Allocate page cache headers and page cache data buffers separately.

The data buffer will be a power of two in size and this gives some
malloc implementation additional optimization opportunitites. (CVS 4409)

FossilOrigin-Name: 2b755defe51a565a2b6ace58381d6e91f6f17db8

manifest

@@ -1,5 +1,5 @@
-C Updated\scomments\son\sjournal.c.\s\sNo\schanges\sto\scode.\s(CVS\s4408)
+C Allocate\spage\scache\sheaders\sand\spage\scache\sdata\sbuffers\sseparately.\nThe\sdata\sbuffer\swill\sbe\sa\spower\sof\stwo\sin\ssize\sand\sthis\sgives\ssome\nmalloc\simplementation\sadditional\soptimization\sopportunitites.\s(CVS\s4409)
-D 2007-09-06T13:49:37
+D 2007-09-06T22:19:15
 F Makefile.in cbfb898945536a8f9ea8b897e1586dd1fdbcc5db
 F Makefile.linux-gcc 65241babba6faf1152bf86574477baab19190499
 F README 9c4e2d6706bdcc3efdd773ce752a8cdab4f90028
@@ -81,7 +81,7 @@ F src/analyze.c 49b4bd45eb286d833793ed6bf72355a5c1974865
 F src/attach.c 02fd8779270b1df1c63e7ba6e6655b960fa0f3d5
 F src/auth.c d41c34f3150b3b8248d364770ef922bbcefbff82
 F src/btmutex.c 442be6f068d77ca9ffd69899cf0a3943c244548c
-F src/btree.c b1e8025ac1ba81d78c2d2bab47af8efd1d143995
+F src/btree.c f37796c6f0306ed61e69e4c54621350572df30d1
 F src/btree.h d0736ebca4b6eafbdd823c46a8de574cea078211
 F src/btreeInt.h 4330c19b8314545fdb209cc77e2a57f6a5290e9c
 F src/build.c 94d0d6dfd1e706c480903fbdda2e77466f21b898
@@ -120,7 +120,7 @@ F src/os_unix.c 522486e24d1a743efc366cb93a727fa9314797b4
 F src/os_unix.h 5768d56d28240d3fe4537fac08cc85e4fb52279e
 F src/os_win.c b663c171abbbfa8ca20e5b9b93f7a0f09df616e9
 F src/os_win.h 41a946bea10f61c158ce8645e7646b29d44f122b
-F src/pager.c 0ce52a22ebfae0c085b68e00e30e9bf9ea462d0d
+F src/pager.c 10a95d9373105af62255a05aac1ffe0400b93a3d
 F src/pager.h d783e7f184afdc33adff37ba58d4e029bd8793b3
 F src/parse.y 2d2ce439dc6184621fb0b86f4fc5aca7f391a590
 F src/pragma.c 363e548dafb52327face8d99757ab56a7b1c1b26
@@ -202,7 +202,7 @@ F test/bigrow.test f0aeb7573dcb8caaafea76454be3ade29b7fc747
 F test/bind.test 261fd1603613e7f877a516d29f281c9d8c2ecf52
 F test/bindxfer.test b9a57f66dbd317feeefa28bd65b6576f1592ee98
 F test/blob.test 28c3b25150684ee3d108bb78cfb67a472deef2f0
-F test/btree.test d394f22b03712f05fca9c59f6f4848a3cad01379
+F test/btree.test eef24f3dd52b5b268321ae5aacfe30d3baca4e84
 F test/btree2.test 4b56a2a4a4f84d68c77aef271223a713bf5ebafc
 F test/btree4.test 3797b4305694c7af6828675b0f4b1424b8ca30e4
 F test/btree5.test 8e5ff32c02e685d36516c6499add9375fe1377f2
@@ -570,7 +570,7 @@ F www/tclsqlite.tcl 8be95ee6dba05eabcd27a9d91331c803f2ce2130
 F www/vdbe.tcl 87a31ace769f20d3627a64fa1fade7fed47b90d0
 F www/version3.tcl 890248cf7b70e60c383b0e84d77d5132b3ead42b
 F www/whentouse.tcl fc46eae081251c3c181bd79c5faef8195d7991a5
-P 79cf4e886cd5f1cd22574ce13135d4e32c1047b6
+P 3298441086330d1d24c30b7c061dfec98e9ea3ac
-R e940832e4042881361230f79ce400588
+R fab7f476269f83c2a9fb525589500817
 U drh
-Z 99beef2eb5ea70985d445f46a130fc96
+Z 091fe80161eedb1e09f57479031bbfb6

