Move prepareClientToWrite out of loop for lrange command to reduce the redundant call. (#860)

## Description 
When I explore the cycles distributions for `lrange` test (
`valkey-benchmark -p 9001 -t lrange -d 100 -r 1000000 -n 1000000 -c 50
--threads 4`). I found the `prepareClientToWrite` and
`clientHasPendingReplies` could be reduced to single call outside
instead of called in a loop, ideally we can gain 3% performance. The
corresponding `LRANG_100`, `LRANG_300`, `LRANGE_500`, `LRANGE_600` have
~2% - 3% performance boost, the benchmark test prove it helps.

This patch try to move the `prepareClientToWrite` and its child
`clientHasPendingReplies` out of the loop to reduce the function
overhead.

---------

Signed-off-by: Lipeng Zhu <lipeng.zhu@intel.com>

src/networking.c

@@ -910,6 +910,16 @@ void setDeferredPushLen(client *c, void *node, long length) {
     setDeferredAggregateLen(c, node, length, '>');
 }
 
+/* Prepare a client for future writes. This is used so that we can
+ * skip a large number of calls to prepareClientToWrite when
+ * a command produces a lot of discrete elements in its output. */
+writePreparedClient *prepareClientForFutureWrites(client *c) {
+    if (prepareClientToWrite(c) == C_OK) {
+        return (writePreparedClient *)c;
+    }
+    return NULL;
+}
+
 /* Add a double as a bulk reply */
 void addReplyDouble(client *c, double d) {
     if (c->resp == 3) {
@@ -1026,6 +1036,11 @@ void addReplyArrayLen(client *c, long length) {
     addReplyAggregateLen(c, length, '*');
 }
 
+void addWritePreparedReplyArrayLen(writePreparedClient *c, long length) {
+    serverAssert(length >= 0);
+    _addReplyLongLongWithPrefix(c, length, '*');
+}
+
 void addReplyMapLen(client *c, long length) {
     int prefix = c->resp == 2 ? '*' : '%';
     if (c->resp == 2) length *= 2;
@@ -1098,6 +1113,12 @@ void addReplyBulkCBuffer(client *c, const void *p, size_t len) {
     _addReplyToBufferOrList(c, "\r\n", 2);
 }
 
+void addWritePreparedReplyBulkCBuffer(writePreparedClient *c, const void *p, size_t len) {
+    _addReplyLongLongWithPrefix(c, len, '$');
+    _addReplyToBufferOrList(c, p, len);
+    _addReplyToBufferOrList(c, "\r\n", 2);
+}
+
 /* Add sds to reply (takes ownership of sds and frees it) */
 void addReplyBulkSds(client *c, sds s) {
     if (prepareClientToWrite(c) != C_OK) {
@@ -1136,6 +1157,14 @@ void addReplyBulkLongLong(client *c, long long ll) {
     addReplyBulkCBuffer(c, buf, len);
 }
 
+void addWritePreparedReplyBulkLongLong(writePreparedClient *c, long long ll) {
+    char buf[64];
+    int len;
+
+    len = ll2string(buf, 64, ll);
+    addWritePreparedReplyBulkCBuffer(c, buf, len);
+}
+
 /* Reply with a verbatim type having the specified extension.
  *
  * The 'ext' is the "extension" of the file, actually just a three

src/server.h

@@ -1390,6 +1390,12 @@ typedef struct client {
         net_output_bytes_curr_cmd; /* Total network output bytes sent to this client, by the current command. */
 } client;
 
+/* When a command generates a lot of discrete elements to the client output buffer, it is much faster to
+ * skip certain types of initialization. This type is used to indicate a client that has been initialized
+ * and can be used with addWritePreparedReply* functions. A client can be cast into this type with
+ * prepareClientForFutureWrites(client *c). */
+typedef client writePreparedClient;
+
 /* ACL information */
 typedef struct aclInfo {
     long long user_auth_failures;       /* Auth failure counts on user level */
@@ -2779,6 +2785,7 @@ int processInputBuffer(client *c);
 void acceptCommonHandler(connection *conn, struct ClientFlags flags, char *ip);
 void readQueryFromClient(connection *conn);
 int prepareClientToWrite(client *c);
+writePreparedClient *prepareClientForFutureWrites(client *c);
 void addReplyNull(client *c);
 void addReplyNullArray(client *c);
 void addReplyBool(client *c, int b);
@@ -2788,7 +2795,9 @@ void AddReplyFromClient(client *c, client *src);
 void addReplyBulk(client *c, robj *obj);
 void addReplyBulkCString(client *c, const char *s);
 void addReplyBulkCBuffer(client *c, const void *p, size_t len);
+void addWritePreparedReplyBulkCBuffer(writePreparedClient *c, const void *p, size_t len);
 void addReplyBulkLongLong(client *c, long long ll);
+void addWritePreparedReplyBulkLongLong(writePreparedClient *c, long long ll);
 void addReply(client *c, robj *obj);
 void addReplyStatusLength(client *c, const char *s, size_t len);
 void addReplySds(client *c, sds s);
@@ -2810,6 +2819,7 @@ void addReplyBigNum(client *c, const char *num, size_t len);
 void addReplyHumanLongDouble(client *c, long double d);
 void addReplyLongLong(client *c, long long ll);
 void addReplyArrayLen(client *c, long length);
+void addWritePreparedReplyArrayLen(writePreparedClient *c, long length);
 void addReplyMapLen(client *c, long length);
 void addReplySetLen(client *c, long length);
 void addReplyAttributeLen(client *c, long length);

src/t_list.c

@@ -650,8 +650,10 @@ void listPopRangeAndReplyWithKey(client *c, robj *o, robj *key, int where, long
  * Note that the purpose is to make the methods small so that the
  * code in the loop can be inlined better to improve performance. */
 void addListQuicklistRangeReply(client *c, robj *o, int from, int rangelen, int reverse) {
+    writePreparedClient *wpc = prepareClientForFutureWrites(c);
+    if (!wpc) return;
     /* Return the result in form of a multi-bulk reply */
-    addReplyArrayLen(c, rangelen);
+    addWritePreparedReplyArrayLen(wpc, rangelen);
 
     int direction = reverse ? AL_START_TAIL : AL_START_HEAD;
     quicklistIter *iter = quicklistGetIteratorAtIdx(o->ptr, direction, from);
@@ -659,9 +661,9 @@ void addListQuicklistRangeReply(client *c, robj *o, int from, int rangelen, int
         quicklistEntry qe;
         serverAssert(quicklistNext(iter, &qe)); /* fail on corrupt data */
         if (qe.value) {
-            addReplyBulkCBuffer(c, qe.value, qe.sz);
+            addWritePreparedReplyBulkCBuffer(wpc, qe.value, qe.sz);
         } else {
-            addReplyBulkLongLong(c, qe.longval);
+            addWritePreparedReplyBulkLongLong(wpc, qe.longval);
         }
     }
     quicklistReleaseIterator(iter);
@@ -671,21 +673,22 @@ void addListQuicklistRangeReply(client *c, robj *o, int from, int rangelen, int
  * Note that the purpose is to make the methods small so that the
  * code in the loop can be inlined better to improve performance. */
 void addListListpackRangeReply(client *c, robj *o, int from, int rangelen, int reverse) {
+    writePreparedClient *wpc = prepareClientForFutureWrites(c);
+    if (!wpc) return;
+    /* Return the result in form of a multi-bulk reply */
+    addWritePreparedReplyArrayLen(wpc, rangelen);
     unsigned char *p = lpSeek(o->ptr, from);
     unsigned char *vstr;
     unsigned int vlen;
     long long lval;
 
-    /* Return the result in form of a multi-bulk reply */
-    addReplyArrayLen(c, rangelen);
-
     while (rangelen--) {
         serverAssert(p); /* fail on corrupt data */
         vstr = lpGetValue(p, &vlen, &lval);
         if (vstr) {
-            addReplyBulkCBuffer(c, vstr, vlen);
+            addWritePreparedReplyBulkCBuffer(wpc, vstr, vlen);
         } else {
-            addReplyBulkLongLong(c, lval);
+            addWritePreparedReplyBulkLongLong(wpc, lval);
         }
         p = reverse ? lpPrev(o->ptr, p) : lpNext(o->ptr, p);
     }