be/benchmark/parquet/parquet_benchmark_scenarios.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 <cstddef>
#include <cstdint>
#include <set>
#include <string>
#include <tuple>
#include <vector>

namespace doris::parquet_benchmark {

enum class Encoding {
    PLAIN,
    DICTIONARY,
    BYTE_STREAM_SPLIT,
    DELTA_BINARY_PACKED,
    DELTA_LENGTH_BYTE_ARRAY,
    DELTA_BYTE_ARRAY
};
enum class ValueType { INT32, INT64, FLOAT, DOUBLE, BYTE_ARRAY, FIXED_LEN_BYTE_ARRAY };
enum class Pattern { CLUSTERED, ALTERNATING };
enum class Projection { PREDICATE_ONLY, PREDICATE_PROJECTED };
enum class SelectionOperation { RESIZE_IDENTITY, ROW_FILTER, CASCADE_FILTER };
enum class ReaderOperation {
    OPEN_TO_FIRST_BLOCK,
    FULL_SCAN,
    PREDICATE_SCAN,
    COMPLEX_RESIDUAL_SCAN,
    LIMIT_1,
    LIMIT_1000
};
enum class Kernel {
    BYTE_STREAM_SPLIT,
    DELTA_PREFIX_SUM,
    DICTIONARY_GATHER,
    NULLABLE_EXPAND,
    RAW_PREDICATE,
    NESTED_SELECTION
};
enum class NestedSelectionImplementation { LEGACY, FUSED };
enum class NullableSelectionImplementation { LEGACY, FUSED };

struct DecoderScenario {
    Encoding encoding;
    ValueType value_type;
};

struct ReaderScenario {
    ReaderOperation operation;
    Encoding encoding;
    int null_percent;
    Pattern null_pattern;
    int selectivity_percent;
    Projection projection;
    int schema_width;
    int predicate_position;
    ValueType value_type = ValueType::INT32;
};

struct KernelScenario {
    Kernel kernel;
    ValueType value_type;
    int selectivity_percent;
    int null_percent;
    Pattern pattern;
    size_t dictionary_entries;
    NestedSelectionImplementation nested_implementation = NestedSelectionImplementation::FUSED;
};

struct SelectionScenario {
    SelectionOperation operation;
    int selectivity_percent;
    Pattern pattern;
};

struct NullableSelectionScenario {
    int selectivity_percent;
    int null_percent;
    Pattern selection_pattern;
    Pattern null_pattern;
    NullableSelectionImplementation implementation;
};

struct SelectionRange {
    size_t first;
    size_t count;
};

struct SelectionPlan {
    size_t total_rows = 0;
    size_t selected_rows = 0;
    std::vector<SelectionRange> ranges;
};

inline std::vector<DecoderScenario> decoder_scenarios() {
    return {
            {Encoding::PLAIN, ValueType::INT32},
            {Encoding::PLAIN, ValueType::INT64},
            {Encoding::PLAIN, ValueType::FLOAT},
            {Encoding::PLAIN, ValueType::DOUBLE},
            {Encoding::PLAIN, ValueType::BYTE_ARRAY},
            {Encoding::PLAIN, ValueType::FIXED_LEN_BYTE_ARRAY},
            {Encoding::DICTIONARY, ValueType::INT32},
            {Encoding::DICTIONARY, ValueType::INT64},
            {Encoding::DICTIONARY, ValueType::FLOAT},
            {Encoding::DICTIONARY, ValueType::DOUBLE},
            {Encoding::DICTIONARY, ValueType::BYTE_ARRAY},
            {Encoding::DICTIONARY, ValueType::FIXED_LEN_BYTE_ARRAY},
            {Encoding::BYTE_STREAM_SPLIT, ValueType::FLOAT},
            {Encoding::BYTE_STREAM_SPLIT, ValueType::DOUBLE},
            {Encoding::BYTE_STREAM_SPLIT, ValueType::FIXED_LEN_BYTE_ARRAY},
            {Encoding::DELTA_BINARY_PACKED, ValueType::INT32},
            {Encoding::DELTA_BINARY_PACKED, ValueType::INT64},
            {Encoding::DELTA_LENGTH_BYTE_ARRAY, ValueType::BYTE_ARRAY},
            {Encoding::DELTA_BYTE_ARRAY, ValueType::BYTE_ARRAY},
    };
}

inline std::vector<KernelScenario> kernel_scenarios() {
    std::vector<KernelScenario> scenarios;
    for (const auto value_type : {ValueType::FLOAT, ValueType::DOUBLE}) {
        scenarios.push_back(
                {Kernel::BYTE_STREAM_SPLIT, value_type, 100, 0, Pattern::CLUSTERED, 256});
    }
    for (const auto value_type : {ValueType::INT32, ValueType::INT64}) {
        scenarios.push_back(
                {Kernel::DELTA_PREFIX_SUM, value_type, 100, 0, Pattern::CLUSTERED, 256});
    }
    for (const auto value_type :
         {ValueType::INT32, ValueType::INT64, ValueType::FLOAT, ValueType::DOUBLE}) {
        for (const size_t dictionary_entries : {32, 4096, 262144}) {
            scenarios.push_back({Kernel::DICTIONARY_GATHER, value_type, 100, 0, Pattern::CLUSTERED,
                                 dictionary_entries});
        }
        for (const int null_percent : {0, 1, 10, 50, 90}) {
            for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
                scenarios.push_back(
                        {Kernel::NULLABLE_EXPAND, value_type, 100, null_percent, pattern, 256});
            }
        }
        for (const int selectivity : {0, 1, 10, 50, 90, 100}) {
            scenarios.push_back(
                    {Kernel::RAW_PREDICATE, value_type, selectivity, 0, Pattern::ALTERNATING, 256});
        }
    }
    for (const int selectivity : {1, 10, 50}) {
        for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
            for (const auto implementation :
                 {NestedSelectionImplementation::LEGACY, NestedSelectionImplementation::FUSED}) {
                scenarios.push_back({Kernel::NESTED_SELECTION, ValueType::INT32, selectivity, 10,
                                     pattern, 256, implementation});
            }
        }
    }
    return scenarios;
}

inline std::vector<SelectionScenario> selection_scenarios() {
    std::vector<SelectionScenario> scenarios {
            {SelectionOperation::RESIZE_IDENTITY, 100, Pattern::CLUSTERED}};
    for (const auto operation :
         {SelectionOperation::ROW_FILTER, SelectionOperation::CASCADE_FILTER}) {
        for (const int selectivity : {0, 1, 10, 50, 90, 100}) {
            for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
                scenarios.push_back({operation, selectivity, pattern});
            }
        }
    }
    return scenarios;
}

inline std::vector<NullableSelectionScenario> nullable_selection_scenarios() {
    std::vector<NullableSelectionScenario> scenarios;
    for (const int selectivity : {1, 10, 50, 90, 99}) {
        for (const int null_percent : {0, 1, 10, 50, 90}) {
            for (const auto selection_pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
                for (const auto null_pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
                    for (const auto implementation : {NullableSelectionImplementation::LEGACY,
                                                      NullableSelectionImplementation::FUSED}) {
                        scenarios.push_back({selectivity, null_percent, selection_pattern,
                                             null_pattern, implementation});
                    }
                }
            }
        }
    }
    return scenarios;
}

inline std::vector<ReaderScenario> reader_scenarios() {
    std::vector<ReaderScenario> scenarios;
    std::set<std::tuple<ReaderOperation, Encoding, int, Pattern, int, Projection, int, int,
                        ValueType>>
            seen;
    const auto add = [&](ReaderScenario scenario) {
        const auto key = std::make_tuple(
                scenario.operation, scenario.encoding, scenario.null_percent, scenario.null_pattern,
                scenario.selectivity_percent, scenario.projection, scenario.schema_width,
                scenario.predicate_position, scenario.value_type);
        if (seen.insert(key).second) {
            scenarios.push_back(scenario);
        }
    };

