Make use of built-in bswap32() and bswap16() functions in GCC/Clang for a

significant performance improvement there.

FossilOrigin-Name: 8bfcda3d10aec864d71d12a1248c37e4db6f8899

manifest

@@ -1,5 +1,5 @@
-C Change\san\sunreachable\sbranch\sinto\san\sassert()\sin\ssqlite3PagerAcquire()\nand\soptimize\ssqlite3PcacheOpenSavepoint()\sby\sfactoring\sout\srarely\sused\scode\ninto\sa\ssubroutine.
-D 2015-06-30T11:07:32.245
+C Make\suse\sof\sbuilt-in\sbswap32()\sand\sbswap16()\sfunctions\sin\sGCC/Clang\sfor\sa\nsignificant\sperformance\simprovement\sthere.
+D 2015-06-30T15:10:29.855
 F Makefile.arm-wince-mingw32ce-gcc d6df77f1f48d690bd73162294bbba7f59507c72f
 F Makefile.in 285a0a234ed7610d431d91671c136098c2bd86a9
 F Makefile.linux-gcc 91d710bdc4998cb015f39edf3cb314ec4f4d7e23
@@ -269,9 +269,9 @@ F src/auth.c b56c78ebe40a2110fd361379f7e8162d23f92240
 F src/backup.c ff743689c4d6c5cb55ad42ed9d174b2b3e71f1e3
 F src/bitvec.c d1f21d7d91690747881f03940584f4cc548c9d3d
 F src/btmutex.c 45a968cc85afed9b5e6cf55bf1f42f8d18107f79
-F src/btree.c 08ae64b0387e72b8201981a2d7b0f58b92fdff29
+F src/btree.c 3a535c0118872c3ee4e198b80a62d09315381dab
 F src/btree.h 969adc948e89e449220ff0ff724c94bb2a52e9f1
-F src/btreeInt.h e4eabc722b1ae017ac9c266a75769d4c6a9afde6
+F src/btreeInt.h 64b5723d8410a1f02c94d99c46b1b74cd547e6ef
 F src/build.c b3f15255d5b16e42dafeaa638fd4f8a47c94ed70
 F src/callback.c 7b44ce59674338ad48b0e84e7b72f935ea4f68b0
 F src/complete.c addcd8160b081131005d5bc2d34adf20c1c5c92f
@@ -332,7 +332,7 @@ F src/shell.c 8af3cced094aebb5f57a8ad739b9dafc7867eed7
 F src/sqlite.h.in 76d2f5637eb795b6300d9dd3c3ec3632ffafd721
 F src/sqlite3.rc 992c9f5fb8285ae285d6be28240a7e8d3a7f2bad
 F src/sqlite3ext.h be1a718b7d2ce40ceba725ae92c8eb5f18003066
-F src/sqliteInt.h 89768198547bdd70a160e47643a0e493f711e8d0
+F src/sqliteInt.h cf643b5a34633de5d4703eba1e2b1a205af53c8a
 F src/sqliteLimit.h 216557999cb45f2e3578ed53ebefe228d779cb46
 F src/status.c f266ad8a2892d659b74f0f50cb6a88b6e7c12179
 F src/table.c 51b46b2a62d1b3a959633d593b89bab5e2c9155e
@@ -388,7 +388,7 @@ F src/treeview.c c84b1a8ebc7f1d00cd76ce4958eeb3ae1021beed
 F src/trigger.c 322f23aad694e8f31d384dcfa386d52a48d3c52f
 F src/update.c 487747b328b7216bb7f6af0695d6937d5c9e605f
 F src/utf.c fc6b889ba0779b7722634cdeaa25f1930d93820c
-F src/util.c a5471ac4834ca3abe9c569745700d3f4dd5f5574
+F src/util.c 89bfe78b4610d456ba431a0865309a20acc115f3
 F src/vacuum.c 2ddd5cad2a7b9cef7f9e431b8c7771634c6b1701
 F src/vdbe.c 5ee4a2bf871418f61d06dc256b9b3a0084b5ec46
 F src/vdbe.h 7a75045d879118b9d3af7e8b3c108f2f27c51473
@@ -1364,7 +1364,8 @@ F tool/vdbe_profile.tcl 67746953071a9f8f2f668b73fe899074e2c6d8c1
 F tool/warnings-clang.sh f6aa929dc20ef1f856af04a730772f59283631d4
 F tool/warnings.sh 48bd54594752d5be3337f12c72f28d2080cb630b
 F tool/win/sqlite.vsix deb315d026cc8400325c5863eef847784a219a2f
-P d9a0481ce0fd9824b1566f182d68ec46ef091b1b
-R 24b6e6de7d9b089958239ab689d72050
+P b406b20ecdf0bff63c5c222fab11cb9acee86765 ce8177e3e64d2ac4cd5d9e7757cdd5fcecd7d0ea
+R e9f634c6724f1e6a3b5c72453520bbb0
+T +closed ce8177e3e64d2ac4cd5d9e7757cdd5fcecd7d0ea
 U drh
-Z 9400732c29ac033e0975f1d6b596266c
+Z 3fa235f76195131887feff91c1f8744e

manifest.uuid

@@ -1 +1 @@
-b406b20ecdf0bff63c5c222fab11cb9acee86765
+8bfcda3d10aec864d71d12a1248c37e4db6f8899

src/btree.c

@@ -971,9 +971,9 @@ static int ptrmapGet(BtShared *pBt, Pgno key, u8 *pEType, Pgno *pPgno){
 ** This routine works only for pages that do not contain overflow cells.
 */
 #define findCell(P,I) \
-  ((P)->aData + ((P)->maskPage & get2byte(&(P)->aCellIdx[2*(I)])))
+  ((P)->aData + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)])))
 #define findCellPastPtr(P,I) \
-  ((P)->aDataOfst + ((P)->maskPage & get2byte(&(P)->aCellIdx[2*(I)])))
+  ((P)->aDataOfst + ((P)->maskPage & get2byteAligned(&(P)->aCellIdx[2*(I)])))
 
