Merge branch 'master' into mvcc_prototype

.github/PULL_REQUEST_TEMPLATE.md

@@ -1,4 +1,4 @@
-Changelog category (leave one):
+### Changelog category (leave one):
 - New Feature
 - Improvement
 - Bug Fix (user-visible misbehaviour in official stable or prestable release)
@@ -9,7 +9,7 @@ Changelog category (leave one):
 - Not for changelog (changelog entry is not required)
 
 
-Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
+### Changelog entry (a user-readable short description of the changes that goes to CHANGELOG.md):
 ...
 
 

contrib/unixodbc

@@ -1 +1 @@
-Subproject commit b0ad30f7f6289c12b76f04bfb9d466374bb32168
+Subproject commit a2cd5395e8c7f7390025ec93af5bfebef3fb5fcd

docker/keeper/Dockerfile

@@ -20,6 +20,8 @@ ENV LANG=en_US.UTF-8 \
 COPY --from=glibc-donor /lib/linux-gnu/libc.so.6 /lib/linux-gnu/libdl.so.2 /lib/linux-gnu/libm.so.6 /lib/linux-gnu/libpthread.so.0 /lib/linux-gnu/librt.so.1 /lib/linux-gnu/libnss_dns.so.2 /lib/linux-gnu/libnss_files.so.2 /lib/linux-gnu/libresolv.so.2 /lib/linux-gnu/ld-2.31.so /lib/
 COPY --from=glibc-donor /etc/nsswitch.conf /etc/
 COPY entrypoint.sh /entrypoint.sh
+
+ARG TARGETARCH
 RUN arch=${TARGETARCH:-amd64} \
     && case $arch in \
         amd64) mkdir -p /lib64 && ln -sf /lib/ld-2.31.so /lib64/ld-linux-x86-64.so.2 ;; \

programs/client/Client.cpp

@@ -810,7 +810,7 @@ void Client::addOptions(OptionsDescription & options_description)
         ("quota_key", po::value<std::string>(), "A string to differentiate quotas when the user have keyed quotas configured on server")
 
         ("max_client_network_bandwidth", po::value<int>(), "the maximum speed of data exchange over the network for the client in bytes per second.")
-        ("compression", po::value<bool>(), "enable or disable compression")
+        ("compression", po::value<bool>(), "enable or disable compression (enabled by default for remote communication and disabled for localhost communication).")
 
         ("query-fuzzer-runs", po::value<int>()->default_value(0), "After executing every SELECT query, do random mutations in it and run again specified number of times. This is used for testing to discover unexpected corner cases.")
         ("interleave-queries-file", po::value<std::vector<std::string>>()->multitoken(),

src/CMakeLists.txt

@@ -49,6 +49,18 @@ if (COMPILER_GCC)
     add_definitions ("-fno-tree-loop-distribute-patterns")
 endif ()
 
+# ClickHouse developers may use platform-dependent code under some macro (e.g. `#ifdef ENABLE_MULTITARGET`).
+# If turned ON, this option defines such macro.
+# See `src/Common/TargetSpecific.h`
+option(ENABLE_MULTITARGET_CODE "Enable platform-dependent code" ON)
+
+if (ENABLE_MULTITARGET_CODE)
+    add_definitions(-DENABLE_MULTITARGET_CODE=1)
+else()
+    add_definitions(-DENABLE_MULTITARGET_CODE=0)
+endif()
+
+
 add_subdirectory (Access)
 add_subdirectory (Backups)
 add_subdirectory (Columns)

src/Common/TargetSpecific.cpp

@@ -1,4 +1,4 @@
-#include <Functions/TargetSpecific.h>
+#include <Common/TargetSpecific.h>
 
 #include <Common/CpuId.h>
 

src/Disks/IO/CachedReadBufferFromRemoteFS.cpp

@@ -334,15 +334,17 @@ SeekableReadBufferPtr CachedReadBufferFromRemoteFS::getImplementationBuffer(File
                 read_buffer_for_file_segment->seek(file_offset_of_buffer_end, SEEK_SET);
             }
 
-            auto impl_range = read_buffer_for_file_segment->getRemainingReadRange();
             auto download_offset = file_segment->getDownloadOffset();
             if (download_offset != static_cast<size_t>(read_buffer_for_file_segment->getPosition()))
+            {
+                auto impl_range = read_buffer_for_file_segment->getRemainingReadRange();
                 throw Exception(
                     ErrorCodes::LOGICAL_ERROR,
                     "Buffer's offsets mismatch; cached buffer offset: {}, download_offset: {}, position: {}, implementation buffer offset: {}, "
                     "implementation buffer reading until: {}, file segment info: {}",
                     file_offset_of_buffer_end, download_offset, read_buffer_for_file_segment->getPosition(),
                     impl_range.left, *impl_range.right, file_segment->getInfoForLog());
+            }
 
             break;
         }
@@ -802,12 +804,14 @@ std::optional<size_t> CachedReadBufferFromRemoteFS::getLastNonDownloadedOffset()
 
 String CachedReadBufferFromRemoteFS::getInfoForLog()
 {
-    auto implementation_buffer_read_range_str =
+    String implementation_buffer_read_range_str;
-        implementation_buffer ?
+    if (implementation_buffer)
-        std::to_string(implementation_buffer->getRemainingReadRange().left)
+    {
-        + '-'
+        auto read_range = implementation_buffer->getRemainingReadRange();
-        + (implementation_buffer->getRemainingReadRange().right ? std::to_string(*implementation_buffer->getRemainingReadRange().right) : "None")
+        implementation_buffer_read_range_str = std::to_string(read_range.left) + '-' + (read_range.right ? std::to_string(*read_range.right) : "None");
-        : "None";
+    }
+    else
+        implementation_buffer_read_range_str = "None";
 
     auto current_file_segment_info = current_file_segment_it == file_segments_holder->file_segments.end() ? "None" : (*current_file_segment_it)->getInfoForLog();
 

src/Functions/CMakeLists.txt

@@ -96,17 +96,6 @@ if (TARGET ch_contrib::rapidjson)
     target_link_libraries(clickhouse_functions PRIVATE ch_contrib::rapidjson)
 endif()
 
-# ClickHouse developers may use platform-dependent code under some macro (e.g. `#ifdef ENABLE_MULTITARGET`).
-# If turned ON, this option defines such macro.
-# See `src/Functions/TargetSpecific.h`
-option(ENABLE_MULTITARGET_CODE "Enable platform-dependent code" ON)
-
-if (ENABLE_MULTITARGET_CODE)
-    add_definitions(-DENABLE_MULTITARGET_CODE=1)
-else()
-    add_definitions(-DENABLE_MULTITARGET_CODE=0)
-endif()
-
 add_subdirectory(GatherUtils)
 target_link_libraries(clickhouse_functions PRIVATE clickhouse_functions_gatherutils)
 

src/Functions/FunctionStartsEndsWith.h

@@ -1,12 +1,12 @@
 #pragma once
 #include <base/map.h>
 
+#include <Common/TargetSpecific.h>
 #include <Functions/FunctionHelpers.h>
 #include <Functions/GatherUtils/GatherUtils.h>
 #include <Functions/GatherUtils/Sources.h>
 #include <Functions/IFunction.h>
 #include <Functions/PerformanceAdaptors.h>
-#include <Functions/TargetSpecific.h>
 #include <DataTypes/DataTypeString.h>
 #include <DataTypes/DataTypesNumber.h>
 #include <DataTypes/getLeastSupertype.h>

src/Functions/FunctionsHashing.h

@@ -38,8 +38,8 @@
 #include <Columns/ColumnTuple.h>
 #include <Functions/IFunction.h>
 #include <Functions/FunctionHelpers.h>
-#include <Functions/TargetSpecific.h>
 #include <Functions/PerformanceAdaptors.h>
+#include <Common/TargetSpecific.h>
 #include <base/range.h>
 #include <base/bit_cast.h>
 

src/Functions/FunctionsRandom.h

@@ -1,9 +1,9 @@
 #pragma once
 
+#include <Common/TargetSpecific.h>
 #include <DataTypes/DataTypesNumber.h>
 #include <Columns/ColumnVector.h>
 #include <Functions/IFunction.h>
-#include <Functions/TargetSpecific.h>
 #include <Functions/PerformanceAdaptors.h>
 #include <IO/WriteHelpers.h>
 

src/Functions/GatherUtils/Algorithms.h

@@ -7,6 +7,8 @@
 #include <Core/AccurateComparison.h>
 #include <base/range.h>
 #include "GatherUtils.h"
+#include "sliceEqualElements.h"
+#include "sliceHasImplAnyAll.h"
 
 
 namespace DB::ErrorCodes
@@ -461,39 +463,19 @@ void NO_INLINE conditional(SourceA && src_a, SourceB && src_b, Sink && sink, con
 }
 
