OmniSciDB  471d68cefb
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros Groups 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 RUNTIME_EXPORT 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 RUNTIME_EXPORT 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 RUNTIME_EXPORT bool array_is_null(int8_t* chunk_iter_,
54  const uint64_t row_pos) {
55  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
56  ArrayDatum ad;
57  bool is_end;
58  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end);
59  return ad.is_null;
60 }
61 
62 extern "C" DEVICE RUNTIME_EXPORT bool point_coord_array_is_null(int8_t* chunk_iter_,
63  const uint64_t row_pos) {
64  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
65  ArrayDatum ad;
66  bool is_end;
67  ChunkIter_get_nth_point_coords(chunk_iter, row_pos, &ad, &is_end);
68  return ad.is_null;
69 }
70 
71 #define ARRAY_AT(type) \
72  extern "C" DEVICE RUNTIME_EXPORT type array_at_##type( \
73  int8_t* chunk_iter_, const uint64_t row_pos, const uint32_t elem_idx) { \
74  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
75  ArrayDatum ad; \
76  bool is_end; \
77  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
78  return reinterpret_cast<type*>(ad.pointer)[elem_idx]; \
79  }
80 
81 ARRAY_AT(int8_t)
82 ARRAY_AT(int16_t)
83 ARRAY_AT(int32_t)
84 ARRAY_AT(int64_t)
85 ARRAY_AT(float)
86 ARRAY_AT(double)
87 
88 #undef ARRAY_AT
89 
90 #define VARLEN_ARRAY_AT(type) \
91  extern "C" DEVICE RUNTIME_EXPORT type varlen_array_at_##type( \
92  int8_t* chunk_iter_, const uint64_t row_pos, const uint32_t elem_idx) { \
93  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
94  ArrayDatum ad; \
95  bool is_end; \
96  ChunkIter_get_nth_varlen(chunk_iter, row_pos, &ad, &is_end); \
97  return reinterpret_cast<type*>(ad.pointer)[elem_idx]; \
98  }
99 
100 VARLEN_ARRAY_AT(int8_t)
101 VARLEN_ARRAY_AT(int16_t)
102 VARLEN_ARRAY_AT(int32_t)
103 VARLEN_ARRAY_AT(int64_t)
104 VARLEN_ARRAY_AT(float)
105 VARLEN_ARRAY_AT(double)
106 
107 #undef VARLEN_ARRAY_AT
108 
109 #define VARLEN_NOTNULL_ARRAY_AT(type) \
110  extern "C" DEVICE RUNTIME_EXPORT type varlen_notnull_array_at_##type( \
111  int8_t* chunk_iter_, const uint64_t row_pos, const uint32_t elem_idx) { \
112  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
113  ArrayDatum ad; \
114  bool is_end; \
115  ChunkIter_get_nth_varlen_notnull(chunk_iter, row_pos, &ad, &is_end); \
116  return reinterpret_cast<type*>(ad.pointer)[elem_idx]; \
117  }
118 
119 VARLEN_NOTNULL_ARRAY_AT(int8_t)
120 VARLEN_NOTNULL_ARRAY_AT(int16_t)
121 VARLEN_NOTNULL_ARRAY_AT(int32_t)
122 VARLEN_NOTNULL_ARRAY_AT(int64_t)
123 VARLEN_NOTNULL_ARRAY_AT(float)
124 VARLEN_NOTNULL_ARRAY_AT(double)
125 
126 #undef VARLEN_NOTNULL_ARRAY_AT
127 
128 #define ARRAY_ANY(type, needle_type, oper_name, oper) \
129  extern "C" DEVICE RUNTIME_EXPORT bool array_any_##oper_name##_##type##_##needle_type( \
130  int8_t* chunk_iter_, \
131  const uint64_t row_pos, \
132  const needle_type needle, \
133  const type null_val) { \
134  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
135  ArrayDatum ad; \
136  bool is_end; \
137  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
138  const size_t elem_count = ad.length / sizeof(type); \
139  for (size_t i = 0; i < elem_count; ++i) { \
140  const needle_type val = reinterpret_cast<type*>(ad.pointer)[i]; \
141  if (val != null_val && val oper needle) { \
142  return true; \
143  } \
144  } \
145  return false; \
146  }
147 
148 #define ARRAY_ALL(type, needle_type, oper_name, oper) \
149  extern "C" DEVICE RUNTIME_EXPORT bool array_all_##oper_name##_##type##_##needle_type( \
150  int8_t* chunk_iter_, \
151  const uint64_t row_pos, \
152  const needle_type needle, \
153  const type null_val) { \
154  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
155  ArrayDatum ad; \
156  bool is_end; \
157  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
158  const size_t elem_count = ad.length / sizeof(type); \
159  for (size_t i = 0; i < elem_count; ++i) { \
160  const needle_type val = reinterpret_cast<type*>(ad.pointer)[i]; \
161  if (!(val != null_val && val oper needle)) { \
162  return false; \
163  } \
164  } \
165  return true; \
166  }
167 
168 #define ARRAY_ALL_ANY_ALL_TYPES(oper_name, oper, needle_type) \
169  ARRAY_ANY(int8_t, needle_type, oper_name, oper) \
170  ARRAY_ALL(int8_t, needle_type, oper_name, oper) \
171  ARRAY_ANY(int16_t, needle_type, oper_name, oper) \
172  ARRAY_ALL(int16_t, needle_type, oper_name, oper) \
173  ARRAY_ANY(int32_t, needle_type, oper_name, oper) \
174  ARRAY_ALL(int32_t, needle_type, oper_name, oper) \
175  ARRAY_ANY(int64_t, needle_type, oper_name, oper) \
176  ARRAY_ALL(int64_t, needle_type, oper_name, oper) \
177  ARRAY_ANY(float, needle_type, oper_name, oper) \
178  ARRAY_ALL(float, needle_type, oper_name, oper) \
179  ARRAY_ANY(double, needle_type, oper_name, oper) \
180  ARRAY_ALL(double, needle_type, oper_name, oper)
181 
182 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int8_t)
183 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int8_t)
184 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int8_t)
185 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int8_t)
186 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int8_t)
187 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int8_t)
188 
189 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int16_t)
190 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int16_t)
191 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int16_t)
192 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int16_t)
193 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int16_t)
194 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int16_t)
195 
196 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int32_t)
197 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int32_t)
198 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int32_t)
199 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int32_t)
200 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int32_t)
201 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int32_t)
202 
203 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, int64_t)
204 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, int64_t)
205 ARRAY_ALL_ANY_ALL_TYPES(lt, <, int64_t)
206 ARRAY_ALL_ANY_ALL_TYPES(le, <=, int64_t)
207 ARRAY_ALL_ANY_ALL_TYPES(gt, >, int64_t)
208 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, int64_t)
209 
210 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, float)
211 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, float)
212 ARRAY_ALL_ANY_ALL_TYPES(lt, <, float)
213 ARRAY_ALL_ANY_ALL_TYPES(le, <=, float)
214 ARRAY_ALL_ANY_ALL_TYPES(gt, >, float)
215 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, float)
216 
217 ARRAY_ALL_ANY_ALL_TYPES(eq, ==, double)
218 ARRAY_ALL_ANY_ALL_TYPES(ne, !=, double)
219 ARRAY_ALL_ANY_ALL_TYPES(lt, <, double)
220 ARRAY_ALL_ANY_ALL_TYPES(le, <=, double)
221 ARRAY_ALL_ANY_ALL_TYPES(gt, >, double)
222 ARRAY_ALL_ANY_ALL_TYPES(ge, >=, double)
223 
224 #undef ARRAY_ALL_ANY_ALL_TYPES
225 #undef ARRAY_ALL
226 #undef ARRAY_ANY
227 
228 #define ARRAY_AT_CHECKED(type) \
229  extern "C" DEVICE RUNTIME_EXPORT type array_at_##type##_checked( \
230  int8_t* chunk_iter_, \
231  const uint64_t row_pos, \
232  const int64_t elem_idx, \
233  const type null_val) { \
234  if (elem_idx <= 0) { \
235  return null_val; \
236  } \
237  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
238  ArrayDatum ad; \
239  bool is_end; \
240  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
241  if (ad.is_null || static_cast<size_t>(elem_idx) > ad.length / sizeof(type)) { \
242  return null_val; \
243  } \
244  return reinterpret_cast<type*>(ad.pointer)[elem_idx - 1]; \
245  }
246 
247 ARRAY_AT_CHECKED(int8_t)
248 ARRAY_AT_CHECKED(int16_t)
249 ARRAY_AT_CHECKED(int32_t)
250 ARRAY_AT_CHECKED(int64_t)
251 ARRAY_AT_CHECKED(float)
252 ARRAY_AT_CHECKED(double)
253 
254 #undef ARRAY_AT_CHECKED
255 
256 extern "C" DEVICE RUNTIME_EXPORT int8_t* allocate_varlen_buffer(int64_t element_count,
257  int64_t element_size) {
258 #ifndef __CUDACC__
259  int8_t* varlen_buffer =
260  reinterpret_cast<int8_t*>(checked_malloc((element_count + 1) * element_size));
261  return varlen_buffer;
262 #else
263  return nullptr;
264 #endif
265 }
266 
267 extern "C" DEVICE RUNTIME_EXPORT ALWAYS_INLINE int32_t
268 fast_fixlen_array_size(int8_t* chunk_iter_, const uint32_t elem_log_sz) {
269  ChunkIter* it = reinterpret_cast<ChunkIter*>(chunk_iter_);
270  return it->skip_size >> elem_log_sz;
271 }
272 
273 extern "C" DEVICE RUNTIME_EXPORT ALWAYS_INLINE int8_t* fast_fixlen_array_buff(
274  int8_t* chunk_iter_,
275  const uint64_t row_pos) {
276  ChunkIter* it = reinterpret_cast<ChunkIter*>(chunk_iter_);
277  auto n = static_cast<int>(row_pos);
278  int8_t* current_pos = it->start_pos + n * it->skip_size;
279  return current_pos;
280 }
281 
282 extern "C" DEVICE RUNTIME_EXPORT int8_t* array_buff(int8_t* chunk_iter_,
283  const uint64_t row_pos) {
284  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_);
285  ArrayDatum ad;
286  bool is_end;
287  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end);
288  return ad.pointer;
289 }
290 
291 #ifndef __CUDACC__
292 
293 #include <set>
294 
295 extern "C" RUNTIME_EXPORT ALWAYS_INLINE int64_t elem_bitcast_int8_t(const int8_t val) {
296  return val;
297 }
298 
299 extern "C" RUNTIME_EXPORT ALWAYS_INLINE int64_t elem_bitcast_int16_t(const int16_t val) {
300  return val;
301 }
302 
303 extern "C" RUNTIME_EXPORT ALWAYS_INLINE int64_t elem_bitcast_int32_t(const int32_t val) {
304  return val;
305 }
306 
307 extern "C" RUNTIME_EXPORT ALWAYS_INLINE int64_t elem_bitcast_int64_t(const int64_t val) {
308  return val;
309 }
310 
311 extern "C" RUNTIME_EXPORT ALWAYS_INLINE int64_t elem_bitcast_float(const float val) {
312  const double dval{val};
313  return *reinterpret_cast<const int64_t*>(may_alias_ptr(&dval));
314 }
315 
316 extern "C" RUNTIME_EXPORT ALWAYS_INLINE int64_t elem_bitcast_double(const double val) {
317  return *reinterpret_cast<const int64_t*>(may_alias_ptr(&val));
318 }
319 
320 #define COUNT_DISTINCT_ARRAY(type) \
321  extern "C" RUNTIME_EXPORT void agg_count_distinct_array_##type( \
322  int64_t* agg, int8_t* chunk_iter_, const uint64_t row_pos, const type null_val) { \
323  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
324  ArrayDatum ad; \
325  bool is_end; \
326  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
327  const size_t elem_count{ad.length / sizeof(type)}; \
328  for (size_t i = 0; i < elem_count; ++i) { \
329  const auto val = reinterpret_cast<type*>(ad.pointer)[i]; \
330  if (val != null_val) { \
331  reinterpret_cast<std::set<int64_t>*>(*agg)->insert(elem_bitcast_##type(val)); \
332  } \
333  } \
334  }
335 
336 COUNT_DISTINCT_ARRAY(int8_t)
337 COUNT_DISTINCT_ARRAY(int16_t)
338 COUNT_DISTINCT_ARRAY(int32_t)
339 COUNT_DISTINCT_ARRAY(int64_t)
340 COUNT_DISTINCT_ARRAY(float)
341 COUNT_DISTINCT_ARRAY(double)
342 
343 #undef COUNT_DISTINCT_ARRAY
344 
345 #include <string>
346 