    const ReaderScenario baseline {.operation = ReaderOperation::FULL_SCAN,
                                   .encoding = Encoding::PLAIN,
                                   .null_percent = 10,
                                   .null_pattern = Pattern::ALTERNATING,
                                   .selectivity_percent = 10,
                                   .projection = Projection::PREDICATE_PROJECTED,
                                   .schema_width = 32,
                                   .predicate_position = 0};
    for (const auto operation :
         {ReaderOperation::OPEN_TO_FIRST_BLOCK, ReaderOperation::FULL_SCAN,
          ReaderOperation::PREDICATE_SCAN, ReaderOperation::COMPLEX_RESIDUAL_SCAN,
          ReaderOperation::LIMIT_1, ReaderOperation::LIMIT_1000}) {
        auto scenario = baseline;
        scenario.operation = operation;
        add(scenario);
    }
    for (const auto encoding : {Encoding::PLAIN, Encoding::DICTIONARY, Encoding::BYTE_STREAM_SPLIT,
                                Encoding::DELTA_BINARY_PACKED}) {
        auto scenario = baseline;
        scenario.encoding = encoding;
        add(scenario);
        scenario.operation = ReaderOperation::PREDICATE_SCAN;
        add(scenario);
    }
    for (const auto encoding : {Encoding::BYTE_STREAM_SPLIT, Encoding::DELTA_BINARY_PACKED}) {
        for (const int selectivity : {1, 10, 50, 90}) {
            for (const auto projection :
                 {Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
                auto scenario = baseline;
                scenario.operation = ReaderOperation::PREDICATE_SCAN;
                scenario.encoding = encoding;
                scenario.selectivity_percent = selectivity;
                scenario.projection = projection;
                add(scenario);
            }
        }
    }
    for (const int selectivity : {1, 10, 50, 90}) {
        for (const auto projection :
             {Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
            auto scenario = baseline;
            scenario.operation = ReaderOperation::PREDICATE_SCAN;
            scenario.encoding = Encoding::DICTIONARY;
            scenario.selectivity_percent = selectivity;
            scenario.projection = projection;
            add(scenario);
        }
    }
    for (const auto value_type : {ValueType::INT64, ValueType::BYTE_ARRAY}) {
        for (const int selectivity : {10, 50}) {
            for (const auto projection :
                 {Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
                auto scenario = baseline;
                scenario.operation = ReaderOperation::PREDICATE_SCAN;
                scenario.encoding = Encoding::DICTIONARY;
                scenario.selectivity_percent = selectivity;
                scenario.projection = projection;
                scenario.value_type = value_type;
                add(scenario);
            }
        }
    }
    for (const int width : {4, 32, 128, 512}) {
        for (const int predicate_position : {0, width - 1}) {
            auto scenario = baseline;
            scenario.operation = ReaderOperation::PREDICATE_SCAN;
            scenario.schema_width = width;
            scenario.predicate_position = predicate_position;
            add(scenario);
        }
    }
    for (const int null_percent : {0, 1, 10, 50, 90}) {
        for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
            for (const int selectivity : {0, 1, 10, 50, 90, 100}) {
                for (const auto projection :
                     {Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
                    auto scenario = baseline;
                    scenario.operation = ReaderOperation::PREDICATE_SCAN;
                    scenario.null_percent = null_percent;
                    scenario.null_pattern = pattern;
                    scenario.selectivity_percent = selectivity;
                    scenario.projection = projection;
                    add(scenario);
                }
            }
        }
    }
    return scenarios;
}

inline SelectionPlan make_selection_plan(size_t total_rows, int selectivity_percent,
                                         Pattern pattern) {
    SelectionPlan plan {.total_rows = total_rows, .selected_rows = 0, .ranges = {}};
    if (total_rows == 0 || selectivity_percent <= 0) {
        return plan;
    }
    if (selectivity_percent >= 100) {
        plan.selected_rows = total_rows;
        plan.ranges.push_back({.first = 0, .count = total_rows});
        return plan;
    }
    plan.selected_rows = total_rows * static_cast<size_t>(selectivity_percent) / 100;
    if (plan.selected_rows == 0) {
        plan.selected_rows = 1;
    }
    if (pattern == Pattern::CLUSTERED) {
        plan.ranges.push_back({.first = 0, .count = plan.selected_rows});
        return plan;
    }

    // Evenly spaced rows deliberately maximize the number of physical ranges. This is the
    // adversarial sparse shape that exposes per-run decoder and cursor overhead.
    for (size_t selected = 0; selected < plan.selected_rows; ++selected) {
        const size_t row = selected * total_rows / plan.selected_rows;
        if (!plan.ranges.empty() && plan.ranges.back().first + plan.ranges.back().count == row) {
            ++plan.ranges.back().count;
        } else {
            plan.ranges.push_back({.first = row, .count = 1});
        }
    }
    return plan;
}

template <typename Visitor>
inline void visit_selected_rows(const SelectionPlan& plan, Visitor visitor) {
    for (const auto& range : plan.ranges) {
        for (size_t offset = 0; offset < range.count; ++offset) {
            visitor(range.first + offset);
        }
    }
}

inline std::string to_string(Encoding value) {
    switch (value) {
    case Encoding::PLAIN:
        return "plain";
    case Encoding::DICTIONARY:
        return "dictionary";
    case Encoding::BYTE_STREAM_SPLIT:
        return "byte_stream_split";
    case Encoding::DELTA_BINARY_PACKED:
        return "delta_binary_packed";
    case Encoding::DELTA_LENGTH_BYTE_ARRAY:
        return "delta_length_byte_array";
    case Encoding::DELTA_BYTE_ARRAY:
        return "delta_byte_array";
    }
    return "unknown";
}

inline std::string to_string(ValueType value) {
    switch (value) {
    case ValueType::INT32:
        return "int32";
    case ValueType::INT64:
        return "int64";
    case ValueType::FLOAT:
        return "float";
    case ValueType::DOUBLE:
        return "double";
    case ValueType::BYTE_ARRAY:
        return "byte_array";
    case ValueType::FIXED_LEN_BYTE_ARRAY:
        return "fixed_len_byte_array";
    }
    return "unknown";
}

inline std::string to_string(Pattern value) {
    return value == Pattern::CLUSTERED ? "clustered" : "alternating";
}

inline std::string to_string(Projection value) {
    return value == Projection::PREDICATE_ONLY ? "predicate_only" : "predicate_projected";
}

inline std::string to_string(SelectionOperation value) {
    switch (value) {
    case SelectionOperation::RESIZE_IDENTITY:
        return "resize_identity";
    case SelectionOperation::ROW_FILTER:
        return "row_filter";
    case SelectionOperation::CASCADE_FILTER:
        return "cascade_filter";
    }
    return "unknown";
}

inline std::string to_string(ReaderOperation value) {
    switch (value) {
    case ReaderOperation::OPEN_TO_FIRST_BLOCK:
        return "open_to_first_block";
    case ReaderOperation::FULL_SCAN:
        return "full_scan";
    case ReaderOperation::PREDICATE_SCAN:
        return "predicate_scan";
    case ReaderOperation::COMPLEX_RESIDUAL_SCAN:
        return "complex_residual_scan";
    case ReaderOperation::LIMIT_1:
        return "limit_1";
    case ReaderOperation::LIMIT_1000:
        return "limit_1000";
    }
    return "unknown";
}

inline std::string reader_scenario_name(const ReaderScenario& scenario) {
    return to_string(scenario.operation) + "/" + to_string(scenario.encoding) + "/" +
           to_string(scenario.value_type) + "/null_" + std::to_string(scenario.null_percent) + "/" +
           to_string(scenario.null_pattern) + "/sel_" +
           std::to_string(scenario.selectivity_percent) + "/" + to_string(scenario.projection) +
           "/width_" + std::to_string(scenario.schema_width) + "/predicate_" +
           std::to_string(scenario.predicate_position);
}

inline std::string to_string(Kernel value) {
    switch (value) {
    case Kernel::BYTE_STREAM_SPLIT:
        return "byte_stream_split";
    case Kernel::DELTA_PREFIX_SUM:
        return "delta_prefix_sum";
    case Kernel::DICTIONARY_GATHER:
        return "dictionary_gather";
    case Kernel::NULLABLE_EXPAND:
        return "nullable_expand";
    case Kernel::RAW_PREDICATE:
        return "raw_predicate";
    case Kernel::NESTED_SELECTION:
        return "nested_selection";
    }
    return "unknown";
}

inline std::string to_string(NestedSelectionImplementation value) {
    switch (value) {
    case NestedSelectionImplementation::LEGACY:
        return "legacy";
    case NestedSelectionImplementation::FUSED:
        return "fused";
    }
    return "unknown";
}

inline std::string to_string(NullableSelectionImplementation value) {
    switch (value) {
    case NullableSelectionImplementation::LEGACY:
        return "legacy";
    case NullableSelectionImplementation::FUSED:
        return "fused";
    }
    return "unknown";
}

