InPlaceSort.cpp File Reference

#include "InPlaceSort.h"
#include "InPlaceSortImpl.h"
#include <Analyzer/Analyzer.h>
#include "DataMgr/Allocators/ThrustAllocator.h"
#include "Descriptors/QueryMemoryDescriptor.h"
#include "Logger/Logger.h"
#include <cstdint>

Include dependency graph for InPlaceSort.cpp:

Go to the source code of this file.

Namespaces
	anonymous_namespace{InPlaceSort.cpp}

Functions
void	sort_groups_cpu (int64_t *val_buff, int32_t *idx_buff, const uint64_t entry_count, const bool desc, const uint32_t chosen_bytes)
void	apply_permutation_cpu (int64_t *val_buff, int32_t *idx_buff, const uint64_t entry_count, int64_t *tmp_buff, const uint32_t chosen_bytes)
void	anonymous_namespace{InPlaceSort.cpp}::sort_groups_gpu (int64_t *val_buff, int32_t *idx_buff, const uint64_t entry_count, const bool desc, const uint32_t chosen_bytes, ThrustAllocator &alloc, const int device_id)
void	anonymous_namespace{InPlaceSort.cpp}::apply_permutation_gpu (int64_t *val_buff, int32_t *idx_buff, const uint64_t entry_count, const uint32_t chosen_bytes, ThrustAllocator &alloc, const int device_id)
void	inplace_sort_gpu (const std::list< Analyzer::OrderEntry > &order_entries, const QueryMemoryDescriptor &query_mem_desc, const GpuGroupByBuffers &group_by_buffers, Data_Namespace::DataMgr *data_mgr, const int device_id)

Function Documentation

void apply_permutation_cpu	(	int64_t *	val_buff,
		int32_t *	idx_buff,
		const uint64_t	entry_count,
		int64_t *	tmp_buff,
		const uint32_t	chosen_bytes
	)

Definition at line 46 of file InPlaceSort.cpp.

References apply_permutation_on_cpu(), and CHECK.

Referenced by ResultSet::radixSortOnCpu().

                                                        {
#ifdef HAVE_CUDA
  switch (chosen_bytes) {
    case 1:
    case 2:
    case 4:
    case 8:
      apply_permutation_on_cpu(val_buff, idx_buff, entry_count, tmp_buff, chosen_bytes);
      break;
    default:
      CHECK(false);
  }
#endif
}

Here is the call graph for this function:

Here is the caller graph for this function:

void inplace_sort_gpu	(	const std::list< Analyzer::OrderEntry > &	order_entries,
		const QueryMemoryDescriptor &	query_mem_desc,
		const GpuGroupByBuffers &	group_by_buffers,
		Data_Namespace::DataMgr *	data_mgr,
		const int	device_id
	)

Definition at line 111 of file InPlaceSort.cpp.

References align_to_int64(), anonymous_namespace{InPlaceSort.cpp}::apply_permutation_gpu(), CHECK_EQ, GpuGroupByBuffers::data, QueryMemoryDescriptor::getColOffInBytes(), QueryMemoryDescriptor::getEntryCount(), QueryMemoryDescriptor::getPaddedSlotWidthBytes(), QueryMemoryDescriptor::getSlotCount(), QueryMemoryDescriptor::hasKeylessHash(), and anonymous_namespace{InPlaceSort.cpp}::sort_groups_gpu().

Referenced by QueryExecutionContext::launchGpuCode(), and ResultSet::radixSortOnGpu().

                                           {
  ThrustAllocator alloc(data_mgr, device_id);
  CHECK_EQ(size_t(1), order_entries.size());
  const auto idx_buff = group_by_buffers.data -
                        align_to_int64(query_mem_desc.getEntryCount() * sizeof(int32_t));
  for (const auto& order_entry : order_entries) {
    const auto target_idx = order_entry.tle_no - 1;
    const auto val_buff =
        group_by_buffers.data + query_mem_desc.getColOffInBytes(target_idx);
    const auto chosen_bytes = query_mem_desc.getPaddedSlotWidthBytes(target_idx);
    sort_groups_gpu(reinterpret_cast<int64_t*>(val_buff),
                    reinterpret_cast<int32_t*>(idx_buff),
                    query_mem_desc.getEntryCount(),
                    order_entry.is_desc,
                    chosen_bytes,
                    alloc,
                    device_id);
    if (!query_mem_desc.hasKeylessHash()) {
      apply_permutation_gpu(reinterpret_cast<int64_t*>(group_by_buffers.data),
                            reinterpret_cast<int32_t*>(idx_buff),
                            query_mem_desc.getEntryCount(),
                            sizeof(int64_t),
                            alloc,
                            device_id);
    }
    for (size_t target_idx = 0; target_idx < query_mem_desc.getSlotCount();
         ++target_idx) {
      if (static_cast<int>(target_idx) == order_entry.tle_no - 1) {
        continue;
      }
      const auto chosen_bytes = query_mem_desc.getPaddedSlotWidthBytes(target_idx);
      const auto val_buff =
          group_by_buffers.data + query_mem_desc.getColOffInBytes(target_idx);
      apply_permutation_gpu(reinterpret_cast<int64_t*>(val_buff),
                            reinterpret_cast<int32_t*>(idx_buff),
                            query_mem_desc.getEntryCount(),
                            chosen_bytes,
                            alloc,
                            device_id);
    }
  }
}

Here is the call graph for this function:

Here is the caller graph for this function:

void sort_groups_cpu	(	int64_t *	val_buff,
		int32_t *	idx_buff,
		const uint64_t	entry_count,
		const bool	desc,
		const uint32_t	chosen_bytes
	)

Definition at line 27 of file InPlaceSort.cpp.

References CHECK, and sort_on_cpu().

Referenced by ResultSet::radixSortOnCpu().

                                                  {
#ifdef HAVE_CUDA
  switch (chosen_bytes) {
    case 1:
    case 2:
    case 4:
    case 8:
      sort_on_cpu(val_buff, idx_buff, entry_count, desc, chosen_bytes);
      break;
    default:
      CHECK(false);
  }
#endif
}

Here is the call graph for this function:

Here is the caller graph for this function: