OmniSciDB  21ac014ffc
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups Pages
QueryMemoryDescriptor Class Reference

#include <QueryMemoryDescriptor.h>

+ Collaboration diagram for QueryMemoryDescriptor:

Public Member Functions

 QueryMemoryDescriptor ()
 
 QueryMemoryDescriptor (const Executor *executor, const RelAlgExecutionUnit &ra_exe_unit, const std::vector< InputTableInfo > &query_infos, const bool allow_multifrag, const bool keyless_hash, const bool interleaved_bins_on_gpu, const int32_t idx_target_as_key, const ColRangeInfo &col_range_info, const ColSlotContext &col_slot_context, const std::vector< int8_t > &group_col_widths, const int8_t group_col_compact_width, const std::vector< int64_t > &target_groupby_indices, const size_t entry_count, const CountDistinctDescriptors count_distinct_descriptors, const bool sort_on_gpu_hint, const bool output_columnar, const bool render_output, const bool must_use_baseline_sort, const bool use_streaming_top_n)
 
 QueryMemoryDescriptor (const Executor *executor, const size_t entry_count, const QueryDescriptionType query_desc_type, const bool is_table_function)
 
 QueryMemoryDescriptor (const QueryDescriptionType query_desc_type, const int64_t min_val, const int64_t max_val, const bool has_nulls, const std::vector< int8_t > &group_col_widths)
 
 QueryMemoryDescriptor (const TResultSetBufferDescriptor &thrift_query_memory_descriptor)
 
bool operator== (const QueryMemoryDescriptor &other) const
 
std::unique_ptr
< QueryExecutionContext
getQueryExecutionContext (const RelAlgExecutionUnit &, const Executor *executor, const ExecutorDeviceType device_type, const ExecutorDispatchMode dispatch_mode, const int device_id, const int64_t num_rows, const std::vector< std::vector< const int8_t * >> &col_buffers, const std::vector< std::vector< uint64_t >> &frag_offsets, std::shared_ptr< RowSetMemoryOwner >, const bool output_columnar, const bool sort_on_gpu, const size_t thread_idx, RenderInfo *) const
 
bool countDistinctDescriptorsLogicallyEmpty () const
 
const ExecutorgetExecutor () const
 
QueryDescriptionType getQueryDescriptionType () const
 
void setQueryDescriptionType (const QueryDescriptionType val)
 
bool isSingleColumnGroupByWithPerfectHash () const
 
bool hasKeylessHash () const
 
void setHasKeylessHash (const bool val)
 
bool hasInterleavedBinsOnGpu () const
 
void setHasInterleavedBinsOnGpu (const bool val)
 
int32_t getTargetIdxForKey () const
 
void setTargetIdxForKey (const int32_t val)
 
int8_t groupColWidth (const size_t key_idx) const
 
size_t getPrependedGroupColOffInBytes (const size_t group_idx) const
 
size_t getPrependedGroupBufferSizeInBytes () const
 
const auto groupColWidthsBegin () const
 
const auto groupColWidthsEnd () const
 
void clearGroupColWidths ()
 
bool isGroupBy () const
 
void setGroupColCompactWidth (const int8_t val)
 
size_t getColCount () const
 
size_t getSlotCount () const
 
const int8_t getPaddedSlotWidthBytes (const size_t slot_idx) const
 
const int8_t getLogicalSlotWidthBytes (const size_t slot_idx) const
 
const int8_t getSlotIndexForSingleSlotCol (const size_t col_idx) const
 
size_t getPaddedColWidthForRange (const size_t offset, const size_t range) const
 
void useConsistentSlotWidthSize (const int8_t slot_width_size)
 
size_t getRowWidth () const
 
int8_t updateActualMinByteWidth (const int8_t actual_min_byte_width) const
 
void addColSlotInfo (const std::vector< std::tuple< int8_t, int8_t >> &slots_for_col)
 
void clearSlotInfo ()
 
void alignPaddedSlots ()
 
int64_t getTargetGroupbyIndex (const size_t target_idx) const
 
void setAllTargetGroupbyIndices (std::vector< int64_t > group_by_indices)
 
size_t targetGroupbyIndicesSize () const
 
size_t targetGroupbyNegativeIndicesSize () const
 
void clearTargetGroupbyIndices ()
 
size_t getEntryCount () const
 
void setEntryCount (const size_t val)
 
int64_t getMinVal () const
 
int64_t getMaxVal () const
 
int64_t getBucket () const
 
bool hasNulls () const
 
const CountDistinctDescriptorgetCountDistinctDescriptor (const size_t idx) const
 
size_t getCountDistinctDescriptorsSize () const
 
bool sortOnGpu () const
 
bool canOutputColumnar () const
 
bool didOutputColumnar () const
 
void setOutputColumnar (const bool val)
 
bool useStreamingTopN () const
 
bool isLogicalSizedColumnsAllowed () const
 
bool mustUseBaselineSort () const
 
bool forceFourByteFloat () const
 
void setForceFourByteFloat (const bool val)
 
size_t getGroupbyColCount () const
 
size_t getKeyCount () const
 
size_t getBufferColSlotCount () const
 
size_t getBufferSizeBytes (const RelAlgExecutionUnit &ra_exe_unit, const unsigned thread_count, const ExecutorDeviceType device_type) const
 
size_t getBufferSizeBytes (const ExecutorDeviceType device_type) const
 
size_t getBufferSizeBytes (const ExecutorDeviceType device_type, const size_t override_entry_count) const
 
const ColSlotContextgetColSlotContext () const
 
bool usesGetGroupValueFast () const
 
bool blocksShareMemory () const
 
bool threadsShareMemory () const
 
bool lazyInitGroups (const ExecutorDeviceType) const
 
bool interleavedBins (const ExecutorDeviceType) const
 
size_t getColOffInBytes (const size_t col_idx) const
 
size_t getColOffInBytesInNextBin (const size_t col_idx) const
 
size_t getNextColOffInBytes (const int8_t *col_ptr, const size_t bin, const size_t col_idx) const
 
size_t getNextColOffInBytesRowOnly (const int8_t *col_ptr, const size_t col_idx) const
 
size_t getColOnlyOffInBytes (const size_t col_idx) const
 
size_t getRowSize () const
 
size_t getColsSize () const
 
size_t getWarpCount () const
 
size_t getCompactByteWidth () const
 
size_t getEffectiveKeyWidth () const
 
bool isWarpSyncRequired (const ExecutorDeviceType) const
 
std::string queryDescTypeToString () const
 
std::string toString () const
 
std::string reductionKey () const
 
bool hasVarlenOutput () const
 
std::optional< size_t > varlenOutputBufferElemSize () const
 
size_t varlenOutputRowSizeToSlot (const size_t slot_idx) const
 
bool slotIsVarlenOutput (const size_t slot_idx) const
 

Static Public Member Functions

static TResultSetBufferDescriptor toThrift (const QueryMemoryDescriptor &)
 
static std::unique_ptr
< QueryMemoryDescriptor
init (const Executor *executor, const RelAlgExecutionUnit &ra_exe_unit, const std::vector< InputTableInfo > &query_infos, const ColRangeInfo &col_range_info, const KeylessInfo &keyless_info, const bool allow_multifrag, const ExecutorDeviceType device_type, const int8_t crt_min_byte_width, const bool sort_on_gpu_hint, const size_t shard_count, const size_t max_groups_buffer_entry_count, RenderInfo *render_info, const CountDistinctDescriptors count_distinct_descriptors, const bool must_use_baseline_sort, const bool output_columnar_hint, const bool streaming_top_n_hint)
 
static bool many_entries (const int64_t max_val, const int64_t min_val, const int64_t bucket)
 
static bool countDescriptorsLogicallyEmpty (const CountDistinctDescriptors &count_distinct_descriptors)
 
static int8_t pick_target_compact_width (const RelAlgExecutionUnit &ra_exe_unit, const std::vector< InputTableInfo > &query_infos, const int8_t crt_min_byte_width)
 

Protected Member Functions

void resetGroupColWidths (const std::vector< int8_t > &new_group_col_widths)
 

Private Member Functions

size_t getTotalBytesOfColumnarBuffers () const
 
size_t getTotalBytesOfColumnarBuffers (const size_t num_entries_per_column) const
 
size_t getTotalBytesOfColumnarProjections (const size_t projection_count) const
 

Private Attributes

const Executorexecutor_
 
bool allow_multifrag_
 
QueryDescriptionType query_desc_type_
 
bool keyless_hash_
 
bool interleaved_bins_on_gpu_
 
int32_t idx_target_as_key_
 
std::vector< int8_t > group_col_widths_
 
int8_t group_col_compact_width_
 
std::vector< int64_t > target_groupby_indices_
 
size_t entry_count_
 
int64_t min_val_
 
int64_t max_val_
 
int64_t bucket_
 
bool has_nulls_
 
CountDistinctDescriptors count_distinct_descriptors_
 
bool sort_on_gpu_
 
bool output_columnar_
 
bool render_output_
 
bool must_use_baseline_sort_
 
bool is_table_function_
 
bool use_streaming_top_n_
 
bool force_4byte_float_
 
ColSlotContext col_slot_context_
 

Friends

class ResultSet
 
class QueryExecutionContext
 

Detailed Description

Definition at line 68 of file QueryMemoryDescriptor.h.

Constructor & Destructor Documentation

QueryMemoryDescriptor::QueryMemoryDescriptor ( )

Definition at line 541 of file QueryMemoryDescriptor.cpp.

References Projection.

542  : executor_(nullptr)
543  , allow_multifrag_(false)
545  , keyless_hash_(false)
546  , interleaved_bins_on_gpu_(false)
547  , idx_target_as_key_(0)
549  , entry_count_(0)
550  , min_val_(0)
551  , max_val_(0)
552  , bucket_(0)
553  , has_nulls_(false)
554  , sort_on_gpu_(false)
555  , output_columnar_(false)
556  , render_output_(false)
557  , must_use_baseline_sort_(false)
558  , is_table_function_(false)
559  , use_streaming_top_n_(false)
560  , force_4byte_float_(false) {}
QueryDescriptionType query_desc_type_
QueryMemoryDescriptor::QueryMemoryDescriptor ( const Executor executor,
const RelAlgExecutionUnit ra_exe_unit,
const std::vector< InputTableInfo > &  query_infos,
const bool  allow_multifrag,
const bool  keyless_hash,
const bool  interleaved_bins_on_gpu,
const int32_t  idx_target_as_key,
const ColRangeInfo col_range_info,
const ColSlotContext col_slot_context,
const std::vector< int8_t > &  group_col_widths,
const int8_t  group_col_compact_width,
const std::vector< int64_t > &  target_groupby_indices,
const size_t  entry_count,
const CountDistinctDescriptors  count_distinct_descriptors,
const bool  sort_on_gpu_hint,
const bool  output_columnar,
const bool  render_output,
const bool  must_use_baseline_sort,
const bool  use_streaming_top_n 
)

Definition at line 444 of file QueryMemoryDescriptor.cpp.

References anonymous_namespace{QueryMemoryDescriptor.cpp}::anyOf(), canOutputColumnar(), CHECK, col_slot_context_, count_distinct_descriptors_, countDescriptorsLogicallyEmpty(), executor_, streaming_top_n::get_heap_size(), getEntryCount(), getRowSize(), GroupByBaselineHash, GroupByPerfectHash, isLogicalSizedColumnsAllowed(), kAPPROX_QUANTILE, keyless_hash_, NonGroupedAggregate, output_columnar_, Projection, query_desc_type_, ColSlotContext::setAllSlotsPaddedSizeToLogicalSize(), ColSlotContext::setAllUnsetSlotsPaddedSize(), sort_on_gpu_, RelAlgExecutionUnit::target_exprs, RelAlgExecutionUnit::use_bump_allocator, use_streaming_top_n_, and ColSlotContext::validate().

464  : executor_(executor)
465  , allow_multifrag_(allow_multifrag)
466  , query_desc_type_(col_range_info.hash_type_)
467  , keyless_hash_(keyless_hash)
468  , interleaved_bins_on_gpu_(interleaved_bins_on_gpu)
469  , idx_target_as_key_(idx_target_as_key)
470  , group_col_widths_(group_col_widths)
471  , group_col_compact_width_(group_col_compact_width)
472  , target_groupby_indices_(target_groupby_indices)
473  , entry_count_(entry_count)
474  , min_val_(col_range_info.min)
475  , max_val_(col_range_info.max)
476  , bucket_(col_range_info.bucket)
477  , has_nulls_(col_range_info.has_nulls)
478  , count_distinct_descriptors_(count_distinct_descriptors)
479  , output_columnar_(false)
480  , render_output_(render_output)
481  , must_use_baseline_sort_(must_use_baseline_sort)
482  , is_table_function_(false)
484  , force_4byte_float_(false)
485  , col_slot_context_(col_slot_context) {
488 
489  sort_on_gpu_ = sort_on_gpu_hint && canOutputColumnar() && !keyless_hash_;
490 
491  if (sort_on_gpu_) {
492  CHECK(!ra_exe_unit.use_bump_allocator);
493  output_columnar_ = true;
494  } else {
495  switch (query_desc_type_) {
497  output_columnar_ = output_columnar_hint;
498  break;
500  output_columnar_ = output_columnar_hint &&
503  !anyOf(ra_exe_unit.target_exprs, kAPPROX_QUANTILE);
504  break;
506  output_columnar_ = output_columnar_hint;
507  break;
509  output_columnar_ = output_columnar_hint &&
512  !anyOf(ra_exe_unit.target_exprs, kAPPROX_QUANTILE);
513  break;
514  default:
515  output_columnar_ = false;
516  break;
517  }
518  }
519 
521  // TODO(adb): Ensure fixed size buffer allocations are correct with all logical column
522  // sizes
523  CHECK(!ra_exe_unit.use_bump_allocator);
526  }
527 
528 #ifdef HAVE_CUDA
529  // Check Streaming Top N heap usage, bail if > max slab size, CUDA ONLY
530  if (use_streaming_top_n_ && executor->getDataMgr()->gpusPresent()) {
531  const auto thread_count = executor->blockSize() * executor->gridSize();
532  const auto total_buff_size =
534  if (total_buff_size > executor_->maxGpuSlabSize()) {
535  throw StreamingTopNOOM(total_buff_size);
536  }
537  }
538 #endif
539 }
std::vector< Analyzer::Expr * > target_exprs
bool isLogicalSizedColumnsAllowed() const
QueryDescriptionType hash_type_
bool use_streaming_top_n(const RelAlgExecutionUnit &ra_exe_unit, const bool output_columnar)
std::vector< int64_t > target_groupby_indices_
CountDistinctDescriptors count_distinct_descriptors_
void validate() const
bool anyOf(std::vector< Analyzer::Expr * > const &target_exprs, SQLAgg const agg_kind)
QueryDescriptionType query_desc_type_
size_t get_heap_size(const size_t row_size, const size_t n, const size_t thread_count)
void setAllSlotsPaddedSizeToLogicalSize()
#define CHECK(condition)
Definition: Logger.h:206
std::vector< int8_t > group_col_widths_
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)
void setAllUnsetSlotsPaddedSize(const int8_t padded_size)

+ Here is the call graph for this function:

QueryMemoryDescriptor::QueryMemoryDescriptor ( const Executor executor,
const size_t  entry_count,
const QueryDescriptionType  query_desc_type,
const bool  is_table_function 
)

Definition at line 562 of file QueryMemoryDescriptor.cpp.

566  : executor_(executor)
567  , allow_multifrag_(false)
568  , query_desc_type_(query_desc_type)
569  , keyless_hash_(false)
570  , interleaved_bins_on_gpu_(false)
571  , idx_target_as_key_(0)
573  , entry_count_(entry_count)
574  , min_val_(0)
575  , max_val_(0)
576  , bucket_(0)
577  , has_nulls_(false)
578  , sort_on_gpu_(false)
579  , output_columnar_(false)
580  , render_output_(false)
581  , must_use_baseline_sort_(false)
582  , is_table_function_(is_table_function)
583  , use_streaming_top_n_(false)
584  , force_4byte_float_(false) {}
QueryDescriptionType query_desc_type_
QueryMemoryDescriptor::QueryMemoryDescriptor ( const QueryDescriptionType  query_desc_type,
const int64_t  min_val,
const int64_t  max_val,
const bool  has_nulls,
const std::vector< int8_t > &  group_col_widths 
)

Definition at line 586 of file QueryMemoryDescriptor.cpp.

591  : executor_(nullptr)
592  , allow_multifrag_(false)
593  , query_desc_type_(query_desc_type)
594  , keyless_hash_(false)
595  , interleaved_bins_on_gpu_(false)
596  , idx_target_as_key_(0)
597  , group_col_widths_(group_col_widths)
599  , entry_count_(0)
600  , min_val_(min_val)
601  , max_val_(max_val)
602  , bucket_(0)
603  , has_nulls_(false)
604  , sort_on_gpu_(false)
605  , output_columnar_(false)
606  , render_output_(false)
607  , must_use_baseline_sort_(false)
608  , is_table_function_(false)
609  , use_streaming_top_n_(false)
610  , force_4byte_float_(false) {}
QueryDescriptionType query_desc_type_
std::vector< int8_t > group_col_widths_
QueryMemoryDescriptor::QueryMemoryDescriptor ( const TResultSetBufferDescriptor &  thrift_query_memory_descriptor)

Member Function Documentation

void QueryMemoryDescriptor::addColSlotInfo ( const std::vector< std::tuple< int8_t, int8_t >> &  slots_for_col)

Definition at line 1156 of file QueryMemoryDescriptor.cpp.

References ColSlotContext::addColumn(), and col_slot_context_.

Referenced by QueryOutputBufferMemoryManager::allocate_output_buffers(), ResultSetLogicalValuesBuilder::create(), and TableFunctionExecutionContext::launchGpuCode().

1157  {
1158  col_slot_context_.addColumn(slots_for_col);
1159 }
void addColumn(const std::vector< std::tuple< int8_t, int8_t >> &slots_for_col)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

void QueryMemoryDescriptor::alignPaddedSlots ( )

Definition at line 1165 of file QueryMemoryDescriptor.cpp.

References ColSlotContext::alignPaddedSlots(), col_slot_context_, and sortOnGpu().

1165  {
1167 }
void alignPaddedSlots(const bool sort_on_gpu)

+ Here is the call graph for this function:

bool QueryMemoryDescriptor::blocksShareMemory ( ) const

Definition at line 1082 of file QueryMemoryDescriptor.cpp.

References bucket_, count_distinct_descriptors_, countDescriptorsLogicallyEmpty(), executor_, g_cluster, getGroupbyColCount(), GroupByBaselineHash, GroupByPerfectHash, is_table_function_, many_entries(), max_val_, min_val_, Projection, query_desc_type_, and render_output_.

Referenced by canOutputColumnar(), ResultSetReductionJIT::codegen(), QueryMemoryInitializer::computeNumberOfBuffers(), copy_group_by_buffers_from_gpu(), create_dev_group_by_buffers(), and toString().

1082  {
1083  if (g_cluster || is_table_function_) {
1084  return true;
1085  }
1087  return true;
1088  }
1089  if (executor_->isCPUOnly() || render_output_ ||
1093  getGroupbyColCount() > 1)) {
1094  return true;
1095  }
1098 }
static bool many_entries(const int64_t max_val, const int64_t min_val, const int64_t bucket)
size_t getGroupbyColCount() const
CountDistinctDescriptors count_distinct_descriptors_
QueryDescriptionType query_desc_type_
bool g_cluster
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::canOutputColumnar ( ) const

Definition at line 1169 of file QueryMemoryDescriptor.cpp.

References blocksShareMemory(), count_distinct_descriptors_, countDescriptorsLogicallyEmpty(), GPU, interleavedBins(), threadsShareMemory(), and usesGetGroupValueFast().

Referenced by QueryMemoryDescriptor().

1169  {
1173 }
CountDistinctDescriptors count_distinct_descriptors_
bool interleavedBins(const ExecutorDeviceType) const
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

void QueryMemoryDescriptor::clearGroupColWidths ( )
inline

Definition at line 194 of file QueryMemoryDescriptor.h.

References group_col_widths_.

194 { group_col_widths_.clear(); }
std::vector< int8_t > group_col_widths_
void QueryMemoryDescriptor::clearSlotInfo ( )

Definition at line 1161 of file QueryMemoryDescriptor.cpp.

References ColSlotContext::clear(), and col_slot_context_.

1161  {
1163 }

+ Here is the call graph for this function:

void QueryMemoryDescriptor::clearTargetGroupbyIndices ( )
inline

Definition at line 243 of file QueryMemoryDescriptor.h.

References target_groupby_indices_.

243 { target_groupby_indices_.clear(); }
std::vector< int64_t > target_groupby_indices_
static bool QueryMemoryDescriptor::countDescriptorsLogicallyEmpty ( const CountDistinctDescriptors count_distinct_descriptors)
inlinestatic

Definition at line 149 of file QueryMemoryDescriptor.h.

References Invalid.

Referenced by blocksShareMemory(), canOutputColumnar(), countDistinctDescriptorsLogicallyEmpty(), lazyInitGroups(), and QueryMemoryDescriptor().

150  {
151  return std::all_of(count_distinct_descriptors.begin(),
152  count_distinct_descriptors.end(),
153  [](const CountDistinctDescriptor& desc) {
154  return desc.impl_type_ == CountDistinctImplType::Invalid;
155  });
156  }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::countDistinctDescriptorsLogicallyEmpty ( ) const
inline

Definition at line 158 of file QueryMemoryDescriptor.h.

References count_distinct_descriptors_, and countDescriptorsLogicallyEmpty().

Referenced by QueryMemoryInitializer::allocateCountDistinctGpuMem().

158  {
160  }
CountDistinctDescriptors count_distinct_descriptors_
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::didOutputColumnar ( ) const
inline

Definition at line 265 of file QueryMemoryDescriptor.h.

References output_columnar_.

Referenced by ResultSetStorage::binSearchRowCount(), TargetExprCodegen::codegen(), ResultSetReductionJIT::codegen(), GroupByAndAggregate::codegen(), GroupByAndAggregate::codegenAggCalls(), GroupByAndAggregate::codegenAggColumnPtr(), TargetExprCodegen::codegenAggregate(), GroupByAndAggregate::codegenGroupBy(), GroupByAndAggregate::codegenMultiColumnBaselineHash(), GroupByAndAggregate::codegenMultiColumnPerfectHash(), GroupByAndAggregate::codegenOutputSlot(), GroupByAndAggregate::codegenSingleColumnPerfectHash(), GroupByAndAggregate::codegenWindowRowPointer(), copy_projection_buffer_from_gpu_columnar(), ResultSetStorage::copyKeyColWise(), ResultSet::createComparator(), ResultSet::didOutputColumnar(), anonymous_namespace{ResultSetReduction.cpp}::fill_slots(), ResultSetStorage::fillOneEntryColWise(), ResultSetStorage::fillOneEntryRowWise(), ResultSet::fixupQueryMemoryDescriptor(), get_cols_ptr(), ResultSet::getTargetValueFromBufferColwise(), GpuReductionHelperJIT::GpuReductionHelperJIT(), GpuSharedMemCodeBuilder::GpuSharedMemCodeBuilder(), ResultSetStorage::initializeBaselineValueSlots(), anonymous_namespace{TargetExprBuilder.cpp}::is_columnar_projection(), ResultSetReductionJIT::isEmpty(), ResultSetStorage::isEmptyEntry(), ResultSetStorage::isEmptyEntryColumnar(), QueryExecutionContext::launchCpuCode(), QueryExecutionContext::launchGpuCode(), ResultSet::makeGeoTargetValue(), ResultSetStorage::moveOneEntryToBuffer(), ResultSetStorage::reduce(), ResultSetStorage::reduceOneEntryBaseline(), ResultSetReductionJIT::reduceOneEntryBaselineIdx(), ResultSetStorage::reduceOneEntrySlotsBaseline(), ResultSetStorage::reduceOneSlotBaseline(), ResultSetStorage::reduceSingleRow(), and ResultSetStorage::rewriteAggregateBufferOffsets().

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::forceFourByteFloat ( ) const
inline

Definition at line 276 of file QueryMemoryDescriptor.h.

References force_4byte_float_.

Referenced by ResultSet::makeTargetValue().

+ Here is the caller graph for this function:

int64_t QueryMemoryDescriptor::getBucket ( ) const
inline

Definition at line 250 of file QueryMemoryDescriptor.h.

References bucket_.

Referenced by GroupByAndAggregate::codegenGroupBy(), and GroupByAndAggregate::codegenSingleColumnPerfectHash().

250 { return bucket_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getBufferColSlotCount ( ) const

Definition at line 1062 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, ColSlotContext::getSlotCount(), i, and target_groupby_indices_.

Referenced by anonymous_namespace{ResultSetIteration.cpp}::advance_col_buff_to_slot(), anonymous_namespace{ResultSetReduction.cpp}::fill_slots(), ResultSetStorage::fillOneEntryColWise(), and ResultSetStorage::fillOneEntryRowWise().

1062  {
1063  size_t total_slot_count = col_slot_context_.getSlotCount();
1064 
1065  if (target_groupby_indices_.empty()) {
1066  return total_slot_count;
1067  }
1068  return total_slot_count - std::count_if(target_groupby_indices_.begin(),
1070  [](const int64_t i) { return i >= 0; });
1071 }
std::vector< int64_t > target_groupby_indices_
size_t getSlotCount() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getBufferSizeBytes ( const RelAlgExecutionUnit ra_exe_unit,
const unsigned  thread_count,
const ExecutorDeviceType  device_type 
) const

Definition at line 993 of file QueryMemoryDescriptor.cpp.

References entry_count_, streaming_top_n::get_heap_size(), getRowSize(), SortInfo::limit, SortInfo::offset, RelAlgExecutionUnit::sort_info, and use_streaming_top_n_.

Referenced by QueryMemoryInitializer::applyStreamingTopNOffsetCpu(), QueryMemoryInitializer::copyGroupByBuffersFromGpu(), create_dev_group_by_buffers(), getBufferSizeBytes(), and QueryMemoryInitializer::QueryMemoryInitializer().

996  {
997  if (use_streaming_top_n_) {
998  const size_t n = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
999  return streaming_top_n::get_heap_size(getRowSize(), n, thread_count);
1000  }
1001  return getBufferSizeBytes(device_type, entry_count_);
1002 }
size_t getBufferSizeBytes(const RelAlgExecutionUnit &ra_exe_unit, const unsigned thread_count, const ExecutorDeviceType device_type) const
const size_t limit
size_t get_heap_size(const size_t row_size, const size_t n, const size_t thread_count)
const size_t offset

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getBufferSizeBytes ( const ExecutorDeviceType  device_type) const

Definition at line 1039 of file QueryMemoryDescriptor.cpp.

References entry_count_, and getBufferSizeBytes().

1040  {
1041  return getBufferSizeBytes(device_type, entry_count_);
1042 }
size_t getBufferSizeBytes(const RelAlgExecutionUnit &ra_exe_unit, const unsigned thread_count, const ExecutorDeviceType device_type) const

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getBufferSizeBytes ( const ExecutorDeviceType  device_type,
const size_t  entry_count 
) const

Returns total amount of output buffer memory for each device (CPU/GPU)

Columnar: if projection: it returns index buffer + columnar buffer (all non-lazy columns) if group by: it returns the amount required for each group column (assumes 64-bit per group) + columnar buffer (all involved agg columns)

Row-wise: returns required memory per row multiplied by number of entries

Definition at line 1015 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), CHECK_GE, executor_, getColsSize(), getRowSize(), getTotalBytesOfColumnarBuffers(), group_col_widths_, interleavedBins(), keyless_hash_, output_columnar_, Projection, and query_desc_type_.

1016  {
1017  if (keyless_hash_ && !output_columnar_) {
1018  CHECK_GE(group_col_widths_.size(), size_t(1));
1019  auto row_bytes = align_to_int64(getColsSize());
1020 
1021  return (interleavedBins(device_type) ? executor_->warpSize() : 1) * entry_count *
1022  row_bytes;
1023  }
1024 
1025  constexpr size_t row_index_width = sizeof(int64_t);
1026  size_t total_bytes{0};
1027  if (output_columnar_) {
1029  ? row_index_width * entry_count
1030  : sizeof(int64_t) * group_col_widths_.size() * entry_count) +
1032  } else {
1033  total_bytes = getRowSize() * entry_count;
1034  }
1035 
1036  return total_bytes;
1037 }
#define CHECK_GE(x, y)
Definition: Logger.h:219
size_t getTotalBytesOfColumnarBuffers() const
QueryDescriptionType query_desc_type_
bool interleavedBins(const ExecutorDeviceType) const
std::vector< int8_t > group_col_widths_
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getColCount ( ) const

Definition at line 1118 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getColCount().

1118  {
1119  return col_slot_context_.getColCount();
1120 }
size_t getColCount() const

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getColOffInBytes ( const size_t  col_idx) const

Definition at line 872 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), CHECK, CHECK_EQ, entry_count_, getColOnlyOffInBytes(), getEffectiveKeyWidth(), getPaddedSlotWidthBytes(), getPrependedGroupBufferSizeInBytes(), getWarpCount(), group_col_widths_, GroupByPerfectHash, keyless_hash_, output_columnar_, and query_desc_type_.

Referenced by TargetExprCodegen::codegen(), GroupByAndAggregate::codegenAggColumnPtr(), GroupByAndAggregate::codegenOutputSlot(), anonymous_namespace{QueryMemoryInitializer.cpp}::compact_projection_buffer_for_cpu_columnar(), copy_projection_buffer_from_gpu_columnar(), get_cols_ptr(), QueryExecutionContext::groupBufferToDeinterleavedResults(), QueryMemoryInitializer::initRowGroups(), inplace_sort_gpu(), and anonymous_namespace{Execute.cpp}::permute_storage_columnar().

872  {
873  const auto warp_count = getWarpCount();
874  if (output_columnar_) {
875  CHECK_EQ(size_t(1), warp_count);
876  size_t offset{0};
877  if (!keyless_hash_) {
879  }
880  for (size_t index = 0; index < col_idx; ++index) {
882  }
883  return offset;
884  }
885 
886  size_t offset{0};
887  if (keyless_hash_) {
888  // ignore, there's no group column in the output buffer
890  } else {
891  offset += group_col_widths_.size() * getEffectiveKeyWidth();
892  offset = align_to_int64(offset);
893  }
894  offset += getColOnlyOffInBytes(col_idx);
895  return offset;
896 }
#define CHECK_EQ(x, y)
Definition: Logger.h:214
size_t getEffectiveKeyWidth() const
size_t getColOnlyOffInBytes(const size_t col_idx) const
size_t getPrependedGroupBufferSizeInBytes() const
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const
QueryDescriptionType query_desc_type_
#define CHECK(condition)
Definition: Logger.h:206
std::vector< int8_t > group_col_widths_
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getColOffInBytesInNextBin ( const size_t  col_idx) const

Definition at line 931 of file QueryMemoryDescriptor.cpp.

References CHECK_EQ, getPaddedSlotWidthBytes(), getRowSize(), getWarpCount(), group_col_widths_, and output_columnar_.

Referenced by QueryExecutionContext::groupBufferToDeinterleavedResults().

931  {
932  auto warp_count = getWarpCount();
933  if (output_columnar_) {
934  CHECK_EQ(size_t(1), group_col_widths_.size());
935  CHECK_EQ(size_t(1), warp_count);
936  return getPaddedSlotWidthBytes(col_idx);
937  }
938 
939  return warp_count * getRowSize();
940 }
#define CHECK_EQ(x, y)
Definition: Logger.h:214
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const
std::vector< int8_t > group_col_widths_

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getColOnlyOffInBytes ( const size_t  col_idx) const

Definition at line 859 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getColOnlyOffInBytes().

Referenced by TargetExprCodegen::codegen(), GroupByAndAggregate::codegenAggColumnPtr(), TargetExprCodegen::codegenAggregate(), getColOffInBytes(), and ResultSetStorage::reduceSingleRow().

859  {
860  return col_slot_context_.getColOnlyOffInBytes(col_idx);
861 }
size_t getColOnlyOffInBytes(const size_t slot_idx) const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

const ColSlotContext& QueryMemoryDescriptor::getColSlotContext ( ) const
inline

Definition at line 291 of file QueryMemoryDescriptor.h.

References col_slot_context_.

Referenced by ResultSetStorage::reduceEntriesNoCollisionsColWise(), and ResultSetReductionJIT::reduceOneEntryTargetsNoCollisions().

291 { return col_slot_context_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getColsSize ( ) const

Definition at line 798 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getAllSlotsAlignedPaddedSize().

Referenced by getBufferSizeBytes(), getRowSize(), and QueryExecutionContext::launchCpuCode().

798  {
800 }
size_t getAllSlotsAlignedPaddedSize() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getCompactByteWidth ( ) const

Definition at line 820 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getCompactByteWidth().

Referenced by anonymous_namespace{TargetExprBuilder.cpp}::get_initial_agg_val(), and init_agg_val_vec().

820  {
822 }
size_t getCompactByteWidth() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

const CountDistinctDescriptor& QueryMemoryDescriptor::getCountDistinctDescriptor ( const size_t  idx) const
inline
size_t QueryMemoryDescriptor::getCountDistinctDescriptorsSize ( ) const
inline
size_t QueryMemoryDescriptor::getEntryCount ( ) const
inline

Definition at line 245 of file QueryMemoryDescriptor.h.

References entry_count_.

Referenced by advance_to_next_columnar_target_buff(), QueryMemoryInitializer::allocateCountDistinctGpuMem(), QueryMemoryInitializer::applyStreamingTopNOffsetCpu(), QueryMemoryInitializer::applyStreamingTopNOffsetGpu(), ResultSetStorage::binSearchRowCount(), anonymous_namespace{QueryMemoryInitializer.cpp}::check_total_bitmap_memory(), ResultSetReductionJIT::codegen(), GroupByAndAggregate::codegenMultiColumnBaselineHash(), GroupByAndAggregate::codegenMultiColumnPerfectHash(), GpuSharedMemCodeBuilder::codegenReduction(), GroupByAndAggregate::codegenWindowRowPointer(), anonymous_namespace{QueryMemoryInitializer.cpp}::compact_projection_buffer_for_cpu_columnar(), QueryMemoryInitializer::compactProjectionBuffersCpu(), QueryMemoryInitializer::compactProjectionBuffersGpu(), copy_group_by_buffers_from_gpu(), create_dev_group_by_buffers(), Executor::createKernels(), ResultSet::entryCount(), Executor::executePlanWithGroupBy(), ResultSetStorage::fillOneEntryColWise(), ResultSetStorage::fillOneEntryRowWise(), anonymous_namespace{ResultSetReduction.cpp}::get_matching_group_value_reduction(), ResultSetStorage::getEntryCount(), getPrependedGroupBufferSizeInBytes(), getPrependedGroupColOffInBytes(), ResultSet::getTargetValueFromBufferColwise(), QueryMemoryInitializer::initColumnarGroups(), QueryMemoryInitializer::initGroupByBuffer(), ResultSetStorage::initializeBaselineValueSlots(), ResultSetStorage::initializeColWise(), ResultSetStorage::initializeRowWise(), inplace_sort_gpu(), QueryExecutionContext::launchGpuCode(), ResultSetStorage::moveEntriesToBuffer(), ResultSetStorage::moveOneEntryToBuffer(), QueryMemoryDescriptor(), QueryMemoryInitializer::QueryMemoryInitializer(), ResultSetStorage::reduce(), ResultSetStorage::reduceOneEntryBaseline(), ResultSetStorage::reduceOneEntrySlotsBaseline(), ResultSetStorage::reduceOneSlotBaseline(), and ResultSetStorage::rewriteAggregateBufferOffsets().

245 { return entry_count_; }

+ Here is the caller graph for this function:

const Executor* QueryMemoryDescriptor::getExecutor ( ) const
inline

Definition at line 167 of file QueryMemoryDescriptor.h.

References executor_.

Referenced by anonymous_namespace{Execute.cpp}::build_row_for_empty_input(), ResultSetReductionJIT::codegen(), anonymous_namespace{Execute.cpp}::fill_entries_for_empty_input(), ResultSet::getVarlenOrderEntry(), ResultSet::makeGeoTargetValue(), and ResultSet::makeVarlenTargetValue().

167 { return executor_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getKeyCount ( ) const
inline

Definition at line 281 of file QueryMemoryDescriptor.h.

References getGroupbyColCount(), and keyless_hash_.

Referenced by anonymous_namespace{Execute.cpp}::permute_storage_columnar().

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

const int8_t QueryMemoryDescriptor::getLogicalSlotWidthBytes ( const size_t  slot_idx) const

Definition at line 1130 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, ColSlotContext::getSlotInfo(), and SlotSize::logical_size.

Referenced by QueryMemoryInitializer::allocateCountDistinctBuffers(), QueryMemoryInitializer::allocateTDigests(), TargetExprCodegen::codegenAggregate(), and ResultSet::getTargetValueFromBufferRowwise().

1131  {
1132  return col_slot_context_.getSlotInfo(slot_idx).logical_size;
1133 }
int8_t logical_size
const SlotSize & getSlotInfo(const size_t slot_idx) const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int64_t QueryMemoryDescriptor::getMaxVal ( ) const
inline

Definition at line 249 of file QueryMemoryDescriptor.h.

References max_val_.

Referenced by GroupByAndAggregate::codegenGroupBy().

249 { return max_val_; }

+ Here is the caller graph for this function:

int64_t QueryMemoryDescriptor::getMinVal ( ) const
inline

Definition at line 248 of file QueryMemoryDescriptor.h.

References min_val_.

Referenced by GroupByAndAggregate::codegenSingleColumnPerfectHash().

248 { return min_val_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getNextColOffInBytes ( const int8_t *  col_ptr,
const size_t  bin,
const size_t  col_idx 
) const

Definition at line 942 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), CHECK, CHECK_EQ, entry_count_, getPaddedSlotWidthBytes(), getSlotCount(), getWarpCount(), group_col_widths_, and output_columnar_.

944  {
946  size_t offset{0};
947  auto warp_count = getWarpCount();
948  const auto chosen_bytes = getPaddedSlotWidthBytes(col_idx);
949  const auto total_slot_count = getSlotCount();
950  if (col_idx + 1 == total_slot_count) {
951  if (output_columnar_) {
952  return (entry_count_ - bin) * chosen_bytes;
953  } else {
954  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
955  }
956  }
957 
958  const auto next_chosen_bytes = getPaddedSlotWidthBytes(col_idx + 1);
959  if (output_columnar_) {
960  CHECK_EQ(size_t(1), group_col_widths_.size());
961  CHECK_EQ(size_t(1), warp_count);
962 
963  offset = align_to_int64(entry_count_ * chosen_bytes);
964 
965  offset += bin * (next_chosen_bytes - chosen_bytes);
966  return offset;
967  }
968 
969  if (next_chosen_bytes == sizeof(int64_t)) {
970  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
971  } else {
972  return chosen_bytes;
973  }
974 }
#define CHECK_EQ(x, y)
Definition: Logger.h:214
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const
#define CHECK(condition)
Definition: Logger.h:206
std::vector< int8_t > group_col_widths_
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getNextColOffInBytesRowOnly ( const int8_t *  col_ptr,
const size_t  col_idx 
) const

Definition at line 976 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), getPaddedSlotWidthBytes(), and getSlotCount().

Referenced by QueryMemoryInitializer::initColumnsPerRow().

977  {
978  const auto chosen_bytes = getPaddedSlotWidthBytes(col_idx);
979  const auto total_slot_count = getSlotCount();
980  if (col_idx + 1 == total_slot_count) {
981  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
982  }
983 
984  const auto next_chosen_bytes = getPaddedSlotWidthBytes(col_idx + 1);
985 
986  if (next_chosen_bytes == sizeof(int64_t)) {
987  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
988  } else {
989  return chosen_bytes;
990  }
991 }
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getPaddedColWidthForRange ( const size_t  offset,
const size_t  range 
) const
inline

Definition at line 208 of file QueryMemoryDescriptor.h.

References getPaddedSlotWidthBytes(), and i.

Referenced by result_set::get_byteoff_of_slot(), and ResultSet::makeGeoTargetValue().

208  {
209  size_t ret = 0;
210  for (size_t i = offset; i < offset + range; i++) {
211  ret += static_cast<size_t>(getPaddedSlotWidthBytes(i));
212  }
213  return ret;
214  }
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

const int8_t QueryMemoryDescriptor::getPaddedSlotWidthBytes ( const size_t  slot_idx) const

Definition at line 1126 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, ColSlotContext::getSlotInfo(), and SlotSize::padded_size.

Referenced by advance_target_ptr_row_wise(), advance_to_next_columnar_target_buff(), TargetExprCodegen::codegen(), anonymous_namespace{GpuSharedMemoryUtils.cpp}::codegen_smem_dest_slot_ptr(), TargetExprCodegen::codegenAggregate(), GroupByAndAggregate::codegenOutputSlot(), compact_init_vals(), anonymous_namespace{QueryMemoryInitializer.cpp}::compact_projection_buffer_for_cpu_columnar(), copy_projection_buffer_from_gpu_columnar(), ResultSet::copyColumnIntoBuffer(), Executor::executePlanWithoutGroupBy(), result_set::get_width_for_slot(), getColOffInBytes(), getColOffInBytesInNextBin(), getNextColOffInBytes(), getNextColOffInBytesRowOnly(), getPaddedColWidthForRange(), ResultSet::getPaddedSlotWidthBytes(), ResultSet::getTargetValueFromBufferColwise(), ResultSet::getTargetValueFromBufferRowwise(), init_agg_val_vec(), QueryMemoryInitializer::initColumnarGroups(), QueryMemoryInitializer::initColumnsPerRow(), inplace_sort_gpu(), ResultSetReductionJIT::isEmpty(), ResultSetStorage::isEmptyEntry(), ResultSetStorage::isEmptyEntryColumnar(), ResultSet::makeGeoTargetValue(), TargetExprCodegenBuilder::operator()(), anonymous_namespace{Execute.cpp}::permute_storage_columnar(), ResultSetStorage::reduceEntriesNoCollisionsColWise(), ResultSetReductionJIT::reduceOneAggregateSlot(), ResultSetReductionJIT::reduceOneEntryTargetsNoCollisions(), ResultSetStorage::reduceOneSlot(), ResultSetStorage::reduceSingleRow(), and ResultSetStorage::rewriteAggregateBufferOffsets().

1126  {
1127  return col_slot_context_.getSlotInfo(slot_idx).padded_size;
1128 }
const SlotSize & getSlotInfo(const size_t slot_idx) const
int8_t padded_size

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getPrependedGroupBufferSizeInBytes ( ) const

Definition at line 920 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), CHECK, getEntryCount(), getGroupbyColCount(), groupColWidth(), and output_columnar_.

Referenced by getColOffInBytes().

920  {
922  size_t buffer_size{0};
923  for (size_t group_idx = 0; group_idx < getGroupbyColCount(); group_idx++) {
924  buffer_size += align_to_int64(
925  std::max(groupColWidth(group_idx), static_cast<int8_t>(sizeof(int64_t))) *
926  getEntryCount());
927  }
928  return buffer_size;
929 }
int8_t groupColWidth(const size_t key_idx) const
size_t getGroupbyColCount() const
#define CHECK(condition)
Definition: Logger.h:206
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getPrependedGroupColOffInBytes ( const size_t  group_idx) const

Definition at line 902 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), CHECK, getEntryCount(), getGroupbyColCount(), groupColWidth(), and output_columnar_.

Referenced by ResultSetStorage::copyKeyColWise(), ResultSetStorage::isEmptyEntryColumnar(), and anonymous_namespace{Execute.cpp}::permute_storage_columnar().

903  {
905  CHECK(group_idx < getGroupbyColCount());
906  size_t offset{0};
907  for (size_t col_idx = 0; col_idx < group_idx; col_idx++) {
908  // TODO(Saman): relax that int64_bit part immediately
909  offset += align_to_int64(
910  std::max(groupColWidth(col_idx), static_cast<int8_t>(sizeof(int64_t))) *
911  getEntryCount());
912  }
913  return offset;
914 }
int8_t groupColWidth(const size_t key_idx) const
size_t getGroupbyColCount() const
#define CHECK(condition)
Definition: Logger.h:206
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

QueryDescriptionType QueryMemoryDescriptor::getQueryDescriptionType ( ) const
inline

Definition at line 169 of file QueryMemoryDescriptor.h.

References query_desc_type_.

Referenced by ResultSetStorage::binSearchRowCount(), ResultSetReductionJIT::codegen(), GroupByAndAggregate::codegen(), GpuReductionHelperJIT::codegen(), GroupByAndAggregate::codegenAggCalls(), GroupByAndAggregate::codegenAggColumnPtr(), GroupByAndAggregate::codegenGroupBy(), GroupByAndAggregate::codegenMultiColumnPerfectHash(), GroupByAndAggregate::codegenOutputSlot(), Executor::collectAllDeviceResults(), copy_projection_buffer_from_gpu_columnar(), Executor::createKernels(), ResultSet::getQueryDescriptionType(), GpuReductionHelperJIT::GpuReductionHelperJIT(), GpuSharedMemCodeBuilder::GpuSharedMemCodeBuilder(), init_agg_val_vec(), QueryMemoryInitializer::initColumnarGroups(), anonymous_namespace{TargetExprBuilder.cpp}::is_columnar_projection(), ResultSetReductionJIT::isEmpty(), ResultSetStorage::isEmptyEntry(), ResultSetStorage::isEmptyEntryColumnar(), isSingleColumnGroupByWithPerfectHash(), QueryExecutionContext::launchCpuCode(), QueryExecutionContext::launchGpuCode(), ResultSetStorage::moveEntriesToBuffer(), TargetExprCodegenBuilder::operator()(), ResultSetStorage::reduce(), Executor::reduceMultiDeviceResultSets(), ResultSetStorage::reduceOneEntryBaseline(), ResultSetReductionJIT::reduceOneEntryBaselineIdx(), ResultSetReductionJIT::reduceOneEntryNoCollisionsIdx(), ExecutionKernel::run(), ExecutionKernel::runImpl(), target_exprs_to_infos(), and ResultSet::updateStorageEntryCount().

169 { return query_desc_type_; }
QueryDescriptionType query_desc_type_

+ Here is the caller graph for this function:

std::unique_ptr< QueryExecutionContext > QueryMemoryDescriptor::getQueryExecutionContext ( const RelAlgExecutionUnit ra_exe_unit,
const Executor executor,
const ExecutorDeviceType  device_type,
const ExecutorDispatchMode  dispatch_mode,
const int  device_id,
const int64_t  num_rows,
const std::vector< std::vector< const int8_t * >> &  col_buffers,
const std::vector< std::vector< uint64_t >> &  frag_offsets,
std::shared_ptr< RowSetMemoryOwner row_set_mem_owner,
const bool  output_columnar,
const bool  sort_on_gpu,
const size_t  thread_idx,
RenderInfo render_info 
) const

Definition at line 676 of file QueryMemoryDescriptor.cpp.

References DEBUG_TIMER, and QueryExecutionContext.

Referenced by ExecutionKernel::runImpl().

689  {
690  auto timer = DEBUG_TIMER(__func__);
691  if (frag_offsets.empty()) {
692  return nullptr;
693  }
694  return std::unique_ptr<QueryExecutionContext>(
695  new QueryExecutionContext(ra_exe_unit,
696  *this,
697  executor,
698  device_type,
699  dispatch_mode,
700  device_id,
701  num_rows,
702  col_buffers,
703  frag_offsets,
704  row_set_mem_owner,
705  output_columnar,
706  sort_on_gpu,
707  thread_idx,
708  render_info));
709 }
#define DEBUG_TIMER(name)
Definition: Logger.h:322
void sort_on_gpu(int64_t *val_buff, int32_t *idx_buff, const uint64_t entry_count, const bool desc, const uint32_t chosen_bytes, ThrustAllocator &alloc)

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getRowSize ( ) const

Definition at line 802 of file QueryMemoryDescriptor.cpp.

References align_to_int64(), CHECK, getColsSize(), getEffectiveKeyWidth(), group_col_widths_, GroupByPerfectHash, keyless_hash_, output_columnar_, and query_desc_type_.

Referenced by QueryMemoryInitializer::applyStreamingTopNOffsetCpu(), QueryMemoryInitializer::applyStreamingTopNOffsetGpu(), ResultSetLogicalValuesBuilder::build(), GroupByAndAggregate::codegenGroupBy(), GroupByAndAggregate::codegenOutputSlot(), GroupByAndAggregate::codegenWindowRowPointer(), QueryMemoryInitializer::copyGroupByBuffersFromGpu(), create_dev_group_by_buffers(), getBufferSizeBytes(), getColOffInBytesInNextBin(), QueryMemoryInitializer::initRowGroups(), anonymous_namespace{Execute.cpp}::permute_storage_row_wise(), QueryMemoryDescriptor(), QueryMemoryInitializer::QueryMemoryInitializer(), and ResultSetStorage::reduceSingleRow().

802  {
804  size_t total_bytes{0};
805  if (keyless_hash_) {
806  // ignore, there's no group column in the output buffer
808  } else {
809  total_bytes += group_col_widths_.size() * getEffectiveKeyWidth();
810  total_bytes = align_to_int64(total_bytes);
811  }
812  total_bytes += getColsSize();
813  return align_to_int64(total_bytes);
814 }
size_t getEffectiveKeyWidth() const
QueryDescriptionType query_desc_type_
#define CHECK(condition)
Definition: Logger.h:206
std::vector< int8_t > group_col_widths_
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getRowWidth ( ) const

Definition at line 1146 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getAllSlotsPaddedSize().

Referenced by get_row_bytes().

1146  {
1147  // Note: Actual row size may include padding (see ResultSetBufferAccessors.h)
1149 }
size_t getAllSlotsPaddedSize() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getSlotCount ( ) const
const int8_t QueryMemoryDescriptor::getSlotIndexForSingleSlotCol ( const size_t  col_idx) const

Definition at line 1135 of file QueryMemoryDescriptor.cpp.

References CHECK_EQ, col_slot_context_, and ColSlotContext::getSlotsForCol().

Referenced by QueryMemoryInitializer::allocateCountDistinctBuffers(), and QueryMemoryInitializer::allocateTDigests().

1136  {
1137  const auto& col_slots = col_slot_context_.getSlotsForCol(col_idx);
1138  CHECK_EQ(col_slots.size(), size_t(1));
1139  return col_slots.front();
1140 }
#define CHECK_EQ(x, y)
Definition: Logger.h:214
const std::vector< size_t > & getSlotsForCol(const size_t col_idx) const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

int64_t QueryMemoryDescriptor::getTargetGroupbyIndex ( const size_t  target_idx) const
inline

Definition at line 227 of file QueryMemoryDescriptor.h.

References CHECK_LT, and target_groupby_indices_.

Referenced by ResultSet::getTargetValueFromBufferColwise(), ResultSet::getTargetValueFromBufferRowwise(), ResultSetReductionJIT::reduceOneEntryBaseline(), ResultSetStorage::reduceOneEntrySlotsBaseline(), ResultSetReductionJIT::reduceOneEntryTargetsNoCollisions(), ResultSetReductionJIT::reduceOneSlot(), ResultSetStorage::reduceOneSlot(), and reductionKey().

227  {
228  CHECK_LT(target_idx, target_groupby_indices_.size());
229  return target_groupby_indices_[target_idx];
230  }
std::vector< int64_t > target_groupby_indices_
#define CHECK_LT(x, y)
Definition: Logger.h:216

+ Here is the caller graph for this function:

int32_t QueryMemoryDescriptor::getTargetIdxForKey ( ) const
inline

Definition at line 182 of file QueryMemoryDescriptor.h.

References idx_target_as_key_.

Referenced by ResultSetReductionJIT::isEmpty(), ResultSetStorage::isEmptyEntry(), ResultSetStorage::isEmptyEntryColumnar(), ResultSetStorage::reduceSingleRow(), and reductionKey().

182 { return idx_target_as_key_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getTotalBytesOfColumnarBuffers ( ) const
private

Returns the maximum total number of bytes (including required paddings) to store all non-lazy columns' results for columnar cases.

Definition at line 829 of file QueryMemoryDescriptor.cpp.

References CHECK, col_slot_context_, entry_count_, ColSlotContext::getTotalBytesOfColumnarBuffers(), and output_columnar_.

Referenced by getBufferSizeBytes(), and getTotalBytesOfColumnarProjections().

829  {
832 }
size_t getTotalBytesOfColumnarBuffers(const size_t entry_count) const
#define CHECK(condition)
Definition: Logger.h:206

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getTotalBytesOfColumnarBuffers ( const size_t  num_entries_per_column) const
private

This is a helper function that returns the total number of bytes (including required paddings) to store all non-lazy columns' results for columnar cases.

Definition at line 838 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getTotalBytesOfColumnarBuffers().

839  {
840  return col_slot_context_.getTotalBytesOfColumnarBuffers(num_entries_per_column);
841 }
size_t getTotalBytesOfColumnarBuffers(const size_t entry_count) const

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getTotalBytesOfColumnarProjections ( const size_t  projection_count) const
private

Returns the effective total number of bytes from columnar projections, which includes 1) total number of bytes used to store all non-lazy columns 2) total number of bytes used to store row indices (for lazy fetches, etc.)

NOTE: this function does not represent the buffer sizes dedicated for the results, but the required memory to fill all valid results into a compact new buffer (with no holes in it)

Definition at line 852 of file QueryMemoryDescriptor.cpp.

References getTotalBytesOfColumnarBuffers().

853  {
854  constexpr size_t row_index_width = sizeof(int64_t);
855  return getTotalBytesOfColumnarBuffers(projection_count) +
856  row_index_width * projection_count;
857 }
size_t getTotalBytesOfColumnarBuffers() const

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getWarpCount ( ) const

Definition at line 816 of file QueryMemoryDescriptor.cpp.

References executor_, and interleaved_bins_on_gpu_.

Referenced by getColOffInBytes(), getColOffInBytesInNextBin(), and getNextColOffInBytes().

816  {
817  return (interleaved_bins_on_gpu_ ? executor_->warpSize() : 1);
818 }

+ Here is the caller graph for this function:

int8_t QueryMemoryDescriptor::groupColWidth ( const size_t  key_idx) const
inline

Definition at line 185 of file QueryMemoryDescriptor.h.

References CHECK_LT, and group_col_widths_.

Referenced by ResultSetStorage::copyKeyColWise(), getPrependedGroupBufferSizeInBytes(), getPrependedGroupColOffInBytes(), ResultSetStorage::isEmptyEntryColumnar(), and anonymous_namespace{Execute.cpp}::permute_storage_columnar().

185  {
186  CHECK_LT(key_idx, group_col_widths_.size());
187  return group_col_widths_[key_idx];
188  }
#define CHECK_LT(x, y)
Definition: Logger.h:216
std::vector< int8_t > group_col_widths_

+ Here is the caller graph for this function:

const auto QueryMemoryDescriptor::groupColWidthsBegin ( ) const
inline

Definition at line 192 of file QueryMemoryDescriptor.h.

References group_col_widths_.

192 { return group_col_widths_.begin(); }
std::vector< int8_t > group_col_widths_
const auto QueryMemoryDescriptor::groupColWidthsEnd ( ) const
inline

Definition at line 193 of file QueryMemoryDescriptor.h.

References group_col_widths_.

193 { return group_col_widths_.end(); }
std::vector< int8_t > group_col_widths_
bool QueryMemoryDescriptor::hasInterleavedBinsOnGpu ( ) const
inline

Definition at line 179 of file QueryMemoryDescriptor.h.

References interleaved_bins_on_gpu_.

bool QueryMemoryDescriptor::hasNulls ( ) const
inline

Definition at line 252 of file QueryMemoryDescriptor.h.

References has_nulls_.

Referenced by GroupByAndAggregate::codegenGroupBy().

252 { return has_nulls_; }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::hasVarlenOutput ( ) const
inline

Definition at line 330 of file QueryMemoryDescriptor.h.

References col_slot_context_, and ColSlotContext::hasVarlenOutput().

Referenced by QueryMemoryInitializer::applyStreamingTopNOffsetCpu(), QueryMemoryInitializer::applyStreamingTopNOffsetGpu(), GroupByAndAggregate::codegenVarlenOutputBuffer(), QueryMemoryInitializer::compactProjectionBuffersCpu(), QueryMemoryInitializer::compactProjectionBuffersGpu(), QueryMemoryInitializer::copyGroupByBuffersFromGpu(), QueryExecutionContext::getRowSet(), query_group_by_template_impl(), and QueryMemoryInitializer::QueryMemoryInitializer().

bool hasVarlenOutput() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

std::unique_ptr< QueryMemoryDescriptor > QueryMemoryDescriptor::init ( const Executor executor,
const RelAlgExecutionUnit ra_exe_unit,
const std::vector< InputTableInfo > &  query_infos,
const ColRangeInfo col_range_info,
const KeylessInfo keyless_info,
const bool  allow_multifrag,
const ExecutorDeviceType  device_type,
const int8_t  crt_min_byte_width,
const bool  sort_on_gpu_hint,
const size_t  shard_count,
const size_t  max_groups_buffer_entry_count,
RenderInfo render_info,
const CountDistinctDescriptors  count_distinct_descriptors,
const bool  must_use_baseline_sort,
const bool  output_columnar_hint,
const bool  streaming_top_n_hint 
)
static

Definition at line 240 of file QueryMemoryDescriptor.cpp.

References anonymous_namespace{QueryMemoryDescriptor.cpp}::get_col_byte_widths(), RelAlgExecutionUnit::groupby_exprs, and RelAlgExecutionUnit::target_exprs.

Referenced by GroupByAndAggregate::initQueryMemoryDescriptorImpl().

256  {
257  auto group_col_widths = get_col_byte_widths(ra_exe_unit.groupby_exprs);
258  const bool is_group_by{!group_col_widths.empty()};
259 
260  auto col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, {});
261 
262  const auto min_slot_size = QueryMemoryDescriptor::pick_target_compact_width(
263  ra_exe_unit, query_infos, crt_min_byte_width);
264 
265  col_slot_context.setAllSlotsPaddedSize(min_slot_size);
266  col_slot_context.validate();
267 
268  if (!is_group_by) {
269  CHECK(!must_use_baseline_sort);
270 
271  return std::make_unique<QueryMemoryDescriptor>(
272  executor,
273  ra_exe_unit,
274  query_infos,
275  allow_multifrag,
276  false,
277  false,
278  -1,
279  ColRangeInfo{ra_exe_unit.estimator ? QueryDescriptionType::Estimator
281  0,
282  0,
283  0,
284  false},
285  col_slot_context,
286  std::vector<int8_t>{},
287  /*group_col_compact_width=*/0,
288  std::vector<int64_t>{},
289  /*entry_count=*/1,
290  count_distinct_descriptors,
291  false,
292  output_columnar_hint,
293  render_info && render_info->isPotentialInSituRender(),
294  must_use_baseline_sort,
295  /*use_streaming_top_n=*/false);
296  }
297 
298  size_t entry_count = 1;
299  auto actual_col_range_info = col_range_info;
300  bool interleaved_bins_on_gpu = false;
301  bool keyless_hash = false;
302  bool streaming_top_n = false;
303  int8_t group_col_compact_width = 0;
304  int32_t idx_target_as_key = -1;
305  auto output_columnar = output_columnar_hint;
306  std::vector<int64_t> target_groupby_indices;
307 
308  switch (col_range_info.hash_type_) {
310  if (render_info) {
311  render_info->setInSituDataIfUnset(false);
312  }
313  // keyless hash: whether or not group columns are stored at the beginning of the
314  // output buffer
315  keyless_hash =
316  (!sort_on_gpu_hint ||
318  col_range_info.max, col_range_info.min, col_range_info.bucket)) &&
319  !col_range_info.bucket && !must_use_baseline_sort && keyless_info.keyless;
320 
321  // if keyless, then this target index indicates wheter an entry is empty or not
322  // (acts as a key)
323  idx_target_as_key = keyless_info.target_index;
324 
325  if (group_col_widths.size() > 1) {
326  // col range info max contains the expected cardinality of the output
327  entry_count = static_cast<size_t>(actual_col_range_info.max);
328  actual_col_range_info.bucket = 0;
329  } else {
330  // single column perfect hash
331  entry_count = std::max(
332  GroupByAndAggregate::getBucketedCardinality(col_range_info), int64_t(1));
333  const size_t interleaved_max_threshold{512};
334 
335  if (must_use_baseline_sort) {
336  target_groupby_indices = target_expr_group_by_indices(ra_exe_unit.groupby_exprs,
337  ra_exe_unit.target_exprs);
338  col_slot_context =
339  ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
340  }
341 
342  bool has_varlen_sample_agg = false;
343  for (const auto& target_expr : ra_exe_unit.target_exprs) {
344  if (target_expr->get_contains_agg()) {
345  const auto agg_expr = dynamic_cast<Analyzer::AggExpr*>(target_expr);
346  CHECK(agg_expr);
347  if (agg_expr->get_aggtype() == kSAMPLE &&
348  agg_expr->get_type_info().is_varlen()) {
349  has_varlen_sample_agg = true;
350  break;
351  }
352  }
353  }
354 
355  interleaved_bins_on_gpu = keyless_hash && !has_varlen_sample_agg &&
356  (entry_count <= interleaved_max_threshold) &&
357  (device_type == ExecutorDeviceType::GPU) &&
359  count_distinct_descriptors) &&
360  !output_columnar;
361  }
362  break;
363  }
365  if (render_info) {
366  render_info->setInSituDataIfUnset(false);
367  }
368  entry_count = shard_count
369  ? (max_groups_buffer_entry_count + shard_count - 1) / shard_count
370  : max_groups_buffer_entry_count;
371  target_groupby_indices = target_expr_group_by_indices(ra_exe_unit.groupby_exprs,
372  ra_exe_unit.target_exprs);
373  col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
374 
375  group_col_compact_width =
376  output_columnar ? 8
377  : pick_baseline_key_width(ra_exe_unit, query_infos, executor);
378 
379  actual_col_range_info =
381  break;
382  }
384  CHECK(!must_use_baseline_sort);
385 
386  if (streaming_top_n_hint && use_streaming_top_n(ra_exe_unit, output_columnar)) {
387  streaming_top_n = true;
388  entry_count = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
389  } else {
390  if (ra_exe_unit.use_bump_allocator) {
391  output_columnar = false;
392  entry_count = 0;
393  } else {
394  entry_count = ra_exe_unit.scan_limit
395  ? static_cast<size_t>(ra_exe_unit.scan_limit)
396  : max_groups_buffer_entry_count;
397  }
398  }
399 
400  const auto catalog = executor->getCatalog();
401  CHECK(catalog);
402  target_groupby_indices = executor->plan_state_->allow_lazy_fetch_
403  ? target_expr_proj_indices(ra_exe_unit, *catalog)
404  : std::vector<int64_t>{};
405 
406  col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
407  break;
408  }
409  default:
410  UNREACHABLE() << "Unknown query type";
411  }
412 
413  return std::make_unique<QueryMemoryDescriptor>(
414  executor,
415  ra_exe_unit,
416  query_infos,
417  allow_multifrag,
418  keyless_hash,
419  interleaved_bins_on_gpu,
420  idx_target_as_key,
421  actual_col_range_info,
422  col_slot_context,
423  group_col_widths,
424  group_col_compact_width,
425  target_groupby_indices,
426  entry_count,
427  count_distinct_descriptors,
428  sort_on_gpu_hint,
429  output_columnar,
430  render_info && render_info->isPotentialInSituRender(),
431  must_use_baseline_sort,
432  streaming_top_n);
433 }
std::vector< Analyzer::Expr * > target_exprs
static bool many_entries(const int64_t max_val, const int64_t min_val, const int64_t bucket)
static int64_t getBucketedCardinality(const ColRangeInfo &col_range_info)
std::vector< int64_t > target_expr_proj_indices(const RelAlgExecutionUnit &ra_exe_unit, const Catalog_Namespace::Catalog &cat)
const bool keyless
std::vector< int8_t > get_col_byte_widths(const T &col_expr_list)
bool setInSituDataIfUnset(const bool is_in_situ_data)
Definition: RenderInfo.cpp:98
QueryDescriptionType hash_type_
#define UNREACHABLE()
Definition: Logger.h:250
bool use_streaming_top_n(const RelAlgExecutionUnit &ra_exe_unit, const bool output_columnar)
const std::list< std::shared_ptr< Analyzer::Expr > > groupby_exprs
static int8_t pick_target_compact_width(const RelAlgExecutionUnit &ra_exe_unit, const std::vector< InputTableInfo > &query_infos, const int8_t crt_min_byte_width)
const int32_t target_index
std::vector< int64_t > target_expr_group_by_indices(const std::list< std::shared_ptr< Analyzer::Expr >> &groupby_exprs, const std::vector< Analyzer::Expr * > &target_exprs)
bool isPotentialInSituRender() const
Definition: RenderInfo.cpp:64
#define CHECK(condition)
Definition: Logger.h:206
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)
if(yyssp >=yyss+yystacksize-1)
int8_t pick_baseline_key_width(const RelAlgExecutionUnit &ra_exe_unit, const std::vector< InputTableInfo > &query_infos, const Executor *executor)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::interleavedBins ( const ExecutorDeviceType  device_type) const

Definition at line 1105 of file QueryMemoryDescriptor.cpp.

References GPU, and interleaved_bins_on_gpu_.

Referenced by canOutputColumnar(), GroupByAndAggregate::codegenSingleColumnPerfectHash(), getBufferSizeBytes(), QueryExecutionContext::groupBufferToResults(), QueryMemoryInitializer::initGroupByBuffer(), and QueryMemoryInitializer::QueryMemoryInitializer().

1105  {
1106  return interleaved_bins_on_gpu_ && device_type == ExecutorDeviceType::GPU;
1107 }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::isGroupBy ( ) const
inline

Definition at line 196 of file QueryMemoryDescriptor.h.

References group_col_widths_.

Referenced by anonymous_namespace{TargetExprBuilder.cpp}::get_initial_agg_val(), init_agg_val_vec(), QueryMemoryInitializer::initColumnsPerRow(), QueryExecutionContext::launchCpuCode(), QueryExecutionContext::launchGpuCode(), and QueryMemoryInitializer::QueryMemoryInitializer().

196 { return !group_col_widths_.empty(); }
std::vector< int8_t > group_col_widths_

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::isLogicalSizedColumnsAllowed ( ) const

Definition at line 1055 of file QueryMemoryDescriptor.cpp.

References g_cluster, output_columnar_, Projection, and query_desc_type_.

Referenced by TargetExprCodegen::codegenAggregate(), TargetExprCodegenBuilder::codegenMultiSlotSampleExpressions(), TargetExprCodegenBuilder::codegenSlotEmptyKey(), init_agg_val_vec(), ResultSet::makeTargetValue(), QueryMemoryDescriptor(), ResultSetStorage::reduceOneSlot(), ResultSetStorage::reduceOneSlotSingleValue(), and setOutputColumnar().

1055  {
1056  // In distributed mode, result sets are serialized using rowwise iterators, so we use
1057  // consistent slot widths for now
1058  return output_columnar_ && !g_cluster &&
1060 }
QueryDescriptionType query_desc_type_
bool g_cluster

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::isSingleColumnGroupByWithPerfectHash ( ) const
inline

Definition at line 171 of file QueryMemoryDescriptor.h.

References getGroupbyColCount(), getQueryDescriptionType(), and GroupByPerfectHash.

Referenced by GroupByAndAggregate::codegenGroupBy(), and ResultSet::getTargetValueFromBufferRowwise().

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::isWarpSyncRequired ( const ExecutorDeviceType  device_type) const

Definition at line 1110 of file QueryMemoryDescriptor.cpp.

References executor_, and GPU.

Referenced by query_group_by_template_impl().

1111  {
1112  if (device_type == ExecutorDeviceType::GPU) {
1113  return executor_->cudaMgr()->isArchVoltaOrGreaterForAll();
1114  }
1115  return false;
1116 }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::lazyInitGroups ( const ExecutorDeviceType  device_type) const

Definition at line 1100 of file QueryMemoryDescriptor.cpp.

References count_distinct_descriptors_, countDescriptorsLogicallyEmpty(), GPU, and render_output_.

Referenced by create_dev_group_by_buffers(), QueryMemoryInitializer::QueryMemoryInitializer(), and toString().

1100  {
1101  return device_type == ExecutorDeviceType::GPU && !render_output_ &&
1103 }
CountDistinctDescriptors count_distinct_descriptors_
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

static bool QueryMemoryDescriptor::many_entries ( const int64_t  max_val,
const int64_t  min_val,
const int64_t  bucket 
)
inlinestatic

Definition at line 143 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory().

145  {
146  return max_val - min_val > 10000 * std::max(bucket, int64_t(1));
147  }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::mustUseBaselineSort ( ) const
inline

Definition at line 272 of file QueryMemoryDescriptor.h.

References must_use_baseline_sort_.

Referenced by GroupByAndAggregate::codegenSingleColumnPerfectHash().

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::operator== ( const QueryMemoryDescriptor other) const

Definition at line 612 of file QueryMemoryDescriptor.cpp.

References bucket_, col_slot_context_, count_distinct_descriptors_, force_4byte_float_, group_col_compact_width_, group_col_widths_, has_nulls_, i, idx_target_as_key_, interleaved_bins_on_gpu_, keyless_hash_, max_val_, min_val_, output_columnar_, query_desc_type_, sort_on_gpu_, and target_groupby_indices_.

612  {
613  // Note that this method does not check ptr reference members (e.g. executor_) or
614  // entry_count_
615  if (query_desc_type_ != other.query_desc_type_) {
616  return false;
617  }
618  if (keyless_hash_ != other.keyless_hash_) {
619  return false;
620  }
622  return false;
623  }
624  if (idx_target_as_key_ != other.idx_target_as_key_) {
625  return false;
626  }
627  if (force_4byte_float_ != other.force_4byte_float_) {
628  return false;
629  }
630  if (group_col_widths_ != other.group_col_widths_) {
631  return false;
632  }
634  return false;
635  }
637  return false;
638  }
639  if (min_val_ != other.min_val_) {
640  return false;
641  }
642  if (max_val_ != other.max_val_) {
643  return false;
644  }
645  if (bucket_ != other.bucket_) {
646  return false;
647  }
648  if (has_nulls_ != other.has_nulls_) {
649  return false;
650  }
652  return false;
653  } else {
654  // Count distinct descriptors can legitimately differ in device only.
655  for (size_t i = 0; i < count_distinct_descriptors_.size(); ++i) {
656  auto ref_count_distinct_desc = other.count_distinct_descriptors_[i];
657  auto count_distinct_desc = count_distinct_descriptors_[i];
658  count_distinct_desc.device_type = ref_count_distinct_desc.device_type;
659  if (ref_count_distinct_desc != count_distinct_desc) {
660  return false;
661  }
662  }
663  }
664  if (sort_on_gpu_ != other.sort_on_gpu_) {
665  return false;
666  }
667  if (output_columnar_ != other.output_columnar_) {
668  return false;
669  }
670  if (col_slot_context_ != other.col_slot_context_) {
671  return false;
672  }
673  return true;
674 }
std::vector< int64_t > target_groupby_indices_
CountDistinctDescriptors count_distinct_descriptors_
QueryDescriptionType query_desc_type_
std::vector< int8_t > group_col_widths_
int8_t QueryMemoryDescriptor::pick_target_compact_width ( const RelAlgExecutionUnit ra_exe_unit,
const std::vector< InputTableInfo > &  query_infos,
const int8_t  crt_min_byte_width 
)
static

Definition at line 711 of file QueryMemoryDescriptor.cpp.

References CHECK, CHECK_EQ, g_bigint_count, anonymous_namespace{QueryMemoryDescriptor.cpp}::get_col_byte_widths(), Analyzer::UOper::get_operand(), Analyzer::Expr::get_type_info(), RelAlgExecutionUnit::groupby_exprs, RelAlgExecutionUnit::input_col_descs, anonymous_namespace{QueryMemoryDescriptor.cpp}::is_int_and_no_bigger_than(), kCOUNT, kENCODING_DICT, kUNNEST, and RelAlgExecutionUnit::target_exprs.

714  {
715  if (g_bigint_count) {
716  return sizeof(int64_t);
717  }
718  int8_t compact_width{0};
719  auto col_it = ra_exe_unit.input_col_descs.begin();
720  int unnest_array_col_id{std::numeric_limits<int>::min()};
721  for (const auto& groupby_expr : ra_exe_unit.groupby_exprs) {
722  const auto uoper = dynamic_cast<Analyzer::UOper*>(groupby_expr.get());
723  if (uoper && uoper->get_optype() == kUNNEST) {
724  const auto& arg_ti = uoper->get_operand()->get_type_info();
725  CHECK(arg_ti.is_array());
726  const auto& elem_ti = arg_ti.get_elem_type();
727  if (elem_ti.is_string() && elem_ti.get_compression() == kENCODING_DICT) {
728  unnest_array_col_id = (*col_it)->getColId();
729  } else {
730  compact_width = crt_min_byte_width;
731  break;
732  }
733  }
734  ++col_it;
735  }
736  if (!compact_width &&
737  (ra_exe_unit.groupby_exprs.size() != 1 || !ra_exe_unit.groupby_exprs.front())) {
738  compact_width = crt_min_byte_width;
739  }
740  if (!compact_width) {
741  col_it = ra_exe_unit.input_col_descs.begin();
742  std::advance(col_it, ra_exe_unit.groupby_exprs.size());
743  for (const auto target : ra_exe_unit.target_exprs) {
744  const auto& ti = target->get_type_info();
745  const auto agg = dynamic_cast<const Analyzer::AggExpr*>(target);
746  if (agg && agg->get_arg()) {
747  compact_width = crt_min_byte_width;
748  break;
749  }
750 
751  if (agg) {
752  CHECK_EQ(kCOUNT, agg->get_aggtype());
753  CHECK(!agg->get_is_distinct());
754  ++col_it;
755  continue;
756  }
757 
758  if (is_int_and_no_bigger_than(ti, 4) ||
759  (ti.is_string() && ti.get_compression() == kENCODING_DICT)) {
760  ++col_it;
761  continue;
762  }
763 
764  const auto uoper = dynamic_cast<Analyzer::UOper*>(target);
765  if (uoper && uoper->get_optype() == kUNNEST &&
766  (*col_it)->getColId() == unnest_array_col_id) {
767  const auto arg_ti = uoper->get_operand()->get_type_info();
768  CHECK(arg_ti.is_array());
769  const auto& elem_ti = arg_ti.get_elem_type();
770  if (elem_ti.is_string() && elem_ti.get_compression() == kENCODING_DICT) {
771  ++col_it;
772  continue;
773  }
774  }
775 
776  compact_width = crt_min_byte_width;
777  break;
778  }
779  }
780  if (!compact_width) {
781  size_t total_tuples{0};
782  for (const auto& qi : query_infos) {
783  total_tuples += qi.info.getNumTuples();
784  }
785  return total_tuples <= static_cast<size_t>(std::numeric_limits<uint32_t>::max()) ||
786  unnest_array_col_id != std::numeric_limits<int>::min()
787  ? 4
788  : crt_min_byte_width;
789  } else {
790  // TODO(miyu): relax this condition to allow more cases just w/o padding
791  for (auto wid : get_col_byte_widths(ra_exe_unit.target_exprs)) {
792  compact_width = std::max(compact_width, wid);
793  }
794  return compact_width;
795  }
796 }
std::vector< Analyzer::Expr * > target_exprs
#define CHECK_EQ(x, y)
Definition: Logger.h:214
std::vector< int8_t > get_col_byte_widths(const T &col_expr_list)
const std::list< std::shared_ptr< Analyzer::Expr > > groupby_exprs
bool g_bigint_count
const SQLTypeInfo & get_type_info() const
Definition: Analyzer.h:78
const Expr * get_operand() const
Definition: Analyzer.h:371
Definition: sqldefs.h:76
bool is_int_and_no_bigger_than(const SQLTypeInfo &ti, const size_t byte_width)
#define CHECK(condition)
Definition: Logger.h:206
std::list< std::shared_ptr< const InputColDescriptor > > input_col_descs

+ Here is the call graph for this function:

std::string QueryMemoryDescriptor::queryDescTypeToString ( ) const

Definition at line 1175 of file QueryMemoryDescriptor.cpp.

References Estimator, GroupByBaselineHash, GroupByPerfectHash, NonGroupedAggregate, Projection, query_desc_type_, and UNREACHABLE.

Referenced by reductionKey().

1175  {
1176  switch (query_desc_type_) {
1178  return "Perfect Hash";
1180  return "Baseline Hash";
1182  return "Projection";
1184  return "Non-grouped Aggregate";
1186  return "Estimator";
1187  default:
1188  UNREACHABLE();
1189  }
1190  return "";
1191 }
#define UNREACHABLE()
Definition: Logger.h:250
QueryDescriptionType query_desc_type_

+ Here is the caller graph for this function:

std::string QueryMemoryDescriptor::reductionKey ( ) const

Definition at line 1216 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, getEffectiveKeyWidth(), getGroupbyColCount(), getTargetGroupbyIndex(), getTargetIdxForKey(), join(), keyless_hash_, queryDescTypeToString(), targetGroupbyIndicesSize(), to_string(), ColSlotContext::toString(), and toString().

Referenced by ResultSetReductionJIT::cacheKey(), and toString().

1216  {
1217  std::string str;
1218  str += "Query Memory Descriptor State\n";
1219  str += "\tQuery Type: " + queryDescTypeToString() + "\n";
1220  str +=
1221  "\tKeyless Hash: " + ::toString(keyless_hash_) +
1222  (keyless_hash_ ? ", target index for key: " + std::to_string(getTargetIdxForKey())
1223  : "") +
1224  "\n";
1225  str += "\tEffective key width: " + std::to_string(getEffectiveKeyWidth()) + "\n";
1226  str += "\tNumber of group columns: " + std::to_string(getGroupbyColCount()) + "\n";
1227  const auto group_indices_size = targetGroupbyIndicesSize();
1228  if (group_indices_size) {
1229  std::vector<std::string> group_indices_strings;
1230  for (size_t target_idx = 0; target_idx < group_indices_size; ++target_idx) {
1231  group_indices_strings.push_back(std::to_string(getTargetGroupbyIndex(target_idx)));
1232  }
1233  str += "\tTarget group by indices: " +
1234  boost::algorithm::join(group_indices_strings, ",") + "\n";
1235  }
1236  str += "\t" + col_slot_context_.toString();
1237  return str;
1238 }
int64_t getTargetGroupbyIndex(const size_t target_idx) const
std::string toString() const
std::string join(T const &container, std::string const &delim)
size_t getEffectiveKeyWidth() const
std::string to_string(char const *&&v)
size_t getGroupbyColCount() const
size_t targetGroupbyIndicesSize() const
std::string toString() const
std::string queryDescTypeToString() const
int32_t getTargetIdxForKey() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

void QueryMemoryDescriptor::resetGroupColWidths ( const std::vector< int8_t > &  new_group_col_widths)
inlineprotected

Definition at line 345 of file QueryMemoryDescriptor.h.

References group_col_widths_.

345  {
346  group_col_widths_ = new_group_col_widths;
347  }
std::vector< int8_t > group_col_widths_
void QueryMemoryDescriptor::setAllTargetGroupbyIndices ( std::vector< int64_t >  group_by_indices)
inline

Definition at line 232 of file QueryMemoryDescriptor.h.

References target_groupby_indices_.

232  {
233  target_groupby_indices_ = group_by_indices;
234  }
std::vector< int64_t > target_groupby_indices_
void QueryMemoryDescriptor::setEntryCount ( const size_t  val)
inline

Definition at line 246 of file QueryMemoryDescriptor.h.

References entry_count_.

Referenced by Executor::executePlanWithGroupBy(), Executor::reduceMultiDeviceResultSets(), ResultSetStorage::updateEntryCount(), and ResultSet::updateStorageEntryCount().

246 { entry_count_ = val; }

+ Here is the caller graph for this function:

void QueryMemoryDescriptor::setForceFourByteFloat ( const bool  val)
inline

Definition at line 277 of file QueryMemoryDescriptor.h.

References force_4byte_float_.

void QueryMemoryDescriptor::setGroupColCompactWidth ( const int8_t  val)
inline

Definition at line 198 of file QueryMemoryDescriptor.h.

References group_col_compact_width_.

void QueryMemoryDescriptor::setHasInterleavedBinsOnGpu ( const bool  val)
inline

Definition at line 180 of file QueryMemoryDescriptor.h.

References interleaved_bins_on_gpu_.

void QueryMemoryDescriptor::setHasKeylessHash ( const bool  val)
inline

Definition at line 177 of file QueryMemoryDescriptor.h.

References keyless_hash_.

void QueryMemoryDescriptor::setOutputColumnar ( const bool  val)

Definition at line 1044 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, isLogicalSizedColumnsAllowed(), output_columnar_, and ColSlotContext::setAllSlotsPaddedSizeToLogicalSize().

Referenced by QueryOutputBufferMemoryManager::allocate_output_buffers(), Executor::executeTableFunction(), and TableFunctionExecutionContext::launchGpuCode().

1044  {
1045  output_columnar_ = val;
1048  }
1049 }
bool isLogicalSizedColumnsAllowed() const
void setAllSlotsPaddedSizeToLogicalSize()

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

void QueryMemoryDescriptor::setQueryDescriptionType ( const QueryDescriptionType  val)
inline

Definition at line 170 of file QueryMemoryDescriptor.h.

References query_desc_type_.

170 { query_desc_type_ = val; }
QueryDescriptionType query_desc_type_
void QueryMemoryDescriptor::setTargetIdxForKey ( const int32_t  val)
inline

Definition at line 183 of file QueryMemoryDescriptor.h.

References idx_target_as_key_.

bool QueryMemoryDescriptor::slotIsVarlenOutput ( const size_t  slot_idx) const
inline

Definition at line 340 of file QueryMemoryDescriptor.h.

References col_slot_context_, and ColSlotContext::slotIsVarlen().

Referenced by advance_target_ptr_row_wise(), and ResultSet::makeGeoTargetValue().

340  {
341  return col_slot_context_.slotIsVarlen(slot_idx);
342  }
bool slotIsVarlen(const size_t slot_idx) const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::sortOnGpu ( ) const
inline

Definition at line 262 of file QueryMemoryDescriptor.h.

References sort_on_gpu_.

Referenced by alignPaddedSlots(), QueryExecutionContext::launchGpuCode(), ExecutionKernel::runImpl(), and use_speculative_top_n().

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::targetGroupbyIndicesSize ( ) const
inline
size_t QueryMemoryDescriptor::targetGroupbyNegativeIndicesSize ( ) const
inline

Definition at line 237 of file QueryMemoryDescriptor.h.

References target_groupby_indices_.

237  {
238  return std::count_if(
239  target_groupby_indices_.begin(),
241  [](const int64_t& target_group_by_index) { return target_group_by_index < 0; });
242  }
std::vector< int64_t > target_groupby_indices_
bool QueryMemoryDescriptor::threadsShareMemory ( ) const
std::string QueryMemoryDescriptor::toString ( ) const

Definition at line 1193 of file QueryMemoryDescriptor.cpp.

References allow_multifrag_, blocksShareMemory(), bucket_, entry_count_, GPU, interleaved_bins_on_gpu_, is_table_function_, lazyInitGroups(), max_val_, min_val_, must_use_baseline_sort_, output_columnar_, reductionKey(), render_output_, sort_on_gpu_, threadsShareMemory(), to_string(), use_streaming_top_n_, and usesGetGroupValueFast().

Referenced by Executor::createKernels(), and reductionKey().

1193  {
1194  auto str = reductionKey();
1195  str += "\tAllow Multifrag: " + ::toString(allow_multifrag_) + "\n";
1196  str += "\tInterleaved Bins on GPU: " + ::toString(interleaved_bins_on_gpu_) + "\n";
1197  str += "\tBlocks Share Memory: " + ::toString(blocksShareMemory()) + "\n";
1198  str += "\tThreads Share Memory: " + ::toString(threadsShareMemory()) + "\n";
1199  str += "\tUses Fast Group Values: " + ::toString(usesGetGroupValueFast()) + "\n";
1200  str +=
1201  "\tLazy Init Groups (GPU): " + ::toString(lazyInitGroups(ExecutorDeviceType::GPU)) +
1202  "\n";
1203  str += "\tEntry Count: " + std::to_string(entry_count_) + "\n";
1204  str += "\tMin Val (perfect hash only): " + std::to_string(min_val_) + "\n";
1205  str += "\tMax Val (perfect hash only): " + std::to_string(max_val_) + "\n";
1206  str += "\tBucket Val (perfect hash only): " + std::to_string(bucket_) + "\n";
1207  str += "\tSort on GPU: " + ::toString(sort_on_gpu_) + "\n";
1208  str += "\tUse Streaming Top N: " + ::toString(use_streaming_top_n_) + "\n";
1209  str += "\tOutput Columnar: " + ::toString(output_columnar_) + "\n";
1210  str += "\tRender Output: " + ::toString(render_output_) + "\n";
1211  str += "\tUse Baseline Sort: " + ::toString(must_use_baseline_sort_) + "\n";
1212  str += "\tIs Table Function: " + ::toString(is_table_function_) + "\n";
1213  return str;
1214 }
std::string toString() const
std::string to_string(char const *&&v)
bool lazyInitGroups(const ExecutorDeviceType) const
std::string reductionKey() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

static TResultSetBufferDescriptor QueryMemoryDescriptor::toThrift ( const QueryMemoryDescriptor )
static
int8_t QueryMemoryDescriptor::updateActualMinByteWidth ( const int8_t  actual_min_byte_width) const

Definition at line 1151 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getMinPaddedByteSize().

1152  {
1153  return col_slot_context_.getMinPaddedByteSize(actual_min_byte_width);
1154 }
int8_t getMinPaddedByteSize(const int8_t actual_min_byte_width) const

+ Here is the call graph for this function:

void QueryMemoryDescriptor::useConsistentSlotWidthSize ( const int8_t  slot_width_size)

Definition at line 1142 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::setAllSlotsSize().

1142  {
1143  col_slot_context_.setAllSlotsSize(slot_width_size);
1144 }
void setAllSlotsSize(const int8_t slot_width_size)

+ Here is the call graph for this function:

bool QueryMemoryDescriptor::usesGetGroupValueFast ( ) const

Definition at line 1073 of file QueryMemoryDescriptor.cpp.

References getGroupbyColCount(), GroupByPerfectHash, and query_desc_type_.

Referenced by canOutputColumnar(), GroupByAndAggregate::codegen(), GroupByAndAggregate::codegenSingleColumnPerfectHash(), and toString().

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::useStreamingTopN ( ) const
inline

Definition at line 268 of file QueryMemoryDescriptor.h.

References use_streaming_top_n_.

Referenced by GroupByAndAggregate::codegen(), GroupByAndAggregate::codegenOutputSlot(), QueryMemoryInitializer::copyGroupByBuffersFromGpu(), QueryMemoryInitializer::initGroupByBuffer(), QueryExecutionContext::launchCpuCode(), and QueryExecutionContext::launchGpuCode().

+ Here is the caller graph for this function:

std::optional< size_t > QueryMemoryDescriptor::varlenOutputBufferElemSize ( ) const

Definition at line 1256 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, ColSlotContext::getSlotCount(), i, and ColSlotContext::varlenOutputElementSize().

Referenced by TargetExprCodegen::codegenAggregate(), create_dev_group_by_buffers(), QueryExecutionContext::launchGpuCode(), and QueryMemoryInitializer::QueryMemoryInitializer().

1256  {
1257  int64_t buffer_element_size{0};
1258  for (size_t i = 0; i < col_slot_context_.getSlotCount(); i++) {
1259  try {
1260  const auto slot_element_size = col_slot_context_.varlenOutputElementSize(i);
1261  if (slot_element_size < 0) {
1262  return std::nullopt;
1263  }
1264  buffer_element_size += slot_element_size;
1265  } catch (...) {
1266  continue;
1267  }
1268  }
1269  return buffer_element_size;
1270 }
int64_t varlenOutputElementSize(const size_t slot_idx) const
size_t getSlotCount() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::varlenOutputRowSizeToSlot ( const size_t  slot_idx) const

Definition at line 1272 of file QueryMemoryDescriptor.cpp.

References CHECK_LT, col_slot_context_, ColSlotContext::getSlotCount(), i, and ColSlotContext::varlenOutputElementSize().

Referenced by TargetExprCodegen::codegenAggregate().

1272  {
1273  int64_t buffer_element_size{0};
1275  for (size_t i = 0; i < slot_idx; i++) {
1276  try {
1277  const auto slot_element_size = col_slot_context_.varlenOutputElementSize(i);
1278  if (slot_element_size < 0) {
1279  continue;
1280  }
1281  buffer_element_size += slot_element_size;
1282  } catch (...) {
1283  continue;
1284  }
1285  }
1286  return buffer_element_size;
1287 }
int64_t varlenOutputElementSize(const size_t slot_idx) const
#define CHECK_LT(x, y)
Definition: Logger.h:216
size_t getSlotCount() const

+ Here is the call graph for this function:

+ Here is the caller graph for this function:

Friends And Related Function Documentation

friend class QueryExecutionContext
friend

Definition at line 387 of file QueryMemoryDescriptor.h.

Referenced by getQueryExecutionContext().

friend class ResultSet
friend

Definition at line 386 of file QueryMemoryDescriptor.h.

Member Data Documentation

bool QueryMemoryDescriptor::allow_multifrag_
private

Definition at line 351 of file QueryMemoryDescriptor.h.

Referenced by toString().

int64_t QueryMemoryDescriptor::bucket_
private

Definition at line 368 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory(), getBucket(), operator==(), and toString().

size_t QueryMemoryDescriptor::entry_count_
private
const Executor* QueryMemoryDescriptor::executor_
private
bool QueryMemoryDescriptor::force_4byte_float_
private

Definition at line 378 of file QueryMemoryDescriptor.h.

Referenced by forceFourByteFloat(), operator==(), and setForceFourByteFloat().

int8_t QueryMemoryDescriptor::group_col_compact_width_
private
bool QueryMemoryDescriptor::has_nulls_
private

Definition at line 369 of file QueryMemoryDescriptor.h.

Referenced by hasNulls(), and operator==().

int32_t QueryMemoryDescriptor::idx_target_as_key_
private

Definition at line 355 of file QueryMemoryDescriptor.h.

Referenced by getTargetIdxForKey(), operator==(), and setTargetIdxForKey().

bool QueryMemoryDescriptor::interleaved_bins_on_gpu_
private
bool QueryMemoryDescriptor::is_table_function_
private

Definition at line 375 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory(), and toString().

bool QueryMemoryDescriptor::keyless_hash_
private
int64_t QueryMemoryDescriptor::max_val_
private

Definition at line 367 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory(), getMaxVal(), operator==(), and toString().

int64_t QueryMemoryDescriptor::min_val_
private

Definition at line 365 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory(), getMinVal(), operator==(), and toString().

bool QueryMemoryDescriptor::must_use_baseline_sort_
private

Definition at line 374 of file QueryMemoryDescriptor.h.

Referenced by mustUseBaselineSort(), and toString().

bool QueryMemoryDescriptor::render_output_
private

Definition at line 373 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory(), lazyInitGroups(), and toString().

bool QueryMemoryDescriptor::sort_on_gpu_
private

Definition at line 371 of file QueryMemoryDescriptor.h.

Referenced by operator==(), QueryMemoryDescriptor(), sortOnGpu(), and toString().

std::vector<int64_t> QueryMemoryDescriptor::target_groupby_indices_
private
bool QueryMemoryDescriptor::use_streaming_top_n_
private

The documentation for this class was generated from the following files: