OmniSciDB  fe05a0c208
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
QueryMemoryDescriptor.cpp
Go to the documentation of this file.
1 /*
2  * Copyright 2018 MapD Technologies, Inc.
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  * http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "QueryMemoryDescriptor.h"
18 
19 #include "../Execute.h"
20 #include "../ExpressionRewrite.h"
21 #include "../GroupByAndAggregate.h"
22 #include "../StreamingTopN.h"
23 #include "../UsedColumnsVisitor.h"
24 #include "ColSlotContext.h"
25 
26 #include <boost/algorithm/cxx11/any_of.hpp>
27 
29 extern bool g_enable_columnar_output;
30 
31 namespace {
32 
33 bool is_int_and_no_bigger_than(const SQLTypeInfo& ti, const size_t byte_width) {
34  if (!ti.is_integer()) {
35  return false;
36  }
37  return get_bit_width(ti) <= (byte_width * 8);
38 }
39 
41  return range.getIntMin() > INT32_MIN && range.getIntMax() < EMPTY_KEY_32 - 1;
42 }
43 
44 std::vector<int64_t> target_expr_group_by_indices(
45  const std::list<std::shared_ptr<Analyzer::Expr>>& groupby_exprs,
46  const std::vector<Analyzer::Expr*>& target_exprs) {
47  std::vector<int64_t> indices(target_exprs.size(), -1);
48  for (size_t target_idx = 0; target_idx < target_exprs.size(); ++target_idx) {
49  const auto target_expr = target_exprs[target_idx];
50  if (dynamic_cast<const Analyzer::AggExpr*>(target_expr)) {
51  continue;
52  }
53  const auto var_expr = dynamic_cast<const Analyzer::Var*>(target_expr);
54  if (var_expr && var_expr->get_which_row() == Analyzer::Var::kGROUPBY) {
55  indices[target_idx] = var_expr->get_varno() - 1;
56  continue;
57  }
58  }
59  return indices;
60 }
61 
62 std::vector<int64_t> target_expr_proj_indices(const RelAlgExecutionUnit& ra_exe_unit,
64  if (ra_exe_unit.input_descs.size() > 1 ||
65  !ra_exe_unit.sort_info.order_entries.empty()) {
66  return {};
67  }
68  std::vector<int64_t> target_indices(ra_exe_unit.target_exprs.size(), -1);
69  UsedColumnsVisitor columns_visitor;
70  std::unordered_set<int> used_columns;
71  for (const auto& simple_qual : ra_exe_unit.simple_quals) {
72  const auto crt_used_columns = columns_visitor.visit(simple_qual.get());
73  used_columns.insert(crt_used_columns.begin(), crt_used_columns.end());
74  }
75  for (const auto& qual : ra_exe_unit.quals) {
76  const auto crt_used_columns = columns_visitor.visit(qual.get());
77  used_columns.insert(crt_used_columns.begin(), crt_used_columns.end());
78  }
79  for (const auto& target : ra_exe_unit.target_exprs) {
80  const auto col_var = dynamic_cast<const Analyzer::ColumnVar*>(target);
81  if (col_var) {
82  const auto cd = get_column_descriptor_maybe(
83  col_var->get_column_id(), col_var->get_table_id(), cat);
84  if (!cd || !cd->isVirtualCol) {
85  continue;
86  }
87  }
88  const auto crt_used_columns = columns_visitor.visit(target);
89  used_columns.insert(crt_used_columns.begin(), crt_used_columns.end());
90  }
91  for (size_t target_idx = 0; target_idx < ra_exe_unit.target_exprs.size();
92  ++target_idx) {
93  const auto target_expr = ra_exe_unit.target_exprs[target_idx];
94  CHECK(target_expr);
95  const auto& ti = target_expr->get_type_info();
96  // TODO: add proper lazy fetch for varlen types in result set
97  if (ti.is_varlen()) {
98  continue;
99  }
100  const auto col_var = dynamic_cast<const Analyzer::ColumnVar*>(target_expr);
101  if (!col_var) {
102  continue;
103  }
104  if (!ti.is_varlen() &&
105  used_columns.find(col_var->get_column_id()) == used_columns.end()) {
106  // setting target index to be zero so that later it can be decoded properly (in lazy
107  // fetch, the zeroth target index indicates the corresponding rowid column for the
108  // projected entry)
109  target_indices[target_idx] = 0;
110  }
111  }
112  return target_indices;
113 }
114 
116  const size_t group_col_width) {
117  if (range.getType() == ExpressionRangeType::Invalid) {
118  return sizeof(int64_t);
119  }
120  switch (range.getType()) {
122  if (group_col_width == sizeof(int64_t) && range.hasNulls()) {
123  return sizeof(int64_t);
124  }
125  return is_valid_int32_range(range) ? sizeof(int32_t) : sizeof(int64_t);
128  return sizeof(int64_t); // No compaction for floating point yet.
129  default:
130  UNREACHABLE();
131  }
132  return sizeof(int64_t);
133 }
134 
135 // TODO(miyu): make sure following setting of compact width is correct in all cases.
137  const std::vector<InputTableInfo>& query_infos,
138  const Executor* executor) {
139  int8_t compact_width{4};
140  for (const auto& groupby_expr : ra_exe_unit.groupby_exprs) {
141  const auto expr_range = getExpressionRange(groupby_expr.get(), query_infos, executor);
142  compact_width = std::max(compact_width,
144  expr_range, groupby_expr->get_type_info().get_size()));
145  }
146  return compact_width;
147 }
148 
149 bool use_streaming_top_n(const RelAlgExecutionUnit& ra_exe_unit,
150  const bool output_columnar) {
151  if (g_cluster) {
152  return false; // TODO(miyu)
153  }
154 
155  for (const auto target_expr : ra_exe_unit.target_exprs) {
156  if (dynamic_cast<const Analyzer::AggExpr*>(target_expr)) {
157  return false;
158  }
159  if (dynamic_cast<const Analyzer::WindowFunction*>(target_expr)) {
160  return false;
161  }
162  }
163 
164  // TODO: Allow streaming top n for columnar output
165  if (!output_columnar && ra_exe_unit.sort_info.order_entries.size() == 1 &&
166  ra_exe_unit.sort_info.limit &&
168  const auto only_order_entry = ra_exe_unit.sort_info.order_entries.front();
169  CHECK_GT(only_order_entry.tle_no, int(0));
170  CHECK_LE(static_cast<size_t>(only_order_entry.tle_no),
171  ra_exe_unit.target_exprs.size());
172  const auto order_entry_expr = ra_exe_unit.target_exprs[only_order_entry.tle_no - 1];
173  const auto n = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
174  if ((order_entry_expr->get_type_info().is_number() ||
175  order_entry_expr->get_type_info().is_time()) &&
176  n <= 100000) { // TODO(miyu): relax?
