changelog

Parallel Processing, Important document for learning about high-performance computing.

	* pthread.dsw (version.rc): Added to MSVS workspace.

2004-11-20  Boudewijn Dekker  <b.dekker at ellipsis.nl>

	* pthread_getspecific.c (pthread_getspecific): Check for
	invalid (NULL) key argument.

2004-11-19  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* config.h (PTW32_THREAD_ID_REUSE_INCREMENT): Added to allow
	building the library for either unique thread IDs like Solaris
	or non-unique thread IDs like Linux; allows application developers
	to override the library's default insensitivity to some apps
	that may not be strictly POSIX compliant.
	* version.rc: New resource module to encode version information
	within the DLL.
	* pthread.h: Added PTW32_VERSION* defines and grouped sections
	required by resource compiler together; bulk of file is skipped
	if RC_INVOKED. Defined some error numbers and other names for
	Borland compiler.

2004-11-02  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_cond_wait.c (ptw32_cond_wait_cleanup): Lock CV mutex at
	start of cleanup handler rather than at the end.
	* implement.h (PTW32_THREAD_REUSE_EMPTY): Renamed from *_BOTTOM.
	(ptw32_threadReuseBottom): New global variable.
	* global.c (ptw32_threadReuseBottom): Declare new variable.
	* ptw32_reuse.c (ptw32_reuse): Change reuse LIFO stack to LILO queue
	to more evenly distribute use of reusable thread IDs; use renamed
	PTW32_THREAD_REUSE_EMPTY.
	* ptw32_processTerminate.c (ptw2_processTerminate): Use renamed
	PTW32_THREAD_REUSE_EMPTY.

2004-10-31  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* implement.h (PThreadState): Add new state value
	'PThreadStateCancelPending'.
	* pthread_testcancel.c (pthread_testcancel): Use new thread
	'PThreadStateCancelPending' state as short cut to avoid entering
	kernel space via WaitForSingleObject() call. This was obviated
	by user space sema acquisition in sem_wait() and sem_timedwait(),
	which are also cancelation points. A call to pthread_testcancel()
	was required, which introduced a kernel call, effectively nullifying
	any gains made by the user space sem acquisition checks.
	* pthread_cancel.c (pthread_cancel): Set new thread
	'PThreadStateCancelPending' state.

2004-10-29  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* implement.h (pthread_t): Renamed to ptw32_thread_t; struct contains
	all thread state.
	* pthread.h (ptw32_handle_t): New general purpose struct to serve
	as a handle for various reusable object IDs - currently only used
	by pthread_t; contains a pointer to ptw32_thread_t (thread state)
	and a general purpose uint for use as a reuse counter or flags etc.
	(pthread_t): typedef'ed to ptw32_handle_t; the uint is the reuse
	counter that allows the library to maintain unique POSIX thread IDs.
	When the pthread struct reuse stack was introduced, threads would
	often acquire an identical ID to a previously destroyed thread. The
	same was true for the pre-reuse stack library, by virtue of pthread_t
	being the address of the thread struct. The new pthread_t retains
	the reuse stack but provides virtually unique thread IDs.
	* sem_wait.c (ptw32_sem_wait_cleanup): New routine used for
	cancelation cleanup.
	* sem_timedwait.c (ptw32_sem_timedwait_cleanup): Likewise.

2004-10-22  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* sem_init.c (sem_init): Introduce a 'lock' element in order to
	replace the interlocked operations with conventional serialisation.
	This is needed in order to be able to atomically modify the sema
	value and perform Win32 sema release operations. Win32 semaphores are
	used instead of events in order to support efficient multiple posting.
	If the whole modify/release isn't atomic, a race between
	sem_timedwait() and sem_post() could result in a release when there is
	no waiting semaphore, which would cause too many threads to proceed.
	* sem_wait.c (sem_wait): Use new 'lock'element.
	* sem_timedwait.c (sem_timedwait): Likewise.
	* sem_trywait.c (sem_trywait): Likewise.
	* sem_post.c (sem_post): Likewise.
	* sem_post_multiple.c (sem_post_multiple): Likewise.
	* sem_getvalue.c (sem_getvalue): Likewise.
	* ptw32_semwait.c (ptw32_semwait): Likewise.
	* sem_destroy.c (sem_destroy): Likewise; also tightened the conditions
	for semaphore destruction; in particular, a semaphore will not be
	destroyed if it has waiters.
	* sem_timedwait.c (sem_timedwait): Added cancel cleanup handler to
	restore sema value when cancelled.
	* sem_wait.c (sem_wait): Likewise.

2004-10-21  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_mutex_unlock.c (pthread_mutex_unlock): Must use PulseEvent()
	rather than SetEvent() to reset the event if there are no waiters.

2004-10-19  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* sem_init.c (sem_init): New semaphore model based on the same idea
	as mutexes, i.e. user space interlocked check to avoid 
	unnecessarily entering kernel space. Wraps the Win32 semaphore and
	keeps it's own counter. Although the motivation to do this has existed
	for a long time, credit goes to Alexander Terekhov for providing
	the logic. I have deviated slightly from AT's logic to add the waiters
	count, which has made the code more complicated by adding cancelation
	cleanup. This also appears to have broken the VCE (C++ EH) version of
	the library (the same problem as previously reported - see BUGS #2),
	only apparently not fixable using the usual workaround, nor by turning
	all optimisation off. The GCE version works fine, so it is presumed to
	be a bug in MSVC++ 6.0. The cancelation exception is thrown and caught
	correctly, but the cleanup class destructor is never called. The failing
	test is tests\semaphore4.c.
	* sem_wait.c (sem_wait): Implemented user space check model.
	* sem_post.c (sem_post): Likewise.
	* sem_trywait.c (sem_trywait): Likewise.
	* sem_timedwait.c (sem_timedwait): Likewise.
	* sem_post_multiple.c (sem_post_multiple): Likewise.
	* sem_getvalue.c (sem_getvalue): Likewise.
	* ptw32_semwait.c (ptw32_semwait): Likewise.
	* implement.h (sem_t_): Add counter element.

2004-10-15  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* implement.h (pthread_mutex_t_): Use an event in place of
	the POSIX semaphore.
	* pthread_mutex_init.c: Create the event; remove semaphore init.
	* pthread_mutex_destroy.c: Delete the event.
	* pthread_mutex_lock.c: Replace the semaphore wait with the event wait.
	* pthread_mutex_trylock.c: Likewise.
	* pthread_mutex_timedlock.c: Likewise.
	* pthread_mutex_unlock.c: Set the event.
	
2004-10-14  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_mutex_lock.c (pthread_mutex_lock): New algorithm using
	Terekhov's xchg based variation of Drepper's cmpxchg model.
	Theoretically, xchg uses fewer clock cycles than cmpxchg (using IA-32
	as a reference), however, in my opinion bus locking dominates the
	equation on smp systems, so the model with the least number of bus
	lock operations in the execution path should win, which is Terekhov's
	variant. On IA-32 uni-processor systems, it's faster to use the
	CMPXCHG instruction without locking the bus than to use the XCHG
	instruction, which always locks the bus. This makes the two variants
	equal for the non-contended lock (fast lane) execution path on up
	IA-32. Testing shows that the xchg variant is faster on up IA-32 as
	well if the test forces higher lock contention frequency, even though
	kernel calls should be dominating the times (on up IA-32, both
	variants used CMPXCHG instructions and neither locked the bus).
	* pthread_mutex_timedlock.c pthread_mutex_timedlock(): Similarly.
	* pthread_mutex_trylock.c (pthread_mutex_trylock): Similarly.
	* pthread_mutex_unlock.c (pthread_mutex_unlock): Similarly.
	* ptw32_InterlockedCompareExchange.c (ptw32_InterlockExchange): New
	function.
	(PTW32_INTERLOCKED_EXCHANGE): Sets up macro to use inlined
	ptw32_InterlockedExchange.
	* implement.h (PTW32_INTERLOCKED_EXCHANGE): Set default to
	InterlockedExchange().
	* Makefile: Building using /Ob2 so that asm sections within inline
	functions are inlined.

2004-10-08  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_mutex_destroy.c (pthread_mutex_destroy): Critical Section
	element is no longer required.
	* pthread_mutex_init.c (pthread_mutex_init): Likewise.
	* pthread_mutex_lock.c (pthread_mutex_lock): New algorithm following
	Drepper's paper at http://people.redhat.com/drepper/futex.pdf, but
	using the existing semaphore in place of the futex described in the
	paper. Idea suggested by Alexander Terekhov - see:
	http://sources.redhat.com/ml/pthreads-win32/2003/msg00108.html
	* pthread_mutex_timedlock.c pthread_mutex_timedlock(): Similarly.
	* pthread_mutex_trylock.c (pthread_mutex_trylock): Similarly.
	* pthread_mutex_unlock.c (pthread_mutex_unlock): Similarly.
	* pthread_barrier_wait.c (pthread_barrier_wait): Use inlined version
	of InterlockedCompareExchange() if possible - determined at
	build-time.
	* pthread_spin_destroy.c pthread_spin_destroy(): Likewise.
	* pthread_spin_lock.c pthread_spin_lock():Likewise.
	* pthread_spin_trylock.c (pthread_spin_trylock):Likewise.
	* pthread_spin_unlock.c (pthread_spin_unlock):Likewise.
	* ptw32_InterlockedCompareExchange.c: Sets up macro for inlined use.
	* implement.h (pthread_mutex_t_): Remove Critical Section element.
	(PTW32_INTERLOCKED_COMPARE_EXCHANGE): Set to default non-inlined
	version of InterlockedCompareExchange().
	* private.c: Include ptw32_InterlockedCompareExchange.c first for
	inlining.
	* GNUmakefile: Add commandline option to use inlined
	InterlockedCompareExchange().
	* Makefile: Likewise.

