Tachyon cross-platform thread creation and management, atomic operations, and CPU feature query APIs. More...

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "threads.h"

Macros
#define	RT_USEINTCPUID 1

Typedefs
typedef void (	RTTHREAD_START_ROUTINE) (void *)
	Typedef to eliminate compiler warning caused by C/C++ linkage conflict. More...

Functions
int	rt_thread_numphysprocessors (void)
	If compiling on Linux, enable the GNU CPU affinity functions in both libc and the libpthreads. More...

int	rt_thread_numprocessors (void)
	number of processors available, subject to user override More...

int	rt_cpu_capability_flags (rt_cpu_caps_t *cpucaps)
	CPU optional instruction set capability flags. More...

int	rt_cpu_smt_depth (void)
	CPU logical processors (SMT depth / aka hyperthreading) More...

int *	rt_cpu_affinitylist (int *cpuaffinitycount)
	query CPU affinity of the calling process (if allowed by host system) More...

int	rt_thread_set_self_cpuaffinity (int cpu)
	set the CPU affinity of the current thread (if allowed by host system) More...

int	rt_thread_setconcurrency (int nthr)
	set the concurrency level and scheduling scope for threads More...

int	rt_thread_create (rt_thread_t thr, void fctn(void ), void arg)
	create a new child thread More...

int	rt_thread_join (rt_thread_t thr, void **stat)
	join (wait for completion of, and merge with) a thread More...

int	rt_mutex_init (rt_mutex_t *mp)
	initialize a mutex More...

int	rt_mutex_lock (rt_mutex_t *mp)
	lock a mutex More...

int	rt_mutex_trylock (rt_mutex_t *mp)
	try to lock a mutex More...

int	rt_mutex_spin_lock (rt_mutex_t *mp)
	lock a mutex by spinning only More...

int	rt_mutex_unlock (rt_mutex_t *mp)
	unlock a mutex More...

int	rt_mutex_destroy (rt_mutex_t *mp)
	destroy a mutex More...

int	rt_cond_init (rt_cond_t *cvp)
	initialize a condition variable More...

int	rt_cond_destroy (rt_cond_t *cvp)
	destroy a condition variable More...

int	rt_cond_wait (rt_cond_t cvp, rt_mutex_t mp)
	wait on a condition variable More...

int	rt_cond_signal (rt_cond_t *cvp)
	signal a condition variable, waking at least one thread More...

int	rt_cond_broadcast (rt_cond_t *cvp)
	signal a condition variable, waking all threads More...

int	rt_atomic_int_init (rt_atomic_int_t *atomp, int val)
	initialize an atomic int variable More...

int	rt_atomic_int_destroy (rt_atomic_int_t *atomp)
	destroy an atomic int variable More...

int	rt_atomic_int_set (rt_atomic_int_t *atomp, int val)
	set an atomic int variable More...

int	rt_atomic_int_get (rt_atomic_int_t *atomp)
	get an atomic int variable More...

int	rt_atomic_int_fetch_and_add (rt_atomic_int_t *atomp, int inc)
	fetch an atomic int and add inc to it, returning original value More...

int	rt_atomic_int_add_and_fetch (rt_atomic_int_t *atomp, int inc)
	fetch an atomic int and add inc to it, returning new value More...

int	rt_rwlock_init (rt_rwlock_t *rwp)
	initialize a reader/writer lock More...

int	rt_rwlock_readlock (rt_rwlock_t *rwp)
	set reader lock More...

int	rt_rwlock_writelock (rt_rwlock_t *rwp)
	set writer lock More...

int	rt_rwlock_unlock (rt_rwlock_t *rwp)
	unlock reader/writer lock More...

rt_barrier_t *	rt_thread_barrier_init (int n_clients)
	initialize counting barrier primitive More...

int	rt_thread_barrier_init_proc_shared (rt_barrier_t *barrier, int n_clients)
	When rendering in the CAVE we use a special synchronization mode so that shared memory mutexes and condition variables will work correctly when accessed from multiple processes. More...

void	rt_thread_barrier_destroy (rt_barrier_t *barrier)
	destroy counting barrier primitive More...

int	rt_thread_barrier (rt_barrier_t *barrier, int increment)
	synchronize on counting barrier primitive More...

int	rt_thread_run_barrier_init (rt_run_barrier_t *barrier, int n_clients)
	initialize thread pool barrier More...

void	rt_thread_run_barrier_destroy (rt_run_barrier_t *barrier)
	destroy thread pool barrier More...

int	rt_thread_run_barrier_poll (rt_run_barrier_t *barrier)
	non-blocking poll to see if peers are already at the barrier More...

int	rt_tilestack_init (rt_tilestack_t *s, int size)
	initialize task tile stack (to empty) More...

void	rt_tilestack_destroy (rt_tilestack_t *s)
	destroy task tile stack More...

int	rt_tilestack_compact (rt_tilestack_t *s)
	shrink memory buffers associated with task tile stack if possible More...

int	rt_tilestack_push (rt_tilestack_t s, const rt_tasktile_t t)
	push a task tile onto the stack More...

int	rt_tilestack_pop (rt_tilestack_t s, rt_tasktile_t t)
	pop a task tile off of the stack More...

