add bench

This commit is contained in:
Shunsuke Kanda 2021-07-02 06:14:55 +09:00
parent 6e8efb6abb
commit b798603930
6 changed files with 188 additions and 8 deletions

View file

@ -70,9 +70,8 @@ template <class Trie>
[[maybe_unused]] std::vector<std::string> load_strings(std::string_view filepath) {
std::ifstream ifs(filepath);
if (!ifs) {
return {};
}
XCDAT_THROW_IF(!ifs.good(), "Cannot open the input file");
std::vector<std::string> strs;
for (std::string str; std::getline(ifs, str);) {
strs.push_back(str);

View file

@ -141,6 +141,14 @@ class bc_vector_7 {
return m_ints_l1.size() / 2;
}
inline std::uint64_t num_free_units() const {
return m_num_frees;
}
inline std::uint64_t num_nodes() const {
return num_units() - num_free_units();
}
inline std::uint64_t num_leaves() const {
return m_leaves.num_ones();
}

View file

@ -109,6 +109,14 @@ class bc_vector_8 {
return m_bytes[0].size() / 2;
}
inline std::uint64_t num_free_units() const {
return m_num_frees;
}
inline std::uint64_t num_nodes() const {
return num_units() - num_free_units();
}
inline std::uint64_t num_leaves() const {
return m_leaves.num_ones();
}

View file

@ -72,6 +72,21 @@ class trie {
return m_table.max_length();
}
//! Get the number of trie nodes.
inline std::uint64_t num_nodes() const {
return m_bcvec.num_nodes();
}
//! Get the number of DA units.
inline std::uint64_t num_units() const {
return m_bcvec.num_units();
}
//! Get the number of unused DA units.
inline std::uint64_t num_free_units() const {
return m_bcvec.num_free_units();
}
//! Lookup the ID of the keyword.
inline std::optional<std::uint64_t> lookup(std::string_view key) const {
std::uint64_t kpos = 0, npos = 0;
@ -98,12 +113,18 @@ class trie {
//! Decode the keyword associated with the ID.
inline std::string decode(std::uint64_t id) const {
if (num_keys() <= id) {
return {};
}
std::string decoded;
decoded.reserve(max_length());
decode(id, decoded);
return decoded;
}
//! Decode the keyword associated with the ID.
inline void decode(std::uint64_t id, std::string& decoded) const {
decoded.clear();
if (num_keys() <= id) {
return;
}
std::uint64_t npos = id_to_npos(id);
std::uint64_t tpos = m_bcvec.is_leaf(npos) ? m_bcvec.link(npos) : UINT64_MAX;
@ -118,7 +139,6 @@ class trie {
if (tpos != 0 && tpos != UINT64_MAX) {
m_tvec.decode(tpos, [&](char c) { decoded.push_back(c); });
}
return decoded;
}
//! An iterator class for common prefix search.

View file

@ -5,6 +5,7 @@ set(XCDAT_FILES
"xcdat_prefix_search"
"xcdat_predictive_search"
"xcdat_enumerate"
"xcdat_benchmark"
)
foreach(XCDAT_FILE ${XCDAT_FILES})

