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

significant performance improvement there.

FossilOrigin-Name: 8bfcda3d10aec864d71d12a1248c37e4db6f8899

4 files changed, 44 insertions, 7 deletions

diff --git a/src/btree.c b/src/btree.c
index ef2a86152..fe2c067db 100644
--- a/src/btree.c
+++ b/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]);
diff --git a/src/btreeInt.h b/src/btreeInt.h
index 92a4d4469..74117848a 100644
--- a/src/btreeInt.h
+++ b/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
diff --git a/src/sqliteInt.h b/src/sqliteInt.h
index aed6557cd..907434889 100644
--- a/src/sqliteInt.h
+++ b/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
diff --git a/src/util.c b/src/util.c
index 8fdaf2678..4e029b0e0 100644
--- a/src/util.c
+++ b/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
 }