From 3806c95073f2453944dd8fbcb6a3a11f5f6c15d1 Mon Sep 17 00:00:00 2001
From: skywind3000 <skywind3000@163.com>
Date: Fri, 3 Jun 2016 20:28:21 +0800
Subject: [PATCH] improve performance & optimize fast-ack

---
 ikcp.c   | 45 ++++++++++++++++++++++++++++++++++++++-------
 test.cpp | 12 +++++++-----
 test.h   |  8 ++++++++
 3 files changed, 53 insertions(+), 12 deletions(-)

diff --git a/ikcp.c b/ikcp.c
index 1335ddf..2824cd8 100644
--- a/ikcp.c
+++ b/ikcp.c
@@ -541,15 +541,14 @@ static void ikcp_parse_ack(ikcpcb *kcp, IUINT32 sn)
 			kcp->nsnd_buf--;
 			break;
 		}
-		else {
-			seg->fastack++;
+		if (_itimediff(sn, seg->sn) < 0) {
+			break;
 		}
 	}
 }
 
 static void ikcp_parse_una(ikcpcb *kcp, IUINT32 una)
 {
-#if 1
 	struct IQUEUEHEAD *p, *next;
 	for (p = kcp->snd_buf.next; p != &kcp->snd_buf; p = next) {
 		IKCPSEG *seg = iqueue_entry(p, IKCPSEG, node);
@@ -562,7 +561,25 @@ static void ikcp_parse_una(ikcpcb *kcp, IUINT32 una)
 			break;
 		}
 	}
-#endif
+}
+
+static void ikcp_parse_fastack(ikcpcb *kcp, IUINT32 sn)
+{
+	struct IQUEUEHEAD *p, *next;
+
+	if (_itimediff(sn, kcp->snd_una) < 0 || _itimediff(sn, kcp->snd_nxt) >= 0)
+		return;
+
+	for (p = kcp->snd_buf.next; p != &kcp->snd_buf; p = next) {
+		IKCPSEG *seg = iqueue_entry(p, IKCPSEG, node);
+		next = p->next;
+		if (_itimediff(sn, seg->sn) < 0) {
+			break;
+		}
+		else if (sn != seg->sn) {
+			seg->fastack++;
+		}
+	}
 }
 
 
@@ -684,6 +701,8 @@ void ikcp_parse_data(ikcpcb *kcp, IKCPSEG *newseg)
 int ikcp_input(ikcpcb *kcp, const char *data, long size)
 {
 	IUINT32 una = kcp->snd_una;
+	IUINT32 maxack = 0;
+	int flag = 0;
 
 	if (ikcp_canlog(kcp, IKCP_LOG_INPUT)) {
 		ikcp_log(kcp, IKCP_LOG_INPUT, "[RI] %d bytes", size);
@@ -728,6 +747,14 @@ int ikcp_input(ikcpcb *kcp, const char *data, long size)
 			}
 			ikcp_parse_ack(kcp, sn);
 			ikcp_shrink_buf(kcp);
+			if (flag == 0) {
+				flag = 1;
+				maxack = sn;
+			}	else {
+				if (_itimediff(sn, maxack) > 0) {
+					maxack = sn;
+				}
+			}
 			if (ikcp_canlog(kcp, IKCP_LOG_IN_ACK)) {
 				ikcp_log(kcp, IKCP_LOG_IN_DATA, 
 					"input ack: sn=%lu rtt=%ld rto=%ld", sn, 
@@ -784,6 +811,10 @@ int ikcp_input(ikcpcb *kcp, const char *data, long size)
 		size -= len;
 	}
 
+	if (flag != 0) {
+		ikcp_parse_fastack(kcp, maxack);
+	}
+
 	if (_itimediff(kcp->snd_una, una) > 0) {
 		if (kcp->cwnd < kcp->rmt_wnd) {
 			IUINT32 mss = kcp->mss;
@@ -865,7 +896,7 @@ void ikcp_flush(ikcpcb *kcp)
 	count = kcp->ackcount;
 	for (i = 0; i < count; i++) {
 		size = (int)(ptr - buffer);
-		if (size + IKCP_OVERHEAD > (int)kcp->mtu) {
+		if (size + (int)IKCP_OVERHEAD > (int)kcp->mtu) {
 			ikcp_output(kcp, buffer, size);
 			ptr = buffer;
 		}
@@ -901,7 +932,7 @@ void ikcp_flush(ikcpcb *kcp)
 	if (kcp->probe & IKCP_ASK_SEND) {
 		seg.cmd = IKCP_CMD_WASK;
 		size = (int)(ptr - buffer);
-		if (size + IKCP_OVERHEAD > (int)kcp->mtu) {
+		if (size + (int)IKCP_OVERHEAD > (int)kcp->mtu) {
 			ikcp_output(kcp, buffer, size);
 			ptr = buffer;
 		}
@@ -912,7 +943,7 @@ void ikcp_flush(ikcpcb *kcp)
 	if (kcp->probe & IKCP_ASK_TELL) {
 		seg.cmd = IKCP_CMD_WINS;
 		size = (int)(ptr - buffer);
-		if (size + IKCP_OVERHEAD > (int)kcp->mtu) {
+		if (size + (int)IKCP_OVERHEAD > (int)kcp->mtu) {
 			ikcp_output(kcp, buffer, size);
 			ptr = buffer;
 		}
diff --git a/test.cpp b/test.cpp
index 4736055..b0fdb98 100644
--- a/test.cpp
+++ b/test.cpp
@@ -71,6 +71,8 @@ void test(int mode)
 		// 第五个参数 为是否禁用常规流控，这里禁止
 		ikcp_nodelay(kcp1, 1, 10, 2, 1);
 		ikcp_nodelay(kcp2, 1, 10, 2, 1);
+		kcp1->rx_minrto = 10;
+		kcp1->fastresend = 1;
 	}
 
 
@@ -87,8 +89,8 @@ void test(int mode)
 
 		// 每隔 20ms，kcp1发送数据
 		for (; current >= slap; slap += 20) {
-			*(IUINT32*)(buffer + 0) = index++;
-			*(IUINT32*)(buffer + 4) = current;
+			((IUINT32*)buffer)[0] = index++;
+			((IUINT32*)buffer)[1] = current;
 
 			// 发送上层协议包
 			ikcp_send(kcp1, buffer, 8);
@@ -151,15 +153,15 @@ void test(int mode)
 
 	const char *names[3] = { "default", "normal", "fast" };
 	printf("%s mode result (%dms):\n", names[mode], (int)ts1);
-	printf("avgrtt=%d maxrtt=%d\n", (int)(sumrtt / count), maxrtt);
+	printf("avgrtt=%d maxrtt=%d tx=%d\n", (int)(sumrtt / count), (int)maxrtt, (int)vnet->tx1);
 	printf("press enter to next ...\n");
 	char ch; scanf("%c", &ch);
 }
 
 int main()
 {
-	test(0);	// 默认模式，类似 TCP：正常模式，无快速重传，常规流控
-	test(1);	// 普通模式，关闭流控等
+//test(0);	// 默认模式，类似 TCP：正常模式，无快速重传，常规流控
+//test(1);	// 普通模式，关闭流控等
 	test(2);	// 快速模式，所有开关都打开，且关闭流控
 	return 0;
 }
diff --git a/test.h b/test.h
index 7db5f98..1dd4319 100644
--- a/test.h
+++ b/test.h
@@ -146,6 +146,7 @@ protected:
 class LatencySimulator
 {
 public:
+
 	virtual ~LatencySimulator() {
 		clear();
 	}
@@ -160,6 +161,7 @@ public:
 		this->rttmin = rttmin / 2;
 		this->rttmax = rttmax / 2;
 		this->nmax = nmax;
+		tx1 = tx2 = 0;
 	}
 
 	// 清除数据
@@ -179,9 +181,11 @@ public:
 	// peer - 端点0/1，从0发送，从1接收；从1发送从0接收
 	void send(int peer, const void *data, int size) {
 		if (peer == 0) {
+			tx1++;
 			if (r12.random() < lostrate) return;
 			if ((int)p12.size() >= nmax) return;
 		}	else {
+			tx2++;
 			if (r21.random() < lostrate) return;
 			if ((int)p21.size() >= nmax) return;
 		}
@@ -222,6 +226,10 @@ public:
 		return maxsize;
 	}
 
+public:
+	int tx1;
+	int tx2;
+
 protected:
 	IUINT32 current;
 	int lostrate;