} // namespace doris::parquet_benchmark

Coverage Report

Created: 2026-08-06 18:25

Line	Count	Source
1		// Licensed to the Apache Software Foundation (ASF) under one
2		// or more contributor license agreements. See the NOTICE file
3		// distributed with this work for additional information
4		// regarding copyright ownership. The ASF licenses this file
5		// to you under the Apache License, Version 2.0 (the
6		// "License"); you may not use this file except in compliance
7		// with the License. You may obtain a copy of the License at
8		//
9		// http://www.apache.org/licenses/LICENSE-2.0
10		//
11		// Unless required by applicable law or agreed to in writing,
12		// software distributed under the License is distributed on an
13		// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14		// KIND, either express or implied. See the License for the
15		// specific language governing permissions and limitations
16		// under the License.
17
18		#pragma once
19
20		#include <cstddef>
21		#include <cstdint>
22		#include <set>
23		#include <string>
24		#include <tuple>
25		#include <vector>
26
27		namespace doris::parquet_benchmark {
28
29		enum class Encoding {
30		PLAIN,
31		DICTIONARY,
32		BYTE_STREAM_SPLIT,
33		DELTA_BINARY_PACKED,
34		DELTA_LENGTH_BYTE_ARRAY,
35		DELTA_BYTE_ARRAY
36		};
37		enum class ValueType { INT32, INT64, FLOAT, DOUBLE, BYTE_ARRAY, FIXED_LEN_BYTE_ARRAY };
38		enum class Pattern { CLUSTERED, ALTERNATING };
39		enum class Projection { PREDICATE_ONLY, PREDICATE_PROJECTED };
40		enum class SelectionOperation { RESIZE_IDENTITY, ROW_FILTER, CASCADE_FILTER };
41		enum class ReaderOperation {
42		OPEN_TO_FIRST_BLOCK,
43		FULL_SCAN,
44		PREDICATE_SCAN,
45		COMPLEX_RESIDUAL_SCAN,
46		LIMIT_1,
47		LIMIT_1000
48		};
49		enum class Kernel {
50		BYTE_STREAM_SPLIT,
51		DELTA_PREFIX_SUM,
52		DICTIONARY_GATHER,
53		NULLABLE_EXPAND,
54		RAW_PREDICATE,
55		NESTED_SELECTION
56		};
57		enum class NestedSelectionImplementation { LEGACY, FUSED };
58		enum class NullableSelectionImplementation { LEGACY, FUSED };
59
60		struct DecoderScenario {
61		Encoding encoding;
62		ValueType value_type;
63		};
64
65		struct ReaderScenario {
66		ReaderOperation operation;
67		Encoding encoding;
68		int null_percent;
69		Pattern null_pattern;
70		int selectivity_percent;
71		Projection projection;
72		int schema_width;
73		int predicate_position;
74		ValueType value_type = ValueType::INT32;
75		};
76
77		struct KernelScenario {
78		Kernel kernel;
79		ValueType value_type;
80		int selectivity_percent;
81		int null_percent;
82		Pattern pattern;
83		size_t dictionary_entries;
84		NestedSelectionImplementation nested_implementation = NestedSelectionImplementation::FUSED;
85		};
86
87		struct SelectionScenario {
88		SelectionOperation operation;
89		int selectivity_percent;
90		Pattern pattern;
91		};
92
93		struct NullableSelectionScenario {
94		int selectivity_percent;
95		int null_percent;
96		Pattern selection_pattern;
97		Pattern null_pattern;
98		NullableSelectionImplementation implementation;
99		};
100
101		struct SelectionRange {
102		size_t first;
103		size_t count;
104		};
105
106		struct SelectionPlan {
107		size_t total_rows = 0;
108		size_t selected_rows = 0;
109		std::vector<SelectionRange> ranges;
110		};
111
112	1	inline std::vector<DecoderScenario> decoder_scenarios() {
113	1	return {
114	1	{Encoding::PLAIN, ValueType::INT32},
115	1	{Encoding::PLAIN, ValueType::INT64},
116	1	{Encoding::PLAIN, ValueType::FLOAT},
117	1	{Encoding::PLAIN, ValueType::DOUBLE},
118	1	{Encoding::PLAIN, ValueType::BYTE_ARRAY},
119	1	{Encoding::PLAIN, ValueType::FIXED_LEN_BYTE_ARRAY},
120	1	{Encoding::DICTIONARY, ValueType::INT32},
121	1	{Encoding::DICTIONARY, ValueType::INT64},
122	1	{Encoding::DICTIONARY, ValueType::FLOAT},
123	1	{Encoding::DICTIONARY, ValueType::DOUBLE},
124	1	{Encoding::DICTIONARY, ValueType::BYTE_ARRAY},
125	1	{Encoding::DICTIONARY, ValueType::FIXED_LEN_BYTE_ARRAY},
126	1	{Encoding::BYTE_STREAM_SPLIT, ValueType::FLOAT},
127	1	{Encoding::BYTE_STREAM_SPLIT, ValueType::DOUBLE},
128	1	{Encoding::BYTE_STREAM_SPLIT, ValueType::FIXED_LEN_BYTE_ARRAY},
129	1	{Encoding::DELTA_BINARY_PACKED, ValueType::INT32},
130	1	{Encoding::DELTA_BINARY_PACKED, ValueType::INT64},
131	1	{Encoding::DELTA_LENGTH_BYTE_ARRAY, ValueType::BYTE_ARRAY},
132	1	{Encoding::DELTA_BYTE_ARRAY, ValueType::BYTE_ARRAY},
133	1	};
134	1	}
135
136	2	inline std::vector<KernelScenario> kernel_scenarios() {
137	2	std::vector<KernelScenario> scenarios;
138	4	for (const auto value_type : {ValueType::FLOAT, ValueType::DOUBLE}) {
139	4	scenarios.push_back(
140	4	{Kernel::BYTE_STREAM_SPLIT, value_type, 100, 0, Pattern::CLUSTERED, 256});
141	4	}
142	4	for (const auto value_type : {ValueType::INT32, ValueType::INT64}) {
143	4	scenarios.push_back(
144	4	{Kernel::DELTA_PREFIX_SUM, value_type, 100, 0, Pattern::CLUSTERED, 256});
145	4	}
146	2	for (const auto value_type :
147	8	{ValueType::INT32, ValueType::INT64, ValueType::FLOAT, ValueType::DOUBLE}) {
148	24	for (const size_t dictionary_entries : {32, 4096, 262144}) {
149	24	scenarios.push_back({Kernel::DICTIONARY_GATHER, value_type, 100, 0, Pattern::CLUSTERED,
150	24	dictionary_entries});
151	24	}
152	40	for (const int null_percent : {0, 1, 10, 50, 90}) {
153	80	for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
154	80	scenarios.push_back(
155	80	{Kernel::NULLABLE_EXPAND, value_type, 100, null_percent, pattern, 256});
156	80	}
157	40	}
158	48	for (const int selectivity : {0, 1, 10, 50, 90, 100}) {
159	48	scenarios.push_back(
160	48	{Kernel::RAW_PREDICATE, value_type, selectivity, 0, Pattern::ALTERNATING, 256});
161	48	}
162	8	}
163	6	for (const int selectivity : {1, 10, 50}) {
164	12	for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
165	12	for (const auto implementation :
166	24	{NestedSelectionImplementation::LEGACY, NestedSelectionImplementation::FUSED}) {
167	24	scenarios.push_back({Kernel::NESTED_SELECTION, ValueType::INT32, selectivity, 10,
168	24	pattern, 256, implementation});
169	24	}
170	12	}
171	6	}
172	2	return scenarios;
173	2	}
174
175	1	inline std::vector<SelectionScenario> selection_scenarios() {
176	1	std::vector<SelectionScenario> scenarios {
177	1	{SelectionOperation::RESIZE_IDENTITY, 100, Pattern::CLUSTERED}};
178	1	for (const auto operation :
179	2	{SelectionOperation::ROW_FILTER, SelectionOperation::CASCADE_FILTER}) {
180	12	for (const int selectivity : {0, 1, 10, 50, 90, 100}) {
181	24	for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
182	24	scenarios.push_back({operation, selectivity, pattern});
183	24	}
184	12	}
185	2	}
186	1	return scenarios;
187	1	}
188
189	1	inline std::vector<NullableSelectionScenario> nullable_selection_scenarios() {
190	1	std::vector<NullableSelectionScenario> scenarios;
191	5	for (const int selectivity : {1, 10, 50, 90, 99}) {
192	25	for (const int null_percent : {0, 1, 10, 50, 90}) {
193	50	for (const auto selection_pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
194	100	for (const auto null_pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
195	100	for (const auto implementation : {NullableSelectionImplementation::LEGACY,
196	200	NullableSelectionImplementation::FUSED}) {
197	200	scenarios.push_back({selectivity, null_percent, selection_pattern,
198	200	null_pattern, implementation});
199	200	}
200	100	}
201	50	}
202	25	}
203	5	}
204	1	return scenarios;
205	1	}
206
207	6	inline std::vector<ReaderScenario> reader_scenarios() {
208	6	std::vector<ReaderScenario> scenarios;
209	6	std::set<std::tuple<ReaderOperation, Encoding, int, Pattern, int, Projection, int, int,
210	6	ValueType>>
211	6	seen;
212	1.04k	const auto add = [&](ReaderScenario scenario) {
213	1.04k	const auto key = std::make_tuple(
214	1.04k	scenario.operation, scenario.encoding, scenario.null_percent, scenario.null_pattern,
215	1.04k	scenario.selectivity_percent, scenario.projection, scenario.schema_width,
216	1.04k	scenario.predicate_position, scenario.value_type);
217	1.04k	if (seen.insert(key).second) {
218	1.00k	scenarios.push_back(scenario);
219	1.00k	}
220	1.04k	};
221
222	6	const ReaderScenario baseline {.operation = ReaderOperation::FULL_SCAN,
223	6	.encoding = Encoding::PLAIN,
224	6	.null_percent = 10,
225	6	.null_pattern = Pattern::ALTERNATING,
226	6	.selectivity_percent = 10,
227	6	.projection = Projection::PREDICATE_PROJECTED,
228	6	.schema_width = 32,
229	6	.predicate_position = 0};
230	6	for (const auto operation :
231	6	{ReaderOperation::OPEN_TO_FIRST_BLOCK, ReaderOperation::FULL_SCAN,
232	6	ReaderOperation::PREDICATE_SCAN, ReaderOperation::COMPLEX_RESIDUAL_SCAN,
233	36	ReaderOperation::LIMIT_1, ReaderOperation::LIMIT_1000}) {
234	36	auto scenario = baseline;
235	36	scenario.operation = operation;
236	36	add(scenario);
237	36	}
238	6	for (const auto encoding : {Encoding::PLAIN, Encoding::DICTIONARY, Encoding::BYTE_STREAM_SPLIT,
239	24	Encoding::DELTA_BINARY_PACKED}) {
240	24	auto scenario = baseline;
241	24	scenario.encoding = encoding;
242	24	add(scenario);
243	24	scenario.operation = ReaderOperation::PREDICATE_SCAN;
244	24	add(scenario);
245	24	}
246	12	for (const auto encoding : {Encoding::BYTE_STREAM_SPLIT, Encoding::DELTA_BINARY_PACKED}) {
247	48	for (const int selectivity : {1, 10, 50, 90}) {
248	48	for (const auto projection :
249	96	{Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
250	96	auto scenario = baseline;
251	96	scenario.operation = ReaderOperation::PREDICATE_SCAN;
252	96	scenario.encoding = encoding;
253	96	scenario.selectivity_percent = selectivity;
254	96	scenario.projection = projection;
255	96	add(scenario);
256	96	}
257	48	}
258	12	}
259	24	for (const int selectivity : {1, 10, 50, 90}) {
260	24	for (const auto projection :
261	48	{Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
262	48	auto scenario = baseline;
263	48	scenario.operation = ReaderOperation::PREDICATE_SCAN;
264	48	scenario.encoding = Encoding::DICTIONARY;
265	48	scenario.selectivity_percent = selectivity;
266	48	scenario.projection = projection;
267	48	add(scenario);
268	48	}
269	24	}
270	12	for (const auto value_type : {ValueType::INT64, ValueType::BYTE_ARRAY}) {
271	24	for (const int selectivity : {10, 50}) {
272	24	for (const auto projection :
273	48	{Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
274	48	auto scenario = baseline;
275	48	scenario.operation = ReaderOperation::PREDICATE_SCAN;
276	48	scenario.encoding = Encoding::DICTIONARY;
277	48	scenario.selectivity_percent = selectivity;
278	48	scenario.projection = projection;
279	48	scenario.value_type = value_type;
280	48	add(scenario);
281	48	}
282	24	}
283	12	}
284	24	for (const int width : {4, 32, 128, 512}) {
285	48	for (const int predicate_position : {0, width - 1}) {
286	48	auto scenario = baseline;
287	48	scenario.operation = ReaderOperation::PREDICATE_SCAN;
288	48	scenario.schema_width = width;
289	48	scenario.predicate_position = predicate_position;
290	48	add(scenario);
291	48	}
292	24	}
293	30	for (const int null_percent : {0, 1, 10, 50, 90}) {
294	60	for (const auto pattern : {Pattern::CLUSTERED, Pattern::ALTERNATING}) {
295	360	for (const int selectivity : {0, 1, 10, 50, 90, 100}) {
296	360	for (const auto projection :
297	720	{Projection::PREDICATE_ONLY, Projection::PREDICATE_PROJECTED}) {
298	720	auto scenario = baseline;
299	720	scenario.operation = ReaderOperation::PREDICATE_SCAN;
300	720	scenario.null_percent = null_percent;
301	720	scenario.null_pattern = pattern;
302	720	scenario.selectivity_percent = selectivity;
303	720	scenario.projection = projection;
304	720	add(scenario);
305	720	}
306	360	}
307	60	}
308	30	}
309	6	return scenarios;
310	6	}
311
312		inline SelectionPlan make_selection_plan(size_t total_rows, int selectivity_percent,
313	4	Pattern pattern) {
314	4	SelectionPlan plan {.total_rows = total_rows, .selected_rows = 0, .ranges = {}};
315	4	if (total_rows == 0 \|\| selectivity_percent <= 0) {
316	1	return plan;
317	1	}
318	3	if (selectivity_percent >= 100) {
319	1	plan.selected_rows = total_rows;
320	1	plan.ranges.push_back({.first = 0, .count = total_rows});
321	1	return plan;
322	1	}
323	2	plan.selected_rows = total_rows * static_cast<size_t>(selectivity_percent) / 100;
324	2	if (plan.selected_rows == 0) {
325	0	plan.selected_rows = 1;
326	0	}
327	2	if (pattern == Pattern::CLUSTERED) {
328	1	plan.ranges.push_back({.first = 0, .count = plan.selected_rows});
329	1	return plan;
330	1	}
331
332		// Evenly spaced rows deliberately maximize the number of physical ranges. This is the
333		// adversarial sparse shape that exposes per-run decoder and cursor overhead.
334	101	for (size_t selected = 0; selected < plan.selected_rows; ++selected) {
335	100	const size_t row = selected * total_rows / plan.selected_rows;
336	100	if (!plan.ranges.empty() && plan.ranges.back().first + plan.ranges.back().count == row) {
337	0	++plan.ranges.back().count;
338	100	} else {
339	100	plan.ranges.push_back({.first = row, .count = 1});
340	100	}
341	100	}
342	1	return plan;
343	2	}
344
345		template <typename Visitor>
346	1	inline void visit_selected_rows(const SelectionPlan& plan, Visitor visitor) {
347	3	for (const auto& range : plan.ranges) {
348	8	for (size_t offset = 0; offset < range.count; ++offset) {
349	5	visitor(range.first + offset);
350	5	}
351	3	}
352	1	}
353
354	167	inline std::string to_string(Encoding value) {
355	167	switch (value) {
356	132	case Encoding::PLAIN:
357	132	return "plain";
358	17	case Encoding::DICTIONARY:
359	17	return "dictionary";
360	9	case Encoding::BYTE_STREAM_SPLIT:
361	9	return "byte_stream_split";
362	9	case Encoding::DELTA_BINARY_PACKED:
363	9	return "delta_binary_packed";
364	0	case Encoding::DELTA_LENGTH_BYTE_ARRAY:
365	0	return "delta_length_byte_array";
366	0	case Encoding::DELTA_BYTE_ARRAY:
367	0	return "delta_byte_array";
368	167	}
369	0	return "unknown";
370	167	}
371
372	167	inline std::string to_string(ValueType value) {
373	167	switch (value) {
374	159	case ValueType::INT32:
375	159	return "int32";
376	4	case ValueType::INT64:
377	4	return "int64";
378	0	case ValueType::FLOAT:
379	0	return "float";
380	0	case ValueType::DOUBLE:
381	0	return "double";
382	4	case ValueType::BYTE_ARRAY:
383	4	return "byte_array";
384	0	case ValueType::FIXED_LEN_BYTE_ARRAY:
385	0	return "fixed_len_byte_array";
386	167	}
387	0	return "unknown";
388	167	}
389
390	167	inline std::string to_string(Pattern value) {
391	167	return value == Pattern::CLUSTERED ? "clustered" : "alternating";
392	167	}
393
394	167	inline std::string to_string(Projection value) {
395	167	return value == Projection::PREDICATE_ONLY ? "predicate_only" : "predicate_projected";
396	167	}
397
398	0	inline std::string to_string(SelectionOperation value) {
399	0	switch (value) {
400	0	case SelectionOperation::RESIZE_IDENTITY:
401	0	return "resize_identity";
402	0	case SelectionOperation::ROW_FILTER:
403	0	return "row_filter";
404	0	case SelectionOperation::CASCADE_FILTER:
405	0	return "cascade_filter";
406	0	}
407	0	return "unknown";
408	0	}
409
410	167	inline std::string to_string(ReaderOperation value) {
411	167	switch (value) {
412	1	case ReaderOperation::OPEN_TO_FIRST_BLOCK:
413	1	return "open_to_first_block";
414	4	case ReaderOperation::FULL_SCAN:
415	4	return "full_scan";
416	159	case ReaderOperation::PREDICATE_SCAN:
417	159	return "predicate_scan";
418	1	case ReaderOperation::COMPLEX_RESIDUAL_SCAN:
419	1	return "complex_residual_scan";
420	1	case ReaderOperation::LIMIT_1:
421	1	return "limit_1";
422	1	case ReaderOperation::LIMIT_1000:
423	1	return "limit_1000";
424	167	}
425	0	return "unknown";
426	167	}
427
428	167	inline std::string reader_scenario_name(const ReaderScenario& scenario) {
429	167	return to_string(scenario.operation) + "/" + to_string(scenario.encoding) + "/" +
430	167	to_string(scenario.value_type) + "/null_" + std::to_string(scenario.null_percent) + "/" +
431	167	to_string(scenario.null_pattern) + "/sel_" +
432	167	std::to_string(scenario.selectivity_percent) + "/" + to_string(scenario.projection) +
433	167	"/width_" + std::to_string(scenario.schema_width) + "/predicate_" +
434	167	std::to_string(scenario.predicate_position);
435	167	}
436
437	0	inline std::string to_string(Kernel value) {
438	0	switch (value) {
439	0	case Kernel::BYTE_STREAM_SPLIT:
440	0	return "byte_stream_split";
441	0	case Kernel::DELTA_PREFIX_SUM:
442	0	return "delta_prefix_sum";
443	0	case Kernel::DICTIONARY_GATHER:
444	0	return "dictionary_gather";
445	0	case Kernel::NULLABLE_EXPAND:
446	0	return "nullable_expand";
447	0	case Kernel::RAW_PREDICATE:
448	0	return "raw_predicate";
449	0	case Kernel::NESTED_SELECTION:
450	0	return "nested_selection";
451	0	}
452	0	return "unknown";
453	0	}
454
455	0	inline std::string to_string(NestedSelectionImplementation value) {
456	0	switch (value) {
457	0	case NestedSelectionImplementation::LEGACY:
458	0	return "legacy";
459	0	case NestedSelectionImplementation::FUSED:
460	0	return "fused";
461	0	}
462	0	return "unknown";
463	0	}
464
465	0	inline std::string to_string(NullableSelectionImplementation value) {
466	0	switch (value) {
467	0	case NullableSelectionImplementation::LEGACY:
468	0	return "legacy";
469	0	case NullableSelectionImplementation::FUSED:
470	0	return "fused";
471	0	}
472	0	return "unknown";
473	0	}
474
475		} // namespace doris::parquet_benchmark