janwas 20d038efda profiling indicates memcpy is a bottleneck, so globally replace all of them with memcpy2 (hand-optimized). depending on transfer size, this is 10%..300% faster. also made memcpy2 into a macro instead of thunk function because vc7 was still generating a jump. ...

This was SVN commit r3057.

2005-10-30 16:19:20 +00:00

..

AllNetMessages.h

…

Client.cpp

…

Client.h

…

JSEvents.h

…

NetLog.cpp

…

NetLog.h

…

NetMessage.cpp

…

NetMessage.h

…

Network.cpp

…

Network.h

…

NetworkInternal.h

…

NMTCreator.h

…

Serialization.h

…

Server.cpp

…

Server.h

…

ServerSession.cpp

…

ServerSession.h

…

ServerSocket.cpp

…

Session.cpp

…

Session.h

…

SessionManager.cpp

…

SessionManager.h

…

SocketBase.cpp

profiling indicates memcpy is a bottleneck, so globally replace all of them with memcpy2 (hand-optimized). depending on transfer size, this is 10%..300% faster. also made memcpy2 into a macro instead of thunk function because vc7 was still generating a jump.

2005-10-30 16:19:20 +00:00

SocketBase.h

…

StreamSocket.cpp

…

StreamSocket.h

…

StringConverters.h

…