1 | #include <roaring/roaring.h> |
2 | |
3 | struct roaring_pq_element_s { |
4 | uint64_t size; |
5 | bool is_temporary; |
6 | roaring_bitmap_t *bitmap; |
7 | }; |
8 | |
9 | typedef struct roaring_pq_element_s roaring_pq_element_t; |
10 | |
11 | struct roaring_pq_s { |
12 | roaring_pq_element_t *elements; |
13 | uint64_t size; |
14 | }; |
15 | |
16 | typedef struct roaring_pq_s roaring_pq_t; |
17 | |
18 | static inline bool compare(roaring_pq_element_t *t1, roaring_pq_element_t *t2) { |
19 | return t1->size < t2->size; |
20 | } |
21 | |
22 | static void pq_add(roaring_pq_t *pq, roaring_pq_element_t *t) { |
23 | uint64_t i = pq->size; |
24 | pq->elements[pq->size++] = *t; |
25 | while (i > 0) { |
26 | uint64_t p = (i - 1) >> 1; |
27 | roaring_pq_element_t ap = pq->elements[p]; |
28 | if (!compare(t, &ap)) break; |
29 | pq->elements[i] = ap; |
30 | i = p; |
31 | } |
32 | pq->elements[i] = *t; |
33 | } |
34 | |
35 | static void pq_free(roaring_pq_t *pq) { |
36 | free(pq->elements); |
37 | pq->elements = NULL; // paranoid |
38 | free(pq); |
39 | } |
40 | |
41 | static void percolate_down(roaring_pq_t *pq, uint32_t i) { |
42 | uint32_t size = (uint32_t)pq->size; |
43 | uint32_t hsize = size >> 1; |
44 | roaring_pq_element_t ai = pq->elements[i]; |
45 | while (i < hsize) { |
46 | uint32_t l = (i << 1) + 1; |
47 | uint32_t r = l + 1; |
48 | roaring_pq_element_t bestc = pq->elements[l]; |
49 | if (r < size) { |
50 | if (compare(pq->elements + r, &bestc)) { |
51 | l = r; |
52 | bestc = pq->elements[r]; |
53 | } |
54 | } |
55 | if (!compare(&bestc, &ai)) { |
56 | break; |
57 | } |
58 | pq->elements[i] = bestc; |
59 | i = l; |
60 | } |
61 | pq->elements[i] = ai; |
62 | } |
63 | |
64 | static roaring_pq_t *create_pq(const roaring_bitmap_t **arr, uint32_t length) { |
65 | roaring_pq_t *answer = (roaring_pq_t *)malloc(sizeof(roaring_pq_t)); |
66 | answer->elements = |
67 | (roaring_pq_element_t *)malloc(sizeof(roaring_pq_element_t) * length); |
68 | answer->size = length; |
69 | for (uint32_t i = 0; i < length; i++) { |
70 | answer->elements[i].bitmap = (roaring_bitmap_t *)arr[i]; |
71 | answer->elements[i].is_temporary = false; |
72 | answer->elements[i].size = |
73 | roaring_bitmap_portable_size_in_bytes(arr[i]); |
74 | } |
75 | for (int32_t i = (length >> 1); i >= 0; i--) { |
76 | percolate_down(answer, i); |
77 | } |
78 | return answer; |
79 | } |
80 | |
81 | static roaring_pq_element_t pq_poll(roaring_pq_t *pq) { |
82 | roaring_pq_element_t ans = *pq->elements; |
83 | if (pq->size > 1) { |
84 | pq->elements[0] = pq->elements[--pq->size]; |
85 | percolate_down(pq, 0); |
86 | } else |
87 | --pq->size; |
88 | // memmove(pq->elements,pq->elements+1,(pq->size-1)*sizeof(roaring_pq_element_t));--pq->size; |
89 | return ans; |
90 | } |
91 | |
92 | // this function consumes and frees the inputs |
93 | static roaring_bitmap_t *lazy_or_from_lazy_inputs(roaring_bitmap_t *x1, |
94 | roaring_bitmap_t *x2) { |
95 | uint8_t container_result_type = 0; |
96 | const int length1 = ra_get_size(&x1->high_low_container), |
97 | length2 = ra_get_size(&x2->high_low_container); |
98 | if (0 == length1) { |
99 | roaring_bitmap_free(x1); |
100 | return x2; |
101 | } |
102 | if (0 == length2) { |
103 | roaring_bitmap_free(x2); |
104 | return x1; |
105 | } |
106 | uint32_t neededcap = length1 > length2 ? length2 : length1; |
107 | roaring_bitmap_t *answer = roaring_bitmap_create_with_capacity(neededcap); |
108 | int pos1 = 0, pos2 = 0; |
109 | uint8_t container_type_1, container_type_2; |
110 | uint16_t s1 = ra_get_key_at_index(&x1->high_low_container, pos1); |
111 | uint16_t s2 = ra_get_key_at_index(&x2->high_low_container, pos2); |
112 | while (true) { |
113 | if (s1 == s2) { |
114 | // todo: unsharing can be inefficient as it may create a clone where |
115 | // none |
116 | // is needed, but it has the benefit of being easy to reason about. |
117 | ra_unshare_container_at_index(&x1->high_low_container, pos1); |
118 | void *c1 = ra_get_container_at_index(&x1->high_low_container, pos1, |
119 | &container_type_1); |
120 | assert(container_type_1 != SHARED_CONTAINER_TYPE_CODE); |
121 | ra_unshare_container_at_index(&x2->high_low_container, pos2); |
122 | void *c2 = ra_get_container_at_index(&x2->high_low_container, pos2, |
123 | &container_type_2); |
124 | assert(container_type_2 != SHARED_CONTAINER_TYPE_CODE); |
125 | void *c; |
126 | |
127 | if ((container_type_2 == BITSET_CONTAINER_TYPE_CODE) && |
