SQLite

Check-in [49d20ede5f]
Login

Many hyperlinks are disabled.
Use anonymous login to enable hyperlinks.

Overview
Comment:Performance improvements for the RowSet object when it undergoes many cycles between RowSetInsert and RowSetTest.
Downloads: Tarball | ZIP archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1: 49d20ede5f4c0895a165126d5cf7c95a0510ba35
User & Date: drh 2012-04-05 01:37:32.902
Context
2012-04-05
20:04
Ignore the value of SQLITE_FCNTL_CHUNK_SIZE if it is negative. (check-in: 1b08fef945 user: drh tags: trunk)
01:37
Performance improvements for the RowSet object when it undergoes many cycles between RowSetInsert and RowSetTest. (check-in: 49d20ede5f user: drh tags: trunk)
2012-04-04
16:56
Add the ".trace" option to the command-line shell. (check-in: b9ac3d7e34 user: drh tags: trunk)
Changes
Unified Diff Ignore Whitespace Patch
Changes to src/rowset.c.
72
73
74
75
76
77
78





79
80
81
82
83
84
85
** The number of rowset entries per allocation chunk.
*/
#define ROWSET_ENTRY_PER_CHUNK  \
                       ((ROWSET_ALLOCATION_SIZE-8)/sizeof(struct RowSetEntry))

/*
** Each entry in a RowSet is an instance of the following object.





*/
struct RowSetEntry {            
  i64 v;                        /* ROWID value for this entry */
  struct RowSetEntry *pRight;   /* Right subtree (larger entries) or list */
  struct RowSetEntry *pLeft;    /* Left subtree (smaller entries) */
};








>
>
>
>
>







72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
** The number of rowset entries per allocation chunk.
*/
#define ROWSET_ENTRY_PER_CHUNK  \
                       ((ROWSET_ALLOCATION_SIZE-8)/sizeof(struct RowSetEntry))

/*
** Each entry in a RowSet is an instance of the following object.
**
** This same object is reused to store a linked list of trees of RowSetEntry
** objects.  In that alternative use, pRight points to the next entry
** in the list, pLeft points to the tree, and v is unused.  The
** RowSet.pForest value points to the head of this forest list.
*/
struct RowSetEntry {            
  i64 v;                        /* ROWID value for this entry */
  struct RowSetEntry *pRight;   /* Right subtree (larger entries) or list */
  struct RowSetEntry *pLeft;    /* Left subtree (smaller entries) */
};

101
102
103
104
105
106
107
108
109
110
111
112
113






114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157


158























159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179

180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
*/
struct RowSet {
  struct RowSetChunk *pChunk;    /* List of all chunk allocations */
  sqlite3 *db;                   /* The database connection */
  struct RowSetEntry *pEntry;    /* List of entries using pRight */
  struct RowSetEntry *pLast;     /* Last entry on the pEntry list */
  struct RowSetEntry *pFresh;    /* Source of new entry objects */
  struct RowSetEntry *pTree;     /* Binary tree of entries */
  u16 nFresh;                    /* Number of objects on pFresh */
  u8 isSorted;                   /* True if pEntry is sorted */
  u8 iBatch;                     /* Current insert batch */
};







/*
** Turn bulk memory into a RowSet object.  N bytes of memory
** are available at pSpace.  The db pointer is used as a memory context
** for any subsequent allocations that need to occur.
** Return a pointer to the new RowSet object.
**
** It must be the case that N is sufficient to make a Rowset.  If not
** an assertion fault occurs.
** 
** If N is larger than the minimum, use the surplus as an initial
** allocation of entries available to be filled.
*/
RowSet *sqlite3RowSetInit(sqlite3 *db, void *pSpace, unsigned int N){
  RowSet *p;
  assert( N >= ROUND8(sizeof(*p)) );
  p = pSpace;
  p->pChunk = 0;
  p->db = db;
  p->pEntry = 0;
  p->pLast = 0;
  p->pTree = 0;
  p->pFresh = (struct RowSetEntry*)(ROUND8(sizeof(*p)) + (char*)p);
  p->nFresh = (u16)((N - ROUND8(sizeof(*p)))/sizeof(struct RowSetEntry));
  p->isSorted = 1;
  p->iBatch = 0;
  return p;
}

/*
** Deallocate all chunks from a RowSet.  This frees all memory that
** the RowSet has allocated over its lifetime.  This routine is
** the destructor for the RowSet.
*/
void sqlite3RowSetClear(RowSet *p){
  struct RowSetChunk *pChunk, *pNextChunk;
  for(pChunk=p->pChunk; pChunk; pChunk = pNextChunk){
    pNextChunk = pChunk->pNextChunk;
    sqlite3DbFree(p->db, pChunk);
  }
  p->pChunk = 0;
  p->nFresh = 0;
  p->pEntry = 0;
  p->pLast = 0;
  p->pTree = 0;


  p->isSorted = 1;























}

/*
** Insert a new value into a RowSet.
**
** The mallocFailed flag of the database connection is set if a
** memory allocation fails.
*/
void sqlite3RowSetInsert(RowSet *p, i64 rowid){
  struct RowSetEntry *pEntry;  /* The new entry */
  struct RowSetEntry *pLast;   /* The last prior entry */
  assert( p!=0 );
  if( p->nFresh==0 ){
    struct RowSetChunk *pNew;
    pNew = sqlite3DbMallocRaw(p->db, sizeof(*pNew));
    if( pNew==0 ){
      return;
    }
    pNew->pNextChunk = p->pChunk;
    p->pChunk = pNew;
    p->pFresh = pNew->aEntry;

    p->nFresh = ROWSET_ENTRY_PER_CHUNK;
  }
  pEntry = p->pFresh++;
  p->nFresh--;
  pEntry->v = rowid;
  pEntry->pRight = 0;
  pLast = p->pLast;
  if( pLast ){
    if( p->isSorted && rowid<=pLast->v ){
      p->isSorted = 0;
    }
    pLast->pRight = pEntry;
  }else{
    assert( p->pEntry==0 ); /* Fires if INSERT after SMALLEST */
    p->pEntry = pEntry;
  }
  p->pLast = pEntry;
}

/*
** Merge two lists of RowSetEntry objects.  Remove duplicates.
**
** The input lists are connected via pRight pointers and are 
** assumed to each already be in sorted order.
*/
static struct RowSetEntry *rowSetMerge(
  struct RowSetEntry *pA,    /* First sorted list to be merged */
  struct RowSetEntry *pB     /* Second sorted list to be merged */
){
  struct RowSetEntry head;
  struct RowSetEntry *pTail;

  pTail = &head;







|

|



>
>
>
>
>
>




















|


|



















|
>
>
|
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>











<
<
<
<
<
<
|
<
<
<
>
|
|
|
|




|
|



<











|







106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205






206



207
208
209
210
211
212
213
214
215
216
217
218
219
220

221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
*/
struct RowSet {
  struct RowSetChunk *pChunk;    /* List of all chunk allocations */
  sqlite3 *db;                   /* The database connection */
  struct RowSetEntry *pEntry;    /* List of entries using pRight */
  struct RowSetEntry *pLast;     /* Last entry on the pEntry list */
  struct RowSetEntry *pFresh;    /* Source of new entry objects */
  struct RowSetEntry *pForest;   /* List of binary trees of entries */
  u16 nFresh;                    /* Number of objects on pFresh */
  u8 rsFlags;                    /* Various flags */
  u8 iBatch;                     /* Current insert batch */
};

/*
** Allowed values for RowSet.rsFlags
*/
#define ROWSET_SORTED  0x01   /* True if RowSet.pEntry is sorted */
#define ROWSET_NEXT    0x02   /* True if sqlite3RowSetNext() has been called */

/*
** Turn bulk memory into a RowSet object.  N bytes of memory
** are available at pSpace.  The db pointer is used as a memory context
** for any subsequent allocations that need to occur.
** Return a pointer to the new RowSet object.
**
** It must be the case that N is sufficient to make a Rowset.  If not
** an assertion fault occurs.
** 
** If N is larger than the minimum, use the surplus as an initial
** allocation of entries available to be filled.
*/
RowSet *sqlite3RowSetInit(sqlite3 *db, void *pSpace, unsigned int N){
  RowSet *p;
  assert( N >= ROUND8(sizeof(*p)) );
  p = pSpace;
  p->pChunk = 0;
  p->db = db;
  p->pEntry = 0;
  p->pLast = 0;
  p->pForest = 0;
  p->pFresh = (struct RowSetEntry*)(ROUND8(sizeof(*p)) + (char*)p);
  p->nFresh = (u16)((N - ROUND8(sizeof(*p)))/sizeof(struct RowSetEntry));
  p->rsFlags = ROWSET_SORTED;
  p->iBatch = 0;
  return p;
}

/*
** Deallocate all chunks from a RowSet.  This frees all memory that
** the RowSet has allocated over its lifetime.  This routine is
** the destructor for the RowSet.
*/
void sqlite3RowSetClear(RowSet *p){
  struct RowSetChunk *pChunk, *pNextChunk;
  for(pChunk=p->pChunk; pChunk; pChunk = pNextChunk){
    pNextChunk = pChunk->pNextChunk;
    sqlite3DbFree(p->db, pChunk);
  }
  p->pChunk = 0;
  p->nFresh = 0;
  p->pEntry = 0;
  p->pLast = 0;
  p->pForest = 0;
  p->rsFlags = ROWSET_SORTED;
}

/*
** Allocate a new RowSetEntry object that is associated with the
** given RowSet.  Return a pointer to the new and completely uninitialized
** objected.
**
** In an OOM situation, the RowSet.db->mallocFailed flag is set and this
** routine returns NULL.
*/
static struct RowSetEntry *rowSetEntryAlloc(RowSet *p){
  assert( p!=0 );
  if( p->nFresh==0 ){
    struct RowSetChunk *pNew;
    pNew = sqlite3DbMallocRaw(p->db, sizeof(*pNew));
    if( pNew==0 ){
      return 0;
    }
    pNew->pNextChunk = p->pChunk;
    p->pChunk = pNew;
    p->pFresh = pNew->aEntry;
    p->nFresh = ROWSET_ENTRY_PER_CHUNK;
  }
  p->nFresh--;
  return p->pFresh++;
}

/*
** Insert a new value into a RowSet.
**
** The mallocFailed flag of the database connection is set if a
** memory allocation fails.
*/
void sqlite3RowSetInsert(RowSet *p, i64 rowid){
  struct RowSetEntry *pEntry;  /* The new entry */
  struct RowSetEntry *pLast;   /* The last prior entry */










  /* This routine is never called after sqlite3RowSetNext() */
  assert( p!=0 && (p->rsFlags & ROWSET_NEXT)==0 );

  pEntry = rowSetEntryAlloc(p);
  if( pEntry==0 ) return;
  pEntry->v = rowid;
  pEntry->pRight = 0;
  pLast = p->pLast;
  if( pLast ){
    if( (p->rsFlags & ROWSET_SORTED)!=0 && rowid<=pLast->v ){
      p->rsFlags &= ~ROWSET_SORTED;
    }
    pLast->pRight = pEntry;
  }else{

    p->pEntry = pEntry;
  }
  p->pLast = pEntry;
}

/*
** Merge two lists of RowSetEntry objects.  Remove duplicates.
**
** The input lists are connected via pRight pointers and are 
** assumed to each already be in sorted order.
*/
static struct RowSetEntry *rowSetEntryMerge(
  struct RowSetEntry *pA,    /* First sorted list to be merged */
  struct RowSetEntry *pB     /* Second sorted list to be merged */
){
  struct RowSetEntry head;
  struct RowSetEntry *pTail;

  pTail = &head;
232
233
234
235
236
237
238
239

240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256

257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
    assert( pB==0 || pB->pRight==0 || pB->v<=pB->pRight->v );
    pTail->pRight = pB;
  }
  return head.pRight;
}

/*
** Sort all elements on the pEntry list of the RowSet into ascending order.

*/ 
static void rowSetSort(RowSet *p){
  unsigned int i;
  struct RowSetEntry *pEntry;
  struct RowSetEntry *aBucket[40];

  assert( p->isSorted==0 );
  memset(aBucket, 0, sizeof(aBucket));
  while( p->pEntry ){
    pEntry = p->pEntry;
    p->pEntry = pEntry->pRight;
    pEntry->pRight = 0;
    for(i=0; aBucket[i]; i++){
      pEntry = rowSetMerge(aBucket[i], pEntry);
      aBucket[i] = 0;
    }
    aBucket[i] = pEntry;

  }
  pEntry = 0;
  for(i=0; i<sizeof(aBucket)/sizeof(aBucket[0]); i++){
    pEntry = rowSetMerge(pEntry, aBucket[i]);
  }
  p->pEntry = pEntry;
  p->pLast = 0;
  p->isSorted = 1;
}


/*
** The input, pIn, is a binary tree (or subtree) of RowSetEntry objects.
** Convert this tree into a linked list connected by the pRight pointers
** and return pointers to the first and last elements of the new list.







|
>

|

|
<

<

|
<
|
|

|


|
>

|

|

|
<
<







259
260
261
262
263
264
265
266
267
268
269
270
271

272

273
274

275
276
277
278
279
280
281
282
283
284
285
286
287
288


289
290
291
292
293
294
295
    assert( pB==0 || pB->pRight==0 || pB->v<=pB->pRight->v );
    pTail->pRight = pB;
  }
  return head.pRight;
}

/*
** Sort all elements on the list of RowSetEntry objects into order of
** increasing v.
*/ 
static struct RowSetEntry *rowSetEntrySort(struct RowSetEntry *pIn){
  unsigned int i;
  struct RowSetEntry *pNext, *aBucket[40];



  memset(aBucket, 0, sizeof(aBucket));
  while( pIn ){

    pNext = pIn->pRight;
    pIn->pRight = 0;
    for(i=0; aBucket[i]; i++){
      pIn = rowSetEntryMerge(aBucket[i], pIn);
      aBucket[i] = 0;
    }
    aBucket[i] = pIn;
    pIn = pNext;
  }
  pIn = 0;
  for(i=0; i<sizeof(aBucket)/sizeof(aBucket[0]); i++){
    pIn = rowSetEntryMerge(pIn, aBucket[i]);
  }
  return pIn;


}


/*
** The input, pIn, is a binary tree (or subtree) of RowSetEntry objects.
** Convert this tree into a linked list connected by the pRight pointers
** and return pointers to the first and last elements of the new list.
351
352
353
354
355
356
357
358
359
360

361
362
363


364


365









366
367
368
369
370
371




372
373
374
375
376
377
378
379
380
381
382



383


384
385
386
387
388
389
390
391
392
393
394
395
396
397
398




399
400
401






402
403


















404



405


406
407

408
409
410




411

412
413
414
415
416
417
418

419
420
421
422
    p->pLeft = pLeft;
    p->pRight = rowSetNDeepTree(&pList, iDepth);
  }
  return p;
}

/*
** Convert the list in p->pEntry into a sorted list if it is not
** sorted already.  If there is a binary tree on p->pTree, then
** convert it into a list too and merge it into the p->pEntry list.

*/
static void rowSetToList(RowSet *p){
  if( !p->isSorted ){


    rowSetSort(p);


  }









  if( p->pTree ){
    struct RowSetEntry *pHead, *pTail;
    rowSetTreeToList(p->pTree, &pHead, &pTail);
    p->pTree = 0;
    p->pEntry = rowSetMerge(p->pEntry, pHead);
  }




}

/*
** Extract the smallest element from the RowSet.
** Write the element into *pRowid.  Return 1 on success.  Return
** 0 if the RowSet is already empty.
**
** After this routine has been called, the sqlite3RowSetInsert()
** routine may not be called again.  
*/
int sqlite3RowSetNext(RowSet *p, i64 *pRowid){



  rowSetToList(p);


  if( p->pEntry ){
    *pRowid = p->pEntry->v;
    p->pEntry = p->pEntry->pRight;
    if( p->pEntry==0 ){
      sqlite3RowSetClear(p);
    }
    return 1;
  }else{
    return 0;
  }
}

/*
** Check to see if element iRowid was inserted into the the rowset as
** part of any insert batch prior to iBatch.  Return 1 or 0.




*/
int sqlite3RowSetTest(RowSet *pRowSet, u8 iBatch, sqlite3_int64 iRowid){
  struct RowSetEntry *p;






  if( iBatch!=pRowSet->iBatch ){
    if( pRowSet->pEntry ){


















      rowSetToList(pRowSet);



      pRowSet->pTree = rowSetListToTree(pRowSet->pEntry);


      pRowSet->pEntry = 0;
      pRowSet->pLast = 0;

    }
    pRowSet->iBatch = iBatch;
  }




  p = pRowSet->pTree;

  while( p ){
    if( p->v<iRowid ){
      p = p->pRight;
    }else if( p->v>iRowid ){
      p = p->pLeft;
    }else{
      return 1;

    }
  }
  return 0;
}







|
|
|
>


|
>
>
|
>
>

>
>
>
>
>
>
>
>
>
|
|
|
<
|
|
>
>
>
>











>
>
>
|
>
>















>
>
>
>


|
>
>
>
>
>
>

|
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
|
>
>
>
|
>
>


>



>
>
>
>
|
>
|
|
|
|
|
|
|
>




375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406

407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
    p->pLeft = pLeft;
    p->pRight = rowSetNDeepTree(&pList, iDepth);
  }
  return p;
}

/*
** Take all the entries on p->pEntry and on the trees in p->pForest and
** sort them all together into one big ordered list on p->pEntry.
**
** This routine should only be called once in the life of a RowSet.
*/
static void rowSetToList(RowSet *p){

  /* This routine is called only once */
  assert( p!=0 && (p->rsFlags & ROWSET_NEXT)==0 );

  if( (p->rsFlags & ROWSET_SORTED)==0 ){
    p->pEntry = rowSetEntrySort(p->pEntry);
  }

  /* While this module could theoretically support it, sqlite3RowSetNext()
  ** is never called after sqlite3RowSetText() for the same RowSet.  So
  ** there is never a forest to deal with.  Should this change, simply
  ** remove the assert() and the #if 0. */
  assert( p->pForest==0 );
#if 0
  while( p->pForest ){
    struct RowSetEntry *pTree = p->pForest->pLeft;
    if( pTree ){
      struct RowSetEntry *pHead, *pTail;
      rowSetTreeToList(pTree, &pHead, &pTail);

      p->pEntry = rowSetEntryMerge(p->pEntry, pHead);
    }
    p->pForest = p->pForest->pRight;
  }
#endif
  p->rsFlags |= ROWSET_NEXT;  /* Verify this routine is never called again */
}

/*
** Extract the smallest element from the RowSet.
** Write the element into *pRowid.  Return 1 on success.  Return
** 0 if the RowSet is already empty.
**
** After this routine has been called, the sqlite3RowSetInsert()
** routine may not be called again.  
*/
int sqlite3RowSetNext(RowSet *p, i64 *pRowid){
  assert( p!=0 );

  /* Merge the forest into a single sorted list on first call */
  if( (p->rsFlags & ROWSET_NEXT)==0 ) rowSetToList(p);

  /* Return the next entry on the list */
  if( p->pEntry ){
    *pRowid = p->pEntry->v;
    p->pEntry = p->pEntry->pRight;
    if( p->pEntry==0 ){
      sqlite3RowSetClear(p);
    }
    return 1;
  }else{
    return 0;
  }
}

/*
** Check to see if element iRowid was inserted into the the rowset as
** part of any insert batch prior to iBatch.  Return 1 or 0.
**
** If this is the first test of a new batch and if there exist entires
** on pRowSet->pEntry, then sort those entires into the forest at
** pRowSet->pForest so that they can be tested.
*/
int sqlite3RowSetTest(RowSet *pRowSet, u8 iBatch, sqlite3_int64 iRowid){
  struct RowSetEntry *p, *pTree;

  /* This routine is never called after sqlite3RowSetNext() */
  assert( pRowSet!=0 && (pRowSet->rsFlags & ROWSET_NEXT)==0 );

  /* Sort entries into the forest on the first test of a new batch 
  */
  if( iBatch!=pRowSet->iBatch ){
    p = pRowSet->pEntry;
    if( p ){
      struct RowSetEntry **ppPrevTree = &pRowSet->pForest;
      if( (pRowSet->rsFlags & ROWSET_SORTED)==0 ){
        p = rowSetEntrySort(p);
      }
      for(pTree = pRowSet->pForest; pTree; pTree=pTree->pRight){
        ppPrevTree = &pTree->pRight;
        if( pTree->pLeft==0 ){
          pTree->pLeft = rowSetListToTree(p);
          break;
        }else{
          struct RowSetEntry *pAux, *pTail;
          rowSetTreeToList(pTree->pLeft, &pAux, &pTail);
          pTree->pLeft = 0;
          p = rowSetEntryMerge(pAux, p);
        }
      }
      if( pTree==0 ){
        *ppPrevTree = pTree = rowSetEntryAlloc(pRowSet);
        if( pTree ){
          pTree->v = 0;
          pTree->pRight = 0;
          pTree->pLeft = rowSetListToTree(p);
        }
      }
      pRowSet->pEntry = 0;
      pRowSet->pLast = 0;
      pRowSet->rsFlags |= ROWSET_SORTED;
    }
    pRowSet->iBatch = iBatch;
  }

  /* Test to see if the iRowid value appears anywhere in the forest.
  ** Return 1 if it does and 0 if not.
  */
  for(pTree = pRowSet->pForest; pTree; pTree=pTree->pRight){
    p = pTree->pLeft;
    while( p ){
      if( p->v<iRowid ){
        p = p->pRight;
      }else if( p->v>iRowid ){
        p = p->pLeft;
      }else{
        return 1;
      }
    }
  }
  return 0;
}