OmniSciDB  06b3bd477c
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Pages
ArrayOps.cpp
Go to the documentation of this file.
1 /*
2  * Copyright 2017 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 
25 #include <cstdint>
26 #include "../Shared/funcannotations.h"
27 #include "../Utils/ChunkIter.h"
28 #include "TypePunning.h"
29 
30 #ifdef EXECUTE_INCLUDE
31 
32 extern "C" DEVICE int32_t array_size(int8_t* chunk_iter_,
33  const uint64_t row_pos,
34  const uint32_t elem_log_sz) {
35  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
36  ArrayDatum ad;
37  bool is_end;
38  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end);
39  return ad.is_null ? 0 : ad.length >> elem_log_sz;
40 }
41 
42 extern "C" DEVICE int32_t array_size_nullable(int8_t* chunk_iter_,
43  const uint64_t row_pos,
44  const uint32_t elem_log_sz,
45  const int32_t null_val) {
46  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
47  ArrayDatum ad;
48  bool is_end;
49  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end);
50  return ad.is_null ? null_val : ad.length >> elem_log_sz;
51 }
52 
53 extern "C" DEVICE bool array_is_null(int8_t* chunk_iter_, const uint64_t row_pos) {
54  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
55  ArrayDatum ad;
56  bool is_end;
57  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end);
58  return ad.is_null;
59 }
60 
61 extern "C" DEVICE bool point_coord_array_is_null(int8_t* chunk_iter_,
62  const uint64_t row_pos) {
63  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
64  ArrayDatum ad;
65  bool is_end;
66  ChunkIter_get_nth_point_coords(chunk_iter, row_pos, &ad, &is_end);
67  return ad.is_null;
68 }
69 
70 #define ARRAY_AT(type) \
71  extern "C" DEVICE type array_at_##type( \
72  int8_t* chunk_iter_, const uint64_t row_pos, const uint32_t elem_idx) { \
73  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
74  ArrayDatum ad; \
75  bool is_end; \
76  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
77  return reinterpret_cast<type*>(ad.pointer)[elem_idx]; \
78  }
79 
80 ARRAY_AT(int8_t)
81 ARRAY_AT(int16_t)
82 ARRAY_AT(int32_t)
83 ARRAY_AT(int64_t)
84 ARRAY_AT(float)
85 ARRAY_AT(double)
86 
87 #undef ARRAY_AT
88 
89 #define VARLEN_ARRAY_AT(type) \
90  extern "C" DEVICE type varlen_array_at_##type( \
91  int8_t* chunk_iter_, const uint64_t row_pos, const uint32_t elem_idx) { \
92  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
93  ArrayDatum ad; \
94  bool is_end; \
95  ChunkIter_get_nth_varlen(chunk_iter, row_pos, &ad, &is_end); \
96  return reinterpret_cast<type*>(ad.pointer)[elem_idx]; \
97  }
98 
99 VARLEN_ARRAY_AT(int8_t)
100 VARLEN_ARRAY_AT(int16_t)
101 VARLEN_ARRAY_AT(int32_t)
102 VARLEN_ARRAY_AT(int64_t)
103 VARLEN_ARRAY_AT(float)
104 VARLEN_ARRAY_AT(double)
105 
106 #undef VARLEN_ARRAY_AT
107 
108 #define VARLEN_NOTNULL_ARRAY_AT(type) \
109  extern "C" DEVICE type varlen_notnull_array_at_##type( \
110  int8_t* chunk_iter_, const uint64_t row_pos, const uint32_t elem_idx) { \
111  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
112  ArrayDatum ad; \
113  bool is_end; \
114  ChunkIter_get_nth_varlen_notnull(chunk_iter, row_pos, &ad, &is_end); \
115  return reinterpret_cast<type*>(ad.pointer)[elem_idx]; \
116  }
117 
118 VARLEN_NOTNULL_ARRAY_AT(int8_t)
119 VARLEN_NOTNULL_ARRAY_AT(int16_t)
120 VARLEN_NOTNULL_ARRAY_AT(int32_t)
121 VARLEN_NOTNULL_ARRAY_AT(int64_t)
122 VARLEN_NOTNULL_ARRAY_AT(float)
123 VARLEN_NOTNULL_ARRAY_AT(double)
124 
125 #undef VARLEN_NOTNULL_ARRAY_AT
126 
127 #define ARRAY_ANY(type, needle_type, oper_name, oper) \
128  extern "C" DEVICE bool array_any_##oper_name##_##type##_##needle_type( \
129  int8_t* chunk_iter_, \
130  const uint64_t row_pos, \
131  const needle_type needle, \
132  const type null_val) { \
133  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
134  ArrayDatum ad; \
135  bool is_end; \
136  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
137  const size_t elem_count = ad.length / sizeof(type); \
138  for (size_t i = 0; i < elem_count; ++i) { \
139  const needle_type val = reinterpret_cast<type*>(ad.pointer)[i]; \
140  if (val != null_val && val oper needle) { \
141  return true; \
142  } \
143  } \
144  return false; \
145  }
146 
147 #define ARRAY_ALL(type, needle_type, oper_name, oper) \
148  extern "C" DEVICE bool array_all_##oper_name##_##type##_##needle_type( \
149  int8_t* chunk_iter_, \
150  const uint64_t row_pos, \
151  const needle_type needle, \
152  const type null_val) { \
153  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
154  ArrayDatum ad; \
155  bool is_end; \
156  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
157  const size_t elem_count = ad.length / sizeof(type); \
158  for (size_t i = 0; i < elem_count; ++i) { \
159  const needle_type val = reinterpret_cast<type*>(ad.pointer)[i]; \
160  if (!(val != null_val && val oper needle)) { \
161  return false; \
162  } \
163  } \
164  return true; \
165  }
166 
167 #define ARRAY_ALL_ANY_ALL_TYPES(oper_name, oper, needle_type) \
168  ARRAY_ANY(int8_t, needle_type, oper_name, oper) \
169  ARRAY_ALL(int8_t, needle_type, oper_name, oper) \
170  ARRAY_ANY(int16_t, needle_type, oper_name, oper) \
171  ARRAY_ALL(int16_t, needle_type, oper_name, oper) \
172  ARRAY_ANY(int32_t, needle_type, oper_name, oper) \
173  ARRAY_ALL(int32_t, needle_type, oper_name, oper) \
174  ARRAY_ANY(int64_t, needle_type, oper_name, oper) \
175  ARRAY_ALL(int64_t, needle_type, oper_name, oper) \
176  ARRAY_ANY(float, needle_type, oper_name, oper) \
177  ARRAY_ALL(float, needle_type, oper_name, oper) \
178  ARRAY_ANY(double, needle_type, oper_name, oper) \
179  ARRAY_ALL(double, needle_type, oper_name, oper)
180 
181 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int8_t)
182 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int8_t)
183 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int8_t)
184 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int8_t)
185 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int8_t)
186 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int8_t)
187 
188 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int16_t)
189 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int16_t)
190 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int16_t)
191 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int16_t)
192 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int16_t)
193 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int16_t)
194 
195 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int32_t)
196 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int32_t)
197 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int32_t)
198 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int32_t)
199 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int32_t)
200 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int32_t)
201 
202 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int64_t)
203 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int64_t)
204 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int64_t)
205 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int64_t)
206 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int64_t)
207 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int64_t)
208 
209 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, float)
210 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, float)
211 ARRAY_ALL_ANY_ALL_TYPES(lt, <, float)
212 ARRAY_ALL_ANY_ALL_TYPES(le, <=, float)
213 ARRAY_ALL_ANY_ALL_TYPES(gt, >, float)
214 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, float)
215 
216 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, double)
217 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, double)
218 ARRAY_ALL_ANY_ALL_TYPES(lt, <, double)
219 ARRAY_ALL_ANY_ALL_TYPES(le, <=, double)
220 ARRAY_ALL_ANY_ALL_TYPES(gt, >, double)
221 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, double)
222 
223 #undef ARRAY_ALL_ANY_ALL_TYPES
224 #undef ARRAY_ALL
225 #undef ARRAY_ANY
226 
227 #define ARRAY_AT_CHECKED(type) \
228  extern "C" DEVICE type array_at_##type##_checked(int8_t* chunk_iter_, \
229  const uint64_t row_pos, \
230  const int64_t elem_idx, \
231  const type null_val) { \
232  if (elem_idx <= 0) { \
233  return null_val; \
234  } \
235  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
236  ArrayDatum ad; \
237  bool is_end; \
238  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
239  if (ad.is_null || static_cast<size_t>(elem_idx) > ad.length / sizeof(type)) { \
240  return null_val; \
241  } \
242  return reinterpret_cast<type*>(ad.pointer)[elem_idx - 1]; \
243  }
244 
245 ARRAY_AT_CHECKED(int8_t)
246 ARRAY_AT_CHECKED(int16_t)
247 ARRAY_AT_CHECKED(int32_t)
248 ARRAY_AT_CHECKED(int64_t)
249 ARRAY_AT_CHECKED(float)
250 ARRAY_AT_CHECKED(double)
251 
252 #undef ARRAY_AT_CHECKED
253 
254 extern "C" DEVICE int8_t* allocate_varlen_buffer(int64_t element_count,
255  int64_t element_size) {
256 #ifndef __CUDACC__
257  int8_t* varlen_buffer =
258  reinterpret_cast<int8_t*>(checked_malloc((element_count + 1) * element_size));
259  return varlen_buffer;
260 #else
261  return nullptr;
262 #endif
263 }
264 
265 extern "C" DEVICE ALWAYS_INLINE int32_t
266 fast_fixlen_array_size(int8_t* chunk_iter_, const uint32_t elem_log_sz) {
267  ChunkIter* it = reinterpret_cast<ChunkIter*>(chunk_iter_);
268  return it->skip_size >> elem_log_sz;
269 }
270 
271 extern "C" DEVICE ALWAYS_INLINE int8_t* fast_fixlen_array_buff(int8_t* chunk_iter_,
272  const uint64_t row_pos) {
273  ChunkIter* it = reinterpret_cast<ChunkIter*>(chunk_iter_);
274  auto n = static_cast<int>(row_pos);
275  int8_t* current_pos = it->start_pos + n * it->skip_size;
276  return current_pos;
277 }
278 
279 extern "C" DEVICE int8_t* array_buff(int8_t* chunk_iter_, const uint64_t row_pos) {
280  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
281  ArrayDatum ad;
282  bool is_end;
283  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end);
284  return ad.pointer;
285 }
286 
287 #ifndef __CUDACC__
288 
289 #include <set>
290 
291 extern "C" ALWAYS_INLINE int64_t elem_bitcast_int8_t(const int8_t val) {
292  return val;
293 }
294 
295 extern "C" ALWAYS_INLINE int64_t elem_bitcast_int16_t(const int16_t val) {
296  return val;
297 }
298 
299 extern "C" ALWAYS_INLINE int64_t elem_bitcast_int32_t(const int32_t val) {
300  return val;
301 }
302 
303 extern "C" ALWAYS_INLINE int64_t elem_bitcast_int64_t(const int64_t val) {
304  return val;
305 }
306 
307 extern "C" ALWAYS_INLINE int64_t elem_bitcast_float(const float val) {
308  const double dval{val};
309  return *reinterpret_cast<const int64_t*>(may_alias_ptr(&dval));
310 }
311 
312 extern "C" ALWAYS_INLINE int64_t elem_bitcast_double(const double val) {
313  return *reinterpret_cast<const int64_t*>(may_alias_ptr(&val));
314 }
315 
316 #define COUNT_DISTINCT_ARRAY(type) \
317  extern "C" void agg_count_distinct_array_##type( \
318  int64_t* agg, int8_t* chunk_iter_, const uint64_t row_pos, const type null_val) { \
319  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
320  ArrayDatum ad; \
321  bool is_end; \
322  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
323  const size_t elem_count{ad.length / sizeof(type)}; \
324  for (size_t i = 0; i < elem_count; ++i) { \
325  const auto val = reinterpret_cast<type*>(ad.pointer)[i]; \
326  if (val != null_val) { \
327  reinterpret_cast<std::set<int64_t>*>(*agg)->insert(elem_bitcast_##type(val)); \
328  } \
329  } \
330  }
331 
332 COUNT_DISTINCT_ARRAY(int8_t)
333 COUNT_DISTINCT_ARRAY(int16_t)
334 COUNT_DISTINCT_ARRAY(int32_t)
335 COUNT_DISTINCT_ARRAY(int64_t)
336 COUNT_DISTINCT_ARRAY(float)
337 COUNT_DISTINCT_ARRAY(double)
338 
339 #undef COUNT_DISTINCT_ARRAY
340 
341 #include <string>
342 
343 extern "C" uint64_t string_decompress(const int32_t string_id,
344  const int64_t string_dict_handle);
345 
346 #define ARRAY_STR_ANY(type, oper_name, oper) \