2004-09-27  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_mutex_lock.c (pthread_mutex_lock): Separate
	PTHREAD_MUTEX_NORMAL logic since we do not need to keep or check some
	state required by other mutex types; do not check mutex pointer arg
	for validity - leave this to the system since we are only checking
	for NULL pointers. This should improve speed of NORMAL mutexes and
	marginally improve speed of other type.
	* pthread_mutex_trylock.c (pthread_mutex_trylock): Likewise.
	* pthread_mutex_unlock.c (pthread_mutex_unlock): Likewise; also avoid
	entering the critical section for the no-waiters case, with approx.
	30% reduction in lock/unlock overhead for this case.
	* pthread_mutex_timedlock.c (pthread_mutex_timedlock): Likewise; also
	no longer keeps mutex if post-timeout second attempt succeeds - this
	will assist applications that wish to impose strict lock deadlines,
	rather than simply to escape from frozen locks.

2004-09-09  Tristan Savatier  <tristan at mpegtv.com>
	* pthread.h (struct pthread_once_t_): Qualify the 'done' element
	as 'volatile'.
	* pthread_once.c: Concerned about possible race condition,
	specifically on MPU systems re concurrent access to multibyte types.
	[Maintainer's note: the race condition is harmless on SPU systems
	and only a problem on MPU systems if concurrent access results in an
	exception (presumably generated by a hardware interrupt). There are
	other instances of similar harmless race conditions that have not
	been identified as issues.]

2004-09-09  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread.h: Declare additional types as volatile.

2004-08-27  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_barrier_wait.c (pthread_barrier_wait): Remove excessive code
	by substituting the internal non-cancelable version of sem_wait
	(ptw32_semwait).

2004-08-25  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_join.c (pthread_join): Rewrite and re-order the conditional
	tests in an attempt to improve efficiency and remove a race
	condition.

2004-08-23  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* create.c (pthread_create): Don't create a thread if the thread
	id pointer location (first arg) is inaccessible. A memory
	protection fault will result if the thread id arg isn't an accessible
	location. This is consistent with GNU/Linux but different to
	Solaris or MKS (and possibly others), which accept NULL as meaning
	'don't return the created thread's ID'. Applications that run
	using pthreads-win32 will run on all other POSIX threads
	implementations, at least w.r.t. this feature.

	It was decided not to copy the Solaris et al behaviour because,
	although it would have simplified some application porting (but only
	from Solaris to Windows), the feature is not technically necessary,
	and the alternative segfault behaviour helps avoid buggy application
	code.

2004-07-01  Anuj Goyal  <anuj.goyal at gmail.com>

	* builddmc.bat: New; Windows bat file to build the library.
	* config.h (__DMC__): Support for Digital Mars compiler.
	* create.c (__DMC__): Likewise.
	* pthread_exit.c (__DMC__): Likewise.
	* pthread_join.c (__DMC__): Likewise.
	* ptw32_threadDestroy.c (__DMC__): Likewise.
	* ptw32_threadStart.c (__DMC__): Likewise.
	* ptw32_throw.c (__DMC__): Likewise.

2004-06-29  Anuj Goyal  <anuj.goyal at gmail.com>

	* pthread.h (__DMC__): Initial support for Digital Mars compiler.

2004-06-29  Will Bryant  <will.bryant at ecosm.com>

	* README.Borland: New; description of Borland changes.
	* Bmakefile: New makefile for the Borland make utility.
	* ptw32_InterlockedCompareExchange.c:
	Add Borland compatible asm code.

2004-06-26  Jason Bard  <BardJA at Npt.NUWC.Navy.Mil>

	* pthread.h (HAVE_STRUCT_TIMESPEC): If undefined, define it
	to avoid timespec struct redefined errors elsewhere in an
	application.

2004-06-21  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread.h (PTHREAD_RECURSIVE_MUTEX_INITIALIZER): Mutex
	initialiser added for compatibility with Linux threads and
	others; currently not included in SUSV3.
	* pthread.h (PTHREAD_ERRORCHECK_MUTEX_INITIALIZER): Likewise.
	* pthread.h (PTHREAD_RECURSIVE_MUTEX_INITIALIZER_NP): Likewise.
	* pthread.h (PTHREAD_ERRORCHECK_MUTEX_INITIALIZER_NP): Likewise.

	* ptw32_mutex_check_need_init.c (ptw32_mutex_check_need_init): 
	Add new initialisers.

	* pthread_mutex_lock.c (pthread_mutex_lock): Check for new
	initialisers.
	* pthread_mutex_trylock.c (pthread_mutex_trylock): Likewise.
	* pthread_mutex_timedlock.c (pthread_mutex_timedlock): Likewise.
	* pthread_mutex_unlock.c (pthread_mutex_unlock): Likewise.
	* pthread_mutex_destroy.c (pthread_mutex_destroy): Likewise.

2004-05-20  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* README.NONPORTABLE: Document pthread_win32_test_features_np().
	* FAQ: Update various answers.

2004-05-19  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* Makefile: Don't define _WIN32_WINNT on compiler command line.
	* GNUmakefile: Likewise.