144
tools/xcdat_benchmark.cpp Normal file
View file

@ -0,0 +1,144 @@
#include <chrono>
#include <random>
#include <xcdat.hpp>
#include "cmd_line_parser/parser.hpp"
#include "tinyformat/tinyformat.h"
static constexpr int num_trials = 10;
cmd_line_parser::parser make_parser(int argc, char** argv) {
cmd_line_parser::parser p(argc, argv);
p.add("input_keys", "Input filepath of data keys");
p.add("num_samples", "Number of sample keys for benchmark (default=1000)", "-n", false);
p.add("random_seed", "Random seed for sampling (default=13)", "-s", false);
return p;
}
auto sample_keys(const std::vector<std::string>& keys, std::uint64_t num_samples, std::uint64_t random_seed) {
std::vector<std::string_view> sampled_keys(num_samples);
std::vector<std::uint64_t> sampled_ids(num_samples);
std::mt19937_64 engine(random_seed);
std::uniform_int_distribution<std::uint64_t> dist(0, keys.size() - 1);
for (std::uint64_t i = 0; i < num_samples; i++) {
sampled_ids[i] = dist(engine);
sampled_keys[i] = std::string_view(keys[sampled_ids[i]]);
}
return std::make_tuple(std::move(sampled_keys), std::move(sampled_ids));
}
template <class Trie>
Trie benchmark_build(const std::vector<std::string>& keys) {
const auto start_tp = std::chrono::high_resolution_clock::now();
Trie trie(keys);
const auto stop_tp = std::chrono::high_resolution_clock::now();
const auto dur_ms = std::chrono::duration_cast<std::chrono::milliseconds>(stop_tp - start_tp);
const double time_in_sec = dur_ms.count() / 1000.0;
const double memory_in_bytes = xcdat::memory_in_bytes(trie);
tfm::printfln("Number of trie nodes: %d", trie.num_nodes());
tfm::printfln("Number of DA units: %d", trie.num_units());
tfm::printfln("Number of free DA units: %d", trie.num_free_units());
tfm::printfln("Memory usage in bytes: %d", memory_in_bytes);
tfm::printfln("Memory usage in MiB: %g", memory_in_bytes / (1024.0 * 1024.0));
tfm::printfln("Construction time in seconds: %g", time_in_sec);
return trie;
}
template <class Trie>
void benchmark_lookup(const Trie& trie, const std::vector<std::string_view>& queries) {
// Warmup
volatile std::uint64_t tmp = 0;
for (const auto& query : queries) {
tmp += trie.lookup(query).value();
}
// Measure
const auto start_tp = std::chrono::high_resolution_clock::now();
for (int r = 0; r < num_trials; r++) {
for (const auto& query : queries) {
tmp += trie.lookup(query).value();
}
}
const auto stop_tp = std::chrono::high_resolution_clock::now();
const auto dur_us = std::chrono::duration_cast<std::chrono::microseconds>(stop_tp - start_tp);
const auto elapsed_us = static_cast<double>(dur_us.count());
tfm::printfln("Lookup time in microsec/query: %g", elapsed_us / (num_trials * queries.size()));
}
template <class Trie>
void benchmark_decode(const Trie& trie, const std::vector<std::uint64_t>& queries) {
// Warmup
std::string tmp;
for (const std::uint64_t query : queries) {
trie.decode(query, tmp);
}
// Measure
const auto start_tp = std::chrono::high_resolution_clock::now();
for (int r = 0; r < num_trials; r++) {
for (const std::uint64_t query : queries) {
trie.decode(query, tmp);
}
}
const auto stop_tp = std::chrono::high_resolution_clock::now();
const auto dur_us = std::chrono::duration_cast<std::chrono::microseconds>(stop_tp - start_tp);
const auto elapsed_us = static_cast<double>(dur_us.count());
tfm::printfln("Decode time in microsec/query: %g", elapsed_us / (num_trials * queries.size()));
}
template <class Trie>
void benchmark(std::vector<std::string> keys, const std::vector<std::string_view>& q_keys,
const std::vector<std::uint64_t>& q_ids) {
const auto trie = benchmark_build<Trie>(keys);
benchmark_lookup(trie, q_keys);
benchmark_decode(trie, q_ids);
}
int main(int argc, char** argv) {
#ifndef NDEBUG
tfm::warnfln("The code is running in debug mode.");
#endif
std::ios::sync_with_stdio(false);
auto p = make_parser(argc, argv);
if (!p.parse()) {
return 1;
}
const auto input_keys = p.get<std::string>("input_keys");
const auto num_samples = p.get<std::uint64_t>("num_samples", 1000);
const auto random_seed = p.get<std::uint64_t>("random_seed", 13);
auto keys = xcdat::load_strings(input_keys);
if (keys.empty()) {
tfm::errorfln("Error: The input dataset is empty.");
return 1;
}
// To unique
std::sort(keys.begin(), keys.end());
keys.erase(std::unique(keys.begin(), keys.end()), keys.end());
tfm::printfln("Number of keys: %d", keys.size());
auto [q_keys, q_ids] = sample_keys(keys, num_samples, random_seed);
tfm::printfln("** xcdat::trie_7_type **");
benchmark<xcdat::trie_7_type>(keys, q_keys, q_ids);
tfm::printfln("** xcdat::trie_8_type **");
benchmark<xcdat::trie_8_type>(keys, q_keys, q_ids);
return 0;
}