347 extern "C" RUNTIME_EXPORT uint64_t string_decompress(const int32_t string_id,
348  const int64_t string_dict_handle);
349 
350 #define ARRAY_STR_ANY(type, oper_name, oper) \
351  extern "C" RUNTIME_EXPORT bool array_any_##oper_name##_str_##type( \
352  int8_t* chunk_iter_, \
353  const uint64_t row_pos, \
354  const char* needle_ptr, \
355  const uint32_t needle_len, \
356  const int64_t string_dict_handle, \
357  const type null_val) { \
358  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
359  ArrayDatum ad; \
360  bool is_end; \
361  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
362  const size_t elem_count = ad.length / sizeof(type); \
363  std::string needle_str(needle_ptr, needle_len); \
364  for (size_t i = 0; i < elem_count; ++i) { \
365  const type val = reinterpret_cast<type*>(ad.pointer)[i]; \
366  if (val != null_val) { \
367  uint64_t str_and_len = string_decompress(val, string_dict_handle); \
368  const char* str = reinterpret_cast<const char*>(str_and_len & 0xffffffffffff); \
369  const uint16_t len = str_and_len >> 48; \
370  std::string val_str(str, len); \
371  if (val_str oper needle_str) { \
372  return true; \
373  } \
374  } \
375  } \
376  return false; \
377  }
378 
379 #define ARRAY_STR_ALL(type, oper_name, oper) \
380  extern "C" RUNTIME_EXPORT bool array_all_##oper_name##_str_##type( \
381  int8_t* chunk_iter_, \
382  const uint64_t row_pos, \
383  const char* needle_ptr, \
384  const uint32_t needle_len, \
385  const int64_t string_dict_handle, \
386  const type null_val) { \
387  ChunkIter* chunk_iter = reinterpret_cast<ChunkIter*>(chunk_iter_); \
388  ArrayDatum ad; \
389  bool is_end; \
390  ChunkIter_get_nth(chunk_iter, row_pos, &ad, &is_end); \
391  const size_t elem_count = ad.length / sizeof(type); \
392  std::string needle_str(needle_ptr, needle_len); \
393  for (size_t i = 0; i < elem_count; ++i) { \
394  const type val = reinterpret_cast<type*>(ad.pointer)[i]; \
395  if (val == null_val) { \
396  return false; \
397  } \
398  uint64_t str_and_len = string_decompress(val, string_dict_handle); \
399  const char* str = reinterpret_cast<const char*>(str_and_len & 0xffffffffffff); \
400  const uint16_t len = str_and_len >> 48; \
401  std::string val_str(str, len); \
402  if (!(val_str oper needle_str)) { \
403  return false; \
404  } \
405  } \
406  return true; \
407  }
408 
409 #define ARRAY_STR_ALL_ANY_ALL_TYPES(oper_name, oper) \
410  ARRAY_STR_ANY(int8_t, oper_name, oper) \
411  ARRAY_STR_ALL(int8_t, oper_name, oper) \
412  ARRAY_STR_ANY(int16_t, oper_name, oper) \
413  ARRAY_STR_ALL(int16_t, oper_name, oper) \
414  ARRAY_STR_ANY(int32_t, oper_name, oper) \
415  ARRAY_STR_ALL(int32_t, oper_name, oper) \
416  ARRAY_STR_ANY(int64_t, oper_name, oper) \
417  ARRAY_STR_ALL(int64_t, oper_name, oper)
418 
419 ARRAY_STR_ALL_ANY_ALL_TYPES(eq, ==)
420 ARRAY_STR_ALL_ANY_ALL_TYPES(ne, !=)
421 ARRAY_STR_ALL_ANY_ALL_TYPES(lt, <)
422 ARRAY_STR_ALL_ANY_ALL_TYPES(le, <=)
423 ARRAY_STR_ALL_ANY_ALL_TYPES(gt, >)
424 ARRAY_STR_ALL_ANY_ALL_TYPES(ge, >=)
425 
426 #undef ARRAY_ALL_ANY_ALL_TYPES
427 #undef ARRAY_STR_ALL
428 #undef ARRAY_STR_ANY
429 
430 #endif
431 
432 #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:208
checked_malloc
void * checked_malloc(const size_t size)
Definition: checked_alloc.h:45
RUNTIME_EXPORT
#define RUNTIME_EXPORT
Definition: funcannotations.h:77
ChunkIter::skip_size
int skip_size
Definition: ChunkIter.h:36
anonymous_namespace{Utm.h}::n
constexpr double n
Definition: Utm.h:46
ALWAYS_INLINE
#define ALWAYS_INLINE
Definition: funcannotations.h:57
string_decompress
RUNTIME_EXPORT uint64_t string_decompress(const int32_t string_id, const int64_t string_dict_handle)
Definition: StringOpsIR.cpp:37