QueryMemoryInitializer.cpp File Reference

#include "QueryMemoryInitializer.h"
#include "Execute.h"
#include "GpuInitGroups.h"
#include "GpuMemUtils.h"
#include "OutputBufferInitialization.h"
#include "ResultSet.h"
#include "Shared/Logger.h"
#include "StreamingTopN.h"
#include <Shared/checked_alloc.h>

Include dependency graph for QueryMemoryInitializer.cpp:

Go to the source code of this file.

Namespaces
	anonymous_namespace{QueryMemoryInitializer.cpp}

Functions
void	anonymous_namespace{QueryMemoryInitializer.cpp}::check_total_bitmap_memory (const QueryMemoryDescriptor &query_mem_desc)
int64_t *	anonymous_namespace{QueryMemoryInitializer.cpp}::alloc_group_by_buffer (const size_t numBytes, RenderAllocatorMap *render_allocator_map)
int64_t	anonymous_namespace{QueryMemoryInitializer.cpp}::get_consistent_frag_size (const std::vector< uint64_t > &frag_offsets)
std::vector< int64_t >	anonymous_namespace{QueryMemoryInitializer.cpp}::get_consistent_frags_sizes (const std::vector< std::vector< uint64_t >> &frag_offsets)
std::vector< int64_t >	anonymous_namespace{QueryMemoryInitializer.cpp}::get_consistent_frags_sizes (const std::vector< Analyzer::Expr * > &target_exprs, const std::vector< int64_t > &table_frag_sizes)
std::vector< std::vector < int64_t > >	anonymous_namespace{QueryMemoryInitializer.cpp}::get_col_frag_offsets (const std::vector< Analyzer::Expr * > &target_exprs, const std::vector< std::vector< uint64_t >> &table_frag_offsets)
	num_buffers_ (1)
	count_distinct_bitmap_mem_ (0)
	count_distinct_bitmap_mem_bytes_ (0)
	count_distinct_bitmap_crt_ptr_ (nullptr)
	count_distinct_bitmap_host_mem_ (nullptr)
	device_allocator_ (device_allocator)
template<typename T >
int8_t *	anonymous_namespace{QueryMemoryInitializer.cpp}::initColumnarBuffer (T *buffer_ptr, const T init_val, const uint32_t entry_count)
void	anonymous_namespace{QueryMemoryInitializer.cpp}::compact_projection_buffer_for_cpu_columnar (const QueryMemoryDescriptor &query_mem_desc, int8_t *projection_buffer, const size_t projection_count)

Variables
	query_mem_desc

Function Documentation

count_distinct_bitmap_crt_ptr_

(

nullptr

)

count_distinct_bitmap_host_mem_

(

nullptr

)

count_distinct_bitmap_mem_

(

0

)

count_distinct_bitmap_mem_bytes_

(

0

)

device_allocator_

(

device_allocator

)

Definition at line 331 of file QueryMemoryInitializer.cpp.

References anonymous_namespace{QueryMemoryInitializer.cpp}::alloc_group_by_buffer(), CHECK_EQ, CHECK_GE, checked_malloc(), anonymous_namespace{QueryMemoryInitializer.cpp}::get_col_frag_offsets(), anonymous_namespace{QueryMemoryInitializer.cpp}::get_consistent_frags_sizes(), GPU, num_buffers_(), query_mem_desc, and target_exprs_to_infos().

Referenced by QueryMemoryInitializer::allocateCountDistinctGpuMem().

                                          {
  // Table functions output columnar, basically treat this as a projection

  const auto& consistent_frag_sizes = get_consistent_frags_sizes(frag_offsets);
  if (consistent_frag_sizes.empty()) {
    // No fragments in the input, no underlying buffers will be needed.
    return;
  }

  size_t group_buffer_size{0};
  // TODO(adb): this is going to give us an index buffer and then the target buffers. this
  // might not be desireable -- revisit
  group_buffer_size = query_mem_desc.getBufferSizeBytes(device_type, num_rows_);
  CHECK_GE(group_buffer_size, size_t(0));

  std::unique_ptr<int64_t, CheckedAllocDeleter> group_by_buffer_template;
  if (!query_mem_desc.lazyInitGroups(device_type)) {
    group_by_buffer_template.reset(
        static_cast<int64_t*>(checked_malloc(group_buffer_size)));
    initColumnarGroups(
        query_mem_desc, group_by_buffer_template.get(), init_agg_vals_, executor);
  }

  const auto index_buffer_qw =
      device_type == ExecutorDeviceType::GPU && query_mem_desc.hasKeylessHash()
          ? query_mem_desc.getEntryCount()
          : size_t(0);
  const auto actual_group_buffer_size =
      group_buffer_size + index_buffer_qw * sizeof(int64_t);
  CHECK_GE(actual_group_buffer_size, group_buffer_size);

  CHECK_EQ(num_buffers_, size_t(1));
  auto group_by_buffer = alloc_group_by_buffer(actual_group_buffer_size, nullptr);
  if (!query_mem_desc.lazyInitGroups(device_type)) {
    memcpy(group_by_buffer + index_buffer_qw,
           group_by_buffer_template.get(),
           group_buffer_size);
  }
  group_by_buffers_.push_back(group_by_buffer);
  row_set_mem_owner_->addGroupByBuffer(group_by_buffer);

  const auto column_frag_offsets =
      get_col_frag_offsets(exe_unit.target_exprs, frag_offsets);
  const auto column_frag_sizes =
      get_consistent_frags_sizes(exe_unit.target_exprs, consistent_frag_sizes);
  result_sets_.emplace_back(
      new ResultSet(target_exprs_to_infos(exe_unit.target_exprs, query_mem_desc),
                    {},
                    col_buffers,
                    column_frag_offsets,
                    column_frag_sizes,
                    device_type,
                    device_id,
                    ResultSet::fixupQueryMemoryDescriptor(query_mem_desc),
                    row_set_mem_owner_,
                    executor));
  result_sets_.back()->allocateStorage(reinterpret_cast<int8_t*>(group_by_buffer),
                                       init_agg_vals_);
}

Here is the call graph for this function:

Here is the caller graph for this function:

num_buffers_

(

1

)

Referenced by device_allocator_().

Here is the caller graph for this function:

Variable Documentation

query_mem_desc

Definition at line 325 of file QueryMemoryInitializer.cpp.

Referenced by SpeculativeTopNMap::asRows(), anonymous_namespace{Execute.cpp}::build_row_for_empty_input(), GroupByAndAggregate::codegen(), QueryCompilationDescriptor::compile(), Executor::ExecutionDispatch::compile(), device_allocator_(), Executor::dispatchFragments(), RelAlgExecutor::executeLogicalValues(), Executor::executeWorkUnitImpl(), ResultSet::fixupQueryMemoryDescriptor(), anonymous_namespace{ResultSetReduction.cpp}::get_matching_group_value_reduction(), GroupByAndAggregate::initQueryMemoryDescriptor(), TableFunctionExecutionContext::launchCpuCode(), TableFunctionExecutionContext::launchGpuCode(), query_group_by_template(), query_group_by_template_impl(), QueryExecutionContext::QueryExecutionContext(), ResultSetManager::reduce(), and Executor::reduceMultiDeviceResultSets().