 
-/// Methods to check if first array has elements from second array, overloaded for various combinations of types.
+template <typename T>
-template <
+bool insliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
-    ArraySearchType search_type,
+                          size_t first_ind [[maybe_unused]],
-    typename FirstSliceType,
+                          size_t second_ind [[maybe_unused]])
-    typename SecondSliceType,
-          bool (*isEqual)(const FirstSliceType &, const SecondSliceType &, size_t, size_t)>
-bool sliceHasImplAnyAll(const FirstSliceType & first, const SecondSliceType & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
 {
-    const bool has_first_null_map = first_null_map != nullptr;
+    if constexpr (is_decimal<T>)
-    const bool has_second_null_map = second_null_map != nullptr;
+        return accurate::equalsOp(first.data[first_ind].value, first.data[second_ind].value);
-
+    else
-    for (size_t i = 0; i < second.size; ++i)
+        return accurate::equalsOp(first.data[first_ind], first.data[second_ind]);
-    {
+}
-        bool has = false;
+inline ALWAYS_INLINE bool insliceEqualElements(const GenericArraySlice & first, size_t first_ind, size_t second_ind)
-        for (size_t j = 0; j < first.size && !has; ++j)
+{
-        {
+    return first.elements->compareAt(first_ind + first.begin, second_ind + first.begin, *first.elements, -1) == 0;
-            const bool is_first_null = has_first_null_map && first_null_map[j];
-            const bool is_second_null = has_second_null_map && second_null_map[i];
-
-            if (is_first_null && is_second_null)
-                has = true;
-
-            if (!is_first_null && !is_second_null && isEqual(first, second, j, i))
-                has = true;
-        }
-
-        if (has && search_type == ArraySearchType::Any)
-            return true;
-
-        if (!has && search_type == ArraySearchType::All)
-            return false;
-    }
-    return search_type == ArraySearchType::All;
 }
 
 template <
@@ -620,55 +602,6 @@ bool sliceHasImpl(const FirstSliceType & first, const SecondSliceType & second,
         return sliceHasImplAnyAll<search_type, FirstSliceType, SecondSliceType, isEqual>(first, second, first_null_map, second_null_map);
 }
 
-
-template <typename T, typename U>
-bool sliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
-                        const NumericArraySlice<U> & second [[maybe_unused]],
-                        size_t first_ind [[maybe_unused]],
-                        size_t second_ind [[maybe_unused]])
-{
-    /// TODO: Decimal scale
-    if constexpr (is_decimal<T> && is_decimal<U>)
-        return accurate::equalsOp(first.data[first_ind].value, second.data[second_ind].value);
-    else if constexpr (is_decimal<T> || is_decimal<U>)
-        return false;
-    else
-        return accurate::equalsOp(first.data[first_ind], second.data[second_ind]);
-}
-
-template <typename T>
-bool sliceEqualElements(const NumericArraySlice<T> &, const GenericArraySlice &, size_t, size_t)
-{
-    return false;
-}
-
-template <typename U>
-bool sliceEqualElements(const GenericArraySlice &, const NumericArraySlice<U> &, size_t, size_t)
-{
-    return false;
-}
-
-inline ALWAYS_INLINE bool sliceEqualElements(const GenericArraySlice & first, const GenericArraySlice & second, size_t first_ind, size_t second_ind)
-{
-    return first.elements->compareAt(first_ind + first.begin, second_ind + second.begin, *second.elements, -1) == 0;
-}
-
-template <typename T>
-bool insliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
-                          size_t first_ind [[maybe_unused]],
-                          size_t second_ind [[maybe_unused]])
-{
-    if constexpr (is_decimal<T>)
-        return accurate::equalsOp(first.data[first_ind].value, first.data[second_ind].value);
-    else
-        return accurate::equalsOp(first.data[first_ind], first.data[second_ind]);
-}
-
-inline ALWAYS_INLINE bool insliceEqualElements(const GenericArraySlice & first, size_t first_ind, size_t second_ind)
-{
-    return first.elements->compareAt(first_ind + first.begin, second_ind + first.begin, *first.elements, -1) == 0;
-}
-
 template <ArraySearchType search_type, typename T, typename U>
 bool sliceHas(const NumericArraySlice<T> & first, const NumericArraySlice<U> & second)
 {
@@ -854,4 +787,3 @@ void resizeConstantSize(ArraySource && array_source, ValueSource && value_source
 }
 
 }
-

src/Functions/GatherUtils/CMakeLists.txt

@@ -1,4 +1,5 @@
 include("${ClickHouse_SOURCE_DIR}/cmake/dbms_glob_sources.cmake")
+
 add_headers_and_sources(clickhouse_functions_gatherutils .)
 add_library(clickhouse_functions_gatherutils ${clickhouse_functions_gatherutils_sources} ${clickhouse_functions_gatherutils_headers})
 target_link_libraries(clickhouse_functions_gatherutils PRIVATE dbms)
@@ -14,3 +15,5 @@ endif()
 if (STRIP_DEBUG_SYMBOLS_FUNCTIONS)
     target_compile_options(clickhouse_functions_gatherutils PRIVATE "-g0")
 endif()
+
+set_target_properties(clickhouse_functions_gatherutils PROPERTIES COMPILE_FLAGS "${X86_INTRINSICS_FLAGS}")

src/Functions/GatherUtils/sliceEqualElements.h

@@ -0,0 +1,41 @@
+#pragma once
+
+#include <Core/AccurateComparison.h>
+#include "Slices.h"
+
+namespace DB::GatherUtils
+{
+
+template <typename T, typename U>
+bool sliceEqualElements(const NumericArraySlice<T> & first [[maybe_unused]],
+                        const NumericArraySlice<U> & second [[maybe_unused]],
+                        size_t first_ind [[maybe_unused]],
+                        size_t second_ind [[maybe_unused]])
+{
+    /// TODO: Decimal scale
+    if constexpr (is_decimal<T> && is_decimal<U>)
+        return accurate::equalsOp(first.data[first_ind].value, second.data[second_ind].value);
+    else if constexpr (is_decimal<T> || is_decimal<U>)
+        return false;
+    else
+        return accurate::equalsOp(first.data[first_ind], second.data[second_ind]);
+}
+
+template <typename T>
+bool sliceEqualElements(const NumericArraySlice<T> &, const GenericArraySlice &, size_t, size_t)
+{
+    return false;
+}
+
+template <typename U>
+bool sliceEqualElements(const GenericArraySlice &, const NumericArraySlice<U> &, size_t, size_t)
+{
+    return false;
+}
+
+inline ALWAYS_INLINE bool sliceEqualElements(const GenericArraySlice & first, const GenericArraySlice & second, size_t first_ind, size_t second_ind)
+{
+    return first.elements->compareAt(first_ind + first.begin, second_ind + second.begin, *second.elements, -1) == 0;
+}
+
+}

src/Functions/GatherUtils/sliceHasImplAnyAll.h

@@ -0,0 +1,943 @@
+#pragma once
+
+#include "GatherUtils.h"
+#include "Slices.h"
+#include "sliceEqualElements.h"
+
+#if defined(__SSE4_2__)
+    #include <emmintrin.h>
+    #include <smmintrin.h>
+    #include <nmmintrin.h>
+#endif
+
+#if defined(__AVX2__)
+    #include <immintrin.h>
+#endif
+
+#include <Common/TargetSpecific.h>
+
+
+namespace DB::GatherUtils
+{
+
+inline ALWAYS_INLINE bool hasNull(const UInt8 * null_map, size_t null_map_size)
+{
+    if (null_map == nullptr)
+        return false;
+
+    for (size_t i = 0; i < null_map_size; ++i)
+    {
+        if (null_map[i])
+            return true;
+    }
+
+    return false;
+}
+
+template<class T>
+inline ALWAYS_INLINE bool hasAllIntegralLoopRemainder(
+    size_t j, const NumericArraySlice<T> & first, const NumericArraySlice<T> & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
+{
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    for (; j < second.size; ++j)
+    {
+        // skip null elements since both have at least one - assuming it was checked earlier that at least one element in 'first' is null
+        if (has_second_null_map && second_null_map[j])
+            continue;
+
+        bool found = false;
+
+        for (size_t i = 0; i < first.size; ++i)
+        {
+            if (has_first_null_map && first_null_map[i])
+                continue;
+
+            if (first.data[i] == second.data[j])
+            {
+                found = true;
+                break;
+            }
+        }
+
+        if (!found)
+            return false;
+    }
+    return true;
+}
+
+
+#if defined(__AVX2__)
+
+DECLARE_AVX2_SPECIFIC_CODE (
+
+// AVX2 Int64, UInt64 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int64> || std::is_same_v<IntType, UInt64>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt64(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr Int64 full = -1, none = 0;
+    const __m256i ones = _mm256_set1_epi64x(full);
+    const __m256i zeros = _mm256_setzero_si256();
+
+    if (second.size > 3 && first.size > 3)
+    {
+        for (; j < second.size - 3 && has_mask; j += 4)
+        {
+            has_mask = 0;
+            const __m256i second_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(second.data + j));
+            // bits of the bitmask are set to one if considered as null in the corresponding null map, 0 otherwise;
+            __m256i bitmask = has_second_null_map ?
+                _mm256_set_epi64x(
+                    (second_null_map[j + 3])? full : none,
+                    (second_null_map[j + 2])? full : none,
+                    (second_null_map[j + 1])? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 3 && !has_mask; has_mask = _mm256_testc_si256(bitmask, ones), i += 4)
+            {
+                const __m256i first_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(first.data + i));
+                const __m256i first_nm_mask = has_first_null_map?
+                    _mm256_set_m128i(
+                        _mm_cvtepi8_epi64(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i + 2))),
+                        _mm_cvtepi8_epi64(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))))
+                    : zeros;
+                bitmask =
+                    _mm256_or_si256(
+                        _mm256_or_si256(
+                            _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        first_nm_mask,
+                                        _mm256_cmpeq_epi64(second_data, first_data)),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(5,4,3,2,1,0,7,6)),
+                                        _mm256_cmpeq_epi64(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(5,4,3,2,1,0,7,6))))),
+
+                            _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(3,2,1,0,7,6,5,4)),
+                                        _mm256_cmpeq_epi64(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(3,2,1,0,7,6,5,4)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(1,0,7,6,5,4,3,2)),
+                                        _mm256_cmpeq_epi64(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(1,0,7,6,5,4,3,2)))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m256i v_i = _mm256_set1_epi64x(first.data[i]);
+                    bitmask = _mm256_or_si256(bitmask, _mm256_cmpeq_epi64(second_data, v_i));
+                    has_mask = _mm256_testc_si256(bitmask, ones);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 3)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// AVX2 Int32, UInt32 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int32> || std::is_same_v<IntType, UInt32>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt32(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int full = -1, none = 0;
+
+    const __m256i ones = _mm256_set1_epi32(full);
+    const __m256i zeros = _mm256_setzero_si256();
+
+    if (second.size > 7 && first.size > 7)
+    {
+        for (; j < second.size - 7 && has_mask; j += 8)
+        {
+            has_mask = 0;
+            const __m256i second_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(second.data + j));
+            // bits of the bitmask are set to one if considered as null in the corresponding null map, 0 otherwise;
+            __m256i bitmask = has_second_null_map ?
+                _mm256_set_epi32(
+                    (second_null_map[j + 7]) ? full : none,
+                    (second_null_map[j + 6]) ? full : none,
+                    (second_null_map[j + 5]) ? full : none,
+                    (second_null_map[j + 4]) ? full : none,
+                    (second_null_map[j + 3]) ? full : none,
+                    (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 7 && !has_mask; has_mask = _mm256_testc_si256(bitmask, ones), i += 8)
+            {
+                const __m256i first_data = _mm256_loadu_si256(reinterpret_cast<const __m256i *>(first.data + i));
+                // Create a mask to avoid to compare null elements
+                // set_m128i takes two arguments: (high segment, low segment) that are two __m128i convert from 8bits to 32bits to match with next operations
+                const __m256i first_nm_mask = has_first_null_map?
+                    _mm256_set_m128i(
+                        _mm_cvtepi8_epi32(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i + 4))),
+                        _mm_cvtepi8_epi32(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))))
+                    : zeros;
+                bitmask =
+                    _mm256_or_si256(
+                        _mm256_or_si256(
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        first_nm_mask,
+                                        _mm256_cmpeq_epi32(second_data, first_data)),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(6,5,4,3,2,1,0,7)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(6,5,4,3,2,1,0,7))))),
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(5,4,3,2,1,0,7,6)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(5,4,3,2,1,0,7,6)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(4,3,2,1,0,7,6,5)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(4,3,2,1,0,7,6,5)))))
+                            ),
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(3,2,1,0,7,6,5,4)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(3,2,1,0,7,6,5,4)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(2,1,0,7,6,5,4,3)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(2,1,0,7,6,5,4,3))))),
+                                _mm256_or_si256(
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(1,0,7,6,5,4,3,2)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(1,0,7,6,5,4,3,2)))),
+                                    _mm256_andnot_si256(
+                                        _mm256_permutevar8x32_epi32(first_nm_mask, _mm256_set_epi32(0,7,6,5,4,3,2,1)),
+                                        _mm256_cmpeq_epi32(second_data, _mm256_permutevar8x32_epi32(first_data, _mm256_set_epi32(0,7,6,5,4,3,2,1))))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m256i v_i = _mm256_set1_epi32(first.data[i]);
+                    bitmask = _mm256_or_si256(bitmask, _mm256_cmpeq_epi32(second_data, v_i));
+                    has_mask = _mm256_testc_si256(bitmask, ones);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 7)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// AVX2 Int16, UInt16 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int16> || std::is_same_v<IntType, UInt16>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt16(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int16_t full = -1, none = 0;
+    const __m256i ones = _mm256_set1_epi16(full);
+    const __m256i zeros = _mm256_setzero_si256();
+    if (second.size > 15 && first.size > 15)
+    {
+        for (; j < second.size - 15 && has_mask; j += 16)
+        {
+            has_mask = 0;
+            const __m256i second_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(second.data + j));
+            __m256i bitmask = has_second_null_map ?
+                _mm256_set_epi16(
+                    (second_null_map[j + 15]) ? full : none, (second_null_map[j + 14]) ? full : none,
+                    (second_null_map[j + 13]) ? full : none, (second_null_map[j + 12]) ? full : none,
+                    (second_null_map[j + 11]) ? full : none, (second_null_map[j + 10]) ? full : none,
+                    (second_null_map[j + 9]) ? full : none, (second_null_map[j + 8])? full : none,
+                    (second_null_map[j + 7]) ? full : none, (second_null_map[j + 6])? full : none,
+                    (second_null_map[j + 5]) ? full : none, (second_null_map[j + 4])? full : none,
+                    (second_null_map[j + 3]) ? full : none, (second_null_map[j + 2])? full : none,
+                    (second_null_map[j + 1]) ? full : none, (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 15 && !has_mask; has_mask = _mm256_testc_si256(bitmask, ones), i += 16)
+            {
+                const __m256i first_data = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(first.data + i));
+                const __m256i first_nm_mask = has_first_null_map?
+                    _mm256_set_m128i(
+                        _mm_cvtepi8_epi16(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i + 8))),
+                        _mm_cvtepi8_epi16(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))))
+                    : zeros;
+
+                bitmask =
+                    _mm256_or_si256(
+                        _mm256_or_si256(
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            first_nm_mask,
+                                            _mm256_cmpeq_epi16(second_data, first_data)),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26)))))
+                                ),
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(first_nm_mask, _mm256_set_epi8(17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(first_data, _mm256_set_epi8(17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18))))))
+                            ),
+                            _mm256_or_si256(
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_permute2x128_si256(first_nm_mask, first_nm_mask,1),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_permute2x128_si256(first_data, first_data, 1))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(9,8,7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10)))))
+                                ),
+                                _mm256_or_si256(
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data ,first_data, 1), _mm256_set_epi8(7,6,5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(5,4,3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6))))),
+                                    _mm256_or_si256(
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data ,first_data ,1), _mm256_set_epi8(3,2,1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4)))),
+                                        _mm256_andnot_si256(
+                                            _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_nm_mask, first_nm_mask, 1), _mm256_set_epi8(1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2)),
+                                            _mm256_cmpeq_epi16(second_data, _mm256_shuffle_epi8(_mm256_permute2x128_si256(first_data, first_data, 1), _mm256_set_epi8(1,0,31,30,29,28,27,26,25,24,23,22,21,20,19,18,17,16,15,14,13,12,11,10,9,8,7,6,5,4,3,2))))))
+                        )
+                    ),
+                    bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m256i v_i = _mm256_set1_epi16(first.data[i]);
+                    bitmask = _mm256_or_si256(bitmask, _mm256_cmpeq_epi16(second_data, v_i));
+                    has_mask = _mm256_testc_si256(bitmask, ones);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 15)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+)
+
+#endif
+
+#if defined(__SSE4_2__)
+
+DECLARE_SSE42_SPECIFIC_CODE (
+
+// SSE4.2 Int64, UInt64 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int64> || std::is_same_v<IntType, UInt64>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt64(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr Int64 full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+    if (second.size > 1 && first.size > 1)
+    {
+        for (; j < second.size - 1 && has_mask; j += 2)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi64x(
+                    (second_null_map[j + 1]) ? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+
+            for (; i < first.size - 1 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 2)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_cvtepi8_epi64(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i)))
+                    : zeros;
+
+                bitmask =
+                    _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_andnot_si128(
+                                    first_nm_mask,
+                                    _mm_cmpeq_epi64(second_data, first_data)),
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(1,0,3,2)),
