Add a new parallel looper for MemPool items to BLI_task.

It merely uses the new thread-safe iterators system of mempool, quite
straight forward.

Note that to avoid possible confusion with two void pointers as
parameters of the callback, a dummy opaque struct pointer is used
instead for the second parameter (pointer generated by iteration over
mempool), callback functions must explicitely convert it to expected
real type.

Also added a basic gtest for this new feature.

source/blender/blenlib/BLI_task.h

@@ -35,6 +35,8 @@ extern "C" {
 #include "BLI_threads.h"
 #include "BLI_utildefines.h"
 
+struct BLI_mempool;
+
 /* Task Scheduler
  * 
  * Central scheduler that holds running threads ready to execute tasks. A single
@@ -150,6 +152,15 @@ void BLI_task_parallel_listbase(
         TaskParallelListbaseFunc func,
         const bool use_threading);
 
+typedef struct MempoolIterData MempoolIterData;
+typedef void (*TaskParallelMempoolFunc)(void *userdata,
+                                        MempoolIterData *iter);
+void BLI_task_parallel_mempool(
+        struct BLI_mempool *mempool,
+        void *userdata,
+        TaskParallelMempoolFunc func,
+        const bool use_threading);
+
 #ifdef __cplusplus
 }
 #endif

source/blender/blenlib/intern/task.c

@@ -32,6 +32,7 @@
 
 #include "BLI_listbase.h"
 #include "BLI_math.h"
+#include "BLI_mempool.h"
 #include "BLI_task.h"
 #include "BLI_threads.h"
 
@@ -1354,3 +1355,89 @@ void BLI_task_parallel_listbase(
 
 	BLI_spin_end(&state.lock);
 }
+
+
+typedef struct ParallelMempoolState {
+	void *userdata;
+	TaskParallelMempoolFunc func;
+} ParallelMempoolState;
+
+static void parallel_mempool_func(
+        TaskPool * __restrict pool,
+        void *taskdata,
+        int UNUSED(threadid))
+{
+	ParallelMempoolState * __restrict state = BLI_task_pool_userdata(pool);
+	BLI_mempool_iter *iter = taskdata;
+	MempoolIterData *item;
+
+	while ((item = BLI_mempool_iterstep(iter)) != NULL) {
+		state->func(state->userdata, item);
+	}
+}
+
+/**
+ * This function allows to parallelize for loops over Mempool items.
+ *
+ * \param pool The iterable BLI_mempool to loop over.
+ * \param userdata Common userdata passed to all instances of \a func.
+ * \param func Callback function.
+ * \param use_threading If \a true, actually split-execute loop in threads, else just do a sequential forloop
+ *                      (allows caller to use any kind of test to switch on parallelization or not).
+ *
+ * \note There is no static scheduling here.
+ */
+void BLI_task_parallel_mempool(
+        BLI_mempool *mempool,
+        void *userdata,
+        TaskParallelMempoolFunc func,
+        const bool use_threading)
+{
+	TaskScheduler *task_scheduler;
+	TaskPool *task_pool;
+	ParallelMempoolState state;
+	int i, num_threads, num_tasks;
+
+	if (BLI_mempool_count(mempool) == 0) {
+		return;
+	}
+
+	if (!use_threading) {
+		BLI_mempool_iter iter;
+		BLI_mempool_iternew(mempool, &iter);
+
+		for (void *item = BLI_mempool_iterstep(&iter); item != NULL; item = BLI_mempool_iterstep(&iter)) {
+			func(userdata, item);
+		}
+		return;
+	}
+
+	task_scheduler = BLI_task_scheduler_get();
+	task_pool = BLI_task_pool_create(task_scheduler, &state);
+	num_threads = BLI_task_scheduler_num_threads(task_scheduler);
+
+	/* 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
+	 * pull next item to be crunched using the threaded-aware BLI_mempool_iter.
+	 */
+	num_tasks = num_threads * 2;
+
+	state.userdata = userdata;
+	state.func = func;
+
+	BLI_mempool_iter *mempool_iterators = BLI_mempool_iter_threadsafe_create(mempool, (size_t)num_tasks);
+
+	for (i = 0; i < num_tasks; i++) {
+		/* Use this pool's pre-allocated tasks. */
+		BLI_task_pool_push_from_thread(task_pool,
+		                               parallel_mempool_func,
+		                               &mempool_iterators[i], false,
+		                               TASK_PRIORITY_HIGH,
+		                               task_pool->thread_id);
+	}
+
+	BLI_task_pool_work_and_wait(task_pool);
+	BLI_task_pool_free(task_pool);
+
+	BLI_mempool_iter_threadsafe_free(mempool_iterators);
+}

