o changed name, OSD_ATOMIC_INLINE to EPICS_ATOMIC_INLINE
o changed supported data types, unsigned removed and int added (per reveiw at codeathon)
o added add/subtract functions (per reveiw at codeathon)
o now presuming that __sync_synchronize available all gcc 4 mingw does not provide windows mem barrier)
o consolodated on one implemention for Microsoft invarient of cmplr intrinisic or win32 by using macros to config a shared header file
o improved doc in epicsAtomic.h
o added overloaded c++ interface in namespace epics :: atomic to epicsAtomic.h
o added epicsAtomicReadMemoryBarrier and epicsAtomicWriteMemoryBarrier interface to epicsAtomic.h
o changed the implementation so that each of the functions can be individually specified for a particular compiler, os, or in the generic implementation (this is accomplished with macros)
o modified the functional and performance test so that they are based on templates so we can easily support new data types
o modified performance tests to repeat function calls and measure performance using a template
o removed function
epicsAtomicTestAndSetUIntT
o added new functions
epicsAtomicSetPtrT
epicsAtomicGetPtrT
epicsAtomicCmpAndSwapUIntT
epicsAtomicCmpAndSwapPtrT
o changed msvc intrinsics to define memory fence
o fixed mutex synchronized version so that its slow, but correct if the c++ compiler doesnt synchronized local scope static initialization
o changed most of the set/get methods to use memory barriers instead of some other primitive
o added additional tests
o removed -march=i586 from CONFIG.linux-x86.linux-x86
o fixed compile errors in epicsAtomicOSD.h (I didnt have the proper version of solaris for initial testing here)
o added CMPLR_SRC_DIRS to ALL_SRC_DIRS, and added compiler/$CMPLR_CLASS) to INSTALL_INCLUDES in CONFIG_COMMON
o added additional compiler specific rule deciding where things are installed to in CONFIG_COMMON
o added compiler specific include install to RULES_BUILD
o changed configure/os/CONFIG.linux-x86.linux-x86 to specify minimum arch is 586 (pentium circa 1995)
o changed configure/os/CONFIG.win32-x86-cygwin.win32-x86-cygwin to specify minimum arch is 586 (pentium circa 1995)
o added epicsAtoimic facility to libCom/Makefile
o added testing and performance measurement to libCom/test/Makefile
* fix libCom exports for things which used to be built differently
* logserver needs winsock
* errlog prefix: fix warnings on win32
* fix dll imports in ioc/db
Refactored logClientSend in libCom/logClient/logClient.c; took the code between the mutex operations and
moved it to a private method - sendLogMessageinChunks. Call this method once for the prefix (if it exists) and once
for the actual message.
Added ioCsh registration code into src/libCom/iocsh/libComRegister.c registering a command called
"iocLogPrefix" that sets this prefix.
Unit tested with and without prefixes.
Performance tested with and without prefixes - without prefix is approx the same. With prefix is about
twice the time (reflecting the two calls to sendLogMessageinChunks I think)
o refactored compilerDependencies.h into code that is compiler specific (one libCom/osi/compiler/xxx/compilerDependent.h for each comiler) and code that isnt libCom/osi/compilerDependencies.h (this is a new location and is no longer in libCom/misc).
This symbol is required to be visible on non-PowerPC (68k) vxWorks
systems for devInterruptInUseVME() to recognize interrupt vectors
that the devDisconnectInterruptVME() routine has marked as not used.
A rare race during shutdown. The contenders are the log thread
coming out of its loop and calling errlogCleanup(), and the
exitHandler signaling waitForWork.
This solution is to move cleanup completely into exitHandler,
which already waits for the log thread to exit.