347  extern "C" bool array_any_##oper_name##_str_##type(int8_t* chunk_iter_, \
348  const uint64_t row_pos, \
349  const char* needle_ptr, \
350  const uint32_t needle_len, \
351  const int64_t string_dict_handle, \
352  const type null_val) { \
353  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
354  ArrayDatum ad; \
355  bool is_end; \
356  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
357  const size_t elem_count = ad.length / sizeof(type); \
358  std::string needle_str(needle_ptr, needle_len); \
359  for (size_t i = 0; i < elem_count; ++i) { \
360  const type val = reinterpret_cast<type*>(ad.pointer)[i]; \
361  if (val != null_val) { \
362  uint64_t str_and_len = string_decompress(val, string_dict_handle); \
363  const char* str = reinterpret_cast<const char*>(str_and_len & 0xffffffffffff); \
364  const uint16_t len = str_and_len >> 48; \
365  std::string val_str(str, len); \
366  if (val_str oper needle_str) { \
367  return true; \
368  } \
369  } \
370  } \
371  return false; \
372  }
373 
374 #define ARRAY_STR_ALL(type, oper_name, oper) \
375  extern "C" bool array_all_##oper_name##_str_##type(int8_t* chunk_iter_, \
376  const uint64_t row_pos, \
377  const char* needle_ptr, \
378  const uint32_t needle_len, \
379  const int64_t string_dict_handle, \
380  const type null_val) { \
381  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
382  ArrayDatum ad; \
383  bool is_end; \
384  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
385  const size_t elem_count = ad.length / sizeof(type); \
386  std::string needle_str(needle_ptr, needle_len); \
387  for (size_t i = 0; i < elem_count; ++i) { \
388  const type val = reinterpret_cast<type*>(ad.pointer)[i]; \
389  if (val == null_val) { \
390  return false; \
391  } \
392  uint64_t str_and_len = string_decompress(val, string_dict_handle); \
393  const char* str = reinterpret_cast<const char*>(str_and_len & 0xffffffffffff); \
394  const uint16_t len = str_and_len >> 48; \
395  std::string val_str(str, len); \
396  if (!(val_str oper needle_str)) { \
397  return false; \
398  } \
399  } \
400  return true; \
401  }
402 
403 #define ARRAY_STR_ALL_ANY_ALL_TYPES(oper_name, oper) \
404  ARRAY_STR_ANY(int8_t, oper_name, oper) \
405  ARRAY_STR_ALL(int8_t, oper_name, oper) \
406  ARRAY_STR_ANY(int16_t, oper_name, oper) \
407  ARRAY_STR_ALL(int16_t, oper_name, oper) \
408  ARRAY_STR_ANY(int32_t, oper_name, oper) \
409  ARRAY_STR_ALL(int32_t, oper_name, oper) \
410  ARRAY_STR_ANY(int64_t, oper_name, oper) \
411  ARRAY_STR_ALL(int64_t, oper_name, oper)
412 
413 ARRAY_STR_ALL_ANY_ALL_TYPES(eq, ==)
414 ARRAY_STR_ALL_ANY_ALL_TYPES(ne, !=)
415 ARRAY_STR_ALL_ANY_ALL_TYPES(lt, <)
416 ARRAY_STR_ALL_ANY_ALL_TYPES(le, <=)
417 ARRAY_STR_ALL_ANY_ALL_TYPES(gt, >)
418 ARRAY_STR_ALL_ANY_ALL_TYPES(ge, >=)
419 
420 #undef ARRAY_ALL_ANY_ALL_TYPES
421 #undef ARRAY_STR_ALL
422 #undef ARRAY_STR_ANY
423 
424 #endif
425 
426 #endif // EXECUTE_INCLUDE
ChunkIter::start_pos
int8_t * start_pos
Definition: ChunkIter.h:33
allocate_varlen_buffer
EXTENSION_NOINLINE int8_t * allocate_varlen_buffer(int64_t element_count, int64_t element_size)
ChunkIter_get_nth_point_coords
DEVICE void ChunkIter_get_nth_point_coords(ChunkIter *it, int n, ArrayDatum *result, bool *is_end)
Definition: ChunkIter.cpp:308
TypePunning.h
ChunkIter_get_nth
DEVICE void ChunkIter_get_nth(ChunkIter *it, int n, bool uncompress, VarlenDatum *result, bool *is_end)
Definition: ChunkIter.cpp:181
DEVICE
#define DEVICE
Definition: funcannotations.h:20
ArrayDatum
std::conditional_t< is_cuda_compiler(), DeviceArrayDatum, HostArrayDatum > ArrayDatum
Definition: sqltypes.h:130
checked_malloc
void * checked_malloc(const size_t size)
Definition: checked_alloc.h:44
ChunkIter::skip_size
int skip_size
Definition: ChunkIter.h:36
string_decompress
uint64_t string_decompress(const int32_t string_id, const int64_t string_dict_handle)
Definition: StringOpsIR.cpp:36
ALWAYS_INLINE
#define ALWAYS_INLINE
Definition: funcannotations.h:54