/root/doris/be/src/olap/comparison_predicate.h

Source
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements.  See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership.  The ASF licenses this file
// to you under the Apache License, Version 2.0 (the
// "License"); you may not use this file except in compliance
// with the License.  You may obtain a copy of the License at
//
//   http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing,
// software distributed under the License is distributed on an
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
// KIND, either express or implied.  See the License for the
// specific language governing permissions and limitations
// under the License.

#pragma once

#include <cstdint>
#include <type_traits>

#include "common/compare.h"
#include "olap/column_predicate.h"
#include "olap/rowset/segment_v2/bloom_filter.h"
#include "olap/rowset/segment_v2/inverted_index_cache.h" // IWYU pragma: keep
#include "olap/rowset/segment_v2/inverted_index_reader.h"
#include "olap/wrapper_field.h"
#include "vec/columns/column_dictionary.h"

namespace doris {
#include "common/compile_check_begin.h"
template <PrimitiveType Type, PredicateType PT>
class ComparisonPredicateBase final : public ColumnPredicate {
public:
    ENABLE_FACTORY_CREATOR(ComparisonPredicateBase);
    using T = typename PrimitiveTypeTraits<Type>::CppType;
    ComparisonPredicateBase(uint32_t column_id, const T& value, bool opposite = false)
            : ColumnPredicate(column_id, Type, opposite), _value(value) {}
    ComparisonPredicateBase(const ComparisonPredicateBase<Type, PT>& other, uint32_t col_id)
            : ColumnPredicate(other, col_id), _value(other._value) {}
    ComparisonPredicateBase(const ComparisonPredicateBase<Type, PT>& other) = delete;
    std::shared_ptr<ColumnPredicate> clone(uint32_t col_id) const override {
        DCHECK(_segment_id_to_cached_code.empty());
        return ComparisonPredicateBase<Type, PT>::create_shared(*this, col_id);
    }
    std::string debug_string() const override {
        fmt::memory_buffer debug_string_buffer;
        fmt::format_to(debug_string_buffer, "ComparisonPredicateBase({})",
                       ColumnPredicate::debug_string());
        return fmt::to_string(debug_string_buffer);
    }

    PredicateType type() const override { return PT; }

    Status evaluate(const vectorized::IndexFieldNameAndTypePair& name_with_type,
                    IndexIterator* iterator, uint32_t num_rows,
                    roaring::Roaring* bitmap) const override {
        if (iterator == nullptr) {
            return Status::Error<ErrorCode::INVERTED_INDEX_EVALUATE_SKIPPED>(
                    "Inverted index evaluate skipped, no inverted index reader can not support "
                    "comparison predicate");
        }

        if (iterator->get_reader(segment_v2::InvertedIndexReaderType::STRING_TYPE) == nullptr &&
            iterator->get_reader(segment_v2::InvertedIndexReaderType::BKD) == nullptr) {
            return Status::Error<ErrorCode::INVERTED_INDEX_EVALUATE_SKIPPED>(
                    "Inverted index evaluate skipped, no inverted index reader can not support "
                    "comparison predicate");
        }

        InvertedIndexQueryType query_type = InvertedIndexQueryType::UNKNOWN_QUERY;
        switch (PT) {
        case PredicateType::EQ:
            query_type = InvertedIndexQueryType::EQUAL_QUERY;
            break;
        case PredicateType::NE:
            query_type = InvertedIndexQueryType::EQUAL_QUERY;
            break;
        case PredicateType::LT:
            query_type = InvertedIndexQueryType::LESS_THAN_QUERY;
            break;
        case PredicateType::LE:
            query_type = InvertedIndexQueryType::LESS_EQUAL_QUERY;
            break;
        case PredicateType::GT:
            query_type = InvertedIndexQueryType::GREATER_THAN_QUERY;
            break;
        case PredicateType::GE:
            query_type = InvertedIndexQueryType::GREATER_EQUAL_QUERY;
            break;
        default:
            return Status::InvalidArgument("invalid comparison predicate type {}", PT);
        }

        std::unique_ptr<InvertedIndexQueryParamFactory> query_param = nullptr;
        RETURN_IF_ERROR(
                InvertedIndexQueryParamFactory::create_query_value<Type>(&_value, query_param));