128 | (container_type_1 != BITSET_CONTAINER_TYPE_CODE)) { |
129 | c = container_lazy_ior(c2, container_type_2, c1, |
130 | container_type_1, |
131 | &container_result_type); |
132 | container_free(c1, container_type_1); |
133 | if (c != c2) { |
134 | container_free(c2, container_type_2); |
135 | } |
136 | } else { |
137 | c = container_lazy_ior(c1, container_type_1, c2, |
138 | container_type_2, |
139 | &container_result_type); |
140 | container_free(c2, container_type_2); |
141 | if (c != c1) { |
142 | container_free(c1, container_type_1); |
143 | } |
144 | } |
145 | // since we assume that the initial containers are non-empty, the |
146 | // result here |
147 | // can only be non-empty |
148 | ra_append(&answer->high_low_container, s1, c, |
149 | container_result_type); |
150 | ++pos1; |
151 | ++pos2; |
152 | if (pos1 == length1) break; |
153 | if (pos2 == length2) break; |
154 | s1 = ra_get_key_at_index(&x1->high_low_container, pos1); |
155 | s2 = ra_get_key_at_index(&x2->high_low_container, pos2); |
156 | |
157 | } else if (s1 < s2) { // s1 < s2 |
158 | void *c1 = ra_get_container_at_index(&x1->high_low_container, pos1, |
159 | &container_type_1); |
160 | ra_append(&answer->high_low_container, s1, c1, container_type_1); |
161 | pos1++; |
162 | if (pos1 == length1) break; |
163 | s1 = ra_get_key_at_index(&x1->high_low_container, pos1); |
164 | |
165 | } else { // s1 > s2 |
166 | void *c2 = ra_get_container_at_index(&x2->high_low_container, pos2, |
167 | &container_type_2); |
168 | ra_append(&answer->high_low_container, s2, c2, container_type_2); |
169 | pos2++; |
170 | if (pos2 == length2) break; |
171 | s2 = ra_get_key_at_index(&x2->high_low_container, pos2); |
172 | } |
173 | } |
174 | if (pos1 == length1) { |
175 | ra_append_move_range(&answer->high_low_container, |
176 | &x2->high_low_container, pos2, length2); |
177 | } else if (pos2 == length2) { |
178 | ra_append_move_range(&answer->high_low_container, |
179 | &x1->high_low_container, pos1, length1); |
180 | } |
181 | ra_clear_without_containers(&x1->high_low_container); |
182 | ra_clear_without_containers(&x2->high_low_container); |
183 | free(x1); |
184 | free(x2); |
185 | return answer; |
186 | } |
187 | |
188 | /** |
189 | * Compute the union of 'number' bitmaps using a heap. This can |
190 | * sometimes be faster than roaring_bitmap_or_many which uses |
191 | * a naive algorithm. Caller is responsible for freeing the |
192 | * result. |
193 | */ |
194 | roaring_bitmap_t *roaring_bitmap_or_many_heap(uint32_t number, |
195 | const roaring_bitmap_t **x) { |
196 | if (number == 0) { |
197 | return roaring_bitmap_create(); |
198 | } |
199 | if (number == 1) { |
200 | return roaring_bitmap_copy(x[0]); |
201 | } |
202 | roaring_pq_t *pq = create_pq(x, number); |
203 | while (pq->size > 1) { |
204 | roaring_pq_element_t x1 = pq_poll(pq); |
205 | roaring_pq_element_t x2 = pq_poll(pq); |
206 | |
207 | if (x1.is_temporary && x2.is_temporary) { |
208 | roaring_bitmap_t *newb = |
209 | lazy_or_from_lazy_inputs(x1.bitmap, x2.bitmap); |
210 | // should normally return a fresh new bitmap *except* that |
211 | // it can return x1.bitmap or x2.bitmap in degenerate cases |
212 | bool temporary = !((newb == x1.bitmap) && (newb == x2.bitmap)); |
213 | uint64_t bsize = roaring_bitmap_portable_size_in_bytes(newb); |
214 | roaring_pq_element_t newelement = { |
215 | .size = bsize, .is_temporary = temporary, .bitmap = newb}; |
216 | pq_add(pq, &newelement); |
217 | } else if (x2.is_temporary) { |
218 | roaring_bitmap_lazy_or_inplace(x2.bitmap, x1.bitmap, false); |
219 | x2.size = roaring_bitmap_portable_size_in_bytes(x2.bitmap); |
220 | pq_add(pq, &x2); |
221 | } else if (x1.is_temporary) { |
222 | roaring_bitmap_lazy_or_inplace(x1.bitmap, x2.bitmap, false); |
223 | x1.size = roaring_bitmap_portable_size_in_bytes(x1.bitmap); |
224 | |
225 | pq_add(pq, &x1); |
226 | } else { |
227 | roaring_bitmap_t *newb = |
228 | roaring_bitmap_lazy_or(x1.bitmap, x2.bitmap, false); |
229 | uint64_t bsize = roaring_bitmap_portable_size_in_bytes(newb); |
230 | roaring_pq_element_t newelement = { |
231 | .size = bsize, .is_temporary = true, .bitmap = newb}; |
232 | |
233 | pq_add(pq, &newelement); |
234 | } |
235 | } |
236 | roaring_pq_element_t X = pq_poll(pq); |
237 | roaring_bitmap_t *answer = X.bitmap; |
238 | roaring_bitmap_repair_after_lazy(answer); |
239 | pq_free(pq); |
240 | return answer; |
241 | } |
242 | |