PerfectJoinHashTable.cpp File Reference

#include "QueryEngine/JoinHashTable/PerfectJoinHashTable.h"
#include <atomic>
#include <future>
#include <numeric>
#include <thread>
#include "Logger/Logger.h"
#include "QueryEngine/CodeGenerator.h"
#include "QueryEngine/ColumnFetcher.h"
#include "QueryEngine/Execute.h"
#include "QueryEngine/ExpressionRewrite.h"
#include "QueryEngine/JoinHashTable/Builders/PerfectHashTableBuilder.h"
#include "QueryEngine/JoinHashTable/Runtime/HashJoinRuntime.h"
#include "QueryEngine/RuntimeFunctions.h"

Include dependency graph for PerfectJoinHashTable.cpp:

Go to the source code of this file.

Namespaces
	anonymous_namespace{PerfectJoinHashTable.cpp}

Functions
std::pair< InnerOuter, InnerOuterStringOpInfos >	anonymous_namespace{PerfectJoinHashTable.cpp}::get_cols (const Analyzer::BinOper *qual_bin_oper, const TemporaryTables *temporary_tables)
BucketizedHashEntryInfo	anonymous_namespace{PerfectJoinHashTable.cpp}::get_bucketized_hash_entry_info (SQLTypeInfo const &context_ti, ExpressionRange const &col_range, bool const is_bw_eq)
size_t	anonymous_namespace{PerfectJoinHashTable.cpp}::get_hash_entry_count (const ExpressionRange &col_range, const bool is_bw_eq)
bool	anonymous_namespace{PerfectJoinHashTable.cpp}::shard_count_less_or_equal_device_count (const shared::TableKey &inner_table_key, const Executor *executor)
size_t	get_shard_count (std::pair< const Analyzer::ColumnVar *, const Analyzer::Expr * > equi_pair, const Executor *executor)
bool	needs_dictionary_translation (const InnerOuter &inner_outer_col_pair, const InnerOuterStringOpInfos &inner_outer_string_op_infos, const Executor *executor)
std::vector < Fragmenter_Namespace::FragmentInfo >	only_shards_for_device (const std::vector< Fragmenter_Namespace::FragmentInfo > &fragments, const int device_id, const int device_count)
const InputTableInfo &	get_inner_query_info (const shared::TableKey &inner_table_key, const std::vector< InputTableInfo > &query_infos)
size_t	get_entries_per_device (const size_t total_entries, const size_t shard_count, const size_t device_count, const Data_Namespace::MemoryLevel memory_level)

Function Documentation

size_t get_entries_per_device	(	const size_t	total_entries,
		const size_t	shard_count,
		const size_t	device_count,
		const Data_Namespace::MemoryLevel	memory_level
	)

Definition at line 1271 of file PerfectJoinHashTable.cpp.

References CHECK_GT, and Data_Namespace::GPU_LEVEL.

Referenced by OverlapsJoinHashTable::computeHashTableCounts(), RangeJoinHashTable::computeRangeHashTableCounts(), and BaselineJoinHashTable::reifyWithLayout().

                                                                            {
  const auto entries_per_shard =
      shard_count ? (total_entries + shard_count - 1) / shard_count : total_entries;
  size_t entries_per_device = entries_per_shard;
  if (memory_level == Data_Namespace::GPU_LEVEL && shard_count) {
    const auto shards_per_device = (shard_count + device_count - 1) / device_count;
    CHECK_GT(shards_per_device, 0u);
    entries_per_device = entries_per_shard * shards_per_device;
  }
  return entries_per_device;
}

Here is the caller graph for this function:

const InputTableInfo& get_inner_query_info	(	const shared::TableKey &	inner_table_key,
		const std::vector< InputTableInfo > &	query_infos
	)

Definition at line 1257 of file PerfectJoinHashTable.cpp.

References CHECK.

Referenced by PerfectJoinHashTable::getInnerQueryInfo(), RangeJoinHashTable::getInstance(), OverlapsJoinHashTable::getInstance(), RangeJoinHashTable::reifyWithLayout(), OverlapsJoinHashTable::reifyWithLayout(), and BaselineJoinHashTable::reifyWithLayout().

                                                  {
  std::optional<size_t> ti_idx;
  for (size_t i = 0; i < query_infos.size(); ++i) {
    if (inner_table_key == query_infos[i].table_key) {
      ti_idx = i;
      break;
    }
  }
  CHECK(ti_idx);
  return query_infos[*ti_idx];
}

Here is the caller graph for this function:

size_t get_shard_count	(	std::pair< const Analyzer::ColumnVar *, const Analyzer::Expr * >	equi_pair,
		const Executor *	executor
	)

Definition at line 112 of file PerfectJoinHashTable.cpp.

References CHECK, Catalog_Namespace::get_metadata_for_table(), and anonymous_namespace{PerfectJoinHashTable.cpp}::shard_count_less_or_equal_device_count().

                              {
  const auto inner_col = equi_pair.first;
  const auto outer_col = dynamic_cast<const Analyzer::ColumnVar*>(equi_pair.second);
  if (!outer_col || inner_col->getColumnKey().table_id < 0 ||
      outer_col->getColumnKey().table_id < 0) {
    return 0;
  }
  if (outer_col->get_rte_idx()) {
    return 0;
  }
  if (inner_col->get_type_info() != outer_col->get_type_info()) {
    return 0;
  }

  const auto inner_td =
      Catalog_Namespace::get_metadata_for_table(inner_col->getTableKey());
  CHECK(inner_td);
  const auto outer_td =
      Catalog_Namespace::get_metadata_for_table(outer_col->getTableKey());
  CHECK(outer_td);
  if (inner_td->shardedColumnId == 0 || outer_td->shardedColumnId == 0 ||
      inner_td->nShards != outer_td->nShards) {
    return 0;
  }
  if (!shard_count_less_or_equal_device_count(inner_col->getTableKey(), executor)) {
    return 0;
  }
  // The two columns involved must be the ones on which the tables have been sharded on.
  return (inner_td->shardedColumnId == inner_col->getColumnKey().column_id &&
          outer_td->shardedColumnId == outer_col->getColumnKey().column_id) ||
                 (outer_td->shardedColumnId == inner_col->getColumnKey().column_id &&
                  inner_td->shardedColumnId == inner_col->getColumnKey().column_id)
             ? inner_td->nShards
             : 0;
}

Here is the call graph for this function:

bool needs_dictionary_translation	(	const InnerOuter &	inner_outer_col_pair,
		const InnerOuterStringOpInfos &	inner_outer_string_op_infos,
		const Executor *	executor
	)

Definition at line 275 of file PerfectJoinHashTable.cpp.

References CHECK, CHECK_EQ, get_column_descriptor_maybe(), and get_column_type().

                              {
  if (inner_outer_string_op_infos.first.size() ||
      inner_outer_string_op_infos.second.size()) {
    return true;
  }
  auto inner_col = inner_outer_col_pair.first;
  auto outer_col_expr = inner_outer_col_pair.second;
  const auto inner_cd = get_column_descriptor_maybe(inner_col->getColumnKey());
  const auto& inner_col_key = inner_col->getColumnKey();
  const auto& inner_ti = get_column_type(inner_col_key.column_id,
                                         inner_col_key.table_id,
                                         inner_cd,
                                         executor->getTemporaryTables());
  // Only strings may need dictionary translation.
  if (!inner_ti.is_string()) {
    return false;
  }
  const auto outer_col = dynamic_cast<const Analyzer::ColumnVar*>(outer_col_expr);
  CHECK(outer_col);
  const auto outer_cd = get_column_descriptor_maybe(outer_col->getColumnKey());
  // Don't want to deal with temporary tables for now, require translation.
  if (!inner_cd || !outer_cd) {
    return true;
  }
  const auto& outer_col_key = outer_col->getColumnKey();
  const auto& outer_ti = get_column_type(outer_col_key.column_id,
                                         outer_col_key.table_id,
                                         outer_cd,
                                         executor->getTemporaryTables());
  CHECK_EQ(inner_ti.is_string(), outer_ti.is_string());
  // If the two columns don't share the dictionary, translation is needed.
  if (outer_ti.getStringDictKey() != inner_ti.getStringDictKey()) {
    return true;
  }
  const auto inner_str_dict_proxy =
      executor->getStringDictionaryProxy(inner_ti.getStringDictKey(), true);
  CHECK(inner_str_dict_proxy);
  const auto outer_str_dict_proxy =
      executor->getStringDictionaryProxy(outer_ti.getStringDictKey(), true);
  CHECK(outer_str_dict_proxy);

  return *inner_str_dict_proxy != *outer_str_dict_proxy;
}

Here is the call graph for this function:

std::vector<Fragmenter_Namespace::FragmentInfo> only_shards_for_device	(	const std::vector< Fragmenter_Namespace::FragmentInfo > &	fragments,
		const int	device_id,
		const int	device_count
	)

Definition at line 322 of file PerfectJoinHashTable.cpp.

References CHECK_GE.

Referenced by PerfectJoinHashTable::reify(), OverlapsJoinHashTable::reifyImpl(), RangeJoinHashTable::reifyWithLayout(), OverlapsJoinHashTable::reifyWithLayout(), and BaselineJoinHashTable::reifyWithLayout().

                            {
  std::vector<Fragmenter_Namespace::FragmentInfo> shards_for_device;
  for (const auto& fragment : fragments) {
    CHECK_GE(fragment.shard, 0);
    if (fragment.shard % device_count == device_id) {
      shards_for_device.push_back(fragment);
    }
  }
  return shards_for_device;
}

Here is the caller graph for this function: