OmniSciDB  bf83d84833
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros 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, 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 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
 

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 470 of file QueryMemoryDescriptor.cpp.

References Projection.

471  : executor_(nullptr)
472  , allow_multifrag_(false)
474  , keyless_hash_(false)
475  , interleaved_bins_on_gpu_(false)
476  , idx_target_as_key_(0)
478  , entry_count_(0)
479  , min_val_(0)
480  , max_val_(0)
481  , bucket_(0)
482  , has_nulls_(false)
483  , sort_on_gpu_(false)
484  , output_columnar_(false)
485  , render_output_(false)
486  , must_use_baseline_sort_(false)
487  , is_table_function_(false)
488  , use_streaming_top_n_(false)
489  , 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 375 of file QueryMemoryDescriptor.cpp.

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

395  : executor_(executor)
396  , allow_multifrag_(allow_multifrag)
397  , query_desc_type_(col_range_info.hash_type_)
398  , keyless_hash_(keyless_hash)
399  , interleaved_bins_on_gpu_(interleaved_bins_on_gpu)
400  , idx_target_as_key_(idx_target_as_key)
401  , group_col_widths_(group_col_widths)
402  , group_col_compact_width_(group_col_compact_width)
403  , target_groupby_indices_(target_groupby_indices)
404  , entry_count_(entry_count)
405  , min_val_(col_range_info.min)
406  , max_val_(col_range_info.max)
407  , bucket_(col_range_info.bucket)
408  , has_nulls_(col_range_info.has_nulls)
409  , count_distinct_descriptors_(count_distinct_descriptors)
410  , output_columnar_(false)
411  , render_output_(render_output)
412  , must_use_baseline_sort_(must_use_baseline_sort)
413  , is_table_function_(false)
415  , force_4byte_float_(false)
416  , col_slot_context_(col_slot_context) {
419 
420  sort_on_gpu_ = sort_on_gpu_hint && canOutputColumnar() && !keyless_hash_;
421 
422  if (sort_on_gpu_) {
423  CHECK(!ra_exe_unit.use_bump_allocator);
424  output_columnar_ = true;
425  } else {
426  switch (query_desc_type_) {
428  output_columnar_ = output_columnar_hint;
429  break;
434  break;
436  output_columnar_ = output_columnar_hint;
437  break;
442  break;
443  default:
444  output_columnar_ = false;
445  break;
446  }
447  }
448 
450  // TODO(adb): Ensure fixed size buffer allocations are correct with all logical column
451  // sizes
452  CHECK(!ra_exe_unit.use_bump_allocator);
455  }
456 
457 #ifdef HAVE_CUDA
458  // Check Streaming Top N heap usage, bail if > max slab size, CUDA ONLY
459  if (use_streaming_top_n_ && executor->catalog_->getDataMgr().gpusPresent()) {
460  const auto thread_count = executor->blockSize() * executor->gridSize();
461  const auto total_buff_size =
463  if (total_buff_size > executor_->maxGpuSlabSize()) {
464  throw StreamingTopNOOM(total_buff_size);
465  }
466  }
467 #endif
468 }
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
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:197
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 491 of file QueryMemoryDescriptor.cpp.

495  : executor_(executor)
496  , allow_multifrag_(false)
497  , query_desc_type_(query_desc_type)
498  , keyless_hash_(false)
499  , interleaved_bins_on_gpu_(false)
500  , idx_target_as_key_(0)
502  , entry_count_(entry_count)
503  , min_val_(0)
504  , max_val_(0)
505  , bucket_(0)
506  , has_nulls_(false)
507  , sort_on_gpu_(false)
508  , output_columnar_(false)
509  , render_output_(false)
510  , must_use_baseline_sort_(false)
511  , is_table_function_(is_table_function)
512  , use_streaming_top_n_(false)
513  , 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 515 of file QueryMemoryDescriptor.cpp.

520  : executor_(nullptr)
521  , allow_multifrag_(false)
522  , query_desc_type_(query_desc_type)
523  , keyless_hash_(false)
524  , interleaved_bins_on_gpu_(false)
525  , idx_target_as_key_(0)
526  , group_col_widths_(group_col_widths)
528  , entry_count_(0)
529  , min_val_(min_val)
530  , max_val_(max_val)
531  , bucket_(0)
532  , has_nulls_(false)
533  , sort_on_gpu_(false)
534  , output_columnar_(false)
535  , render_output_(false)
536  , must_use_baseline_sort_(false)
537  , is_table_function_(false)
538  , use_streaming_top_n_(false)
539  , 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 1069 of file QueryMemoryDescriptor.cpp.

References ColSlotContext::addColumn(), and col_slot_context_.

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

1070  {
1071  col_slot_context_.addColumn(slots_for_col);
1072 }
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 1078 of file QueryMemoryDescriptor.cpp.

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

1078  {
1080 }
void alignPaddedSlots(const bool sort_on_gpu)

+ Here is the call graph for this function:

bool QueryMemoryDescriptor::blocksShareMemory ( ) const

Definition at line 992 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().

992  {
993  if (g_cluster || is_table_function_) {
994  return true;
995  }
997  return true;
998  }
999  if (executor_->isCPUOnly() || render_output_ ||
1003  getGroupbyColCount() > 1)) {
1004  return true;
1005  }
1008 }
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 1082 of file QueryMemoryDescriptor.cpp.

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

Referenced by QueryMemoryDescriptor().

1082  {
1086 }
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 193 of file QueryMemoryDescriptor.h.

References group_col_widths_.

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

Definition at line 1074 of file QueryMemoryDescriptor.cpp.

References ColSlotContext::clear(), and col_slot_context_.

1074  {
1076 }

+ Here is the call graph for this function:

void QueryMemoryDescriptor::clearTargetGroupbyIndices ( )
inline

Definition at line 242 of file QueryMemoryDescriptor.h.

References target_groupby_indices_.

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

Definition at line 148 of file QueryMemoryDescriptor.h.

References Invalid.

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

149  {
150  return std::all_of(count_distinct_descriptors.begin(),
151  count_distinct_descriptors.end(),
152  [](const CountDistinctDescriptor& desc) {
153  return desc.impl_type_ == CountDistinctImplType::Invalid;
154  });
155  }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::countDistinctDescriptorsLogicallyEmpty ( ) const
inline

Definition at line 157 of file QueryMemoryDescriptor.h.

References count_distinct_descriptors_, and countDescriptorsLogicallyEmpty().

Referenced by QueryMemoryInitializer::allocateCountDistinctGpuMem().

157  {
159  }
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 264 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 275 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 249 of file QueryMemoryDescriptor.h.

References bucket_.

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

249 { return bucket_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getBufferColSlotCount ( ) const

Definition at line 972 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, ColSlotContext::getSlotCount(), 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().

972  {
973  size_t total_slot_count = col_slot_context_.getSlotCount();
974 
975  if (target_groupby_indices_.empty()) {
976  return total_slot_count;
977  }
978  return total_slot_count - std::count_if(target_groupby_indices_.begin(),
980  [](const int64_t i) { return i >= 0; });
981 }
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 903 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().

906  {
907  if (use_streaming_top_n_) {
908  const size_t n = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
909  return streaming_top_n::get_heap_size(getRowSize(), n, thread_count);
910  }
911  return getBufferSizeBytes(device_type, entry_count_);
912 }
size_t getBufferSizeBytes(const RelAlgExecutionUnit &ra_exe_unit, const unsigned thread_count, const ExecutorDeviceType device_type) const
const size_t limit
const SortInfo sort_info
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 949 of file QueryMemoryDescriptor.cpp.

References entry_count_, and getBufferSizeBytes().

950  {
951  return getBufferSizeBytes(device_type, entry_count_);
952 }
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 925 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_.

926  {
928  CHECK_GE(group_col_widths_.size(), size_t(1));
929  auto row_bytes = align_to_int64(getColsSize());
930 
931  return (interleavedBins(device_type) ? executor_->warpSize() : 1) * entry_count *
932  row_bytes;
933  }
934 
935  constexpr size_t row_index_width = sizeof(int64_t);
936  size_t total_bytes{0};
937  if (output_columnar_) {
939  ? row_index_width * entry_count
940  : sizeof(int64_t) * group_col_widths_.size() * entry_count) +
942  } else {
943  total_bytes = getRowSize() * entry_count;
944  }
945 
946  return total_bytes;
947 }
#define CHECK_GE(x, y)
Definition: Logger.h:210
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 1031 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getColCount().

1031  {
1032  return col_slot_context_.getColCount();
1033 }
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 799 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::initGroups(), inplace_sort_gpu(), and anonymous_namespace{Execute.cpp}::permute_storage_columnar().

799  {
800  const auto warp_count = getWarpCount();
801  if (output_columnar_) {
802  CHECK_EQ(size_t(1), warp_count);
803  size_t offset{0};
804  if (!keyless_hash_) {
806  }
807  for (size_t index = 0; index < col_idx; ++index) {
809  }
810  return offset;
811  }
812 
813  size_t offset{0};
814  if (keyless_hash_) {
815  // ignore, there's no group column in the output buffer
817  } else {
818  offset += group_col_widths_.size() * getEffectiveKeyWidth();
819  offset = align_to_int64(offset);
820  }
821  offset += getColOnlyOffInBytes(col_idx);
822  return offset;
823 }
#define CHECK_EQ(x, y)
Definition: Logger.h:205
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:197
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 858 of file QueryMemoryDescriptor.cpp.

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

Referenced by QueryExecutionContext::groupBufferToDeinterleavedResults().

858  {
859  auto warp_count = getWarpCount();
860  if (output_columnar_) {
861  CHECK_EQ(size_t(1), group_col_widths_.size());
862  CHECK_EQ(size_t(1), warp_count);
863  return getPaddedSlotWidthBytes(col_idx);
864  }
865 
866  return warp_count * getRowSize();
867 }
#define CHECK_EQ(x, y)
Definition: Logger.h:205
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 786 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getColOnlyOffInBytes().

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

786  {
787  return col_slot_context_.getColOnlyOffInBytes(col_idx);
788 }
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 290 of file QueryMemoryDescriptor.h.

References col_slot_context_.

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

290 { return col_slot_context_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getColsSize ( ) const

Definition at line 725 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getAllSlotsAlignedPaddedSize().

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

725  {
727 }
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 747 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().

747  {
749 }
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

Definition at line 257 of file QueryMemoryDescriptor.h.

References count_distinct_descriptors_.

Referenced by QueryMemoryInitializer::allocateCountDistinctGpuMem(), anonymous_namespace{QueryMemoryInitializer.cpp}::check_total_bitmap_memory(), ResultSetReductionJIT::reduceOneApproxMedianSlot(), ResultSetStorage::reduceOneApproxMedianSlot(), ResultSetReductionJIT::reduceOneCountDistinctSlot(), and ResultSetStorage::reduceOneCountDistinctSlot().

257  {
258  return count_distinct_descriptors_.size();
259  }
CountDistinctDescriptors count_distinct_descriptors_

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getEntryCount ( ) const
inline

Definition at line 244 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(), GroupByAndAggregate::codegenOutputSlot(), 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().

244 { return entry_count_; }

+ Here is the caller graph for this function:

const Executor* QueryMemoryDescriptor::getExecutor ( ) const
inline

Definition at line 166 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().

166 { return executor_; }

+ Here is the caller graph for this function:

size_t QueryMemoryDescriptor::getKeyCount ( ) const
inline

Definition at line 280 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 1043 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().

1044  {
1045  return col_slot_context_.getSlotInfo(slot_idx).logical_size;
1046 }
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 248 of file QueryMemoryDescriptor.h.

References max_val_.

Referenced by GroupByAndAggregate::codegenGroupBy().

248 { return max_val_; }

+ Here is the caller graph for this function:

int64_t QueryMemoryDescriptor::getMinVal ( ) const
inline

Definition at line 247 of file QueryMemoryDescriptor.h.

References min_val_.

Referenced by GroupByAndAggregate::codegenSingleColumnPerfectHash().

247 { 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 869 of file QueryMemoryDescriptor.cpp.

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

Referenced by QueryMemoryInitializer::initColumnPerRow().

871  {
873  size_t offset{0};
874  auto warp_count = getWarpCount();
875  const auto chosen_bytes = getPaddedSlotWidthBytes(col_idx);
876  const auto total_slot_count = getSlotCount();
877  if (col_idx + 1 == total_slot_count) {
878  if (output_columnar_) {
879  return (entry_count_ - bin) * chosen_bytes;
880  } else {
881  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
882  }
883  }
884 
885  const auto next_chosen_bytes = getPaddedSlotWidthBytes(col_idx + 1);
886  if (output_columnar_) {
887  CHECK_EQ(size_t(1), group_col_widths_.size());
888  CHECK_EQ(size_t(1), warp_count);
889 
890  offset = align_to_int64(entry_count_ * chosen_bytes);
891 
892  offset += bin * (next_chosen_bytes - chosen_bytes);
893  return offset;
894  }
895 
896  if (next_chosen_bytes == sizeof(int64_t)) {
897  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
898  } else {
899  return chosen_bytes;
900  }
901 }
#define CHECK_EQ(x, y)
Definition: Logger.h:205
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const
#define CHECK(condition)
Definition: Logger.h:197
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::getPaddedColWidthForRange ( const size_t  offset,
const size_t  range 
) const
inline

Definition at line 207 of file QueryMemoryDescriptor.h.

References getPaddedSlotWidthBytes().

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

207  {
208  size_t ret = 0;
209  for (size_t i = offset; i < offset + range; i++) {
210  ret += static_cast<size_t>(getPaddedSlotWidthBytes(i));
211  }
212  return ret;
213  }
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 1039 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(), getPaddedColWidthForRange(), ResultSet::getPaddedSlotWidthBytes(), ResultSet::getTargetValueFromBufferColwise(), ResultSet::getTargetValueFromBufferRowwise(), init_agg_val_vec(), QueryMemoryInitializer::initColumnarGroups(), QueryMemoryInitializer::initColumnPerRow(), 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().

1039  {
1040  return col_slot_context_.getSlotInfo(slot_idx).padded_size;
1041 }
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 847 of file QueryMemoryDescriptor.cpp.

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

Referenced by getColOffInBytes().

847  {
849  size_t buffer_size{0};
850  for (size_t group_idx = 0; group_idx < getGroupbyColCount(); group_idx++) {
851  buffer_size += align_to_int64(
852  std::max(groupColWidth(group_idx), static_cast<int8_t>(sizeof(int64_t))) *
853  getEntryCount());
854  }
855  return buffer_size;
856 }
int8_t groupColWidth(const size_t key_idx) const
size_t getGroupbyColCount() const
#define CHECK(condition)
Definition: Logger.h:197
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 829 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().

830  {
832  CHECK(group_idx < getGroupbyColCount());
833  size_t offset{0};
834  for (size_t col_idx = 0; col_idx < group_idx; col_idx++) {
835  // TODO(Saman): relax that int64_bit part immediately
836  offset += align_to_int64(
837  std::max(groupColWidth(col_idx), static_cast<int8_t>(sizeof(int64_t))) *
838  getEntryCount());
839  }
840  return offset;
841 }
int8_t groupColWidth(const size_t key_idx) const
size_t getGroupbyColCount() const
#define CHECK(condition)
Definition: Logger.h:197
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 168 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().

168 { 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,
RenderInfo render_info 
) const

Definition at line 605 of file QueryMemoryDescriptor.cpp.

References DEBUG_TIMER, and QueryExecutionContext.

Referenced by ExecutionKernel::runImpl().

617  {
618  auto timer = DEBUG_TIMER(__func__);
619  if (frag_offsets.empty()) {
620  return nullptr;
621  }
622  return std::unique_ptr<QueryExecutionContext>(
623  new QueryExecutionContext(ra_exe_unit,
624  *this,
625  executor,
626  device_type,
627  dispatch_mode,
628  device_id,
629  num_rows,
630  col_buffers,
631  frag_offsets,
632  row_set_mem_owner,
633  output_columnar,
634  sort_on_gpu,
635  render_info));
636 }
#define DEBUG_TIMER(name)
Definition: Logger.h:313
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 729 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::initGroups(), anonymous_namespace{Execute.cpp}::permute_storage_row_wise(), QueryMemoryDescriptor(), QueryMemoryInitializer::QueryMemoryInitializer(), and ResultSetStorage::reduceSingleRow().

729  {
731  size_t total_bytes{0};
732  if (keyless_hash_) {
733  // ignore, there's no group column in the output buffer
735  } else {
736  total_bytes += group_col_widths_.size() * getEffectiveKeyWidth();
737  total_bytes = align_to_int64(total_bytes);
738  }
739  total_bytes += getColsSize();
740  return align_to_int64(total_bytes);
741 }
size_t getEffectiveKeyWidth() const
QueryDescriptionType query_desc_type_
#define CHECK(condition)
Definition: Logger.h:197
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 1059 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getAllSlotsPaddedSize().

Referenced by get_row_bytes().

1059  {
1060  // Note: Actual row size may include padding (see ResultSetBufferAccessors.h)
1062 }
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 1048 of file QueryMemoryDescriptor.cpp.

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

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

1049  {
1050  const auto& col_slots = col_slot_context_.getSlotsForCol(col_idx);
1051  CHECK_EQ(col_slots.size(), size_t(1));
1052  return col_slots.front();
1053 }
#define CHECK_EQ(x, y)
Definition: Logger.h:205
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 226 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().

226  {
227  CHECK_LT(target_idx, target_groupby_indices_.size());
228  return target_groupby_indices_[target_idx];
229  }
std::vector< int64_t > target_groupby_indices_
#define CHECK_LT(x, y)
Definition: Logger.h:207

+ Here is the caller graph for this function:

int32_t QueryMemoryDescriptor::getTargetIdxForKey ( ) const
inline

Definition at line 181 of file QueryMemoryDescriptor.h.

References idx_target_as_key_.

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

181 { 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 756 of file QueryMemoryDescriptor.cpp.

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

Referenced by getBufferSizeBytes(), and getTotalBytesOfColumnarProjections().

756  {
759 }
size_t getTotalBytesOfColumnarBuffers(const size_t entry_count) const
#define CHECK(condition)
Definition: Logger.h:197

+ 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 765 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getTotalBytesOfColumnarBuffers().

766  {
767  return col_slot_context_.getTotalBytesOfColumnarBuffers(num_entries_per_column);
768 }
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 779 of file QueryMemoryDescriptor.cpp.

References getTotalBytesOfColumnarBuffers().

780  {
781  constexpr size_t row_index_width = sizeof(int64_t);
782  return getTotalBytesOfColumnarBuffers(projection_count) +
783  row_index_width * projection_count;
784 }
size_t getTotalBytesOfColumnarBuffers() const

+ Here is the call graph for this function:

size_t QueryMemoryDescriptor::getWarpCount ( ) const

Definition at line 743 of file QueryMemoryDescriptor.cpp.

References executor_, and interleaved_bins_on_gpu_.

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

743  {
744  return (interleaved_bins_on_gpu_ ? executor_->warpSize() : 1);
745 }

+ Here is the caller graph for this function:

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

Definition at line 184 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().

184  {
185  CHECK_LT(key_idx, group_col_widths_.size());
186  return group_col_widths_[key_idx];
187  }
#define CHECK_LT(x, y)
Definition: Logger.h:207
std::vector< int8_t > group_col_widths_

+ Here is the caller graph for this function:

const auto QueryMemoryDescriptor::groupColWidthsBegin ( ) const
inline

Definition at line 191 of file QueryMemoryDescriptor.h.

References group_col_widths_.

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

Definition at line 192 of file QueryMemoryDescriptor.h.

References group_col_widths_.

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

Definition at line 178 of file QueryMemoryDescriptor.h.

References interleaved_bins_on_gpu_.

bool QueryMemoryDescriptor::hasNulls ( ) const
inline

Definition at line 251 of file QueryMemoryDescriptor.h.

References has_nulls_.

Referenced by GroupByAndAggregate::codegenGroupBy().

251 { return has_nulls_; }

+ 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 180 of file QueryMemoryDescriptor.cpp.

References get_col_byte_widths(), RelAlgExecutionUnit::groupby_exprs, and RelAlgExecutionUnit::target_exprs.

Referenced by GroupByAndAggregate::initQueryMemoryDescriptorImpl().

196  {
197  auto group_col_widths = get_col_byte_widths(ra_exe_unit.groupby_exprs);
198  const bool is_group_by{!group_col_widths.empty()};
199 
200  auto col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, {});
201 
202  const auto min_slot_size = QueryMemoryDescriptor::pick_target_compact_width(
203  ra_exe_unit, query_infos, crt_min_byte_width);
204 
205  col_slot_context.setAllSlotsPaddedSize(min_slot_size);
206  col_slot_context.validate();
207 
208  if (!is_group_by) {
209  CHECK(!must_use_baseline_sort);
210 
211  return std::make_unique<QueryMemoryDescriptor>(
212  executor,
213  ra_exe_unit,
214  query_infos,
215  allow_multifrag,
216  false,
217  false,
218  -1,
219  ColRangeInfo{ra_exe_unit.estimator ? QueryDescriptionType::Estimator
221  0,
222  0,
223  0,
224  false},
225  col_slot_context,
226  std::vector<int8_t>{},
227  /*group_col_compact_width=*/0,
228  std::vector<int64_t>{},
229  /*entry_count=*/1,
230  count_distinct_descriptors,
231  false,
232  output_columnar_hint,
233  render_info && render_info->isPotentialInSituRender(),
234  must_use_baseline_sort,
235  /*use_streaming_top_n=*/false);
236  }
237 
238  size_t entry_count = 1;
239  auto actual_col_range_info = col_range_info;
240  bool interleaved_bins_on_gpu = false;
241  bool keyless_hash = false;
242  bool streaming_top_n = false;
243  int8_t group_col_compact_width = 0;
244  int32_t idx_target_as_key = -1;
245  auto output_columnar = output_columnar_hint;
246  std::vector<int64_t> target_groupby_indices;
247 
248  switch (col_range_info.hash_type_) {
250  if (render_info) {
251  render_info->setInSituDataIfUnset(false);
252  }
253  // keyless hash: whether or not group columns are stored at the beginning of the
254  // output buffer
255  keyless_hash =
256  (!sort_on_gpu_hint ||
258  col_range_info.max, col_range_info.min, col_range_info.bucket)) &&
259  !col_range_info.bucket && !must_use_baseline_sort && keyless_info.keyless;
260 
261  // if keyless, then this target index indicates wheter an entry is empty or not
262  // (acts as a key)
263  idx_target_as_key = keyless_info.target_index;
264 
265  if (group_col_widths.size() > 1) {
266  // col range info max contains the expected cardinality of the output
267  entry_count = static_cast<size_t>(actual_col_range_info.max);
268  actual_col_range_info.bucket = 0;
269  } else {
270  // single column perfect hash
271  entry_count = std::max(
272  GroupByAndAggregate::getBucketedCardinality(col_range_info), int64_t(1));
273  const size_t interleaved_max_threshold{512};
274 
275  if (must_use_baseline_sort) {
276  target_groupby_indices = target_expr_group_by_indices(ra_exe_unit.groupby_exprs,
277  ra_exe_unit.target_exprs);
278  col_slot_context =
279  ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
280  }
281 
282  bool has_varlen_sample_agg = false;
283  for (const auto& target_expr : ra_exe_unit.target_exprs) {
284  if (target_expr->get_contains_agg()) {
285  const auto agg_expr = dynamic_cast<Analyzer::AggExpr*>(target_expr);
286  CHECK(agg_expr);
287  if (agg_expr->get_aggtype() == kSAMPLE &&
288  agg_expr->get_type_info().is_varlen()) {
289  has_varlen_sample_agg = true;
290  break;
291  }
292  }
293  }
294 
295  interleaved_bins_on_gpu = keyless_hash && !has_varlen_sample_agg &&
296  (entry_count <= interleaved_max_threshold) &&
297  (device_type == ExecutorDeviceType::GPU) &&
299  count_distinct_descriptors) &&
300  !output_columnar;
301  }
302  break;
303  }
305  if (render_info) {
306  render_info->setInSituDataIfUnset(false);
307  }
308  entry_count = shard_count
309  ? (max_groups_buffer_entry_count + shard_count - 1) / shard_count
310  : max_groups_buffer_entry_count;
311  target_groupby_indices = target_expr_group_by_indices(ra_exe_unit.groupby_exprs,
312  ra_exe_unit.target_exprs);
313  col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
314 
315  group_col_compact_width =
316  output_columnar ? 8
317  : pick_baseline_key_width(ra_exe_unit, query_infos, executor);
318 
319  actual_col_range_info =
321  break;
322  }
324  CHECK(!must_use_baseline_sort);
325 
326  if (streaming_top_n_hint && use_streaming_top_n(ra_exe_unit, output_columnar)) {
327  streaming_top_n = true;
328  entry_count = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
329  } else {
330  if (ra_exe_unit.use_bump_allocator) {
331  output_columnar = false;
332  entry_count = 0;
333  } else {
334  entry_count = ra_exe_unit.scan_limit
335  ? static_cast<size_t>(ra_exe_unit.scan_limit)
336  : max_groups_buffer_entry_count;
337  }
338  }
339 
340  const auto catalog = executor->getCatalog();
341  CHECK(catalog);
342  target_groupby_indices = executor->plan_state_->allow_lazy_fetch_
343  ? target_expr_proj_indices(ra_exe_unit, *catalog)
344  : std::vector<int64_t>{};
345 
346  col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
347  break;
348  }
349  default:
350  UNREACHABLE() << "Unknown query type";
351  }
352 
353  return std::make_unique<QueryMemoryDescriptor>(
354  executor,
355  ra_exe_unit,
356  query_infos,
357  allow_multifrag,
358  keyless_hash,
359  interleaved_bins_on_gpu,
360  idx_target_as_key,
361  actual_col_range_info,
362  col_slot_context,
363  group_col_widths,
364  group_col_compact_width,
365  target_groupby_indices,
366  entry_count,
367  count_distinct_descriptors,
368  sort_on_gpu_hint,
369  output_columnar,
370  render_info && render_info->isPotentialInSituRender(),
371  must_use_baseline_sort,
372  streaming_top_n);
373 }
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
bool setInSituDataIfUnset(const bool is_in_situ_data)
Definition: RenderInfo.cpp:98
QueryDescriptionType hash_type_
#define UNREACHABLE()
Definition: Logger.h:241
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)
std::vector< int8_t > get_col_byte_widths(const T &col_expr_list)
bool isPotentialInSituRender() const
Definition: RenderInfo.cpp:64
#define CHECK(condition)
Definition: Logger.h:197
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)
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 1015 of file QueryMemoryDescriptor.cpp.

References GPU, and interleaved_bins_on_gpu_.

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

1015  {
1016  return interleaved_bins_on_gpu_ && device_type == ExecutorDeviceType::GPU;
1017 }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::isGroupBy ( ) const
inline

Definition at line 195 of file QueryMemoryDescriptor.h.

References group_col_widths_.

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

195 { 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 965 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().

965  {
966  // In distributed mode, result sets are serialized using rowwise iterators, so we use
967  // consistent slot widths for now
968  return output_columnar_ && !g_cluster &&
970 }
QueryDescriptionType query_desc_type_
bool g_cluster

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::isSingleColumnGroupByWithPerfectHash ( ) const
inline

Definition at line 170 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 1020 of file QueryMemoryDescriptor.cpp.

References CHECK, executor_, and GPU.

Referenced by query_group_by_template_impl().

1021  {
1022  if (device_type != ExecutorDeviceType::GPU) {
1023  return false;
1024  } else {
1025  auto cuda_mgr = executor_->getCatalog()->getDataMgr().getCudaMgr();
1026  CHECK(cuda_mgr);
1027  return cuda_mgr->isArchVoltaOrGreaterForAll();
1028  }
1029 }
#define CHECK(condition)
Definition: Logger.h:197

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::lazyInitGroups ( const ExecutorDeviceType  device_type) const

Definition at line 1010 of file QueryMemoryDescriptor.cpp.

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

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

1010  {
1011  return device_type == ExecutorDeviceType::GPU && !render_output_ &&
1013 }
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 142 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory().

144  {
145  return max_val - min_val > 10000 * std::max(bucket, int64_t(1));
146  }

+ Here is the caller graph for this function:

bool QueryMemoryDescriptor::mustUseBaselineSort ( ) const
inline

Definition at line 271 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 541 of file QueryMemoryDescriptor.cpp.

References bucket_, col_slot_context_, count_distinct_descriptors_, force_4byte_float_, group_col_compact_width_, group_col_widths_, has_nulls_, 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_.

541  {
542  // Note that this method does not check ptr reference members (e.g. executor_) or
543  // entry_count_
544  if (query_desc_type_ != other.query_desc_type_) {
545  return false;
546  }
547  if (keyless_hash_ != other.keyless_hash_) {
548  return false;
549  }
551  return false;
552  }
553  if (idx_target_as_key_ != other.idx_target_as_key_) {
554  return false;
555  }
556  if (force_4byte_float_ != other.force_4byte_float_) {
557  return false;
558  }
559  if (group_col_widths_ != other.group_col_widths_) {
560  return false;
561  }
563  return false;
564  }
566  return false;
567  }
568  if (min_val_ != other.min_val_) {
569  return false;
570  }
571  if (max_val_ != other.max_val_) {
572  return false;
573  }
574  if (bucket_ != other.bucket_) {
575  return false;
576  }
577  if (has_nulls_ != other.has_nulls_) {
578  return false;
579  }
581  return false;
582  } else {
583  // Count distinct descriptors can legitimately differ in device only.
584  for (size_t i = 0; i < count_distinct_descriptors_.size(); ++i) {
585  auto ref_count_distinct_desc = other.count_distinct_descriptors_[i];
586  auto count_distinct_desc = count_distinct_descriptors_[i];
587  count_distinct_desc.device_type = ref_count_distinct_desc.device_type;
588  if (ref_count_distinct_desc != count_distinct_desc) {
589  return false;
590  }
591  }
592  }
593  if (sort_on_gpu_ != other.sort_on_gpu_) {
594  return false;
595  }
596  if (output_columnar_ != other.output_columnar_) {
597  return false;
598  }
599  if (col_slot_context_ != other.col_slot_context_) {
600  return false;
601  }
602  return true;
603 }
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 638 of file QueryMemoryDescriptor.cpp.

References CHECK, CHECK_EQ, g_bigint_count, 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.

641  {
642  if (g_bigint_count) {
643  return sizeof(int64_t);
644  }
645  int8_t compact_width{0};
646  auto col_it = ra_exe_unit.input_col_descs.begin();
647  int unnest_array_col_id{std::numeric_limits<int>::min()};
648  for (const auto& groupby_expr : ra_exe_unit.groupby_exprs) {
649  const auto uoper = dynamic_cast<Analyzer::UOper*>(groupby_expr.get());
650  if (uoper && uoper->get_optype() == kUNNEST) {
651  const auto& arg_ti = uoper->get_operand()->get_type_info();
652  CHECK(arg_ti.is_array());
653  const auto& elem_ti = arg_ti.get_elem_type();
654  if (elem_ti.is_string() && elem_ti.get_compression() == kENCODING_DICT) {
655  unnest_array_col_id = (*col_it)->getColId();
656  } else {
657  compact_width = crt_min_byte_width;
658  break;
659  }
660  }
661  ++col_it;
662  }
663  if (!compact_width &&
664  (ra_exe_unit.groupby_exprs.size() != 1 || !ra_exe_unit.groupby_exprs.front())) {
665  compact_width = crt_min_byte_width;
666  }
667  if (!compact_width) {
668  col_it = ra_exe_unit.input_col_descs.begin();
669  std::advance(col_it, ra_exe_unit.groupby_exprs.size());
670  for (const auto target : ra_exe_unit.target_exprs) {
671  const auto& ti = target->get_type_info();
672  const auto agg = dynamic_cast<const Analyzer::AggExpr*>(target);
673  if (agg && agg->get_arg()) {
674  compact_width = crt_min_byte_width;
675  break;
676  }
677 
678  if (agg) {
679  CHECK_EQ(kCOUNT, agg->get_aggtype());
680  CHECK(!agg->get_is_distinct());
681  ++col_it;
682  continue;
683  }
684 
685  if (is_int_and_no_bigger_than(ti, 4) ||
686  (ti.is_string() && ti.get_compression() == kENCODING_DICT)) {
687  ++col_it;
688  continue;
689  }
690 
691  const auto uoper = dynamic_cast<Analyzer::UOper*>(target);
692  if (uoper && uoper->get_optype() == kUNNEST &&
693  (*col_it)->getColId() == unnest_array_col_id) {
694  const auto arg_ti = uoper->get_operand()->get_type_info();
695  CHECK(arg_ti.is_array());
696  const auto& elem_ti = arg_ti.get_elem_type();
697  if (elem_ti.is_string() && elem_ti.get_compression() == kENCODING_DICT) {
698  ++col_it;
699  continue;
700  }
701  }
702 
703  compact_width = crt_min_byte_width;
704  break;
705  }
706  }
707  if (!compact_width) {
708  size_t total_tuples{0};
709  for (const auto& qi : query_infos) {
710  total_tuples += qi.info.getNumTuples();
711  }
712  return total_tuples <= static_cast<size_t>(std::numeric_limits<uint32_t>::max()) ||
713  unnest_array_col_id != std::numeric_limits<int>::min()
714  ? 4
715  : crt_min_byte_width;
716  } else {
717  // TODO(miyu): relax this condition to allow more cases just w/o padding
718  for (auto wid : get_col_byte_widths(ra_exe_unit.target_exprs)) {
719  compact_width = std::max(compact_width, wid);
720  }
721  return compact_width;
722  }
723 }
std::vector< Analyzer::Expr * > target_exprs
#define CHECK_EQ(x, y)
Definition: Logger.h:205
const std::list< std::shared_ptr< Analyzer::Expr > > groupby_exprs
bool g_bigint_count
const SQLTypeInfo & get_type_info() const
Definition: Analyzer.h:78
std::vector< int8_t > get_col_byte_widths(const T &col_expr_list)
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:197
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 1088 of file QueryMemoryDescriptor.cpp.

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

Referenced by reductionKey().

1088  {
1089  switch (query_desc_type_) {
1091  return "Perfect Hash";
1093  return "Baseline Hash";
1095  return "Projection";
1097  return "Non-grouped Aggregate";
1099  return "Estimator";
1100  default:
1101  UNREACHABLE();
1102  }
1103  return "";
1104 }
#define UNREACHABLE()
Definition: Logger.h:241
QueryDescriptionType query_desc_type_

+ Here is the caller graph for this function:

std::string QueryMemoryDescriptor::reductionKey ( ) const

Definition at line 1128 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().

1128  {
1129  std::string str;
1130  str += "Query Memory Descriptor State\n";
1131  str += "\tQuery Type: " + queryDescTypeToString() + "\n";
1132  str +=
1133  "\tKeyless Hash: " + ::toString(keyless_hash_) +
1134  (keyless_hash_ ? ", target index for key: " + std::to_string(getTargetIdxForKey())
1135  : "") +
1136  "\n";
1137  str += "\tEffective key width: " + std::to_string(getEffectiveKeyWidth()) + "\n";
1138  str += "\tNumber of group columns: " + std::to_string(getGroupbyColCount()) + "\n";
1139  const auto group_indices_size = targetGroupbyIndicesSize();
1140  if (group_indices_size) {
1141  std::vector<std::string> group_indices_strings;
1142  for (size_t target_idx = 0; target_idx < group_indices_size; ++target_idx) {
1143  group_indices_strings.push_back(std::to_string(getTargetGroupbyIndex(target_idx)));
1144  }
1145  str += "\tTarget group by indices: " +
1146  boost::algorithm::join(group_indices_strings, ",") + "\n";
1147  }
1148  str += "\t" + col_slot_context_.toString();
1149  return str;
1150 }
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 326 of file QueryMemoryDescriptor.h.

References group_col_widths_.

326  {
327  group_col_widths_ = new_group_col_widths;
328  }
std::vector< int8_t > group_col_widths_
void QueryMemoryDescriptor::setAllTargetGroupbyIndices ( std::vector< int64_t >  group_by_indices)
inline

Definition at line 231 of file QueryMemoryDescriptor.h.

References target_groupby_indices_.

231  {
232  target_groupby_indices_ = group_by_indices;
233  }
std::vector< int64_t > target_groupby_indices_
void QueryMemoryDescriptor::setEntryCount ( const size_t  val)
inline

Definition at line 245 of file QueryMemoryDescriptor.h.

References entry_count_.

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

245 { entry_count_ = val; }

+ Here is the caller graph for this function:

void QueryMemoryDescriptor::setForceFourByteFloat ( const bool  val)
inline

Definition at line 276 of file QueryMemoryDescriptor.h.

References force_4byte_float_.

void QueryMemoryDescriptor::setGroupColCompactWidth ( const int8_t  val)
inline

Definition at line 197 of file QueryMemoryDescriptor.h.

References group_col_compact_width_.

void QueryMemoryDescriptor::setHasInterleavedBinsOnGpu ( const bool  val)
inline

Definition at line 179 of file QueryMemoryDescriptor.h.

References interleaved_bins_on_gpu_.

void QueryMemoryDescriptor::setHasKeylessHash ( const bool  val)
inline

Definition at line 176 of file QueryMemoryDescriptor.h.

References keyless_hash_.

void QueryMemoryDescriptor::setOutputColumnar ( const bool  val)

Definition at line 954 of file QueryMemoryDescriptor.cpp.

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

Referenced by TableFunctionExecutionContext::launchCpuCode(), and TableFunctionExecutionContext::launchGpuCode().

+ 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 169 of file QueryMemoryDescriptor.h.

References query_desc_type_.

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

Definition at line 182 of file QueryMemoryDescriptor.h.

References idx_target_as_key_.

bool QueryMemoryDescriptor::sortOnGpu ( ) const
inline

Definition at line 261 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 236 of file QueryMemoryDescriptor.h.

References target_groupby_indices_.

236  {
237  return std::count_if(
238  target_groupby_indices_.begin(),
240  [](const int64_t& target_group_by_index) { return target_group_by_index < 0; });
241  }
std::vector< int64_t > target_groupby_indices_
bool QueryMemoryDescriptor::threadsShareMemory ( ) const
std::string QueryMemoryDescriptor::toString ( ) const

Definition at line 1106 of file QueryMemoryDescriptor.cpp.

References allow_multifrag_, blocksShareMemory(), bucket_, entry_count_, GPU, interleaved_bins_on_gpu_, 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().

1106  {
1107  auto str = reductionKey();
1108  str += "\tAllow Multifrag: " + ::toString(allow_multifrag_) + "\n";
1109  str += "\tInterleaved Bins on GPU: " + ::toString(interleaved_bins_on_gpu_) + "\n";
1110  str += "\tBlocks Share Memory: " + ::toString(blocksShareMemory()) + "\n";
1111  str += "\tThreads Share Memory: " + ::toString(threadsShareMemory()) + "\n";
1112  str += "\tUses Fast Group Values: " + ::toString(usesGetGroupValueFast()) + "\n";
1113  str +=
1114  "\tLazy Init Groups (GPU): " + ::toString(lazyInitGroups(ExecutorDeviceType::GPU)) +
1115  "\n";
1116  str += "\tEntry Count: " + std::to_string(entry_count_) + "\n";
1117  str += "\tMin Val (perfect hash only): " + std::to_string(min_val_) + "\n";
1118  str += "\tMax Val (perfect hash only): " + std::to_string(max_val_) + "\n";
1119  str += "\tBucket Val (perfect hash only): " + std::to_string(bucket_) + "\n";
1120  str += "\tSort on GPU: " + ::toString(sort_on_gpu_) + "\n";
1121  str += "\tUse Streaming Top N: " + ::toString(use_streaming_top_n_) + "\n";
1122  str += "\tOutput Columnar: " + ::toString(output_columnar_) + "\n";
1123  str += "\tRender Output: " + ::toString(render_output_) + "\n";
1124  str += "\tUse Baseline Sort: " + ::toString(must_use_baseline_sort_) + "\n";
1125  return str;
1126 }
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 1064 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::getMinPaddedByteSize().

1065  {
1066  return col_slot_context_.getMinPaddedByteSize(actual_min_byte_width);
1067 }
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 1055 of file QueryMemoryDescriptor.cpp.

References col_slot_context_, and ColSlotContext::setAllSlotsSize().

1055  {
1056  col_slot_context_.setAllSlotsSize(slot_width_size);
1057 }
void setAllSlotsSize(const int8_t slot_width_size)

+ Here is the call graph for this function:

bool QueryMemoryDescriptor::usesGetGroupValueFast ( ) const

Definition at line 983 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 267 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:

Friends And Related Function Documentation

friend class QueryExecutionContext
friend

Definition at line 368 of file QueryMemoryDescriptor.h.

Referenced by getQueryExecutionContext().

friend class ResultSet
friend

Definition at line 367 of file QueryMemoryDescriptor.h.

Member Data Documentation

bool QueryMemoryDescriptor::allow_multifrag_
private

Definition at line 332 of file QueryMemoryDescriptor.h.

Referenced by toString().

int64_t QueryMemoryDescriptor::bucket_
private

Definition at line 349 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 359 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 350 of file QueryMemoryDescriptor.h.

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

int32_t QueryMemoryDescriptor::idx_target_as_key_
private

Definition at line 336 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 356 of file QueryMemoryDescriptor.h.

Referenced by blocksShareMemory().

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

Definition at line 348 of file QueryMemoryDescriptor.h.

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

int64_t QueryMemoryDescriptor::min_val_
private

Definition at line 346 of file QueryMemoryDescriptor.h.

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

bool QueryMemoryDescriptor::must_use_baseline_sort_
private

Definition at line 355 of file QueryMemoryDescriptor.h.

Referenced by mustUseBaselineSort(), and toString().

bool QueryMemoryDescriptor::render_output_
private

Definition at line 354 of file QueryMemoryDescriptor.h.

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

bool QueryMemoryDescriptor::sort_on_gpu_
private

Definition at line 352 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: