Performance optimization on byte-swapping in R-Tree.

FossilOrigin-Name: 444084fd620fc3f45cfb87b83f532d76bd2744e7

diff --git a/ext/rtree/rtree.c b/ext/rtree/rtree.c
index 53414af7ee618c9d54d607189c8da6c2a168afae..620c15828887882ae4ed78099a0ce58843e0b1ad 100644 (file)
--- a/ext/rtree/rtree.c
+++ b/ext/rtree/rtree.c
@@ -243,6 +243,7 @@ struct RtreeCursor {
 union RtreeCoord {
   RtreeValue f;      /* Floating point value */
   int i;             /* Integer value */
+  u32 u;             /* Unsigned for byte-order conversions */
 };
 
 /*
@@ -902,19 +903,32 @@ static int rtreeEof(sqlite3_vtab_cursor *cur){
 }
 
 /*
-** Convert raw bits from the on-disk RTree record into a coordinate value
-** The on-disk record stores integer coordinates if eInt is true and it
-** stores 32-bit floating point records if eInt is false.  a[] is the four
+** Convert raw bits from the on-disk RTree record into a coordinate value.
+** The on-disk format is big-endian and needs to be converted for little-endian
+** platforms.  The on-disk record stores integer coordinates if eInt is true
+** and it stores 32-bit floating point records if eInt is false.  a[] is the four
 ** bytes of the on-disk record to be decoded.  Store the results in "r".
+**
+** The first version of this macro is fast on x86, x86_64 and ARM, all of which
+** are little-endian.  The second version of this macro is cross-platform but
+** takes twice as long, according to valgrind on linux x64.
 */
+#if defined(__x86) || defined(__x86_64) || defined(__arm__) || defined(_MSC_VER)
 #define RTREE_DECODE_COORD(eInt, a, r) {                        \
-    u32 x;           /* Raw bits of the coordinate value */     \
     RtreeCoord c;    /* Coordinate decoded */                   \
-    x = ((u32)a[0]<<24) + ((u32)a[1]<<16)                       \
+    memcpy(&c.u,a,4);                                           \
+    c.u = ((c.u>>24)&0xff)|((c.u>>8)&0xff00)|                   \
+          ((c.u&0xff)<<24)|((c.u&0xff00)<<8);                   \
+    r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \
+}
+#else
+#define RTREE_DECODE_COORD(eInt, a, r) {                        \
+    RtreeCoord c;    /* Coordinate decoded */                   \
+    c.u = ((u32)a[0]<<24) + ((u32)a[1]<<16)                     \
            +((u32)a[2]<<8) + a[3];                              \
-    c.i = *(int*)&x;                                            \
     r = eInt ? (sqlite3_rtree_dbl)c.i : (sqlite3_rtree_dbl)c.f; \
 }
+#endif
    
 
 /*

diff --git a/manifest b/manifest
index 6eb3fbb2651f6d220c4b9333a46373b22fa7c912..f00e6668a6283bd6d5fb7b117df4f3eeb5aa826f 100644 (file)
--- a/manifest
+++ b/manifest
@@ -1,5 +1,5 @@
-C More\stest\scases\swith\svery\slong\spriority\squeues.
-D 2014-04-17T15:34:58.372
+C Performance\soptimization\son\sbyte-swapping\sin\sR-Tree.
+D 2014-04-17T23:23:29.676
 F Makefile.arm-wince-mingw32ce-gcc d6df77f1f48d690bd73162294bbba7f59507c72f
 F Makefile.in e4ee6d36cdf6136aee0158675a3b24dd3bf31a5a
 F Makefile.linux-gcc 91d710bdc4998cb015f39edf3cb314ec4f4d7e23
@@ -120,7 +120,7 @@ F ext/misc/vfslog.c fe40fab5c077a40477f7e5eba994309ecac6cc95
 F ext/misc/vtshim.c babb0dc2bf116029e3e7c9a618b8a1377045303e
 F ext/misc/wholenumber.c 784b12543d60702ebdd47da936e278aa03076212
 F ext/rtree/README 6315c0d73ebf0ec40dedb5aa0e942bc8b54e3761
-F ext/rtree/rtree.c 6a47918e44697dd32f5bba8a79d3490e56bd76c9
+F ext/rtree/rtree.c 5cf5ae21ca1742be863f025ce64e4262fc6def4c
 F ext/rtree/rtree.h 834dbcb82dc85b2481cde6a07cdadfddc99e9b9e
 F ext/rtree/rtree1.test cf679265ecafff494a768ac9c2f43a70915a6290
 F ext/rtree/rtree2.test acbb3a4ce0f4fbc2c304d2b4b784cfa161856bba
@@ -1176,7 +1176,7 @@ F tool/vdbe_profile.tcl 67746953071a9f8f2f668b73fe899074e2c6d8c1
 F tool/warnings-clang.sh f6aa929dc20ef1f856af04a730772f59283631d4
 F tool/warnings.sh d1a6de74685f360ab718efda6265994b99bbea01
 F tool/win/sqlite.vsix 030f3eeaf2cb811a3692ab9c14d021a75ce41fff
-P 1ccaaed6b516ec2ce953c1b31025a82ba76d00e7
-R 449eb9a5ce303ca34184a4306bd9d328
+P 71692aa97c78676f0ba80eaeec0ad9ac225f4427
+R f1ebd50a7930430ccdb2dd222ebc5021
 U drh
-Z 39039e877d4cb95ab625ec95bd5cc650
+Z 8e204d3bb697cd291df3586d6cea66ab

diff --git a/manifest.uuid b/manifest.uuid
index f923c4ba43c5cf6eba7f8363948395ea024b4649..41369a2d985fccb0ecb23b59b3ec9383f85a9be9 100644 (file)
--- a/manifest.uuid
+++ b/manifest.uuid
@@ -1 +1 @@
-71692aa97c78676f0ba80eaeec0ad9ac225f4427
\ No newline at end of file
+444084fd620fc3f45cfb87b83f532d76bd2744e7
\ No newline at end of file