177  return true;
178  }
179  }
180 
181  return false;
182 }
183 
184 template <class T>
185 inline std::vector<int8_t> get_col_byte_widths(const T& col_expr_list) {
186  std::vector<int8_t> col_widths;
187  size_t col_expr_idx = 0;
188  for (const auto col_expr : col_expr_list) {
189  if (!col_expr) {
190  // row index
191  col_widths.push_back(sizeof(int64_t));
192  } else {
193  const auto agg_info = get_target_info(col_expr, g_bigint_count);
194  const auto chosen_type = get_compact_type(agg_info);
195  if ((chosen_type.is_string() && chosen_type.get_compression() == kENCODING_NONE) ||
196  chosen_type.is_array()) {
197  col_widths.push_back(sizeof(int64_t));
198  col_widths.push_back(sizeof(int64_t));
199  ++col_expr_idx;
200  continue;
201  }
202  if (chosen_type.is_geometry()) {
203  for (auto i = 0; i < chosen_type.get_physical_coord_cols(); ++i) {
204  col_widths.push_back(sizeof(int64_t));
205  col_widths.push_back(sizeof(int64_t));
206  }
207  ++col_expr_idx;
208  continue;
209  }
210  const auto col_expr_bitwidth = get_bit_width(chosen_type);
211  CHECK_EQ(size_t(0), col_expr_bitwidth % 8);
212  col_widths.push_back(static_cast<int8_t>(col_expr_bitwidth >> 3));
213  // for average, we'll need to keep the count as well
214  if (agg_info.agg_kind == kAVG) {
215  CHECK(agg_info.is_agg);
216  col_widths.push_back(sizeof(int64_t));
217  }
218  }
219  ++col_expr_idx;
220  }
221  return col_widths;
222 }
223 
224 } // namespace
225 
226 std::unique_ptr<QueryMemoryDescriptor> QueryMemoryDescriptor::init(
227  const Executor* executor,
228  const RelAlgExecutionUnit& ra_exe_unit,
229  const std::vector<InputTableInfo>& query_infos,
230  const ColRangeInfo& col_range_info,
231  const KeylessInfo& keyless_info,
232  const bool allow_multifrag,
233  const ExecutorDeviceType device_type,
234  const int8_t crt_min_byte_width,
235  const bool sort_on_gpu_hint,
236  const size_t shard_count,
237  const size_t max_groups_buffer_entry_count,
238  RenderInfo* render_info,
239  const CountDistinctDescriptors count_distinct_descriptors,
240  const bool must_use_baseline_sort,
241  const bool output_columnar_hint,
242  const bool streaming_top_n_hint) {
243  auto group_col_widths = get_col_byte_widths(ra_exe_unit.groupby_exprs);
244  const bool is_group_by{!group_col_widths.empty()};
245 
246  auto col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, {});
247 
248  const auto min_slot_size = QueryMemoryDescriptor::pick_target_compact_width(
249  ra_exe_unit, query_infos, crt_min_byte_width);
250 
251  col_slot_context.setAllSlotsPaddedSize(min_slot_size);
252  col_slot_context.validate();
253 
254  if (!is_group_by) {
255  CHECK(!must_use_baseline_sort);
256 
257  return std::make_unique<QueryMemoryDescriptor>(
258  executor,
259  ra_exe_unit,
260  query_infos,
261  allow_multifrag,
262  false,
263  false,
264  -1,
265  ColRangeInfo{ra_exe_unit.estimator ? QueryDescriptionType::Estimator
267  0,
268  0,
269  0,
270  false},
271  col_slot_context,
272  std::vector<int8_t>{},
273  /*group_col_compact_width=*/0,
274  std::vector<int64_t>{},
275  /*entry_count=*/1,
276  count_distinct_descriptors,
277  false,
278  output_columnar_hint,
279  render_info && render_info->isPotentialInSituRender(),
280  must_use_baseline_sort,
281  /*use_streaming_top_n=*/false);
282  }
283 
284  size_t entry_count = 1;
285  auto actual_col_range_info = col_range_info;
286  bool interleaved_bins_on_gpu = false;
287  bool keyless_hash = false;
288  bool streaming_top_n = false;
289  int8_t group_col_compact_width = 0;
290  int32_t idx_target_as_key = -1;
291  auto output_columnar = output_columnar_hint;
292  std::vector<int64_t> target_groupby_indices;
293 
294  switch (col_range_info.hash_type_) {
296  if (render_info) {
297  render_info->setInSituDataIfUnset(false);
298  }
299  // keyless hash: whether or not group columns are stored at the beginning of the
300  // output buffer
301  keyless_hash =
302  (!sort_on_gpu_hint ||
304  col_range_info.max, col_range_info.min, col_range_info.bucket)) &&
305  !col_range_info.bucket && !must_use_baseline_sort && keyless_info.keyless;
306 
307  // if keyless, then this target index indicates wheter an entry is empty or not
308  // (acts as a key)
309  idx_target_as_key = keyless_info.target_index;
310 
311  if (group_col_widths.size() > 1) {
312  // col range info max contains the expected cardinality of the output
313  entry_count = static_cast<size_t>(actual_col_range_info.max);
314  actual_col_range_info.bucket = 0;
315  } else {
316  // single column perfect hash
317  entry_count = std::max(
318  GroupByAndAggregate::getBucketedCardinality(col_range_info), int64_t(1));
319  const size_t interleaved_max_threshold{512};
320 
321  if (must_use_baseline_sort) {
322  target_groupby_indices = target_expr_group_by_indices(ra_exe_unit.groupby_exprs,
323  ra_exe_unit.target_exprs);
324  col_slot_context =
325  ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
326  }
327 
328  bool has_varlen_sample_agg = false;
329  for (const auto& target_expr : ra_exe_unit.target_exprs) {
330  if (target_expr->get_contains_agg()) {
331  const auto agg_expr = dynamic_cast<Analyzer::AggExpr*>(target_expr);
332  CHECK(agg_expr);
333  if (agg_expr->get_aggtype() == kSAMPLE &&
334  agg_expr->get_type_info().is_varlen()) {
335  has_varlen_sample_agg = true;
336  break;
337  }
338  }
339  }
340 
341  interleaved_bins_on_gpu = keyless_hash && !has_varlen_sample_agg &&
342  (entry_count <= interleaved_max_threshold) &&
343  (device_type == ExecutorDeviceType::GPU) &&
345  count_distinct_descriptors) &&
346  !output_columnar;
347  }
348  break;
349  }
351  if (render_info) {
352  render_info->setInSituDataIfUnset(false);
353  }
354  entry_count = shard_count
355  ? (max_groups_buffer_entry_count + shard_count - 1) / shard_count
356  : max_groups_buffer_entry_count;
357  target_groupby_indices = target_expr_group_by_indices(ra_exe_unit.groupby_exprs,
358  ra_exe_unit.target_exprs);
359  col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
360 
361  group_col_compact_width =
362  output_columnar ? 8
363  : pick_baseline_key_width(ra_exe_unit, query_infos, executor);
364 
365  actual_col_range_info =
367  break;
368  }
370  CHECK(!must_use_baseline_sort);
371 
372  if (streaming_top_n_hint && use_streaming_top_n(ra_exe_unit, output_columnar)) {
373  streaming_top_n = true;
374  entry_count = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
375  } else {
376  if (ra_exe_unit.use_bump_allocator) {
377  output_columnar = false;
378  entry_count = 0;
379  } else {
380  entry_count = ra_exe_unit.scan_limit
381  ? static_cast<size_t>(ra_exe_unit.scan_limit)
382  : max_groups_buffer_entry_count;
383  }
384  }
385 
386  const auto catalog = executor->getCatalog();
387  CHECK(catalog);
388  target_groupby_indices = executor->plan_state_->allow_lazy_fetch_
389  ? target_expr_proj_indices(ra_exe_unit, *catalog)
390  : std::vector<int64_t>{};
391 
392  col_slot_context = ColSlotContext(ra_exe_unit.target_exprs, target_groupby_indices);
393  break;
394  }
395  default:
396  UNREACHABLE() << "Unknown query type";
397  }
398 
399  return std::make_unique<QueryMemoryDescriptor>(
400  executor,
401  ra_exe_unit,
402  query_infos,
403  allow_multifrag,
404  keyless_hash,
405  interleaved_bins_on_gpu,
406  idx_target_as_key,
407  actual_col_range_info,
408  col_slot_context,
409  group_col_widths,
410  group_col_compact_width,
411  target_groupby_indices,
412  entry_count,
413  count_distinct_descriptors,
414  sort_on_gpu_hint,
415  output_columnar,
416  render_info && render_info->isPotentialInSituRender(),
417  must_use_baseline_sort,
418  streaming_top_n);
419 }
420 
421 namespace {
422 bool anyOf(std::vector<Analyzer::Expr*> const& target_exprs, SQLAgg const agg_kind) {
423  return boost::algorithm::any_of(target_exprs, [agg_kind](Analyzer::Expr const* expr) {
424  auto const* const agg = dynamic_cast<Analyzer::AggExpr const*>(expr);
425  return agg && agg->get_aggtype() == agg_kind;
426  });
427 }
428 } // namespace
429 
431  const Executor* executor,
432  const RelAlgExecutionUnit& ra_exe_unit,
433  const std::vector<InputTableInfo>& query_infos,
434  const bool allow_multifrag,
435  const bool keyless_hash,
436  const bool interleaved_bins_on_gpu,
437  const int32_t idx_target_as_key,
438  const ColRangeInfo& col_range_info,
439  const ColSlotContext& col_slot_context,
440  const std::vector<int8_t>& group_col_widths,
441  const int8_t group_col_compact_width,
442  const std::vector<int64_t>& target_groupby_indices,
443  const size_t entry_count,
444  const CountDistinctDescriptors count_distinct_descriptors,
445  const bool sort_on_gpu_hint,
446  const bool output_columnar_hint,
447  const bool render_output,
448  const bool must_use_baseline_sort,
449  const bool use_streaming_top_n)
450  : executor_(executor)
451  , allow_multifrag_(allow_multifrag)
452  , query_desc_type_(col_range_info.hash_type_)
453  , keyless_hash_(keyless_hash)
454  , interleaved_bins_on_gpu_(interleaved_bins_on_gpu)
455  , idx_target_as_key_(idx_target_as_key)
456  , group_col_widths_(group_col_widths)
457  , group_col_compact_width_(group_col_compact_width)
458  , target_groupby_indices_(target_groupby_indices)
459  , entry_count_(entry_count)
460  , min_val_(col_range_info.min)
461  , max_val_(col_range_info.max)
462  , bucket_(col_range_info.bucket)
463  , has_nulls_(col_range_info.has_nulls)
464  , count_distinct_descriptors_(count_distinct_descriptors)
465  , output_columnar_(false)
466  , render_output_(render_output)
467  , must_use_baseline_sort_(must_use_baseline_sort)
468  , is_table_function_(false)
469  , use_streaming_top_n_(use_streaming_top_n)
470  , force_4byte_float_(false)
471  , col_slot_context_(col_slot_context) {
474 
475  sort_on_gpu_ = sort_on_gpu_hint && canOutputColumnar() && !keyless_hash_;
476 
477  if (sort_on_gpu_) {
478  CHECK(!ra_exe_unit.use_bump_allocator);
479  output_columnar_ = true;
480  } else {
481  switch (query_desc_type_) {
483  output_columnar_ = output_columnar_hint;
484  break;
486  output_columnar_ = output_columnar_hint &&
489  !anyOf(ra_exe_unit.target_exprs, kAPPROX_MEDIAN);
490  break;
492  output_columnar_ = output_columnar_hint;
493  break;
495  output_columnar_ = output_columnar_hint &&
498  !anyOf(ra_exe_unit.target_exprs, kAPPROX_MEDIAN);
499  break;
500  default:
501  output_columnar_ = false;
502  break;
503  }
504  }
505 
507  // TODO(adb): Ensure fixed size buffer allocations are correct with all logical column
508  // sizes
509  CHECK(!ra_exe_unit.use_bump_allocator);
512  }
513 
514 #ifdef HAVE_CUDA
515  // Check Streaming Top N heap usage, bail if > max slab size, CUDA ONLY
516  if (use_streaming_top_n_ && executor->catalog_->getDataMgr().gpusPresent()) {
517  const auto thread_count = executor->blockSize() * executor->gridSize();
518  const auto total_buff_size =
520  if (total_buff_size > executor_->maxGpuSlabSize()) {
521  throw StreamingTopNOOM(total_buff_size);
522  }
523  }
524 #endif
525 }
526 
528  : executor_(nullptr)
529  , allow_multifrag_(false)
530  , query_desc_type_(QueryDescriptionType::Projection)
531  , keyless_hash_(false)
532  , interleaved_bins_on_gpu_(false)
533  , idx_target_as_key_(0)
534  , group_col_compact_width_(0)
535  , entry_count_(0)
536  , min_val_(0)
537  , max_val_(0)
538  , bucket_(0)
539  , has_nulls_(false)
540  , sort_on_gpu_(false)
541  , output_columnar_(false)
542  , render_output_(false)
543  , must_use_baseline_sort_(false)
544  , is_table_function_(false)
545  , use_streaming_top_n_(false)
546  , force_4byte_float_(false) {}
547 
549  const size_t entry_count,
550  const QueryDescriptionType query_desc_type,
551  const bool is_table_function)
552  : executor_(executor)
553  , allow_multifrag_(false)
554  , query_desc_type_(query_desc_type)
555  , keyless_hash_(false)
556  , interleaved_bins_on_gpu_(false)
557  , idx_target_as_key_(0)
558  , group_col_compact_width_(0)
559  , entry_count_(entry_count)
560  , min_val_(0)
561  , max_val_(0)
562  , bucket_(0)
563  , has_nulls_(false)
564  , sort_on_gpu_(false)
565  , output_columnar_(false)
566  , render_output_(false)
567  , must_use_baseline_sort_(false)
568  , is_table_function_(is_table_function)
569  , use_streaming_top_n_(false)
570  , force_4byte_float_(false) {}
571 
573  const int64_t min_val,
574  const int64_t max_val,
575  const bool has_nulls,
576  const std::vector<int8_t>& group_col_widths)
577  : executor_(nullptr)
578  , allow_multifrag_(false)
579  , query_desc_type_(query_desc_type)
580  , keyless_hash_(false)
581  , interleaved_bins_on_gpu_(false)
582  , idx_target_as_key_(0)
583  , group_col_widths_(group_col_widths)
584  , group_col_compact_width_(0)
585  , entry_count_(0)
586  , min_val_(min_val)
587  , max_val_(max_val)
588  , bucket_(0)
589  , has_nulls_(false)
590  , sort_on_gpu_(false)
591  , output_columnar_(false)
592  , render_output_(false)
593  , must_use_baseline_sort_(false)
594  , is_table_function_(false)
595  , use_streaming_top_n_(false)
596  , force_4byte_float_(false) {}
597 
599  // Note that this method does not check ptr reference members (e.g. executor_) or
600  // entry_count_
601  if (query_desc_type_ != other.query_desc_type_) {
602  return false;
603  }
604  if (keyless_hash_ != other.keyless_hash_) {
605  return false;
606  }
608  return false;
609  }
610  if (idx_target_as_key_ != other.idx_target_as_key_) {
611  return false;
612  }
613  if (force_4byte_float_ != other.force_4byte_float_) {
614  return false;
615  }
616  if (group_col_widths_ != other.group_col_widths_) {
617  return false;
618  }
620  return false;
621  }
623  return false;
624  }
625  if (min_val_ != other.min_val_) {
626  return false;
627  }
628  if (max_val_ != other.max_val_) {
629  return false;
630  }
631  if (bucket_ != other.bucket_) {
632  return false;
633  }
634  if (has_nulls_ != other.has_nulls_) {
635  return false;
636  }
638  return false;
639  } else {
640  // Count distinct descriptors can legitimately differ in device only.
641  for (size_t i = 0; i < count_distinct_descriptors_.size(); ++i) {
642  auto ref_count_distinct_desc = other.count_distinct_descriptors_[i];
643  auto count_distinct_desc = count_distinct_descriptors_[i];
644  count_distinct_desc.device_type = ref_count_distinct_desc.device_type;
645  if (ref_count_distinct_desc != count_distinct_desc) {
646  return false;
647  }
648  }
649  }
650  if (sort_on_gpu_ != other.sort_on_gpu_) {
651  return false;
652  }
653  if (output_columnar_ != other.output_columnar_) {
654  return false;
655  }
656  if (col_slot_context_ != other.col_slot_context_) {
657  return false;
658  }
659  return true;
660 }
661 
662 std::unique_ptr<QueryExecutionContext> QueryMemoryDescriptor::getQueryExecutionContext(
663  const RelAlgExecutionUnit& ra_exe_unit,
664  const Executor* executor,
665  const ExecutorDeviceType device_type,
666  const ExecutorDispatchMode dispatch_mode,
667  const int device_id,
668  const int64_t num_rows,
669  const std::vector<std::vector<const int8_t*>>& col_buffers,
670  const std::vector<std::vector<uint64_t>>& frag_offsets,
671  std::shared_ptr<RowSetMemoryOwner> row_set_mem_owner,
672  const bool output_columnar,
673  const bool sort_on_gpu,
674  const size_t thread_idx,
675  RenderInfo* render_info) const {
676  auto timer = DEBUG_TIMER(__func__);
677  if (frag_offsets.empty()) {
678  return nullptr;
679  }
680  return std::unique_ptr<QueryExecutionContext>(
681  new QueryExecutionContext(ra_exe_unit,
682  *this,
683  executor,
684  device_type,
685  dispatch_mode,
686  device_id,
687  num_rows,
688  col_buffers,
689  frag_offsets,
690  row_set_mem_owner,
691  output_columnar,
692  sort_on_gpu,
693  thread_idx,
694  render_info));
695 }
696 
698  const RelAlgExecutionUnit& ra_exe_unit,
699  const std::vector<InputTableInfo>& query_infos,
700  const int8_t crt_min_byte_width) {
701  if (g_bigint_count) {
702  return sizeof(int64_t);
703  }
704  int8_t compact_width{0};
705  auto col_it = ra_exe_unit.input_col_descs.begin();
706  int unnest_array_col_id{std::numeric_limits<int>::min()};
707  for (const auto& groupby_expr : ra_exe_unit.groupby_exprs) {
708  const auto uoper = dynamic_cast<Analyzer::UOper*>(groupby_expr.get());
709  if (uoper && uoper->get_optype() == kUNNEST) {
710  const auto& arg_ti = uoper->get_operand()->get_type_info();
711  CHECK(arg_ti.is_array());
712  const auto& elem_ti = arg_ti.get_elem_type();
713  if (elem_ti.is_string() && elem_ti.get_compression() == kENCODING_DICT) {
714  unnest_array_col_id = (*col_it)->getColId();
715  } else {
716  compact_width = crt_min_byte_width;
717  break;
718  }
719  }
720  ++col_it;
721  }
722  if (!compact_width &&
723  (ra_exe_unit.groupby_exprs.size() != 1 || !ra_exe_unit.groupby_exprs.front())) {
724  compact_width = crt_min_byte_width;
725  }
726  if (!compact_width) {
727  col_it = ra_exe_unit.input_col_descs.begin();
728  std::advance(col_it, ra_exe_unit.groupby_exprs.size());
729  for (const auto target : ra_exe_unit.target_exprs) {
730  const auto& ti = target->get_type_info();
731  const auto agg = dynamic_cast<const Analyzer::AggExpr*>(target);
732  if (agg && agg->get_arg()) {
733  compact_width = crt_min_byte_width;
734  break;
735  }
736 
737  if (agg) {
738  CHECK_EQ(kCOUNT, agg->get_aggtype());
739  CHECK(!agg->get_is_distinct());
740  ++col_it;
741  continue;
742  }
743 
744  if (is_int_and_no_bigger_than(ti, 4) ||
745  (ti.is_string() && ti.get_compression() == kENCODING_DICT)) {
746  ++col_it;
747  continue;
748  }
749 
750  const auto uoper = dynamic_cast<Analyzer::UOper*>(target);
751  if (uoper && uoper->get_optype() == kUNNEST &&
752  (*col_it)->getColId() == unnest_array_col_id) {
753  const auto arg_ti = uoper->get_operand()->get_type_info();
754  CHECK(arg_ti.is_array());
755  const auto& elem_ti = arg_ti.get_elem_type();
756  if (elem_ti.is_string() && elem_ti.get_compression() == kENCODING_DICT) {
757  ++col_it;
758  continue;
759  }
760  }
761 
762  compact_width = crt_min_byte_width;
763  break;
764  }
765  }
766  if (!compact_width) {
767  size_t total_tuples{0};
768  for (const auto& qi : query_infos) {
769  total_tuples += qi.info.getNumTuples();
770  }
771  return total_tuples <= static_cast<size_t>(std::numeric_limits<uint32_t>::max()) ||
772  unnest_array_col_id != std::numeric_limits<int>::min()
773  ? 4
774  : crt_min_byte_width;
775  } else {
776  // TODO(miyu): relax this condition to allow more cases just w/o padding
777  for (auto wid : get_col_byte_widths(ra_exe_unit.target_exprs)) {
778  compact_width = std::max(compact_width, wid);
779  }
780  return compact_width;
781  }
782 }
783 
786 }
787 
790  size_t total_bytes{0};
791  if (keyless_hash_) {
792  // ignore, there's no group column in the output buffer
794  } else {
795  total_bytes += group_col_widths_.size() * getEffectiveKeyWidth();
796  total_bytes = align_to_int64(total_bytes);
797  }
798  total_bytes += getColsSize();
799  return align_to_int64(total_bytes);
800 }
801 
803  return (interleaved_bins_on_gpu_ ? executor_->warpSize() : 1);
804 }
805 
808 }
809 
818 }
819 
825  const size_t num_entries_per_column) const {
826  return col_slot_context_.getTotalBytesOfColumnarBuffers(num_entries_per_column);
827 }
828 
839  const size_t projection_count) const {
840  constexpr size_t row_index_width = sizeof(int64_t);
841  return getTotalBytesOfColumnarBuffers(projection_count) +
842  row_index_width * projection_count;
843 }
844 
845 size_t QueryMemoryDescriptor::getColOnlyOffInBytes(const size_t col_idx) const {
846  return col_slot_context_.getColOnlyOffInBytes(col_idx);
847 }
848 
849 /*
850  * Returns the memory offset in bytes for a specific agg column in the output
851  * memory buffer. Depending on the query type, there may be some extra portion
852  * of memory prepended at the beginning of the buffer. A brief description of
853  * the memory layout is as follows:
854  * 1. projections: index column (64bit) + all target columns
855  * 2. group by: all group columns (64-bit each) + all agg columns
856  * 2a. if keyless, there is no prepending group column stored at the beginning
857  */
858 size_t QueryMemoryDescriptor::getColOffInBytes(const size_t col_idx) const {
859  const auto warp_count = getWarpCount();
860  if (output_columnar_) {
861  CHECK_EQ(size_t(1), warp_count);
862  size_t offset{0};
863  if (!keyless_hash_) {
865  }
866  for (size_t index = 0; index < col_idx; ++index) {
868  }
869  return offset;
870  }
871 
872  size_t offset{0};
873  if (keyless_hash_) {
874  // ignore, there's no group column in the output buffer
876  } else {
877  offset += group_col_widths_.size() * getEffectiveKeyWidth();
878  offset = align_to_int64(offset);
879  }
880  offset += getColOnlyOffInBytes(col_idx);
881  return offset;
882 }
883 
884 /*
885  * Returns the memory offset for a particular group column in the prepended group
886  * columns portion of the memory.
887  */
889  const size_t group_idx) const {
891  CHECK(group_idx < getGroupbyColCount());
892  size_t offset{0};
893  for (size_t col_idx = 0; col_idx < group_idx; col_idx++) {
894  // TODO(Saman): relax that int64_bit part immediately
895  offset += align_to_int64(
896  std::max(groupColWidth(col_idx), static_cast<int8_t>(sizeof(int64_t))) *
897  getEntryCount());
898  }
899  return offset;
900 }
901 
902 /*
903  * Returns total amount of memory prepended at the beginning of the output memory
904  * buffer.
905  */
908  size_t buffer_size{0};
909  for (size_t group_idx = 0; group_idx < getGroupbyColCount(); group_idx++) {
910  buffer_size += align_to_int64(
911  std::max(groupColWidth(group_idx), static_cast<int8_t>(sizeof(int64_t))) *
912  getEntryCount());
913  }
914  return buffer_size;
915 }
916 
917 size_t QueryMemoryDescriptor::getColOffInBytesInNextBin(const size_t col_idx) const {
918  auto warp_count = getWarpCount();
919  if (output_columnar_) {
920  CHECK_EQ(size_t(1), group_col_widths_.size());
921  CHECK_EQ(size_t(1), warp_count);
922  return getPaddedSlotWidthBytes(col_idx);
923  }
924 
925  return warp_count * getRowSize();
926 }
927 
928 size_t QueryMemoryDescriptor::getNextColOffInBytes(const int8_t* col_ptr,
929  const size_t bin,
930  const size_t col_idx) const {
932  size_t offset{0};
933  auto warp_count = getWarpCount();
934  const auto chosen_bytes = getPaddedSlotWidthBytes(col_idx);
935  const auto total_slot_count = getSlotCount();
936  if (col_idx + 1 == total_slot_count) {
937  if (output_columnar_) {
938  return (entry_count_ - bin) * chosen_bytes;
939  } else {
940  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
941  }
942  }
943 
944  const auto next_chosen_bytes = getPaddedSlotWidthBytes(col_idx + 1);
945  if (output_columnar_) {
946  CHECK_EQ(size_t(1), group_col_widths_.size());
947  CHECK_EQ(size_t(1), warp_count);
948 
949  offset = align_to_int64(entry_count_ * chosen_bytes);
950 
951  offset += bin * (next_chosen_bytes - chosen_bytes);
952  return offset;
953  }
954 
955  if (next_chosen_bytes == sizeof(int64_t)) {
956  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
957  } else {
958  return chosen_bytes;
959  }
960 }
961 
963  const size_t col_idx) const {
964  const auto chosen_bytes = getPaddedSlotWidthBytes(col_idx);
965  const auto total_slot_count = getSlotCount();
966  if (col_idx + 1 == total_slot_count) {
967  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
968  }
969 
970  const auto next_chosen_bytes = getPaddedSlotWidthBytes(col_idx + 1);
971 
972  if (next_chosen_bytes == sizeof(int64_t)) {
973  return static_cast<size_t>(align_to_int64(col_ptr + chosen_bytes) - col_ptr);
974  } else {
975  return chosen_bytes;
976  }
977 }
978 
980  const RelAlgExecutionUnit& ra_exe_unit,
981  const unsigned thread_count,
982  const ExecutorDeviceType device_type) const {
983  if (use_streaming_top_n_) {
984  const size_t n = ra_exe_unit.sort_info.offset + ra_exe_unit.sort_info.limit;
985  return streaming_top_n::get_heap_size(getRowSize(), n, thread_count);
986  }
987  return getBufferSizeBytes(device_type, entry_count_);
988 }
989 
1002  const size_t entry_count) const {
1003  if (keyless_hash_ && !output_columnar_) {
1004  CHECK_GE(group_col_widths_.size(), size_t(1));
1005  auto row_bytes = align_to_int64(getColsSize());
1006 
1007  return (interleavedBins(device_type) ? executor_->warpSize() : 1) * entry_count *
1008  row_bytes;
1009  }
1010 
1011  constexpr size_t row_index_width = sizeof(int64_t);
1012  size_t total_bytes{0};
1013  if (output_columnar_) {
1015  ? row_index_width * entry_count
1016  : sizeof(int64_t) * group_col_widths_.size() * entry_count) +
1018  } else {
1019  total_bytes = getRowSize() * entry_count;
1020  }
1021 
1022  return total_bytes;
1023 }
1024 
1026  const ExecutorDeviceType device_type) const {
1027  return getBufferSizeBytes(device_type, entry_count_);
1028 }
1029 
1031  output_columnar_ = val;
1034  }
1035 }
1036 
1037 /*
1038  * Indicates the query types that are currently allowed to use the logical
1039  * sized columns instead of padded sized ones.
1040  */
1042  // In distributed mode, result sets are serialized using rowwise iterators, so we use
1043  // consistent slot widths for now
1044  return output_columnar_ && !g_cluster &&
1046 }
1047 
1049  size_t total_slot_count = col_slot_context_.getSlotCount();
1050 
1051  if (target_groupby_indices_.empty()) {
1052  return total_slot_count;
1053  }
1054  return total_slot_count - std::count_if(target_groupby_indices_.begin(),
1056  [](const int64_t i) { return i >= 0; });
1057 }
1058 
1061  getGroupbyColCount() == 1);
1062 }
1063 
1066 }
1067 
1069  if (g_cluster || is_table_function_) {
1070  return true;
1071  }
1073  return true;
1074  }
1075  if (executor_->isCPUOnly() || render_output_ ||
1079  getGroupbyColCount() > 1)) {
1080  return true;
1081  }
1084 }
1085 
1087  return device_type == ExecutorDeviceType::GPU && !render_output_ &&
1089 }
1090 
1092  return interleaved_bins_on_gpu_ && device_type == ExecutorDeviceType::GPU;
1093 }
1094 
1095 // TODO(Saman): an implementation detail, so move this out of QMD
1097  const ExecutorDeviceType device_type) const {
1098  if (device_type != ExecutorDeviceType::GPU) {
1099  return false;
1100  } else {
1101  auto cuda_mgr = executor_->getCatalog()->getDataMgr().getCudaMgr();
1102  CHECK(cuda_mgr);
1103  return cuda_mgr->isArchVoltaOrGreaterForAll();
1104  }
1105 }
1106 
1108  return col_slot_context_.getColCount();
1109 }
1110 
1113 }
1114 
1115 const int8_t QueryMemoryDescriptor::getPaddedSlotWidthBytes(const size_t slot_idx) const {
1116  return col_slot_context_.getSlotInfo(slot_idx).padded_size;
1117 }
1118 
1120  const size_t slot_idx) const {
1121  return col_slot_context_.getSlotInfo(slot_idx).logical_size;
1122 }
1123 
1125  const size_t col_idx) const {
1126  const auto& col_slots = col_slot_context_.getSlotsForCol(col_idx);
1127  CHECK_EQ(col_slots.size(), size_t(1));
1128  return col_slots.front();
1129 }
1130 
1131 void QueryMemoryDescriptor::useConsistentSlotWidthSize(const int8_t slot_width_size) {
1132  col_slot_context_.setAllSlotsSize(slot_width_size);
1133 }
1134 
1136  // Note: Actual row size may include padding (see ResultSetBufferAccessors.h)
1138 }
1139 
1141  const int8_t actual_min_byte_width) const {
1142  return col_slot_context_.getMinPaddedByteSize(actual_min_byte_width);
1143 }
1144 
1146  const std::vector<std::tuple<int8_t, int8_t>>& slots_for_col) {
1147  col_slot_context_.addColumn(slots_for_col);
1148 }
1149 
1152 }
1153 
1156 }
1157 
1162 }
1163 
1165  switch (query_desc_type_) {
1167  return "Perfect Hash";
1169  return "Baseline Hash";
1171  return "Projection";
1173  return "Non-grouped Aggregate";
1175  return "Estimator";
1176  default:
1177  UNREACHABLE();
1178  }
1179  return "";
1180 }
1181 
1182 std::string QueryMemoryDescriptor::toString() const {
1183  auto str = reductionKey();
1184  str += "\tAllow Multifrag: " + ::toString(allow_multifrag_) + "\n";
1185  str += "\tInterleaved Bins on GPU: " + ::toString(interleaved_bins_on_gpu_) + "\n";
1186  str += "\tBlocks Share Memory: " + ::toString(blocksShareMemory()) + "\n";
1187  str += "\tThreads Share Memory: " + ::toString(threadsShareMemory()) + "\n";
1188  str += "\tUses Fast Group Values: " + ::toString(usesGetGroupValueFast()) + "\n";
1189  str +=
1190  "\tLazy Init Groups (GPU): " + ::toString(lazyInitGroups(ExecutorDeviceType::GPU)) +
1191  "\n";
1192  str += "\tEntry Count: " + std::to_string(entry_count_) + "\n";
1193  str += "\tMin Val (perfect hash only): " + std::to_string(min_val_) + "\n";
1194  str += "\tMax Val (perfect hash only): " + std::to_string(max_val_) + "\n";
1195  str += "\tBucket Val (perfect hash only): " + std::to_string(bucket_) + "\n";
1196  str += "\tSort on GPU: " + ::toString(sort_on_gpu_) + "\n";
1197  str += "\tUse Streaming Top N: " + ::toString(use_streaming_top_n_) + "\n";
1198  str += "\tOutput Columnar: " + ::toString(output_columnar_) + "\n";
1199  str += "\tRender Output: " + ::toString(render_output_) + "\n";
1200  str += "\tUse Baseline Sort: " + ::toString(must_use_baseline_sort_) + "\n";
1201  str += "\tIs Table Function: " + ::toString(is_table_function_) + "\n";
1202  return str;
1203 }
1204 
1206  std::string str;
1207  str += "Query Memory Descriptor State\n";
1208  str += "\tQuery Type: " + queryDescTypeToString() + "\n";
1209  str +=
1210  "\tKeyless Hash: " + ::toString(keyless_hash_) +
1211  (keyless_hash_ ? ", target index for key: " + std::to_string(getTargetIdxForKey())
1212  : "") +
1213  "\n";
1214  str += "\tEffective key width: " + std::to_string(getEffectiveKeyWidth()) + "\n";
1215  str += "\tNumber of group columns: " + std::to_string(getGroupbyColCount()) + "\n";
1216  const auto group_indices_size = targetGroupbyIndicesSize();
1217  if (group_indices_size) {
1218  std::vector<std::string> group_indices_strings;
1219  for (size_t target_idx = 0; target_idx < group_indices_size; ++target_idx) {
1220  group_indices_strings.push_back(std::to_string(getTargetGroupbyIndex(target_idx)));
1221  }
1222  str += "\tTarget group by indices: " +
1223  boost::algorithm::join(group_indices_strings, ",") + "\n";
1224  }
1225  str += "\t" + col_slot_context_.toString();
1226  return str;
1227 }
1228 
1229 std::vector<TargetInfo> target_exprs_to_infos(
1230  const std::vector<Analyzer::Expr*>& targets,
1232  std::vector<TargetInfo> target_infos;
1233  for (const auto target_expr : targets) {
1234  auto target = get_target_info(target_expr, g_bigint_count);
1235  if (query_mem_desc.getQueryDescriptionType() ==
1237  set_notnull(target, false);
1238  target.sql_type.set_notnull(false);
1239  }
1240  target_infos.push_back(target);
1241  }
1242  return target_infos;
1243 }
int8_t getMinPaddedByteSize(const int8_t actual_min_byte_width) const
std::vector< Analyzer::Expr * > target_exprs
static bool many_entries(const int64_t max_val, const int64_t min_val, const int64_t bucket)
int64_t getIntMin() const
SQLAgg
Definition: sqldefs.h:71
#define CHECK_EQ(x, y)
Definition: Logger.h:211
size_t getBufferSizeBytes(const RelAlgExecutionUnit &ra_exe_unit, const unsigned thread_count, const ExecutorDeviceType device_type) const
bool g_enable_smem_group_by
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)
void alignPaddedSlots(const bool sort_on_gpu)
std::string cat(Ts &&...args)
int8_t logical_size
class for a per-database catalog. also includes metadata for the current database and the current use...