        InvertedIndexParam param;
        param.column_name = name_with_type.first;
        param.column_type = name_with_type.second;
        param.query_value = query_param->get_value();
        param.query_type = query_type;
        param.num_rows = num_rows;
        param.roaring = std::make_shared<roaring::Roaring>();
        RETURN_IF_ERROR(iterator->read_from_index(segment_v2::IndexParam {&param}));

        // mask out null_bitmap, since NULL cmp VALUE will produce NULL
        //  and be treated as false in WHERE
        // keep it after query, since query will try to read null_bitmap and put it to cache
        if (iterator->has_null()) {
            InvertedIndexQueryCacheHandle null_bitmap_cache_handle;
            RETURN_IF_ERROR(iterator->read_null_bitmap(&null_bitmap_cache_handle));
            std::shared_ptr<roaring::Roaring> null_bitmap = null_bitmap_cache_handle.get_bitmap();
            if (null_bitmap) {
                *bitmap -= *null_bitmap;
            }
        }

        if constexpr (PT == PredicateType::NE) {
            *bitmap -= *param.roaring;
        } else {
            *bitmap &= *param.roaring;
        }

        return Status::OK();
    }

    void evaluate_and(const vectorized::IColumn& column, const uint16_t* sel, uint16_t size,
                      bool* flags) const override {
        _evaluate_bit<true>(column, sel, size, flags);
    }

    bool evaluate_and(const std::pair<WrapperField*, WrapperField*>& statistic) const override {
        if (statistic.first->is_null() && statistic.second->is_null()) {
            return false;
        }

        T tmp_min_value = get_zone_map_value<Type, T>(statistic.first->cell_ptr());
        T tmp_max_value = get_zone_map_value<Type, T>(statistic.second->cell_ptr());

        if constexpr (PT == PredicateType::EQ) {
            return _operator(Compare::less_equal(tmp_min_value, _value) &&
                                     Compare::greater_equal(tmp_max_value, _value),
                             true);
        } else if constexpr (PT == PredicateType::NE) {
            return _operator(
                    Compare::equal(tmp_min_value, _value) && Compare::equal(tmp_max_value, _value),
                    true);
        } else if constexpr (PT == PredicateType::LT || PT == PredicateType::LE) {
            return _operator(tmp_min_value, _value);
        } else {
            static_assert(PT == PredicateType::GT || PT == PredicateType::GE);
            return _operator(tmp_max_value, _value);
        }
    }

    /**
     * To figure out whether this page is matched partially or completely.
     *
     * 1. EQ: if `_value` belongs to the interval [min, max], return true to further compute each value in this page.
     * 2. NE: return true to further compute each value in this page if some values not equal to `_value`.
     * 3. LT|LE: if `_value` is greater than min, return true to further compute each value in this page.
     * 4. GT|GE: if `_value` is less than max, return true to further compute each value in this page.
     */

    bool camp_field(const vectorized::Field& min_field, const vectorized::Field& max_field) const {
        T min_value;
        T max_value;
        if constexpr (is_int_or_bool(Type) || is_float_or_double(Type)) {
            min_value =
                    (typename PrimitiveTypeTraits<Type>::CppType)min_field
                            .template get<typename PrimitiveTypeTraits<Type>::NearestFieldType>();
            max_value =
                    (typename PrimitiveTypeTraits<Type>::CppType)max_field
                            .template get<typename PrimitiveTypeTraits<Type>::NearestFieldType>();
        } else {
            min_value = min_field.template get<typename PrimitiveTypeTraits<Type>::CppType>();
            max_value = max_field.template get<typename PrimitiveTypeTraits<Type>::CppType>();
        }

        if constexpr (PT == PredicateType::EQ) {
            return Compare::less_equal(min_value, _value) &&
                   Compare::greater_equal(max_value, _value);
        } else if constexpr (PT == PredicateType::NE) {
            return !Compare::equal(min_value, _value) || !Compare::equal(max_value, _value);
        } else if constexpr (PT == PredicateType::LT || PT == PredicateType::LE) {
            return Compare::less_equal(min_value, _value);
        } else {
            static_assert(PT == PredicateType::GT || PT == PredicateType::GE);
            return Compare::greater_equal(max_value, _value);
        }
    }

    bool evaluate_and(vectorized::ParquetPredicate::ColumnStat* statistic) const override {
        bool result = true;
        if ((*statistic->get_stat_func)(statistic, column_id())) {
            vectorized::Field min_field;
            vectorized::Field max_field;
            if (!vectorized::ParquetPredicate::parse_min_max_value(
                         statistic->col_schema, statistic->encoded_min_value,
                         statistic->encoded_max_value, *statistic->ctz, &min_field, &max_field)
                         .ok()) [[unlikely]] {
                result = true;
            } else {
                result = camp_field(min_field, max_field);
            }
        }

        if constexpr (PT == PredicateType::EQ) {
            if (result && statistic->get_bloom_filter_func != nullptr &&
                (*statistic->get_bloom_filter_func)(statistic, column_id())) {
                if (!statistic->bloom_filter) {
                    return result;
                }
                return evaluate_and(statistic->bloom_filter.get());
            }
        }
        return result;
    }

    bool evaluate_and(vectorized::ParquetPredicate::CachedPageIndexStat* statistic,
                      RowRanges* row_ranges) const override {
        vectorized::ParquetPredicate::PageIndexStat* stat = nullptr;
        if (!(statistic->get_stat_func)(&stat, column_id())) {
            return true;
        }

        for (int page_id = 0; page_id < stat->num_of_pages; page_id++) {
            if (stat->is_all_null[page_id]) {
                // all null page, not need read.
                continue;
            }

            vectorized::Field min_field;
            vectorized::Field max_field;
            if (!vectorized::ParquetPredicate::parse_min_max_value(
                         stat->col_schema, stat->encoded_min_value[page_id],
                         stat->encoded_max_value[page_id], *statistic->ctz, &min_field, &max_field)
                         .ok()) [[unlikely]] {
                row_ranges->add(stat->ranges[page_id]);
                continue;
            };

            if (camp_field(min_field, max_field)) {
                row_ranges->add(stat->ranges[page_id]);
            }
        };
        return row_ranges->count() > 0;
    }

    bool is_always_true(const std::pair<WrapperField*, WrapperField*>& statistic) const override {
        if (statistic.first->is_null() || statistic.second->is_null()) {
            return false;
        }

        T tmp_min_value = get_zone_map_value<Type, T>(statistic.first->cell_ptr());
        T tmp_max_value = get_zone_map_value<Type, T>(statistic.second->cell_ptr());

        if constexpr (PT == PredicateType::LT) {
            return _value > tmp_max_value;
        } else if constexpr (PT == PredicateType::LE) {
            return _value >= tmp_max_value;
        } else if constexpr (PT == PredicateType::GT) {
            return _value < tmp_min_value;
        } else if constexpr (PT == PredicateType::GE) {
            return _value <= tmp_min_value;
        }

        return false;
    }

    bool evaluate_del(const std::pair<WrapperField*, WrapperField*>& statistic) const override {
        if (statistic.first->is_null() || statistic.second->is_null()) {
            return false;
        }

        T tmp_min_value = get_zone_map_value<Type, T>(statistic.first->cell_ptr());
        T tmp_max_value = get_zone_map_value<Type, T>(statistic.second->cell_ptr());

        if constexpr (PT == PredicateType::EQ) {
            return tmp_min_value == _value && tmp_max_value == _value;
        } else if constexpr (PT == PredicateType::NE) {
            return tmp_min_value > _value || tmp_max_value < _value;
        } else if constexpr (PT == PredicateType::LT || PT == PredicateType::LE) {
            return _operator(tmp_max_value, _value);
        } else {
            static_assert(PT == PredicateType::GT || PT == PredicateType::GE);
            return _operator(tmp_min_value, _value);
        }
    }

    bool evaluate_and(const segment_v2::BloomFilter* bf) const override {
        if constexpr (PT == PredicateType::EQ) {
            // EQ predicate can not use ngram bf, just return true to accept
            if (bf->is_ngram_bf()) {
                return true;
            }
            if constexpr (std::is_same_v<T, StringRef>) {
                return bf->test_bytes(_value.data, _value.size);
            } else {
                // DecimalV2 using decimal12_t in bloom filter, should convert value to decimal12_t
                if constexpr (Type == PrimitiveType::TYPE_DECIMALV2) {
                    decimal12_t decimal12_t_val(_value.int_value(), _value.frac_value());
                    return bf->test_bytes(reinterpret_cast<const char*>(&decimal12_t_val),
                                          sizeof(decimal12_t));
                    // Datev1 using uint24_t in bloom filter
                } else if constexpr (Type == PrimitiveType::TYPE_DATE) {
                    uint24_t date_value(uint32_t(_value.to_olap_date()));
                    return bf->test_bytes(reinterpret_cast<const char*>(&date_value),
                                          sizeof(uint24_t));
                    // DatetimeV1 using int64_t in bloom filter
                } else if constexpr (Type == PrimitiveType::TYPE_DATETIME) {
                    int64_t datetime_value(_value.to_olap_datetime());
                    return bf->test_bytes(reinterpret_cast<const char*>(&datetime_value),
                                          sizeof(int64_t));
                } else {
                    return bf->test_bytes(reinterpret_cast<const char*>(&_value), sizeof(T));
                }
            }
        } else {
            LOG(FATAL) << "Bloom filter is not supported by predicate type.";
            return true;
        }
    }

    bool evaluate_and(const StringRef* dict_words, const size_t count) const override {
        if constexpr (std::is_same_v<T, StringRef>) {
            for (size_t i = 0; i != count; ++i) {
                if (_operator(dict_words[i], _value) ^ _opposite) {
                    return true;
                }
            }
            return false;
        }

        return true;
    }

    bool can_do_bloom_filter(bool ngram) const override {
        return PT == PredicateType::EQ && !ngram;
    }

    bool evaluate_and(const vectorized::ParquetBlockSplitBloomFilter* bf) const override {
        if constexpr (PT == PredicateType::EQ) {
            auto test_bytes = [&]<typename V>(const V& value) {
                return bf->test_bytes(const_cast<char*>(reinterpret_cast<const char*>(&value)),
                                      sizeof(V));
            };

            // Only support Parquet native types where physical == logical representation
            // BOOLEAN -> hash as int32 (Parquet bool stored as int32)
            if constexpr (Type == PrimitiveType::TYPE_BOOLEAN) {
                int32_t int32_value = static_cast<int32_t>(_value);
                return test_bytes(int32_value);
            } else if constexpr (Type == PrimitiveType::TYPE_INT) {
                // INT -> hash as int32
                return test_bytes(_value);
            } else if constexpr (Type == PrimitiveType::TYPE_BIGINT) {
                // BIGINT -> hash as int64
                return test_bytes(_value);
            } else if constexpr (Type == PrimitiveType::TYPE_FLOAT) {
                // FLOAT -> hash as float
                return test_bytes(_value);
            } else if constexpr (Type == PrimitiveType::TYPE_DOUBLE) {
                // DOUBLE -> hash as double
                return test_bytes(_value);
            } else if constexpr (std::is_same_v<T, StringRef>) {
                // VARCHAR/STRING -> hash bytes
                return bf->test_bytes(_value.data, _value.size);
            } else {
                // Unsupported types: return true (accept)
                return true;
            }
        } else {
            LOG(FATAL) << "Bloom filter is not supported by predicate type.";
            return true;
        }
    }

    void evaluate_or(const vectorized::IColumn& column, const uint16_t* sel, uint16_t size,
                     bool* flags) const override {
        _evaluate_bit<false>(column, sel, size, flags);
    }

    template <bool is_and>
    void __attribute__((flatten))
    _evaluate_vec_internal(const vectorized::IColumn& column, uint16_t size, bool* flags) const {
        uint16_t current_evaluated_rows = 0;
        uint16_t current_passed_rows = 0;
        if (_can_ignore()) {
            if (is_and) {
                for (uint16_t i = 0; i < size; i++) {
                    current_evaluated_rows += flags[i];
                }
            } else {
                current_evaluated_rows += size;
            }
        }

        // defer is created after its reference args are created.
        // so defer will be destroyed BEFORE the reference args.
        // so reference here is safe.
        // https://stackoverflow.com/questions/14688285/c-local-variable-destruction-order
        Defer defer([&]() {
            update_filter_info(current_evaluated_rows - current_passed_rows, current_evaluated_rows,
                               0);
            try_reset_judge_selectivity();
        });

        if (column.is_nullable()) {
            const auto* nullable_column_ptr =
                    vectorized::check_and_get_column<vectorized::ColumnNullable>(column);
            const auto& nested_column = nullable_column_ptr->get_nested_column();
            const auto& null_map = assert_cast<const vectorized::ColumnUInt8&>(
                                           nullable_column_ptr->get_null_map_column())
                                           .get_data();

            if (nested_column.is_column_dictionary()) {
                if constexpr (std::is_same_v<T, StringRef>) {
                    const auto* dict_column_ptr =
                            vectorized::check_and_get_column<vectorized::ColumnDictI32>(
                                    nested_column);

                    auto dict_code = _find_code_from_dictionary_column(*dict_column_ptr);
                    do {
                        if constexpr (PT == PredicateType::EQ) {
                            if (dict_code == -2) {
                                memset(flags, 0, size);
                                break;
                            }
                        }
                        const auto* data_array = dict_column_ptr->get_data().data();

                        _base_loop_vec<true, is_and>(size, flags, null_map.data(), data_array,
                                                     dict_code);
                    } while (false);
                } else {
                    LOG(FATAL) << "column_dictionary must use StringRef predicate.";
                    __builtin_unreachable();
                }
            } else {
                auto* data_array =
                        vectorized::check_and_get_column<
                                const vectorized::PredicateColumnType<PredicateEvaluateType<Type>>>(
                                nested_column)
                                ->get_data()
                                .data();

                _base_loop_vec<true, is_and>(size, flags, null_map.data(), data_array, _value);
            }
        } else {
            if (column.is_column_dictionary()) {
                if constexpr (std::is_same_v<T, StringRef>) {
                    const auto* dict_column_ptr =
                            vectorized::check_and_get_column<vectorized::ColumnDictI32>(column);
                    auto dict_code = _find_code_from_dictionary_column(*dict_column_ptr);
                    do {
                        if constexpr (PT == PredicateType::EQ) {
                            if (dict_code == -2) {
                                memset(flags, 0, size);
                                break;
                            }
                        }
                        const auto* data_array = dict_column_ptr->get_data().data();

                        _base_loop_vec<false, is_and>(size, flags, nullptr, data_array, dict_code);
                    } while (false);
                } else {
                    LOG(FATAL) << "column_dictionary must use StringRef predicate.";
                    __builtin_unreachable();
                }
            } else {
                auto* data_array =
                        vectorized::check_and_get_column<
                                vectorized::PredicateColumnType<PredicateEvaluateType<Type>>>(
                                column)
                                ->get_data()
                                .data();

                _base_loop_vec<false, is_and>(size, flags, nullptr, data_array, _value);
            }
        }

        if (_opposite) {
            for (uint16_t i = 0; i < size; i++) {
                flags[i] = !flags[i];
            }
        }

        if (_can_ignore()) {
            for (uint16_t i = 0; i < size; i++) {
                current_passed_rows += flags[i];
            }
            do_judge_selectivity(current_evaluated_rows - current_passed_rows,
                                 current_evaluated_rows);
        }
    }

    void evaluate_vec(const vectorized::IColumn& column, uint16_t size,
                      bool* flags) const override {
        _evaluate_vec_internal<false>(column, size, flags);
    }

    void evaluate_and_vec(const vectorized::IColumn& column, uint16_t size,
                          bool* flags) const override {
        _evaluate_vec_internal<true>(column, size, flags);
    }

    double get_ignore_threshold() const override { return get_comparison_ignore_thredhold(); }

private:
    uint16_t _evaluate_inner(const vectorized::IColumn& column, uint16_t* sel,
                             uint16_t size) const override {
        if (column.is_nullable()) {
            const auto* nullable_column_ptr =
                    vectorized::check_and_get_column<vectorized::ColumnNullable>(column);
            const auto& nested_column = nullable_column_ptr->get_nested_column();
            const auto& null_map = assert_cast<const vectorized::ColumnUInt8&>(
                                           nullable_column_ptr->get_null_map_column())
                                           .get_data();

            return _base_evaluate<true>(&nested_column, null_map.data(), sel, size);
        } else {
            return _base_evaluate<false>(&column, nullptr, sel, size);
        }
    }

    template <typename LeftT, typename RightT>
    bool _operator(const LeftT& lhs, const RightT& rhs) const {
        if constexpr (PT == PredicateType::EQ) {
            return Compare::equal(lhs, rhs);
        } else if constexpr (PT == PredicateType::NE) {
            return Compare::not_equal(lhs, rhs);
        } else if constexpr (PT == PredicateType::LT) {
            return Compare::less(lhs, rhs);
        } else if constexpr (PT == PredicateType::LE) {
            return Compare::less_equal(lhs, rhs);
        } else if constexpr (PT == PredicateType::GT) {
            return Compare::greater(lhs, rhs);
        } else if constexpr (PT == PredicateType::GE) {
            return Compare::greater_equal(lhs, rhs);
        }
    }

    constexpr bool _is_range() const { return PredicateTypeTraits::is_range(PT); }

    constexpr bool _is_greater() const { return _operator(1, 0); }

    constexpr bool _is_eq() const { return _operator(1, 1); }

    template <bool is_and>
    void _evaluate_bit(const vectorized::IColumn& column, const uint16_t* sel, uint16_t size,
                       bool* flags) const {
        if (column.is_nullable()) {
            const auto* nullable_column_ptr =
                    vectorized::check_and_get_column<vectorized::ColumnNullable>(column);
            const auto& nested_column = nullable_column_ptr->get_nested_column();
            const auto& null_map = assert_cast<const vectorized::ColumnUInt8&>(
                                           nullable_column_ptr->get_null_map_column())
                                           .get_data();

            _base_evaluate_bit<true, is_and>(&nested_column, null_map.data(), sel, size, flags);
        } else {
            _base_evaluate_bit<false, is_and>(&column, nullptr, sel, size, flags);
        }
    }

    template <bool is_nullable, bool is_and, typename TArray, typename TValue>
    void __attribute__((flatten))
    _base_loop_vec(uint16_t size, bool* __restrict bflags, const uint8_t* __restrict null_map,
                   const TArray* __restrict data_array, const TValue& value) const {
        //uint8_t helps compiler to generate vectorized code
        auto* flags = reinterpret_cast<uint8_t*>(bflags);
        if constexpr (is_and) {
            for (uint16_t i = 0; i < size; i++) {
                if constexpr (is_nullable) {
                    flags[i] &= (uint8_t)(!null_map[i] && _operator(data_array[i], value));
                } else {
                    flags[i] &= (uint8_t)_operator(data_array[i], value);
                }
            }
        } else {
            for (uint16_t i = 0; i < size; i++) {
                if constexpr (is_nullable) {
                    flags[i] = !null_map[i] && _operator(data_array[i], value);
                } else {
                    flags[i] = _operator(data_array[i], value);
                }
            }
        }
    }

    template <bool is_nullable, bool is_and, typename TArray, typename TValue>
    void _base_loop_bit(const uint16_t* sel, uint16_t size, bool* flags,
                        const uint8_t* __restrict null_map, const TArray* __restrict data_array,
                        const TValue& value) const {
        for (uint16_t i = 0; i < size; i++) {
            if (is_and ^ flags[i]) {
                continue;
            }
            if constexpr (is_nullable) {
                if (_opposite ^ is_and ^
                    (!null_map[sel[i]] && _operator(data_array[sel[i]], value))) {
                    flags[i] = !is_and;
                }
            } else {
                if (_opposite ^ is_and ^ _operator(data_array[sel[i]], value)) {
                    flags[i] = !is_and;
                }
            }
        }
    }

    template <bool is_nullable, bool is_and>
    void _base_evaluate_bit(const vectorized::IColumn* column, const uint8_t* null_map,
                            const uint16_t* sel, uint16_t size, bool* flags) const {
        if (column->is_column_dictionary()) {
            if constexpr (std::is_same_v<T, StringRef>) {
                const auto* dict_column_ptr =
                        vectorized::check_and_get_column<vectorized::ColumnDictI32>(column);
                const auto* data_array = dict_column_ptr->get_data().data();
                auto dict_code = _find_code_from_dictionary_column(*dict_column_ptr);
                _base_loop_bit<is_nullable, is_and>(sel, size, flags, null_map, data_array,
                                                    dict_code);
            } else {
                LOG(FATAL) << "column_dictionary must use StringRef predicate.";
                __builtin_unreachable();
            }
        } else {
            auto* data_array =
                    vectorized::check_and_get_column<
                            vectorized::PredicateColumnType<PredicateEvaluateType<Type>>>(column)
                            ->get_data()
                            .data();

            _base_loop_bit<is_nullable, is_and>(sel, size, flags, null_map, data_array, _value);
        }
    }

    template <bool is_nullable>
    uint16_t _base_evaluate(const vectorized::IColumn* column, const uint8_t* null_map,
                            uint16_t* sel, uint16_t size) const {
        if (column->is_column_dictionary()) {
            if constexpr (std::is_same_v<T, StringRef>) {
                const auto* dict_column_ptr =
                        vectorized::check_and_get_column<vectorized::ColumnDictI32>(column);
                const auto& pred_col = dict_column_ptr->get_data();
                const auto* pred_col_data = pred_col.data();
                auto dict_code = _find_code_from_dictionary_column(*dict_column_ptr);

                if constexpr (PT == PredicateType::EQ) {
                    if (dict_code == -2) {
                        return _opposite ? size : 0;
                    }
                }
                uint16_t new_size = 0;
#define EVALUATE_WITH_NULL_IMPL(IDX) \
    _opposite ^ (!null_map[IDX] && _operator(pred_col_data[IDX], dict_code))
#define EVALUATE_WITHOUT_NULL_IMPL(IDX) _opposite ^ _operator(pred_col_data[IDX], dict_code)
                EVALUATE_BY_SELECTOR(EVALUATE_WITH_NULL_IMPL, EVALUATE_WITHOUT_NULL_IMPL)
#undef EVALUATE_WITH_NULL_IMPL
#undef EVALUATE_WITHOUT_NULL_IMPL

                return new_size;
            } else {
                LOG(FATAL) << "column_dictionary must use StringRef predicate.";
                return 0;
            }
        } else {
            auto& pred_col =
                    vectorized::check_and_get_column<
                            vectorized::PredicateColumnType<PredicateEvaluateType<Type>>>(column)
                            ->get_data();
            auto pred_col_data = pred_col.data();
            uint16_t new_size = 0;
#define EVALUATE_WITH_NULL_IMPL(IDX) \
    _opposite ^ (!null_map[IDX] && _operator(pred_col_data[IDX], _value))
#define EVALUATE_WITHOUT_NULL_IMPL(IDX) _opposite ^ _operator(pred_col_data[IDX], _value)
            EVALUATE_BY_SELECTOR(EVALUATE_WITH_NULL_IMPL, EVALUATE_WITHOUT_NULL_IMPL)
#undef EVALUATE_WITH_NULL_IMPL
#undef EVALUATE_WITHOUT_NULL_IMPL
            return new_size;
        }
    }

    int32_t __attribute__((flatten))
    _find_code_from_dictionary_column(const vectorized::ColumnDictI32& column) const {
        int32_t code = 0;
        if (_segment_id_to_cached_code.if_contains(
                    column.get_rowset_segment_id(),
                    [&code](const auto& pair) { code = pair.second; })) {
            return code;
        }
        code = _is_range() ? column.find_code_by_bound(_value, _is_greater(), _is_eq())
                           : column.find_code(_value);
        // Sometimes the dict is not initialized when run comparison predicate here, for example,
        // the full page is null, then the reader will skip read, so that the dictionary is not
        // inited. The cached code is wrong during this case, because the following page maybe not
        // null, and the dict should have items in the future.
        //
        // Cached code may have problems, so that add a config here, if not opened, then
        // we will return the code and not cache it.
        if (!column.is_dict_empty() && config::enable_low_cardinality_cache_code) {
            _segment_id_to_cached_code.emplace(std::pair {column.get_rowset_segment_id(), code});
        }

        return code;
    }

    mutable phmap::parallel_flat_hash_map<
            std::pair<RowsetId, uint32_t>, int32_t,
            phmap::priv::hash_default_hash<std::pair<RowsetId, uint32_t>>,
            phmap::priv::hash_default_eq<std::pair<RowsetId, uint32_t>>,
            std::allocator<std::pair<const std::pair<RowsetId, uint32_t>, int32_t>>, 4,
            std::shared_mutex>
            _segment_id_to_cached_code;
    T _value;
};
#include "common/compile_check_end.h"
} //namespace doris

Coverage Report

Created: 2025-12-26 18:42