manifest.uuid

@@ -1 +1 @@
-3298441086330d1d24c30b7c061dfec98e9ea3ac
+2b755defe51a565a2b6ace58381d6e91f6f17db8

src/btree.c

@@ -9,7 +9,7 @@
 **    May you share freely, never taking more than you give.
 **
 *************************************************************************
-** $Id: btree.c,v 1.422 2007/09/03 22:00:39 drh Exp $
+** $Id: btree.c,v 1.423 2007/09/06 22:19:15 drh Exp $
 **
 ** This file implements a external (disk-based) database using BTrees.
 ** See the header comment on "btreeInt.h" for additional information.
@@ -902,7 +902,8 @@ int sqlite3BtreeInitPage(
   assert( pParent==0 || pParent->pBt==pBt );
   assert( sqlite3_mutex_held(pBt->mutex) );
   assert( pPage->pgno==sqlite3PagerPagenumber(pPage->pDbPage) );
-  assert( pPage->aData == &((unsigned char*)pPage)[-pBt->pageSize] );
+  assert( pPage == sqlite3PagerGetExtra(pPage->pDbPage) );
+  assert( pPage->aData == sqlite3PagerGetData(pPage->pDbPage) );
   if( pPage->pParent!=pParent && (pPage->pParent!=0 || pPage->isInit) ){
     /* The parent page should never change unless the file is corrupt */
     return SQLITE_CORRUPT_BKPT;
@@ -969,7 +970,8 @@ static void zeroPage(MemPage *pPage, int flags){
   int first;
 
   assert( sqlite3PagerPagenumber(pPage->pDbPage)==pPage->pgno );
-  assert( &data[pBt->pageSize] == (unsigned char*)pPage );
+  assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage );
+  assert( sqlite3PagerGetData(pPage->pDbPage) == data );
   assert( sqlite3PagerIswriteable(pPage->pDbPage) );
   assert( sqlite3_mutex_held(pBt->mutex) );
   memset(&data[hdr], 0, pBt->usableSize - hdr);
@@ -1053,7 +1055,8 @@ static void releasePage(MemPage *pPage){
   if( pPage ){
     assert( pPage->aData );
     assert( pPage->pBt );
-    assert( &pPage->aData[pPage->pBt->pageSize]==(unsigned char*)pPage );
+    assert( sqlite3PagerGetExtra(pPage->pDbPage) == (void*)pPage );
+    assert( sqlite3PagerGetData(pPage->pDbPage)==pPage->aData );
     assert( sqlite3_mutex_held(pPage->pBt->mutex) );
     sqlite3PagerUnref(pPage->pDbPage);
   }
@@ -1731,7 +1734,7 @@ static void unlockBtreeIfUnused(BtShared *pBt){
     if( sqlite3PagerRefcount(pBt->pPager)>=1 ){
       if( pBt->pPage1->aData==0 ){
         MemPage *pPage = pBt->pPage1;
-        pPage->aData = &((u8*)pPage)[-pBt->pageSize];
+        pPage->aData = sqlite3PagerGetData(pPage->pDbPage);
         pPage->pBt = pBt;
         pPage->pgno = 1;
       }
@@ -4348,7 +4351,7 @@ static int reparentPage(BtShared *pBt, Pgno pgno, MemPage *pNewParent, int idx){
   if( pDbPage ){
     pThis = (MemPage *)sqlite3PagerGetExtra(pDbPage);
     if( pThis->isInit ){
-      assert( pThis->aData==(sqlite3PagerGetData(pDbPage)) );
+      assert( pThis->aData==sqlite3PagerGetData(pDbPage) );
       if( pThis->pParent!=pNewParent ){
         if( pThis->pParent ) sqlite3PagerUnref(pThis->pParent->pDbPage);
         pThis->pParent = pNewParent;
@@ -4899,12 +4902,10 @@ static int balance_nonroot(MemPage *pPage){
   ** process of being overwritten.
   */
   for(i=0; i<nOld; i++){
-    MemPage *p = apCopy[i] = (MemPage*)&aCopy[i][pBt->pageSize];
+    MemPage *p = apCopy[i] = (MemPage*)aCopy[i];
-    p->aData = &((u8*)p)[-pBt->pageSize];
+    memcpy(p, apOld[i], sizeof(MemPage));
-    memcpy(p->aData, apOld[i]->aData, pBt->pageSize + sizeof(MemPage));
+    p->aData = (void*)&p[1];
-    /* The memcpy() above changes the value of p->aData so we have to
+    memcpy(p->aData, apOld[i]->aData, pBt->pageSize);
-    ** set it again. */
-    p->aData = &((u8*)p)[-pBt->pageSize];
   }
 