Definition: Catalog.h:102
size_t getTotalBytesOfColumnarProjections(const size_t projection_count) const
int64_t getTargetGroupbyIndex(const size_t target_idx) const
ExecutorDeviceType
std::string toString() const
bool isLogicalSizedColumnsAllowed() const
#define const
TargetInfo get_target_info(const PointerType target_expr, const bool bigint_count)
Definition: TargetInfo.h:79
std::vector< int8_t > get_col_byte_widths(const T &col_expr_list)
int8_t pick_baseline_key_component_width(const ExpressionRange &range, const size_t group_col_width)
const std::list< Analyzer::OrderEntry > order_entries
std::string join(T const &container, std::string const &delim)
std::vector< InputDescriptor > input_descs
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)
#define UNREACHABLE()
Definition: Logger.h:247
void setOutputColumnar(const bool val)
const SortAlgorithm algorithm
#define CHECK_GE(x, y)
Definition: Logger.h:216
size_t getAllSlotsPaddedSize() const
size_t getAllSlotsAlignedPaddedSize() const
size_t getNextColOffInBytes(const int8_t *col_ptr, const size_t bin, const size_t col_idx) const
size_t getEffectiveKeyWidth() const
bool use_streaming_top_n(const RelAlgExecutionUnit &ra_exe_unit, const bool output_columnar)
const std::list< std::shared_ptr< Analyzer::Expr > > groupby_exprs
T visit(const Analyzer::Expr *expr) const
#define CHECK_GT(x, y)
Definition: Logger.h:215
void setAllSlotsSize(const int8_t slot_width_size)
std::string to_string(char const *&&v)
void useConsistentSlotWidthSize(const int8_t slot_width_size)
const SlotSize & getSlotInfo(const size_t slot_idx) const
size_t getColOnlyOffInBytes(const size_t col_idx) const
ExecutorDispatchMode
size_t getColOnlyOffInBytes(const size_t slot_idx) const
const SQLTypeInfo get_compact_type(const TargetInfo &target)
const size_t limit
bool g_enable_columnar_output
Definition: Execute.cpp:93
int8_t groupColWidth(const size_t key_idx) const
size_t get_bit_width(const SQLTypeInfo &ti)
#define INT32_MIN
const ColumnDescriptor * get_column_descriptor_maybe(const int col_id, const int table_id, const Catalog_Namespace::Catalog &cat)
Definition: Execute.h:221
std::vector< CountDistinctDescriptor > CountDistinctDescriptors
Definition: CountDistinct.h:35
const SortInfo sort_info
size_t getCompactByteWidth() const
Provides column info and slot info for the output buffer and some metadata helpers.
size_t getGroupbyColCount() const
bool is_integer() const
Definition: sqltypes.h:491
bool lazyInitGroups(const ExecutorDeviceType) const
size_t targetGroupbyIndicesSize() const
size_t getPrependedGroupBufferSizeInBytes() 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, const size_t thread_idx, RenderInfo *) const
size_t getTotalBytesOfColumnarBuffers() const
std::vector< int64_t > target_groupby_indices_
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)
bool g_bigint_count
CountDistinctDescriptors count_distinct_descriptors_
bool is_valid_int32_range(const ExpressionRange &range)
void validate() const
int get_varno() const
Definition: Analyzer.h:275
const int8_t getPaddedSlotWidthBytes(const size_t slot_idx) const
ExpressionRange getExpressionRange(const Analyzer::BinOper *expr, const std::vector< InputTableInfo > &query_infos, const Executor *, boost::optional< std::list< std::shared_ptr< Analyzer::Expr >>> simple_quals)
bool hasNulls() const
const SQLTypeInfo & get_type_info() const
Definition: Analyzer.h:78
QueryDescriptionType getQueryDescriptionType() const
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)
bool anyOf(std::vector< Analyzer::Expr * > const &target_exprs, SQLAgg const agg_kind)
void addColumn(const std::vector< std::tuple< int8_t, int8_t >> &slots_for_col)
#define CHECK_LE(x, y)
Definition: Logger.h:214
size_t getNextColOffInBytesRowOnly(const int8_t *col_ptr, const size_t col_idx) const
const Expr * get_operand() const
Definition: Analyzer.h:371
QueryDescriptionType query_desc_type_
int8_t padded_size
Definition: sqldefs.h:76
int8_t updateActualMinByteWidth(const int8_t actual_min_byte_width) const
size_t getTotalBytesOfColumnarBuffers(const size_t entry_count) const
bool operator==(const QueryMemoryDescriptor &other) const
Descriptor for the result set buffer layout.
bool is_int_and_no_bigger_than(const SQLTypeInfo &ti, const size_t byte_width)
std::list< std::shared_ptr< Analyzer::Expr > > quals
ExpressionRangeType getType() const
size_t get_heap_size(const size_t row_size, const size_t n, const size_t thread_count)
int64_t getIntMax() const
bool isWarpSyncRequired(const ExecutorDeviceType) const
std::string toString() const
size_t getSlotCount() const
void setAllSlotsPaddedSizeToLogicalSize()
bool interleavedBins(const ExecutorDeviceType) const
bool g_enable_watchdog false
Definition: Execute.cpp:76
#define CHECK(condition)
Definition: Logger.h:203
#define DEBUG_TIMER(name)
Definition: Logger.h:319
size_t getColCount() const
std::vector< int8_t > group_col_widths_
#define EMPTY_KEY_32
QueryDescriptionType
Definition: Types.h:26
bool g_cluster
std::vector< TargetInfo > target_exprs_to_infos(const std::vector< Analyzer::Expr * > &targets, const QueryMemoryDescriptor &query_mem_desc)
const std::vector< size_t > & getSlotsForCol(const size_t col_idx) const
std::string queryDescTypeToString() const
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)
std::list< std::shared_ptr< const InputColDescriptor > > input_col_descs
void addColSlotInfo(const std::vector< std::tuple< int8_t, int8_t >> &slots_for_col)
const size_t offset
static bool countDescriptorsLogicallyEmpty(const CountDistinctDescriptors &count_distinct_descriptors)
void setAllUnsetSlotsPaddedSize(const int8_t padded_size)
const int8_t getSlotIndexForSingleSlotCol(const size_t col_idx) const
const int8_t getLogicalSlotWidthBytes(const size_t slot_idx) const
Definition: sqldefs.h:72
size_t getColOffInBytes(const size_t col_idx) const
size_t getColOffInBytesInNextBin(const size_t col_idx) const
if(yyssp >=yyss+yystacksize-1)
FORCE_INLINE HOST DEVICE T align_to_int64(T addr)
int8_t pick_baseline_key_width(const RelAlgExecutionUnit &ra_exe_unit, const std::vector< InputTableInfo > &query_infos, const Executor *executor)
std::string reductionKey() const
std::list< std::shared_ptr< Analyzer::Expr > > simple_quals
void set_notnull(TargetInfo &target, const bool not_null)
int32_t getTargetIdxForKey() const
size_t getPrependedGroupColOffInBytes(const size_t group_idx) const