/ConcurrentHashJoin.cpp
index 8d91375e661..fd802a49008 100644
--- a/src/Interpreters/ConcurrentHashJoin.cpp
+++ b/src/Interpreters/ConcurrentHashJoin.cpp
@@ -33,7 +33,7 @@ static UInt32 toPowerOfTwo(UInt32 x)
 {
     if (x <= 1)
         return 1;
-    return static_cast<UInt32>(1) << (32 - __builtin_clz(x - 1));
+    return static_cast<UInt32>(1) << (32 - std::countl_zero(x - 1));
 }
 
 ConcurrentHashJoin::ConcurrentHashJoin(ContextPtr context_, std::shared_ptr<TableJoin> table_join_, size_t slots_, const Block & right_sample_block, bool any_take_last_row_)
diff --git a/src/Storages/MergeTree/MergeTreeRangeReader.cpp b/src/Storages/MergeTree/MergeTreeRangeReader.cpp
index a10192c4cc1..79854785016 100644
--- a/src/Storages/MergeTree/MergeTreeRangeReader.cpp
+++ b/src/Storages/MergeTree/MergeTreeRangeReader.cpp
@@ -8,6 +8,7 @@
 #include <base/range.h>
 #include <Interpreters/castColumn.h>
 #include <DataTypes/DataTypeNothing.h>
+#include <bit>
 
 #ifdef __SSE2__
 #include <emmintrin.h>
@@ -473,7 +474,7 @@ size_t numZerosInTail(const UInt8 * begin, const UInt8 * end)
             count += 64;
         else
         {
-            count += __builtin_clzll(val);
+            count += std::countl_zero(val);
             return count;
         }
     }
@@ -507,7 +508,7 @@ size_t numZerosInTail(const UInt8 * begin, const UInt8 * end)
             count += 64;
         else
         {
-            count += __builtin_clzll(val);
+            count += std::countl_zero(val);
             return count;
         }
     }
@@ -531,7 +532,7 @@ size_t MergeTreeRangeReader::ReadResult::numZerosInTail(const UInt8 * begin, con
 
     size_t count = 0;
 
-#if defined(__SSE2__) && defined(__POPCNT__)
+#if defined(__SSE2__)
     const __m128i zero16 = _mm_setzero_si128();
     while (end - begin >= 64)
     {
@@ -555,7 +556,7 @@ size_t MergeTreeRangeReader::ReadResult::numZerosInTail(const UInt8 * begin, con
             count += 64;
         else
         {
-            count += __builtin_clzll(val);
+            count += std::countl_zero(val);
             return count;
         }
     }
@@ -583,7 +584,7 @@ size_t MergeTreeRangeReader::ReadResult::numZerosInTail(const UInt8 * begin, con
             count += 64;
         else
         {
-            count += __builtin_clzll(val);
+            count += std::countl_zero(val);
             return count;
         }
     }