   /*

src/pager.c

@@ -18,7 +18,7 @@
 ** file simultaneously, or one process from reading the database while
 ** another is writing.
 **
-** @(#) $Id: pager.c,v 1.385 2007/09/03 15:19:35 drh Exp $
+** @(#) $Id: pager.c,v 1.386 2007/09/06 22:19:15 drh Exp $
 */
 #ifndef SQLITE_OMIT_DISKIO
 #include "sqliteInt.h"
@@ -271,12 +271,11 @@ struct PgHdr {
 #ifdef SQLITE_ENABLE_MEMORY_MANAGEMENT
   PagerLruLink gfree;            /* Global list of nRef==0 pages */
 #endif
-  u32 notUsed;                   /* Buffer space */
 #ifdef SQLITE_CHECK_PAGES
   u32 pageHash;
 #endif
-  /* pPager->pageSize bytes of page data follow this header */
+  void *pData;                   /* Page data */
-  /* Pager.nExtra bytes of local data follow the page data */
+  /* Pager.nExtra bytes of local data appended to this header */
 };
 
 /*
@@ -313,11 +312,10 @@ struct PgHistory {
 ** Convert a pointer to a PgHdr into a pointer to its data
 ** and back again.
 */
-#define PGHDR_TO_DATA(P)  ((void*)(&(P)[1]))
+#define PGHDR_TO_DATA(P)    ((P)->pData)
-#define DATA_TO_PGHDR(D)  (&((PgHdr*)(D))[-1])
+#define PGHDR_TO_EXTRA(G,P) ((void*)&((G)[1]))
-#define PGHDR_TO_EXTRA(G,P) ((void*)&((char*)(&(G)[1]))[(P)->pageSize])
 #define PGHDR_TO_HIST(P,PGR)  \
-            ((PgHistory*)&((char*)(&(P)[1]))[(PGR)->pageSize+(PGR)->nExtra])
+            ((PgHistory*)&((char*)(&(P)[1]))[(PGR)->nExtra])
 
 /*
 ** A open page cache is an instance of the following structure.
@@ -1245,6 +1243,7 @@ static void pager_reset(Pager *pPager){
     PAGER_INCR(sqlite3_pager_pgfree_count);
     pNext = pPg->pNextAll;
     lruListRemove(pPg);
+    sqlite3_free(pPg->pData);
     sqlite3_free(pPg);
   }
   assert(pPager->lru.pFirst==0);
@@ -2495,6 +2494,7 @@ static void pager_truncate_cache(Pager *pPager){
       PAGER_INCR(sqlite3_pager_pgfree_count);
       unlinkPage(pPg);
       makeClean(pPg);
+      sqlite3_free(pPg->pData);
       sqlite3_free(pPg);
       pPager->nPage--;
     }
@@ -3158,6 +3158,7 @@ int sqlite3PagerReleaseMemory(int nReq){
       );
       IOTRACE(("PGFREE %p %d *\n", pPager, pPg->pgno));
       PAGER_INCR(sqlite3_pager_pgfree_count);
+      sqlite3_free(pPg->pData);
       sqlite3_free(pPg);
       pPager->nPage--;
     }else{
@@ -3394,6 +3395,7 @@ static int pagerSharedLock(Pager *pPager){
 static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){
   int rc = SQLITE_OK;
   PgHdr *pPg;
+  void *pData;
 
   /* Create a new PgHdr if any of the four conditions defined 
   ** above are met: */
@@ -3411,9 +3413,15 @@ static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){
       }
     }
     pagerLeave(pPager);
-    pPg = sqlite3_malloc( sizeof(*pPg) + pPager->pageSize
+    pPg = sqlite3_malloc( sizeof(*pPg) + sizeof(u32) + pPager->nExtra
-                            + sizeof(u32) + pPager->nExtra
                             + MEMDB*sizeof(PgHistory) );
+    if( pPg ){
+      pData = sqlite3_malloc( pPager->pageSize );
+      if( pData==0 ){
+        sqlite3_free(pPg);
+        pPg = 0;
+      }
+    }
     pagerEnter(pPager);
     if( pPg==0 ){
       rc = SQLITE_NOMEM;
@@ -3423,6 +3431,7 @@ static int pagerAllocatePage(Pager *pPager, PgHdr **ppPg){
     if( MEMDB ){
       memset(PGHDR_TO_HIST(pPg, pPager), 0, sizeof(PgHistory));
     }
+    pPg->pData = pData;
     pPg->pPager = pPager;
     pPg->pNextAll = pPager->pAll;
     pPager->pAll = pPg;
@@ -3987,7 +3996,6 @@ static int pager_write(PgHdr *pPg){
     */
     if( !pPg->inJournal && (pPager->useJournal || MEMDB) ){
       if( (int)pPg->pgno <= pPager->origDbSize ){
-        int szPg;
         if( MEMDB ){
           PgHistory *pHist = PGHDR_TO_HIST(pPg, pPager);
           PAGERTRACE3("JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
@@ -3997,8 +4005,8 @@ static int pager_write(PgHdr *pPg){
             memcpy(pHist->pOrig, PGHDR_TO_DATA(pPg), pPager->pageSize);
           }
         }else{
-          u32 cksum, saved;
+          u32 cksum;
-          char *pData2, *pEnd;
+          char *pData2;
 