int	rt_tilestack_popall (rt_tilestack_t *s)
	pop all of the task tiles off of the stack More...

int	rt_tilestack_empty (rt_tilestack_t *s)
	query if the task tile stack is empty or not More...

int	rt_shared_iterator_init (rt_shared_iterator_t *it)
	initialize a shared iterator More...

int	rt_shared_iterator_destroy (rt_shared_iterator_t *it)
	destroy a shared iterator More...

int	rt_shared_iterator_set (rt_shared_iterator_t it, rt_tasktile_t tile)
	set shared iterator parameters More...

int	rt_shared_iterator_next_tile (rt_shared_iterator_t it, int reqsize, rt_tasktile_t tile)
	iterate the shared iterator, over a requested half-open interval More...

int	rt_shared_iterator_setfatalerror (rt_shared_iterator_t *it)
	worker thread calls this to indicate a fatal error More...

int	rt_shared_iterator_getfatalerror (rt_shared_iterator_t *it)
	master thread calls this to query for fatal errors More...

rt_threadpool_t *	rt_threadpool_create (int workercount, int *devlist)
	create a thread pool with a specified number of worker threads More...

int	rt_threadpool_launch (rt_threadpool_t thrpool, void fctn(void ), void parms, int blocking)
	launch threads onto a new function, with associated parms More...

int	rt_threadpool_wait (rt_threadpool_t *thrpool)
	wait for all worker threads to complete their work More...

int	rt_threadpool_poll (rt_threadpool_t *thrpool)

int	rt_threadpool_destroy (rt_threadpool_t *thrpool)
	join all worker threads and free resources More...

int	rt_threadpool_get_workercount (rt_threadpool_t *thrpool)
	return the number of worker threads currently in the pool More...

int	rt_threadpool_worker_getid (void voiddata, int threadid, int *threadcount)
	worker thread can call this to get its ID and number of peers More...

int	rt_threadpool_worker_getdevid (void voiddata, int devid)
	worker thread can call this to get its CPU/GPU device ID More...

int	rt_threadpool_worker_setdevspeed (void *voiddata, float speed)
	worker thread calls this to set relative speed of this device as determined by the SM/core count and clock rate Note: this should only be called once, during the worker's device initialization process More...

int	rt_threadpool_worker_getdevspeed (void voiddata, float speed)
	worker thread calls this to get relative speed of this device as determined by the SM/core count and clock rate More...

int	rt_threadpool_worker_devscaletile (void voiddata, int tilesize)
	worker thread calls this to scale max tile size by worker speed as determined by the SM/core count and clock rate More...

int	rt_threadpool_worker_getdata (void voiddata, void *clientdata)
	worker thread can call this to get its client data pointer More...

int	rt_threadpool_sched_dynamic (rt_threadpool_t thrpool, rt_tasktile_t tile)
	Set shared iterator state to half-open interval defined by tile. More...

int	rt_threadpool_next_tile (void voidparms, int reqsize, rt_tasktile_t tile)
	iterate the shared iterator over the requested half-open interval More...

int	rt_threadpool_tile_failed (void voidparms, rt_tasktile_t tile)
	worker thread calls this when a failure occurs on a tile it has already taken from the scheduler More...

int	rt_threadpool_setfatalerror (void *voidparms)
	worker thread calls this to indicate that an unrecoverable error occured More...

int	rt_threadpool_getfatalerror (void *voidparms)
	master thread calls this to query for fatal errors More...

int	rt_threadlaunch (int numprocs, void clientdata, void fctn(void ), rt_tasktile_t tile)
	launch up to numprocs threads using shared iterator as a load balancer More...

int	rt_threadlaunch_getid (void voidparms, int threadid, int *threadcount)
	worker thread can call this to get its ID and number of peers More...

int	rt_threadlaunch_getdata (void voidparms, void *clientdata)
	worker thread can call this to get its client data pointer More...

int	rt_threadlaunch_next_tile (void voidparms, int reqsize, rt_tasktile_t tile)
	iterate the shared iterator over the requested half-open interval More...

int	rt_threadlaunch_setfatalerror (void *voidparms)
	worker thread calls this to indicate that an unrecoverable error occured More...

Variables
void ()(void *)	rt_thread_run_barrier (rt_run_barrier_t barrier, void fctn(void ), void parms, void **rsltparms)
	Wait until all threads reach barrier, and return the function pointer passed in by the master thread. More...

Detailed Description

Tachyon cross-platform thread creation and management, atomic operations, and CPU feature query APIs.

Definition in file threads.c.

Macro Definition Documentation

◆ RT_USEINTCPUID

#define RT_USEINTCPUID 1

Definition at line 233 of file threads.c.

Typedef Documentation

◆ RTTHREAD_START_ROUTINE

typedef void*(* RTTHREAD_START_ROUTINE) (void *)

Typedef to eliminate compiler warning caused by C/C++ linkage conflict.

Definition at line 643 of file threads.c.

Function Documentation

◆ rt_atomic_int_add_and_fetch()

int rt_atomic_int_add_and_fetch	(	rt_atomic_int_t *	atomp,
		int	inc
	)

fetch an atomic int and add inc to it, returning new value

Definition at line 1180 of file threads.c.

References atomic_int_struct::lock, rt_mutex_lock(), rt_mutex_unlock(), and atomic_int_struct::val.

◆ rt_atomic_int_destroy()

int rt_atomic_int_destroy ( rt_atomic_int_t * atomp )

destroy an atomic int variable

Definition at line 1066 of file threads.c.

References atomic_int_struct::lock, and rt_mutex_destroy().

Referenced by destroy_render_threads().

◆ rt_atomic_int_fetch_and_add()

int rt_atomic_int_fetch_and_add	(	rt_atomic_int_t *	atomp,
		int	inc
	)

fetch an atomic int and add inc to it, returning original value

Definition at line 1152 of file threads.c.

References atomic_int_struct::lock, rt_mutex_lock(), rt_mutex_unlock(), and atomic_int_struct::val.

Referenced by thread_trace().

◆ rt_atomic_int_get()

int rt_atomic_int_get ( rt_atomic_int_t * atomp )

get an atomic int variable

Definition at line 1123 of file threads.c.

References atomic_int_struct::lock, rt_mutex_lock(), rt_mutex_unlock(), and atomic_int_struct::val.

◆ rt_atomic_int_init()

int rt_atomic_int_init	(	rt_atomic_int_t *	atomp,
		int	val
	)

initialize an atomic int variable

Definition at line 1043 of file threads.c.

References atomic_int_struct::lock, rt_mutex_init(), and atomic_int_struct::val.

Referenced by create_render_threads().

◆ rt_atomic_int_set()

int rt_atomic_int_set	(	rt_atomic_int_t *	atomp,
		int	val
	)

set an atomic int variable

Definition at line 1087 of file threads.c.

References atomic_int_struct::lock, rt_mutex_lock(), rt_mutex_unlock(), and atomic_int_struct::val.

Referenced by renderscene().

◆ rt_cond_broadcast()

int rt_cond_broadcast ( rt_cond_t * cvp )

signal a condition variable, waking all threads

Definition at line 1003 of file threads.c.

Referenced by rt_thread_barrier().

◆ rt_cond_destroy()

int rt_cond_destroy ( rt_cond_t * cvp )

destroy a condition variable

Definition at line 888 of file threads.c.

Referenced by rt_thread_barrier_destroy(), and rt_thread_run_barrier_destroy().

◆ rt_cond_init()

int rt_cond_init ( rt_cond_t * cvp )

initialize a condition variable

Definition at line 852 of file threads.c.

Referenced by rt_rwlock_init(), rt_thread_barrier_init(), and rt_thread_run_barrier_init().

◆ rt_cond_signal()

int rt_cond_signal ( rt_cond_t * cvp )

signal a condition variable, waking at least one thread

Definition at line 971 of file threads.c.

Referenced by rt_rwlock_unlock().

◆ rt_cond_wait()

int rt_cond_wait	(	rt_cond_t *	cvp,
		rt_mutex_t *	mp
	)

wait on a condition variable

Definition at line 912 of file threads.c.

Referenced by rt_rwlock_readlock(), rt_rwlock_writelock(), and rt_thread_barrier().

◆ rt_cpu_affinitylist()

int* rt_cpu_affinitylist ( int * cpuaffinitycount )

query CPU affinity of the calling process (if allowed by host system)

Definition at line 483 of file threads.c.

◆ rt_cpu_capability_flags()

int rt_cpu_capability_flags ( rt_cpu_caps_t * cpucaps )

CPU optional instruction set capability flags.

Definition at line 281 of file threads.c.

References CPU_ARM64_AES, CPU_ARM64_ASIMD, CPU_ARM64_ASIMDDP, CPU_ARM64_ASIMDFHM, CPU_ARM64_ASIMDHP, CPU_ARM64_ASIMDRDM, CPU_ARM64_CRC32, CPU_ARM64_FP, CPU_ARM64_SHA1, CPU_ARM64_SHA2, CPU_ARM64_SHA3, CPU_ARM64_SHA512, CPU_ARM64_SVE, CPU_AVX, CPU_AVX2, CPU_AVX512CD, CPU_AVX512ER, CPU_AVX512F, CPU_AVX512PF, CPU_F16C, CPU_FMA, CPU_HT, CPU_HYPERVISOR, CPU_SMTDEPTH_UNKNOWN, CPU_SSE2, CPU_SSE4_1, CPU_UNKNOWN, rt_cpu_caps_struct::flags, and rt_cpu_caps_struct::smtdepth.

Referenced by rt_cpu_smt_depth(), and rt_par_getcpuinfo().

◆ rt_cpu_smt_depth()

int rt_cpu_smt_depth ( void )

CPU logical processors (SMT depth / aka hyperthreading)

Definition at line 466 of file threads.c.

References CPU_SMTDEPTH_UNKNOWN, rt_cpu_capability_flags(), and rt_cpu_caps_struct::smtdepth.

◆ rt_mutex_destroy()

int rt_mutex_destroy ( rt_mutex_t * mp )

destroy a mutex

Definition at line 828 of file threads.c.

Referenced by rt_atomic_int_destroy(), rt_shared_iterator_destroy(), rt_thread_barrier_destroy(), rt_thread_run_barrier_destroy(), and rt_tilestack_destroy().

◆ rt_mutex_init()

int rt_mutex_init ( rt_mutex_t * mp )

initialize a mutex

Definition at line 721 of file threads.c.

Referenced by rt_atomic_int_init(), rt_rwlock_init(), rt_shared_iterator_init(), rt_thread_barrier_init(), rt_thread_run_barrier_init(), and rt_tilestack_init().

◆ rt_mutex_lock()

int rt_mutex_lock ( rt_mutex_t * mp )

lock a mutex

Definition at line 742 of file threads.c.

Referenced by rt_atomic_int_add_and_fetch(), rt_atomic_int_fetch_and_add(), rt_atomic_int_get(), rt_atomic_int_set(), rt_rwlock_readlock(), rt_rwlock_unlock(), rt_rwlock_writelock(), rt_shared_iterator_getfatalerror(), rt_shared_iterator_set(), rt_thread_barrier(), rt_thread_run_barrier_poll(), rt_tilestack_compact(), rt_tilestack_empty(), rt_tilestack_pop(), rt_tilestack_popall(), and rt_tilestack_push().

◆ rt_mutex_spin_lock()

int rt_mutex_spin_lock ( rt_mutex_t * mp )

lock a mutex by spinning only

Definition at line 784 of file threads.c.

Referenced by rt_shared_iterator_next_tile(), and rt_shared_iterator_setfatalerror().

◆ rt_mutex_trylock()

int rt_mutex_trylock ( rt_mutex_t * mp )

try to lock a mutex

Definition at line 763 of file threads.c.

◆ rt_mutex_unlock()

int rt_mutex_unlock ( rt_mutex_t * mp )

◆ rt_rwlock_init()

int rt_rwlock_init ( rt_rwlock_t * rwp )

initialize a reader/writer lock

Definition at line 1212 of file threads.c.

References rt_cond_init(), and rt_mutex_init().

◆ rt_rwlock_readlock()

int rt_rwlock_readlock ( rt_rwlock_t * rwp )

set reader lock

Definition at line 1241 of file threads.c.

References rt_cond_wait(), rt_mutex_lock(), and rt_mutex_unlock().

◆ rt_rwlock_unlock()

int rt_rwlock_unlock ( rt_rwlock_t * rwp )

unlock reader/writer lock

Definition at line 1305 of file threads.c.

References rt_cond_signal(), rt_mutex_lock(), and rt_mutex_unlock().

◆ rt_rwlock_writelock()

int rt_rwlock_writelock ( rt_rwlock_t * rwp )

set writer lock

Definition at line 1270 of file threads.c.

References rt_cond_wait(), rt_mutex_lock(), and rt_mutex_unlock().

◆ rt_shared_iterator_destroy()

int rt_shared_iterator_destroy ( rt_shared_iterator_t * it )

destroy a shared iterator

Definition at line 1710 of file threads.c.

References rt_shared_iterator_struct::mtx, and rt_mutex_destroy().

Referenced by rt_threadlaunch(), and rt_threadpool_destroy().

◆ rt_shared_iterator_getfatalerror()

int rt_shared_iterator_getfatalerror ( rt_shared_iterator_t * it )

master thread calls this to query for fatal errors

Definition at line 1785 of file threads.c.

References rt_shared_iterator_struct::fatalerror, rt_shared_iterator_struct::mtx, rt_mutex_lock(), and rt_mutex_unlock().

Referenced by rt_threadlaunch(), and rt_threadpool_getfatalerror().

◆ rt_shared_iterator_init()

int rt_shared_iterator_init ( rt_shared_iterator_t * it )

initialize a shared iterator

Definition at line 1700 of file threads.c.

References rt_shared_iterator_struct::mtx, and rt_mutex_init().

Referenced by rt_threadlaunch(), and rt_threadpool_create().

◆ rt_shared_iterator_next_tile()

int rt_shared_iterator_next_tile	(	rt_shared_iterator_t *	it,
		int	reqsize,
		rt_tasktile_t *	tile
	)

iterate the shared iterator, over a requested half-open interval

iterate the shared iterator with a requested tile size, returns the tile received, and a return code of -1 if no iterations left or a fatal error has occured during processing, canceling all worker threads.

Definition at line 1736 of file threads.c.

References rt_shared_iterator_struct::current, rt_tasktile_struct::end, rt_shared_iterator_struct::end, rt_shared_iterator_struct::fatalerror, rt_shared_iterator_struct::mtx, rt_mutex_spin_lock(), rt_mutex_unlock(), RT_SCHED_CONTINUE, RT_SCHED_DONE, and rt_tasktile_struct::start.

Referenced by rt_threadlaunch_next_tile(), and rt_threadpool_next_tile().

◆ rt_shared_iterator_set()

int rt_shared_iterator_set	(	rt_shared_iterator_t *	it,
		rt_tasktile_t *	tile
	)

set shared iterator parameters

Set shared iterator state to half-open interval defined by tile.

Definition at line 1719 of file threads.c.

References rt_shared_iterator_struct::current, rt_tasktile_struct::end, rt_shared_iterator_struct::end, rt_shared_iterator_struct::fatalerror, rt_shared_iterator_struct::mtx, rt_mutex_lock(), rt_mutex_unlock(), rt_tasktile_struct::start, and rt_shared_iterator_struct::start.

Referenced by rt_threadlaunch(), and rt_threadpool_sched_dynamic().

◆ rt_shared_iterator_setfatalerror()

int rt_shared_iterator_setfatalerror ( rt_shared_iterator_t * it )

worker thread calls this to indicate a fatal error

Definition at line 1772 of file threads.c.

References rt_shared_iterator_struct::fatalerror, rt_shared_iterator_struct::mtx, rt_mutex_spin_lock(), and rt_mutex_unlock().

Referenced by rt_threadlaunch_setfatalerror(), and rt_threadpool_setfatalerror().

◆ rt_thread_barrier()

int rt_thread_barrier	(	rt_barrier_t *	barrier,
		int	increment
	)

synchronize on counting barrier primitive

Definition at line 1425 of file threads.c.

References barrier_struct::lock, barrier_struct::n_clients, barrier_struct::n_waiting, barrier_struct::phase, barrier_struct::result, rt_cond_broadcast(), rt_cond_wait(), rt_mutex_lock(), rt_mutex_unlock(), barrier_struct::sum, and barrier_struct::wait_cv.

Referenced by destroy_render_threads(), renderscene(), thread_trace(), and thread_worker().

◆ rt_thread_barrier_destroy()

void rt_thread_barrier_destroy ( rt_barrier_t * barrier )

destroy counting barrier primitive

Definition at line 1416 of file threads.c.

References barrier_struct::lock, rt_cond_destroy(), rt_mutex_destroy(), and barrier_struct::wait_cv.

Referenced by destroy_render_threads().

◆ rt_thread_barrier_init()

rt_barrier_t* rt_thread_barrier_init ( int n_clients )

initialize counting barrier primitive

Definition at line 1355 of file threads.c.

References barrier_struct::lock, barrier_struct::n_clients, barrier_struct::n_waiting, barrier_struct::phase, rt_cond_init(), rt_mutex_init(), barrier_struct::sum, and barrier_struct::wait_cv.

Referenced by create_render_threads().

◆ rt_thread_barrier_init_proc_shared()

int rt_thread_barrier_init_proc_shared	(	rt_barrier_t *	,
		int	n_clients
	)

When rendering in the CAVE we use a special synchronization mode so that shared memory mutexes and condition variables will work correctly when accessed from multiple processes.

Inter-process synchronization involves the kernel to a greater degree, so these barriers are substantially more costly to use than the ones designed for use within a single-process.

Definition at line 1379 of file threads.c.

References barrier_struct::lock, barrier_struct::n_clients, barrier_struct::n_waiting, barrier_struct::phase, barrier_struct::sum, and barrier_struct::wait_cv.

◆ rt_thread_create()

int rt_thread_create	(	rt_thread_t *	thr,
		void *	fctnvoid *,
		void *	arg
	)

create a new child thread

Definition at line 645 of file threads.c.

Referenced by create_render_threads(), rt_threadlaunch(), and rt_threadpool_create().

◆ rt_thread_join()

int rt_thread_join	(	rt_thread_t	thr,
		void **	stat
	)

join (wait for completion of, and merge with) a thread

Definition at line 688 of file threads.c.

Referenced by destroy_render_threads(), rt_threadlaunch(), and rt_threadpool_destroy().

◆ rt_thread_numphysprocessors()

int rt_thread_numphysprocessors ( void )

If compiling on Linux, enable the GNU CPU affinity functions in both libc and the libpthreads.

number of physical processors available

Definition at line 114 of file threads.c.

Referenced by rt_thread_numprocessors().

◆ rt_thread_numprocessors()

int rt_thread_numprocessors ( void )

number of processors available, subject to user override

Definition at line 202 of file threads.c.

References rt_thread_numphysprocessors().

Referenced by rt_par_getcpuinfo(), and rt_set_numthreads().

◆ rt_thread_run_barrier_destroy()

void rt_thread_run_barrier_destroy ( rt_run_barrier_t * barrier )

destroy thread pool barrier

Definition at line 1478 of file threads.c.

References rt_run_barrier_struct::lock, rt_cond_destroy(), rt_mutex_destroy(), and rt_run_barrier_struct::wait_cv.

Referenced by rt_threadpool_destroy().

◆ rt_thread_run_barrier_init()

int rt_thread_run_barrier_init	(	rt_run_barrier_t *	barrier,
		int	n_clients
	)

initialize thread pool barrier

Definition at line 1462 of file threads.c.

References rt_run_barrier_struct::fctn, rt_run_barrier_struct::lock, rt_run_barrier_struct::n_clients, rt_run_barrier_struct::n_waiting, rt_run_barrier_struct::phase, rt_cond_init(), rt_mutex_init(), and rt_run_barrier_struct::wait_cv.

Referenced by rt_threadpool_create().

◆ rt_thread_run_barrier_poll()

int rt_thread_run_barrier_poll ( rt_run_barrier_t * barrier )

non-blocking poll to see if peers are already at the barrier

Definition at line 1536 of file threads.c.

References rt_run_barrier_struct::lock, rt_run_barrier_struct::n_clients, rt_run_barrier_struct::n_waiting, rt_mutex_lock(), and rt_mutex_unlock().

Referenced by rt_threadpool_poll().

◆ rt_thread_set_self_cpuaffinity()

int rt_thread_set_self_cpuaffinity ( int cpu )

set the CPU affinity of the current thread (if allowed by host system)

Definition at line 572 of file threads.c.

Referenced by thread_worker().

◆ rt_thread_setconcurrency()

int rt_thread_setconcurrency ( int nthr )

set the concurrency level and scheduling scope for threads

Definition at line 618 of file threads.c.

Referenced by rt_set_numthreads().

◆ rt_threadlaunch()

int rt_threadlaunch	(	int	numprocs,
		void *	clientdata,
		void *	fctnvoid *,
		rt_tasktile_t *	tile
	)

launch up to numprocs threads using shared iterator as a load balancer

Definition at line 2085 of file threads.c.

References rt_threadlaunch_struct::clientdata, rt_threadlaunch_struct::iter, rt_shared_iterator_destroy(), rt_shared_iterator_getfatalerror(), rt_shared_iterator_init(), rt_shared_iterator_set(), rt_thread_create(), rt_thread_join(), rt_threadlaunch_struct::threadcount, and rt_threadlaunch_struct::threadid.

◆ rt_threadlaunch_getdata()

int rt_threadlaunch_getdata	(	void *	voidparms,
		void **	clientdata
	)

worker thread can call this to get its client data pointer

Definition at line 2175 of file threads.c.

References rt_threadlaunch_struct::clientdata.

◆ rt_threadlaunch_getid()

int rt_threadlaunch_getid	(	void *	voidparms,
		int *	threadid,
		int *	threadcount
	)

worker thread can call this to get its ID and number of peers

Definition at line 2162 of file threads.c.

References rt_threadlaunch_struct::threadcount, and rt_threadlaunch_struct::threadid.

◆ rt_threadlaunch_next_tile()

int rt_threadlaunch_next_tile	(	void *	voidparms,
		int	reqsize,
		rt_tasktile_t *	tile
	)

iterate the shared iterator over the requested half-open interval

worker thread calls this to get its next work unit iterate the shared iterator, returns -1 if no iterations left

Definition at line 2185 of file threads.c.

References rt_threadlaunch_struct::iter, and rt_shared_iterator_next_tile().

◆ rt_threadlaunch_setfatalerror()

int rt_threadlaunch_setfatalerror ( void * voidparms )

worker thread calls this to indicate that an unrecoverable error occured

Definition at line 2193 of file threads.c.

References rt_threadlaunch_struct::iter, and rt_shared_iterator_setfatalerror().

◆ rt_threadpool_create()

rt_threadpool_t* rt_threadpool_create	(	int	workercount,
		int *	devlist
	)

◆ rt_threadpool_destroy()

int rt_threadpool_destroy ( rt_threadpool_t * thrpool )

join all worker threads and free resources

Definition at line 1917 of file threads.c.

References rt_threadpool_struct::devlist, rt_threadpool_struct::errorstack, rt_threadpool_struct::iter, rt_shared_iterator_destroy(), rt_thread_join(), rt_thread_run_barrier, rt_thread_run_barrier_destroy(), rt_tilestack_destroy(), rt_threadpool_struct::runbar, rt_threadpool_struct::threads, rt_threadpool_struct::workercount, and rt_threadpool_struct::workerdata.

◆ rt_threadpool_get_workercount()

int rt_threadpool_get_workercount ( rt_threadpool_t * thrpool )

return the number of worker threads currently in the pool

query number of worker threads in the pool

Definition at line 1951 of file threads.c.

References rt_threadpool_struct::workercount.

◆ rt_threadpool_getfatalerror()

int rt_threadpool_getfatalerror ( void * voidparms )

master thread calls this to query for fatal errors

Definition at line 2077 of file threads.c.

References rt_threadpool_workerdata_struct::iter, and rt_shared_iterator_getfatalerror().

◆ rt_threadpool_launch()

int rt_threadpool_launch	(	rt_threadpool_t *	thrpool,
		void *	fctnvoid *,
		void *	parms,
		int	blocking
	)

launch threads onto a new function, with associated parms

Definition at line 1882 of file threads.c.

References rt_threadpool_workerdata_struct::parms, rt_thread_run_barrier, rt_threadpool_struct::runbar, and rt_threadpool_struct::workerdata.

◆ rt_threadpool_next_tile()

int rt_threadpool_next_tile	(	void *	voidparms,
		int	reqsize,
		rt_tasktile_t *	tile
	)

iterate the shared iterator over the requested half-open interval

worker thread calls this to get its next work unit iterate the shared iterator, returns -1 if no iterations left

Definition at line 2042 of file threads.c.

References rt_threadpool_workerdata_struct::errorstack, rt_threadpool_workerdata_struct::iter, RT_SCHED_CONTINUE, RT_SCHED_DONE, rt_shared_iterator_next_tile(), RT_TILESTACK_EMPTY, and rt_tilestack_pop().

