Differential D7475 Diff 23908 source/blender/blenlib/intern/task_iterator.c

Changeset View

Standalone View

source/blender/blenlib/intern/task_iterator.c

Show All 11 Lines
* You should have received a copy of the GNU General Public License		* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software Foundation,		* along with this program; if not, write to the Free Software Foundation,
* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.		* Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*/		*/

/** \file		/** \file
* \ingroup bli		* \ingroup bli
*		*
* A generic task system which can be used for any task based subsystem.		* Parallel tasks over all elements in a container.
*/		*/

#include <stdlib.h>		#include <stdlib.h>

#include "MEM_guardedalloc.h"		#include "MEM_guardedalloc.h"

#include "DNA_listBase.h"		#include "DNA_listBase.h"

#include "BLI_listbase.h"		#include "BLI_listbase.h"
#include "BLI_math.h"		#include "BLI_math.h"
#include "BLI_mempool.h"		#include "BLI_mempool.h"
#include "BLI_task.h"		#include "BLI_task.h"
#include "BLI_threads.h"		#include "BLI_threads.h"

#include "atomic_ops.h"		#include "atomic_ops.h"

/* Parallel range routines */

/**
*
* Main functions:
* - #BLI_task_parallel_range
* - #BLI_task_parallel_listbase (#ListBase - double linked list)
*
* TODO:
* - #BLI_task_parallel_foreach_link (#Link - single linked list)
* - #BLI_task_parallel_foreach_ghash/gset (#GHash/#GSet - hash & set)
* - #BLI_task_parallel_foreach_mempool (#BLI_mempool - iterate over mempools)
*/

/* Allows to avoid using malloc for userdata_chunk in tasks, when small enough. */		/* Allows to avoid using malloc for userdata_chunk in tasks, when small enough. */
#define MALLOCA(_size) ((_size) <= 8192) ? alloca((_size)) : MEM_mallocN((_size), __func__)		#define MALLOCA(_size) ((_size) <= 8192) ? alloca((_size)) : MEM_mallocN((_size), __func__)
#define MALLOCA_FREE(_mem, _size) \		#define MALLOCA_FREE(_mem, _size) \
if (((_mem) != NULL) && ((_size) > 8192)) \		if (((_mem) != NULL) && ((_size) > 8192)) \
MEM_freeN((_mem))		MEM_freeN((_mem))

		#if 0
/* Stores all needed data to perform a parallelized iteration,		/* Stores all needed data to perform a parallelized iteration,
* with a same operation (callback function).		* with a same operation (callback function).
* It can be chained with other tasks in a single-linked list way. */		* It can be chained with other tasks in a single-linked list way. */
typedef struct TaskParallelRangeState {		typedef struct TaskParallelRangeState {
struct TaskParallelRangeState *next;		struct TaskParallelRangeState *next;

/* Start and end point of integer value iteration. */		/* Start and end point of integer value iteration. */
int start, stop;		int start, stop;
Show All 40 Lines	typedef struct TaskParallelRangePool {
/* Linked list of range tasks to process. */		/* Linked list of range tasks to process. */
TaskParallelRangeState *parallel_range_states;		TaskParallelRangeState *parallel_range_states;
/* Current range task beeing processed, swapped atomically. */		/* Current range task beeing processed, swapped atomically. */
TaskParallelRangeState *current_state;		TaskParallelRangeState *current_state;
/* Scheduling settings common to all tasks. */		/* Scheduling settings common to all tasks. */
TaskParallelSettings *settings;		TaskParallelSettings *settings;
} TaskParallelRangePool;		} TaskParallelRangePool;

		#endif

BLI_INLINE void task_parallel_calc_chunk_size(const TaskParallelSettings *settings,		BLI_INLINE void task_parallel_calc_chunk_size(const TaskParallelSettings *settings,
const int tot_items,		const int tot_items,
int num_tasks,		int num_tasks,
int *r_chunk_size)		int *r_chunk_size)
{		{
int chunk_size = 0;		int chunk_size = 0;

if (!settings->use_threading) {		if (!settings->use_threading) {
Show All 28 Lines	else {
/* Basic heuristic to avoid threading on low amount of items.		/* Basic heuristic to avoid threading on low amount of items.
* We could make that limit configurable in settings too. */		* We could make that limit configurable in settings too. */
if (tot_items > 0 && tot_items < max_ii(256, chunk_size * 2)) {		if (tot_items > 0 && tot_items < max_ii(256, chunk_size * 2)) {
chunk_size = tot_items;		chunk_size = tot_items;
}		}
}		}

BLI_assert(chunk_size > 0);		BLI_assert(chunk_size > 0);
		#if 0

if (tot_items > 0) {		if (tot_items > 0) {
switch (settings->scheduling_mode) {		switch (settings->scheduling_mode) {
case TASK_SCHEDULING_STATIC:		case TASK_SCHEDULING_STATIC:
*r_chunk_size = max_ii(chunk_size, tot_items / num_tasks);		*r_chunk_size = max_ii(chunk_size, tot_items / num_tasks);
break;		break;
case TASK_SCHEDULING_DYNAMIC:		case TASK_SCHEDULING_DYNAMIC:
*r_chunk_size = chunk_size;		*r_chunk_size = chunk_size;
▲ Show 20 Lines • Show All 173 Lines • ▼ Show 20 Lines	void BLI_task_parallel_range(const int start,
range_pool.num_tasks = num_tasks = min_ii(num_tasks,		range_pool.num_tasks = num_tasks = min_ii(num_tasks,
max_ii(1, (stop - start) / range_pool.chunk_size));		max_ii(1, (stop - start) / range_pool.chunk_size));

if (num_tasks == 1) {		if (num_tasks == 1) {
parallel_range_single_thread(&range_pool);		parallel_range_single_thread(&range_pool);
return;		return;
}		}

TaskPool *task_pool = range_pool.pool = BLI_task_pool_create_suspended(		TaskPool *task_pool = range_pool.pool = BLI_task_pool_create(
task_scheduler, &range_pool, TASK_PRIORITY_HIGH);		task_scheduler, &range_pool, TASK_PRIORITY_HIGH);

range_pool.current_state = &state;		range_pool.current_state = &state;

if (use_tls_data) {		if (use_tls_data) {
state.flatten_tls_storage = flatten_tls_storage = MALLOCA(tls_data_size * (size_t)num_tasks);		state.flatten_tls_storage = flatten_tls_storage = MALLOCA(tls_data_size * (size_t)num_tasks);
state.tls_data_size = tls_data_size;		state.tls_data_size = tls_data_size;
}		}
Show All 18 Lines	for (i = 0; i < num_tasks; i++) {
if (settings->func_reduce) {		if (settings->func_reduce) {
settings->func_reduce(userdata, tls_data, userdata_chunk_local);		settings->func_reduce(userdata, tls_data, userdata_chunk_local);
}		}
if (settings->func_free) {		if (settings->func_free) {
/* `func_free` should only free data that was created during execution of `func`. */		/* `func_free` should only free data that was created during execution of `func`. */
settings->func_free(userdata, userdata_chunk_local);		settings->func_free(userdata, userdata_chunk_local);
}		}
}		}
MALLOCA_FREE(flatten_tls_storage, tls_data_size * (size_t)num_tasks);
}		}
		MALLOCA_FREE(flatten_tls_storage, tls_data_size * (size_t)num_tasks);
}		}

/**		/**
* Initialize a task pool to parallelize several for loops at the same time.		* Initialize a task pool to parallelize several for loops at the same time.
*		*
* See public API doc of ParallelRangeSettings for description of all settings.		* See public API doc of ParallelRangeSettings for description of all settings.
* Note that loop-specific settings (like 'tls' data or reduce/free functions) must be left NULL		* Note that loop-specific settings (like 'tls' data or reduce/free functions) must be left NULL
* here. Only settings controlling how iteration is parallelized must be defined, as those will		* here. Only settings controlling how iteration is parallelized must be defined, as those will
▲ Show 20 Lines • Show All 46 Lines • ▼ Show 20 Lines	void BLI_task_parallel_range_pool_push(TaskParallelRangePool *range_pool,
state->tls_data_size = settings->userdata_chunk_size;		state->tls_data_size = settings->userdata_chunk_size;
state->func_reduce = settings->func_reduce;		state->func_reduce = settings->func_reduce;
state->func_free = settings->func_free;		state->func_free = settings->func_free;

state->next = range_pool->parallel_range_states;		state->next = range_pool->parallel_range_states;
range_pool->parallel_range_states = state;		range_pool->parallel_range_states = state;
}		}

static void parallel_range_func_finalize(TaskPool *__restrict pool,		static void parallel_range_func_reduce(TaskPool *__restrict pool,
void *v_state,		void *v_state,
int UNUSED(thread_id))		int UNUSED(thread_id))
{		{
TaskParallelRangePool *__restrict range_pool = BLI_task_pool_userdata(pool);		TaskParallelRangePool *__restrict range_pool = BLI_task_pool_userdata(pool);
TaskParallelRangeState *state = v_state;		TaskParallelRangeState *state = v_state;

for (int i = 0; i < range_pool->num_tasks; i++) {		for (int i = 0; i < range_pool->num_tasks; i++) {
void tls_data = (char )state->flatten_tls_storage + (state->tls_data_size * (size_t)i);		void tls_data = (char )state->flatten_tls_storage + (state->tls_data_size * (size_t)i);
if (state->func_reduce != NULL) {		if (state->func_reduce != NULL) {
state->func_reduce(state->userdata_shared, state->initial_tls_memory, tls_data);		state->func_reduce(state->userdata_shared, state->initial_tls_memory, tls_data);
▲ Show 20 Lines • Show All 57 Lines • ▼ Show 20 Lines	state->flatten_tls_storage = userdata_chunk_array = MALLOCA(userdata_chunk_size *
(size_t)num_tasks);		(size_t)num_tasks);
for (int i = 0; i < num_tasks; i++) {		for (int i = 0; i < num_tasks; i++) {
void userdata_chunk_local = (char )userdata_chunk_array +		void userdata_chunk_local = (char )userdata_chunk_array +
(userdata_chunk_size * (size_t)i);		(userdata_chunk_size * (size_t)i);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);		memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
}		}
}		}

TaskPool *task_pool = range_pool->pool = BLI_task_pool_create_suspended(		TaskPool *task_pool = range_pool->pool = BLI_task_pool_create(
task_scheduler, range_pool, TASK_PRIORITY_HIGH);		task_scheduler, range_pool, TASK_PRIORITY_HIGH);

range_pool->current_state = range_pool->parallel_range_states;		range_pool->current_state = range_pool->parallel_range_states;
const int thread_id = BLI_task_pool_creator_thread_id(task_pool);		const int thread_id = BLI_task_pool_creator_thread_id(task_pool);
for (int i = 0; i < num_tasks; i++) {		for (int i = 0; i < num_tasks; i++) {
BLI_task_pool_push_from_thread(		BLI_task_pool_push_from_thread(
task_pool, parallel_range_func, POINTER_FROM_INT(i), false, NULL, thread_id);		task_pool, parallel_range_func, POINTER_FROM_INT(i), false, NULL, thread_id);
}		}
Show All 10 Lines	for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL;
UNUSED_VARS_NDEBUG(userdata_chunk_array);		UNUSED_VARS_NDEBUG(userdata_chunk_array);
if (userdata_chunk_size == 0) {		if (userdata_chunk_size == 0) {
BLI_assert(userdata_chunk_array == NULL);		BLI_assert(userdata_chunk_array == NULL);
continue;		continue;
}		}

if (state->func_reduce != NULL \|\| state->func_free != NULL) {		if (state->func_reduce != NULL \|\| state->func_free != NULL) {
BLI_task_pool_push_from_thread(		BLI_task_pool_push_from_thread(
task_pool, parallel_range_func_finalize, state, false, NULL, thread_id);		task_pool, parallel_range_func_reduce, state, false, NULL, thread_id);
}		}
}		}

BLI_task_pool_work_and_wait(task_pool);		BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);		BLI_task_pool_free(task_pool);
range_pool->pool = NULL;		range_pool->pool = NULL;

/* Cleanup all tasks. */		/* Cleanup all tasks. */
Show All 22 Lines	void BLI_task_parallel_range_pool_free(TaskParallelRangePool *range_pool)
TaskParallelRangeState *state_next = NULL;		TaskParallelRangeState *state_next = NULL;
for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL;		for (TaskParallelRangeState *state = range_pool->parallel_range_states; state != NULL;
state = state_next) {		state = state_next) {
state_next = state->next;		state_next = state->next;
MEM_freeN(state);		MEM_freeN(state);
}		}
MEM_freeN(range_pool->settings);		MEM_freeN(range_pool->settings);
MEM_freeN(range_pool);		MEM_freeN(range_pool);
		#else
		*r_chunk_size = chunk_size;
		#endif
}		}

typedef struct TaskParallelIteratorState {		typedef struct TaskParallelIteratorState {
void *userdata;		void *userdata;
TaskParallelIteratorIterFunc iter_func;		TaskParallelIteratorIterFunc iter_func;
TaskParallelIteratorFunc func;		TaskParallelIteratorFunc func;

/* * Data used to 'acquire' chunks of items from the iterator. * */		/* * Data used to 'acquire' chunks of items from the iterator. * */
/* Common data also passed to the generator callback. */		/* Common data also passed to the generator callback. */
TaskParallelIteratorStateShared iter_shared;		TaskParallelIteratorStateShared iter_shared;
/* Total number of items. If unknown, set it to a negative number. */		/* Total number of items. If unknown, set it to a negative number. */
int tot_items;		int tot_items;
} TaskParallelIteratorState;		} TaskParallelIteratorState;

BLI_INLINE void task_parallel_iterator_calc_chunk_size(const TaskParallelSettings *settings,
const int num_tasks,
TaskParallelIteratorState *state)
{
task_parallel_calc_chunk_size(
settings, state->tot_items, num_tasks, &state->iter_shared.chunk_size);
}

static void parallel_iterator_func_do(TaskParallelIteratorState *__restrict state,		static void parallel_iterator_func_do(TaskParallelIteratorState *__restrict state,
void *userdata_chunk,		void *userdata_chunk)
int threadid)
{		{
TaskParallelTLS tls = {		TaskParallelTLS tls = {
.thread_id = threadid,
.userdata_chunk = userdata_chunk,		.userdata_chunk = userdata_chunk,
};		};

void **current_chunk_items;		void **current_chunk_items;
int *current_chunk_indices;		int *current_chunk_indices;
int current_chunk_size;		int current_chunk_size;

const size_t items_size = sizeof(current_chunk_items) (size_t)state->iter_shared.chunk_size;		const size_t items_size = sizeof(current_chunk_items) (size_t)state->iter_shared.chunk_size;
Show All 36 Lines	for (i = 0; i < current_chunk_size; ++i) {
state->func(state->userdata, current_chunk_items[i], current_chunk_indices[i], &tls);		state->func(state->userdata, current_chunk_items[i], current_chunk_indices[i], &tls);
}		}
}		}

MALLOCA_FREE(current_chunk_items, items_size);		MALLOCA_FREE(current_chunk_items, items_size);
MALLOCA_FREE(current_chunk_indices, indices_size);		MALLOCA_FREE(current_chunk_indices, indices_size);
}		}

static void parallel_iterator_func(TaskPool __restrict pool, void userdata_chunk, int threadid)		static void parallel_iterator_func(TaskPool __restrict pool, void userdata_chunk)
{		{
TaskParallelIteratorState *__restrict state = BLI_task_pool_userdata(pool);		TaskParallelIteratorState *__restrict state = BLI_task_pool_user_data(pool);

parallel_iterator_func_do(state, userdata_chunk, threadid);		parallel_iterator_func_do(state, userdata_chunk);
}		}

static void task_parallel_iterator_no_threads(const TaskParallelSettings *settings,		static void task_parallel_iterator_no_threads(const TaskParallelSettings *settings,
TaskParallelIteratorState *state)		TaskParallelIteratorState *state)
{		{
/* Prepare user's TLS data. */		/* Prepare user's TLS data. */
void *userdata_chunk = settings->userdata_chunk;		void *userdata_chunk = settings->userdata_chunk;
const size_t userdata_chunk_size = settings->userdata_chunk_size;		const size_t userdata_chunk_size = settings->userdata_chunk_size;
void *userdata_chunk_local = NULL;		void *userdata_chunk_local = NULL;
const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);		const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);
if (use_userdata_chunk) {		if (use_userdata_chunk) {
userdata_chunk_local = MALLOCA(userdata_chunk_size);		userdata_chunk_local = MALLOCA(userdata_chunk_size);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);		memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
}		}

/* Also marking it as non-threaded for the iterator callback. */		/* Also marking it as non-threaded for the iterator callback. */
state->iter_shared.spin_lock = NULL;		state->iter_shared.spin_lock = NULL;

parallel_iterator_func_do(state, userdata_chunk, 0);		parallel_iterator_func_do(state, userdata_chunk);

if (use_userdata_chunk && settings->func_free != NULL) {		if (use_userdata_chunk && settings->func_free != NULL) {
/* `func_free` should only free data that was created during execution of `func`. */		/* `func_free` should only free data that was created during execution of `func`. */
settings->func_free(state->userdata, userdata_chunk_local);		settings->func_free(state->userdata, userdata_chunk_local);
}		}
}		}

static void task_parallel_iterator_do(const TaskParallelSettings *settings,		static void task_parallel_iterator_do(const TaskParallelSettings *settings,
TaskParallelIteratorState *state)		TaskParallelIteratorState *state)
{		{
TaskScheduler *task_scheduler = BLI_task_scheduler_get();		const int num_threads = BLI_task_scheduler_num_threads();
const int num_threads = BLI_task_scheduler_num_threads(task_scheduler);

task_parallel_iterator_calc_chunk_size(settings, num_threads, state);		task_parallel_calc_chunk_size(
		settings, state->tot_items, num_threads, &state->iter_shared.chunk_size);

if (!settings->use_threading) {		if (!settings->use_threading) {
task_parallel_iterator_no_threads(settings, state);		task_parallel_iterator_no_threads(settings, state);
return;		return;
}		}

const int chunk_size = state->iter_shared.chunk_size;		const int chunk_size = state->iter_shared.chunk_size;
const int tot_items = state->tot_items;		const int tot_items = state->tot_items;
Show All 12 Lines	static void task_parallel_iterator_do(const TaskParallelSettings *settings,
state->iter_shared.spin_lock = &spin_lock;		state->iter_shared.spin_lock = &spin_lock;

void *userdata_chunk = settings->userdata_chunk;		void *userdata_chunk = settings->userdata_chunk;
const size_t userdata_chunk_size = settings->userdata_chunk_size;		const size_t userdata_chunk_size = settings->userdata_chunk_size;
void *userdata_chunk_local = NULL;		void *userdata_chunk_local = NULL;
void *userdata_chunk_array = NULL;		void *userdata_chunk_array = NULL;
const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);		const bool use_userdata_chunk = (userdata_chunk_size != 0) && (userdata_chunk != NULL);

TaskPool *task_pool = BLI_task_pool_create_suspended(task_scheduler, state, TASK_PRIORITY_HIGH);		TaskPool *task_pool = BLI_task_pool_create(state, TASK_PRIORITY_HIGH);

if (use_userdata_chunk) {		if (use_userdata_chunk) {
userdata_chunk_array = MALLOCA(userdata_chunk_size * num_tasks);		userdata_chunk_array = MALLOCA(userdata_chunk_size * num_tasks);
}		}

const int thread_id = BLI_task_pool_creator_thread_id(task_pool);
for (size_t i = 0; i < num_tasks; i++) {		for (size_t i = 0; i < num_tasks; i++) {
if (use_userdata_chunk) {		if (use_userdata_chunk) {
userdata_chunk_local = (char )userdata_chunk_array + (userdata_chunk_size i);		userdata_chunk_local = (char )userdata_chunk_array + (userdata_chunk_size i);
memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);		memcpy(userdata_chunk_local, userdata_chunk, userdata_chunk_size);
}		}
/* Use this pool's pre-allocated tasks. */		/* Use this pool's pre-allocated tasks. */
BLI_task_pool_push_from_thread(		BLI_task_pool_push(task_pool, parallel_iterator_func, userdata_chunk_local, false, NULL);
task_pool, parallel_iterator_func, userdata_chunk_local, false, NULL, thread_id);
}		}

BLI_task_pool_work_and_wait(task_pool);		BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);		BLI_task_pool_free(task_pool);

if (use_userdata_chunk && (settings->func_reduce != NULL \|\| settings->func_free != NULL)) {		if (use_userdata_chunk && (settings->func_reduce != NULL \|\| settings->func_free != NULL)) {
for (size_t i = 0; i < num_tasks; i++) {		for (size_t i = 0; i < num_tasks; i++) {
userdata_chunk_local = (char )userdata_chunk_array + (userdata_chunk_size i);		userdata_chunk_local = (char )userdata_chunk_array + (userdata_chunk_size i);
▲ Show 20 Lines • Show All 99 Lines • ▼ Show 20 Lines
#undef MALLOCA		#undef MALLOCA
#undef MALLOCA_FREE		#undef MALLOCA_FREE

typedef struct ParallelMempoolState {		typedef struct ParallelMempoolState {
void *userdata;		void *userdata;
TaskParallelMempoolFunc func;		TaskParallelMempoolFunc func;
} ParallelMempoolState;		} ParallelMempoolState;

static void parallel_mempool_func(TaskPool __restrict pool, void taskdata, int UNUSED(threadid))		static void parallel_mempool_func(TaskPool __restrict pool, void taskdata)
{		{
ParallelMempoolState *__restrict state = BLI_task_pool_userdata(pool);		ParallelMempoolState *__restrict state = BLI_task_pool_user_data(pool);
BLI_mempool_iter *iter = taskdata;		BLI_mempool_iter *iter = taskdata;
MempoolIterData *item;		MempoolIterData *item;

while ((item = BLI_mempool_iterstep(iter)) != NULL) {		while ((item = BLI_mempool_iterstep(iter)) != NULL) {
state->func(state->userdata, item);		state->func(state->userdata, item);
}		}
}		}

Show All 9 Lines
*		*
* \note There is no static scheduling here.		* \note There is no static scheduling here.
*/		*/
void BLI_task_parallel_mempool(BLI_mempool *mempool,		void BLI_task_parallel_mempool(BLI_mempool *mempool,
void *userdata,		void *userdata,
TaskParallelMempoolFunc func,		TaskParallelMempoolFunc func,
const bool use_threading)		const bool use_threading)
{		{
TaskScheduler *task_scheduler;
TaskPool *task_pool;		TaskPool *task_pool;
ParallelMempoolState state;		ParallelMempoolState state;
int i, num_threads, num_tasks;		int i, num_threads, num_tasks;

if (BLI_mempool_len(mempool) == 0) {		if (BLI_mempool_len(mempool) == 0) {
return;		return;
}		}

if (!use_threading) {		if (!use_threading) {
BLI_mempool_iter iter;		BLI_mempool_iter iter;
BLI_mempool_iternew(mempool, &iter);		BLI_mempool_iternew(mempool, &iter);

for (void *item = BLI_mempool_iterstep(&iter); item != NULL;		for (void *item = BLI_mempool_iterstep(&iter); item != NULL;
item = BLI_mempool_iterstep(&iter)) {		item = BLI_mempool_iterstep(&iter)) {
func(userdata, item);		func(userdata, item);
}		}
return;		return;
}		}

task_scheduler = BLI_task_scheduler_get();		task_pool = BLI_task_pool_create(&state, TASK_PRIORITY_HIGH);
task_pool = BLI_task_pool_create_suspended(task_scheduler, &state, TASK_PRIORITY_HIGH);		num_threads = BLI_task_scheduler_num_threads();
num_threads = BLI_task_scheduler_num_threads(task_scheduler);

/* The idea here is to prevent creating task for each of the loop iterations		/* The idea here is to prevent creating task for each of the loop iterations
* and instead have tasks which are evenly distributed across CPU cores and		* and instead have tasks which are evenly distributed across CPU cores and
* pull next item to be crunched using the threaded-aware BLI_mempool_iter.		* pull next item to be crunched using the threaded-aware BLI_mempool_iter.
*/		*/
num_tasks = num_threads + 2;		num_tasks = num_threads + 2;

state.userdata = userdata;		state.userdata = userdata;
state.func = func;		state.func = func;

BLI_mempool_iter *mempool_iterators = BLI_mempool_iter_threadsafe_create(mempool,		BLI_mempool_iter *mempool_iterators = BLI_mempool_iter_threadsafe_create(mempool,
(size_t)num_tasks);		(size_t)num_tasks);

const int thread_id = BLI_task_pool_creator_thread_id(task_pool);
for (i = 0; i < num_tasks; i++) {		for (i = 0; i < num_tasks; i++) {
/* Use this pool's pre-allocated tasks. */		/* Use this pool's pre-allocated tasks. */
BLI_task_pool_push_from_thread(		BLI_task_pool_push(task_pool, parallel_mempool_func, &mempool_iterators[i], false, NULL);
task_pool, parallel_mempool_func, &mempool_iterators[i], false, NULL, thread_id);
}		}

BLI_task_pool_work_and_wait(task_pool);		BLI_task_pool_work_and_wait(task_pool);
BLI_task_pool_free(task_pool);		BLI_task_pool_free(task_pool);

BLI_mempool_iter_threadsafe_free(mempool_iterators);		BLI_mempool_iter_threadsafe_free(mempool_iterators);
}		}