           /* We should never write to the journal file the page that
           ** contains the database locks.  The following assert verifies
@@ -4006,20 +4014,21 @@ static int pager_write(PgHdr *pPg){
           assert( pPg->pgno!=PAGER_MJ_PGNO(pPager) );
           pData2 = CODEC2(pPager, pData, pPg->pgno, 7);
           cksum = pager_cksum(pPager, (u8*)pData2);
-          pEnd = pData2 + pPager->pageSize;
+          rc = write32bits(pPager->jfd, pPager->journalOff, pPg->pgno);
-          pData2 -= 4;
+          if( rc==SQLITE_OK ){
-          saved = *(u32*)pEnd;
+            rc = sqlite3OsWrite(pPager->jfd, pData2, pPager->pageSize,
-          put32bits(pEnd, cksum);
+                                pPager->journalOff + 4);
-          szPg = pPager->pageSize+8;
+            pPager->journalOff += pPager->pageSize+4;
-          put32bits(pData2, pPg->pgno);
+          }
-          rc = sqlite3OsWrite(pPager->jfd, pData2, szPg, pPager->journalOff);
+          if( rc==SQLITE_OK ){
-          IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno,
+            rc = write32bits(pPager->jfd, pPager->journalOff, cksum);
+            pPager->journalOff += 4;
+          }
+          IOTRACE(("JOUT %p %d %lld %d\n", pPager, pPg->pgno, 
                    pPager->journalOff, szPg));
           PAGER_INCR(sqlite3_pager_writej_count);
-          pPager->journalOff += szPg;
           PAGERTRACE5("JOURNAL %d page %d needSync=%d hash(%08x)\n",
                PAGERID(pPager), pPg->pgno, pPg->needSync, pager_pagehash(pPg));
-          *(u32*)pEnd = saved;
 
           /* An error has occured writing to the journal file. The 
           ** transaction will be rolled back by the layer above.
@@ -4068,9 +4077,11 @@ static int pager_write(PgHdr *pPg){
         page_add_to_stmt_list(pPg);
       }else{
         i64 offset = pPager->stmtNRec*(4+pPager->pageSize);
-        char *pData2 = CODEC2(pPager, pData, pPg->pgno, 7)-4;
+        char *pData2 = CODEC2(pPager, pData, pPg->pgno, 7);
-        put32bits(pData2, pPg->pgno);
+        rc = write32bits(pPager->stfd, offset, pPg->pgno);
-        rc = sqlite3OsWrite(pPager->stfd, pData2, pPager->pageSize+4, offset);
+        if( rc==SQLITE_OK ){
+          rc = sqlite3OsWrite(pPager->stfd, pData2, pPager->pageSize, offset+4);
+        }
         PAGERTRACE3("STMT-JOURNAL %d page %d\n", PAGERID(pPager), pPg->pgno);
         if( rc!=SQLITE_OK ){
           return rc;

test/btree.test

@@ -11,7 +11,7 @@
 # This file implements regression tests for SQLite library.  The
 # focus of this script is btree database backend
 #
-# $Id: btree.test,v 1.40 2007/06/25 08:16:58 danielk1977 Exp $
+# $Id: btree.test,v 1.41 2007/09/06 22:19:15 drh Exp $
 
 
 set testdir [file dirname $argv0]
@@ -782,6 +782,7 @@ do_test btree-10.2 {
   lindex [btree_pager_stats $::b1] 1
 } {2}
 do_test btree-10.3 {
+btree_breakpoint
   for {set i 1} {$i<=30} {incr i} {
     set key [format %03d $i]
     set data "*** $key *** $key *** $key *** $key ***"