├── Concurrent
    ├── .gitignore
    ├── CMakeLists.txt
    ├── include
    │   ├── common.h
    │   └── tlbtree.h
    ├── src
    │   ├── CMakeLists.txt
    │   ├── fixtree.h
    │   ├── flush.h
    │   ├── pmallocator.h
    │   ├── spinlock.h
    │   ├── tlbtree_impl.cc
    │   ├── tlbtree_impl.h
    │   ├── wotree256.cc
    │   └── wotree256.h
    └── test
    │   ├── CMakeLists.txt
    │   ├── datagen.cc
    │   ├── main.cc
    │   ├── preload.cc
    │   └── zipfian.h
├── LICENSE
├── README.md
└── Single
    ├── .gitignore
    ├── CMakeLists.txt
    ├── include
        ├── common.h
        └── tlbtree.h
    ├── src
        ├── CMakeLists.txt
        ├── fixtree.h
        ├── flush.h
        ├── pmallocator.h
        ├── spinlock.h
        ├── tlbtree_impl.cc
        ├── tlbtree_impl.h
        ├── wotree256.cc
        └── wotree256.h
    └── test
        ├── CMakeLists.txt
        ├── datagen.cc
        ├── gen.sh
        ├── main.cc
        ├── preload.cc
        ├── test.cc
        └── zipfian.h


/Concurrent/.gitignore:
--------------------------------------------------------------------------------
1 | build
2 | .vscode


--------------------------------------------------------------------------------
/Concurrent/CMakeLists.txt:
--------------------------------------------------------------------------------
 1 | project(TLBtree)
 2 | 
 3 | cmake_minimum_required(VERSION 3.16)
 4 | 
 5 | add_compile_options(-mclwb -fmax-errors=5 -fopenmp)
 6 | add_compile_options(-O3)
 7 | link_libraries(/usr/lib/x86_64-linux-gnu/libpmemobj.so)
 8 | add_link_options(-pthread -fopenmp)
 9 | 
10 | include_directories(include)
11 | 
12 | add_subdirectory(src)
13 | add_subdirectory(test)


--------------------------------------------------------------------------------
/Concurrent/include/common.h:
--------------------------------------------------------------------------------
 1 | /*
 2 |     Copyright (c) Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
 3 |     USE AT YOUR OWN RISK!
 4 | */
 5 | #ifndef __COMMON_H__
 6 | #define __COMMON_H__
 7 | 
 8 | #include <cstdlib>
 9 | #include <limits>
10 | #include <cstdint>
11 | #include <sys/stat.h>
12 | 
13 | #define LOADSCALE 8
14 | 
15 | #define KILO 1024
16 | #define MILLION (KILO * KILO)
17 | #define CACHE_LINE_SIZE 64
18 | 
19 | #define DOFLUSH
20 | 
21 | #ifndef KEYTYPE
22 |     using _key_t = int64_t;
23 | #else
24 |     using _key_t = KEYTYPE;
25 | #endif
26 | 
27 | const _key_t MAX_KEY = std::numeric_limits<_key_t>::max();
28 | const _key_t MIN_KEY = typeid(_key_t) == typeid(double) || typeid(_key_t) == typeid(float) 
29 |                             ? -1 * MAX_KEY : std::numeric_limits<_key_t>::min();
30 | 
31 | struct Record {
32 |     _key_t key;
33 |     char * val; 
34 |     Record(_key_t k=MAX_KEY, char * v=NULL) : key(k), val(v) {}
35 |     bool operator < (const Record & other) {
36 |         return key < other.key;
37 |     }
38 | };
39 | 
40 | enum OperationType {READ = 0, INSERT, UPDATE, DELETE};
41 | 
42 | struct QueryType {
43 |     OperationType op;
44 |     int64_t key;
45 | };
46 | 
47 | struct res_t { // a result type use to pass info when split and search
48 |     bool flag; 
49 |     Record rec;
50 |     res_t(bool f, Record e):flag(f), rec(e) {}
51 | };
52 | 
53 | #include <sys/time.h>
54 | inline double seconds()
55 | {
56 |   timeval now;
57 |   gettimeofday(&now, NULL);
58 |   return now.tv_sec + now.tv_usec/1000000.0;
59 | }
60 | 
61 | inline int getRandom() {
62 |     timeval now;
63 |     gettimeofday(&now, NULL);
64 |     return now.tv_usec;
65 | }
66 | 
67 | inline bool file_exist(const char *pool_path) {
68 |     struct stat buffer;
69 |     return (stat(pool_path, &buffer) == 0);
70 | }
71 | 
72 | #endif //__COMMON_H__


--------------------------------------------------------------------------------
/Concurrent/include/tlbtree.h:
--------------------------------------------------------------------------------
 1 | #ifndef __TLBTREE_H__
 2 | #define __TLBTREE_H__
 3 | 
 4 | #include "../src/tlbtree_impl.h"
 5 | 
 6 | using tlbtree::TLBtreeImpl;
 7 | 
 8 | // configure the PMEM file and file size
 9 | static constexpr uint64_t POOL_SIZE = 512UL * 1024 * 1024;
10 | 
11 | class TLBtree {
12 | public:
13 |     TLBtree(std::string tlbname, uint64_t poolsize = POOL_SIZE) {
14 |         bool recover = file_exist(tlbname.c_str());
15 |         tree_ = new TLBtreeImpl<2,2>(tlbname, recover, poolsize);
16 |     }
17 | 
18 |     ~TLBtree() {
19 |         delete tree_;
20 |     }
21 | 
22 |     inline void insert(_key_t key, uint64_t val) {
23 |         tree_->insert(key, val);
24 |     }
25 | 
26 |     inline bool update(_key_t key, uint64_t val) {
27 |         return tree_->update(key, val);
28 |     }
29 | 
30 |     inline uint64_t lookup(_key_t key) {
31 |         uint64_t val;
32 |         bool found = tree_->find(key, val);
33 | 
34 |         if(found)
35 |             return val;
36 |         else 
37 |             return 0;
38 |     }
39 | 
40 |     inline bool remove(_key_t key) {
41 |         return tree_->remove(key);
42 |     }
43 | 
44 | private:
45 |     TLBtreeImpl <2,2> * tree_;
46 | };
47 | 
48 | #endif //__TLBTREE_H__


--------------------------------------------------------------------------------
/Concurrent/src/CMakeLists.txt:
--------------------------------------------------------------------------------
1 | add_library(tlbtree tlbtree_impl.cc wotree256.cc)


--------------------------------------------------------------------------------
/Concurrent/src/fixtree.h:
--------------------------------------------------------------------------------
  1 | /*  fixtree.h - A search-optimized fixed tree
  2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  3 |     USE AT YOUR OWN RISK!
  4 | */
  5 | 
  6 | #ifndef __FIXTREE__
  7 | #define __FIXTREE__
  8 | 
  9 | #include <cstdio>
 10 | #include <cstdint>
 11 | #include <cmath>
 12 | #include <vector>
 13 | #include <cassert>
 14 | #include <cstring>
 15 | 
 16 | #include "flush.h"
 17 | #include "pmallocator.h"
 18 | 
 19 | namespace fixtree {
 20 |     const int INNER_CARD = 32; // node size: 256B, the fanout of inner node is 32
 21 |     const int LEAF_CARD = 15;  // node size: 256B, the fanout of leaf node is 16
 22 |     const int LEAF_REBUILD_CARD = 8;
 23 |     const int MAX_HEIGHT = 10;
 24 | 
 25 |     // the entrance of fixtree that stores its persistent tree metadata
 26 |     struct entrance_t {
 27 |         void * leaf_buff;
 28 |         void * inner_buff;
 29 |         uint32_t height; // the tree height
 30 |         uint32_t leaf_cnt; 
 31 |     };
 32 | 
 33 | /*  Fixtree: 
 34 |         a search-optimized linearize tree structure which can absort moderate insertions: 
 35 | */
 36 | class Fixtree {
 37 |     public:
 38 |         struct INNode { // inner node is packed keys, which is very compact
 39 |             _key_t keys[INNER_CARD];
 40 |         } __attribute__((aligned(CACHE_LINE_SIZE)));
 41 | 
 42 |         struct LFNode { // leaf node is packed key-ptr along with a header. 
 43 |                         // leaf node has some gap to absort insert
 44 |             Spinlock mtx;
 45 |             uint64_t node_version;
 46 |             _key_t keys[LEAF_CARD];
 47 |             char * vals[LEAF_CARD];
 48 |         } __attribute__((aligned(CACHE_LINE_SIZE)));
 49 | 
 50 |     public:
 51 |         // volatile structures
 52 |         INNode * inner_nodes_;
 53 |         LFNode * leaf_nodes_;
 54 |         uint32_t height_;
 55 |         uint32_t leaf_cnt_;
 56 |         entrance_t * entrance_;
 57 |         uint32_t level_offset_[MAX_HEIGHT];
 58 |     
 59 |     public:
 60 |         Fixtree(entrance_t * ent) { // recovery the tree from the entrance
 61 |             inner_nodes_ = (INNode *)galc->absolute(ent->inner_buff);
 62 |             leaf_nodes_ = (LFNode *)galc->absolute(ent->leaf_buff);
 63 |             height_ = ent->height;
 64 |             leaf_cnt_ = ent->leaf_cnt;
 65 |             entrance_ = ent;
 66 | 
 67 |             uint32_t tmp = 0;
 68 |             for(int l = 0; l < height_; l++) {
 69 |                 level_offset_[l] = tmp;
 70 |                 tmp += std::pow(INNER_CARD, l);
 71 |             }
 72 |             level_offset_[height_] = tmp;
 73 |         }
 74 | 
 75 |         Fixtree(std::vector<Record> records) {
 76 |             const int lfary = LEAF_REBUILD_CARD;
 77 |             int record_count = records.size();
 78 |             
 79 |             uint32_t lfnode_cnt = std::ceil((float)record_count / lfary);
 80 |             leaf_nodes_ = (LFNode *) galc->malloc(std::max((size_t)4096, lfnode_cnt * sizeof(LFNode)));
 81 | 
 82 |             height_ = std::ceil(std::log(std::max((uint32_t)INNER_CARD, lfnode_cnt)) / std::log(INNER_CARD));
 83 |             uint32_t innode_cnt = (std::pow(INNER_CARD, height_) - 1) / (INNER_CARD - 1);
 84 |             inner_nodes_ = (INNode *) galc->malloc(std::max((size_t)4096, innode_cnt * sizeof(INNode)));
 85 | 
 86 |             // fill leaf nodes
 87 |             for(int i = 0; i < lfnode_cnt; i++) {
 88 |                 for(int j = 0; j < lfary; j++) {
 89 |                     auto idx = i * lfary + j;
 90 |                     leaf_nodes_[i].keys[j] = idx < record_count ? records[idx].key : MAX_KEY; 
 91 |                     leaf_nodes_[i].vals[j] = idx < record_count ? records[idx].val : 0;
 92 |                 }
 93 |                 for(int j = lfary; j < LEAF_CARD; j++) { // intialized key
 94 |                     leaf_nodes_[i].keys[j] = MAX_KEY;
 95 |                 }
 96 |                 clwb(leaf_nodes_ + i, sizeof(LFNode));
 97 |             }
 98 |             
 99 |             int cur_level_cnt = lfnode_cnt;
100 |             int cur_level_off = innode_cnt - std::pow(INNER_CARD, height_ - 1);
101 |             int last_level_off = 0;
102 |             
103 |             // fill parent innodes of leaf nodes
104 |             for(int i = 0; i < cur_level_cnt; i++)
105 |                 inner_insert(cur_level_off + i / INNER_CARD, i % INNER_CARD, leaf_nodes_[i].keys[0]);
106 |             if(cur_level_cnt % INNER_CARD)
107 |                 inner_insert(cur_level_off + cur_level_cnt / INNER_CARD, cur_level_cnt % INNER_CARD, MAX_KEY);
108 |             clwb(&inner_nodes_[cur_level_off], sizeof(INNode) * (cur_level_cnt / INNER_CARD + 1));
109 |             
110 |             cur_level_cnt = std::ceil((float)cur_level_cnt / INNER_CARD);
111 |             last_level_off = cur_level_off;
112 |             cur_level_off = cur_level_off - std::pow(INNER_CARD, height_ - 2);
113 | 
114 |             // fill other inner nodes
115 |             for(int l = height_ - 2; l >= 0; l--) { // level by level
116 |                 for(int i = 0; i < cur_level_cnt; i++)
117 |                     inner_insert(cur_level_off + i / INNER_CARD, i % INNER_CARD, inner_nodes_[last_level_off + i].keys[0]);
118 |                 if(cur_level_cnt % INNER_CARD)
119 |                     inner_insert(cur_level_off + cur_level_cnt / INNER_CARD, cur_level_cnt % INNER_CARD, MAX_KEY);
120 |                 clwb(&inner_nodes_[cur_level_off], sizeof(INNode) * (cur_level_cnt / INNER_CARD + 1));
121 | 
122 |                 cur_level_cnt = std::ceil((float)cur_level_cnt / INNER_CARD);
123 |                 last_level_off = cur_level_off;
124 |                 cur_level_off = cur_level_off - std::pow(INNER_CARD, l - 1);
125 |             }
126 |             
127 |             leaf_cnt_ = lfnode_cnt;
128 |             entrance_ = (entrance_t *)galc->malloc(4096); // the allocator is not thread_safe, allocate a large entrance
129 |             uint32_t tmp = 0;
130 |             for(int l = 0; l < height_; l++) {
131 |                 level_offset_[l] = tmp;
132 |                 tmp += std::pow(INNER_CARD, l);
133 |             }
134 |             level_offset_[height_] = tmp;
135 | 
136 |             persist_assign(&(entrance_->leaf_buff), (void *) galc->relative(leaf_nodes_));
137 |             persist_assign(&(entrance_->inner_buff), (void *) galc->relative(inner_nodes_));
138 |             persist_assign(&(entrance_->height), height_);
139 |             persist_assign(&(entrance_->leaf_cnt), lfnode_cnt);
140 | 
141 |             return ;
142 |         }
143 | 
144 |     public:
145 |         char ** find_lower(_key_t key) const { 
146 |             /* linear search, return the position of the stored value */
147 |             int cur_idx = level_offset_[0];
148 |             for(int l = 0; l < height_; l++) {
149 |                 #ifdef DEBUG
150 |                     INNode * inner = inner_nodes_ + cur_idx; 
151 |                 #endif
152 |                 cur_idx = level_offset_[l + 1] + (cur_idx - level_offset_[l]) * INNER_CARD + inner_search(cur_idx, key);
153 |             }
154 |             cur_idx -= level_offset_[height_];
155 | 
156 |             return leaf_search(cur_idx, key);
157 |         }
158 | 
159 |         bool insert(_key_t key, uint64_t val) {
160 |             uint32_t cur_idx = level_offset_[0];
161 |             for(int l = 0; l < height_; l++) {
162 |                 #ifdef DEBUG
163 |                     INNode * inner = inner_nodes_ + cur_idx; 
164 |                 #endif
165 |                 cur_idx = level_offset_[l + 1] + (cur_idx - level_offset_[l]) * INNER_CARD + inner_search(cur_idx, key);
166 |             }
167 |             cur_idx -= level_offset_[height_];
168 |             
169 |             LFNode * cur_leaf = leaf_nodes_ + cur_idx;
170 | 
171 |             for(int i = 0; i < LEAF_CARD; i++) {
172 |                 if (cur_leaf->keys[i] == MAX_KEY) { // empty slot
173 |                     cur_leaf->mtx.lock();
174 |                         leaf_insert(cur_idx, i, {key, (char *)val});
175 |                     cur_leaf->node_version++;
176 |                     cur_leaf->mtx.unlock();
177 |                     return true;
178 |                 }
179 |             }
180 |             return false;
181 |         }
182 | 
183 |         bool try_remove(_key_t key) {
184 |             int cur_idx = level_offset_[0];
185 |             for(int l = 0; l < height_; l++) {
186 |                 #ifdef DEBUG
187 |                     INNode * inner = inner_nodes_ + cur_idx; 
188 |                 #endif
189 |                 cur_idx = level_offset_[l + 1] + (cur_idx - level_offset_[l]) * INNER_CARD + inner_search(cur_idx, key);
190 |             }
191 |             cur_idx -= level_offset_[height_];
192 |             
193 |             LFNode * cur_leaf = leaf_nodes_ + cur_idx;
194 | 
195 |             cur_leaf->mtx.lock();
196 | 
197 |             _key_t max_leqkey = cur_leaf->keys[0];
198 |             int8_t max_leqi = 0;
199 |             int8_t rec_cnt = 1;
200 |             for(int i = 1; i < LEAF_CARD; i++) {
201 |                 if(cur_leaf->keys[i] != MAX_KEY) {
202 |                     rec_cnt += 1;
203 |                     if (cur_leaf->keys[i] <= key && cur_leaf->keys[i] > max_leqkey) {
204 |                         max_leqkey = cur_leaf->keys[i];
205 |                         max_leqi = i;
206 |                     }
207 |                 }
208 |             }
209 | 
210 |             /* There are three cases:
211 |                   1. | k1 | --- | kx |, delete k1, leaf is not empty if k1 is deleted (fail)
212 |                   2. | k1 | ---      |, delete k1, leaf is empty if k1 is deleted (success)
213 |                   3. | k1 | --- | kx |, delete kx, leaf is not empty if kx is deleted (success)
214 |             */
215 |             if(max_leqi == 0 && rec_cnt > 1) { // case 1
216 |                 cur_leaf->mtx.unlock();
217 |                 return false;
218 |             } else { // case 2, 3
219 |                 persist_assign(&(cur_leaf->keys[max_leqi]), MAX_KEY);
220 |                 
221 |                 cur_leaf->node_version++;
222 |                 cur_leaf->mtx.unlock();
223 |                 return true;
224 |             }
225 |         }
226 | 
227 |         void printAll() {
228 |             for(int l = 0; l < height_; l++) {
229 |                 printf("level: %d =>", l);
230 |                 
231 |                 for(int i = level_offset_[l]; i < level_offset_[l + 1]; i++) {
232 |                     inner_print(i);
233 |                 }
234 |                 printf("\n");
235 |             }
236 | 
237 |             printf("leafs");
238 | 
239 |             for(int i = 0; i < leaf_cnt_; i++) {
240 |                 leaf_print(i);
241 |             }
242 |         }
243 | 
244 |         char ** find_first() {
245 |             return (char **)&(leaf_nodes_[0].vals[0]);
246 |         }
247 | 
248 |         void merge(std::vector<Record> & in, std::vector<Record> & out) { // merge the records with in to out
249 |             uint32_t insize = in.size();
250 | 
251 |             uint32_t incur = 0, innode_pos = 0, cur_lfcnt = 0;
252 |             Record tmp[LEAF_CARD];
253 |             load_node(tmp, &leaf_nodes_[0]);
254 |             _key_t k1 = in[0].key, k2 = tmp[0].key;
255 |             while(incur < insize && cur_lfcnt < leaf_cnt_) {
256 |                 if(k1 == k2) { 
257 |                     out.push_back(in[incur]);
258 | 
259 |                     incur += 1;
260 |                     innode_pos += 1;
261 |                     if(innode_pos == LEAF_CARD || tmp[innode_pos].key == MAX_KEY) {
262 |                         cur_lfcnt += 1;
263 |                         load_node(tmp, &leaf_nodes_[cur_lfcnt]);
264 |                         innode_pos = 0;
265 |                     }
266 | 
267 |                     k1 = in[incur].key;
268 |                     k2 = tmp[innode_pos].key;
269 |                 } else if(k1 > k2) {
270 |                     out.push_back(tmp[innode_pos]);
271 | 
272 |                     innode_pos += 1;
273 |                     if(innode_pos == LEAF_CARD || tmp[innode_pos].key == MAX_KEY) {
274 |                         cur_lfcnt += 1;
275 |                         load_node(tmp, &leaf_nodes_[cur_lfcnt]);
276 |                         innode_pos = 0;
277 |                     }
278 | 
279 |                     k2 = tmp[innode_pos].key;;
280 |                 } else {
281 |                     out.push_back(in[incur]);
282 |                     
283 |                     incur += 1;
284 |                     k1 = in[incur].key;
285 |                 }
286 |             }
287 | 
288 |             if(incur < insize) {
289 |                 for(int i = incur; i < insize; i++) 
290 |                     out.push_back(in[i]);
291 |             }
292 | 
293 |             if(cur_lfcnt < leaf_cnt_) {
294 |                 while(cur_lfcnt < leaf_cnt_) {
295 |                     out.push_back(tmp[innode_pos]);
296 | 
297 |                     innode_pos += 1;
298 |                     if(innode_pos == LEAF_CARD || tmp[innode_pos].key == MAX_KEY) {
299 |                         cur_lfcnt += 1;
300 |                         load_node(tmp, &leaf_nodes_[cur_lfcnt]);
301 |                         innode_pos = 0;
302 |                     }
303 |                 }
304 |             }
305 |         }
306 | 
307 |     private:
308 |         int inner_search(int node_idx, _key_t key) const{
309 |             INNode * cur_inner = inner_nodes_ + node_idx;
310 |             for(int i = 0; i < INNER_CARD; i++) {
311 |                 if (cur_inner->keys[i] > key) {
312 |                     return i - 1;
313 |                 }
314 |             }
315 | 
316 |             return INNER_CARD - 1;
317 |         }
318 |         
319 |         char **leaf_search(int node_idx, _key_t key) const {
320 |             LFNode * cur_leaf = leaf_nodes_ + node_idx;
321 | 
322 |             retry:
323 |             auto old_version = cur_leaf->node_version;
324 | 
325 |             _key_t max_leqkey = cur_leaf->keys[0];
326 |             int8_t max_leqi = 0;
327 |             for(int i = 1; i < LEAF_CARD; i++) {
328 |                 if (cur_leaf->keys[i] <= key && cur_leaf->keys[i] > max_leqkey) {
329 |                     max_leqkey = cur_leaf->keys[i];
330 |                     max_leqi = i;
331 |                 }
332 |             }
333 | 
334 |             if (old_version != cur_leaf->node_version) goto retry;
335 |             
336 |             return (char **) &(cur_leaf->vals[max_leqi]);
337 |         }
338 | 
339 |         void leaf_insert(int node_idx, int off, Record rec) { // TODO: should do it in a CAS way
340 |             leaf_nodes_[node_idx].vals[off] = rec.val;
341 |             clwb(&leaf_nodes_[node_idx].vals[off], 8);
342 |             mfence();
343 | 
344 |             leaf_nodes_[node_idx].keys[off] = rec.key;
345 |             clwb(&leaf_nodes_[node_idx].keys[off], 8);
346 |             mfence();
347 |         }
348 | 
349 |         inline void inner_insert(int node_idx, int off, _key_t key) {
350 |             inner_nodes_[node_idx].keys[off] = key;
351 |         }
352 | 
353 |         void inner_print(int node_idx) {
354 |             printf("(");
355 |             for(int i = 0; i < INNER_CARD; i++) {
356 |                 printf("%lu ", inner_nodes_[node_idx].keys[i]);
357 |             }
358 |             printf(") ");
359 |         }
360 | 
361 |         void leaf_print(int node_idx) {
362 |             printf("(");
363 |             for(int i = 0; i < LEAF_CARD; i++) {
364 |                 printf("[%lu, %lu] ", leaf_nodes_[node_idx].keys[i], (long unsigned int)leaf_nodes_[node_idx].vals[i]);
365 |             }
366 |             printf(") \n");
367 |         }
368 | 
369 |         static void load_node(Record * to, LFNode * from) {
370 |             for(int i = 0; i < LEAF_CARD; i++) {
371 |                 to[i].key = from->keys[i];
372 |                 to[i].val = from->vals[i];
373 |             }
374 | 
375 |             std::sort(to, to + LEAF_CARD);
376 |         }
377 | };
378 | 
379 | inline entrance_t * get_entrance(Fixtree * tree) {
380 |     return tree->entrance_;
381 | }
382 | 
383 | inline void free(Fixtree * tree) {
384 |     entrance_t * upent = get_entrance(tree);
385 |     delete tree;
386 | 
387 |     galc->free(galc->absolute(upent->inner_buff));
388 |     galc->free(galc->absolute(upent->leaf_buff));
389 |     galc->free(upent);
390 | 
391 |     return ;
392 | }
393 | 
394 | typedef Fixtree uptree_t;
395 | 
396 | } // namespace fixtree
397 | 
398 | #endif //__FIXTREE__


--------------------------------------------------------------------------------
/Concurrent/src/flush.h:
--------------------------------------------------------------------------------
 1 | #ifndef __FLUSH_H__
 2 | #define __FLUSH_H__
 3 | 
 4 | #include <x86intrin.h>
 5 | 
 6 | #include "common.h"
 7 | 
 8 | static inline void mfence() {
 9 |     asm volatile("sfence" ::: "memory");
10 | }
11 | 
12 | static inline void flush(void * ptr) {
13 | #ifdef CLWB
14 |     _mm_clwb(ptr);
15 | #elif defined(CLFLUSHOPT)
16 |     _mm_clflushopt(ptr);
17 | #else
18 |     _mm_clflush(ptr);
19 | #endif
20 | }
21 | 
22 | inline void clwb(void *data, int len) {
23 | #ifdef DOFLUSH
24 |     volatile char *ptr = (char *)((unsigned long long)data &~(CACHE_LINE_SIZE-1));
25 |     for(; ptr < (char *)data + len; ptr += CACHE_LINE_SIZE) {
26 |         flush((void *)ptr);
27 |     }
28 | #endif //DOFLUSH
29 | }
30 | 
31 | inline void clflush(void *data, int len, bool fence=true)
32 | {
33 | #ifdef DOFLUSH
34 |     volatile char *ptr = (char *)((unsigned long long)data &~(CACHE_LINE_SIZE-1));
35 |     if(fence) mfence();
36 |     for(; ptr < (char *)data + len; ptr += CACHE_LINE_SIZE){
37 |         flush((void *)ptr);
38 |     }
39 |     if(fence) mfence();
40 | #endif //DOFLUSH
41 | }
42 | 
43 | template<typename T>
44 | inline void persist_assign(T* addr, const T &v) { // To ensure atomicity, the size of T should be less equal than 8
45 |     *addr = v;
46 |     clwb(addr, sizeof(T));
47 | }
48 | 
49 | #endif // __FLUSH_H__


--------------------------------------------------------------------------------
/Concurrent/src/pmallocator.h:
--------------------------------------------------------------------------------
  1 | /*  
  2 |     A wrapper for using PMDK allocator easily (refer to pmwcas)
  3 |     Copyright (c) Luo Yongping  All Rights Reserved!
  4 | */
  5 | 
  6 | #ifndef __BLKALLOCATOR_H__
  7 | #define __BLKALLOCATOR_H__
  8 | 
  9 | #include <cassert>
 10 | #include <cstdio>
 11 | #include <libpmemobj.h>
 12 | 
 13 | #include "common.h"
 14 | #include "flush.h"
 15 | #include "spinlock.h"
 16 | 
 17 | POBJ_LAYOUT_BEGIN(pmallocator);
 18 | POBJ_LAYOUT_TOID(pmallocator, char)
 19 | POBJ_LAYOUT_END(pmallocator)
 20 | 
 21 | /*
 22 |     Persistent Memory Allocator: a wrapper of PMDK allocation lib: https://pmem.io/pmdk/
 23 | 
 24 |     PMAllocator allocates persistent memory from a pool file that resides on NVM file system. 
 25 |     It uses malloc() and free() as the allocation and reclaiment interfaces. 
 26 |     Other public interfaces like get_root(), absolute() and relative() are essential to memory
 27 |     management in persistent environment. 
 28 | */
 29 | class PMAllocator {
 30 | private:
 31 |     static const int PEICE_CNT = 64;
 32 |     static const size_t ALIGN_SIZE = 256;
 33 |     
 34 |     struct MetaType {
 35 |         char * buffer[PEICE_CNT];
 36 |         size_t blk_per_piece;
 37 |         size_t cur_blk;
 38 |         // entrance of DS in buffer
 39 |         void * entrance;
 40 |     };
 41 |     MetaType * meta_;
 42 | 
 43 |     // volatile domain
 44 |     PMEMobjpool *pop_;
 45 | 
 46 |     char * buff_[PEICE_CNT];
 47 |     char * buff_aligned_[PEICE_CNT];
 48 |     size_t piece_size_;
 49 |     size_t max_blk_;
 50 |     Spinlock alloc_mtx;
 51 | 
 52 | public: 
 53 |     /*
 54 |      *  Construct a PM allocator, map a pool file into virtual memory
 55 |      *  @param filename     pool file name
 56 |      *  @param recover      if doing recover, false for the first time 
 57 |      *  @param layout_name  ID of a group of allocations (in characters), each ID corresponding to a root entry
 58 |      *  @param pool_size    pool size of the pool file, vaild if the file doesn't exist
 59 |      */
 60 |     PMAllocator(const char *file_name, bool recover, const char *layout_name, uint64_t pool_size) {
 61 |         PMEMobjpool *tmp_pool = nullptr;
 62 |         pool_size = pool_size + ((pool_size & ((1 << 23) - 1)) > 0 ? (1 << 23) : 0); // align to 8MB
 63 | 	    if(recover == false) {
 64 |             if(file_exist(file_name)) {
 65 |                 printf("[CAUTIOUS]: The pool file already exists\n");
 66 |                 printf("Try (1) remove the pool file %s\nOr  (2) set the recover parameter to be true\n", file_name);
 67 |                 exit(-1);
 68 |             }
 69 |             pop_ = pmemobj_create(file_name, layout_name, pool_size, S_IWUSR | S_IRUSR);
 70 |             meta_ = (MetaType *)pmemobj_direct(pmemobj_root(pop_, sizeof(MetaType)));
 71 |             
 72 |             // maintain volatile domain
 73 |             uint64_t alloc_size = (pool_size >> 1) + (pool_size >> 2) + (pool_size >> 3); // 7/8 of the pool is used as block alloction
 74 |             for(int i = 0; i < PEICE_CNT; i++) {
 75 |                 buff_[i] = (char *)mem_alloc(alloc_size / PEICE_CNT);
 76 |                 buff_aligned_[i] = (char *) ((uint64_t)buff_[i] + ((uint64_t) buff_[i] % ALIGN_SIZE == 0 ? 0 : (ALIGN_SIZE - (uint64_t) buff_[i] % ALIGN_SIZE)));
 77 |             }
 78 |             piece_size_ = (alloc_size / PEICE_CNT) / ALIGN_SIZE - 1;
 79 |             max_blk_ = piece_size_ * PEICE_CNT;
 80 |             
 81 |             // initialize meta_
 82 |             for(int i = 0; i < PEICE_CNT; i++) 
 83 |                 meta_->buffer[i] = relative(buff_[i]);
 84 |             meta_->blk_per_piece = piece_size_;
 85 |             meta_->cur_blk = 0;
 86 |             meta_->entrance = NULL;
 87 |             clwb(meta_, sizeof(MetaType));
 88 |         } else {
 89 |             if(!file_exist(file_name)) {
 90 |                 printf("Pool File Not Exist\n");
 91 | 		        exit(-1);
 92 | 	        }
 93 |             pop_ = pmemobj_open(file_name, layout_name);
 94 |             meta_ = (MetaType *)pmemobj_direct(pmemobj_root(pop_, sizeof(MetaType)));
 95 |             // maintain volatile domain
 96 |             for(int i = 0; i < PEICE_CNT; i++) {
 97 |                 buff_[i] = absolute(meta_->buffer[i]);
 98 |                 buff_aligned_[i] = (char *) ((uint64_t)buff_[i] + ((uint64_t) buff_[i] % ALIGN_SIZE == 0 ? 0 : (ALIGN_SIZE - (uint64_t) buff_[i] % ALIGN_SIZE)));
 99 |             }
100 |             piece_size_ = meta_->blk_per_piece;
101 |             max_blk_ = piece_size_ * PEICE_CNT;
102 |         }
103 |     }
104 | 
105 |     ~PMAllocator() {
106 |         pmemobj_close(pop_);
107 |     }
108 | 
109 | public:
110 |     /*  
111 |      *  Get/allocate the root entry of the allocator.
112 |      *  
113 |      *  The root entry is the entrance of one group of allocation, each group is
114 |      *  identified by the layout_name when constructing it.
115 |      * 
116 |      *  Each group of allocations is a independent, self-contained in-memory structure in the pool
117 |      *  such as b-tree or link-list
118 |      */
119 |     void * get_root(size_t nsize) { // the root of DS stored in buff_ is recorded at meta_->entrance
120 |         if(meta_->entrance == NULL) {
121 |             meta_->entrance = relative(malloc(nsize));
122 |             clwb(meta_, sizeof(MetaType));
123 |         }
124 |         return absolute(meta_->entrance);
125 |     }
126 | 
127 |     /*
128 |      *  Allocate a non-root piece of persistent memory from the mapped pool
129 |      *  return the virtual memory address
130 |      */
131 |     void * malloc(size_t nsize) { 
132 |         if(nsize >= (1 << 12)) { // large than 4KB, make sure it is atomic
133 |             void * mem = mem_alloc(nsize + ALIGN_SIZE); // not aligned
134 |             //  |  UNUSED    |HEADER|       memory you can use     |
135 |             // mem             (mem + off)
136 |             uint64_t offset = ALIGN_SIZE - (uint64_t)mem % ALIGN_SIZE;
137 |             // store a header in the front
138 |             uint64_t * header = (uint64_t *)((uint64_t)mem + offset - 8);
139 |             *header = offset;
140 | 
141 |             return (void *)((uint64_t)mem + offset);
142 |         }
143 |         
144 |         retry_malloc:
145 |         uint64_t old_cur_blk = meta_->cur_blk;
146 | 
147 |         int blk_demand = (nsize + ALIGN_SIZE - 1) / ALIGN_SIZE;
148 |         // case 1: not enough in the buffer
149 |         if(blk_demand + meta_->cur_blk > max_blk_) {
150 |             printf("run out of memory\n");
151 |             exit(-1);
152 |         }
153 |         // case 2: current piece can not accommdate this allocation
154 |         int piece_id = meta_->cur_blk / piece_size_;
155 |         if((meta_->cur_blk % piece_size_ + blk_demand) > piece_size_) {
156 |             void * mem = buff_aligned_[piece_id + 1]; // allocate from a new peice
157 | 
158 |             uint64_t new_cur_blk = piece_size_ * (piece_id + 1) + blk_demand;
159 |             if(__sync_bool_compare_and_swap(&(meta_->cur_blk), old_cur_blk, new_cur_blk) == false) 
160 |                 goto retry_malloc;
161 |             clwb(&(meta_->cur_blk), 8);
162 | 
163 |             return mem;
164 |         } 
165 |         // case 3: current piece has enough space
166 |         else {
167 |             void * mem = buff_aligned_[piece_id] + ALIGN_SIZE * (meta_->cur_blk % piece_size_);
168 | 
169 |             uint64_t new_cur_blk = old_cur_blk + blk_demand;
170 |             if(__sync_bool_compare_and_swap(&(meta_->cur_blk), old_cur_blk, new_cur_blk) == false) 
171 |                 goto retry_malloc;
172 |             clwb(&(meta_->cur_blk), 8);
173 | 
174 |             return mem;
175 |         }
176 |     }
177 | 
178 |     void free(void* addr) {
179 |         for(int i = 0; i < PEICE_CNT; i++) {
180 |             uint64_t offset = (uint64_t)addr - (uint64_t)buff_aligned_[i];
181 |             if(offset > 0 && offset < piece_size_ * ALIGN_SIZE) {
182 |                 // the addr is in this piece, do not reclaim it
183 |                 return ;
184 |             }
185 |         }
186 | 
187 |         // larger than 4KB, reclaim it 
188 |         uint64_t * header = (uint64_t *)((uint64_t)addr - 8);
189 |         uint64_t offset = *header; 
190 | 
191 |         alloc_mtx.lock();
192 |         auto oid_ptr = pmemobj_oid((void *)((uint64_t)addr - offset));
193 |         alloc_mtx.unlock();
194 | 
195 |         TOID(char) ptr_cpy;
196 |         TOID_ASSIGN(ptr_cpy, oid_ptr);
197 |         POBJ_FREE(&ptr_cpy);
198 |     }  
199 | 
200 |     /*
201 |      *  Distinguish from virtual memory address and offset in the pool
202 |      *  Each memory piece allocated from the pool has an in-pool offset, which remains unchanged
203 |      *  until reclaiment. We cannot ensure that the pool file is mapped at the same position at 
204 |      *  any time, so it may locate at different virtual memory addresses next time. 
205 |      *  
206 |      *  So the rule is that, using virtual memory when doing normal operations like to DRAM
207 |      *  space, using offset to store link relationship, for exmaple, next pointer in linklist
208 |      * /
209 | 
210 |     /*
211 |      *  convert the virtual memory address to an offset
212 |      */
213 |     template<typename T>
214 |     inline T *absolute(T *pmem_offset) {
215 |         if(pmem_offset == NULL)
216 |             return NULL;
217 |         return reinterpret_cast<T *>(reinterpret_cast<uint64_t>(pmem_offset) + reinterpret_cast<char *>(pop_));
218 |     }
219 |     
220 |     template<typename T>
221 |     inline T *relative(T *pmem_direct) {
222 |         if(pmem_direct == NULL)
223 |             return NULL;
224 |         return reinterpret_cast<T *>(reinterpret_cast<char *>(pmem_direct) - reinterpret_cast<char *>(pop_));
225 |     }
226 | 
227 | private:
228 |     void * mem_alloc(size_t nsize) {
229 |         PMEMoid tmp;
230 | 
231 |         alloc_mtx.lock();
232 |         pmemobj_alloc(pop_, &tmp, nsize, TOID_TYPE_NUM(char), NULL, NULL);
233 |         alloc_mtx.unlock();
234 |         
235 |         void * mem = pmemobj_direct(tmp);
236 |         assert(mem != nullptr);
237 |         return mem;
238 |     }
239 | };
240 | 
241 | extern PMAllocator * galc;
242 | 
243 | #endif // __BLKALLOCATOR_H__
244 | 


--------------------------------------------------------------------------------
/Concurrent/src/spinlock.h:
--------------------------------------------------------------------------------
 1 | /* Copyright(c): Guo Zhongming 
 2 | */
 3 | 
 4 | #ifndef __SPINLOCK_H__
 5 | #define __SPINLOCK_H__
 6 | 
 7 | #include <atomic>
 8 | #include <emmintrin.h>
 9 | #include <thread>
10 | 
11 | class Spinlock {
12 | public:
13 |     Spinlock() {
14 |         atomic_val.store(0, std::memory_order_relaxed);
15 |     }
16 | 
17 |     Spinlock(const Spinlock &) = delete;
18 |     Spinlock & operator = (const Spinlock &) = delete;
19 | 
20 | public:
21 |     inline void lock() {
22 |         while(atomic_val.exchange(1, std::memory_order_acquire) == 1){
23 |             while(1) {
24 |                 _mm_pause(); // delay for 140 cycle
25 | 
26 |                 if(atomic_val.load(std::memory_order_relaxed) == 0) // check the atomic_val
27 |                     break;
28 |                 
29 |                 std::this_thread::yield(); // delay for 113ns
30 | 
31 |                 if(atomic_val.load(std::memory_order_relaxed) == 0) // check the atomic_val
32 |                     break;
33 |             }
34 | 
35 |             // if at here, the atomic_val must be just be 0
36 |         }
37 | 
38 |         return ;
39 |     }
40 | 
41 |     inline void unlock() {
42 |         atomic_val.exchange(0, std::memory_order_acquire);
43 |     }
44 | 
45 |     inline bool trylock() {
46 |         return atomic_val.exchange(1, std::memory_order_acquire) == 0;
47 |     }
48 | 
49 | private:
50 |     std::atomic_short atomic_val;
51 | }; 
52 | 
53 | #endif // __SPINLOCK_H__


--------------------------------------------------------------------------------
/Concurrent/src/tlbtree_impl.cc:
--------------------------------------------------------------------------------
1 | #include "tlbtree_impl.h"
2 | 
3 | PMAllocator * galc;


--------------------------------------------------------------------------------
/Concurrent/src/tlbtree_impl.h:
--------------------------------------------------------------------------------
  1 | /*  tlbtree.h - A two level btree for persistent memory
  2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  3 |     USE AT YOUR OWN RISK!
  4 | */
  5 | 
  6 | #ifndef __TLBTREEIMPL_H__
  7 | #define __TLBTREEIMPL_H__
  8 | 
  9 | #include <string>
 10 | #include <thread>
 11 | #include <algorithm>
 12 | #include <unistd.h>
 13 | 
 14 | #include "pmallocator.h"
 15 | #include "fixtree.h"
 16 | #include "spinlock.h"
 17 | #include "wotree256.h"
 18 | 
 19 | extern PMAllocator * galc;
 20 | 
 21 | #define BACKGROUND_REBUILD
 22 | // choose uptree type, providing interfaces: insert, remove, update, find, merge, free_uptree
 23 | #define UPTREE_NS   fixtree
 24 | // choose downtree type, providing interfaces: insert, find_lower, remove_lower
 25 | #define DOWNTREE_NS wotree256
 26 | 
 27 | namespace tlbtree {
 28 | 
 29 | using std::string;
 30 | using std::vector;
 31 | using Node = DOWNTREE_NS::Node;
 32 | 
 33 | template<int DOWNLEVEL, int REBUILD_THRESHOLD=2>
 34 | class TLBtreeImpl {
 35 | private:
 36 |     typedef TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD> SelfType;
 37 |     
 38 |     // the entrance of TLBtree that stores its persistent tree metadata
 39 |     struct tlbtree_entrance_t {
 40 |         UPTREE_NS::entrance_t * upent; // the entrance of the top layer
 41 |         Record * restore;              // restore subroots that fails to insert into the top layer
 42 |         int restore_size;
 43 |         bool is_clean;                 // is TLBtree shutdown expectedly
 44 |         bool use_rebuild_recover;      // whether to use recover rebuilding next time
 45 |     };
 46 |     
 47 |     // volatile domain
 48 |     UPTREE_NS::uptree_t * uptree_;
 49 |     tlbtree_entrance_t * entrance_;
 50 |     vector<Record> * mutable_;
 51 |     Spinlock rebuild_mtx_;
 52 |     Spinlock mutable_mtx_;
 53 |     bool is_rebuilding_;
 54 | 
 55 | public:
 56 |     TLBtreeImpl(string path, bool recover=true, uint64_t pool_size=10 * (1024UL * 1024 * 1024));
 57 | 
 58 |     ~TLBtreeImpl();
 59 | 
 60 |     void insert(const _key_t & k, uint64_t v);
 61 | 
 62 |     bool find(const _key_t & k, uint64_t & v) const ;
 63 | 
 64 |     bool update(const _key_t & k, const uint64_t & v);
 65 | 
 66 |     bool remove(const _key_t & k);
 67 | 
 68 |     inline void printAll() { uptree_->printAll();}
 69 | 
 70 | private:
 71 |     void rebuild_fast();
 72 | 
 73 |     void rebuild_recover();
 74 | };
 75 | 
 76 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
 77 | TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::TLBtreeImpl(string path, bool recover, uint64_t pool_size) {
 78 |     mutable_ = new vector<Record>();
 79 |     mutable_->reserve(0xfff);
 80 |     bool is_rebuilding_ = false;
 81 |     
 82 |     if(recover == false) {
 83 |         galc = new PMAllocator(path.c_str(), false, "tlbtree", pool_size);
 84 |         // initialize entrance_
 85 |         entrance_ = (tlbtree_entrance_t *) galc->get_root(sizeof(tlbtree_entrance_t));
 86 |         entrance_->upent = NULL;
 87 |         entrance_->restore = NULL;
 88 |         entrance_->restore_size = 0;
 89 |         entrance_->is_clean = false;
 90 |         entrance_->use_rebuild_recover = true;
 91 |         clwb(entrance_, sizeof(tlbtree_entrance_t));
 92 |         
 93 |         //allocate a entrance_ to the fixtree
 94 |         std::vector<Record> init = {Record(MIN_KEY, (char *)galc->relative(new Node()))}; 
 95 |         uptree_ = new UPTREE_NS::uptree_t(init);
 96 |         persist_assign(&(entrance_->upent), galc->relative(UPTREE_NS::get_entrance(uptree_)));
 97 |         persist_assign(&(entrance_->use_rebuild_recover), false); // use fast rebuilding next time
 98 |     } else {
 99 |         galc = new PMAllocator(path.c_str(), true, "tlbtree", pool_size);
100 | 
101 |         entrance_ = (tlbtree_entrance_t *) galc->get_root(sizeof(tlbtree_entrance_t));
102 |         if(entrance_ == NULL || entrance_->upent == NULL) { // empty tree
103 |             printf("the tree is empty\n");
104 |             exit(-1);
105 |         }
106 | 
107 |         if(entrance_->is_clean == false) { // TLBtree crashed at last usage
108 |             persist_assign(&(entrance_->use_rebuild_recover), true); // use recover rebuilding next time
109 |         } else { // normal shutdown
110 |             // recover all subroots from PM back to mutable_, within miliseconds
111 |             if(entrance_->restore != NULL) {
112 |                 Record * rec = galc->absolute(entrance_->restore);
113 |                 for(int i = 0; i < entrance_->restore_size; i++) {
114 |                     mutable_->push_back(rec[i]);
115 |                 }
116 |                 entrance_->restore = NULL;
117 |                 entrance_->restore_size = 0;
118 |                 clwb(&entrance_->restore, 16);
119 |                 galc->free(rec);
120 |             }
121 |         }
122 | 
123 |         uptree_ = new UPTREE_NS::uptree_t (galc->absolute(entrance_->upent));
124 |     }
125 | 
126 |     persist_assign(&(entrance_->is_clean), false); // set the TLBtree state to be dirty
127 | }
128 | 
129 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
130 | TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::~TLBtreeImpl() {
131 |     if(entrance_->use_rebuild_recover == false) { // fast rebuilding next time
132 |         // save all subroots in mutable_ into PM
133 |         Record * rec = (Record *) galc->malloc(std::max((size_t)4096, mutable_->size() * sizeof(Record)));
134 |         for(int i = 0; i < mutable_->size(); i++) {
135 |             rec[i] = (*mutable_)[i];
136 |         }
137 |         clwb(rec, mutable_->size() * sizeof(Record));
138 |         mfence();
139 |         entrance_->restore = galc->relative(rec);
140 |         entrance_->restore_size = mutable_->size();
141 |         clwb(&entrance_->restore, 16);
142 |     }
143 | 
144 |     //printf("%lx %d\n", entrance_->restore, entrance_->restore_size);
145 | 
146 |     persist_assign(&(entrance_->is_clean), true); // a intended shutdown
147 | 
148 |     delete uptree_;
149 |     delete mutable_;
150 |     delete galc;
151 | }
152 | 
153 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
154 | void TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::insert(const _key_t & k, uint64_t v) { 
155 |     Node ** root_ptr = (Node **)uptree_->find_lower(k);
156 |     Node * downroot = (Node *)galc->absolute(*root_ptr);
157 | 
158 |     // travese in sibling chain
159 |     int8_t goes_steps = 0;
160 |     _key_t splitkey; Node ** sibling_ptr;
161 |     downroot->get_sibling(splitkey, sibling_ptr);
162 |     while(splitkey < k) { // the splitkey 
163 |         root_ptr = sibling_ptr; // where is current root store
164 |         downroot = (Node *)galc->absolute(*root_ptr);
165 |         downroot->get_sibling(splitkey, sibling_ptr);
166 |         goes_steps += 1;
167 |     }
168 |     res_t insert_res = DOWNTREE_NS::insert(root_ptr, k, v, DOWNLEVEL);
169 | 
170 |     // we rebuild if the searching in the linklist is too long 
171 |     if(goes_steps > REBUILD_THRESHOLD && rebuild_mtx_.trylock()) {
172 |         if(entrance_->use_rebuild_recover == true) {
173 |             #ifdef BACKGROUND_REBUILD
174 |                 std::thread rebuild_thread(&SelfType::rebuild_recover, this);
175 |                 rebuild_thread.detach();
176 |             #else
177 |                 rebuild_recover();
178 |             #endif
179 |         } else {
180 |             #ifdef BACKGROUND_REBUILD
181 |                 std::thread rebuild_thread(&SelfType::rebuild_fast, this);
182 |                 rebuild_thread.detach();
183 |             #else
184 |                 rebuild_fast();
185 |             #endif
186 |         } 
187 |     }
188 | 
189 |     if(insert_res.flag == true) { // a sub-index tree is splitted
190 |         // try save the sub-indices root into the top layer
191 |         bool succ = uptree_->insert(insert_res.rec.key, (uint64_t)galc->relative(insert_res.rec.val));
192 |         
193 |         // save these records into mutable_
194 |         if(is_rebuilding_ == true || succ == false) {
195 |             mutable_mtx_.lock();
196 |                 mutable_->push_back({insert_res.rec.key, (char *)galc->relative(insert_res.rec.val)});
197 |             mutable_mtx_.unlock();
198 |         }
199 |     }
200 | }
201 | 
202 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
203 | bool TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::find(const _key_t & k, uint64_t & v) const {
204 |     Node ** root_ptr = (Node **)uptree_->find_lower(k);
205 |     Node * downroot = (Node *)galc->absolute(*root_ptr);
206 | 
207 |     // traverse in sibling chain
208 |     _key_t splitkey; Node ** sibling_ptr;
209 |     downroot->get_sibling(splitkey, sibling_ptr);
210 |     while(splitkey <= k) { // the splitkey 
211 |         root_ptr = sibling_ptr; // where is current root store
212 |         downroot = (Node *)galc->absolute(*root_ptr);
213 |         downroot->get_sibling(splitkey, sibling_ptr);
214 |     }
215 | 
216 |     return DOWNTREE_NS::find(root_ptr, k, v);
217 | }
218 | 
219 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
220 | bool TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::remove(const _key_t & k) {
221 |     Node ** root_ptr = (Node **)uptree_->find_lower(k);
222 |     Node ** last_root_ptr = NULL; // record the last root ptr for laster use
223 |     Node *downroot = (Node *)galc->absolute(*root_ptr);
224 | 
225 |     // travese in sibling chain
226 |     _key_t splitkey; Node ** sibling_ptr;
227 |     downroot->get_sibling(splitkey, sibling_ptr);
228 |     while(splitkey < k) { // the splitkey 
229 |         root_ptr = sibling_ptr; // where is current root store
230 |         downroot = (Node *)galc->absolute(*root_ptr);
231 |         downroot->get_sibling(splitkey, sibling_ptr);
232 |     }
233 |     
234 |     bool emptyif = DOWNTREE_NS::remove(root_ptr, k);
235 |     if(emptyif) { // the DOWNTREE_NS is empty now
236 |         uptree_->try_remove(k); // TODO: rebuilding should also be triggered when the top layer is too empty
237 |     }
238 | 
239 |     return true;
240 | }
241 | 
242 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
243 | bool TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::update(const _key_t & k, const uint64_t & v) {
244 |     Node ** root_ptr = (Node **)uptree_->find_lower(k);
245 |     Node * downroot = (Node *)galc->absolute(*root_ptr);
246 | 
247 |     // travese in sibling chain
248 |     _key_t splitkey; Node ** sibling_ptr;
249 |     downroot->get_sibling(splitkey, sibling_ptr);
250 |     while(splitkey < k) { // the splitkey 
251 |         root_ptr = sibling_ptr; // where is current root store
252 |         downroot = (Node *)galc->absolute(*root_ptr);
253 |         downroot->get_sibling(splitkey, sibling_ptr);
254 |     }
255 | 
256 |     return DOWNTREE_NS::update(root_ptr, k, v);
257 | }
258 | 
259 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
260 | void TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::rebuild_fast() { // fast rebuilding function
261 |     // switch the restore to be immutable
262 |     vector<Record> * new_mutable = new vector<Record>;
263 |     new_mutable->reserve(0xffff);
264 |     
265 |     mutable_mtx_.lock();
266 |         vector<Record> * immutable = mutable_; // make the restore vector be immutable
267 |         mutable_ = new_mutable; // before this line, the mutable_ is still the old one
268 |     mutable_mtx_.unlock();
269 | 
270 |     is_rebuilding_ = true;
271 | 
272 |     std::sort(immutable->begin(), immutable->end());
273 |     // get the snapshot of all sub-index trees by combining the top layer with immutable
274 |     std::vector<Record> subroots;
275 |     subroots.reserve(0x2ffff);
276 |     uptree_->merge(*immutable, subroots);
277 | 
278 |     /* rebuild the top layer with immutable */  
279 |     UPTREE_NS::uptree_t * old_tree = uptree_;
280 |     UPTREE_NS::entrance_t * old_upent = galc->absolute(entrance_->upent);
281 |     UPTREE_NS::uptree_t * new_tree = new UPTREE_NS::uptree_t(subroots);
282 |     UPTREE_NS::entrance_t * new_upent = UPTREE_NS::get_entrance(new_tree);
283 |     
284 |     // install the new top layer
285 |     persist_assign(&(entrance_->upent), galc->relative(new_upent));
286 |     uptree_ = new_tree;
287 |     
288 |     /* free the old top layer */
289 |     #ifdef BACKGROUND_REBUILD
290 |         usleep(50); // TODO: unsafe memory reclamation
291 |     #endif
292 |     UPTREE_NS::free(old_tree); // free the old_tree
293 | 
294 |     is_rebuilding_ = false;
295 |     asm volatile("" ::: "memory");
296 |     rebuild_mtx_.unlock();
297 | 
298 |     delete immutable;
299 | }
300 | 
301 | template<int DOWNLEVEL, int REBUILD_THRESHOLD>
302 | void TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD>::rebuild_recover() { // slow rebuilding function 
303 |     is_rebuilding_ = true;
304 |     // get the snapshot of all sub-index trees by traverse in the down layer
305 |     std::vector<Record> subroots;
306 |     subroots.reserve(0x2fffff);
307 |     
308 |     _key_t split_key = 0; 
309 |     Node ** sibling_ptr = (Node **)uptree_->find_first();
310 |     Node * cur_root = (Node *)galc->absolute(*sibling_ptr);
311 |     while (cur_root != NULL) {
312 |         subroots.emplace_back(split_key, (char *)(*sibling_ptr));
313 |         // get next sibling
314 |         cur_root->get_sibling(split_key, sibling_ptr);
315 |         cur_root = galc->absolute(*sibling_ptr);
316 |     }
317 | 
318 |     /* rebuild the top layer with immutable */  
319 |     UPTREE_NS::uptree_t * old_tree = uptree_;
320 |     UPTREE_NS::entrance_t * old_upent = galc->absolute(entrance_->upent);
321 |     UPTREE_NS::uptree_t * new_tree = new UPTREE_NS::uptree_t(subroots);
322 |     UPTREE_NS::entrance_t * new_upent = UPTREE_NS::get_entrance(new_tree);
323 |     
324 |     // install the new top layer
325 |     persist_assign(&(entrance_->upent), galc->relative(new_upent));
326 |     uptree_ = new_tree;
327 |     
328 |     /* free the old top layer */
329 |     #ifdef BACKGROUND_REBUILD
330 |         usleep(50); // TODO: unsafe memory reclamation
331 |     #endif
332 |     UPTREE_NS::free(old_tree); // free the old_tree
333 | 
334 |     is_rebuilding_ = false;
335 |     asm volatile("" ::: "memory");
336 |     rebuild_mtx_.unlock();
337 | 
338 |     persist_assign(&(entrance_->use_rebuild_recover), false); // use fast rebuilding next time
339 | }
340 | 
341 | } // tlbtree namespace
342 | 
343 | #endif //__TLBTREEIMPL_H__


--------------------------------------------------------------------------------
/Concurrent/src/wotree256.cc:
--------------------------------------------------------------------------------
  1 | #include "wotree256.h"
  2 | 
  3 | namespace wotree256 {
  4 | 
  5 | bool insert_recursive(Node * n, _key_t k, uint64_t v, _key_t &split_k, Node * &split_node, int8_t &level) {
  6 |     if(n->leftmost_ptr_ == NULL) {
  7 |         return n->store(k, v, split_k, split_node);
  8 |     } else {
  9 |         level++;
 10 |         Node * child = (Node *) galc->absolute(n->get_child(k));
 11 |         
 12 |         _key_t split_k_child;
 13 |         Node * split_node_child;
 14 |         bool splitIf = insert_recursive(child, k, v, split_k_child, split_node_child, level);
 15 | 
 16 |         if(splitIf) { 
 17 |             return n->store(split_k_child, (uint64_t)galc->relative(split_node_child), split_k, split_node);
 18 |         } 
 19 |         return false;
 20 |     }
 21 | }
 22 | 
 23 | bool remove_recursive(Node * n, _key_t k) {
 24 |     if(n->leftmost_ptr_ == NULL) {
 25 |         n->remove(k);
 26 |         return n->state_.unpack.count < UNDERFLOW_CARD;
 27 |     }
 28 |     else {
 29 |         Node * child = (Node *) galc->absolute(n->get_child(k));
 30 | 
 31 |         bool shouldMrg = remove_recursive(child, k);
 32 | 
 33 |         if(shouldMrg) {
 34 |             Node *leftsib = NULL, *rightsib = NULL;
 35 |             n->get_lrchild(k, leftsib, rightsib);
 36 | 
 37 |             if(leftsib != NULL && (child->state_.unpack.count + leftsib->state_.unpack.count) < CARDINALITY) {
 38 |                 // merge with left node
 39 |                 int8_t slotid = child->state_.read(0);
 40 |                 n->remove(child->recs_[slotid].key);
 41 |                 Node::merge(leftsib, child);
 42 | 
 43 |                 return n->state_.unpack.count < UNDERFLOW_CARD;
 44 |             } else if (rightsib != NULL && (child->state_.unpack.count + rightsib->state_.unpack.count) < CARDINALITY) {
 45 |                 // merge with right node
 46 |                 int8_t slotid = rightsib->state_.read(0);
 47 |                 n->remove(rightsib->recs_[slotid].key);
 48 |                 Node::merge(child, rightsib);
 49 | 
 50 |                 return n->state_.unpack.count < UNDERFLOW_CARD;
 51 |             }
 52 |         }
 53 |         return false;
 54 |     }
 55 | }
 56 | 
 57 | bool find(Node ** rootPtr, _key_t key, uint64_t &val) {
 58 |     Node * cur = galc->absolute(*rootPtr);
 59 |     while(cur->leftmost_ptr_ != NULL) { // no prefetch here
 60 |         char * child_ptr = cur->get_child(key);
 61 |         cur = (Node *)galc->absolute(child_ptr);
 62 |     }
 63 | 
 64 |     val = (uint64_t) cur->get_child(key);
 65 | 
 66 |     if((char *)val == NULL)
 67 |         return false;
 68 |     else
 69 |         return true;
 70 | }
 71 | 
 72 | res_t insert(Node ** rootPtr, _key_t key, uint64_t val, int threshold) {
 73 |     Node *root_= galc->absolute(*rootPtr);
 74 |     
 75 |     int8_t level = 1;
 76 |     _key_t split_k;
 77 |     Node * split_node;
 78 |     bool splitIf = insert_recursive(root_, key, val, split_k, split_node, level);
 79 | 
 80 |     if(splitIf) {
 81 |         if(level < threshold) {
 82 |             Node *new_root = new Node;
 83 |             new_root->leftmost_ptr_ = (char *)galc->relative(root_);
 84 |             new_root->append({split_k, (char *)galc->relative(split_node)}, 0, 0);
 85 |             new_root->state_.unpack.count = 1;
 86 | 
 87 |             clwb(new_root, 64);
 88 | 
 89 |             mfence(); // a barrier to make sure the new node is persisted
 90 |             persist_assign(rootPtr, (Node *)galc->relative(new_root));
 91 | 
 92 |             return res_t(false, {0, NULL});
 93 |         } else {
 94 |             return res_t(true, {split_k, (char *)split_node});
 95 |         }   
 96 |     }
 97 |     else {
 98 |         return res_t(false, {0, NULL});
 99 |     }
100 | }
101 | 
102 | bool update(Node ** rootPtr, _key_t key, uint64_t val) {
103 |     Node * cur = galc->absolute(*rootPtr);
104 |     while(cur->leftmost_ptr_ != NULL) { // no prefetch here
105 |         char * child_ptr = cur->get_child(key);
106 |         cur = (Node *)galc->absolute(child_ptr);
107 |     }
108 | 
109 |     val = (uint64_t) cur->update(key, val);
110 |     return true;
111 | }
112 | 
113 | bool remove(Node ** rootPtr, _key_t key) {   
114 |     Node *root_= galc->absolute(*rootPtr);
115 |     if(root_->leftmost_ptr_ == NULL) {
116 |         root_->remove(key);
117 | 
118 |         return root_->state_.unpack.count == 0;
119 |     }
120 |     else {
121 |         Node * child = (Node *) galc->absolute(root_->get_child(key));
122 | 
123 |         bool shouldMrg = remove_recursive(child, key);
124 | 
125 |         if(shouldMrg) {
126 |             Node *leftsib = NULL, *rightsib = NULL;
127 |             root_->get_lrchild(key, leftsib, rightsib);
128 | 
129 |             if(leftsib != NULL && (child->state_.unpack.count + leftsib->state_.unpack.count) < CARDINALITY) {
130 |                 // merge with left node
131 |                 int8_t slotid = child->state_.read(0);
132 |                 root_->remove(child->recs_[slotid].key);
133 |                 Node::merge(leftsib, child);
134 |             } 
135 |             else if (rightsib != NULL && (child->state_.unpack.count + rightsib->state_.unpack.count) < CARDINALITY) {
136 |                 // merge with right node
137 |                 int8_t slotid = rightsib->state_.read(0);
138 |                 root_->remove(rightsib->recs_[slotid].key);
139 |                 Node::merge(child, rightsib);
140 |             }
141 |             
142 |             if(root_->state_.unpack.count == 0) { // the root is empty
143 |                 Node * old_root = root_;
144 | 
145 |                 persist_assign(rootPtr, (Node *)root_->leftmost_ptr_);
146 | 
147 |                 galc->free(old_root);
148 |             }
149 |         }
150 | 
151 |         return false;
152 |     } 
153 | }
154 | 
155 | void printAll(Node ** rootPtr) {
156 |     Node *root= galc->absolute(*rootPtr);
157 |     root->print("", true);
158 | }
159 | 
160 | } // namespace wotree256


--------------------------------------------------------------------------------
/Concurrent/src/wotree256.h:
--------------------------------------------------------------------------------
  1 | /*
  2 |     log free wbtree with 256B nodesize, nonclass version of wbtree_slotonly
  3 |     Copyright(c) Luo Yongping All rights reserved!
  4 | */
  5 | 
  6 | #ifndef __DOWNTREE256__
  7 | #define __DOWNTREE256__
  8 | 
  9 | #include <cstdint>
 10 | #include <cstring>
 11 | #include <string>
 12 | #include <cstdio>
 13 | #include <thread>
 14 | 
 15 | #include "flush.h"
 16 | #include "pmallocator.h"
 17 | 
 18 | namespace wotree256 {
 19 | 
 20 | inline void barrier() { __asm__ __volatile__("": : :"memory");}
 21 | 
 22 | using std::string;
 23 | constexpr int CARDINALITY = 13;
 24 | constexpr int UNDERFLOW_CARD = 4;
 25 | 
 26 | struct state_t {
 27 |     struct statefield_t { // totally 8 bytes
 28 |         uint64_t slotArray       : 52;
 29 |         uint64_t count           : 4;
 30 |         uint64_t sibling_version : 1;
 31 |         uint64_t latch           : 1;
 32 |         uint64_t node_version    : 6; 
 33 |     };
 34 |     // the real data field of state
 35 |     union {
 36 |         uint64_t pack;      // to do 8 bytes assignment
 37 |         statefield_t unpack;// to facilite accessing subfields
 38 |     };
 39 |     
 40 | public:
 41 |     state_t(uint64_t s = 0): pack(s) {}
 42 | 
 43 |     inline int8_t read(int8_t idx) const {
 44 |         uint64_t p = this->unpack.slotArray << 12;
 45 |         return (p & ((uint64_t)0xf << ((15 - idx) * 4))) >> ((15 - idx) * 4);
 46 |     }
 47 | 
 48 |     inline int8_t alloc() {
 49 |         int8_t occupy[CARDINALITY] = {0};
 50 |         for(int8_t i = 0; i < unpack.count; i++) {
 51 |             occupy[read(i)] = 1; 
 52 |         }
 53 |         for(int i = 0; i < CARDINALITY; i++) {
 54 |             if(occupy[i] == 0) return i;
 55 |         }
 56 |         return CARDINALITY; // never be here
 57 |     }
 58 |     
 59 |     void lock(bool change_version = true) {
 60 |         state_t new_state = pack;
 61 |         new_state.unpack.latch = 0;
 62 |         uint64_t old = new_state.pack;
 63 |         new_state.unpack.latch = 1;
 64 |         if(change_version) new_state.unpack.node_version++;
 65 |         uint64_t desired = new_state.pack;
 66 | 
 67 |         while(!__atomic_compare_exchange(&(this->pack), &old, &desired,
 68 |                 false, __ATOMIC_ACQUIRE, __ATOMIC_ACQUIRE)){
 69 |             // the latch is hold by other thread or the state is changed
 70 |             while(1) { // if the latch is not released, wait it
 71 |                 _mm_pause(); // delay for 140 cycle
 72 |                 if(unpack.latch == 0) // check if the latch is released
 73 |                     break;
 74 |             }
 75 | 
 76 |             state_t new_state = pack;
 77 |             new_state.unpack.latch = 0;
 78 |             old = new_state.pack;
 79 |             new_state.unpack.latch = 1;
 80 |             if(change_version) new_state.unpack.node_version++;
 81 |             desired = new_state.pack;
 82 |         }
 83 |     }
 84 |     
 85 |     void unlock(bool change_version = true) {
 86 |         state_t new_state = pack;
 87 |         new_state.unpack.latch = 0;
 88 |         if(change_version) new_state.unpack.node_version++;
 89 |         uint64_t desired = new_state.pack;
 90 |         uint64_t old;
 91 | 
 92 |         __atomic_exchange(&(this->pack), &desired, &old, __ATOMIC_ACQUIRE);
 93 |     }
 94 | 
 95 |     inline uint64_t add(int8_t idx, int8_t slot) {
 96 |         state_t new_state(this->pack);
 97 | 
 98 |         // update bit fields
 99 |         uint64_t p      = this->unpack.slotArray << 12;
100 |         uint64_t mask   = 0xffffffffffffffff >> (idx * 4);
101 |         uint64_t add_value = (uint64_t)slot << ((15 - idx) * 4);
102 |         new_state.unpack.slotArray = ((p & (~mask)) + add_value + ((p & mask) >> 4)) >> 12;
103 |         new_state.unpack.count++;
104 | 
105 |         return new_state.pack;
106 |     }
107 | 
108 |     inline uint64_t remove(int idx) { // delete a slot id at position idx
109 |         state_t new_state(this->pack);
110 |         // update bit fields
111 |         uint64_t p      = this->unpack.slotArray << 12;
112 |         uint64_t mask   = 0xffffffffffffffff >> (idx * 4);
113 |         new_state.unpack.slotArray = ((p & ~mask) + ((p & (mask>>4)) << 4)) >> 12;
114 |         new_state.unpack.count--;
115 | 
116 |         return new_state.pack;
117 |     }
118 | 
119 |     inline uint64_t append(int8_t idx, int8_t slot) {
120 |         // Append the record at slotid, append the slotArray entry, but DO NOT modify the count
121 |         state_t new_state(this->pack);
122 | 
123 |         // update bit fields
124 |         uint64_t p      = this->unpack.slotArray << 12;
125 |         uint64_t mask   = 0xffffffffffffffff >> (idx * 4);
126 |         uint64_t add_value = (uint64_t)slot << ((15 - idx) * 4);
127 |         new_state.unpack.slotArray = ((p & (~mask)) + add_value + ((p & mask) >> 4)) >> 12;
128 | 
129 |         return new_state.pack;
130 |     }
131 | };
132 | 
133 | class Node {
134 | public:
135 |     // First Cache Line
136 |     state_t state_;      // a very complex and compact state field
137 |     char * leftmost_ptr_;// the left most child of current node
138 |     Record siblings_[2]; // shadow sibling of current node
139 |     // Slots 
140 |     Record recs_[CARDINALITY];
141 | 
142 |     friend class wbtree;
143 | 
144 | public:
145 |     Node(bool isleaf = false): state_(0), leftmost_ptr_(NULL) {
146 |         siblings_[0] = {MAX_KEY, NULL};
147 |         siblings_[1] = {MAX_KEY, NULL};
148 |     }
149 | 
150 |     void *operator new(size_t size) {
151 |         void * ret = galc->malloc(size);
152 |         return ret;
153 |     }
154 | 
155 |     bool store(_key_t k, uint64_t v, _key_t & split_k, Node * & split_node) {
156 |         // there is one exclusive writer 
157 |         state_.lock();
158 | 
159 |         Record &sibling = siblings_[state_.unpack.sibling_version]; // the sibling is updated atomically, we are safe here
160 |         if(k >= sibling.key) { // if the node has splitted and k to find is in next node 
161 |             Node * sib_node = (Node *)galc->absolute(sibling.val);
162 |             state_.unlock();
163 |             return sib_node->store(k, v, split_k, split_node);
164 |         }
165 | 
166 |         if(state_.unpack.count == CARDINALITY) { // should split the node
167 |             uint64_t m = state_.unpack.count / 2;
168 |             split_k = recs_[state_.read(m)].key;
169 | 
170 |             // copy half of the records into split node
171 |             int8_t j = 0;
172 |             state_t new_state = state_;
173 |             if(leftmost_ptr_ == NULL) {
174 |                 split_node = new Node;
175 |                 split_node->state_.lock();
176 |                 for(int i = m; i < state_.unpack.count; i++) {
177 |                     int8_t slotid = state_.read(i);
178 |                     split_node->append(recs_[slotid], j, j);
179 |                     j += 1;
180 |                 }
181 | 
182 |                 new_state.unpack.count -= j;
183 |             } else {
184 |                 int8_t slotid = state_.read(m);
185 |                 split_node = new Node();
186 |                 split_node->state_.lock();
187 |                 split_node->leftmost_ptr_ = recs_[slotid].val;
188 | 
189 |                 for(int i = m + 1; i < state_.unpack.count; i++) {
190 |                     slotid = state_.read(i);
191 |                     split_node->append(recs_[slotid], j, j);
192 |                     j += 1;
193 |                 }
194 | 
195 |                 new_state.unpack.count -= (j + 1);
196 |             }
197 |             split_node->state_.unpack.count = j;
198 |             split_node->state_.unpack.sibling_version = 0;
199 |             // the sibling node of current node pointed by split_node
200 |             split_node->siblings_[0] = siblings_[state_.unpack.sibling_version];
201 |             clwb(split_node, 64); // persist header
202 |             clwb(&split_node->recs_[1], sizeof(Record) * (j - 1)); // persist all the inserted records
203 |             
204 |             // the split node is installed as the shadow sibling of current node
205 |             siblings_[(state_.unpack.sibling_version + 1) % 2] = {split_k, (char *)galc->relative(split_node)};
206 |             // persist_assign the state field
207 |             new_state.unpack.sibling_version = (state_.unpack.sibling_version + 1) % 2;
208 |             mfence(); // a barrier here to make sure all the update is persisted to storage
209 | 
210 |             persist_assign(&(state_.pack), new_state.pack);
211 |             
212 |             // go on the insertion
213 |             if(k < split_k) {
214 |                 insertone(k, (char *)v);
215 |             } else {
216 |                 split_node->insertone(k, (char *)v);                
217 |             }
218 |             split_node->state_.unlock();
219 |             state_.unlock();
220 |             return true;
221 |         } else {
222 |             insertone(k, (char *)v);
223 | 
224 |             state_.unlock();
225 |             return false;
226 |         }
227 |     }
228 | 
229 |     char * get_child(_key_t k) { 
230 |         // use optimized lock to coordinate reader with writer
231 |         get_retry:
232 |         uint64_t old_version = state_.unpack.node_version;
233 |         barrier();
234 | 
235 |         Record &sibling = siblings_[state_.unpack.sibling_version]; // the sibling is updated atomically, we are safe here
236 |         if(k >= sibling.key) { // if the node has splitted and k to find is in next node 
237 |             Node * sib_node = (Node *)galc->absolute(sibling.val);
238 |             
239 |             barrier();
240 |             if(old_version != state_.unpack.node_version || old_version % 2 != 0) {
241 |                 goto get_retry;
242 |             }
243 |             return sib_node->get_child(k);
244 |         }
245 | 
246 |         if(leftmost_ptr_ == NULL) {
247 |             int8_t slotid = 0;
248 |             for(int i = 0; i < state_.unpack.count; i++) {
249 |                 slotid = state_.read(i);
250 |                 if(recs_[slotid].key >= k) {
251 |                     break;
252 |                 }
253 |             }
254 | 
255 |             char * ret;
256 |             if (recs_[slotid].key == k && state_.unpack.count > 0)
257 |                 ret = recs_[slotid].val;
258 |             else 
259 |                 ret =  NULL;
260 |             
261 |             barrier();
262 |             if(old_version != state_.unpack.node_version || old_version % 2 != 0) {
263 |                 goto get_retry;
264 |             }
265 |             return ret;
266 |         } else {
267 |             int8_t slotid = 0, pos = state_.unpack.count;
268 |             for(int i = 0; i < state_.unpack.count; i++) {
269 |                 slotid = state_.read(i);
270 |                 if(recs_[slotid].key > k) {
271 |                     pos = i;
272 |                     break;
273 |                 }
274 |             }
275 | 
276 |             char * ret;
277 |             if (pos == 0) // all the key is bigger than k
278 |                 ret = leftmost_ptr_;
279 |             else 
280 |                 ret = recs_[state_.read(pos - 1)].val;
281 |             
282 |             barrier();
283 |             if(old_version != state_.unpack.node_version || old_version % 2 != 0) {
284 |                 goto get_retry;
285 |             }
286 |             return ret;
287 |         }
288 |     }
289 | 
290 |     bool update(_key_t k, uint64_t v) {
291 |         state_.lock(false);
292 | 
293 |         Record &sibling = siblings_[state_.unpack.sibling_version]; // the sibling is updated atomically, we are safe here
294 |         if(k >= sibling.key) { // if the node has splitted and k to find is in next node 
295 |             Node * sib_node = (Node *)galc->absolute(sibling.val);
296 |             state_.unlock(false);
297 |             return sib_node->update(k, v);
298 |         }
299 | 
300 |         uint64_t slotid = 0;
301 |         for(int i = 0; i < state_.unpack.count; i++) {
302 |             slotid = state_.read(i);
303 |             if(recs_[slotid].key >= k) {
304 |                 break;
305 |             }
306 |         }
307 | 
308 |         bool found = false;
309 |         if (recs_[slotid].key == k) {
310 |             recs_[slotid].val = (char *)v;
311 |             clwb(&recs_[slotid], sizeof(Record));
312 |             found = true;
313 |         }
314 | 
315 |         state_.unlock(false);
316 |         return found;
317 |     }
318 | 
319 |     bool remove(_key_t k) {
320 |         // Non-SMO delete takes only one clwb 
321 |         state_.lock();
322 |         Record &sibling = siblings_[state_.unpack.sibling_version];
323 |         if(k >= sibling.key) { // if the node has splitted and k to find is in next node 
324 |             Node * sib_node = (Node *)galc->absolute(sibling.val);
325 |             state_.unlock();
326 |             return sib_node->remove(k);
327 |         }
328 | 
329 |         if(leftmost_ptr_ == NULL) {
330 |             int8_t idx, slotid;
331 |             for(idx = 0; idx < state_.unpack.count; idx++) {
332 |                 slotid = state_.read(idx);
333 |                 if(recs_[slotid].key >= k)
334 |                     break;
335 |             }
336 | 
337 |             if(recs_[slotid].key == k) {
338 |                 uint64_t newpack = state_.remove(idx);
339 |                 persist_assign(&(state_.pack), newpack);
340 |                 state_.unlock();
341 |                 return true;
342 |             } else {
343 |                 state_.unlock();
344 |                 return false;
345 |             }
346 |         } else {
347 |             int8_t idx;
348 |             for(idx = 0; idx < state_.unpack.count; idx++) {
349 |                 int8_t slotid = state_.read(idx);
350 |                 if(recs_[slotid].key > k) 
351 |                     break;
352 |             }
353 |             /* NOTICE: 
354 |                 * We will never remove the leftmost child in our wbtree design
355 |                 * So the idx here must be larger than 0
356 |                 */
357 |             uint64_t newpack = state_.remove(idx - 1);
358 |             persist_assign(&(state_.pack), newpack);
359 |             
360 |             state_.unlock();
361 |             return true;
362 |         }
363 |     }
364 | 
365 |     void print(string prefix, bool recursively) const {
366 |         printf("%s[%lx(%ld) ", prefix.c_str(), state_.unpack.slotArray, state_.unpack.count);
367 | 
368 |         for(int i = 0; i < state_.unpack.count; i++) {
369 |             printf("%d ", state_.read(i));
370 |         }
371 | 
372 |         for(int i = 0; i < state_.unpack.count; i++) {
373 |             int8_t slotid = state_.read(i);
374 |             printf("(%ld 0x%lx) ", recs_[slotid].key, (uint64_t)recs_[slotid].val);
375 |         }
376 |         printf("]\n");
377 | 
378 |         if(recursively && leftmost_ptr_ != NULL) {
379 |             Node * child = (Node *)galc->absolute(leftmost_ptr_);
380 |             child->print(prefix + "    ", recursively);
381 | 
382 |             for(int i = 0; i < state_.unpack.count; i++) {
383 |                 Node * child = (Node *)galc->absolute(recs_[state_.read(i)].val);
384 |                 child->print(prefix + "    ", recursively);
385 |             }
386 |         }
387 |     }
388 | 
389 |     void get_sibling(_key_t & k, Node ** &sibling) {
390 |         Record &sib = siblings_[state_.unpack.sibling_version];
391 |         k = sib.key;
392 |         sibling = (Node **)&(sib.val);
393 |     }
394 | 
395 | public:
396 |     void insertone(_key_t key, char * right) {
397 |         int8_t idx;
398 |         for(idx = 0; idx < state_.unpack.count; idx++) {
399 |             int8_t slotid = state_.read(idx);
400 |             if(key < recs_[slotid].key) {
401 |                 break;
402 |             }
403 |         }
404 |         
405 |         // insert and flush the kv
406 |         int8_t slotid = state_.alloc(); // alloc a slot in the node
407 |         recs_[slotid] = {key, (char *) right};
408 |         clwb(&recs_[slotid], sizeof(Record));
409 |         mfence();
410 | 
411 |         // atomically update the state
412 |         uint64_t new_pack = state_.add(idx, slotid);
413 |         persist_assign(&(state_.pack), new_pack);
414 |     }
415 | 
416 |     void append(Record r, int8_t slotid, int8_t pos) {
417 |         recs_[slotid] = r;
418 |         state_.pack = state_.append(pos, slotid);
419 |     }
420 | 
421 |     static void merge(Node * left, Node * right) {
422 |         left->state_.lock();
423 |         right->state_.lock();
424 | 
425 |         Record & sibling = left->siblings_[left->state_.unpack.sibling_version];
426 | 
427 |         state_t new_state = left->state_;
428 |         if(left->leftmost_ptr_ != NULL) { // insert the leftmost_ptr of the right node
429 |             int8_t slotid = new_state.alloc();
430 |             left->append({sibling.key, right->leftmost_ptr_}, slotid, new_state.unpack.count);
431 |             new_state.pack = new_state.add(new_state.unpack.count, slotid);
432 |         }
433 |         for(int i = 0; i < right->state_.unpack.count; i++) {
434 |             int8_t slotid = new_state.alloc();
435 |             left->append(right->recs_[right->state_.read(i)], slotid, new_state.unpack.count);
436 |             new_state.pack = new_state.add(new_state.unpack.count, slotid);;
437 |         }
438 |         
439 |         Record tmp = right->siblings_[right->state_.unpack.sibling_version];
440 |         left->siblings_[(left->state_.unpack.sibling_version + 1) % 2] = tmp;
441 |         new_state.unpack.sibling_version = (left->state_.unpack.sibling_version + 1) % 2;
442 |         clwb(left, sizeof(Node)); // persist the whole leaf node
443 | 
444 |         // persist_assign the state_ field
445 |         mfence();
446 |         left->state_.pack = new_state.pack;
447 |         clwb(left, 64);
448 | 
449 |         left->state_.unlock();
450 | 
451 |         galc->free(right); // WARNING: persistent memory leak here
452 |     }
453 | 
454 |     void get_lrchild(_key_t k, Node * & left, Node * & right) {
455 |         get_retry:
456 |         uint64_t old_version = state_.unpack.node_version;
457 |         barrier();
458 |         if(old_version != state_.unpack.node_version || old_version % 2 != 0) {
459 |             goto get_retry;
460 |         }
461 | 
462 |         int16_t i = 0;
463 |         for( ; i < state_.unpack.count; i++) {
464 |             int8_t slotid = state_.read(i);
465 |             if(recs_[slotid].key > k)
466 |                 break;
467 |         }
468 | 
469 |         if(i == 0) {
470 |             left = NULL;
471 |         } else if(i == 1) {
472 |             left = (Node *)galc->absolute(leftmost_ptr_);
473 |         } else {
474 |             left = (Node *)galc->absolute(recs_[state_.read(i - 2)].val);
475 |         }
476 | 
477 |         if(i == state_.unpack.count) {
478 |             right = NULL;
479 |         } else {
480 |             right = (Node *)galc->absolute(recs_[state_.read(i)].val);
481 |         }
482 | 
483 |         barrier();
484 |         if(old_version != state_.unpack.node_version || old_version % 2 != 0) {
485 |             goto get_retry;
486 |         }
487 |         return ;
488 |     }
489 | };
490 | 
491 | extern bool insert_recursive(Node * n, _key_t k, uint64_t v, _key_t &split_k, 
492 |                                 Node * &split_node, int8_t &level);
493 | extern bool remove_recursive(Node * n, _key_t k);
494 | extern bool find(Node ** rootPtr, _key_t key, uint64_t &val);
495 | extern res_t insert(Node ** rootPtr, _key_t key, uint64_t val, int threshold);
496 | extern bool update(Node ** rootPtr, _key_t key, uint64_t val);
497 | extern bool remove(Node ** rootPtr, _key_t key);
498 | extern void printAll(Node ** rootPtr);
499 | 
500 | } // namespace wotree256
501 | 
502 | #endif // __DOWNTREE256__


--------------------------------------------------------------------------------
/Concurrent/test/CMakeLists.txt:
--------------------------------------------------------------------------------
1 | add_executable(datagen "datagen.cc")
2 | 
3 | add_executable(main "main.cc")
4 | target_link_libraries(main tlbtree)
5 | 
6 | add_executable(preload "preload.cc")
7 | target_link_libraries(preload tlbtree)


--------------------------------------------------------------------------------
/Concurrent/test/datagen.cc:
--------------------------------------------------------------------------------
  1 | #include <iostream>
  2 | #include <fstream>
  3 | #include <string>
  4 | #include <random>
  5 | #include <cstdlib>
  6 | #include <unistd.h>
  7 | #include <algorithm>
  8 | 
  9 | #include "common.h"
 10 | #include "zipfian.h"
 11 | 
 12 | using std::cout;
 13 | using std::endl;
 14 | using std::string;
 15 | using std::ofstream;
 16 | using std::ifstream;
 17 | 
 18 | enum DistributionType {RAND = 0, ZIPFIAN};
 19 | 
 20 | struct WorkloadType {
 21 |     int operations = MILLION;
 22 |     float read = 1.0;
 23 |     float insert = 0;
 24 |     float update = 0;
 25 |     float remove = 0;
 26 |     DistributionType dist = RAND;
 27 |     float skewness = 0.8;
 28 |     bool valid() {
 29 |         return read + insert + update + remove == 1.0 && skewness > 0 && skewness < 1.0;
 30 |     }
 31 |     void print() {
 32 |         cout << "=========WORKLOAD TYPE=========" << endl;
 33 |         cout << "Operations  : " << operations << endl;
 34 |         cout << "Read Ratio  : " << read << endl;
 35 |         cout << "Insert Ratio: " << insert << endl;
 36 |         cout << "Update Ratio: " << update << endl;
 37 |         cout << "Remove Ratio: " << remove << endl;
 38 |         cout << "Distribution: " << (dist == RAND ? "random" : "Zipfian") << endl;
 39 |         if (dist == ZIPFIAN) {
 40 |             cout << "Skewness " << skewness << endl;
 41 |         }
 42 |         cout << "===============================" << endl;
 43 |     }
 44 | };
 45 | 
 46 | class OperationGenerator {
 47 | public :
 48 |     OperationType mappings_[100];
 49 |     std::default_random_engine gen_;
 50 |     std::uniform_int_distribution<uint32_t> dist_;
 51 | 
 52 |     OperationGenerator(WorkloadType &w) : gen_(getRandom()) {
 53 |         int read_end = 100 * w.read;
 54 |         int insert_end = read_end + 100 * w.insert;
 55 |         int update_end = insert_end + 100 * w.update;
 56 |         int remove_end = update_end + 100 * w.remove;
 57 | 
 58 |         for(int i = 0; i < read_end; i++)
 59 |             mappings_[i] = OperationType::READ;
 60 | 
 61 |         for(int i = read_end; i < insert_end; i++)
 62 |             mappings_[i] = OperationType::INSERT;
 63 | 
 64 |         for(int i = insert_end; i < update_end; i++)
 65 |             mappings_[i] = OperationType::UPDATE;
 66 | 
 67 |         for(int i = update_end; i < remove_end; i++)
 68 |             mappings_[i] = OperationType::DELETE;
 69 |     }
 70 | 
 71 |     OperationType next() {
 72 |         return mappings_[dist_(gen_) % 100];
 73 |     }
 74 | };
 75 | 
 76 | // try not to generate duplicate keys, because some tree indices may not get used to it
 77 | 
 78 | void gen_dataset(int64_t *arr, uint64_t scale, bool random) {
 79 |     std::mt19937 gen(10007);
 80 |     if(random) {
 81 |         uint64_t step = MAX_KEY / scale;
 82 |         std::uniform_int_distribution<int64_t> dist(0, MAX_KEY);
 83 |         for(int64_t i = 0; i < scale; i++) {
 84 |             #ifdef DEBUG
 85 |                 arr[i] = i + 1;
 86 |             #else  
 87 |                 arr[i] = i * step + 1; // non duplicated keys, and no zero key 
 88 |             #endif
 89 |         }
 90 |         std::shuffle(arr, arr + scale - 1, gen);
 91 |     } else {
 92 |         std::normal_distribution<double> dist(MAX_KEY / 2, MAX_KEY / 8);
 93 |         int64_t i = 0;
 94 |         while (i < scale) {
 95 |             double val = (uint64_t)dist(gen);
 96 |             if(val < 0 || val > (double) MAX_KEY) {
 97 |                 continue;
 98 |             } else {
 99 |                 arr[i++] = (int64_t)std::round(val);
100 |             }
101 |         }
102 |     }
103 |     return ;
104 | }
105 | 
106 | void gen_workload(int64_t *arr, uint64_t scale, QueryType * querys, WorkloadType w) {
107 |     std::mt19937 gen(getRandom());
108 |     std::uniform_int_distribution<int64_t> idx1_dist(0, scale - 1);
109 |     zipfian_int_distribution<int64_t> idx2_dist(0, scale - 1, w.skewness);
110 |     OperationGenerator op_gen(w);
111 | 
112 |     for(int i = 0; i < w.operations; i++) {
113 |         OperationType op = op_gen.next();
114 |         int idx = (w.dist == RAND ? idx1_dist(gen) : idx2_dist(gen));
115 |         // for insert operations, we should make sure the key does not exist in the dataset
116 |         _key_t key = (op == OperationType::INSERT ? arr[idx] + getRandom(): arr[idx]); 
117 | 
118 |         querys[i] = {op, key};
119 |     }
120 | }
121 | 
122 | int main(int argc, char ** argv) {
123 |     static const bool DATASET_RANDOM = true;
124 |     bool opt_zipfian  = false;
125 |     WorkloadType w;
126 | 
127 |     static const char * optstr = "r:i:u:d:o:s:hz"; 
128 |     opterr = 0;
129 |     char opt;
130 |     while((opt = getopt(argc, argv, optstr)) != -1) {
131 |         switch(opt) {
132 |         case 'o':
133 |             w.operations = atoi(optarg) * MILLION;
134 |             break;
135 |         case 's':
136 |             w.skewness = atof(optarg);
137 |             break;
138 |         case 'r':
139 |             w.read = atof(optarg);
140 |             break;
141 |         case 'i':
142 |             w.insert = atof(optarg);
143 |             break;
144 |         case 'd':
145 |             w.remove = atof(optarg);
146 |             break;
147 |         case 'u':
148 |             w.update = atof(optarg);
149 |             break;
150 |         case 'z':
151 |             opt_zipfian = true;
152 |             break;
153 |         case '?':
154 |         case 'h':
155 |         default:
156 |             cout << "USAGE: "<< argv[0] << "[option]" << endl;
157 |             cout << "\t -h: " << "Print the USAGE" << endl;
158 |             cout << "\t -z: " << "Use zipfian distribution (Not specified: random distribution)" << endl;
159 |             cout << "\t -o: " << "The number of operations" << endl;
160 |             cout << "\t -s: " << "The skewness of query workload(0 - 1)" << endl;
161 |             cout << "\t -r: " << "Read ratio" << endl;
162 |             cout << "\t -i: " << "Insert ratio" << endl;
163 |             cout << "\t -u: " << "update ratio" << endl;
164 |             cout << "\t -d: " << "Delete ratio" << endl;
165 |             exit(-1);
166 |         }
167 |     }
168 | 
169 |     if(!w.valid()) {
170 |         cout << "Invalid workload configuration" << endl;
171 |         exit(-1);
172 |     }
173 |     if(opt_zipfian == true) {
174 |         w.dist = ZIPFIAN;
175 |     }
176 | 
177 |     w.print();
178 | 
179 |     #ifdef DEBUG
180 |         uint64_t scale = LOADSCALE * KILO;
181 |     #else
182 |         uint64_t scale = LOADSCALE * MILLION;
183 |     #endif
184 | 
185 |     int64_t * arr = new int64_t[scale];
186 |     if(!file_exist("dataset.dat")) {
187 |         gen_dataset(arr, scale, DATASET_RANDOM);
188 |         ofstream fout("dataset.dat", std::ios::binary);
189 |         fout.write((char *) arr, sizeof(int64_t) * scale);
190 |         fout.close();
191 |         cout << "generate a dataset file" << endl;
192 |     } else {
193 |         ifstream fin("dataset.dat", std::ios::binary);
194 |         fin.read((char *)arr, sizeof(int64_t) * scale);
195 |         fin.close();
196 |     }
197 | 
198 |     QueryType * querys = new QueryType[w.operations];
199 |     gen_workload(arr, scale, querys, w);
200 |     
201 |     ofstream fout("workload.txt");
202 |     for(int i = 0; i < w.operations; i++) {
203 |         fout << querys[i].op << " " << querys[i].key << endl;
204 |     }
205 |     fout.close();
206 |     cout << "generate a query workload file" << endl;
207 | 
208 |     delete [] arr;
209 |     delete [] querys;
210 |     return 0;
211 | }


--------------------------------------------------------------------------------
/Concurrent/test/main.cc:
--------------------------------------------------------------------------------
  1 | #include <iostream>
  2 | #include <fstream>
  3 | #include <cstdint>
  4 | #include <cstdlib>
  5 | #include <vector>
  6 | #include <random>
  7 | #include <algorithm>
  8 | #include <omp.h>
  9 | 
 10 | #include "tlbtree.h"
 11 | 
 12 | using std::cout;
 13 | using std::endl;
 14 | using std::ifstream;
 15 | using std::string;
 16 | 
 17 | template<typename BtreeType>
 18 | double run_test(std::vector<QueryType> querys, int thread_cnt) {
 19 |     // construct a Btree
 20 |     BtreeType tree("/mnt/pmem/tlbtree.pool");
 21 |     
 22 |     std::atomic_int cur_pos(0);
 23 |     int small_noise = getRandom() & 0xff; // each time we run, we will insert different keys
 24 |     
 25 |     // set the timer
 26 |     #pragma omp barrier
 27 |     auto start = seconds();
 28 | 
 29 |     // start the section of parallel 
 30 |     #pragma omp parallel num_threads(thread_cnt)
 31 |     {
 32 |         #pragma omp for schedule(static)
 33 |         for (size_t i = 0; i < querys.size(); ++i) {
 34 |             // get a unique query from querys
 35 |             int obtain_pos = cur_pos.fetch_add(1, std::memory_order_relaxed);
 36 |             OperationType op = querys[obtain_pos].op;
 37 |             _key_t key = querys[obtain_pos].key;
 38 |             uint64_t val = (uint64_t)key;
 39 | 
 40 |             switch (op) {
 41 |                 case OperationType::READ: {
 42 |                     auto val = tree.lookup(key);
 43 |                     assert(val != 0);
 44 |                     break;
 45 |                 }
 46 |                 case OperationType::INSERT: {
 47 |                     tree.insert(key + small_noise, uint64_t((uint64_t)val + small_noise));
 48 |                     break;
 49 |                 }
 50 |                 case OperationType::UPDATE: {
 51 |                     auto r = tree.update(key, val);
 52 |                     assert(r);
 53 |                     break;
 54 |                 }
 55 |                 case OperationType::DELETE: {
 56 |                     auto r = tree.remove(key);
 57 |                     assert(r);
 58 |                     break;
 59 |                 }
 60 |                 default:
 61 |                     std::cout << "Error: unknown operation!" << std::endl;
 62 |                     exit(0);
 63 |                     break;
 64 |             }
 65 |         }
 66 |     }
 67 | 
 68 |     #pragma omp barrier
 69 |     auto end = seconds();
 70 | 
 71 |     return end - start;
 72 | }
 73 | 
 74 | 
 75 | int main(int argc, char ** argv) {
 76 |     string opt_fname = "../build/workload.txt";
 77 |     int opt_num_thread = 1;
 78 | 
 79 |     static const char * optstr = "f:t:h";
 80 |     opterr = 0;
 81 |     char opt;
 82 |     while((opt = getopt(argc, argv, optstr)) != -1) {
 83 |         switch(opt) {
 84 |         case 'f':
 85 |             opt_fname = string(optarg);
 86 |             break;
 87 |         case 't':
 88 |             if(atoi(optarg) > 0)
 89 |                 opt_num_thread = atoi(optarg);
 90 |             break;
 91 |         case '?':
 92 |         case 'h':
 93 |         default:
 94 |             cout << "USAGE: "<< argv[0] << "[option]" << endl;
 95 |             cout << "\t -h: " << "Print the USAGE" << endl;
 96 |             cout << "\t -f: " << "Filename of the workload" << endl;
 97 |             cout << "\t -t: " << "Number of Threads to excute the workload" << endl;
 98 |             cout << "\t -i: " << "The index tree type" << endl;
 99 |             exit(-1);
100 |             break;
101 |         }
102 |     }
103 | 
104 |     ifstream fin(opt_fname.c_str());
105 |     if(!fin) {
106 |         cout << "workload file not openned" << endl;
107 |         exit(-1);
108 |     }
109 |     std::vector<QueryType> querys;
110 |     int op;
111 |     _key_t key;
112 |     while(fin >> op >> key) {
113 |         querys.push_back({(OperationType)op, key});
114 |     }
115 |     double time = run_test<TLBtree>(querys, opt_num_thread);
116 | 
117 |     cout << time << endl;
118 | 
119 |     return 0;
120 | }


--------------------------------------------------------------------------------
/Concurrent/test/preload.cc:
--------------------------------------------------------------------------------
 1 | #include <iostream>
 2 | #include <fstream>
 3 | #include <cstdint>
 4 | #include <cstdlib>
 5 | #include <vector>
 6 | #include <omp.h>
 7 | 
 8 | #include "tlbtree.h"
 9 | 
10 | using std::cout;
11 | using std::endl;
12 | using std::ifstream;
13 | 
14 | typedef double mytime_t;
15 | 
16 | _key_t *keys;
17 | 
18 | template <typename BTreeType>
19 | void preload(BTreeType &tree, int64_t load_size, ifstream & fin, int thread_cnt) {
20 |     #pragma omp parallel num_threads(thread_cnt)
21 |     {
22 |         #pragma omp for schedule(static)
23 |         for(int64_t i = 0; i < load_size; i++) {
24 |             _key_t key = keys[i];
25 |             tree.insert((_key_t)key, (uint64_t)key);
26 |         }
27 |     }
28 |     return ;
29 | }
30 | 
31 | int main(int argc, char ** argv) {
32 |     int num_threads = 4;
33 | 
34 |     if(argc > 1 && atoi(argv[1]) > 0) {
35 |         num_threads = atoi(argv[1]);
36 |     }
37 |     // open the data file
38 |     std::string filename = "/home/lyp/TLBtree/Concurrent/build/dataset.dat";
39 |     std::ifstream fin(filename.c_str(), std::ios::binary);
40 |     if(!fin) {
41 |         cout << "File not exists or Open error\n";
42 |         exit(-1);
43 |     }
44 | 
45 |     // read all the key into vector keys
46 |     keys = new _key_t[sizeof(_key_t) * LOADSCALE * MILLION];
47 |     fin.read((char *)keys, sizeof(_key_t) * LOADSCALE * MILLION);
48 |     
49 |     cout << "tlbtree" << endl;
50 |     TLBtree tree("/mnt/pmem/tlbtree.pool");
51 |     preload(tree, LOADSCALE * MILLION, fin, num_threads);
52 | 
53 |     delete keys;
54 |     fin.close();
55 | 
56 |     return 0;
57 | }
58 | 


--------------------------------------------------------------------------------
/Concurrent/test/zipfian.h:
--------------------------------------------------------------------------------
  1 | /* Implementation derived from:
  2 |  * "Quickly Generating Billion-Record Synthetic Databases", Jim Gray et al,
  3 |  * SIGMOD 1994
  4 |  *
  5 |  * The zipfian_int_distribution class is intended to be compatible with other
  6 |  * distributions introduced in #include <random> by the C++11 standard.
  7 |  *
  8 |  * Usage example:
  9 |  * #include <random>
 10 |  * #include "zipfian_int_distribution.h"
 11 |  * int main()
 12 |  * {
 13 |  *   std::default_random_engine generator;
 14 |  *   zipfian_int_distribution<int> distribution(1, 10, 0.99);
 15 |  *   int i = distribution(generator);
 16 |  * }
 17 |  */
 18 | 
 19 | /*
 20 |  * IMPORTANT: constructing the distribution object requires calculating the zeta
 21 |  * value which becomes prohibetively expensive for very large ranges. As an
 22 |  * alternative for such cases, the user can pass the pre-calculated values and
 23 |  * avoid the calculation every time.
 24 |  *
 25 |  * Usage example:
 26 |  * #include <random>
 27 |  * #include "zipfian_int_distribution.h"
 28 |  * int main()
 29 |  * {
 30 |  *   std::default_random_engine generator;
 31 |  *   zipfian_int_distribution<int>::param_type p(1, 1e6, 0.99, 27.000);
 32 |  *   zipfian_int_distribution<int> distribution(p);
 33 |  *   int i = distribution(generator);
 34 |  * }
 35 |  */
 36 | 
 37 | #include <cassert>
 38 | #include <cmath>
 39 | #include <limits>
 40 | #include <random>
 41 | 
 42 | template <typename _IntType = int>
 43 | class zipfian_int_distribution
 44 | {
 45 |     static_assert(std::is_integral<_IntType>::value, "Template argument not an integral type.");
 46 | 
 47 | public:
 48 |     /** The type of the range of the distribution. */
 49 |     typedef _IntType result_type;
 50 |     /** Parameter type. */
 51 |     struct param_type
 52 |     {
 53 |         typedef zipfian_int_distribution<_IntType> distribution_type;
 54 | 
 55 |         explicit param_type(_IntType __a = 0, _IntType __b = std::numeric_limits<_IntType>::max(), double __theta = 0.99)
 56 |             : _M_a(__a), _M_b(__b), _M_theta(__theta),
 57 |               _M_zeta(zeta(_M_b - _M_a + 1, __theta)), _M_zeta2theta(zeta(2, __theta))
 58 |         {
 59 |             assert(_M_a <= _M_b && _M_theta > 0.0 && _M_theta < 1.0);
 60 |         }
 61 | 
 62 |         explicit param_type(_IntType __a, _IntType __b, double __theta, double __zeta)
 63 |             : _M_a(__a), _M_b(__b), _M_theta(__theta), _M_zeta(__zeta),
 64 |               _M_zeta2theta(zeta(2, __theta))
 65 |         {
 66 |             __glibcxx_assert(_M_a <= _M_b && _M_theta > 0.0 && _M_theta < 1.0);
 67 |         }
 68 | 
 69 |         result_type a() const { return _M_a; }
 70 | 
 71 |         result_type b() const { return _M_b; }
 72 | 
 73 |         double theta() const { return _M_theta; }
 74 | 
 75 |         double zeta() const { return _M_zeta; }
 76 | 
 77 |         double zeta2theta() const { return _M_zeta2theta; }
 78 | 
 79 |         friend bool operator==(const param_type& __p1, const param_type& __p2)
 80 |         {
 81 |             return __p1._M_a == __p2._M_a
 82 |                 && __p1._M_b == __p2._M_b
 83 |                 && __p1._M_theta == __p2._M_theta
 84 |                 && __p1._M_zeta == __p2._M_zeta
 85 |                 && __p1._M_zeta2theta == __p2._M_zeta2theta;
 86 |         }
 87 | 
 88 |     private:
 89 |         _IntType _M_a;
 90 |         _IntType _M_b;
 91 |         double _M_theta;
 92 |         double _M_zeta;
 93 |         double _M_zeta2theta;
 94 | 
 95 |         /**
 96 |          * @brief Calculates zeta.
 97 |          *
 98 |          * @param __n [IN]  The size of the domain.
 99 |          * @param __theta [IN]  The skew factor of the distribution.
100 |          */
101 |         double zeta(unsigned long __n, double __theta)
102 |         {
103 |             double ans = 0.0;
104 |             for (unsigned long i=1; i<=__n; ++i)
105 |                 ans += std::pow(1.0 / i, __theta);
106 |             return ans;
107 |         }
108 |     };
109 | 
110 | public:
111 |     /**
112 |      * @brief Constructs a zipfian_int_distribution object.
113 |      *
114 |      * @param __a [IN]  The lower bound of the distribution.
115 |      * @param __b [IN]  The upper bound of the distribution.
116 |      * @param __theta [IN]  The skew factor of the distribution.
117 |      */
118 |     explicit zipfian_int_distribution(_IntType __a = _IntType(0), _IntType __b = _IntType(1), double __theta = 0.99)
119 |         : _M_param(__a, __b, __theta)
120 |     {
121 |     }
122 | 
123 |     explicit zipfian_int_distribution(const param_type& __p) : _M_param(__p)
124 |     {
125 |     }
126 | 
127 |     /**
128 |      * @brief Resets the distribution state.
129 |      *
130 |      * Does nothing for the zipfian int distribution.
131 |      */
132 |     void reset() {}
133 | 
134 |     result_type a() const { return _M_param.a(); }
135 | 
136 |     result_type b() const { return _M_param.b(); }
137 | 
138 |     double theta() const { return _M_param.theta(); }
139 | 
140 |     /**
141 |      * @brief Returns the parameter set of the distribution.
142 |      */
143 |     param_type param() const { return _M_param; }
144 | 
145 |     /**
146 |      * @brief Sets the parameter set of the distribution.
147 |      * @param __param The new parameter set of the distribution.
148 |      */
149 |     void param(const param_type& __param) { _M_param = __param; }
150 | 
151 |     /**
152 |      * @brief Returns the inclusive lower bound of the distribution range.
153 |      */
154 |     result_type min() const { return this->a(); }
155 | 
156 |     /**
157 |      * @brief Returns the inclusive upper bound of the distribution range.
158 |      */
159 |     result_type max() const { return this->b(); }
160 | 
161 |     /**
162 |      * @brief Generating functions.
163 |      */
164 |     template <typename _UniformRandomNumberGenerator>
165 |     result_type operator()(_UniformRandomNumberGenerator& __urng)
166 |     {
167 |         return this->operator()(__urng, _M_param);
168 |     }
169 | 
170 |     template <typename _UniformRandomNumberGenerator>
171 |     result_type operator()(_UniformRandomNumberGenerator& __urng, const param_type& __p)
172 |     {
173 |         double alpha = 1 / (1 - __p.theta());
174 |         double eta = (1 - std::pow(2.0 / (__p.b() - __p.a() + 1), 1 - __p.theta())) / (1 - __p.zeta2theta() / __p.zeta());
175 | 
176 |         double u = std::generate_canonical<double, std::numeric_limits<double>::digits, _UniformRandomNumberGenerator>(__urng);
177 | 
178 |         double uz = u * __p.zeta();
179 |         if (uz < 1.0)
180 |             return __p.a();
181 |         if (uz < 1.0 + std::pow(0.5, __p.theta()))
182 |             return __p.a() + 1;
183 | 
184 |         return __p.a() + ((__p.b() - __p.a() + 1) * std::pow(eta * u - eta + 1, alpha));
185 |     }
186 | 
187 |     /**
188 |      * @brief Return true if two zipfian int distributions have
189 |      *        the same parameters.
190 |      */
191 |     friend bool operator==(const zipfian_int_distribution& __d1, const zipfian_int_distribution& __d2)
192 |     {
193 |         return __d1._M_param == __d2._M_param;
194 |     }
195 | 
196 | private:
197 |     param_type _M_param;
198 | };


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
  1 |                     GNU GENERAL PUBLIC LICENSE
  2 |                        Version 3, 29 June 2007
  3 | 
  4 |  Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
  5 |  Everyone is permitted to copy and distribute verbatim copies
  6 |  of this license document, but changing it is not allowed.
  7 | 
  8 |                             Preamble
  9 | 
 10 |   The GNU General Public License is a free, copyleft license for
 11 | software and other kinds of works.
 12 | 
 13 |   The licenses for most software and other practical works are designed
 14 | to take away your freedom to share and change the works.  By contrast,
 15 | the GNU General Public License is intended to guarantee your freedom to
 16 | share and change all versions of a program--to make sure it remains free
 17 | software for all its users.  We, the Free Software Foundation, use the
 18 | GNU General Public License for most of our software; it applies also to
 19 | any other work released this way by its authors.  You can apply it to
 20 | your programs, too.
 21 | 
 22 |   When we speak of free software, we are referring to freedom, not
 23 | price.  Our General Public Licenses are designed to make sure that you
 24 | have the freedom to distribute copies of free software (and charge for
 25 | them if you wish), that you receive source code or can get it if you
 26 | want it, that you can change the software or use pieces of it in new
 27 | free programs, and that you know you can do these things.
 28 | 
 29 |   To protect your rights, we need to prevent others from denying you
 30 | these rights or asking you to surrender the rights.  Therefore, you have
 31 | certain responsibilities if you distribute copies of the software, or if
 32 | you modify it: responsibilities to respect the freedom of others.
 33 | 
 34 |   For example, if you distribute copies of such a program, whether
 35 | gratis or for a fee, you must pass on to the recipients the same
 36 | freedoms that you received.  You must make sure that they, too, receive
 37 | or can get the source code.  And you must show them these terms so they
 38 | know their rights.
 39 | 
 40 |   Developers that use the GNU GPL protect your rights with two steps:
 41 | (1) assert copyright on the software, and (2) offer you this License
 42 | giving you legal permission to copy, distribute and/or modify it.
 43 | 
 44 |   For the developers' and authors' protection, the GPL clearly explains
 45 | that there is no warranty for this free software.  For both users' and
 46 | authors' sake, the GPL requires that modified versions be marked as
 47 | changed, so that their problems will not be attributed erroneously to
 48 | authors of previous versions.
 49 | 
 50 |   Some devices are designed to deny users access to install or run
 51 | modified versions of the software inside them, although the manufacturer
 52 | can do so.  This is fundamentally incompatible with the aim of
 53 | protecting users' freedom to change the software.  The systematic
 54 | pattern of such abuse occurs in the area of products for individuals to
 55 | use, which is precisely where it is most unacceptable.  Therefore, we
 56 | have designed this version of the GPL to prohibit the practice for those
 57 | products.  If such problems arise substantially in other domains, we
 58 | stand ready to extend this provision to those domains in future versions
 59 | of the GPL, as needed to protect the freedom of users.
 60 | 
 61 |   Finally, every program is threatened constantly by software patents.
 62 | States should not allow patents to restrict development and use of
 63 | software on general-purpose computers, but in those that do, we wish to
 64 | avoid the special danger that patents applied to a free program could
 65 | make it effectively proprietary.  To prevent this, the GPL assures that
 66 | patents cannot be used to render the program non-free.
 67 | 
 68 |   The precise terms and conditions for copying, distribution and
 69 | modification follow.
 70 | 
 71 |                        TERMS AND CONDITIONS
 72 | 
 73 |   0. Definitions.
 74 | 
 75 |   "This License" refers to version 3 of the GNU General Public License.
 76 | 
 77 |   "Copyright" also means copyright-like laws that apply to other kinds of
 78 | works, such as semiconductor masks.
 79 | 
 80 |   "The Program" refers to any copyrightable work licensed under this
 81 | License.  Each licensee is addressed as "you".  "Licensees" and
 82 | "recipients" may be individuals or organizations.
 83 | 
 84 |   To "modify" a work means to copy from or adapt all or part of the work
 85 | in a fashion requiring copyright permission, other than the making of an
 86 | exact copy.  The resulting work is called a "modified version" of the
 87 | earlier work or a work "based on" the earlier work.
 88 | 
 89 |   A "covered work" means either the unmodified Program or a work based
 90 | on the Program.
 91 | 
 92 |   To "propagate" a work means to do anything with it that, without
 93 | permission, would make you directly or secondarily liable for
 94 | infringement under applicable copyright law, except executing it on a
 95 | computer or modifying a private copy.  Propagation includes copying,
 96 | distribution (with or without modification), making available to the
 97 | public, and in some countries other activities as well.
 98 | 
 99 |   To "convey" a work means any kind of propagation that enables other
100 | parties to make or receive copies.  Mere interaction with a user through
101 | a computer network, with no transfer of a copy, is not conveying.
102 | 
103 |   An interactive user interface displays "Appropriate Legal Notices"
104 | to the extent that it includes a convenient and prominently visible
105 | feature that (1) displays an appropriate copyright notice, and (2)
106 | tells the user that there is no warranty for the work (except to the
107 | extent that warranties are provided), that licensees may convey the
108 | work under this License, and how to view a copy of this License.  If
109 | the interface presents a list of user commands or options, such as a
110 | menu, a prominent item in the list meets this criterion.
111 | 
112 |   1. Source Code.
113 | 
114 |   The "source code" for a work means the preferred form of the work
115 | for making modifications to it.  "Object code" means any non-source
116 | form of a work.
117 | 
118 |   A "Standard Interface" means an interface that either is an official
119 | standard defined by a recognized standards body, or, in the case of
120 | interfaces specified for a particular programming language, one that
121 | is widely used among developers working in that language.
122 | 
123 |   The "System Libraries" of an executable work include anything, other
124 | than the work as a whole, that (a) is included in the normal form of
125 | packaging a Major Component, but which is not part of that Major
126 | Component, and (b) serves only to enable use of the work with that
127 | Major Component, or to implement a Standard Interface for which an
128 | implementation is available to the public in source code form.  A
129 | "Major Component", in this context, means a major essential component
130 | (kernel, window system, and so on) of the specific operating system
131 | (if any) on which the executable work runs, or a compiler used to
132 | produce the work, or an object code interpreter used to run it.
133 | 
134 |   The "Corresponding Source" for a work in object code form means all
135 | the source code needed to generate, install, and (for an executable
136 | work) run the object code and to modify the work, including scripts to
137 | control those activities.  However, it does not include the work's
138 | System Libraries, or general-purpose tools or generally available free
139 | programs which are used unmodified in performing those activities but
140 | which are not part of the work.  For example, Corresponding Source
141 | includes interface definition files associated with source files for
142 | the work, and the source code for shared libraries and dynamically
143 | linked subprograms that the work is specifically designed to require,
144 | such as by intimate data communication or control flow between those
145 | subprograms and other parts of the work.
146 | 
147 |   The Corresponding Source need not include anything that users
148 | can regenerate automatically from other parts of the Corresponding
149 | Source.
150 | 
151 |   The Corresponding Source for a work in source code form is that
152 | same work.
153 | 
154 |   2. Basic Permissions.
155 | 
156 |   All rights granted under this License are granted for the term of
157 | copyright on the Program, and are irrevocable provided the stated
158 | conditions are met.  This License explicitly affirms your unlimited
159 | permission to run the unmodified Program.  The output from running a
160 | covered work is covered by this License only if the output, given its
161 | content, constitutes a covered work.  This License acknowledges your
162 | rights of fair use or other equivalent, as provided by copyright law.
163 | 
164 |   You may make, run and propagate covered works that you do not
165 | convey, without conditions so long as your license otherwise remains
166 | in force.  You may convey covered works to others for the sole purpose
167 | of having them make modifications exclusively for you, or provide you
168 | with facilities for running those works, provided that you comply with
169 | the terms of this License in conveying all material for which you do
170 | not control copyright.  Those thus making or running the covered works
171 | for you must do so exclusively on your behalf, under your direction
172 | and control, on terms that prohibit them from making any copies of
173 | your copyrighted material outside their relationship with you.
174 | 
175 |   Conveying under any other circumstances is permitted solely under
176 | the conditions stated below.  Sublicensing is not allowed; section 10
177 | makes it unnecessary.
178 | 
179 |   3. Protecting Users' Legal Rights From Anti-Circumvention Law.
180 | 
181 |   No covered work shall be deemed part of an effective technological
182 | measure under any applicable law fulfilling obligations under article
183 | 11 of the WIPO copyright treaty adopted on 20 December 1996, or
184 | similar laws prohibiting or restricting circumvention of such
185 | measures.
186 | 
187 |   When you convey a covered work, you waive any legal power to forbid
188 | circumvention of technological measures to the extent such circumvention
189 | is effected by exercising rights under this License with respect to
190 | the covered work, and you disclaim any intention to limit operation or
191 | modification of the work as a means of enforcing, against the work's
192 | users, your or third parties' legal rights to forbid circumvention of
193 | technological measures.
194 | 
195 |   4. Conveying Verbatim Copies.
196 | 
197 |   You may convey verbatim copies of the Program's source code as you
198 | receive it, in any medium, provided that you conspicuously and
199 | appropriately publish on each copy an appropriate copyright notice;
200 | keep intact all notices stating that this License and any
201 | non-permissive terms added in accord with section 7 apply to the code;
202 | keep intact all notices of the absence of any warranty; and give all
203 | recipients a copy of this License along with the Program.
204 | 
205 |   You may charge any price or no price for each copy that you convey,
206 | and you may offer support or warranty protection for a fee.
207 | 
208 |   5. Conveying Modified Source Versions.
209 | 
210 |   You may convey a work based on the Program, or the modifications to
211 | produce it from the Program, in the form of source code under the
212 | terms of section 4, provided that you also meet all of these conditions:
213 | 
214 |     a) The work must carry prominent notices stating that you modified
215 |     it, and giving a relevant date.
216 | 
217 |     b) The work must carry prominent notices stating that it is
218 |     released under this License and any conditions added under section
219 |     7.  This requirement modifies the requirement in section 4 to
220 |     "keep intact all notices".
221 | 
222 |     c) You must license the entire work, as a whole, under this
223 |     License to anyone who comes into possession of a copy.  This
224 |     License will therefore apply, along with any applicable section 7
225 |     additional terms, to the whole of the work, and all its parts,
226 |     regardless of how they are packaged.  This License gives no
227 |     permission to license the work in any other way, but it does not
228 |     invalidate such permission if you have separately received it.
229 | 
230 |     d) If the work has interactive user interfaces, each must display
231 |     Appropriate Legal Notices; however, if the Program has interactive
232 |     interfaces that do not display Appropriate Legal Notices, your
233 |     work need not make them do so.
234 | 
235 |   A compilation of a covered work with other separate and independent
236 | works, which are not by their nature extensions of the covered work,
237 | and which are not combined with it such as to form a larger program,
238 | in or on a volume of a storage or distribution medium, is called an
239 | "aggregate" if the compilation and its resulting copyright are not
240 | used to limit the access or legal rights of the compilation's users
241 | beyond what the individual works permit.  Inclusion of a covered work
242 | in an aggregate does not cause this License to apply to the other
243 | parts of the aggregate.
244 | 
245 |   6. Conveying Non-Source Forms.
246 | 
247 |   You may convey a covered work in object code form under the terms
248 | of sections 4 and 5, provided that you also convey the
249 | machine-readable Corresponding Source under the terms of this License,
250 | in one of these ways:
251 | 
252 |     a) Convey the object code in, or embodied in, a physical product
253 |     (including a physical distribution medium), accompanied by the
254 |     Corresponding Source fixed on a durable physical medium
255 |     customarily used for software interchange.
256 | 
257 |     b) Convey the object code in, or embodied in, a physical product
258 |     (including a physical distribution medium), accompanied by a
259 |     written offer, valid for at least three years and valid for as
260 |     long as you offer spare parts or customer support for that product
261 |     model, to give anyone who possesses the object code either (1) a
262 |     copy of the Corresponding Source for all the software in the
263 |     product that is covered by this License, on a durable physical
264 |     medium customarily used for software interchange, for a price no
265 |     more than your reasonable cost of physically performing this
266 |     conveying of source, or (2) access to copy the
267 |     Corresponding Source from a network server at no charge.
268 | 
269 |     c) Convey individual copies of the object code with a copy of the
270 |     written offer to provide the Corresponding Source.  This
271 |     alternative is allowed only occasionally and noncommercially, and
272 |     only if you received the object code with such an offer, in accord
273 |     with subsection 6b.
274 | 
275 |     d) Convey the object code by offering access from a designated
276 |     place (gratis or for a charge), and offer equivalent access to the
277 |     Corresponding Source in the same way through the same place at no
278 |     further charge.  You need not require recipients to copy the
279 |     Corresponding Source along with the object code.  If the place to
280 |     copy the object code is a network server, the Corresponding Source
281 |     may be on a different server (operated by you or a third party)
282 |     that supports equivalent copying facilities, provided you maintain
283 |     clear directions next to the object code saying where to find the
284 |     Corresponding Source.  Regardless of what server hosts the
285 |     Corresponding Source, you remain obligated to ensure that it is
286 |     available for as long as needed to satisfy these requirements.
287 | 
288 |     e) Convey the object code using peer-to-peer transmission, provided
289 |     you inform other peers where the object code and Corresponding
290 |     Source of the work are being offered to the general public at no
291 |     charge under subsection 6d.
292 | 
293 |   A separable portion of the object code, whose source code is excluded
294 | from the Corresponding Source as a System Library, need not be
295 | included in conveying the object code work.
296 | 
297 |   A "User Product" is either (1) a "consumer product", which means any
298 | tangible personal property which is normally used for personal, family,
299 | or household purposes, or (2) anything designed or sold for incorporation
300 | into a dwelling.  In determining whether a product is a consumer product,
301 | doubtful cases shall be resolved in favor of coverage.  For a particular
302 | product received by a particular user, "normally used" refers to a
303 | typical or common use of that class of product, regardless of the status
304 | of the particular user or of the way in which the particular user
305 | actually uses, or expects or is expected to use, the product.  A product
306 | is a consumer product regardless of whether the product has substantial
307 | commercial, industrial or non-consumer uses, unless such uses represent
308 | the only significant mode of use of the product.
309 | 
310 |   "Installation Information" for a User Product means any methods,
311 | procedures, authorization keys, or other information required to install
312 | and execute modified versions of a covered work in that User Product from
313 | a modified version of its Corresponding Source.  The information must
314 | suffice to ensure that the continued functioning of the modified object
315 | code is in no case prevented or interfered with solely because
316 | modification has been made.
317 | 
318 |   If you convey an object code work under this section in, or with, or
319 | specifically for use in, a User Product, and the conveying occurs as
320 | part of a transaction in which the right of possession and use of the
321 | User Product is transferred to the recipient in perpetuity or for a
322 | fixed term (regardless of how the transaction is characterized), the
323 | Corresponding Source conveyed under this section must be accompanied
324 | by the Installation Information.  But this requirement does not apply
325 | if neither you nor any third party retains the ability to install
326 | modified object code on the User Product (for example, the work has
327 | been installed in ROM).
328 | 
329 |   The requirement to provide Installation Information does not include a
330 | requirement to continue to provide support service, warranty, or updates
331 | for a work that has been modified or installed by the recipient, or for
332 | the User Product in which it has been modified or installed.  Access to a
333 | network may be denied when the modification itself materially and
334 | adversely affects the operation of the network or violates the rules and
335 | protocols for communication across the network.
336 | 
337 |   Corresponding Source conveyed, and Installation Information provided,
338 | in accord with this section must be in a format that is publicly
339 | documented (and with an implementation available to the public in
340 | source code form), and must require no special password or key for
341 | unpacking, reading or copying.
342 | 
343 |   7. Additional Terms.
344 | 
345 |   "Additional permissions" are terms that supplement the terms of this
346 | License by making exceptions from one or more of its conditions.
347 | Additional permissions that are applicable to the entire Program shall
348 | be treated as though they were included in this License, to the extent
349 | that they are valid under applicable law.  If additional permissions
350 | apply only to part of the Program, that part may be used separately
351 | under those permissions, but the entire Program remains governed by
352 | this License without regard to the additional permissions.
353 | 
354 |   When you convey a copy of a covered work, you may at your option
355 | remove any additional permissions from that copy, or from any part of
356 | it.  (Additional permissions may be written to require their own
357 | removal in certain cases when you modify the work.)  You may place
358 | additional permissions on material, added by you to a covered work,
359 | for which you have or can give appropriate copyright permission.
360 | 
361 |   Notwithstanding any other provision of this License, for material you
362 | add to a covered work, you may (if authorized by the copyright holders of
363 | that material) supplement the terms of this License with terms:
364 | 
365 |     a) Disclaiming warranty or limiting liability differently from the
366 |     terms of sections 15 and 16 of this License; or
367 | 
368 |     b) Requiring preservation of specified reasonable legal notices or
369 |     author attributions in that material or in the Appropriate Legal
370 |     Notices displayed by works containing it; or
371 | 
372 |     c) Prohibiting misrepresentation of the origin of that material, or
373 |     requiring that modified versions of such material be marked in
374 |     reasonable ways as different from the original version; or
375 | 
376 |     d) Limiting the use for publicity purposes of names of licensors or
377 |     authors of the material; or
378 | 
379 |     e) Declining to grant rights under trademark law for use of some
380 |     trade names, trademarks, or service marks; or
381 | 
382 |     f) Requiring indemnification of licensors and authors of that
383 |     material by anyone who conveys the material (or modified versions of
384 |     it) with contractual assumptions of liability to the recipient, for
385 |     any liability that these contractual assumptions directly impose on
386 |     those licensors and authors.
387 | 
388 |   All other non-permissive additional terms are considered "further
389 | restrictions" within the meaning of section 10.  If the Program as you
390 | received it, or any part of it, contains a notice stating that it is
391 | governed by this License along with a term that is a further
392 | restriction, you may remove that term.  If a license document contains
393 | a further restriction but permits relicensing or conveying under this
394 | License, you may add to a covered work material governed by the terms
395 | of that license document, provided that the further restriction does
396 | not survive such relicensing or conveying.
397 | 
398 |   If you add terms to a covered work in accord with this section, you
399 | must place, in the relevant source files, a statement of the
400 | additional terms that apply to those files, or a notice indicating
401 | where to find the applicable terms.
402 | 
403 |   Additional terms, permissive or non-permissive, may be stated in the
404 | form of a separately written license, or stated as exceptions;
405 | the above requirements apply either way.
406 | 
407 |   8. Termination.
408 | 
409 |   You may not propagate or modify a covered work except as expressly
410 | provided under this License.  Any attempt otherwise to propagate or
411 | modify it is void, and will automatically terminate your rights under
412 | this License (including any patent licenses granted under the third
413 | paragraph of section 11).
414 | 
415 |   However, if you cease all violation of this License, then your
416 | license from a particular copyright holder is reinstated (a)
417 | provisionally, unless and until the copyright holder explicitly and
418 | finally terminates your license, and (b) permanently, if the copyright
419 | holder fails to notify you of the violation by some reasonable means
420 | prior to 60 days after the cessation.
421 | 
422 |   Moreover, your license from a particular copyright holder is
423 | reinstated permanently if the copyright holder notifies you of the
424 | violation by some reasonable means, this is the first time you have
425 | received notice of violation of this License (for any work) from that
426 | copyright holder, and you cure the violation prior to 30 days after
427 | your receipt of the notice.
428 | 
429 |   Termination of your rights under this section does not terminate the
430 | licenses of parties who have received copies or rights from you under
431 | this License.  If your rights have been terminated and not permanently
432 | reinstated, you do not qualify to receive new licenses for the same
433 | material under section 10.
434 | 
435 |   9. Acceptance Not Required for Having Copies.
436 | 
437 |   You are not required to accept this License in order to receive or
438 | run a copy of the Program.  Ancillary propagation of a covered work
439 | occurring solely as a consequence of using peer-to-peer transmission
440 | to receive a copy likewise does not require acceptance.  However,
441 | nothing other than this License grants you permission to propagate or
442 | modify any covered work.  These actions infringe copyright if you do
443 | not accept this License.  Therefore, by modifying or propagating a
444 | covered work, you indicate your acceptance of this License to do so.
445 | 
446 |   10. Automatic Licensing of Downstream Recipients.
447 | 
448 |   Each time you convey a covered work, the recipient automatically
449 | receives a license from the original licensors, to run, modify and
450 | propagate that work, subject to this License.  You are not responsible
451 | for enforcing compliance by third parties with this License.
452 | 
453 |   An "entity transaction" is a transaction transferring control of an
454 | organization, or substantially all assets of one, or subdividing an
455 | organization, or merging organizations.  If propagation of a covered
456 | work results from an entity transaction, each party to that
457 | transaction who receives a copy of the work also receives whatever
458 | licenses to the work the party's predecessor in interest had or could
459 | give under the previous paragraph, plus a right to possession of the
460 | Corresponding Source of the work from the predecessor in interest, if
461 | the predecessor has it or can get it with reasonable efforts.
462 | 
463 |   You may not impose any further restrictions on the exercise of the
464 | rights granted or affirmed under this License.  For example, you may
465 | not impose a license fee, royalty, or other charge for exercise of
466 | rights granted under this License, and you may not initiate litigation
467 | (including a cross-claim or counterclaim in a lawsuit) alleging that
468 | any patent claim is infringed by making, using, selling, offering for
469 | sale, or importing the Program or any portion of it.
470 | 
471 |   11. Patents.
472 | 
473 |   A "contributor" is a copyright holder who authorizes use under this
474 | License of the Program or a work on which the Program is based.  The
475 | work thus licensed is called the contributor's "contributor version".
476 | 
477 |   A contributor's "essential patent claims" are all patent claims
478 | owned or controlled by the contributor, whether already acquired or
479 | hereafter acquired, that would be infringed by some manner, permitted
480 | by this License, of making, using, or selling its contributor version,
481 | but do not include claims that would be infringed only as a
482 | consequence of further modification of the contributor version.  For
483 | purposes of this definition, "control" includes the right to grant
484 | patent sublicenses in a manner consistent with the requirements of
485 | this License.
486 | 
487 |   Each contributor grants you a non-exclusive, worldwide, royalty-free
488 | patent license under the contributor's essential patent claims, to
489 | make, use, sell, offer for sale, import and otherwise run, modify and
490 | propagate the contents of its contributor version.
491 | 
492 |   In the following three paragraphs, a "patent license" is any express
493 | agreement or commitment, however denominated, not to enforce a patent
494 | (such as an express permission to practice a patent or covenant not to
495 | sue for patent infringement).  To "grant" such a patent license to a
496 | party means to make such an agreement or commitment not to enforce a
497 | patent against the party.
498 | 
499 |   If you convey a covered work, knowingly relying on a patent license,
500 | and the Corresponding Source of the work is not available for anyone
501 | to copy, free of charge and under the terms of this License, through a
502 | publicly available network server or other readily accessible means,
503 | then you must either (1) cause the Corresponding Source to be so
504 | available, or (2) arrange to deprive yourself of the benefit of the
505 | patent license for this particular work, or (3) arrange, in a manner
506 | consistent with the requirements of this License, to extend the patent
507 | license to downstream recipients.  "Knowingly relying" means you have
508 | actual knowledge that, but for the patent license, your conveying the
509 | covered work in a country, or your recipient's use of the covered work
510 | in a country, would infringe one or more identifiable patents in that
511 | country that you have reason to believe are valid.
512 | 
513 |   If, pursuant to or in connection with a single transaction or
514 | arrangement, you convey, or propagate by procuring conveyance of, a
515 | covered work, and grant a patent license to some of the parties
516 | receiving the covered work authorizing them to use, propagate, modify
517 | or convey a specific copy of the covered work, then the patent license
518 | you grant is automatically extended to all recipients of the covered
519 | work and works based on it.
520 | 
521 |   A patent license is "discriminatory" if it does not include within
522 | the scope of its coverage, prohibits the exercise of, or is
523 | conditioned on the non-exercise of one or more of the rights that are
524 | specifically granted under this License.  You may not convey a covered
525 | work if you are a party to an arrangement with a third party that is
526 | in the business of distributing software, under which you make payment
527 | to the third party based on the extent of your activity of conveying
528 | the work, and under which the third party grants, to any of the
529 | parties who would receive the covered work from you, a discriminatory
530 | patent license (a) in connection with copies of the covered work
531 | conveyed by you (or copies made from those copies), or (b) primarily
532 | for and in connection with specific products or compilations that
533 | contain the covered work, unless you entered into that arrangement,
534 | or that patent license was granted, prior to 28 March 2007.
535 | 
536 |   Nothing in this License shall be construed as excluding or limiting
537 | any implied license or other defenses to infringement that may
538 | otherwise be available to you under applicable patent law.
539 | 
540 |   12. No Surrender of Others' Freedom.
541 | 
542 |   If conditions are imposed on you (whether by court order, agreement or
543 | otherwise) that contradict the conditions of this License, they do not
544 | excuse you from the conditions of this License.  If you cannot convey a
545 | covered work so as to satisfy simultaneously your obligations under this
546 | License and any other pertinent obligations, then as a consequence you may
547 | not convey it at all.  For example, if you agree to terms that obligate you
548 | to collect a royalty for further conveying from those to whom you convey
549 | the Program, the only way you could satisfy both those terms and this
550 | License would be to refrain entirely from conveying the Program.
551 | 
552 |   13. Use with the GNU Affero General Public License.
553 | 
554 |   Notwithstanding any other provision of this License, you have
555 | permission to link or combine any covered work with a work licensed
556 | under version 3 of the GNU Affero General Public License into a single
557 | combined work, and to convey the resulting work.  The terms of this
558 | License will continue to apply to the part which is the covered work,
559 | but the special requirements of the GNU Affero General Public License,
560 | section 13, concerning interaction through a network will apply to the
561 | combination as such.
562 | 
563 |   14. Revised Versions of this License.
564 | 
565 |   The Free Software Foundation may publish revised and/or new versions of
566 | the GNU General Public License from time to time.  Such new versions will
567 | be similar in spirit to the present version, but may differ in detail to
568 | address new problems or concerns.
569 | 
570 |   Each version is given a distinguishing version number.  If the
571 | Program specifies that a certain numbered version of the GNU General
572 | Public License "or any later version" applies to it, you have the
573 | option of following the terms and conditions either of that numbered
574 | version or of any later version published by the Free Software
575 | Foundation.  If the Program does not specify a version number of the
576 | GNU General Public License, you may choose any version ever published
577 | by the Free Software Foundation.
578 | 
579 |   If the Program specifies that a proxy can decide which future
580 | versions of the GNU General Public License can be used, that proxy's
581 | public statement of acceptance of a version permanently authorizes you
582 | to choose that version for the Program.
583 | 
584 |   Later license versions may give you additional or different
585 | permissions.  However, no additional obligations are imposed on any
586 | author or copyright holder as a result of your choosing to follow a
587 | later version.
588 | 
589 |   15. Disclaimer of Warranty.
590 | 
591 |   THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
592 | APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
593 | HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
594 | OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
595 | THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
596 | PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
597 | IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
598 | ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
599 | 
600 |   16. Limitation of Liability.
601 | 
602 |   IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
603 | WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
604 | THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
605 | GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
606 | USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
607 | DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
608 | PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
609 | EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
610 | SUCH DAMAGES.
611 | 
612 |   17. Interpretation of Sections 15 and 16.
613 | 
614 |   If the disclaimer of warranty and limitation of liability provided
615 | above cannot be given local legal effect according to their terms,
616 | reviewing courts shall apply local law that most closely approximates
617 | an absolute waiver of all civil liability in connection with the
618 | Program, unless a warranty or assumption of liability accompanies a
619 | copy of the Program in return for a fee.
620 | 
621 |                      END OF TERMS AND CONDITIONS
622 | 
623 |             How to Apply These Terms to Your New Programs
624 | 
625 |   If you develop a new program, and you want it to be of the greatest
626 | possible use to the public, the best way to achieve this is to make it
627 | free software which everyone can redistribute and change under these terms.
628 | 
629 |   To do so, attach the following notices to the program.  It is safest
630 | to attach them to the start of each source file to most effectively
631 | state the exclusion of warranty; and each file should have at least
632 | the "copyright" line and a pointer to where the full notice is found.
633 | 
634 |     <one line to give the program's name and a brief idea of what it does.>
635 |     Copyright (C) <year>  <name of author>
636 | 
637 |     This program is free software: you can redistribute it and/or modify
638 |     it under the terms of the GNU General Public License as published by
639 |     the Free Software Foundation, either version 3 of the License, or
640 |     (at your option) any later version.
641 | 
642 |     This program is distributed in the hope that it will be useful,
643 |     but WITHOUT ANY WARRANTY; without even the implied warranty of
644 |     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
645 |     GNU General Public License for more details.
646 | 
647 |     You should have received a copy of the GNU General Public License
648 |     along with this program.  If not, see <https://www.gnu.org/licenses/>.
649 | 
650 | Also add information on how to contact you by electronic and paper mail.
651 | 
652 |   If the program does terminal interaction, make it output a short
653 | notice like this when it starts in an interactive mode:
654 | 
655 |     <program>  Copyright (C) <year>  <name of author>
656 |     This program comes with ABSOLUTELY NO WARRANTY; for details type `show w'.
657 |     This is free software, and you are welcome to redistribute it
658 |     under certain conditions; type `show c' for details.
659 | 
660 | The hypothetical commands `show w' and `show c' should show the appropriate
661 | parts of the General Public License.  Of course, your program's commands
662 | might be different; for a GUI interface, you would use an "about box".
663 | 
664 |   You should also get your employer (if you work as a programmer) or school,
665 | if any, to sign a "copyright disclaimer" for the program, if necessary.
666 | For more information on this, and how to apply and follow the GNU GPL, see
667 | <https://www.gnu.org/licenses/>.
668 | 
669 |   The GNU General Public License does not permit incorporating your program
670 | into proprietary programs.  If your program is a subroutine library, you
671 | may consider it more useful to permit linking proprietary applications with
672 | the library.  If this is what you want to do, use the GNU Lesser General
673 | Public License instead of this License.  But first, please read
674 | <https://www.gnu.org/licenses/why-not-lgpl.html>.
675 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | Implementation of the paper "TLBtree: A Read/Write-Optimized Tree Index for Non-Volatile Memory (ICDE'21)" and paper "Two Birds with One Stone: Boosting Both Search and Write Performance for Tree Indices on Persistent Memory (TECS'21)".
 2 | 
 3 | TLBtree is a read/write-optimized persistent index for NVM-only Memory System. It is composed of a read-optimized top layer and a NVM-friendly write-optimized down layer. The top layer is read frequently, so we adpot a less-mutable and cache-friendly design; The down layer is write frequently and should be carefully guarded by persistent instructions (e.g. clwb and sfence), so we employ a structure with low persistent cost on real NVM enviroment.
 4 | 
 5 | Our idea is motivated by the observation that B+-tree-like range indices tend to absorb 99% of writes in the bottom 2-3 levels, while reads are evenly distributed in each level. According to the observation, we propose a Two Layer Persistent Index (TLPI) achitecture first. TLPI divides a pesistent tree index into two layers, so it enables to put together specific optimizations which may be too complicated to coexist in an individual index. TLBtree is a tailored instance of TLPI, you can also add new read optimizations into the top layer and write optimizations to the down layer, for further performance potential.
 6 | 
 7 | 
 8 | #### Dependence
 9 | We test our project on ubuntu LTS 2020. 
10 | 1. This project depends on libpmemobj from [PMDK](https://pmem.io/pmdk/libpmemobj/). Install it using command
11 |     ```shell
12 |     sudo apt-get install make
13 |     sudo apt-get install libpmemobj-dev
14 |     ``` 
15 | 2. Make sure you are avaliable with Optane memory (or you can use volatile memory to simulate pmem device on latest ubuntu version). The project assumes that your pmem device is mounted at address `/mnt/pmem` and you have write permission to it.
16 | 
17 |     (a). Avaiable to real Optane DC Memory, configure it in this [way](https://software.intel.com/content/www/us/en/develop/articles/qsg-part2-linux-provisioning-with-optane-pmem.html).
18 |     
19 |     (b). Not avaiable to Optane, try simluate it with DRAM following this [link](https://software.intel.com/content/www/us/en/develop/articles/how-to-emulate-persistent-memory-on-an-intel-architecture-server.html).
20 | 
21 | 
22 | #### Usage
23 | 1. Configure your PMEM file address and file size threshold in *include/tlbtree.h*
24 | 2. Compile the program with following commands (the same in Single or Concurrent)
25 |     ```sh
26 |     mkdir build
27 |     cd build; cmake ..
28 |     make
29 |     ```
30 | 3. Play with TLBtree using provided test modules:
31 |     
32 |     (a). generate data with `datagen` (type `datagen -h` if needed)
33 | 
34 |     (b). populate the TLBtree with some inital key value pairs
35 | 
36 |     (c). doing CUID operations with `main`
37 | 
38 | #### Limitations
39 | Currently TLBtree supports only 8-byte integer key and payload
40 | 


--------------------------------------------------------------------------------
/Single/.gitignore:
--------------------------------------------------------------------------------
1 | build
2 | .vscode
3 | 


--------------------------------------------------------------------------------
/Single/CMakeLists.txt:
--------------------------------------------------------------------------------
 1 | project(TLBtree)
 2 | 
 3 | cmake_minimum_required(VERSION 3.16)
 4 | 
 5 | add_compile_options(-mclwb -fmax-errors=5)
 6 | add_compile_options(-O3)
 7 | link_libraries(/usr/lib/x86_64-linux-gnu/libpmemobj.so)
 8 | add_link_options(-pthread)
 9 | 
10 | include_directories(include)
11 | 
12 | add_subdirectory(src)
13 | add_subdirectory(test)


--------------------------------------------------------------------------------
/Single/include/common.h:
--------------------------------------------------------------------------------
 1 | /*
 2 |     Copyright (c) Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
 3 |     USE AT YOUR OWN RISK!
 4 | */
 5 | #ifndef __COMMON_H__
 6 | #define __COMMON_H__
 7 | 
 8 | #include <cstdint>
 9 | #include <limits>
10 | #include <sys/stat.h>
11 | 
12 | #define LOADSCALE 8
13 | #define DOFLUSH
14 | 
15 | #define KILO 1024
16 | #define MILLION (KILO * KILO)
17 | #define CACHE_LINE_SIZE 64
18 | 
19 | #ifndef KEYTYPE
20 |     using _key_t = int64_t;
21 | #else
22 |     using _key_t = KEYTYPE;
23 | #endif
24 | 
25 | const _key_t MAX_KEY = std::numeric_limits<_key_t>::max();
26 | const _key_t MIN_KEY = typeid(_key_t) == typeid(double) || typeid(_key_t) == typeid(float) 
27 |                             ? -1 * MAX_KEY : std::numeric_limits<_key_t>::min();
28 | 
29 | struct Record {
30 |     _key_t key;
31 |     char * val; 
32 |     Record(_key_t k=MAX_KEY, char * v=NULL) : key(k), val(v) {}
33 |     bool operator < (const Record & other) {
34 |         return key < other.key;
35 |     }
36 | };
37 | 
38 | enum OperationType {READ = 0, INSERT, UPDATE, DELETE};
39 | 
40 | struct res_t { // a result type use to pass info when split and search
41 |     bool flag; 
42 |     Record rec;
43 |     res_t(bool f, Record e):flag(f), rec(e) {}
44 | };
45 | 
46 | #include <sys/time.h>
47 | inline double seconds()
48 | {
49 |   timeval now;
50 |   gettimeofday(&now, NULL);
51 |   return now.tv_sec + now.tv_usec/1000000.0;
52 | }
53 | 
54 | inline int getRandom() {
55 |     timeval now;
56 |     gettimeofday(&now, NULL);
57 |     return now.tv_usec;
58 | }
59 | 
60 | inline bool file_exist(const char *pool_path) {
61 |     struct stat buffer;
62 |     return (stat(pool_path, &buffer) == 0);
63 | }
64 | 
65 | #endif //__COMMON_H__


--------------------------------------------------------------------------------
/Single/include/tlbtree.h:
--------------------------------------------------------------------------------
 1 | #ifndef __TLBTREE_H__
 2 | #define __TLBTREE_H__
 3 | 
 4 | #include "../src/tlbtree_impl.h"
 5 | 
 6 | using tlbtree::TLBtreeImpl;
 7 | 
 8 | // configure the PMEM file and file size
 9 | static constexpr uint64_t POOL_SIZE = 512UL * 1024 * 1024;
10 | 
11 | class TLBtree {
12 | public:
13 |     TLBtree(std::string tlbname, uint64_t poolsize = POOL_SIZE) {
14 |         bool recover = file_exist(tlbname.c_str());
15 |         tree_ = new TLBtreeImpl<2,2>(tlbname, recover, poolsize);
16 |     }
17 | 
18 |     ~TLBtree() {
19 |         delete tree_;
20 |     }
21 | 
22 |     inline void insert(_key_t key, uint64_t val) {
23 |         tree_->insert(key, val);
24 |     }
25 | 
26 |     inline bool update(_key_t key, uint64_t val) {
27 |         return tree_->update(key, val);
28 |     }
29 | 
30 |     inline uint64_t lookup(_key_t key) {
31 |         uint64_t val;
32 |         bool found = tree_->find(key, val);
33 | 
34 |         if(found)
35 |             return val;
36 |         else 
37 |             return 0;
38 |     }
39 | 
40 |     inline bool remove(_key_t key) {
41 |         return tree_->remove(key);
42 |     }
43 | 
44 | private:
45 |     TLBtreeImpl <2,2> * tree_;
46 | };
47 | 
48 | #endif //__TLBTREE_H__


--------------------------------------------------------------------------------
/Single/src/CMakeLists.txt:
--------------------------------------------------------------------------------
1 | add_library(tlbtree tlbtree_impl.cc wotree256.cc)


--------------------------------------------------------------------------------
/Single/src/fixtree.h:
--------------------------------------------------------------------------------
  1 | /*  fixtree.h - A search-optimized fixed tree
  2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  3 |     USE AT YOUR OWN RISK!
  4 | */
  5 | 
  6 | #ifndef __FIXTREE__
  7 | #define __FIXTREE__
  8 | 
  9 | #include <cstdio>
 10 | #include <cstdint>
 11 | #include <cmath>
 12 | #include <vector>
 13 | #include <cassert>
 14 | #include <cstring>
 15 | 
 16 | #include "flush.h"
 17 | #include "pmallocator.h"
 18 | 
 19 | namespace fixtree {
 20 |     const int INNER_CARD = 32; // node size: 256B, the fanout of inner node is 32
 21 |     const int LEAF_CARD = 16;  // node size: 256B, the fanout of leaf node is 16
 22 |     const int LEAF_REBUILD_CARD = 8;
 23 |     const int MAX_HEIGHT = 10;
 24 | 
 25 |     // the entrance of fixtree that stores its persistent tree metadata
 26 |     struct entrance_t {
 27 |         void * leaf_buff;
 28 |         void * inner_buff;
 29 |         uint32_t height; // the tree height
 30 |         uint32_t leaf_cnt; 
 31 |     };
 32 | 
 33 | /*  Fixtree: 
 34 |         a search-optimized linearize tree structure which can absort moderate insertions: 
 35 | */
 36 | class Fixtree {
 37 |     public:
 38 |         struct INNode { // inner node is packed keys, which is very compact
 39 |             _key_t keys[INNER_CARD];
 40 |         } __attribute__((aligned(CACHE_LINE_SIZE)));
 41 | 
 42 |         struct LFNode { // leaf node is packed key-ptr along with a header. 
 43 |                         // leaf node has some gap to absort insert
 44 |             _key_t keys[LEAF_CARD];
 45 |             char * vals[LEAF_CARD];
 46 |         } __attribute__((aligned(CACHE_LINE_SIZE)));
 47 | 
 48 |     public:
 49 |         // volatile structures
 50 |         INNode * inner_nodes_;
 51 |         LFNode * leaf_nodes_;
 52 |         uint32_t height_;
 53 |         uint32_t leaf_cnt_;
 54 |         entrance_t * entrance_;
 55 |         uint32_t level_offset_[MAX_HEIGHT];
 56 |     
 57 |     public:
 58 |         Fixtree(entrance_t * ent) { // recovery the tree from the entrance
 59 |             inner_nodes_ = (INNode *)galc->absolute(ent->inner_buff);
 60 |             leaf_nodes_ = (LFNode *)galc->absolute(ent->leaf_buff);
 61 |             height_ = ent->height;
 62 |             leaf_cnt_ = ent->leaf_cnt;
 63 |             entrance_ = ent;
 64 | 
 65 |             uint32_t tmp = 0;
 66 |             for(int l = 0; l < height_; l++) {
 67 |                 level_offset_[l] = tmp;
 68 |                 tmp += std::pow(INNER_CARD, l);
 69 |             }
 70 |             level_offset_[height_] = tmp;
 71 |         }
 72 | 
 73 |         Fixtree(std::vector<Record> records) {
 74 |             const int lfary = LEAF_REBUILD_CARD;
 75 |             int record_count = records.size();
 76 |             
 77 |             uint32_t lfnode_cnt = std::ceil((float)record_count / lfary);
 78 |             leaf_nodes_ = (LFNode *) galc->malloc(std::max((size_t)4096, lfnode_cnt * sizeof(LFNode)));
 79 | 
 80 |             height_ = std::ceil(std::log(std::max((uint32_t)INNER_CARD, lfnode_cnt)) / std::log(INNER_CARD));
 81 |             uint32_t innode_cnt = (std::pow(INNER_CARD, height_) - 1) / (INNER_CARD - 1);
 82 |             inner_nodes_ = (INNode *) galc->malloc(std::max((size_t)4096, innode_cnt * sizeof(INNode)));
 83 | 
 84 |             // fill leaf nodes
 85 |             for(int i = 0; i < lfnode_cnt; i++) {
 86 |                 for(int j = 0; j < lfary; j++) {
 87 |                     auto idx = i * lfary + j;
 88 |                     leaf_nodes_[i].keys[j] = idx < record_count ? records[idx].key : MAX_KEY; 
 89 |                     leaf_nodes_[i].vals[j] = idx < record_count ? records[idx].val : 0;
 90 |                 }
 91 |                 for(int j = lfary; j < LEAF_CARD; j++) { // intialized key
 92 |                     leaf_nodes_[i].keys[j] = MAX_KEY;
 93 |                 }
 94 |                 clwb(leaf_nodes_ + i, sizeof(LFNode));
 95 |             }
 96 |             
 97 |             int cur_level_cnt = lfnode_cnt;
 98 |             int cur_level_off = innode_cnt - std::pow(INNER_CARD, height_ - 1);
 99 |             int last_level_off = 0;
100 |             
101 |             // fill parent innodes of leaf nodes
102 |             for(int i = 0; i < cur_level_cnt; i++)
103 |                 inner_insert(cur_level_off + i / INNER_CARD, i % INNER_CARD, leaf_nodes_[i].keys[0]);
104 |             if(cur_level_cnt % INNER_CARD)
105 |                 inner_insert(cur_level_off + cur_level_cnt / INNER_CARD, cur_level_cnt % INNER_CARD, MAX_KEY);
106 |             clwb(&inner_nodes_[cur_level_off], sizeof(INNode) * (cur_level_cnt / INNER_CARD + 1));
107 |             
108 |             cur_level_cnt = std::ceil((float)cur_level_cnt / INNER_CARD);
109 |             last_level_off = cur_level_off;
110 |             cur_level_off = cur_level_off - std::pow(INNER_CARD, height_ - 2);
111 | 
112 |             // fill other inner nodes
113 |             for(int l = height_ - 2; l >= 0; l--) { // level by level
114 |                 for(int i = 0; i < cur_level_cnt; i++)
115 |                     inner_insert(cur_level_off + i / INNER_CARD, i % INNER_CARD, inner_nodes_[last_level_off + i].keys[0]);
116 |                 if(cur_level_cnt % INNER_CARD)
117 |                     inner_insert(cur_level_off + cur_level_cnt / INNER_CARD, cur_level_cnt % INNER_CARD, MAX_KEY);
118 |                 clwb(&inner_nodes_[cur_level_off], sizeof(INNode) * (cur_level_cnt / INNER_CARD + 1));
119 | 
120 |                 cur_level_cnt = std::ceil((float)cur_level_cnt / INNER_CARD);
121 |                 last_level_off = cur_level_off;
122 |                 cur_level_off = cur_level_off - std::pow(INNER_CARD, l - 1);
123 |             }
124 |             
125 |             leaf_cnt_ = lfnode_cnt;
126 |             entrance_ = (entrance_t *)galc->malloc(4096); // the allocator is not thread_safe, allocate a large entrance
127 |             uint32_t tmp = 0;
128 |             for(int l = 0; l < height_; l++) {
129 |                 level_offset_[l] = tmp;
130 |                 tmp += std::pow(INNER_CARD, l);
131 |             }
132 |             level_offset_[height_] = tmp;
133 | 
134 |             persist_assign(&(entrance_->leaf_buff), (void *) galc->relative(leaf_nodes_));
135 |             persist_assign(&(entrance_->inner_buff), (void *) galc->relative(inner_nodes_));
136 |             persist_assign(&(entrance_->height), height_);
137 |             persist_assign(&(entrance_->leaf_cnt), lfnode_cnt);
138 | 
139 |             return ;
140 |         }
141 | 
142 |     public:
143 |         char ** find_lower(_key_t key) { 
144 |             /* linear search, return the position of the stored value */
145 |             int cur_idx = level_offset_[0];
146 |             for(int l = 0; l < height_; l++) {
147 |                 #ifdef DEBUG
148 |                     INNode * inner = inner_nodes_ + cur_idx; 
149 |                 #endif
150 |                 cur_idx = level_offset_[l + 1] + (cur_idx - level_offset_[l]) * INNER_CARD + inner_search(cur_idx, key);
151 |             }
152 |             cur_idx -= level_offset_[height_];
153 | 
154 |             return leaf_search(cur_idx, key);
155 |         }
156 | 
157 |         bool insert(_key_t key, uint64_t val) {
158 |             uint32_t cur_idx = level_offset_[0];
159 |             for(int l = 0; l < height_; l++) {
160 |                 #ifdef DEBUG
161 |                     INNode * inner = inner_nodes_ + cur_idx; 
162 |                 #endif
163 |                 cur_idx = level_offset_[l + 1] + (cur_idx - level_offset_[l]) * INNER_CARD + inner_search(cur_idx, key);
164 |             }
165 |             cur_idx -= level_offset_[height_];
166 |             
167 |             LFNode * cur_leaf = leaf_nodes_ + cur_idx;
168 | 
169 |             for(int i = 0; i < LEAF_CARD; i++) {
170 |                 if (cur_leaf->keys[i] == MAX_KEY) { // empty slot
171 |                     leaf_insert(cur_idx, i, {key, (char *)val});
172 |                     return true;
173 |                 }
174 |             }
175 |             return false;
176 |         }
177 | 
178 |         bool try_remove(_key_t key) {
179 |             int cur_idx = level_offset_[0];
180 |             for(int l = 0; l < height_; l++) {
181 |                 #ifdef DEBUG
182 |                     INNode * inner = inner_nodes_ + cur_idx; 
183 |                 #endif
184 |                 cur_idx = level_offset_[l + 1] + (cur_idx - level_offset_[l]) * INNER_CARD + inner_search(cur_idx, key);
185 |             }
186 |             cur_idx -= level_offset_[height_];
187 |             
188 |             LFNode * cur_leaf = leaf_nodes_ + cur_idx;
189 | 
190 |             _key_t max_leqkey = cur_leaf->keys[0];
191 |             int8_t max_leqi = 0;
192 |             int8_t rec_cnt = 1;
193 |             for(int i = 1; i < LEAF_CARD; i++) {
194 |                 if(cur_leaf->keys[i] != MAX_KEY) {
195 |                     rec_cnt += 1;
196 |                     if (cur_leaf->keys[i] <= key && cur_leaf->keys[i] > max_leqkey) {
197 |                         max_leqkey = cur_leaf->keys[i];
198 |                         max_leqi = i;
199 |                     }
200 |                 }
201 |             }
202 | 
203 |             /* There are three cases:
204 |                   1. | k1 | --- | kx |, delete k1, leaf is not empty if k1 is deleted (fail)
205 |                   2. | k1 | ---      |, delete k1, leaf is empty if k1 is deleted (success)
206 |                   3. | k1 | --- | kx |, delete kx, leaf is not empty if kx is deleted (success)
207 |             */
208 |             if(max_leqi == 0 && rec_cnt > 1) { // case 1
209 |                 return false;
210 |             } else { // case 2, 3
211 |                 persist_assign(&(cur_leaf->keys[max_leqi]), MAX_KEY);
212 |                 return true;
213 |             }
214 |         }
215 | 
216 |         void printAll() {
217 |             for(int l = 0; l < height_; l++) {
218 |                 printf("level: %d =>", l);
219 |                 
220 |                 for(int i = level_offset_[l]; i < level_offset_[l + 1]; i++) {
221 |                     inner_print(i);
222 |                 }
223 |                 printf("\n");
224 |             }
225 | 
226 |             printf("leafs");
227 | 
228 |             for(int i = 0; i < leaf_cnt_; i++) {
229 |                 leaf_print(i);
230 |             }
231 |         }
232 | 
233 |         char ** find_first() {
234 |             return (char **)&(leaf_nodes_[0].vals[0]);
235 |         }
236 | 
237 |         void merge(std::vector<Record> & in, std::vector<Record> & out) { // merge the records with in to out
238 |             uint32_t insize = in.size();
239 | 
240 |             uint32_t incur = 0, innode_pos = 0, cur_lfcnt = 0;
241 |             Record tmp[LEAF_CARD];
242 |             load_node(tmp, &leaf_nodes_[0]);
243 |             _key_t k1 = in[0].key, k2 = tmp[0].key;
244 |             while(incur < insize && cur_lfcnt < leaf_cnt_) {
245 |                 if(k1 == k2) { 
246 |                     out.push_back(in[incur]);
247 | 
248 |                     incur += 1;
249 |                     innode_pos += 1;
250 |                     if(innode_pos == LEAF_CARD || tmp[innode_pos].key == MAX_KEY) {
251 |                         cur_lfcnt += 1;
252 |                         load_node(tmp, &leaf_nodes_[cur_lfcnt]);
253 |                         innode_pos = 0;
254 |                     }
255 | 
256 |                     k1 = in[incur].key;
257 |                     k2 = tmp[innode_pos].key;
258 |                 } else if(k1 > k2) {
259 |                     out.push_back(tmp[innode_pos]);
260 | 
261 |                     innode_pos += 1;
262 |                     if(innode_pos == LEAF_CARD || tmp[innode_pos].key == MAX_KEY) {
263 |                         cur_lfcnt += 1;
264 |                         load_node(tmp, &leaf_nodes_[cur_lfcnt]);
265 |                         innode_pos = 0;
266 |                     }
267 | 
268 |                     k2 = tmp[innode_pos].key;;
269 |                 } else {
270 |                     out.push_back(in[incur]);
271 |                     
272 |                     incur += 1;
273 |                     k1 = in[incur].key;
274 |                 }
275 |             }
276 | 
277 |             if(incur < insize) {
278 |                 for(int i = incur; i < insize; i++) 
279 |                     out.push_back(in[i]);
280 |             }
281 | 
282 |             if(cur_lfcnt < leaf_cnt_) {
283 |                 while(cur_lfcnt < leaf_cnt_) {
284 |                     out.push_back(tmp[innode_pos]);
285 | 
286 |                     innode_pos += 1;
287 |                     if(innode_pos == LEAF_CARD || tmp[innode_pos].key == MAX_KEY) {
288 |                         cur_lfcnt += 1;
289 |                         load_node(tmp, &leaf_nodes_[cur_lfcnt]);
290 |                         innode_pos = 0;
291 |                     }
292 |                 }
293 |             }
294 |         }
295 | 
296 |     private:
297 |         int inner_search(int node_idx, _key_t key) const{
298 |             INNode * cur_inner = inner_nodes_ + node_idx;
299 |             for(int i = 0; i < INNER_CARD; i++) {
300 |                 if (cur_inner->keys[i] > key) {
301 |                     return i - 1;
302 |                 }
303 |             }
304 | 
305 |             return INNER_CARD - 1;
306 |         }
307 |         
308 |         char **leaf_search(int node_idx, _key_t key) const {
309 |             LFNode * cur_leaf = leaf_nodes_ + node_idx;
310 | 
311 |             _key_t max_leqkey = cur_leaf->keys[0];
312 |             int8_t max_leqi = 0;
313 |             for(int i = 1; i < LEAF_CARD; i++) {
314 |                 if (cur_leaf->keys[i] <= key && cur_leaf->keys[i] > max_leqkey) {
315 |                     max_leqkey = cur_leaf->keys[i];
316 |                     max_leqi = i;
317 |                 }
318 |             }
319 | 
320 |             return (char **) &(cur_leaf->vals[max_leqi]);
321 |         }
322 | 
323 |         bool leaf_insert(int node_idx, int off, Record rec) { // TODO: should do it in a CAS way
324 |             leaf_nodes_[node_idx].vals[off] = rec.val;
325 |             clwb(&leaf_nodes_[node_idx].vals[off], 8);
326 |             mfence();
327 | 
328 |             leaf_nodes_[node_idx].keys[off] = rec.key;
329 |             clwb(&leaf_nodes_[node_idx].keys[off], 8);
330 |             mfence();
331 |             return true;
332 |         }
333 | 
334 |         inline void inner_insert(int node_idx, int off, _key_t key) {
335 |             inner_nodes_[node_idx].keys[off] = key;
336 |         }
337 | 
338 |         void inner_print(int node_idx) {
339 |             printf("(");
340 |             for(int i = 0; i < INNER_CARD; i++) {
341 |                 printf("%lu ", inner_nodes_[node_idx].keys[i]);
342 |             }
343 |             printf(") ");
344 |         }
345 | 
346 |         void leaf_print(int node_idx) {
347 |             printf("(");
348 |             for(int i = 0; i < LEAF_CARD; i++) {
349 |                 printf("[%lu, %lu] ", leaf_nodes_[node_idx].keys[i], (long unsigned int)leaf_nodes_[node_idx].vals[i]);
350 |             }
351 |             printf(") \n");
352 |         }
353 | 
354 |         static void load_node(Record * to, LFNode * from) {
355 |             for(int i = 0; i < LEAF_CARD; i++) {
356 |                 to[i].key = from->keys[i];
357 |                 to[i].val = from->vals[i];
358 |             }
359 | 
360 |             std::sort(to, to + LEAF_CARD);
361 |         }
362 | };
363 | 
364 | inline entrance_t * get_entrance(Fixtree * tree) {
365 |     return tree->entrance_;
366 | }
367 | 
368 | inline void free(Fixtree * tree) {
369 |     entrance_t * upent = get_entrance(tree);
370 |     delete tree;
371 | 
372 |     galc->free(galc->absolute(upent->inner_buff));
373 |     galc->free(galc->absolute(upent->leaf_buff));
374 |     galc->free(upent);
375 | 
376 |     return ;
377 | }
378 | 
379 | typedef Fixtree uptree_t;
380 | 
381 | } // namespace fixtree
382 | 
383 | #endif //__FIXTREE__


--------------------------------------------------------------------------------
/Single/src/flush.h:
--------------------------------------------------------------------------------
 1 | /*
 2 |     Copyright (c) Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
 3 |     USE AT YOUR OWN RISK!
 4 | */
 5 | 
 6 | #ifndef __FLUSH_H__
 7 | #define __FLUSH_H__
 8 | 
 9 | #include <x86intrin.h>
10 | 
11 | #include "common.h"
12 | 
13 | static inline void mfence() {
14 |     asm volatile("sfence" ::: "memory");
15 | }
16 | 
17 | static inline void flush(void * ptr) {
18 | #ifdef CLWB
19 |     _mm_clwb(ptr);
20 | #elif defined(CLFLUSHOPT)
21 |     _mm_clflushopt(ptr);
22 | #else
23 |     _mm_clflush(ptr);
24 | #endif
25 | }
26 | 
27 | inline void clwb(void *data, int len) {
28 | #ifdef DOFLUSH
29 |     volatile char *ptr = (char *)((unsigned long long)data &~(CACHE_LINE_SIZE-1));
30 |     for(; ptr < (char *)data + len; ptr += CACHE_LINE_SIZE) {
31 |         flush((void *)ptr);
32 |     }
33 | #endif //DOFLUSH
34 | }
35 | 
36 | inline void clflush(void *data, int len, bool fence=true)
37 | {
38 | #ifdef DOFLUSH
39 |     volatile char *ptr = (char *)((unsigned long long)data &~(CACHE_LINE_SIZE-1));
40 |     if(fence) mfence();
41 |     for(; ptr < (char *)data + len; ptr += CACHE_LINE_SIZE){
42 |         flush((void *)ptr);
43 |     }
44 |     if(fence) mfence();
45 | #endif //DOFLUSH
46 | }
47 | 
48 | template<typename T>
49 | inline void persist_assign(T* addr, const T &v) { // To ensure atomicity, the size of T should be less equal than 8
50 |     *addr = v;
51 |     clwb(addr, sizeof(T));
52 | }
53 | 
54 | #endif // __FLUSH_H__


--------------------------------------------------------------------------------
/Single/src/pmallocator.h:
--------------------------------------------------------------------------------
  1 | /*  
  2 |     A wrapper for using PMDK allocator easily (refer to pmwcas)
  3 |     Copyright (c) Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  4 |     USE AT YOUR OWN RISK!
  5 | */
  6 | 
  7 | #ifndef __BLKALLOCATOR_H__
  8 | #define __BLKALLOCATOR_H__
  9 | 
 10 | #include <cassert>
 11 | #include <cstdio>
 12 | #include <libpmemobj.h>
 13 | 
 14 | #include "common.h"
 15 | #include "flush.h"
 16 | 
 17 | POBJ_LAYOUT_BEGIN(pmallocator);
 18 | POBJ_LAYOUT_TOID(pmallocator, char)
 19 | POBJ_LAYOUT_END(pmallocator)
 20 | 
 21 | /*
 22 |     Persistent Memory Allocator: a wrapper of PMDK allocation lib: https://pmem.io/pmdk/
 23 | 
 24 |     PMAllocator allocates persistent memory from a pool file that resides on NVM file system. 
 25 |     It uses malloc() and free() as the allocation and reclaiment interfaces. 
 26 |     Other public interfaces like get_root(), absolute() and relative() are essential to memory
 27 |     management in persistent environment. 
 28 | */
 29 | class PMAllocator {
 30 | private:
 31 |     static const int PEICE_CNT = 64;
 32 |     static const size_t ALIGN_SIZE = 256;
 33 |     
 34 |     struct MetaType {
 35 |         char * buffer[PEICE_CNT];
 36 |         size_t blk_per_piece;
 37 |         size_t cur_blk;
 38 |         // entrance of DS in buffer
 39 |         void * entrance;
 40 |     };
 41 |     MetaType * meta_;
 42 | 
 43 |     // volatile domain
 44 |     PMEMobjpool *pop_;
 45 | 
 46 |     char * buff_[PEICE_CNT];
 47 |     char * buff_aligned_[PEICE_CNT];
 48 |     size_t piece_size_;
 49 |     size_t cur_blk_;
 50 |     size_t max_blk_;
 51 | 
 52 | public: 
 53 |     /*
 54 |      *  Construct a PM allocator, map a pool file into virtual memory
 55 |      *  @param filename     pool file name
 56 |      *  @param recover      if doing recover, false for the first time 
 57 |      *  @param layout_name  ID of a group of allocations (in characters), each ID corresponding to a root entry
 58 |      *  @param pool_size    pool size of the pool file, vaild if the file doesn't exist
 59 |      */
 60 |     PMAllocator(const char *file_name, bool recover, const char *layout_name, uint64_t pool_size) {
 61 |         PMEMobjpool *tmp_pool = nullptr;
 62 |         pool_size = pool_size + ((pool_size & ((1 << 23) - 1)) > 0 ? (1 << 23) : 0); // align to 8MB
 63 | 	    if(recover == false) {
 64 |             if(file_exist(file_name)) {
 65 |                 printf("[CAUTIOUS]: The pool file already exists\n");
 66 |                 printf("Try (1) remove the pool file %s\nOr  (2) set the recover parameter to be true\n", file_name);
 67 |                 exit(-1);
 68 |             }
 69 |             pop_ = pmemobj_create(file_name, layout_name, pool_size, S_IWUSR | S_IRUSR);
 70 |             meta_ = (MetaType *)pmemobj_direct(pmemobj_root(pop_, sizeof(MetaType)));
 71 |             
 72 |             // maintain volatile domain
 73 |             uint64_t alloc_size = (pool_size >> 1) + (pool_size >> 2) + (pool_size >> 3); // 7/8 of the pool is used as block alloction
 74 |             for(int i = 0; i < PEICE_CNT; i++) {
 75 |                 buff_[i] = (char *)mem_alloc(alloc_size / PEICE_CNT);
 76 |                 buff_aligned_[i] = (char *) ((uint64_t)buff_[i] + ((uint64_t) buff_[i] % ALIGN_SIZE == 0 ? 0 : (ALIGN_SIZE - (uint64_t) buff_[i] % ALIGN_SIZE)));
 77 |             }
 78 |             piece_size_ = (alloc_size / PEICE_CNT) / ALIGN_SIZE - 1;
 79 |             cur_blk_ = 0;
 80 |             max_blk_ = piece_size_ * PEICE_CNT;
 81 |             
 82 |             // initialize meta_
 83 |             for(int i = 0; i < PEICE_CNT; i++) 
 84 |                 meta_->buffer[i] = relative(buff_[i]);
 85 |             meta_->blk_per_piece = piece_size_;
 86 |             meta_->cur_blk = 0;
 87 |             meta_->entrance = NULL;
 88 |             clwb(meta_, sizeof(MetaType));
 89 |         } else {
 90 |             if(!file_exist(file_name)) {
 91 |                 printf("Pool File Not Exist\n");
 92 | 		        exit(-1);
 93 | 	        }
 94 |             pop_ = pmemobj_open(file_name, layout_name);
 95 |             meta_ = (MetaType *)pmemobj_direct(pmemobj_root(pop_, sizeof(MetaType)));
 96 |             // maintain volatile domain
 97 |             for(int i = 0; i < PEICE_CNT; i++) {
 98 |                 buff_[i] = absolute(meta_->buffer[i]);
 99 |                 buff_aligned_[i] = (char *) ((uint64_t)buff_[i] + ((uint64_t) buff_[i] % ALIGN_SIZE == 0 ? 0 : (ALIGN_SIZE - (uint64_t) buff_[i] % ALIGN_SIZE)));
100 |             }
101 |             cur_blk_ = meta_->cur_blk;
102 |             piece_size_ = meta_->blk_per_piece;
103 |             max_blk_ = piece_size_ * PEICE_CNT;
104 |         }
105 |     }
106 | 
107 |     ~PMAllocator() {
108 |         pmemobj_close(pop_);
109 |     }
110 | 
111 | public:
112 |     /*  
113 |      *  Get/allocate the root entry of the allocator.
114 |      *  
115 |      *  The root entry is the entrance of one group of allocation, each group is
116 |      *  identified by the layout_name when constructing it.
117 |      * 
118 |      *  Each group of allocations is a independent, self-contained in-memory structure in the pool
119 |      *  such as b-tree or link-list
120 |      */
121 |     void * get_root(size_t nsize) { // the root of DS stored in buff_ is recorded at meta_->entrance
122 |         if(meta_->entrance == NULL) {
123 |             meta_->entrance = relative(malloc(nsize));
124 |             clwb(meta_, sizeof(MetaType));
125 |         }
126 |         return absolute(meta_->entrance);
127 |     }
128 | 
129 |     /*
130 |      *  Allocate a non-root piece of persistent memory from the mapped pool
131 |      *  return the virtual memory address
132 |      */
133 |     void * malloc(size_t nsize) { 
134 |         if(nsize >= (1 << 12)) { // large than 4KB
135 |             void * mem = mem_alloc(nsize + ALIGN_SIZE); // not aligned
136 |             //  |  UNUSED    |HEADER|       memory you can use     |
137 |             // mem             (mem + off)
138 |             uint64_t offset = ALIGN_SIZE - (uint64_t)mem % ALIGN_SIZE;
139 |             // store a header in the front
140 |             uint64_t * header = (uint64_t *)((uint64_t)mem + offset - 8);
141 |             *header = offset;
142 | 
143 |             return (void *)((uint64_t)mem + offset);
144 |         }
145 |         
146 |         int blk_demand = (nsize + ALIGN_SIZE - 1) / ALIGN_SIZE;
147 |         // case 1: not enough in the buffer
148 |         if(blk_demand + cur_blk_ > max_blk_) {
149 |             printf("run out of memory\n");
150 |             exit(-1);
151 |         }
152 |         // case 2: current piece can not accommdate this allocation
153 |         int piece_id = cur_blk_ / piece_size_;
154 |         if((cur_blk_ % piece_size_ + blk_demand) > piece_size_) {
155 |             void * mem = buff_aligned_[piece_id + 1]; // allocate from a new peice
156 | 
157 |             cur_blk_ = piece_size_ * (piece_id + 1) + blk_demand;
158 |             meta_->cur_blk = cur_blk_;
159 |             clwb(&(meta_->cur_blk), 8);
160 | 
161 |             return mem;
162 |         } 
163 |         // case 3: current piece has enough space
164 |         else {
165 |             void * mem = buff_aligned_[piece_id] + ALIGN_SIZE * (cur_blk_ % piece_size_);
166 | 
167 |             cur_blk_ = cur_blk_ + blk_demand;
168 |             meta_->cur_blk = cur_blk_;
169 |             clwb(&(meta_->cur_blk), 8);
170 | 
171 |             return mem;
172 |         }
173 |     }
174 | 
175 |     void free(void* addr) {
176 |         for(int i = 0; i < PEICE_CNT; i++) {
177 |             uint64_t offset = (uint64_t)addr - (uint64_t)buff_aligned_[i];
178 |             if(offset > 0 && offset < piece_size_ * ALIGN_SIZE) {
179 |                 // the addr is in this piece, do not reclaim it
180 |                 return ;
181 |             }
182 |         }
183 | 
184 |         // larger than 4KB, reclaim it 
185 |         uint64_t * header = (uint64_t *)((uint64_t)addr - 8);
186 |         uint64_t offset = *header; 
187 | 
188 |         auto oid_ptr = pmemobj_oid((void *)((uint64_t)addr - offset));
189 |         TOID(char) ptr_cpy;
190 |         TOID_ASSIGN(ptr_cpy, oid_ptr);
191 |         POBJ_FREE(&ptr_cpy);
192 |     }  
193 | 
194 |     /*
195 |      *  Distinguish from virtual memory address and offset in the pool
196 |      *  Each memory piece allocated from the pool has an in-pool offset, which remains unchanged
197 |      *  until reclaiment. We cannot ensure that the pool file is mapped at the same position at 
198 |      *  any time, so it may locate at different virtual memory addresses next time. 
199 |      *  
200 |      *  So the rule is that, using virtual memory when doing normal operations like to DRAM
201 |      *  space, using offset to store link relationship, for exmaple, next pointer in linklist
202 |      * /
203 | 
204 |     /*
205 |      *  convert the virtual memory address to an offset
206 |      */
207 |     template<typename T>
208 |     inline T *absolute(T *pmem_offset) {
209 |         if(pmem_offset == NULL)
210 |             return NULL;
211 |         return reinterpret_cast<T *>(reinterpret_cast<uint64_t>(pmem_offset) + reinterpret_cast<char *>(pop_));
212 |     }
213 |     
214 |     template<typename T>
215 |     inline T *relative(T *pmem_direct) {
216 |         if(pmem_direct == NULL)
217 |             return NULL;
218 |         return reinterpret_cast<T *>(reinterpret_cast<char *>(pmem_direct) - reinterpret_cast<char *>(pop_));
219 |     }
220 | 
221 | private:
222 |     void * mem_alloc(size_t nsize) {
223 |         PMEMoid tmp;
224 |         pmemobj_alloc(pop_, &tmp, nsize, TOID_TYPE_NUM(char), NULL, NULL);
225 |         
226 |         void * mem = pmemobj_direct(tmp);
227 |         assert(mem != nullptr);
228 |         return mem;
229 |     }
230 | };
231 | 
232 | extern PMAllocator * galc;
233 | 
234 | #endif // __BLKALLOCATOR_H__
235 | 


--------------------------------------------------------------------------------
/Single/src/spinlock.h:
--------------------------------------------------------------------------------
 1 | /* Copyright(c): Guo Zhongming 
 2 | */
 3 | 
 4 | #ifndef __SPINLOCK_H__
 5 | #define __SPINLOCK_H__
 6 | 
 7 | #include <atomic>
 8 | #include <emmintrin.h>
 9 | #include <thread>
10 | 
11 | class Spinlock {
12 | public:
13 |     Spinlock() {
14 |         atomic_val.store(0, std::memory_order_relaxed);
15 |     }
16 | 
17 |     Spinlock(const Spinlock &) = delete;
18 |     Spinlock & operator = (const Spinlock &) = delete;
19 | 
20 | public:
21 |     inline void lock() {
22 |         while(atomic_val.exchange(1, std::memory_order_acquire) == 1){
23 |             while(1) {
24 |                 _mm_pause(); // delay for 140 cycle
25 | 
26 |                 if(atomic_val.load(std::memory_order_relaxed) == 0) // check the atomic_val
27 |                     break;
28 |                 
29 |                 std::this_thread::yield(); // delay for 113ns
30 | 
31 |                 if(atomic_val.load(std::memory_order_relaxed) == 0) // check the atomic_val
32 |                     break;
33 |             }
34 | 
35 |             // if at here, the atomic_val must be just be 0
36 |         }
37 | 
38 |         return ;
39 |     }
40 | 
41 |     inline void unlock() {
42 |         atomic_val.exchange(0, std::memory_order_acquire);
43 |     }
44 | 
45 |     inline bool trylock() {
46 |         return atomic_val.exchange(1, std::memory_order_acquire) == 0;
47 |     }
48 | 
49 | private:
50 |     std::atomic_short atomic_val;
51 | }; 
52 | 
53 | #endif // __SPINLOCK_H__


--------------------------------------------------------------------------------
/Single/src/tlbtree_impl.cc:
--------------------------------------------------------------------------------
1 | #include "tlbtree_impl.h"
2 | 
3 | PMAllocator * galc;


--------------------------------------------------------------------------------
/Single/src/tlbtree_impl.h:
--------------------------------------------------------------------------------
  1 | /*  tlbtree.h - A two level btree for persistent memory
  2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  3 |     USE AT YOUR OWN RISK!
  4 | */
  5 | 
  6 | #ifndef __TLBTREEIMPL_H__
  7 | #define __TLBTREEIMPL_H__
  8 | 
  9 | #include <string>
 10 | #include <thread>
 11 | #include <algorithm>
 12 | #include <unistd.h>
 13 | 
 14 | #include "pmallocator.h"
 15 | #include "fixtree.h"
 16 | #include "spinlock.h"
 17 | #include "wotree256.h"
 18 | 
 19 | #define BACKGROUND_REBUILD true
 20 | #define REBUILD_RECOVER false
 21 | 
 22 | // choose uptree type, providing interfaces: insert, remove, update, find, merge, free_uptree
 23 | #define UPTREE_NS   fixtree
 24 | // choose downtree type, providing interfaces: insert, find_lower, remove_lower
 25 | #define DOWNTREE_NS wotree256
 26 | 
 27 | namespace tlbtree {
 28 | using std::string;
 29 | using std::vector;
 30 | using Node = DOWNTREE_NS::Node;
 31 | 
 32 | template<int DOWNLEVEL, int REBUILD_THRESHOLD=2>
 33 | class TLBtreeImpl {
 34 | private:
 35 |     typedef TLBtreeImpl<DOWNLEVEL, REBUILD_THRESHOLD> SelfType;
 36 |     
 37 |     // the entrance of TLBtree that stores its persistent tree metadata
 38 |     struct tlbtree_entrance_t {
 39 |         UPTREE_NS::entrance_t * upent; // the entrance of the top layer
 40 |         Record * restore;              // restore subroots that fails to insert into the top layer
 41 |         int restore_size;
 42 |         bool is_clean;                 // is TLBtree shutdown expectedly
 43 |         bool use_rebuild_recover;      // whether to use recover rebuilding next time
 44 |     };
 45 |     
 46 |     // volatile domain
 47 |     UPTREE_NS::uptree_t * uptree_;
 48 |     tlbtree_entrance_t * entrance_;
 49 |     vector<Record> * mutable_;
 50 |     Spinlock rebuild_mtx_;
 51 |     bool is_rebuilding_;
 52 | 
 53 | public:
 54 |     TLBtreeImpl(string path, bool recover, uint64_t pool_size) {
 55 |         mutable_ = new vector<Record>();
 56 |         mutable_->reserve(0xfff);
 57 |         bool is_rebuilding_ = false;
 58 |         
 59 |         if(recover == false) {
 60 |             galc = new PMAllocator(path.c_str(), false, "tlbtree", pool_size);
 61 |             // initialize entrance_
 62 |             entrance_ = (tlbtree_entrance_t *) galc->get_root(sizeof(tlbtree_entrance_t));
 63 |             entrance_->upent = NULL;
 64 |             entrance_->restore = NULL;
 65 |             entrance_->restore_size = 0;
 66 |             entrance_->is_clean = false;
 67 |             entrance_->use_rebuild_recover = true;
 68 |             clwb(entrance_, sizeof(tlbtree_entrance_t));
 69 |             
 70 |             //allocate a entrance_ to the fixtree
 71 |             std::vector<Record> init = {Record(MIN_KEY, (char *)galc->relative(new Node()))}; 
 72 |             uptree_ = new UPTREE_NS::uptree_t(init);
 73 |             persist_assign(&(entrance_->upent), galc->relative(UPTREE_NS::get_entrance(uptree_)));
 74 |             persist_assign(&(entrance_->use_rebuild_recover), REBUILD_RECOVER); // use fast rebuilding next time
 75 |         } else {
 76 |             galc = new PMAllocator(path.c_str(), true, "tlbtree", pool_size);
 77 | 
 78 |             entrance_ = (tlbtree_entrance_t *) galc->get_root(sizeof(tlbtree_entrance_t));
 79 |             if(entrance_ == NULL || entrance_->upent == NULL) { // empty tree
 80 |                 printf("the tree is empty\n");
 81 |                 exit(-1);
 82 |             }
 83 | 
 84 |             if(entrance_->is_clean == false) { // TLBtree crashed at last usage
 85 |                 persist_assign(&(entrance_->use_rebuild_recover), true); // use recover rebuilding next time
 86 |             } else { // normal shutdown
 87 |                 // recover all subroots from PM back to mutable_, within miliseconds
 88 |                 if(entrance_->restore != NULL) {
 89 |                     Record * rec = galc->absolute(entrance_->restore);
 90 |                     for(int i = 0; i < entrance_->restore_size; i++) {
 91 |                         mutable_->push_back(rec[i]);
 92 |                     }
 93 |                     entrance_->restore = NULL;
 94 |                     entrance_->restore_size = 0;
 95 |                     clwb(&entrance_->restore, 16);
 96 |                     galc->free(rec);
 97 |                 }
 98 |             }
 99 | 
100 |             uptree_ = new UPTREE_NS::uptree_t (galc->absolute(entrance_->upent));
101 |         }
102 | 
103 |         persist_assign(&(entrance_->is_clean), false); // set the TLBtree state to be dirty
104 |     }
105 | 
106 |     ~TLBtreeImpl() {
107 |         if(entrance_->use_rebuild_recover == false) { // fast rebuilding next time
108 |             // save all subroots in mutable_ into PM
109 |             Record * rec = (Record *) galc->malloc(std::max((size_t)4096, mutable_->size() * sizeof(Record)));
110 |             for(int i = 0; i < mutable_->size(); i++) {
111 |                 rec[i] = (*mutable_)[i];
112 |             }
113 |             clwb(rec, mutable_->size() * sizeof(Record));
114 |             mfence();
115 |             entrance_->restore = galc->relative(rec);
116 |             entrance_->restore_size = mutable_->size();
117 |             clwb(&entrance_->restore, 16);
118 |         }
119 | 
120 | 
121 |         persist_assign(&(entrance_->is_clean), true); // a intended shutdown
122 | 
123 |         delete uptree_;
124 |         delete mutable_;
125 |         delete galc;
126 |     }
127 | 
128 | public: // public interface
129 |     void insert(const _key_t & k, uint64_t v) { 
130 |         Node ** root_ptr = (Node **)uptree_->find_lower(k);
131 |         Node * downroot = (Node *)galc->absolute(*root_ptr);
132 |  
133 |         // travese in sibling chain
134 |         int8_t goes_steps = 0;
135 |         _key_t splitkey; Node ** sibling_ptr;
136 |         downroot->get_sibling(splitkey, sibling_ptr);
137 |         while(splitkey < k) { // the splitkey 
138 |             root_ptr = sibling_ptr; // where is current root store
139 |             downroot = (Node *)galc->absolute(*root_ptr);
140 |             downroot->get_sibling(splitkey, sibling_ptr);
141 |             goes_steps += 1;
142 |         }
143 |         res_t insert_res = DOWNTREE_NS::insert(root_ptr, k, v, DOWNLEVEL);
144 | 
145 |         // we rebuild if the searching in the linklist is too long 
146 |         if(goes_steps > REBUILD_THRESHOLD && rebuild_mtx_.trylock()) {
147 |             if(entrance_->use_rebuild_recover == true) {
148 |                 #ifdef BACKGROUND_REBUILD
149 |                     std::thread rebuild_thread(&SelfType::rebuild_recover, this);
150 |                     rebuild_thread.detach();
151 |                 #else
152 |                     rebuild_recover();
153 |                 #endif
154 |             } else {
155 |                 #ifdef BACKGROUND_REBUILD
156 |                     std::thread rebuild_thread(&SelfType::rebuild_fast, this);
157 |                     rebuild_thread.detach();
158 |                 #else
159 |                     rebuild_fast();
160 |                 #endif
161 |             } 
162 |         }
163 | 
164 |         if(insert_res.flag == true) { // a sub-index tree is splitted
165 |             // try save the sub-indices root into the top layer
166 |             bool succ = uptree_->insert(insert_res.rec.key, (uint64_t)galc->relative(insert_res.rec.val));
167 |             
168 |             // save these records into mutable_
169 |             if(is_rebuilding_ == true || succ == false) {
170 |                 mutable_->push_back({insert_res.rec.key, (char *)galc->relative(insert_res.rec.val)});
171 |             }
172 |         }
173 |     }
174 | 
175 |     bool find(const _key_t & k, uint64_t & v) {
176 |         Node ** root_ptr = (Node **)uptree_->find_lower(k);
177 |         Node * downroot = (Node *)galc->absolute(*root_ptr);
178 | 
179 |         // traverse in sibling chain
180 |         _key_t splitkey; Node ** sibling_ptr;
181 |         downroot->get_sibling(splitkey, sibling_ptr);
182 |         while(splitkey <= k) { // the splitkey 
183 |             root_ptr = sibling_ptr; // where is current root store
184 |             downroot = (Node *)galc->absolute(*root_ptr);
185 |             downroot->get_sibling(splitkey, sibling_ptr);
186 |         }
187 | 
188 |         return DOWNTREE_NS::find(root_ptr, k, v);
189 |     }
190 | 
191 |     bool remove(const _key_t & k) {
192 |         Node ** root_ptr = (Node **)uptree_->find_lower(k);
193 |         Node ** last_root_ptr = NULL; // record the last root ptr for laster use
194 |         Node *downroot = (Node *)galc->absolute(*root_ptr);
195 | 
196 |         // travese in sibling chain
197 |         _key_t splitkey; Node ** sibling_ptr;
198 |         downroot->get_sibling(splitkey, sibling_ptr);
199 |         while(splitkey < k) { // the splitkey 
200 |             root_ptr = sibling_ptr; // where is current root store
201 |             downroot = (Node *)galc->absolute(*root_ptr);
202 |             downroot->get_sibling(splitkey, sibling_ptr);
203 |         }
204 |         
205 |         bool emptyif = DOWNTREE_NS::remove(root_ptr, k);
206 |         if(emptyif) { // the DOWNTREE_NS is empty now
207 |            uptree_->try_remove(k); // TODO: rebuilding should also be triggered when the top layer is too empty
208 |         }
209 | 
210 |         return true;
211 |     }
212 | 
213 |     bool update(const _key_t & k, const uint64_t & v) {
214 |         Node ** root_ptr = (Node **)uptree_->find_lower(k);
215 |         Node * downroot = (Node *)galc->absolute(*root_ptr);
216 | 
217 |         // travese in sibling chain
218 |         _key_t splitkey; Node ** sibling_ptr;
219 |         downroot->get_sibling(splitkey, sibling_ptr);
220 |         while(splitkey < k) { // the splitkey 
221 |             root_ptr = sibling_ptr; // where is current root store
222 |             downroot = (Node *)galc->absolute(*root_ptr);
223 |             downroot->get_sibling(splitkey, sibling_ptr);
224 |         }
225 | 
226 |         return DOWNTREE_NS::update(root_ptr, k, v);
227 |     }
228 | 
229 |     void printAll() {
230 |         uptree_->printAll();
231 |     }
232 | 
233 | private:
234 |     void rebuild_fast() { // fast rebuilding function
235 |         // switch the restore to be immutable
236 |         vector<Record> * new_mutable = new vector<Record>;
237 |         new_mutable->reserve(0xffff);
238 |         vector<Record> * immutable = mutable_; // make the restore vector be immutable
239 |         mutable_ = new_mutable; // before this line, the mutable_ is still the old one
240 | 
241 |         is_rebuilding_ = true;
242 | 
243 |         std::sort(immutable->begin(), immutable->end());
244 |         // get the snapshot of all sub-index trees by combining the top layer with immutable
245 |         std::vector<Record> subroots;
246 |         subroots.reserve(0x2ffff);
247 |         uptree_->merge(*immutable, subroots);
248 | 
249 |         /* rebuild the top layer with immutable */  
250 |         UPTREE_NS::uptree_t * old_tree = uptree_;
251 |         UPTREE_NS::entrance_t * old_upent = galc->absolute(entrance_->upent);
252 |         UPTREE_NS::uptree_t * new_tree = new UPTREE_NS::uptree_t(subroots);
253 |         UPTREE_NS::entrance_t * new_upent = UPTREE_NS::get_entrance(new_tree);
254 |         
255 |         // install the new top layer
256 |         persist_assign(&(entrance_->upent), galc->relative(new_upent));
257 |         uptree_ = new_tree;
258 |         
259 |         /* free the old top layer */
260 |         #ifdef BACKGROUND_REBUILD
261 |             usleep(50); // TODO: wait until on-going readers finish
262 |         #endif
263 |         UPTREE_NS::free(old_tree); // free the old_tree
264 |     
265 |         is_rebuilding_ = false;
266 |         rebuild_mtx_.unlock();
267 | 
268 |         delete immutable;
269 |     }
270 | 
271 |     void rebuild_recover() { // slow rebuilding function 
272 |         is_rebuilding_ = true;
273 |         // get the snapshot of all sub-index trees by traverse in the down layer
274 |         std::vector<Record> subroots;
275 |         subroots.reserve(0x2fffff);
276 |         
277 |         _key_t split_key = 0; 
278 |         Node ** sibling_ptr = (Node **)uptree_->find_first();
279 |         Node * cur_root = (Node *)galc->absolute(*sibling_ptr);
280 |         while (cur_root != NULL) {
281 |             subroots.emplace_back(split_key, (char *)(*sibling_ptr));
282 |             // get next sibling
283 |             cur_root->get_sibling(split_key, sibling_ptr);
284 |             cur_root = galc->absolute(*sibling_ptr);
285 |         }
286 | 
287 |         /* rebuild the top layer with immutable */  
288 |         UPTREE_NS::uptree_t * old_tree = uptree_;
289 |         UPTREE_NS::entrance_t * old_upent = galc->absolute(entrance_->upent);
290 |         UPTREE_NS::uptree_t * new_tree = new UPTREE_NS::uptree_t(subroots);
291 |         UPTREE_NS::entrance_t * new_upent = UPTREE_NS::get_entrance(new_tree);
292 |         
293 |         // install the new top layer
294 |         persist_assign(&(entrance_->upent), galc->relative(new_upent));
295 |         uptree_ = new_tree;
296 |         
297 |         /* free the old top layer */
298 |         #ifdef BACKGROUND_REBUILD
299 |             usleep(50); // TODO: wait until on-going readers finish
300 |         #endif
301 |         UPTREE_NS::free(old_tree); // free the old_tree
302 | 
303 |         is_rebuilding_ = false;
304 |         rebuild_mtx_.unlock();
305 | 
306 |         persist_assign(&(entrance_->use_rebuild_recover), REBUILD_RECOVER); // use fast rebuilding next time
307 |     }
308 | };
309 | 
310 | } // tlbtree namespace
311 | 
312 | #endif //__TLBTREEIMPL_H__


--------------------------------------------------------------------------------
/Single/src/wotree256.cc:
--------------------------------------------------------------------------------
  1 | #include "wotree256.h"
  2 | 
  3 | namespace wotree256 {
  4 | 
  5 | bool insert_recursive(Node * n, _key_t k, uint64_t v, _key_t &split_k, Node * &split_node, int8_t &level) {
  6 |     if(n->leftmost_ptr_ == NULL) {
  7 |         return n->store(k, v, split_k, split_node);
  8 |     } else {
  9 |         level++;
 10 |         Node * child = (Node *) galc->absolute(n->get_child(k));
 11 |         
 12 |         _key_t split_k_child;
 13 |         Node * split_node_child;
 14 |         bool splitIf = insert_recursive(child, k, v, split_k_child, split_node_child, level);
 15 | 
 16 |         if(splitIf) { 
 17 |             return n->store(split_k_child, (uint64_t)galc->relative(split_node_child), split_k, split_node);
 18 |         } 
 19 |         return false;
 20 |     }
 21 | }
 22 | 
 23 | bool remove_recursive(Node * n, _key_t k) {
 24 |     if(n->leftmost_ptr_ == NULL) {
 25 |         n->remove(k);
 26 |         return n->state_.unpack.count < UNDERFLOW_CARD;
 27 |     }
 28 |     else {
 29 |         Node * child = (Node *) galc->absolute(n->get_child(k));
 30 | 
 31 |         bool shouldMrg = remove_recursive(child, k);
 32 | 
 33 |         if(shouldMrg) {
 34 |             Node *leftsib = NULL, *rightsib = NULL;
 35 |             n->get_lrchild(k, leftsib, rightsib);
 36 | 
 37 |             if(leftsib != NULL && (child->state_.unpack.count + leftsib->state_.unpack.count) < CARDINALITY) {
 38 |                 // merge with left node
 39 |                 int8_t slotid = child->state_.read(0);
 40 |                 n->remove(child->recs_[slotid].key);
 41 |                 Node::merge(leftsib, child);
 42 | 
 43 |                 return n->state_.unpack.count < UNDERFLOW_CARD;
 44 |             } else if (rightsib != NULL && (child->state_.unpack.count + rightsib->state_.unpack.count) < CARDINALITY) {
 45 |                 // merge with right node
 46 |                 int8_t slotid = rightsib->state_.read(0);
 47 |                 n->remove(rightsib->recs_[slotid].key);
 48 |                 Node::merge(child, rightsib);
 49 | 
 50 |                 return n->state_.unpack.count < UNDERFLOW_CARD;
 51 |             }
 52 |         }
 53 |         return false;
 54 |     }
 55 | }
 56 | 
 57 | bool find(Node ** rootPtr, _key_t key, uint64_t &val) {
 58 |     Node * cur = galc->absolute(*rootPtr);
 59 |     while(cur->leftmost_ptr_ != NULL) { // no prefetch here
 60 |         char * child_ptr = cur->get_child(key);
 61 |         cur = (Node *)galc->absolute(child_ptr);
 62 |     }
 63 | 
 64 |     val = (uint64_t) cur->get_child(key);
 65 | 
 66 |     if((char *)val == NULL)
 67 |         return false;
 68 |     else
 69 |         return true;
 70 | }
 71 | 
 72 | res_t insert(Node ** rootPtr, _key_t key, uint64_t val, int threshold) {
 73 |     Node *root_= galc->absolute(*rootPtr);
 74 |     
 75 |     int8_t level = 1;
 76 |     _key_t split_k;
 77 |     Node * split_node;
 78 |     bool splitIf = insert_recursive(root_, key, val, split_k, split_node, level);
 79 | 
 80 |     if(splitIf) {
 81 |         if(level < threshold) {
 82 |             Node *new_root = new Node;
 83 |             new_root->leftmost_ptr_ = (char *)galc->relative(root_);
 84 |             new_root->append({split_k, (char *)galc->relative(split_node)}, 0, 0);
 85 |             new_root->state_.unpack.count = 1;
 86 | 
 87 |             clwb(new_root, 64);
 88 | 
 89 |             mfence(); // a barrier to make sure the new node is persisted
 90 |             persist_assign(rootPtr, (Node *)galc->relative(new_root));
 91 | 
 92 |             return res_t(false, {0, NULL});
 93 |         } else {
 94 |             return res_t(true, {split_k, (char *)split_node});
 95 |         }   
 96 |     }
 97 |     else {
 98 |         return res_t(false, {0, NULL});
 99 |     }
100 | }
101 | 
102 | bool update(Node ** rootPtr, _key_t key, uint64_t val) {
103 |     Node * cur = galc->absolute(*rootPtr);
104 |     while(cur->leftmost_ptr_ != NULL) { // no prefetch here
105 |         char * child_ptr = cur->get_child(key);
106 |         cur = (Node *)galc->absolute(child_ptr);
107 |     }
108 | 
109 |     val = (uint64_t) cur->update(key, val);
110 |     return true;
111 | }
112 | 
113 | bool remove(Node ** rootPtr, _key_t key) {   
114 |     Node *root_= galc->absolute(*rootPtr);
115 |     if(root_->leftmost_ptr_ == NULL) {
116 |         root_->remove(key);
117 | 
118 |         return root_->state_.unpack.count == 0;
119 |     }
120 |     else {
121 |         Node * child = (Node *) galc->absolute(root_->get_child(key));
122 | 
123 |         bool shouldMrg = remove_recursive(child, key);
124 | 
125 |         if(shouldMrg) {
126 |             Node *leftsib = NULL, *rightsib = NULL;
127 |             root_->get_lrchild(key, leftsib, rightsib);
128 | 
129 |             if(leftsib != NULL && (child->state_.unpack.count + leftsib->state_.unpack.count) < CARDINALITY) {
130 |                 // merge with left node
131 |                 int8_t slotid = child->state_.read(0);
132 |                 root_->remove(child->recs_[slotid].key);
133 |                 Node::merge(leftsib, child);
134 |             } 
135 |             else if (rightsib != NULL && (child->state_.unpack.count + rightsib->state_.unpack.count) < CARDINALITY) {
136 |                 // merge with right node
137 |                 int8_t slotid = rightsib->state_.read(0);
138 |                 root_->remove(rightsib->recs_[slotid].key);
139 |                 Node::merge(child, rightsib);
140 |             }
141 |             
142 |             if(root_->state_.unpack.count == 0) { // the root is empty
143 |                 Node * old_root = root_;
144 | 
145 |                 persist_assign(rootPtr, (Node *)root_->leftmost_ptr_);
146 | 
147 |                 galc->free(old_root);
148 |             }
149 |         }
150 | 
151 |         return false;
152 |     } 
153 | }
154 | 
155 | void printAll(Node ** rootPtr) {
156 |     Node *root= galc->absolute(*rootPtr);
157 |     root->print("", true);
158 | }
159 | 
160 | } // namespace wotree256


--------------------------------------------------------------------------------
/Single/src/wotree256.h:
--------------------------------------------------------------------------------
  1 | /*
  2 |     log free wbtree with 256B nodesize, nonclass version of wbtree_slotonly
  3 |     Copyright(c) Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  4 |     USE AT YOUR OWN RISK!
  5 | */
  6 | 
  7 | #ifndef __WOTREE256__
  8 | #define __WOTREE256__
  9 | 
 10 | #include <cstdint>
 11 | #include <cstring>
 12 | #include <string>
 13 | #include <cstdio>
 14 | 
 15 | #include "flush.h"
 16 | #include "pmallocator.h"
 17 | 
 18 | namespace wotree256 {
 19 | using std::string;
 20 | constexpr int CARDINALITY = 13;
 21 | constexpr int UNDERFLOW_CARD = 4;
 22 | 
 23 | struct state_t {
 24 |     struct statefield_t { // totally 8 bytes
 25 |         uint64_t slotArray       : 52;
 26 |         uint64_t count           : 4;
 27 |         uint64_t sibling_version : 1;
 28 |         uint64_t latch           : 1;
 29 |         uint64_t node_version    : 6; 
 30 |     };
 31 |     // the real data field of state
 32 |     union {
 33 |         uint64_t pack;      // to do 8 bytes assignment
 34 |         statefield_t unpack;// to facilite accessing subfields
 35 |     };
 36 |     
 37 | public:
 38 |     state_t(uint64_t s = 0): pack(s) {}
 39 | 
 40 |     inline int8_t read(int8_t idx) const {
 41 |         uint64_t p = this->unpack.slotArray << 12;
 42 |         return (p & ((uint64_t)0xf << ((15 - idx) * 4))) >> ((15 - idx) * 4);
 43 |     }
 44 | 
 45 |     inline int8_t alloc() {
 46 |         int8_t occupy[CARDINALITY] = {0};
 47 |         for(int64_t i = 0; i < unpack.count; i++) {
 48 |             occupy[read(i)] = 1; 
 49 |         }
 50 |         for(int i = 0; i < CARDINALITY; i++) {
 51 |             if(occupy[i] == 0) return i;
 52 |         }
 53 |         #ifdef DEBUG
 54 |             assert(false);
 55 |         #endif
 56 |         return CARDINALITY; // never be here
 57 |     }
 58 |     
 59 |     inline uint64_t add(int8_t idx, int8_t slot) {
 60 |         state_t new_state(this->pack);
 61 | 
 62 |         // update bit fields
 63 |         uint64_t p      = this->unpack.slotArray << 12;
 64 |         uint64_t mask   = 0xffffffffffffffff >> (idx * 4);
 65 |         uint64_t add_value = (uint64_t)slot << ((15 - idx) * 4);
 66 |         new_state.unpack.slotArray = ((p & (~mask)) + add_value + ((p & mask) >> 4)) >> 12;
 67 |         new_state.unpack.count++;
 68 | 
 69 |         return new_state.pack;
 70 |     }
 71 | 
 72 |     inline uint64_t remove(int idx) { // delete a slot id at position idx
 73 |         state_t new_state(this->pack);
 74 |         // update bit fields
 75 |         uint64_t p      = this->unpack.slotArray << 12;
 76 |         uint64_t mask   = 0xffffffffffffffff >> (idx * 4);
 77 |         new_state.unpack.slotArray = ((p & ~mask) + ((p & (mask>>4)) << 4)) >> 12;
 78 |         new_state.unpack.count--;
 79 | 
 80 |         return new_state.pack;
 81 |     }
 82 | 
 83 |     inline uint64_t append(int8_t idx, int8_t slot) {
 84 |         // Append the record at slotid, append the slotArray entry, but DO NOT modify the count
 85 |         state_t new_state(this->pack);
 86 | 
 87 |         // update bit fields
 88 |         uint64_t p      = this->unpack.slotArray << 12;
 89 |         uint64_t mask   = 0xffffffffffffffff >> (idx * 4);
 90 |         uint64_t add_value = (uint64_t)slot << ((15 - idx) * 4);
 91 |         new_state.unpack.slotArray = ((p & (~mask)) + add_value + ((p & mask) >> 4)) >> 12;
 92 | 
 93 |         return new_state.pack;
 94 |     }
 95 | };
 96 | 
 97 | class Node {
 98 | public:
 99 |     // First Cache Line
100 |     state_t state_;      // a very complex and compact state field
101 |     char * leftmost_ptr_;// the left most child of current node
102 |     Record siblings_[2]; // shadow sibling of current node
103 |     // Slots 
104 |     Record recs_[CARDINALITY];
105 | 
106 |     friend class wbtree;
107 | 
108 | public:
109 |     Node(bool isleaf = false): state_(0), leftmost_ptr_(NULL) {
110 |         siblings_[0] = {MAX_KEY, NULL};
111 |         siblings_[1] = {MAX_KEY, NULL};
112 |     }
113 | 
114 |     void *operator new(size_t size) {
115 |         void * ret = galc->malloc(size);
116 |         return ret;
117 |     }
118 | 
119 |     bool store(_key_t k, uint64_t v, _key_t & split_k, Node * & split_node) {
120 |         if(state_.unpack.count == CARDINALITY) { // should split the node
121 |             uint64_t m = state_.unpack.count / 2;
122 |             split_k = recs_[state_.read(m)].key;
123 | 
124 |             // copy half of the records into split node
125 |             int8_t j = 0;
126 |             state_t new_state = state_;
127 |             if(leftmost_ptr_ == NULL) {
128 |                 split_node = new Node();
129 |                 for(int i = m; i < state_.unpack.count; i++) {
130 |                     int8_t slotid = state_.read(i);
131 |                     split_node->append(recs_[slotid], j, j);
132 |                     j += 1;
133 |                 }
134 | 
135 |                 new_state.unpack.count -= j;
136 |             } else {
137 |                 int8_t slotid = state_.read(m);
138 |                 split_node = new Node();
139 |                 split_node->leftmost_ptr_ = recs_[slotid].val;
140 | 
141 |                 for(int i = m + 1; i < state_.unpack.count; i++) {
142 |                     slotid = state_.read(i);
143 |                     split_node->append(recs_[slotid], j, j);
144 |                     j += 1;
145 |                 }
146 | 
147 |                 new_state.unpack.count -= (j + 1);
148 |             }
149 |             split_node->state_.unpack.count = j;
150 |             split_node->state_.unpack.sibling_version = 0;
151 |             // the sibling node of current node pointed by split_node
152 |             split_node->siblings_[0] = siblings_[state_.unpack.sibling_version];
153 |             clwb(split_node, 64); // persist header
154 |             clwb(&split_node->recs_[1], sizeof(Record) * (j - 1)); // persist all the inserted records
155 |             
156 |             // the split node is installed as the shadow sibling of current node
157 |             siblings_[(state_.unpack.sibling_version + 1) % 2] = {split_k, (char *)galc->relative(split_node)};
158 |             // persist_assign the state field
159 |             new_state.unpack.sibling_version = (state_.unpack.sibling_version + 1) % 2;
160 |             
161 |             mfence(); // a barrier here to make sure all the update is persisted to storage
162 | 
163 |             persist_assign(&(state_.pack), new_state.pack);
164 |             
165 |             // go on the insertion
166 |             if(k < split_k) {
167 |                 insertone(k, (char *)v);
168 |             } else {
169 |                 split_node->insertone(k, (char *)v);
170 |             }
171 |             return true;
172 |         } else {
173 |             insertone(k, (char *)v);
174 |             return false;
175 |         }
176 |     }
177 | 
178 |     char * get_child(_key_t k) {
179 |         Record &sibling = siblings_[state_.unpack.sibling_version];
180 | 
181 |         if(k >= sibling.key) { // if the node has splitted and k to find is in next node 
182 |             Node * sib_node = (Node *)galc->absolute(sibling.val);
183 |             return sib_node->get_child(k);
184 |         }
185 | 
186 |         if(leftmost_ptr_ == NULL) {
187 |             int8_t slotid = 0;
188 |             for(int i = 0; i < state_.unpack.count; i++) {
189 |                 slotid = state_.read(i);
190 |                 if(recs_[slotid].key >= k) {
191 |                     break;
192 |                 }
193 |             }
194 | 
195 |             if (recs_[slotid].key == k && state_.unpack.count > 0)
196 |                 return recs_[slotid].val;
197 |             else 
198 |                 return NULL;
199 |         } else {
200 |             int8_t slotid = 0, pos = state_.unpack.count;
201 |             for(int i = 0; i < state_.unpack.count; i++) {
202 |                 slotid = state_.read(i);
203 |                 if(recs_[slotid].key > k) {
204 |                     pos = i;
205 |                     break;
206 |                 }
207 |             }
208 | 
209 |             if (pos == 0) // all the key is bigger than k
210 |                 return leftmost_ptr_;
211 |             else 
212 |                 return recs_[state_.read(pos - 1)].val;
213 |         }
214 |     }
215 | 
216 |     bool update(_key_t k, uint64_t v) {
217 |         uint64_t slotid = 0;
218 |         for(int i = 0; i < state_.unpack.count; i++) {
219 |             slotid = state_.read(i);
220 |             if(recs_[slotid].key >= k) {
221 |                 break;
222 |             }
223 |         }
224 | 
225 |         if (recs_[slotid].key == k) {
226 |             recs_[slotid].val = (char *)v;
227 |             clwb(&recs_[slotid], sizeof(Record));
228 |             return true;
229 |         } else {
230 |             return false;
231 |         }
232 |     }
233 | 
234 |     bool remove(_key_t k) {
235 |         // Non-SMO delete takes only one clwb 
236 |         Record &sibling = siblings_[state_.unpack.sibling_version];
237 |         if(k >= sibling.key) { // if the node has splitted and k to find is in next node 
238 |             Node * sib_node = (Node *)galc->absolute(sibling.val);
239 |             return sib_node->remove(k);
240 |         }
241 | 
242 |         if(leftmost_ptr_ == NULL) {
243 |             int8_t idx, slotid;
244 |             for(idx = 0; idx < state_.unpack.count; idx++) {
245 |                 slotid = state_.read(idx);
246 |                 if(recs_[slotid].key >= k)
247 |                     break;
248 |             }
249 | 
250 |             if(recs_[slotid].key == k) {
251 |                 uint64_t newpack = state_.remove(idx);
252 |                 persist_assign(&(state_.pack), newpack);
253 |                 return true;
254 |             } else {
255 |                 return false;
256 |             }
257 |         } else {
258 |             int8_t idx;
259 |             for(idx = 0; idx < state_.unpack.count; idx++) {
260 |                 int8_t slotid = state_.read(idx);
261 |                 if(recs_[slotid].key > k) 
262 |                     break;
263 |             }
264 |             /* NOTICE: 
265 |                 * We will never remove the leftmost child in our wbtree design
266 |                 * So the idx here must be larger than 0
267 |                 */
268 |             uint64_t newpack = state_.remove(idx - 1);
269 |             persist_assign(&(state_.pack), newpack);
270 | 
271 |             return true;
272 |         }
273 |     }
274 | 
275 |     void print(string prefix, bool recursively) const {
276 |         printf("%s[%lx(%ld) ", prefix.c_str(), state_.unpack.slotArray, state_.unpack.count);
277 | 
278 |         for(int i = 0; i < state_.unpack.count; i++) {
279 |             printf("%d ", state_.read(i));
280 |         }
281 | 
282 |         for(int i = 0; i < state_.unpack.count; i++) {
283 |             int8_t slotid = state_.read(i);
284 |             printf("(%ld 0x%lx) ", recs_[slotid].key, (uint64_t)recs_[slotid].val);
285 |         }
286 |         printf("]\n");
287 | 
288 |         if(recursively && leftmost_ptr_ != NULL) {
289 |             Node * child = (Node *)galc->absolute(leftmost_ptr_);
290 |             child->print(prefix + "    ", recursively);
291 | 
292 |             for(int i = 0; i < state_.unpack.count; i++) {
293 |                 Node * child = (Node *)galc->absolute(recs_[state_.read(i)].val);
294 |                 child->print(prefix + "    ", recursively);
295 |             }
296 |         }
297 |     }
298 | 
299 |     void get_sibling(_key_t & k, Node ** &sibling) {
300 |         Record &sib = siblings_[state_.unpack.sibling_version];
301 |         k = sib.key;
302 |         sibling = (Node **)&(sib.val);
303 |     }
304 | 
305 | public:
306 |     void insertone(_key_t key, char * right) {
307 |         int8_t idx;
308 |         for(idx = 0; idx < state_.unpack.count; idx++) {
309 |             int8_t slotid = state_.read(idx);
310 |             if(key < recs_[slotid].key) {
311 |                 break;
312 |             }
313 |         }
314 |         
315 |         // insert and flush the kv
316 |         int8_t slotid = state_.alloc(); // alloc a slot in the node
317 |         recs_[slotid] = {key, (char *) right};
318 |         clwb(&recs_[slotid], sizeof(Record));
319 |         mfence();
320 | 
321 |         //write_version++;
322 | 
323 |         // atomically update the state
324 |         uint64_t new_pack = state_.add(idx, slotid);
325 |         persist_assign(&(state_.pack), new_pack);
326 |     }
327 | 
328 |     void append(Record r, int8_t slotid, int8_t pos) {
329 |         recs_[slotid] = r;
330 |         state_.pack = state_.append(pos, slotid);
331 |     }
332 | 
333 |     static void merge(Node * left, Node * right) {
334 |         Record & sibling = left->siblings_[left->state_.unpack.sibling_version];
335 | 
336 |         state_t new_state = left->state_;
337 |         if(left->leftmost_ptr_ != NULL) { // insert the leftmost_ptr of the right node
338 |             int8_t slotid = new_state.alloc();
339 |             left->append({sibling.key, right->leftmost_ptr_}, slotid, new_state.unpack.count);
340 |             new_state.pack = new_state.add(new_state.unpack.count, slotid);
341 |         }
342 |         for(int i = 0; i < right->state_.unpack.count; i++) {
343 |             int8_t slotid = new_state.alloc();
344 |             left->append(right->recs_[right->state_.read(i)], slotid, new_state.unpack.count);
345 |             new_state.pack = new_state.add(new_state.unpack.count, slotid);;
346 |         }
347 |         
348 |         Record tmp = right->siblings_[right->state_.unpack.sibling_version];
349 |         left->siblings_[(left->state_.unpack.sibling_version + 1) % 2] = tmp;
350 |         new_state.unpack.sibling_version = (left->state_.unpack.sibling_version + 1) % 2;
351 |         
352 |         clwb(left, sizeof(Node)); // persist the whole leaf node
353 | 
354 |         // persist_assign the state_ field
355 |         mfence();
356 |         left->state_.pack = new_state.pack;
357 |         clwb(left, 64);
358 | 
359 |         galc->free(right); // WARNING: persistent memory leak here
360 |     }
361 | 
362 |     void get_lrchild(_key_t k, Node * & left, Node * & right) {
363 |         int16_t i = 0;
364 |         for( ; i < state_.unpack.count; i++) {
365 |             int8_t slotid = state_.read(i);
366 |             if(recs_[slotid].key > k)
367 |                 break;
368 |         }
369 | 
370 |         if(i == 0) {
371 |             left = NULL;
372 |         } else if(i == 1) {
373 |             left = (Node *)galc->absolute(leftmost_ptr_);
374 |         } else {
375 |             left = (Node *)galc->absolute(recs_[state_.read(i - 2)].val);
376 |         }
377 | 
378 |         if(i == state_.unpack.count) {
379 |             right = NULL;
380 |         } else {
381 |             right = (Node *)galc->absolute(recs_[state_.read(i)].val);
382 |         }
383 |     }
384 | };
385 | 
386 | extern bool insert_recursive(Node * n, _key_t k, uint64_t v, _key_t &split_k, 
387 |                                 Node * &split_node, int8_t &level);
388 | extern bool remove_recursive(Node * n, _key_t k);
389 | extern bool find(Node ** rootPtr, _key_t key, uint64_t &val);
390 | extern res_t insert(Node ** rootPtr, _key_t key, uint64_t val, int threshold);
391 | extern bool update(Node ** rootPtr, _key_t key, uint64_t val);
392 | extern bool remove(Node ** rootPtr, _key_t key);
393 | extern void printAll(Node ** rootPtr);
394 | 
395 | }
396 | #endif // __WOTREE256__


--------------------------------------------------------------------------------
/Single/test/CMakeLists.txt:
--------------------------------------------------------------------------------
1 | add_executable(datagen "datagen.cc")
2 | 
3 | add_executable(main "main.cc")
4 | target_link_libraries(main tlbtree)
5 | 
6 | add_executable(preload "preload.cc")
7 | target_link_libraries(preload tlbtree)


--------------------------------------------------------------------------------
/Single/test/datagen.cc:
--------------------------------------------------------------------------------
  1 | /*  
  2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  3 |     USE AT YOUR OWN RISK!
  4 | */
  5 | #include <iostream>
  6 | #include <fstream>
  7 | #include <string>
  8 | #include <random>
  9 | #include <cstdlib>
 10 | #include <unistd.h>
 11 | #include <algorithm>
 12 | 
 13 | #include "common.h"
 14 | #include "zipfian.h"
 15 | 
 16 | using std::cout;
 17 | using std::endl;
 18 | using std::string;
 19 | using std::ofstream;
 20 | using std::ifstream;
 21 | 
 22 | enum DistributionType {RAND = 0, ZIPFIAN};
 23 | 
 24 | struct WorkloadType {
 25 |     int operations = KILO;
 26 |     float read = 1.0;
 27 |     float insert = 0;
 28 |     float update = 0;
 29 |     float remove = 0;
 30 |     DistributionType dist = RAND;
 31 |     float skewness = 0.8;
 32 |     bool valid() {
 33 |         return read + insert + update + remove == 1.0 && skewness > 0 && skewness < 1.0;
 34 |     }
 35 |     void print() {
 36 |         cout << "=========WORKLOAD TYPE=========" << endl;
 37 |         cout << "Operations  : " << operations << endl;
 38 |         cout << "Read Ratio  : " << read << endl;
 39 |         cout << "Insert Ratio: " << insert << endl;
 40 |         cout << "Update Ratio: " << update << endl;
 41 |         cout << "Remove Ratio: " << remove << endl;
 42 |         cout << "Distribution: " << (dist == RAND ? "random" : "Zipfian") << endl;
 43 |         if (dist == ZIPFIAN) {
 44 |             cout << "Skewness " << skewness << endl;
 45 |         }
 46 |         cout << "===============================" << endl;
 47 |     }
 48 | };
 49 | 
 50 | struct QueryType {
 51 |     OperationType op;
 52 |     int64_t key;
 53 | };
 54 | 
 55 | class OperationGenerator {
 56 | public :
 57 |     OperationType mappings_[100];
 58 |     std::default_random_engine gen_;
 59 |     std::uniform_int_distribution<uint32_t> dist_;
 60 | 
 61 |     OperationGenerator(WorkloadType &w) : gen_(getRandom()) {
 62 |         int read_end = 100 * w.read;
 63 |         int insert_end = read_end + 100 * w.insert;
 64 |         int update_end = insert_end + 100 * w.update;
 65 |         int remove_end = update_end + 100 * w.remove;
 66 | 
 67 |         for(int i = 0; i < read_end; i++)
 68 |             mappings_[i] = OperationType::READ;
 69 | 
 70 |         for(int i = read_end; i < insert_end; i++)
 71 |             mappings_[i] = OperationType::INSERT;
 72 | 
 73 |         for(int i = insert_end; i < update_end; i++)
 74 |             mappings_[i] = OperationType::UPDATE;
 75 | 
 76 |         for(int i = update_end; i < remove_end; i++)
 77 |             mappings_[i] = OperationType::DELETE;
 78 |     }
 79 | 
 80 |     OperationType next() {
 81 |         return mappings_[dist_(gen_) % 100];
 82 |     }
 83 | };
 84 | 
 85 | // try not to generate duplicate keys, because some tree indices may not get used to it
 86 | 
 87 | void gen_dataset(int64_t *arr, int64_t scale, bool random) {
 88 |     std::mt19937 gen(10007);
 89 |     if(random) {
 90 |         uint64_t step = MAX_KEY / scale;
 91 |         std::uniform_int_distribution<int64_t> dist(0, MAX_KEY);
 92 |         for(int64_t i = 0; i < scale; i++) {
 93 |             #ifdef DEBUG
 94 |                 arr[i] = i + 1;
 95 |             #else  
 96 |                 arr[i] = i * step + 1; // non duplicated keys, and no zero key 
 97 |             #endif
 98 |         }
 99 |         std::shuffle(arr, arr + scale - 1, gen);
100 |     } else {
101 |         std::normal_distribution<double> dist(MAX_KEY / 2, MAX_KEY / 8);
102 |         int64_t i = 0;
103 |         while (i < scale) {
104 |             double val = (int64_t)dist(gen);
105 |             if(val < 0 || val > (double) MAX_KEY) {
106 |                 continue;
107 |             } else {
108 |                 arr[i++] = (int64_t)std::round(val);
109 |             }
110 |         }
111 |     }
112 |     return ;
113 | }
114 | 
115 | void gen_workload(int64_t *arr, int64_t scale, QueryType * querys, WorkloadType w) {
116 |     std::mt19937 gen(getRandom());
117 |     std::uniform_int_distribution<int64_t> idx1_dist(0, scale - 1);
118 |     zipfian_int_distribution<int64_t> idx2_dist(0, scale - 1, w.skewness);
119 |     OperationGenerator op_gen(w);
120 | 
121 |     for(int i = 0; i < w.operations; i++) {
122 |         OperationType op = op_gen.next();
123 |         int idx = (w.dist == RAND ? idx1_dist(gen) : idx2_dist(gen));
124 |         // for insert operations, we should make sure the key does not exist in the dataset
125 |         int64_t key = (op == OperationType::INSERT ? arr[idx] + getRandom(): arr[idx]); 
126 | 
127 |         querys[i] = {op, key};
128 |     }
129 | }
130 | 
131 | int main(int argc, char ** argv) {
132 |     static const bool DATASET_RANDOM = true;
133 |     bool opt_zipfian  = false;
134 |     WorkloadType w;
135 | 
136 |     static const char * optstr = "r:i:u:d:o:s:hz"; 
137 |     opterr = 0;
138 |     char opt;
139 |     while((opt = getopt(argc, argv, optstr)) != -1) {
140 |         switch(opt) {
141 |         case 'o':
142 |             w.operations = atoi(optarg);
143 |             break;
144 |         case 's':
145 |             w.skewness = atof(optarg);
146 |             break;
147 |         case 'r':
148 |             w.read = atof(optarg);
149 |             break;
150 |         case 'i':
151 |             w.insert = atof(optarg);
152 |             break;
153 |         case 'd':
154 |             w.remove = atof(optarg);
155 |             break;
156 |         case 'u':
157 |             w.update = atof(optarg);
158 |             break;
159 |         case 'z':
160 |             opt_zipfian = true;
161 |             break;
162 |         case '?':
163 |         case 'h':
164 |         default:
165 |             cout << "USAGE: "<< argv[0] << "[option]" << endl;
166 |             cout << "\t -h: " << "Print the USAGE" << endl;
167 |             cout << "\t -z: " << "Use zipfian distribution (Not specified: random distribution)" << endl;
168 |             cout << "\t -o: " << "The number of operations" << endl;
169 |             cout << "\t -s: " << "The skewness of query workload(0 - 1)" << endl;
170 |             cout << "\t -r: " << "Read ratio" << endl;
171 |             cout << "\t -i: " << "Insert ratio" << endl;
172 |             cout << "\t -u: " << "update ratio" << endl;
173 |             cout << "\t -d: " << "Delete ratio" << endl;
174 |             exit(-1);
175 |         }
176 |     }
177 | 
178 |     if(!w.valid()) {
179 |         cout << "Invalid workload configuration" << endl;
180 |         exit(-1);
181 |     }
182 |     if(opt_zipfian == true) {
183 |         w.dist = ZIPFIAN;
184 |     }
185 | 
186 |     w.print();
187 | 
188 |     #ifdef DEBUG
189 |         uint64_t scale = LOADSCALE * KILO;
190 |     #else
191 |         uint64_t scale = LOADSCALE * MILLION;
192 |     #endif
193 | 
194 |     int64_t * arr = new int64_t[scale];
195 |     if(!file_exist("dataset.dat")) {
196 |         gen_dataset(arr, scale, DATASET_RANDOM);
197 |         ofstream fout("dataset.dat", std::ios::binary);
198 |         fout.write((char *) arr, sizeof(int64_t) * scale);
199 |         fout.close();
200 |         cout << "generate a dataset file" << endl;
201 |     } else {
202 |         ifstream fin("dataset.dat", std::ios::binary);
203 |         fin.read((char *)arr, sizeof(int64_t) * scale);
204 |         fin.close();
205 |     }
206 | 
207 |     QueryType * querys = new QueryType[w.operations];
208 |     gen_workload(arr, scale, querys, w);
209 |     
210 |     ofstream fout("workload.txt");
211 |     for(int i = 0; i < w.operations; i++) {
212 |         fout << querys[i].op << " " << querys[i].key << endl;
213 |     }
214 |     fout.close();
215 |     cout << "generate a query workload file" << endl;
216 | 
217 |     delete [] arr;
218 |     delete [] querys;
219 |     return 0;
220 | }


--------------------------------------------------------------------------------
/Single/test/gen.sh:
--------------------------------------------------------------------------------
 1 | #!/bin/bash
 2 | 
 3 | scale=$1
 4 | if [ $# -lt 1 ]; then
 5 |     scale=10000
 6 | fi
 7 | 
 8 | # Random RO
 9 | ./datagen -o $scale -r 1 -i 0
10 | mv workload.txt workload1.txt
11 | 
12 | # Random RW
13 | ./datagen -o $scale -r 0.5 -i 0.5
14 | mv workload.txt workload2.txt
15 | 
16 | # Random WO
17 | ./datagen -o $scale -r 0 -i 1
18 | mv workload.txt workload3.txt
19 | 
20 | # Zipfian RO
21 | ./datagen -o $scale -r 1 -i 0 -z -s 0.7
22 | mv workload.txt workload4.txt
23 | 
24 | # Zipfian RW
25 | ./datagen -o $scale -r 0.5 -i 0.5 -z -s 0.7
26 | mv workload.txt workload5.txt
27 | 
28 | # Zipfian WO
29 | ./datagen -o $scale -r 0 -i 1 -z -s 0.7
30 | mv workload.txt workload6.txt


--------------------------------------------------------------------------------
/Single/test/main.cc:
--------------------------------------------------------------------------------
 1 | /*  
 2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
 3 |     USE AT YOUR OWN RISK!
 4 | */
 5 | #include <iostream>
 6 | #include <fstream>
 7 | #include <cstdint>
 8 | #include <cstdlib>
 9 | #include <vector>
10 | #include <random>
11 | #include <algorithm>
12 | 
13 | #include "tlbtree.h"
14 | 
15 | using std::cout;
16 | using std::endl;
17 | using std::ifstream;
18 | using std::string;
19 | 
20 | template<typename BTreeType>
21 | double run_test(ifstream & fin) {
22 |     int small_noise = getRandom() & 0x3ff; // each time we run, we will insert different keys
23 |     
24 |     auto start = seconds();
25 | 
26 |     BTreeType tree("/mnt/pmem/tlbtree.pool");
27 |     int op_id, notfound = 0;
28 |     _key_t key;
29 |     uint64_t val;
30 |     for(int i = 0; fin >> op_id >> key; i++) {
31 |         switch (op_id) {
32 |             case OperationType::INSERT:
33 |                 tree.insert(key + small_noise, uint64_t(key + small_noise));
34 |                 break;
35 |             case OperationType::READ:
36 |                 val = tree.lookup(key);
37 |                 if(val == 0) 
38 |                     notfound++; // optimizer killer
39 |                 break;
40 |             case OperationType::UPDATE:
41 |                 tree.update(key, uint64_t(key * 2));
42 |                 break;
43 |             case OperationType::DELETE:
44 |                 tree.remove(key);
45 |                 break;
46 |             default:
47 |                 cout << "wrong operation id" << endl;
48 |                 break;
49 |         }
50 |     }
51 |     if(notfound > 0) cout << "somthing not found" << endl; // optimizer killer
52 | 
53 |     auto end = seconds();
54 |     return double(end - start);
55 | }
56 | 
57 | int main(int argc, char ** argv) {
58 |     int opt_testid = 1;
59 |     string opt_fname = "workload.txt";
60 | 
61 |     if(argc > 1) {
62 |         if(file_exist(argv[1]))
63 |             opt_fname = argv[1];
64 |         else
65 |             cout << "workload file "<< argv[1] << " not exist" << endl;
66 |     }
67 | 
68 |     ifstream fin(opt_fname.c_str());
69 |     double time = 0;
70 |     time = run_test<TLBtree>(fin);
71 |     
72 |     cout << time << endl;
73 |     fin.close();
74 |     return 0;
75 | }


--------------------------------------------------------------------------------
/Single/test/preload.cc:
--------------------------------------------------------------------------------
 1 | /*  
 2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
 3 |     USE AT YOUR OWN RISK!
 4 | */
 5 | #include <iostream>
 6 | #include <fstream>
 7 | #include <cstdint>
 8 | #include <cstdlib>
 9 | #include <vector>
10 | 
11 | #include "tlbtree.h"
12 | 
13 | using std::cout;
14 | using std::endl;
15 | using std::ifstream;
16 | 
17 | typedef double mytime_t;
18 | 
19 | _key_t *keys;
20 | PMAllocator * galc;
21 | 
22 | template <typename BTreeType>
23 | void preload(BTreeType &tree, uint64_t load_size, ifstream & fin) {
24 |     #ifdef DEBUG
25 |         fin.read((char *)keys, sizeof(_key_t) * MILLION);
26 |         for(int i = 0; i < LOADSCALE * KILO; i++) {
27 |             _key_t key = keys[i];
28 |             cout << key << endl;
29 |             tree.insert((_key_t)key, key);
30 |         }
31 |         //tree.printAll();
32 |     #else 
33 |         for(uint64_t t = 0; t < load_size; t++) {
34 |             fin.read((char *)keys, sizeof(_key_t) * MILLION);
35 | 
36 |             for(int i = 0; i < MILLION; i++) {
37 |                 _key_t key = keys[i];
38 |                 tree.insert((_key_t)key, uint64_t(key));
39 |             }
40 |         }
41 |     #endif
42 | 
43 |     return ;
44 | }
45 | 
46 | int main(int argc, char ** argv) {
47 |     // open the data file
48 |     std::string filename = "dataset.dat";
49 |     std::ifstream fin(filename.c_str(), std::ios::binary);
50 |     if(!fin) {
51 |         cout << "File not exists or Open error\n";
52 |         exit(-1);
53 |     }
54 | 
55 |     // read all the key into vector keys
56 |     keys = new _key_t[sizeof(_key_t) * MILLION];
57 |     
58 |     cout << "tlbtree" << endl;
59 |     TLBtree tree("/mnt/pmem/tlbtree.pool");
60 |     preload(tree, LOADSCALE, fin);
61 | 
62 |     delete keys;
63 |     fin.close();
64 | 
65 |     return 0;
66 | }
67 | 


--------------------------------------------------------------------------------
/Single/test/test.cc:
--------------------------------------------------------------------------------
  1 | /*  
  2 |     Copyright(c) 2020 Luo Yongping. THIS SOFTWARE COMES WITH NO WARRANTIES, 
  3 |     USE AT YOUR OWN RISK!
  4 | */
  5 | #include <iostream>
  6 | #include <fstream>
  7 | #include <cstdint>
  8 | #include <cstdlib>
  9 | #include <vector>
 10 | #include <random>
 11 | #include <algorithm>
 12 | #include <unistd.h>
 13 | 
 14 | #include "tlbtree.h"
 15 | 
 16 | using std::cout;
 17 | using std::endl;
 18 | using std::ifstream;
 19 | using std::string;
 20 | 
 21 | _key_t * keys;
 22 | uint32_t seed;
 23 | 
 24 | template <typename BTreeType>
 25 | double put_throughput(BTreeType &tree, uint32_t req_cnt) {
 26 |     auto start = seconds();
 27 |     for(int i = 0; i < req_cnt; i++) {
 28 |         _key_t key = keys[i];
 29 |         int64_t val = key;
 30 |         //cout << "insert " << key + seed << endl;
 31 |         tree.insert(key + seed, val);
 32 |     }
 33 |     auto end = seconds();
 34 |     return double(end - start);
 35 | }
 36 | 
 37 | template <typename BTreeType>
 38 | double get_throughput(BTreeType &tree, uint32_t req_cnt) {
 39 |     auto start = seconds();
 40 |     int64_t val;
 41 |     for(int i = 0; i < req_cnt; i++) {
 42 |         _key_t key = keys[i];
 43 |         bool foundif = tree.find(key, val);
 44 |         //cout << key << " " << val << endl;
 45 |         if(foundif == false) {
 46 |             cout << "Not Found " << key << endl;
 47 |         }
 48 |     }
 49 |     auto end = seconds();
 50 |     return double(end - start);
 51 | }
 52 | 
 53 | template <typename BTreeType>
 54 | double del_throughput(BTreeType &tree, uint32_t req_cnt) {
 55 |     auto start = seconds();
 56 |     int64_t val;
 57 |     for(int i = 0; i < req_cnt; i++) {
 58 |         cout << "delete " << keys[i] << endl;
 59 |         tree.remove(keys[i]);
 60 |     }
 61 |     auto end = seconds();
 62 |     return double(end - start);
 63 | }
 64 | 
 65 | template <typename BTreeType>
 66 | double update_throughput(BTreeType &tree, uint32_t req_cnt) {
 67 |     auto start = seconds();
 68 |     for(int i = 0; i < req_cnt; i++) {
 69 |         _key_t key = keys[i]; 
 70 |         //cout << "update " << key << endl;
 71 |         bool foundif = tree.update(key, (key * 2));
 72 |     }
 73 |     auto end = seconds();
 74 |     return double(end - start);
 75 | }
 76 | 
 77 | template <typename BtreeType>
 78 | void run_test(int opt_loadid, int opt_scale){
 79 |     switch(opt_loadid) {
 80 |     case 1: { // test insert
 81 |             BtreeType tree(true);
 82 |             double dur1 = put_throughput(tree, opt_scale);
 83 |             cout << dur1 << endl;
 84 |             break;
 85 |         }
 86 |     case 2: { // test read
 87 |             BtreeType tree(true);
 88 |             double dur1 = get_throughput(tree, opt_scale);
 89 |             cout << dur1 << endl;
 90 |             break;
 91 |         }
 92 |     case 3: { // test update
 93 |             BtreeType tree(true);
 94 |             double dur1 = update_throughput(tree, opt_scale);
 95 |             cout << dur1 << endl;
 96 |             break;
 97 |         }
 98 |     case 4: { // test delete
 99 |             BtreeType tree(true);
100 |             double dur1 = del_throughput(tree, opt_scale);
101 |             cout << dur1 << endl;
102 |             break;
103 |         }
104 |     default : { // basic test
105 |             seed = 0; // the tree is empty
106 |             BtreeType tree(false);
107 |             put_throughput(tree, opt_scale);
108 |             get_throughput(tree, opt_scale);
109 |             //del_throughput(tree, opt_scale);
110 |             //get_throughput(tree, opt_scale);
111 |             break;    
112 |         }
113 |     }
114 | }
115 | 
116 | void print_help(const char * name) {
117 |     cout << "USAGE: "<< name << "[option]" << endl;
118 |     cout << "\t -h: " << "Print the USAGE" << endl;
119 |     cout << "\t -s: " << "scale of the test" << endl;
120 |     cout << "\t -l: " << "The workload of the test (1:input 2:get: 3:update 4:delete other: multiple test)" << endl;
121 | }
122 | 
123 | int main(int argc, char ** argv) {
124 |     int opt_loadid = 2;
125 |     int opt_scale = KILO;
126 |     
127 |     static const char * optstr = "s:t:l:dh";
128 |     opterr = 0;
129 |     char opt;
130 | 
131 |     while((opt = getopt(argc, argv, optstr)) != -1) {
132 |         switch(opt) {
133 |         case 's':
134 |             if(atoi(optarg) > 0)
135 |                 opt_scale = KILO * atoi(optarg);
136 |             break;
137 |         case 'l':
138 |             if(atoi(optarg) > 0)
139 |                 opt_loadid = atoi(optarg);
140 |             break;
141 |         case '?':
142 |         case 'h':
143 |         default:
144 |             print_help(argv[0]);
145 |             exit(-1);
146 |             break;
147 |         }
148 |     }
149 | 
150 |     if(argc == 1) {
151 |         print_help(argv[0]);
152 |         exit(-1);
153 |     }
154 | 
155 |     // open the data file
156 |     std::string data_name = "dataset.dat";
157 |     std::ifstream fin(data_name.c_str(), std::ios::binary);
158 |     if(!fin) {
159 |         cout << "File not exists or Open error\n";
160 |         exit(-1);
161 |     }
162 |     // initialization of the test
163 |     keys = new _key_t[opt_scale];
164 |     seed = getRandom(); 
165 |     std::default_random_engine rd(seed);
166 |     std::uniform_int_distribution<uint64_t> dist(0, LOADSCALE * MILLION - opt_scale);
167 |     #ifdef DEBUG
168 |         // seek to the start of the file
169 |         fin.seekg(0, std::ios_base::beg);
170 |     #else
171 |         // seek to a random position in the file
172 |         fin.seekg(dist(rd) * sizeof(_key_t), std::ios_base::beg);
173 |     #endif
174 |     // read data from data file
175 |     fin.read((char *)keys, sizeof(_key_t) * opt_scale);
176 | 
177 |     cout << "tlbtree" << endl;
178 |     run_test<TLBtree>("/mnt/pmem/tlbtree.pool", opt_loadid, opt_scale);
179 | 
180 |     fin.close();
181 |     delete keys;
182 |     return 0;
183 | }


--------------------------------------------------------------------------------
/Single/test/zipfian.h:
--------------------------------------------------------------------------------
  1 | /* Implementation derived from:
  2 |  * "Quickly Generating Billion-Record Synthetic Databases", Jim Gray et al,
  3 |  * SIGMOD 1994
  4 |  *
  5 |  * The zipfian_int_distribution class is intended to be compatible with other
  6 |  * distributions introduced in #include <random> by the C++11 standard.
  7 |  *
  8 |  * Usage example:
  9 |  * #include <random>
 10 |  * #include "zipfian_int_distribution.h"
 11 |  * int main()
 12 |  * {
 13 |  *   std::default_random_engine generator;
 14 |  *   zipfian_int_distribution<int> distribution(1, 10, 0.99);
 15 |  *   int i = distribution(generator);
 16 |  * }
 17 |  */
 18 | 
 19 | /*
 20 |  * IMPORTANT: constructing the distribution object requires calculating the zeta
 21 |  * value which becomes prohibetively expensive for very large ranges. As an
 22 |  * alternative for such cases, the user can pass the pre-calculated values and
 23 |  * avoid the calculation every time.
 24 |  *
 25 |  * Usage example:
 26 |  * #include <random>
 27 |  * #include "zipfian_int_distribution.h"
 28 |  * int main()
 29 |  * {
 30 |  *   std::default_random_engine generator;
 31 |  *   zipfian_int_distribution<int>::param_type p(1, 1e6, 0.99, 27.000);
 32 |  *   zipfian_int_distribution<int> distribution(p);
 33 |  *   int i = distribution(generator);
 34 |  * }
 35 |  */
 36 | 
 37 | #include <cassert>
 38 | #include <cmath>
 39 | #include <limits>
 40 | #include <random>
 41 | 
 42 | template <typename _IntType = int>
 43 | class zipfian_int_distribution
 44 | {
 45 |     static_assert(std::is_integral<_IntType>::value, "Template argument not an integral type.");
 46 | 
 47 | public:
 48 |     /** The type of the range of the distribution. */
 49 |     typedef _IntType result_type;
 50 |     /** Parameter type. */
 51 |     struct param_type
 52 |     {
 53 |         typedef zipfian_int_distribution<_IntType> distribution_type;
 54 | 
 55 |         explicit param_type(_IntType __a = 0, _IntType __b = std::numeric_limits<_IntType>::max(), double __theta = 0.99)
 56 |             : _M_a(__a), _M_b(__b), _M_theta(__theta),
 57 |               _M_zeta(zeta(_M_b - _M_a + 1, __theta)), _M_zeta2theta(zeta(2, __theta))
 58 |         {
 59 |             assert(_M_a <= _M_b && _M_theta > 0.0 && _M_theta < 1.0);
 60 |         }
 61 | 
 62 |         explicit param_type(_IntType __a, _IntType __b, double __theta, double __zeta)
 63 |             : _M_a(__a), _M_b(__b), _M_theta(__theta), _M_zeta(__zeta),
 64 |               _M_zeta2theta(zeta(2, __theta))
 65 |         {
 66 |             __glibcxx_assert(_M_a <= _M_b && _M_theta > 0.0 && _M_theta < 1.0);
 67 |         }
 68 | 
 69 |         result_type a() const { return _M_a; }
 70 | 
 71 |         result_type b() const { return _M_b; }
 72 | 
 73 |         double theta() const { return _M_theta; }
 74 | 
 75 |         double zeta() const { return _M_zeta; }
 76 | 
 77 |         double zeta2theta() const { return _M_zeta2theta; }
 78 | 
 79 |         friend bool operator==(const param_type& __p1, const param_type& __p2)
 80 |         {
 81 |             return __p1._M_a == __p2._M_a
 82 |                 && __p1._M_b == __p2._M_b
 83 |                 && __p1._M_theta == __p2._M_theta
 84 |                 && __p1._M_zeta == __p2._M_zeta
 85 |                 && __p1._M_zeta2theta == __p2._M_zeta2theta;
 86 |         }
 87 | 
 88 |     private:
 89 |         _IntType _M_a;
 90 |         _IntType _M_b;
 91 |         double _M_theta;
 92 |         double _M_zeta;
 93 |         double _M_zeta2theta;
 94 | 
 95 |         /**
 96 |          * @brief Calculates zeta.
 97 |          *
 98 |          * @param __n [IN]  The size of the domain.
 99 |          * @param __theta [IN]  The skew factor of the distribution.
100 |          */
101 |         double zeta(unsigned long __n, double __theta)
102 |         {
103 |             double ans = 0.0;
104 |             for (unsigned long i=1; i<=__n; ++i)
105 |                 ans += std::pow(1.0 / i, __theta);
106 |             return ans;
107 |         }
108 |     };
109 | 
110 | public:
111 |     /**
112 |      * @brief Constructs a zipfian_int_distribution object.
113 |      *
114 |      * @param __a [IN]  The lower bound of the distribution.
115 |      * @param __b [IN]  The upper bound of the distribution.
116 |      * @param __theta [IN]  The skew factor of the distribution.
117 |      */
118 |     explicit zipfian_int_distribution(_IntType __a = _IntType(0), _IntType __b = _IntType(1), double __theta = 0.99)
119 |         : _M_param(__a, __b, __theta)
120 |     {
121 |     }
122 | 
123 |     explicit zipfian_int_distribution(const param_type& __p) : _M_param(__p)
124 |     {
125 |     }
126 | 
127 |     /**
128 |      * @brief Resets the distribution state.
129 |      *
130 |      * Does nothing for the zipfian int distribution.
131 |      */
132 |     void reset() {}
133 | 
134 |     result_type a() const { return _M_param.a(); }
135 | 
136 |     result_type b() const { return _M_param.b(); }
137 | 
138 |     double theta() const { return _M_param.theta(); }
139 | 
140 |     /**
141 |      * @brief Returns the parameter set of the distribution.
142 |      */
143 |     param_type param() const { return _M_param; }
144 | 
145 |     /**
146 |      * @brief Sets the parameter set of the distribution.
147 |      * @param __param The new parameter set of the distribution.
148 |      */
149 |     void param(const param_type& __param) { _M_param = __param; }
150 | 
151 |     /**
152 |      * @brief Returns the inclusive lower bound of the distribution range.
153 |      */
154 |     result_type min() const { return this->a(); }
155 | 
156 |     /**
157 |      * @brief Returns the inclusive upper bound of the distribution range.
158 |      */
159 |     result_type max() const { return this->b(); }
160 | 
161 |     /**
162 |      * @brief Generating functions.
163 |      */
164 |     template <typename _UniformRandomNumberGenerator>
165 |     result_type operator()(_UniformRandomNumberGenerator& __urng)
166 |     {
167 |         return this->operator()(__urng, _M_param);
168 |     }
169 | 
170 |     template <typename _UniformRandomNumberGenerator>
171 |     result_type operator()(_UniformRandomNumberGenerator& __urng, const param_type& __p)
172 |     {
173 |         double alpha = 1 / (1 - __p.theta());
174 |         double eta = (1 - std::pow(2.0 / (__p.b() - __p.a() + 1), 1 - __p.theta())) / (1 - __p.zeta2theta() / __p.zeta());
175 | 
176 |         double u = std::generate_canonical<double, std::numeric_limits<double>::digits, _UniformRandomNumberGenerator>(__urng);
177 | 
178 |         double uz = u * __p.zeta();
179 |         if (uz < 1.0)
180 |             return __p.a();
181 |         if (uz < 1.0 + std::pow(0.5, __p.theta()))
182 |             return __p.a() + 1;
183 | 
184 |         return __p.a() + ((__p.b() - __p.a() + 1) * std::pow(eta * u - eta + 1, alpha));
185 |     }
186 | 
187 |     /**
188 |      * @brief Return true if two zipfian int distributions have
189 |      *        the same parameters.
190 |      */
191 |     friend bool operator==(const zipfian_int_distribution& __d1, const zipfian_int_distribution& __d2)
192 |     {
193 |         return __d1._M_param == __d2._M_param;
194 |     }
195 | 
196 | private:
197 |     param_type _M_param;
198 | };


--------------------------------------------------------------------------------