changelog

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

2004-05-16  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_cancel.c (pthread_cancel): Adapted to use auto-detected
	QueueUserAPCEx features at run-time.
	(ptw32_RegisterCancelation): Drop in replacement for QueueUserAPCEx()
	if it can't be used. Provides older style non-preemptive async
	cancelation.
	* pthread_win32_attach_detach_np.c (pthread_win32_attach_np):
	Auto-detect quserex.dll and the availability of alertdrv.sys;
	initialise and close on process attach/detach.
	* global.c (ptw32_register_cancelation): Pointer to either
	QueueUserAPCEx() or ptw32_RegisterCancelation() depending on
	availability. QueueUserAPCEx makes pre-emptive async cancelation
	possible.
	* implement.h: Add definitions and prototypes related to QueueUserAPC.

2004-05-16  Panagiotis E. Hadjidoukas <peh at hpclab.ceid.upatras.gr>

	* QueueUserAPCEx (separate contributed package): Provides preemptive
	APC feature.
	* pthread_cancel.c (pthread_cancel): Initial integration of
	QueueUserAPCEx into pthreads-win32 to provide true pre-emptive
	async cancelation of threads, including blocked threads.

2004-05-06  Makoto Kato  <raven at oldskool.jp>

	* pthread.h (DWORD_PTR): Define typedef for older MSVC.
	* pthread_cancel.c (AMD64): Add architecture specific Context register.
	* ptw32_getprocessors.c: Use correct types (DWORD_PTR) for mask
	variables.

2004-04-06  P. van Bruggen  <pietvb at newbridges.nl>

	* ptw32_threadDestroy.c: Destroy threadLock mutex to
	close a memory leak.

2004-02-13  Gustav Hallberg  <gustav at virtutech.com>

	* pthread_equal.c: Remove redundant equality logic.

2003-12-10  Philippe Di Cristo  <philipped at voicebox.com>

	* sem_timedwait.c (sem_timedwait): Fix timeout calculations.

2003-10-20  Alexander Terekhov  <TEREKHOV at de.ibm.com>

	* pthread_mutex_timedlock.c (ptw32_semwait): Move to individual module.
	* ptw32_semwait.c: New module.
	* pthread_cond_wait.c (ptw32_cond_wait_cleanup): Replace cancelable
	sem_wait() call with non-cancelable ptw32_semwait() call.
	* pthread.c (private.c): Re-order for inlining. GNU C warned that
	function ptw32_semwait() was defined 'inline' after it was called.
	* pthread_cond_signal.c (ptw32_cond_unblock): Likewise.
	* pthread_delay_np.c: Disable Watcom warning with comment.
	* *.c (process.h): Remove include from .c files. This is conditionally
	included by the common project include files.

2003-10-20  James Ewing  <james.ewing at sveasoft.com>

	* ptw32_getprocessors.c: Some Win32 environments don't have
	GetProcessAffinityMask(), so always return CPU count = 1 for them.
	* config.h (NEED_PROCESSOR_AFFINITY_MASK): Define for WinCE.
	
2003-10-15  Ross Johnson  <ross at callisto.canberra.edu.au>

	* Re-indented all .c files using default GNU style to remove assorted
	editor ugliness (used GNU indent utility in default style).

2003-10-15  Alex Blanco  <Alex.Blanco at motorola.com>

	* sem_init.c (sem_init): Would call CreateSemaphore even if the sema
	struct calloc failed; was not freeing calloced memory if either
	CreateSemaphore or CreateEvent failed.

2003-10-14  Ross Johnson  <ross at callisto.canberra.edu.au>

	* pthread.h: Add Watcom compiler compatibility. Esssentially just add
	the cdecl attribute to all exposed function prototypes so that Watcom
	generates function call code compatible with non-Watcom built libraries.
	By default, Watcom uses registers to pass function args if possible rather
	than pushing to stack.
	* semaphore.h: Likewise.
	* sched.h: Likewise.
	* pthread_cond_wait.c (ptw32_cond_wait_cleanup): Define with cdecl attribute
	for Watcom compatibility. This routine is called via pthread_cleanup_push so
	it had to match function arg definition.
	* Wmakefile: New makefile for Watcom builds.

2003-09-14  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_setschedparam.c (pthread_setschedparam): Attempt to map
	all priority levels between max and min (as returned by
	sched_get_priority_min/max) to reasonable Win32 priority levels - i.e.
	levels between THREAD_PRIORITY_LOWEST/IDLE to THREAD_PRIORITY_LOWEST and
	between THREAD_PRIORITY_HIGHEST/TIME_CRITICAL to THREAD_PRIORITY_HIGHEST
	while others remain unchanged; record specified thread priority level
	for return by pthread_getschedparam.

	Note that, previously, specified levels not matching Win32 priority levels
	would silently leave the current thread priority unaltered.

	* pthread_getschedparam.c (pthread_getschedparam): Return the priority
	level specified by the latest pthread_setschedparam or pthread_create rather
	than the actual running thread priority as returned by GetThreadPriority - as
	required by POSIX. I.e. temporary or adjusted actual priority levels are not
	returned by this routine.

	* pthread_create.c (pthread_create): For priority levels specified via
	pthread attributes, attempt to map all priority levels between max and
	min (as returned by sched_get_priority_min/max) to reasonable Win32
	priority levels; record priority level given via attributes, or
	inherited from parent thread, for later return by pthread_getschedparam.

	* ptw32_new.c (ptw32_new): Initialise pthread_t_ sched_priority element.

	* pthread_self.c (pthread_self): Set newly created implicit POSIX thread
	sched_priority to Win32 thread's current actual priority. Temporarily
	altered priorities can't be avoided in this case.

	* implement.h (struct pthread_t_): Add new sched_priority element.

2003-09-12  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* sched_get_priority_min.c (sched_get_priority_min): On error should return -1
	with errno set.
	* sched_get_priority_max.c (sched_get_priority_max): Likewise.

2003-09-03  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* w32_cancelableWait.c (ptw32_cancelable_wait): Allow cancelation
	of implicit POSIX threads as well.

2003-09-02  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_win32_attach_detach_np.c (pthread_win32_thread_detach_np):
	Add comment.

	* pthread_exit.c (pthread_exit): Fix to recycle the POSIX thread handle in
	addition to calling user TSD destructors. Move the implicit POSIX thread exit
	handling to ptw32_throw to centralise the logic.

	* ptw32_throw.c (ptw32_throw): Implicit POSIX threads have no point
	to jump or throw to, so cleanup and exit the thread here in this case. For
	processes using the C runtime, the exit code will be set to the POSIX
	reason for the throw (i.e. PTHREAD_CANCEL or the value given to pthread_exit).
	Note that pthread_exit() already had similar logic, which has been moved to
	here.

	* ptw32_threadDestroy.c (ptw32_threadDestroy): Don't close the Win32 handle
	of implicit POSIX threads - expect this to be done by Win32?

2003-09-01  Ross Johnson  <rpj at callisto.canberra.edu.au>

	* pthread_self.c (pthread_self): The newly aquired pthread_t must be
	assigned to the reuse stack, not freed, if the routine fails somehow.

2003-08-13  Ross Johnson  <rpj at ise.canberra.edu.au>

	* pthread_getschedparam.c (pthread_getschedparam): An invalid thread ID
	parameter was returning an incorrect error value; now uses a more exhaustive
	check for validity.

	* pthread_setschedparam.c (pthread_setschedparam): Likewise.

	* pthread_join.c (pthread_join): Now uses a more exhaustive
	check for validity.

	* pthread_detach.c (pthread_detach): Likewise.

	* pthread_cancel.c (pthread_cancel): Likewise.

	* ptw32_threadDestroy.c (ptw32_threadDestroy): pthread_t structs are
	never freed - push them onto a stack for reuse.

	* ptw32_new.c (ptw32_new): Check for reusable pthread_t before dynamically
	allocating new memory for the struct.

	* pthread_kill.c (pthread_kill): New file; new routine; takes only a zero
	signal arg so that applications can check the thread arg for validity; checks
	that the underlying Win32 thread HANDLE is valid.

	* pthread.h (pthread_kill): Add prototype.

	* ptw32_reuse.c (ptw32_threadReusePop): New file; new routine; pop a
	pthread_t off the reuse stack. pthread_t_ structs that have been destroyed, i.e.
	have exited detached or have been joined, are cleaned up and put onto a reuse
	stack. Consequently, thread IDs are no longer freed once calloced. The library
	will attempt to get a struct off this stack before asking the system to alloc
	new memory when creating threads. The stack is guarded by a global mutex.
	(ptw32_threadReusePush): New routine; push a pthread_t onto the reuse stack.

	* implement.h (ptw32_threadReusePush): Add new prototype.
	(ptw32_threadReusePop): Likewise.
	(pthread_t): Add new element.

	* ptw32_processTerminate.c (ptw32_processTerminate): Delete the thread
	reuse lock; free all thread ID structs on the thread reuse stack.

	* ptw32_processInitialize.c (ptw32_processInitialize): Initialise the
	thread reuse lock.

2003-07-19  Ross Johnson  <rpj at ise.canberra.edu.au>

	* GNUmakefile: modified to work under MsysDTK environment.
	* pthread_spin_lock.c (pthread_spin_lock): Check for NULL arg.
	* pthread_spin_unlock.c (pthread_spin_unlock): Likewise.
	* pthread_spin_trylock.c (pthread_spin_trylock): Likewise;
	fix incorrect pointer value if lock is dynamically initialised by
	this function.
	* sem_init.c (sem_init): Initialise sem_t value to quell compiler warning.
	* sem_destroy.c (sem_destroy): Likewise.
	* ptw32_threadStart.c (non-MSVC code sections): Include <exception> rather
	than old-style <new.h>; fix all std:: namespace entities such as
	std::terminate_handler instances and associated methods.
	* ptw32_callUserDestroyRoutines.c (non-MSVC code sections): Likewise.

2003-06-24  Piet van Bruggen  <pietvb at newbridges.nl>

	* pthread_spin_destroy.c (pthread_spin_destroy): Was not freeing the
	spinlock struct.

2003-06-22  Nicolas Barry  <boozai at yahoo.com>

	* pthread_mutex_destroy.c (pthread_mutex_destroy): When called
	with a recursive mutex that was locked by the current thread, the
	function was failing with a success return code.

2003-05-15  Steven Reddie  <Steven.Reddie at ca.com>

	* pthread_win32_attach_detach_np.c (pthread_win32_process_detach_np):
	NULLify ptw32_selfThreadKey after the thread is destroyed, otherwise
	destructors calling pthreads routines might resurrect it again, creating
	memory leaks. Call the underlying Win32 Tls routine directly rather than
	pthread_setspecific().
	(pthread_win32_thread_detach_np): Likewise.

2003-05-14  Viv  <vcotirlea at hotmail.com>

	* pthread.dsp: Change /MT compile flag to /MD.

2003-03-04  Alexander Terekhov  <TEREKHOV at de.ibm.com>

	* pthread_mutex_timedlock.c (pthread_mutex_timedlock): Fix failure to
	set ownership of mutex on second grab after abstime timeout.
	- bug reported by Robert Strycek <strycek at posam.sk>

2002-12-17  Thomas Pfaff  <tpfaff at gmx.net>

	* pthread_mutex_lock.c (ptw32_semwait): New static routine to provide
	a non-cancelable sem_wait() function. This is consistent with the
	way that pthread_mutex_timedlock.c does it.
	(pthread_mutex_lock): Use ptw32_semwait() instead of sem_wait().

2002-12-11  Thomas Pfaff  <tpfaff at gmx.net>

	* pthread_mutex_trylock.c: Should return EBUSY rather than EDEADLK.
	* pthread_mutex_destroy.c: Remove redundant ownership test (the
	trylock call does this for us); do not destroy a recursively locked
	mutex.

2002-09-20  Michael Johnson  <michaelj at maine.rr.com>

	* pthread_cond_destroy.c (pthread_cond_destroy): 
	When two different threads exist, and one is attempting to
	destroy a condition variable while the other is attempting to
	initialize a condition variable that was created with
	PTHREAD_COND_INITIALIZER, a deadlock can occur. Shrink
	the ptw32_cond_list_lock critical section to fix it.

2002-07-31  Ross Johnson  <rpj at special.ise.canberra.edu.au>

	* ptw32_threadStart.c (ptw32_threadStart): Thread cancelLock
	destruction moved to ptw32_threadDestroy().

	* ptw32_threadDestroy.c (ptw32_threadDestroy):  Destroy
	the thread's cancelLock. Moved here from ptw32_threadStart.c
	to cleanup implicit threads as well.

2002-07-30  Alexander Terekhov  <TEREKHOV at de.ibm.com>

	* pthread_cond_wait.c (ptw32_cond_wait_cleanup): 
	Remove code designed to avoid/prevent spurious wakeup
	problems. It is believed that the sem_timedwait() call
	is consuming a CV signal that it shouldn't and this is
	breaking the avoidance logic.

2002-07-30  Ross Johnson  <rpj at ise.canberra.edu.au>

	* sem_timedwait.c (sem_timedwait): Tighten checks for
	unreasonable abstime values - that would result in
	unexpected timeout values.

	* w32_CancelableWait.c (ptw32_cancelable_wait):
	Tighten up return value checking and add comments.


2002-06-08  Ross Johnson  <rpj at special.ise.canberra.edu.au>

	* sem_getvalue.c (sem_getvalue): Now returns a value for the
	NEED_SEM version (i.e. earlier versions of WinCE).


2002-06-04  Rob Fanner  <rfanner at stonethree.com>

	* sem_getvalue.c (sem_getvalue): The Johnson M. Hart
	approach didn't work - we are forced to take an
	intrusive approach. We try to decrement the sema
	and then immediately release it again to get the