ResultSetSortImpl.h File Reference

#include "../Shared/TargetInfo.h"
#include "CompilationOptions.h"

Include dependency graph for ResultSetSortImpl.h:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	PodOrderEntry
struct	GroupByBufferLayoutInfo

Namespaces
	Data_Namespace

Functions
template<class K >
std::vector< uint32_t >	baseline_sort (const ExecutorDeviceType device_type, const int device_id, Data_Namespace::DataMgr *data_mgr, const int8_t *groupby_buffer, const PodOrderEntry &oe, const GroupByBufferLayoutInfo &layout, const size_t top_n, const size_t start, const size_t step)

Function Documentation

template<class K >

std::vector<uint32_t> baseline_sort	(	const ExecutorDeviceType	device_type,
		const int	device_id,
		Data_Namespace::DataMgr *	data_mgr,
		const int8_t *	groupby_buffer,
		const PodOrderEntry &	oe,
		const GroupByBufferLayoutInfo &	layout,
		const size_t	top_n,
		const size_t	start,
		const size_t	step
	)

Definition at line 353 of file ResultSetSortImpl.cu.

References CHECK, CPU, get_compact_type(), anonymous_namespace{ResultSetSortImpl.cu}::get_device_copy_ptr(), GPU, PodOrderEntry::is_desc, kAVG, and PodOrderEntry::nulls_first.

                                                       {
  auto oe_col_buffer = collect_order_entry_column<K>(groupby_buffer, layout, start, step);
  const auto& entry_ti = get_compact_type(layout.oe_target_info);
  CHECK(entry_ti.is_number());
  if (entry_ti.is_fp() || layout.oe_target_info.agg_kind == kAVG) {
    return baseline_sort_fp<K>(device_type,
                               device_id,
                               data_mgr,
                               groupby_buffer,
                               oe_col_buffer,
                               oe,
                               layout,
                               top_n,
                               start,
                               step);
  }
  // Because of how we represent nulls for integral types, they'd be at the
  // wrong position in these two cases. Separate them into a different buffer.
  if ((oe.is_desc && oe.nulls_first) || (!oe.is_desc && !oe.nulls_first)) {
    return baseline_sort_int<K>(device_type,
                                device_id,
                                data_mgr,
                                groupby_buffer,
                                oe_col_buffer,
                                oe,
                                layout,
                                top_n,
                                start,
                                step);
  }
  ThrustAllocator thrust_allocator(data_mgr, device_id);
  // Fastest path, no need to separate nulls away since they'll end up at the
  // right place as a side effect of how we're representing nulls.
  if (device_type == ExecutorDeviceType::GPU) {
    if (oe_col_buffer.empty()) {
      return {};
    }
    const auto dev_idx_buff =
        get_device_ptr<uint32_t>(oe_col_buffer.size(), thrust_allocator);
    thrust::sequence(dev_idx_buff, dev_idx_buff + oe_col_buffer.size(), start, step);
    const auto dev_oe_col_buffer = get_device_copy_ptr(oe_col_buffer, thrust_allocator);
    return do_radix_sort<K>(device_type,
                            device_id,
                            thrust_allocator,
                            groupby_buffer,
                            dev_oe_col_buffer,
                            dev_oe_col_buffer + oe_col_buffer.size(),
                            dev_idx_buff,
                            oe_col_buffer.size(),
                            oe,
                            layout,
                            top_n);
  }
  CHECK(device_type == ExecutorDeviceType::CPU);
  thrust::host_vector<uint32_t> host_idx_buff(oe_col_buffer.size());
  thrust::sequence(host_idx_buff.begin(), host_idx_buff.end(), start, step);
  return do_radix_sort<K>(device_type,
                          device_id,
                          thrust_allocator,
                          groupby_buffer,
                          oe_col_buffer.begin(),
                          oe_col_buffer.end(),
                          host_idx_buff.begin(),
                          host_idx_buff.size(),
                          oe,
                          layout,
                          top_n);
}

Here is the call graph for this function: