Class QueryMemoryDescriptor¶

class QueryMemoryDescriptor¶

Public 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 &ra_exe_unit, const Executor *executor, const ExecutorDeviceType device_type, const ExecutorDispatchMode dispatch_mode, const int device_id, const int outer_table_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¶

bool countDistinctDescriptorsLogicallyEmpty() const¶

const Executor *getExecutor() 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¶

void setPaddedSlotWidthBytes(const size_t slot_idx, const int8_t bytes)¶

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 CountDistinctDescriptor &getCountDistinctDescriptor(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¶

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 table function: only the columnar buffer 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

const ColSlotContext &getColSlotContext() const¶

bool usesGetGroupValueFast() const¶

bool blocksShareMemory() const¶

bool threadsShareMemory() const¶

bool lazyInitGroups(const ExecutorDeviceType device_type) const¶

bool interleavedBins(const ExecutorDeviceType device_type) 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 device_type) 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¶

Public Static Functions

static TResultSetBufferDescriptor toThrift(const QueryMemoryDescriptor&)¶

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)¶

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 Functions

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

Private Functions

size_t getTotalBytesOfColumnarBuffers() const¶: Returns the maximum total number of bytes (including required paddings) to store all non-lazy columns’ results for columnar cases.

size_t getTotalBytesOfColumnarBuffers(const size_t num_entries_per_column) const¶: 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.

size_t getTotalBytesOfColumnarProjections(const size_t projection_count) const¶

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)

Private Members

const Executor *executor_¶

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

friend QueryMemoryDescriptor::ResultSet

friend QueryMemoryDescriptor::QueryExecutionContext