 
 /*
@@ -1754,7 +1754,7 @@ static int btreeInitPage(MemPage *pPage){
 
       if( !pPage->leaf ) iCellLast--;
       for(i=0; i<pPage->nCell; i++){
-        pc = get2byte(&data[cellOffset+i*2]);
+        pc = get2byteAligned(&data[cellOffset+i*2]);
         testcase( pc==iCellFirst );
         testcase( pc==iCellLast );
         if( pc<iCellFirst || pc>iCellLast ){
@@ -6625,7 +6625,7 @@ static int editPage(
 #ifdef SQLITE_DEBUG
   for(i=0; i<nNew && !CORRUPT_DB; i++){
     u8 *pCell = pCArray->apCell[i+iNew];
-    int iOff = get2byte(&pPg->aCellIdx[i*2]);
+    int iOff = get2byteAligned(&pPg->aCellIdx[i*2]);
     if( pCell>=aData && pCell<&aData[pPg->pBt->usableSize] ){
       pCell = &pTmp[pCell - aData];
     }
@@ -7127,7 +7127,7 @@ static int balance_nonroot(
       memset(&b.szCell[b.nCell+limit], 0, sizeof(b.szCell[0])*pOld->nOverflow);
       limit = pOld->aiOvfl[0];
       for(j=0; j<limit; j++){
-        b.apCell[b.nCell] = aData + (maskPage & get2byte(piCell));
+        b.apCell[b.nCell] = aData + (maskPage & get2byteAligned(piCell));
         piCell += 2;
         b.nCell++;
       }
@@ -7140,7 +7140,7 @@ static int balance_nonroot(
     piEnd = aData + pOld->cellOffset + 2*pOld->nCell;
     while( piCell<piEnd ){
       assert( b.nCell<nMaxCells );
-      b.apCell[b.nCell] = aData + (maskPage & get2byte(piCell));
+      b.apCell[b.nCell] = aData + (maskPage & get2byteAligned(piCell));
       piCell += 2;
       b.nCell++;
     }
@@ -9105,7 +9105,7 @@ static int checkTreePage(
     /* EVIDENCE-OF: R-02776-14802 The cell pointer array consists of K 2-byte
     ** integer offsets to the cell contents. */
     for(i=0; i<nCell; i++){
-      int pc = get2byte(&data[cellStart+i*2]);
+      int pc = get2byteAligned(&data[cellStart+i*2]);
       u32 size = 65536;
       if( pc<=usableSize-4 ){
         size = pPage->xCellSize(pPage, &data[pc]);

src/btreeInt.h

@@ -691,3 +691,16 @@ struct IntegrityCk {
 #define put2byte(p,v) ((p)[0] = (u8)((v)>>8), (p)[1] = (u8)(v))
 #define get4byte sqlite3Get4byte
 #define put4byte sqlite3Put4byte
+
+/*
+** get2byteAligned(), unlike get2byte(), requires that its argument point to a
+** two-byte aligned address.  get2bytea() is only used for accessing the
+** cell addresses in a btree header.
+*/
+#if SQLITE_BYTEORDER==4321
+# define get2byteAligned(x)  (*(u16*)(x))
+#elif SQLITE_BYTEORDER==1234 && GCC_VERSION>=4008000
+# define get2byteAligned(x)  __builtin_bswap16(*(u16*)(x))
+#else
+# define get2byteAligned(x)  ((x)[0]<<8 | (x)[1])
+#endif

src/sqliteInt.h

@@ -60,6 +60,13 @@
 # define _LARGEFILE_SOURCE 1
 #endif
 
+/* What version of GCC is being used.  0 means GCC is not being used */
+#ifdef __GNUC__
+# define GCC_VERSION (__GNUC__*1000000+__GNUC_MINOR__*1000+__GNUC_PATCHLEVEL__)
+#else
+# define GCC_VERSION 0
+#endif
+
 /* Needed for various definitions... */
 #if defined(__GNUC__) && !defined(_GNU_SOURCE)
 # define _GNU_SOURCE

src/util.c

@@ -1078,14 +1078,31 @@ int sqlite3VarintLen(u64 v){
 ** Read or write a four-byte big-endian integer value.
 */
 u32 sqlite3Get4byte(const u8 *p){
+#if SQLITE_BYTEORDER==4321
+  u32 x;
+  memcpy(&x,p,4);
+  return x;
+#elif SQLITE_BYTEORDER==1234 && defined(__GNUC__)
+  u32 x;
+  memcpy(&x,p,4);
+  return __builtin_bswap32(x);
+#else
   testcase( p[0]&0x80 );
   return ((unsigned)p[0]<<24) | (p[1]<<16) | (p[2]<<8) | p[3];
+#endif
 }
 void sqlite3Put4byte(unsigned char *p, u32 v){
+#if SQLITE_BYTEORDER==4321
+  memcpy(p,&v,4);
+#elif SQLITE_BYTEORDER==1234 && defined(__GNUC__)
+  u32 x = __builtin_bswap32(v);
+  memcpy(p,&x,4);
+#else
   p[0] = (u8)(v>>24);
   p[1] = (u8)(v>>16);
   p[2] = (u8)(v>>8);
   p[3] = (u8)v;
+#endif
 }