+                                    _mm_cmpeq_epi64(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(1,0,3,2))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m128i v_i = _mm_set1_epi64x(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi64(second_data, v_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 1)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// SSE4.2 Int32, UInt32 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int32> || std::is_same_v<IntType, UInt32>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt32(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+    if (second.size > 3 && first.size > 3)
+    {
+        for (; j < second.size - 3 && has_mask; j += 4)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi32(
+                    (second_null_map[j + 3]) ? full : none,
+                    (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none,
+                    (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 3 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 4)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_cvtepi8_epi32(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i)))
+                    : zeros;
+
+                bitmask =
+                    _mm_or_si128(
+                        _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_andnot_si128(
+                                        first_nm_mask,
+                                        _mm_cmpeq_epi32(second_data, first_data)),
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(2,1,0,3)),
+                                    _mm_cmpeq_epi32(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(2,1,0,3))))),
+                            _mm_or_si128(
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(1,0,3,2)),
+                                    _mm_cmpeq_epi32(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(1,0,3,2)))),
+                                _mm_andnot_si128(
+                                    _mm_shuffle_epi32(first_nm_mask, _MM_SHUFFLE(0,3,2,1)),
+                                    _mm_cmpeq_epi32(second_data, _mm_shuffle_epi32(first_data, _MM_SHUFFLE(0,3,2,1)))))
+                        ),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+                    __m128i r_i = _mm_set1_epi32(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi32(second_data, r_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 3)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// SSE4.2 Int16, UInt16 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int16> || std::is_same_v<IntType, UInt16>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt16(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map  != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int16_t full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+    if (second.size > 6 && first.size > 6)
+    {
+        for (; j < second.size - 7 && has_mask; j += 8)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi16(
+                    (second_null_map[j + 7]) ? full : none, (second_null_map[j + 6]) ? full : none,
+                    (second_null_map[j + 5]) ? full : none, (second_null_map[j + 4]) ? full : none,
+                    (second_null_map[j + 3]) ? full : none, (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none, (second_null_map[j]) ? full: none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size-7 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 8)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_cvtepi8_epi16(_mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i)))
+                    : zeros;
+                bitmask =
+                    _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            first_nm_mask,
+                                            _mm_cmpeq_epi16(second_data, first_data)),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)))))
+                                ),
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6)),
+                                        _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2)),
+                                            _mm_cmpeq_epi16(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2))))))
+                        ),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+                    __m128i v_i = _mm_set1_epi16(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi16(second_data, v_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 6)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+// Int8/UInt8 version is faster with SSE than with AVX2
+// SSE2 Int8, UInt8 specialization
+template<typename IntType>
+requires (std::is_same_v<IntType, Int8> || std::is_same_v<IntType, UInt8>)
+inline ALWAYS_INLINE bool sliceHasImplAnyAllImplInt8(
+    const NumericArraySlice<IntType> & first,
+    const NumericArraySlice<IntType> & second,
+    const UInt8 * first_null_map,
+    const UInt8 * second_null_map)
+{
+    if (second.size == 0)
+        return true;
+
+    if (!hasNull(first_null_map, first.size) && hasNull(second_null_map, second.size))
+        return false;
+
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    size_t j = 0;
+    int has_mask = 1;
+    static constexpr int8_t full = -1, none = 0;
+    const __m128i zeros = _mm_setzero_si128();
+
+    if (second.size > 15 && first.size > 15)
+    {
+        for (; j < second.size - 15 && has_mask; j += 16)
+        {
+            has_mask = 0;
+            const __m128i second_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(second.data + j));
+            __m128i bitmask = has_second_null_map ?
+                _mm_set_epi8(
+                    (second_null_map[j + 15]) ? full : none, (second_null_map[j + 14]) ? full : none,
+                    (second_null_map[j + 13]) ? full : none, (second_null_map[j + 12]) ? full : none,
+                    (second_null_map[j + 11]) ? full : none, (second_null_map[j + 10]) ? full : none,
+                    (second_null_map[j + 9]) ? full : none, (second_null_map[j + 8]) ? full : none,
+                    (second_null_map[j + 7]) ? full : none, (second_null_map[j + 6]) ? full : none,
+                    (second_null_map[j + 5]) ? full : none, (second_null_map[j + 4]) ? full : none,
+                    (second_null_map[j + 3]) ? full : none, (second_null_map[j + 2]) ? full : none,
+                    (second_null_map[j + 1]) ? full : none, (second_null_map[j]) ? full : none)
+                : zeros;
+
+            size_t i = 0;
+            for (; i < first.size - 15 && !has_mask; has_mask = _mm_test_all_ones(bitmask), i += 16)
+            {
+                const __m128i first_data = _mm_loadu_si128(reinterpret_cast<const __m128i *>(first.data + i));
+                const __m128i first_nm_mask = has_first_null_map ?
+                    _mm_loadu_si128(reinterpret_cast<const __m128i *>(first_null_map + i))
+                    : zeros;
+                bitmask =
+                    _mm_or_si128(
+                        _mm_or_si128(
+                            _mm_or_si128(
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            first_nm_mask,
+                                            _mm_cmpeq_epi8(second_data, first_data)),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,15)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(14,13,12,11,10,9,8,7,6,5,4,3,2,1,0,15))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(13,12,11,10,9,8,7,6,5,4,3,2,1,0,15,14)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(12,11,10,9,8,7,6,5,4,3,2,1,0,15,14,13)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(12,11,10,9,8,7,6,5,4,3,2,1,0,15,14,13)))))
+                                ),
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(11,10,9,8,7,6,5,4,3,2,1,0,15,14,13,12)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(10,9,8,7,6,5,4,3,2,1,0,15,14,13,12,11)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(10,9,8,7,6,5,4,3,2,1,0,15,14,13,12,11))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(9,8,7,6,5,4,3,2,1,0,15,14,13,12,11,10)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(8,7,6,5,4,3,2,1,0,15,14,13,12,11,10,9)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(8,7,6,5,4,3,2,1,0,15,14,13,12,11,10,9))))))),
+                            _mm_or_si128(
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(7,6,5,4,3,2,1,0,15,14,13,12,11,10,9,8)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,7)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(6,5,4,3,2,1,0,15,14,13,12,11,10,9,8,7))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(5,4,3,2,1,0,15,14,13,12,11,10,9,8,7,6)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(4,3,2,1,0,15,14,13,12,11,10,9,8,7,6,5)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(4,3,2,1,0,15,14,13,12,11,10,9,8,7,6,5)))))),
+                                _mm_or_si128(
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(3,2,1,0,15,14,13,12,11,10,9,8,7,6,5,4)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(2,1,0,15,14,13,12,11,10,9,8,7,6,5,4,3)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(2,1,0,15,14,13,12,11,10,9,8,7,6,5,4,3))))),
+                                    _mm_or_si128(
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(1,0,15,14,13,12,11,10,9,8,7,6,5,4,3,2)))),
+                                        _mm_andnot_si128(
+                                            _mm_shuffle_epi8(first_nm_mask, _mm_set_epi8(0,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1)),
+                                            _mm_cmpeq_epi8(second_data, _mm_shuffle_epi8(first_data, _mm_set_epi8(0,15,14,13,12,11,10,9,8,7,6,5,4,3,2,1)))))))),
+                        bitmask);
+            }
+
+            if (i < first.size)
+            {
+                for (; i < first.size && !has_mask; ++i)
+                {
+                    if (has_first_null_map && first_null_map[i])
+                        continue;
+
+                    __m128i v_i = _mm_set1_epi8(first.data[i]);
+                    bitmask = _mm_or_si128(bitmask, _mm_cmpeq_epi8(second_data, v_i));
+                    has_mask = _mm_test_all_ones(bitmask);
+                }
+            }
+        }
+    }
+
+    if (!has_mask && second.size > 15)
+        return false;
+
+    return hasAllIntegralLoopRemainder(j, first, second, first_null_map, second_null_map);
+}
+
+)
+
+#endif
+
+template <
+    ArraySearchType search_type,
+    typename FirstSliceType,
+    typename SecondSliceType,
+    bool (*isEqual)(const FirstSliceType &, const SecondSliceType &, size_t, size_t)>
+bool sliceHasImplAnyAllGenericImpl(const FirstSliceType & first, const SecondSliceType & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
+{
+    const bool has_first_null_map = first_null_map != nullptr;
+    const bool has_second_null_map = second_null_map != nullptr;
+
+    const bool has_second_null = hasNull(second_null_map, second.size);
+    if (has_second_null)
+    {
+        const bool has_first_null = hasNull(first_null_map, first.size);
+
+        if (has_first_null && search_type == ArraySearchType::Any)
+            return true;
+
+        if (!has_first_null && search_type == ArraySearchType::All)
+            return false;
+    }
+
+    for (size_t i = 0; i < second.size; ++i)
+    {
+        if (has_second_null_map && second_null_map[i])
+            continue;
+
+        bool has = false;
+
+        for (size_t j = 0; j < first.size && !has; ++j)
+        {
+            if (has_first_null_map && first_null_map[j])
+                continue;
+
+            if (isEqual(first, second, j, i))
+            {
+                has = true;
+                break;
+            }
+        }
+
+        if (has && search_type == ArraySearchType::Any)
+            return true;
+
+        if (!has && search_type == ArraySearchType::All)
+            return false;
+    }
+
+    return search_type == ArraySearchType::All;
+}
+
+/// Methods to check if first array has elements from second array, overloaded for various combinations of types.
+template <
+    ArraySearchType search_type,
+    typename FirstSliceType,
+    typename SecondSliceType,
+    bool (*isEqual)(const FirstSliceType &, const SecondSliceType &, size_t, size_t)>
+inline ALWAYS_INLINE bool sliceHasImplAnyAll(const FirstSliceType & first, const SecondSliceType & second, const UInt8 * first_null_map, const UInt8 * second_null_map)
+{
+#if USE_MULTITARGET_CODE
+    if constexpr (search_type == ArraySearchType::All && std::is_same_v<FirstSliceType, SecondSliceType>)
+    {
+
+#if defined(__AVX2__)
+        if (isArchSupported(TargetArch::AVX2))
+        {
+            if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int16>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt16>>)
+            {
+                return GatherUtils::TargetSpecific::AVX2::sliceHasImplAnyAllImplInt16(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int32>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt32>>)
+            {
+                return GatherUtils::TargetSpecific::AVX2::sliceHasImplAnyAllImplInt32(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int64>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt64>>)
+            {
+                return GatherUtils::TargetSpecific::AVX2::sliceHasImplAnyAllImplInt64(first, second, first_null_map, second_null_map);
+            }
+        }
+#endif
+
+        if (isArchSupported(TargetArch::SSE42))
+        {
+            if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int8>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt8>>)
+            {
+                return TargetSpecific::SSE42::sliceHasImplAnyAllImplInt8(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int16>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt16>>)
+            {
+                return GatherUtils::TargetSpecific::SSE42::sliceHasImplAnyAllImplInt16(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int32>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt32>>)
+            {
+                return GatherUtils::TargetSpecific::SSE42::sliceHasImplAnyAllImplInt32(first, second, first_null_map, second_null_map);
+            }
+            else if constexpr (std::is_same_v<FirstSliceType, NumericArraySlice<Int64>> || std::is_same_v<FirstSliceType, NumericArraySlice<UInt64>>)
+            {
+                return GatherUtils::TargetSpecific::SSE42::sliceHasImplAnyAllImplInt64(first, second, first_null_map, second_null_map);
+            }
+        }
+    }
+#endif
+
+    return sliceHasImplAnyAllGenericImpl<search_type, FirstSliceType, SecondSliceType, isEqual>(first, second, first_null_map, second_null_map);
+}
+
+
+}

