/ Check-in [3e96d6ef]
Login

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

Overview
Comment:Minor simplification and performance optimization for Direct Overflow Read.
Downloads: Tarball | ZIP archive | SQL archive
Timelines: family | ancestors | descendants | both | trunk
Files: files | file ages | folders
SHA1: 3e96d6efa867b765c8acf1454014b1e71b2e4f21
User & Date: drh 2017-01-26 16:27:32
Context
2017-01-26
16:54
Performance optimization to sqlite3_blob_read(). check-in: 7459f4b7 user: drh tags: trunk
16:27
Minor simplification and performance optimization for Direct Overflow Read. check-in: 3e96d6ef user: drh tags: trunk
02:26
Make SQLITE_DIRECT_OVERFLOW_READ work for in WAL mode as long as the page being read is not in the the -wal file. check-in: 9879be18 user: drh tags: trunk
Changes
Hide Diffs Unified Diffs Ignore Whitespace Patch

Changes to src/btree.c.

4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
....
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589
4590
4591
4592
4593
4594
4595

4596
4597
4598
4599
4600
4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
4616
4617
4618
4619
4620
4621
){
  unsigned char *aPayload;
  int rc = SQLITE_OK;
  int iIdx = 0;
  MemPage *pPage = pCur->apPage[pCur->iPage]; /* Btree page of current entry */
  BtShared *pBt = pCur->pBt;                  /* Btree this cursor belongs to */
#ifdef SQLITE_DIRECT_OVERFLOW_READ
  unsigned char * const pBufStart = pBuf;
  int bEnd;                                 /* True if reading to end of data */
#endif

  assert( pPage );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->aiIdx[pCur->iPage]<pPage->nCell );
  assert( cursorHoldsMutex(pCur) );
  assert( eOp!=2 || offset==0 );    /* Always start from beginning for eOp==2 */

  getCellInfo(pCur);
  aPayload = pCur->info.pPayload;
#ifdef SQLITE_DIRECT_OVERFLOW_READ
  bEnd = offset+amt==pCur->info.nPayload;
#endif
  assert( offset+amt <= pCur->info.nPayload );

  assert( aPayload > pPage->aData );
  if( (uptr)(aPayload - pPage->aData) > (pBt->usableSize - pCur->info.nLocal) ){
    /* Trying to read or write past the end of the data is an error.  The
    ** conditional above is really:
    **    &aPayload[pCur->info.nLocal] > &pPage->aData[pBt->usableSize]
................................................................................
        }
        offset -= ovflSize;
      }else{
        /* Need to read this page properly. It contains some of the
        ** range of data that is being read (eOp==0) or written (eOp!=0).
        */
#ifdef SQLITE_DIRECT_OVERFLOW_READ
        sqlite3_file *fd;
#endif
        int a = amt;
        if( a + offset > ovflSize ){
          a = ovflSize - offset;
        }

#ifdef SQLITE_DIRECT_OVERFLOW_READ
        /* If all the following are true:
        **
        **   1) this is a read operation, and 
        **   2) data is required from the start of this overflow page, and
        **   3) the database is file-backed, and
        **   4) there is no open write-transaction, and

        **   5) the page is not in the WAL file
        **   6) all data from the page is being read.
        **   7) at least 4 bytes have already been read into the output buffer 
        **
        ** then data can be read directly from the database file into the
        ** output buffer, bypassing the page-cache altogether. This speeds
        ** up loading large records that span many overflow pages.
        */
        if( (eOp&0x01)==0                                      /* (1) */
         && offset==0                                          /* (2) */
         && (bEnd || a==ovflSize)                              /* (6) */
         && pBt->inTransaction==TRANS_READ                     /* (4) */
         && (fd = sqlite3PagerFile(pBt->pPager))->pMethods     /* (3) */
         && 0==sqlite3PagerUseWal(pBt->pPager, nextPage)       /* (5) */
         && &pBuf[-4]>=pBufStart                               /* (7) */
        ){
          u8 aSave[4];
          u8 *aWrite = &pBuf[-4];
          assert( aWrite>=pBufStart );                         /* hence (7) */
          memcpy(aSave, aWrite, 4);
          rc = sqlite3OsRead(fd, aWrite, a+4, (i64)pBt->pageSize*(nextPage-1));
          nextPage = get4byte(aWrite);
          memcpy(aWrite, aSave, 4);
        }else
#endif








|
<










<
<
<







 







|











<
|
>

<
|







<
|
|

|



|







4456
4457
4458
4459
4460
4461
4462
4463

4464
4465
4466
4467
4468
4469
4470
4471
4472
4473



4474
4475
4476
4477
4478
4479
4480
....
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
4587
4588
4589

4590
4591
4592

4593
4594
4595
4596
4597
4598
4599
4600

4601
4602
4603
4604
4605
4606
4607
4608
4609
4610
4611
4612
4613
4614
4615
){
  unsigned char *aPayload;
  int rc = SQLITE_OK;
  int iIdx = 0;
  MemPage *pPage = pCur->apPage[pCur->iPage]; /* Btree page of current entry */
  BtShared *pBt = pCur->pBt;                  /* Btree this cursor belongs to */
#ifdef SQLITE_DIRECT_OVERFLOW_READ
  unsigned char * const pBufStart = pBuf;     /* Start of original out buffer */

#endif

  assert( pPage );
  assert( pCur->eState==CURSOR_VALID );
  assert( pCur->aiIdx[pCur->iPage]<pPage->nCell );
  assert( cursorHoldsMutex(pCur) );
  assert( eOp!=2 || offset==0 );    /* Always start from beginning for eOp==2 */

  getCellInfo(pCur);
  aPayload = pCur->info.pPayload;



  assert( offset+amt <= pCur->info.nPayload );

  assert( aPayload > pPage->aData );
  if( (uptr)(aPayload - pPage->aData) > (pBt->usableSize - pCur->info.nLocal) ){
    /* Trying to read or write past the end of the data is an error.  The
    ** conditional above is really:
    **    &aPayload[pCur->info.nLocal] > &pPage->aData[pBt->usableSize]
................................................................................
        }
        offset -= ovflSize;
      }else{
        /* Need to read this page properly. It contains some of the
        ** range of data that is being read (eOp==0) or written (eOp!=0).
        */
#ifdef SQLITE_DIRECT_OVERFLOW_READ
        sqlite3_file *fd;      /* File from which to do direct overflow read */
#endif
        int a = amt;
        if( a + offset > ovflSize ){
          a = ovflSize - offset;
        }

#ifdef SQLITE_DIRECT_OVERFLOW_READ
        /* If all the following are true:
        **
        **   1) this is a read operation, and 
        **   2) data is required from the start of this overflow page, and

        **   3) there is no open write-transaction, and
        **   4) the database is file-backed, and
        **   5) the page is not in the WAL file

        **   6) at least 4 bytes have already been read into the output buffer 
        **
        ** then data can be read directly from the database file into the
        ** output buffer, bypassing the page-cache altogether. This speeds
        ** up loading large records that span many overflow pages.
        */
        if( (eOp&0x01)==0                                      /* (1) */
         && offset==0                                          /* (2) */

         && pBt->inTransaction==TRANS_READ                     /* (3) */
         && (fd = sqlite3PagerFile(pBt->pPager))->pMethods     /* (4) */
         && 0==sqlite3PagerUseWal(pBt->pPager, nextPage)       /* (5) */
         && &pBuf[-4]>=pBufStart                               /* (6) */
        ){
          u8 aSave[4];
          u8 *aWrite = &pBuf[-4];
          assert( aWrite>=pBufStart );                         /* due to (6) */
          memcpy(aSave, aWrite, 4);
          rc = sqlite3OsRead(fd, aWrite, a+4, (i64)pBt->pageSize*(nextPage-1));
          nextPage = get4byte(aWrite);
          memcpy(aWrite, aSave, 4);
        }else
#endif