◆ rt_threadpool_poll()

int rt_threadpool_poll ( rt_threadpool_t * thrpool )

Definition at line 1908 of file threads.c.

References rt_thread_run_barrier_poll(), and rt_threadpool_struct::runbar.

◆ rt_threadpool_sched_dynamic()

int rt_threadpool_sched_dynamic	(	rt_threadpool_t *	thrpool,
		rt_tasktile_t *	tile
	)

Set shared iterator state to half-open interval defined by tile.

Set dynamic scheduler state to half-open interval defined by tile.

Definition at line 2034 of file threads.c.

References rt_threadpool_struct::iter, and rt_shared_iterator_set().

◆ rt_threadpool_setfatalerror()

int rt_threadpool_setfatalerror ( void * voidparms )

worker thread calls this to indicate that an unrecoverable error occured

Definition at line 2069 of file threads.c.

References rt_threadpool_workerdata_struct::iter, and rt_shared_iterator_setfatalerror().

◆ rt_threadpool_tile_failed()

int rt_threadpool_tile_failed	(	void *	voidparms,
		rt_tasktile_t *	tile
	)

worker thread calls this when a failure occurs on a tile it has already taken from the scheduler

worker thread calls this when it fails computing a tile after it has already taken it from the scheduler

Definition at line 2062 of file threads.c.

References rt_threadpool_workerdata_struct::errorstack, and rt_tilestack_push().

◆ rt_threadpool_wait()

int rt_threadpool_wait ( rt_threadpool_t * thrpool )

wait for all worker threads to complete their work

Definition at line 1900 of file threads.c.

References rt_thread_run_barrier, and rt_threadpool_struct::runbar.

◆ rt_threadpool_worker_devscaletile()

int rt_threadpool_worker_devscaletile	(	void *	voiddata,
		int *	tilesize
	)

worker thread calls this to scale max tile size by worker speed as determined by the SM/core count and clock rate

Definition at line 2008 of file threads.c.

References rt_threadpool_workerdata_struct::devspeed.

◆ rt_threadpool_worker_getdata()

int rt_threadpool_worker_getdata	(	void *	voiddata,
		void **	clientdata
	)

worker thread can call this to get its client data pointer

Definition at line 2024 of file threads.c.

References rt_threadpool_workerdata_struct::parms.

◆ rt_threadpool_worker_getdevid()

int rt_threadpool_worker_getdevid	(	void *	voiddata,
		int *	devid
	)

worker thread can call this to get its CPU/GPU device ID

Definition at line 1970 of file threads.c.

References rt_threadpool_workerdata_struct::devid.

◆ rt_threadpool_worker_getdevspeed()

int rt_threadpool_worker_getdevspeed	(	void *	voiddata,
		float *	speed
	)

worker thread calls this to get relative speed of this device as determined by the SM/core count and clock rate

Worker thread calls this to get relative speed of this device as determined by the SM/core count and clock rate.

Definition at line 1996 of file threads.c.

References rt_threadpool_workerdata_struct::devspeed.

◆ rt_threadpool_worker_getid()

int rt_threadpool_worker_getid	(	void *	voiddata,
		int *	threadid,
		int *	threadcount
	)

worker thread can call this to get its ID and number of peers

Definition at line 1957 of file threads.c.

References rt_threadpool_workerdata_struct::threadcount, and rt_threadpool_workerdata_struct::threadid.

◆ rt_threadpool_worker_setdevspeed()

int rt_threadpool_worker_setdevspeed	(	void *	voiddata,
		float	speed
	)

worker thread calls this to set relative speed of this device as determined by the SM/core count and clock rate Note: this should only be called once, during the worker's device initialization process

Worker thread calls this to set relative speed of this device as determined by the SM/core count and clock rate Note: this should only be called once, during the worker's device initialization process.

Definition at line 1985 of file threads.c.

References rt_threadpool_workerdata_struct::devspeed.

◆ rt_tilestack_compact()

int rt_tilestack_compact ( rt_tilestack_t * s )

shrink memory buffers associated with task tile stack if possible

Definition at line 1584 of file threads.c.

References rt_tilestack_t::mtx, rt_mutex_lock(), rt_mutex_unlock(), rt_tilestack_t::s, rt_tilestack_t::size, and rt_tilestack_t::top.

◆ rt_tilestack_destroy()

void rt_tilestack_destroy ( rt_tilestack_t * s )

destroy task tile stack

Definition at line 1575 of file threads.c.

References rt_tilestack_t::mtx, rt_mutex_destroy(), and rt_tilestack_t::s.

Referenced by rt_threadpool_destroy().

◆ rt_tilestack_empty()

int rt_tilestack_empty ( rt_tilestack_t * s )

query if the task tile stack is empty or not

Definition at line 1675 of file threads.c.

References rt_tilestack_t::mtx, rt_mutex_lock(), rt_mutex_unlock(), and rt_tilestack_t::top.

◆ rt_tilestack_init()

int rt_tilestack_init	(	rt_tilestack_t *	s,
		int	size
	)

initialize task tile stack (to empty)

Definition at line 1552 of file threads.c.