src/Functions/PerformanceAdaptors.h

@@ -1,8 +1,8 @@
 #pragma once
 
-#include <Functions/TargetSpecific.h>
 #include <Functions/IFunction.h>
 
+#include <Common/TargetSpecific.h>
 #include <Common/Stopwatch.h>
 #include <Interpreters/Context.h>
 

src/Functions/greatCircleDistance.cpp

@@ -6,8 +6,8 @@
 #include <Functions/IFunction.h>
 #include <Functions/FunctionHelpers.h>
 #include <Functions/FunctionFactory.h>
-#include <Functions/TargetSpecific.h>
 #include <Functions/PerformanceAdaptors.h>
+#include <Common/TargetSpecific.h>
 #include <base/range.h>
 #include <cmath>
 

src/Functions/tests/gtest_has_all.cpp

@@ -0,0 +1,139 @@
+#include <random>
+#include <gtest/gtest.h>
+#include <Functions/GatherUtils/Algorithms.h>
+
+using namespace DB::GatherUtils;
+
+
+auto uni_int_dist(int min, int max)
+{
+    std::random_device rd;
+    std::mt19937 mt(rd());
+    std::uniform_int_distribution<> dist(min, max);
+    return std::make_pair(dist, mt);
+}
+
+template<class T>
+void arrayInit(T* elements_to_have, size_t nb_elements_to_have, T* array_elements, size_t array_size, bool all_elements_present)
+{
+    for (size_t i = 0; i < array_size; ++i)
+    {
+        array_elements[i] = i;
+    }
+    auto [dist, gen] = uni_int_dist(0, array_size - 1);
+    for (size_t i = 0; i < nb_elements_to_have; ++i)
+    {
+        elements_to_have[i] = array_elements[dist(gen)];
+    }
+    if (!all_elements_present)
+    {
+        /// make one element to be searched for missing from the target array
+        elements_to_have[nb_elements_to_have - 1] = array_size + 1;
+    }
+}
+
+void nullMapInit(UInt8 * null_map, size_t null_map_size, size_t nb_null_elements)
+{
+    /// -2 to keep the last element of the array non-null
+    auto [dist, gen] = uni_int_dist(0, null_map_size - 2);
+    for (size_t i = 0; i < null_map_size; ++i)
+    {
+        null_map[i] = 0;
+    }
+    for (size_t i = 0; i < null_map_size - 1 && i < nb_null_elements; ++i)
+    {
+        null_map[dist(gen)] = 1;
+    }
+}
+
+template<class T>
+bool testHasAll(size_t nb_elements_to_have, size_t array_size, bool with_null_maps, bool all_elements_present)
+{
+    auto array_elements = std::make_unique<T[]>(array_size);
+    auto elements_to_have = std::make_unique<T[]>(nb_elements_to_have);
+
+    std::unique_ptr<UInt8[]> first_nm = nullptr, second_nm = nullptr;
+    if (with_null_maps)
+    {
+        first_nm = std::make_unique<UInt8[]>(array_size);
+        second_nm = std::make_unique<UInt8[]>(nb_elements_to_have);
+        /// add a null to elements to have, but not to the target array, making the answer negative
+        nullMapInit(first_nm.get(), array_size, 0);
+        nullMapInit(second_nm.get(), nb_elements_to_have, 1);
+    }
+
+    arrayInit(elements_to_have.get(), nb_elements_to_have, array_elements.get(), array_size, all_elements_present);
+
+    NumericArraySlice<T> first = {array_elements.get(), array_size};
+    NumericArraySlice<T> second = {elements_to_have.get(), nb_elements_to_have};
+
+    /// check whether all elements of the second array are also elements of the first array, overloaded for various combinations of types.
+    return sliceHasImplAnyAll<ArraySearchType::All, NumericArraySlice<T>, NumericArraySlice<T>, sliceEqualElements<T,T> >(
+        first, second, first_nm.get(), second_nm.get());
+}
+
+TEST(HasAll, integer)
+{
+    bool test1 = testHasAll<int>(4, 100, false, true);
+    bool test2 = testHasAll<int>(4, 100, false, false);
+    bool test3 = testHasAll<int>(100, 4096, false, true);
+    bool test4 = testHasAll<int>(100, 4096, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+
+TEST(HasAll, int64)
+{
+    bool test1 = testHasAll<int64_t>(2, 100, false, true);
+    bool test2 = testHasAll<int64_t>(2, 100, false, false);
+    bool test3 = testHasAll<int64_t>(100, 4096, false, true);
+    bool test4 = testHasAll<int64_t>(100, 4096, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+TEST(HasAll, int16)
+{
+    bool test1 = testHasAll<int16_t>(2, 100, false, true);
+    bool test2 = testHasAll<int16_t>(2, 100, false, false);
+    bool test3 = testHasAll<int16_t>(100, 4096, false, true);
+    bool test4 = testHasAll<int16_t>(100, 4096, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+TEST(HasAll, int8)
+{
+    bool test1 = testHasAll<int8_t>(2, 100, false, true);
+    bool test2 = testHasAll<int8_t>(2, 100, false, false);
+    bool test3 = testHasAll<int8_t>(50, 125, false, true);
+    bool test4 = testHasAll<int8_t>(50, 125, false, false);
+
+    ASSERT_EQ(test1, true);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, true);
+    ASSERT_EQ(test4, false);
+}
+
+TEST(HasAllSingleNullElement, all)
+{
+    bool test1 = testHasAll<int>(4, 100, true, true);
+    bool test2 = testHasAll<int64_t>(4, 100, true, true);
+    bool test3 = testHasAll<int16_t>(4, 100, true, true);
+    bool test4 = testHasAll<int8_t>(4, 100, true, true);
+
+    ASSERT_EQ(test1, false);
+    ASSERT_EQ(test2, false);
+    ASSERT_EQ(test3, false);
+    ASSERT_EQ(test4, false);
+}

src/IO/ReadBufferFromS3.cpp

@@ -240,7 +240,7 @@ void ReadBufferFromS3::setReadUntilPosition(size_t position)
 
 SeekableReadBuffer::Range ReadBufferFromS3::getRemainingReadRange() const
 {
-    return Range{.left = static_cast<size_t>(offset), .right = read_until_position ? std::optional{read_until_position - 1} : std::nullopt};
+    return Range{ .left = static_cast<size_t>(offset), .right = read_until_position ? std::optional{read_until_position - 1} : std::nullopt };
 }
 
 std::unique_ptr<ReadBuffer> ReadBufferFromS3::initialize()

src/IO/ReadBufferFromS3.h

@@ -33,8 +33,11 @@ private:
     String key;
     UInt64 max_single_read_retries;
 
-    off_t offset = 0;
+    /// These variables are atomic because they can be used for `logging only`
-    off_t read_until_position = 0;
+    /// (where it is not important to get consistent result)
+    /// from separate thread other than the one which uses the buffer for s3 reading.
+    std::atomic<off_t> offset = 0;
+    std::atomic<off_t> read_until_position = 0;
 
     Aws::S3::Model::GetObjectResult read_result;
     std::unique_ptr<ReadBuffer> impl;

src/Interpreters/ActionsDAG.cpp

@@ -1527,6 +1527,21 @@ ActionsDAG::SplitResult ActionsDAG::splitActionsBeforeArrayJoin(const NameSet &
     return res;
 }
 
+ActionsDAG::SplitResult ActionsDAG::splitActionsBySortingDescription(const NameSet & sort_columns) const
+{
+    std::unordered_set<const Node *> split_nodes;
+    for (const auto & sort_column : sort_columns)
+        if (const auto * node = tryFindInIndex(sort_column))
+            split_nodes.insert(node);
+        else
+            throw Exception(
+                ErrorCodes::LOGICAL_ERROR, "Sorting column {} wasn't found in the ActionsDAG's index. DAG:\n{}", sort_column, dumpDAG());
+
+    auto res = split(split_nodes);
+    res.second->project_input = project_input;
+    return res;
+}
+
 ActionsDAG::SplitResult ActionsDAG::splitActionsForFilter(const std::string & column_name) const
 {
     const auto * node = tryFindInIndex(column_name);

src/Interpreters/ActionsDAG.h

@@ -274,6 +274,10 @@ public:
     /// Index of initial actions must contain column_name.
     SplitResult splitActionsForFilter(const std::string & column_name) const;
 
+    /// Splits actions into two parts. The first part contains all the calculations required to calculate sort_columns.
+    /// The second contains the rest.
+    SplitResult splitActionsBySortingDescription(const NameSet & sort_columns) const;
+
     /// Create actions which may calculate part of filter using only available_inputs.
     /// If nothing may be calculated, returns nullptr.
     /// Otherwise, return actions which inputs are from available_inputs.

src/Interpreters/ExternalDictionariesLoader.cpp

@@ -136,7 +136,7 @@ std::string ExternalDictionariesLoader::resolveDictionaryNameFromDatabaseCatalog
 
     if (qualified_name->database.empty())
     {
-        /// Ether database name is not specified and we should use current one
+        /// Either database name is not specified and we should use current one
         /// or it's an XML dictionary.
         bool is_xml_dictionary = has(name);
         if (is_xml_dictionary)

src/Parsers/ASTPartition.h

@@ -15,6 +15,7 @@ public:
     size_t fields_count = 0;
 
     String id;
+    bool all = false;
 
     String getID(char) const override;
     ASTPtr clone() const override;

src/Parsers/ParserPartition.cpp

@@ -5,6 +5,7 @@
 #include <Parsers/ASTPartition.h>
 #include <Parsers/ASTLiteral.h>
 #include <Parsers/ASTFunction.h>
+#include <Parsers/ASTIdentifier.h>
 #include <Common/typeid_cast.h>
 
 namespace DB
@@ -13,6 +14,7 @@ namespace DB
 bool ParserPartition::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
 {
     ParserKeyword s_id("ID");
+    ParserKeyword s_all("ALL");
     ParserStringLiteral parser_string_literal;
     ParserExpression parser_expr;
 
@@ -28,6 +30,14 @@ bool ParserPartition::parseImpl(Pos & pos, ASTPtr & node, Expected & expected)
 
         partition->id = partition_id->as<ASTLiteral &>().value.get<String>();
     }
+    else if (s_all.ignore(pos, expected))
+    {
+        ASTPtr value = makeASTFunction("tuple");
+        partition->value = value;
+        partition->children.push_back(value);
+        partition->fields_count = 0;
+        partition->all = true;
+    }
     else
     {
         ASTPtr value;

src/Processors/QueryPlan/Optimizations/Optimizations.h

@@ -44,16 +44,20 @@ size_t tryMergeExpressions(QueryPlan::Node * parent_node, QueryPlan::Nodes &);
 /// May split FilterStep and push down only part of it.
 size_t tryPushDownFilter(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes);
 
+/// Move ExpressionStep after SortingStep if possible.
+/// May split ExpressionStep and lift up only a part of it.
+size_t tryExecuteFunctionsAfterSorting(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes);
+
 inline const auto & getOptimizations()
 {
-    static const std::array<Optimization, 5> optimizations =
+    static const std::array<Optimization, 6> optimizations = {{
-    {{
         {tryLiftUpArrayJoin, "liftUpArrayJoin", &QueryPlanOptimizationSettings::optimize_plan},
         {tryPushDownLimit, "pushDownLimit", &QueryPlanOptimizationSettings::optimize_plan},
         {trySplitFilter, "splitFilter", &QueryPlanOptimizationSettings::optimize_plan},
         {tryMergeExpressions, "mergeExpressions", &QueryPlanOptimizationSettings::optimize_plan},
         {tryPushDownFilter, "pushDownFilter", &QueryPlanOptimizationSettings::filter_push_down},
-     }};
+        {tryExecuteFunctionsAfterSorting, "liftUpFunctions", &QueryPlanOptimizationSettings::optimize_plan},
+    }};
 
     return optimizations;
 }

src/Processors/QueryPlan/Optimizations/liftUpFunctions.cpp

@@ -0,0 +1,77 @@
+#include <Interpreters/ActionsDAG.h>
+#include <Processors/QueryPlan/ExpressionStep.h>
+#include <Processors/QueryPlan/Optimizations/Optimizations.h>
+#include <Processors/QueryPlan/SortingStep.h>
+#include <Common/Exception.h>
+
+namespace DB
+{
+namespace ErrorCodes
+{
+    extern const int LOGICAL_ERROR;
+}
+}
+
+namespace
+{
+
+const DB::DataStream & getChildOutputStream(DB::QueryPlan::Node & node)
+{
+    if (node.children.size() != 1)
+        throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Node \"{}\" is expected to have only one child.", node.step->getName());
+    return node.children.front()->step->getOutputStream();
+}
+
+}
+
+namespace DB::QueryPlanOptimizations
+{
+
+size_t tryExecuteFunctionsAfterSorting(QueryPlan::Node * parent_node, QueryPlan::Nodes & nodes)
+{
+    if (parent_node->children.size() != 1)
+        return 0;
+
+    QueryPlan::Node * child_node = parent_node->children.front();
+
+    auto & parent_step = parent_node->step;
+    auto & child_step = child_node->step;
+    auto * sorting_step = typeid_cast<SortingStep *>(parent_step.get());
+    auto * expression_step = typeid_cast<ExpressionStep *>(child_step.get());
+
+    if (!sorting_step || !expression_step)
+        return 0;
+
+    NameSet sort_columns;
+    for (const auto & col : sorting_step->getSortDescription())
+        sort_columns.insert(col.column_name);
+    auto [needed_for_sorting, unneeded_for_sorting] = expression_step->getExpression()->splitActionsBySortingDescription(sort_columns);
+
+    // No calculations can be postponed.
+    if (unneeded_for_sorting->trivial())
+        return 0;
+
+    // Sorting (parent_node) -> Expression (child_node)
+    auto & node_with_needed = nodes.emplace_back();
+    std::swap(node_with_needed.children, child_node->children);
+    child_node->children = {&node_with_needed};
+
+    node_with_needed.step = std::make_unique<ExpressionStep>(getChildOutputStream(node_with_needed), std::move(needed_for_sorting));
+    node_with_needed.step->setStepDescription(child_step->getStepDescription());
+    // Sorting (parent_node) -> so far the origin Expression (child_node) -> NeededCalculations (node_with_needed)
+
+    std::swap(parent_step, child_step);
+    // so far the origin Expression (parent_node) -> Sorting (child_node) -> NeededCalculations (node_with_needed)
+
+    sorting_step->updateInputStream(getChildOutputStream(*child_node));
+    auto input_header = sorting_step->getInputStreams().at(0).header;
+    sorting_step->updateOutputStream(std::move(input_header));
+
+    auto description = parent_step->getStepDescription();
+    parent_step = std::make_unique<DB::ExpressionStep>(child_step->getOutputStream(), std::move(unneeded_for_sorting));
+    parent_step->setStepDescription(description + " [lifted up part]");
+    // UneededCalculations (parent_node) -> Sorting (child_node) -> NeededCalculations (node_with_needed)
+
+    return 3;
+}
+}

src/Processors/QueryPlan/SortingStep.cpp

@@ -1,11 +1,12 @@
+#include <stdexcept>
+#include <IO/Operators.h>
+#include <Processors/Merges/MergingSortedTransform.h>
 #include <Processors/QueryPlan/SortingStep.h>
-#include <QueryPipeline/QueryPipelineBuilder.h>
+#include <Processors/Transforms/FinishSortingTransform.h>
+#include <Processors/Transforms/LimitsCheckingTransform.h>
 #include <Processors/Transforms/MergeSortingTransform.h>
 #include <Processors/Transforms/PartialSortingTransform.h>
-#include <Processors/Transforms/FinishSortingTransform.h>
+#include <QueryPipeline/QueryPipelineBuilder.h>
-#include <Processors/Merges/MergingSortedTransform.h>
-#include <Processors/Transforms/LimitsCheckingTransform.h>
-#include <IO/Operators.h>
 #include <Common/JSONBuilder.h>
 
 namespace DB
@@ -88,6 +89,18 @@ SortingStep::SortingStep(
     output_stream->sort_mode = DataStream::SortMode::Stream;
 }
 
+void SortingStep::updateInputStream(DataStream input_stream)
+{
+    input_streams.clear();
+    input_streams.emplace_back(std::move(input_stream));
+}
+
+void SortingStep::updateOutputStream(Block result_header)
+{
+    output_stream = createOutputStream(input_streams.at(0), std::move(result_header), getDataStreamTraits());
+    updateDistinctColumns(output_stream->header, output_stream->distinct_columns);
+}
+
 void SortingStep::updateLimit(size_t limit_)
 {
     if (limit_ && (limit == 0 || limit_ < limit))

src/Processors/QueryPlan/SortingStep.h

@@ -49,6 +49,11 @@ public:
     /// Add limit or change it to lower value.
     void updateLimit(size_t limit_);
 
+    void updateInputStream(DataStream input_stream);
+    void updateOutputStream(Block result_header);
+
+    SortDescription getSortDescription() const { return result_description; }
+
 private:
 
     enum class Type

src/Storages/MergeTree/MergeTreeData.cpp

@@ -3110,7 +3110,8 @@ void MergeTreeData::tryRemovePartImmediately(DataPartPtr && part)
     {
         auto lock = lockParts();
 
-        LOG_TRACE(log, "Trying to immediately remove part {}", part->getNameWithState());
+        auto part_name_with_state = part->getNameWithState();
+        LOG_TRACE(log, "Trying to immediately remove part {}", part_name_with_state);
 
         if (part->getState() != DataPartState::Temporary)
         {
@@ -3121,7 +3122,16 @@ void MergeTreeData::tryRemovePartImmediately(DataPartPtr && part)
             part.reset();
 
             if (!((*it)->getState() == DataPartState::Outdated && it->unique()))
+            {
+                if ((*it)->getState() != DataPartState::Outdated)
+                    LOG_WARNING(log, "Cannot immediately remove part {} because it's not in Outdated state "
+                             "usage counter {}", part_name_with_state, it->use_count());
+
+                if (!it->unique())
+                    LOG_WARNING(log, "Cannot immediately remove part {} because someone using it right now "
+                             "usage counter {}", part_name_with_state, it->use_count());
                 return;
+            }
 
             modifyPartState(it, DataPartState::Deleting);
 
@@ -3566,7 +3576,12 @@ void MergeTreeData::checkAlterPartitionIsPossible(
 void MergeTreeData::checkPartitionCanBeDropped(const ASTPtr & partition, ContextPtr local_context)
 {
     const String partition_id = getPartitionIDFromQuery(partition, local_context);
-    auto parts_to_remove = getVisibleDataPartsVectorInPartition(local_context, partition_id);
+    DataPartsVector parts_to_remove;
+    const auto * partition_ast = partition->as<ASTPartition>();
+    if (partition_ast && partition_ast->all)
+        parts_to_remove = getVisibleDataPartsVector(local_context);
+    else
+        parts_to_remove = getVisibleDataPartsVectorInPartition(local_context, partition_id);
 
     UInt64 partition_size = 0;
 
@@ -4020,6 +4035,8 @@ String MergeTreeData::getPartitionIDFromQuery(const ASTPtr & ast, ContextPtr loc
 
     auto metadata_snapshot = getInMemoryMetadataPtr();
     const Block & key_sample_block = metadata_snapshot->getPartitionKey().sample_block;
+    if (partition_ast.all)
+        return "ALL";
     size_t fields_count = key_sample_block.columns();
     if (partition_ast.fields_count != fields_count)
         throw Exception(ErrorCodes::INVALID_PARTITION_VALUE,

src/Storages/StorageMergeTree.cpp

@@ -1474,7 +1474,11 @@ void StorageMergeTree::dropPartition(const ASTPtr & partition, bool detach, Cont
         /// This protects against "revival" of data for a removed partition after completion of merge.
         auto merge_blocker = stopMergesAndWait();
         String partition_id = getPartitionIDFromQuery(partition, local_context);
-        parts_to_remove = getVisibleDataPartsVectorInPartition(local_context, partition_id);
+        const auto * partition_ast = partition->as<ASTPartition>();
+        if (partition_ast && partition_ast->all)
+            parts_to_remove = getVisibleDataPartsVector(local_context);
+        else
+            parts_to_remove = getVisibleDataPartsVectorInPartition(local_context, partition_id);
 
         /// TODO should we throw an exception if parts_to_remove is empty?
         removePartsFromWorkingSet(local_context->getCurrentTransaction().get(), parts_to_remove, true);

src/Storages/StorageReplicatedMergeTree.cpp

@@ -43,6 +43,7 @@
 #include <Parsers/ASTDropQuery.h>
 #include <Parsers/ASTFunction.h>
 #include <Parsers/ASTOptimizeQuery.h>
+#include <Parsers/ASTPartition.h>
 #include <Parsers/ASTLiteral.h>
 #include <Parsers/queryToString.h>
 #include <Parsers/ASTCheckQuery.h>
@@ -3322,7 +3323,7 @@ void StorageReplicatedMergeTree::removePartAndEnqueueFetch(const String & part_n
         if (!broken_part_info.contains(part->info))
             continue;
 
-        /// Broken part itself ether already moved to detached or does not exist.
+        /// Broken part itself either already moved to detached or does not exist.
         assert(broken_part_info != part->info);
         part->makeCloneInDetached("covered-by-broken", getInMemoryMetadataPtr());
     }
@@ -4943,15 +4944,37 @@ void StorageReplicatedMergeTree::dropPartition(const ASTPtr & partition, bool de
         throw Exception("DROP PARTITION cannot be done on this replica because it is not a leader", ErrorCodes::NOT_A_LEADER);
 
     zkutil::ZooKeeperPtr zookeeper = getZooKeeperAndAssertNotReadonly();
-    LogEntry entry;
 
-    String partition_id = getPartitionIDFromQuery(partition, query_context);
+    const auto * partition_ast = partition->as<ASTPartition>();
-    bool did_drop = dropAllPartsInPartition(*zookeeper, partition_id, entry, query_context, detach);
+    if (partition_ast && partition_ast->all)
-
-    if (did_drop)
     {
-        waitForLogEntryToBeProcessedIfNecessary(entry, query_context);
+        Strings partitions = zookeeper->getChildren(fs::path(zookeeper_path) / "block_numbers");
-        cleanLastPartNode(partition_id);
+
+        std::vector<std::pair<String, std::unique_ptr<LogEntry>>> entries_with_partitionid_to_drop;
+        entries_with_partitionid_to_drop.reserve(partitions.size());
+        for (String & partition_id : partitions)
+        {
+            auto entry = std::make_unique<LogEntry>();
+            if (dropAllPartsInPartition(*zookeeper, partition_id, *entry, query_context, detach))
+                entries_with_partitionid_to_drop.emplace_back(partition_id, std::move(entry));
+        }
+
+        for (const auto & entry : entries_with_partitionid_to_drop)
+        {
+            waitForLogEntryToBeProcessedIfNecessary(*entry.second, query_context);
+            cleanLastPartNode(entry.first);
+        }
+    }
+    else
+    {
+        LogEntry entry;
+        String partition_id = getPartitionIDFromQuery(partition, query_context);
+        bool did_drop = dropAllPartsInPartition(*zookeeper, partition_id, entry, query_context, detach);
+        if (did_drop)
+        {
+            waitForLogEntryToBeProcessedIfNecessary(entry, query_context);
+            cleanLastPartNode(partition_id);
+        }
     }
 }
 

tests/ci/run_check.py

@@ -162,7 +162,7 @@ def check_pr_description(pr_info):
 
     i = 0
     while i < len(lines):
-        if re.match(r"(?i)^[>*_ ]*change\s*log\s*category", lines[i]):
+        if re.match(r"(?i)^[#>*_ ]*change\s*log\s*category", lines[i]):
             i += 1
             if i >= len(lines):
                 break
@@ -191,7 +191,7 @@ def check_pr_description(pr_info):
                 return result_status[:140], category
 
         elif re.match(
-            r"(?i)^[>*_ ]*(short\s*description|change\s*log\s*entry)", lines[i]
+            r"(?i)^[#>*_ ]*(short\s*description|change\s*log\s*entry)", lines[i]
         ):
             i += 1
             # Can have one empty line between header and the entry itself.
@@ -262,9 +262,14 @@ if __name__ == "__main__":
         remove_labels(gh, pr_info, pr_labels_to_remove)
 
     if description_report:
-        print("::notice ::Cannot run, description does not match the template")
+        print(
+            "::error ::Cannot run, PR description does not match the template: "
+            f"{description_report}"
+        )
         logging.info(
-            "PR body doesn't match the template: (start)\n%s\n(end)", pr_info.body
+            "PR body doesn't match the template: (start)\n%s\n(end)\n" "Reason: %s",
+            pr_info.body,
+            description_report,
         )
         url = (
             f"{GITHUB_SERVER_URL}/{GITHUB_REPOSITORY}/"

tests/performance/function_calculation_after_sorting_and_limit.xml

@@ -0,0 +1,4 @@
+<test>
+    <query>SELECT sipHash64(number) FROM numbers(1e8) ORDER BY number LIMIT 5</query>
+    <query>SELECT sipHash64(number) FROM numbers(1e8) ORDER BY number + 1 LIMIT 5</query>
+</test>

tests/performance/has_all.xml

@@ -0,0 +1,53 @@
+<test>
+    <substitutions>
+        <substitution>
+            <name>array_type</name>
+            <values>
+                <value>Int8</value>
+                <value>Int16</value>
+                <value>Int32</value>
+                <value>Int64</value>
+            </values>
+        </substitution>
+    </substitutions>
+
+    <create_query>
+        CREATE TABLE test_table_small_{array_type}
+        (
+            `set` Array({array_type}),
+            `subset` Array ({array_type})
+        )
+        ENGINE = MergeTree ORDER BY set;
+    </create_query>
+
+    <create_query>
+        CREATE TABLE test_table_medium_{array_type}
+        (
+            `set` Array({array_type}),
+            `subset` Array ({array_type})
+        )
+        ENGINE = MergeTree ORDER BY set;
+    </create_query>
+
+    <create_query>
+        CREATE TABLE test_table_large_{array_type}
+        (
+            `set` Array({array_type}),
+            `subset` Array ({array_type})
+        )
+        ENGINE = MergeTree ORDER BY set;
+    </create_query>
+
+
+    <fill_query>INSERT INTO test_table_small_{array_type} SELECT groupArraySample(5000)(rand64()) AS set, groupArraySample(500)(rand64()) AS subset FROM numbers(10000000) GROUP BY number % 5000;</fill_query>
+    <fill_query>INSERT INTO test_table_medium_{array_type} SELECT groupArraySample(50000)(rand64()) AS set, groupArraySample(5000)(rand64()) AS subset FROM numbers(25000000) GROUP BY number % 50000;</fill_query>
+    <fill_query>INSERT INTO test_table_large_{array_type} SELECT groupArraySample(500000)(rand64()) AS set, groupArraySample(500000)(rand64()) AS subset FROM numbers(50000000) GROUP BY number % 500000;</fill_query>
+
+    <query>SELECT hasAll(set, subset) FROM test_table_small_{array_type} FORMAT Null</query>
+    <query>SELECT hasAll(set, subset) FROM test_table_medium_{array_type} FORMAT Null</query>
+    <query>SELECT hasAll(set, subset) FROM test_table_large_{array_type} FORMAT Null</query>
+
+    <drop_query>DROP TABLE IF EXISTS test_table_small_{array_type}</drop_query>
+    <drop_query>DROP TABLE IF EXISTS test_table_medium_{array_type}</drop_query>
+    <drop_query>DROP TABLE IF EXISTS test_table_large_{array_type}</drop_query>
+</test>

tests/queries/0_stateless/00753_alter_attach.reference

@@ -10,3 +10,15 @@
 5	2
 6	3
 7	3
+4	2
+5	2
+1	1
+2	1
+3	1
+1	1
+2	1
+3	1
+1	1
+2	2
+1	1
+1	1

tests/queries/0_stateless/00753_alter_attach.sql

@@ -19,4 +19,53 @@ INSERT INTO alter_attach VALUES (6, 3), (7, 3);
 ALTER TABLE alter_attach ATTACH PARTITION 2;
 SELECT * FROM alter_attach ORDER BY x;
 
+ALTER TABLE alter_attach DETACH PARTITION ALL;
+SELECT * FROM alter_attach ORDER BY x;
+
+ALTER TABLE alter_attach ATTACH PARTITION 2;
+SELECT * FROM alter_attach ORDER BY x;
+
+DROP TABLE IF EXISTS detach_all_no_partition;
+CREATE TABLE detach_all_no_partition (x UInt64, p UInt8) ENGINE = MergeTree ORDER BY tuple();
+INSERT INTO detach_all_no_partition VALUES (1, 1), (2, 1), (3, 1);
+SELECT * FROM detach_all_no_partition ORDER BY x;
+
+ALTER TABLE detach_all_no_partition DETACH PARTITION ALL;
+SELECT * FROM detach_all_no_partition ORDER BY x;
+
+ALTER TABLE detach_all_no_partition ATTACH PARTITION tuple();
+SELECT * FROM detach_all_no_partition ORDER BY x;
+
 DROP TABLE alter_attach;
+DROP TABLE detach_all_no_partition;
+
+DROP TABLE IF EXISTS replicated_table_detach_all1;
+DROP TABLE IF EXISTS replicated_table_detach_all2;
+
+CREATE TABLE replicated_table_detach_all1 (
+  id UInt64,
+  Data String
+) ENGINE = ReplicatedMergeTree('/clickhouse/tables/test_00753_{database}/replicated_table_detach_all', '1') ORDER BY id PARTITION BY id;
+
+CREATE TABLE replicated_table_detach_all2 (
+  id UInt64,
+  Data String
+) ENGINE = ReplicatedMergeTree('/clickhouse/tables/test_00753_{database}/replicated_table_detach_all', '2') ORDER BY id PARTITION BY id;
+
+
+INSERT INTO replicated_table_detach_all1 VALUES (1, '1'), (2, '2');
+select * from replicated_table_detach_all1 order by id;
+
+ALTER TABLE replicated_table_detach_all1 DETACH PARTITION ALL;
+select * from replicated_table_detach_all1 order by id;
+SYSTEM SYNC REPLICA replicated_table_detach_all2;
+select * from replicated_table_detach_all2 order by id;
+
+ALTER TABLE replicated_table_detach_all1 ATTACH PARTITION tuple(1);
+select * from replicated_table_detach_all1 order by id;
+SYSTEM SYNC REPLICA replicated_table_detach_all2;
+select * from replicated_table_detach_all2 order by id;
+
+DROP TABLE replicated_table_detach_all1;
+DROP TABLE replicated_table_detach_all2;
+

tests/queries/0_stateless/01015_attach_part.reference

@@ -1,3 +1,4 @@
 1000
 0
 1000
+0

tests/queries/0_stateless/01015_attach_part.sql

@@ -21,4 +21,8 @@ ALTER TABLE table_01 ATTACH PART '20191001_1_1_0';
 
 SELECT COUNT() FROM table_01;
 
+ALTER TABLE table_01 DETACH PARTITION ALL;
+
+SELECT COUNT() FROM table_01;
+
 DROP TABLE IF EXISTS table_01;

tests/queries/0_stateless/01576_alias_column_rewrite.reference

@@ -35,10 +35,11 @@ Expression (Projection)
           ReadFromMergeTree (default.test_table)
 Expression (Projection)
   Limit (preliminary LIMIT (without OFFSET))
-    Sorting
+    Expression (Before ORDER BY [lifted up part])
-      Expression (Before ORDER BY)
+      Sorting
-        SettingQuotaAndLimits (Set limits and quota after reading from storage)
+        Expression (Before ORDER BY)
-          ReadFromMergeTree (default.test_table)
+          SettingQuotaAndLimits (Set limits and quota after reading from storage)
+            ReadFromMergeTree (default.test_table)
 optimize_aggregation_in_order
 Expression ((Projection + Before ORDER BY))
   Aggregating

tests/queries/0_stateless/01591_window_functions.reference

@@ -925,10 +925,11 @@ Expression ((Projection + Before ORDER BY))
   Window (Window step for window \'ORDER BY o ASC, number ASC\')
     Sorting (Sorting for window \'ORDER BY o ASC, number ASC\')
       Window (Window step for window \'ORDER BY number ASC\')
-        Sorting (Sorting for window \'ORDER BY number ASC\')
+        Expression ((Before window functions + (Projection + Before ORDER BY)) [lifted up part])
-          Expression ((Before window functions + (Projection + Before ORDER BY)))
+          Sorting (Sorting for window \'ORDER BY number ASC\')
-            SettingQuotaAndLimits (Set limits and quota after reading from storage)
+            Expression ((Before window functions + (Projection + Before ORDER BY)))
-              ReadFromStorage (SystemNumbers)
+              SettingQuotaAndLimits (Set limits and quota after reading from storage)
+                ReadFromStorage (SystemNumbers)
 -- A test case for the sort comparator found by fuzzer.
 SELECT
     max(number) OVER (ORDER BY number DESC NULLS FIRST),

tests/queries/0_stateless/01600_remerge_sort_lowered_memory_bytes_ratio.sql

@@ -10,8 +10,8 @@ set max_block_size=40960;
 --     MergeSortingTransform: Re-merging intermediate ORDER BY data (20 blocks with 819200 rows) to save memory consumption
 --     MergeSortingTransform: Memory usage is lowered from 186.25 MiB to 95.00 MiB
 --     MergeSortingTransform: Re-merging is not useful (memory usage was not lowered by remerge_sort_lowered_memory_bytes_ratio=2.0)
-select number k, repeat(toString(number), 11) v1, repeat(toString(number), 12) v2 from numbers(3e6) order by k limit 400e3 format Null; -- { serverError 241 }
+select number k, repeat(toString(number), 11) v1, repeat(toString(number), 12) v2 from numbers(3e6) order by v1, v2 limit 400e3 format Null; -- { serverError 241 }
-select number k, repeat(toString(number), 11) v1, repeat(toString(number), 12) v2 from numbers(3e6) order by k limit 400e3 settings remerge_sort_lowered_memory_bytes_ratio=2. format Null; -- { serverError 241 }
+select number k, repeat(toString(number), 11) v1, repeat(toString(number), 12) v2 from numbers(3e6) order by v1, v2 limit 400e3 settings remerge_sort_lowered_memory_bytes_ratio=2. format Null; -- { serverError 241 }
 
 -- remerge_sort_lowered_memory_bytes_ratio 1.9 is good (need at least 1.91/0.98=1.94)
 --     MergeSortingTransform: Re-merging intermediate ORDER BY data (20 blocks with 819200 rows) to save memory consumption

tests/queries/0_stateless/01655_plan_optimizations.reference

@@ -142,3 +142,12 @@ Filter
 Filter
 2	3
 2	3
+> function calculation should be done after sorting and limit (if possible)
+> Expression should be divided into two subexpressions and only one of them should be moved after Sorting
+Expression (Before ORDER BY [lifted up part])
+FUNCTION sipHash64
+Sorting
+Expression (Before ORDER BY)
+FUNCTION plus
+> this query should be executed without throwing an exception
+0

tests/queries/0_stateless/01655_plan_optimizations.sh

@@ -196,3 +196,12 @@ $CLICKHOUSE_CLIENT -q "
     select a, b from (
         select number + 1 as a, number + 2 as b from numbers(2) union all select number + 1 as b, number + 2 as a from numbers(2)
     ) where a != 1 settings enable_optimize_predicate_expression = 0"
+
+echo "> function calculation should be done after sorting and limit (if possible)"
+echo "> Expression should be divided into two subexpressions and only one of them should be moved after Sorting"
+$CLICKHOUSE_CLIENT -q "
+    explain actions = 1 select number as n, sipHash64(n) from numbers(100) order by number + 1 limit 5" |
+    sed 's/^ *//g' | grep -o "^ *\(Expression (Before ORDER BY.*)\|Sorting\|FUNCTION \w\+\)"
+echo "> this query should be executed without throwing an exception"
+$CLICKHOUSE_CLIENT -q "
+    select throwIf(number = 5) from (select * from numbers(10)) order by number limit 1"

tests/queries/0_stateless/02149_read_in_order_fixed_prefix.reference

@@ -7,13 +7,15 @@
 ExpressionTransform
   (Limit)
   Limit
-    (Sorting)
+    (Expression)
-    MergingSortedTransform 2 → 1
+    ExpressionTransform
-      (Expression)
+      (Sorting)
-      ExpressionTransform × 2
+      MergingSortedTransform 2 → 1
-        (SettingQuotaAndLimits)
+        (Expression)
-          (ReadFromMergeTree)
+        ExpressionTransform × 2
-          MergeTreeInOrder × 2 0 → 1
+          (SettingQuotaAndLimits)
+            (ReadFromMergeTree)
+            MergeTreeInOrder × 2 0 → 1
 2020-10-01	9
 2020-10-01	9
 2020-10-01	9
@@ -23,16 +25,18 @@ ExpressionTransform
 ExpressionTransform
   (Limit)
   Limit
-    (Sorting)
+    (Expression)
-    MergingSortedTransform 2 → 1
+    ExpressionTransform
-      (Expression)
+      (Sorting)
-      ExpressionTransform × 2
+      MergingSortedTransform 2 → 1
-        (SettingQuotaAndLimits)
+        (Expression)
-          (ReadFromMergeTree)
+        ExpressionTransform × 2
-          ReverseTransform
+          (SettingQuotaAndLimits)
-            MergeTreeReverse 0 → 1
+            (ReadFromMergeTree)
-              ReverseTransform
+            ReverseTransform
-                MergeTreeReverse 0 → 1
+              MergeTreeReverse 0 → 1
+                ReverseTransform
+                  MergeTreeReverse 0 → 1
 2020-10-01	9
 2020-10-01	9
 2020-10-01	9
@@ -42,15 +46,17 @@ ExpressionTransform
 ExpressionTransform
   (Limit)
   Limit
-    (Sorting)
+    (Expression)
-    FinishSortingTransform
+    ExpressionTransform
-      PartialSortingTransform
+      (Sorting)
-        MergingSortedTransform 2 → 1
+      FinishSortingTransform
-          (Expression)
+        PartialSortingTransform
-          ExpressionTransform × 2
+          MergingSortedTransform 2 → 1
-            (SettingQuotaAndLimits)
+            (Expression)
-              (ReadFromMergeTree)
+            ExpressionTransform × 2
-              MergeTreeInOrder × 2 0 → 1
+              (SettingQuotaAndLimits)
+                (ReadFromMergeTree)
+                MergeTreeInOrder × 2 0 → 1
 2020-10-11	0
 2020-10-11	0
 2020-10-11	0