Win: QUdpSocket: Use QVarLengthArray for retrieving size

Increased size of the peek buffer to 2048 from 1500 and now
uses QVarLengthArray with space for 10 stack-allocated WSABUF instances,
but still growing at the pace of 5.

In benchmarking (created for and included in this patch) this shows
better performance when retrieving the datagram size for larger
datagrams, and the same performance as before for smaller datagrams
(at the cost of 2048 - 1500 + 16 * 10 = 708 bytes extra stack space).

Benchmarks:

With changes:

********* Start testing of tst_QUdpSocket *********
Config: Using QtTest library 5.13.1, Qt 5.13.1 (x86_64-little_endian-llp64 shared (dynamic) release build; by MSVC 2019)
PASS   : tst_QUdpSocket::initTestCase()
PASS   : tst_QUdpSocket::pendingDatagramSize(52)
RESULT : tst_QUdpSocket::pendingDatagramSize():"52":
     0.0038 msecs per iteration (total: 63, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(1024)
RESULT : tst_QUdpSocket::pendingDatagramSize():"1024":
     0.0039 msecs per iteration (total: 64, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(2049)
RESULT : tst_QUdpSocket::pendingDatagramSize():"2049":
     0.0038 msecs per iteration (total: 63, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(4500)
RESULT : tst_QUdpSocket::pendingDatagramSize():"4500":
     0.0039 msecs per iteration (total: 64, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(4098)
RESULT : tst_QUdpSocket::pendingDatagramSize():"4098":
     0.0040 msecs per iteration (total: 66, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(8192)
RESULT : tst_QUdpSocket::pendingDatagramSize():"8192":
     0.0040 msecs per iteration (total: 67, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(12000)
RESULT : tst_QUdpSocket::pendingDatagramSize():"12000":
     0.010 msecs per iteration (total: 90, iterations: 8192)
PASS   : tst_QUdpSocket::pendingDatagramSize(25000)
RESULT : tst_QUdpSocket::pendingDatagramSize():"25000":
     0.021 msecs per iteration (total: 88, iterations: 4096)
PASS   : tst_QUdpSocket::pendingDatagramSize(32768)
RESULT : tst_QUdpSocket::pendingDatagramSize():"32768":
     0.033 msecs per iteration (total: 69, iterations: 2048)
PASS   : tst_QUdpSocket::pendingDatagramSize(64512)
RESULT : tst_QUdpSocket::pendingDatagramSize():"64512":
     0.088 msecs per iteration (total: 91, iterations: 1024)
PASS   : tst_QUdpSocket::cleanupTestCase()
Totals: 12 passed, 0 failed, 0 skipped, 0 blacklisted, 3090ms
********* Finished testing of tst_QUdpSocket *********

Without changes:

********* Start testing of tst_QUdpSocket *********
Config: Using QtTest library 5.13.1, Qt 5.13.1
(x86_64-little_endian-llp64 shared (dynamic) release build; by MSVC 2019)
PASS   : tst_QUdpSocket::initTestCase()
PASS   : tst_QUdpSocket::pendingDatagramSize(52)
RESULT : tst_QUdpSocket::pendingDatagramSize():"52":
     0.0039 msecs per iteration (total: 65, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(1024)
RESULT : tst_QUdpSocket::pendingDatagramSize():"1024":
     0.0039 msecs per iteration (total: 65, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(2049)
RESULT : tst_QUdpSocket::pendingDatagramSize():"2049":
     0.0040 msecs per iteration (total: 66, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(4500)
RESULT : tst_QUdpSocket::pendingDatagramSize():"4500":
     0.0040 msecs per iteration (total: 67, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(4098)
RESULT : tst_QUdpSocket::pendingDatagramSize():"4098":
     0.0040 msecs per iteration (total: 67, iterations: 16384)
PASS   : tst_QUdpSocket::pendingDatagramSize(8192)
RESULT : tst_QUdpSocket::pendingDatagramSize():"8192":
     0.010 msecs per iteration (total: 90, iterations: 8192)
PASS   : tst_QUdpSocket::pendingDatagramSize(12000)
RESULT : tst_QUdpSocket::pendingDatagramSize():"12000":
     0.010 msecs per iteration (total: 90, iterations: 8192)
PASS   : tst_QUdpSocket::pendingDatagramSize(25000)
RESULT : tst_QUdpSocket::pendingDatagramSize():"25000":
     0.033 msecs per iteration (total: 69, iterations: 2048)
PASS   : tst_QUdpSocket::pendingDatagramSize(32768)
RESULT : tst_QUdpSocket::pendingDatagramSize():"32768":
     0.0502 msecs per iteration (total: 103, iterations: 2048)
PASS   : tst_QUdpSocket::pendingDatagramSize(64512)
RESULT : tst_QUdpSocket::pendingDatagramSize():"64512":
     0.13 msecs per iteration (total: 70, iterations: 512)
PASS   : tst_QUdpSocket::cleanupTestCase()
Totals: 12 passed, 0 failed, 0 skipped, 0 blacklisted, 3192ms
********* Finished testing of tst_QUdpSocket *********

Fixes: QTBUG-78275
Change-Id: If86a226620244aa4e470600c6c1db4a7863b5617
Reviewed-by: Timur Pocheptsov <timur.pocheptsov@qt.io>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>

1 files changed, 9 insertions, 6 deletions

diff --git a/src/network/socket/qnativesocketengine_win.cpp b/src/network/socket/qnativesocketengine_win.cpp
index 9edabd7822..dd115c33dc 100644
--- a/src/network/socket/qnativesocketengine_win.cpp
+++ b/src/network/socket/qnativesocketengine_win.cpp
@@ -53,6 +53,8 @@
 #include <qnetworkinterface.h>
 #include <qoperatingsystemversion.h>
 
+#include <algorithm>
+
 //#define QNATIVESOCKETENGINE_DEBUG
 #if defined(QNATIVESOCKETENGINE_DEBUG)
 #   include <qstring.h>
@@ -1141,13 +1143,14 @@ qint64 QNativeSocketEnginePrivate::nativePendingDatagramSize() const
     qint64 ret = -1;
     int recvResult = 0;
     DWORD flags;
-    // We start at 1500 bytes (the MTU for Ethernet V2), which should catch
-    // almost all uses (effective MTU for UDP under IPv4 is 1468), except
-    // for localhost datagrams and those reassembled by the IP layer.
-    char udpMessagePeekBuffer[1500];
-    std::vector<WSABUF> buf;
+    // We increase the amount we peek by 2048 * 5 on each iteration
+    // Grabs most cases fast and early.
+    char udpMessagePeekBuffer[2048];
+    const int increments = 5;
+    QVarLengthArray<WSABUF, 10> buf;
     for (;;) {
-        buf.resize(buf.size() + 5, {sizeof(udpMessagePeekBuffer), udpMessagePeekBuffer});
+        buf.reserve(buf.size() + increments);
+        std::fill_n(std::back_inserter(buf), increments, WSABUF{sizeof(udpMessagePeekBuffer), udpMessagePeekBuffer});
 
         flags = MSG_PEEK;
         DWORD bytesRead = 0;