References rt_tilestack_t::growthrate, rt_tilestack_t::mtx, rt_mutex_init(), rt_tilestack_t::s, rt_tilestack_t::size, and rt_tilestack_t::top.

Referenced by rt_threadpool_create().

◆ rt_tilestack_pop()

int rt_tilestack_pop	(	rt_tilestack_t *	s,
		rt_tasktile_t *	t
	)

pop a task tile off of the stack

Definition at line 1637 of file threads.c.

References rt_tilestack_t::mtx, rt_mutex_lock(), rt_mutex_unlock(), RT_TILESTACK_EMPTY, rt_tilestack_t::s, and rt_tilestack_t::top.

Referenced by rt_threadpool_next_tile().

◆ rt_tilestack_popall()

int rt_tilestack_popall ( rt_tilestack_t * s )

pop all of the task tiles off of the stack

Definition at line 1660 of file threads.c.

References rt_tilestack_t::mtx, rt_mutex_lock(), rt_mutex_unlock(), and rt_tilestack_t::top.

◆ rt_tilestack_push()

int rt_tilestack_push	(	rt_tilestack_t *	s,
		const rt_tasktile_t *	t
	)

push a task tile onto the stack

Definition at line 1608 of file threads.c.

References rt_tilestack_t::growthrate, rt_tilestack_t::mtx, rt_mutex_lock(), rt_mutex_unlock(), rt_tilestack_t::s, rt_tilestack_t::size, and rt_tilestack_t::top.

Referenced by rt_threadpool_tile_failed().

Variable Documentation

◆ rt_thread_run_barrier

void*(*)(void *) rt_thread_run_barrier(rt_run_barrier_t *barrier, void *fctn(void *), void *parms, void **rsltparms)

Wait until all threads reach barrier, and return the function pointer passed in by the master thread.

Definition at line 1490 of file threads.c.

Referenced by rt_threadpool_destroy(), rt_threadpool_launch(), and rt_threadpool_wait().

Macros

Typedefs

Functions

Variables

Detailed Description

Macro Definition Documentation

◆ RT_USEINTCPUID

Typedef Documentation

◆ RTTHREAD_START_ROUTINE

Function Documentation

◆ rt_atomic_int_add_and_fetch()

◆ rt_atomic_int_destroy()

◆ rt_atomic_int_fetch_and_add()

◆ rt_atomic_int_get()

◆ rt_atomic_int_init()

◆ rt_atomic_int_set()

◆ rt_cond_broadcast()

◆ rt_cond_destroy()

◆ rt_cond_init()

◆ rt_cond_signal()

◆ rt_cond_wait()

◆ rt_cpu_affinitylist()

◆ rt_cpu_capability_flags()

◆ rt_cpu_smt_depth()

◆ rt_mutex_destroy()

◆ rt_mutex_init()

◆ rt_mutex_lock()

◆ rt_mutex_spin_lock()

◆ rt_mutex_trylock()

◆ rt_mutex_unlock()

◆ rt_rwlock_init()

◆ rt_rwlock_readlock()

◆ rt_rwlock_unlock()

◆ rt_rwlock_writelock()

◆ rt_shared_iterator_destroy()

◆ rt_shared_iterator_getfatalerror()

◆ rt_shared_iterator_init()

◆ rt_shared_iterator_next_tile()

◆ rt_shared_iterator_set()

◆ rt_shared_iterator_setfatalerror()

◆ rt_thread_barrier()

◆ rt_thread_barrier_destroy()

◆ rt_thread_barrier_init()

◆ rt_thread_barrier_init_proc_shared()

◆ rt_thread_create()

◆ rt_thread_join()

◆ rt_thread_numphysprocessors()

◆ rt_thread_numprocessors()

◆ rt_thread_run_barrier_destroy()

◆ rt_thread_run_barrier_init()

◆ rt_thread_run_barrier_poll()

◆ rt_thread_set_self_cpuaffinity()

◆ rt_thread_setconcurrency()

◆ rt_threadlaunch()

◆ rt_threadlaunch_getdata()

◆ rt_threadlaunch_getid()

◆ rt_threadlaunch_next_tile()

◆ rt_threadlaunch_setfatalerror()

◆ rt_threadpool_create()

◆ rt_threadpool_destroy()

◆ rt_threadpool_get_workercount()

◆ rt_threadpool_getfatalerror()

◆ rt_threadpool_launch()

◆ rt_threadpool_next_tile()

◆ rt_threadpool_poll()

◆ rt_threadpool_sched_dynamic()

◆ rt_threadpool_setfatalerror()

◆ rt_threadpool_tile_failed()

◆ rt_threadpool_wait()

◆ rt_threadpool_worker_devscaletile()

◆ rt_threadpool_worker_getdata()

◆ rt_threadpool_worker_getdevid()

◆ rt_threadpool_worker_getdevspeed()

◆ rt_threadpool_worker_getid()

◆ rt_threadpool_worker_setdevspeed()

◆ rt_tilestack_compact()

◆ rt_tilestack_destroy()

◆ rt_tilestack_empty()

◆ rt_tilestack_init()

◆ rt_tilestack_pop()