Optimize sds free

Eliminates cache-miss on free of String. SdsHdrSize is retieved from
calculating %8 operation on sds pointer. Added padding in 2 Sds Headers
to have unique value of %8 operation.

src/object.c

@@ -281,6 +281,12 @@ robj *createModuleObject(moduleType *mt, void *value) {
     return createObject(OBJ_MODULE,mv);
 }
 
+void freeStringObjectOptim(robj *o) {
+    if (o->encoding == OBJ_ENCODING_RAW) {
+        sdsfreeOptim(o->ptr);
+    }
+}
+
 void freeStringObject(robj *o) {
     if (o->encoding == OBJ_ENCODING_RAW) {
         sdsfree(o->ptr);
@@ -356,7 +362,7 @@ void incrRefCount(robj *o) {
 static void _decrRefCount(robj *o, int on_dram) {
     if (o->refcount == 1) {
         switch(o->type) {
-        case OBJ_STRING: freeStringObject(o); break;
+        case OBJ_STRING: freeStringObjectOptim(o); break;
         case OBJ_LIST: freeListObject(o); break;
         case OBJ_SET: freeSetObject(o); break;
         case OBJ_ZSET: freeZsetObject(o); break;

src/sds.c

@@ -60,6 +60,24 @@ static inline int sdsHdrSize(char type) {
     return 0;
 }
 
+/* Returns size of sdsHdr by checking pointer alignment (expects 8 byte
+ * alignment). Optimizes retriving sdsHdr size by not refering to sds data */
+static inline int sdsHdrSizeOptim(char* s) {
+    switch((uintptr_t)s&7) {
+        case SDS_MOD8(5):
+            return sizeof(struct sdshdr5);
+        case SDS_MOD8(8):
+            return sizeof(struct sdshdr8);
+        case SDS_MOD8(16):
+            return sizeof(struct sdshdr16);
+        case SDS_MOD8(32):
+            return sizeof(struct sdshdr32);
+        case SDS_MOD8(64):
+            return sizeof(struct sdshdr64);
+    }
+    return 0;
+}
+
 static inline char sdsReqType(size_t string_size) {
     if (string_size < 1<<5)
         return SDS_TYPE_5;
@@ -185,6 +203,11 @@ void sdsfree(sds s) {
     s_free((char*)s-sdsHdrSize(s[-1]));
 }
 
+void sdsfreeOptim(sds s) {
+    if (s == NULL) return;
+    s_free((char*)s-sdsHdrSizeOptim(s));
+}
+
 /* Set the sds string length to the length as obtained with strlen(), so
  * considering as content only up to the first null term character.
  *

src/sds.h

@@ -64,12 +64,14 @@ struct __attribute__ ((__packed__)) sdshdr32 {
     uint32_t len; /* used */
     uint32_t alloc; /* excluding the header and null terminator */
     unsigned char flags; /* 3 lsb of type, 5 unused bits */
+    char padding; /* Added to have unique value for %8 operation*/
     char buf[];
 };
 struct __attribute__ ((__packed__)) sdshdr64 {
     uint64_t len; /* used */
     uint64_t alloc; /* excluding the header and null terminator */
     unsigned char flags; /* 3 lsb of type, 5 unused bits */
+    char padding[3]; /* Added to have unique value for %8 operation */
     char buf[];
 };
 
@@ -83,6 +85,7 @@ struct __attribute__ ((__packed__)) sdshdr64 {
 #define SDS_HDR_VAR(T,s) struct sdshdr##T *sh = (void*)((s)-(sizeof(struct sdshdr##T)));
 #define SDS_HDR(T,s) ((struct sdshdr##T *)((s)-(sizeof(struct sdshdr##T))))
 #define SDS_TYPE_5_LEN(f) ((f)>>SDS_TYPE_BITS)
+#define SDS_MOD8(T) (sizeof(struct sdshdr##T)&7)
 
 static inline size_t sdslen(const sds s) {
     unsigned char flags = s[-1];
@@ -221,6 +224,7 @@ sds sdsempty(void);
 sds sdsdramempty(void);
 sds sdsdup(const sds s);
 void sdsfree(sds s);
+void sdsfreeOptim(sds s);
 sds sdsgrowzero(sds s, size_t len);
 sds sdscatlen(sds s, const void *t, size_t len);
 sds sdscat(sds s, const char *t);

src/server.h

@@ -1724,6 +1724,7 @@ void incrRefCount(robj *o);
 robj *makeObjectShared(robj *o);
 robj *resetRefCount(robj *obj);
 void freeStringObject(robj *o);
+void freeStringObjectOptim(robj *o);
 void freeListObject(robj *o);
 void freeSetObject(robj *o);
 void freeZsetObject(robj *o);