tests/gtests/blenlib/BLI_task_test.cc

@@ -0,0 +1,76 @@
+/* Apache License, Version 2.0 */
+
+#include "testing/testing.h"
+#include <string.h>
+
+#include "atomic_ops.h"
+
+extern "C" {
+#include "BLI_mempool.h"
+#include "BLI_task.h"
+#include "BLI_utildefines.h"
+};
+
+#define NUM_ITEMS 10000
+
+static void task_mempool_iter_func(void *userdata, MempoolIterData *item) {
+	int *data = (int *)item;
+	int *count = (int *)userdata;
+
+	EXPECT_TRUE(data != NULL);
+
+	*data += 1;
+	atomic_sub_and_fetch_uint32((uint32_t *)count, 1);
+}
+
+TEST(task, MempoolIter)
+{
+	int *data[NUM_ITEMS];
+	BLI_mempool *mempool = BLI_mempool_create(sizeof(*data[0]), NUM_ITEMS, 32, BLI_MEMPOOL_ALLOW_ITER);
+
+	int i;
+
+	/* 'Randomly' add and remove some items from mempool, to create a non-homogenous one. */
+	int num_items = 0;
+	for (i = 0; i < NUM_ITEMS; i++) {
+		data[i] = (int *)BLI_mempool_alloc(mempool);
+		*data[i] = i - 1;
+		num_items++;
+	}
+
+	for (i = 0; i < NUM_ITEMS; i += 3) {
+		BLI_mempool_free(mempool, data[i]);
+		data[i] = NULL;
+		num_items--;
+	}
+
+	for (i = 0; i < NUM_ITEMS; i += 7) {
+		if (data[i] == NULL) {
+			data[i] = (int *)BLI_mempool_alloc(mempool);
+			*data[i] = i - 1;
+			num_items++;
+		}
+	}
+
+	for (i = 0; i < NUM_ITEMS - 5; i += 23) {
+		for (int j = 0; j < 5; j++) {
+			if (data[i + j] != NULL) {
+				BLI_mempool_free(mempool, data[i + j]);
+				data[i + j] = NULL;
+				num_items--;
+			}
+		}
+	}
+
+	BLI_task_parallel_mempool(mempool, &num_items, task_mempool_iter_func, true);
+
+	/* Those checks should ensure us all items of the mempool were processed once, and only once - as expected. */
+	EXPECT_EQ(num_items, 0);
+	for (i = 0; i < NUM_ITEMS; i++) {
+		if (data[i] != NULL) {
+			EXPECT_EQ(*data[i], i);
+		}
+	}
+
+	BLI_mempool_destroy(mempool);
+}

tests/gtests/blenlib/CMakeLists.txt

@@ -27,6 +27,7 @@ set(INC
 	../../../source/blender/blenlib
 	../../../source/blender/makesdna
 	../../../intern/guardedalloc
+	../../../intern/atomic
 )
 
 include_directories(${INC})
@@ -55,6 +56,7 @@ BLENDER_TEST(BLI_polyfill2d "bf_blenlib")
 BLENDER_TEST(BLI_stack "bf_blenlib")
 BLENDER_TEST(BLI_string "bf_blenlib")
 BLENDER_TEST(BLI_string_utf8 "bf_blenlib")
+BLENDER_TEST(BLI_task "bf_blenlib")
 
 BLENDER_TEST_PERFORMANCE(BLI_ghash_performance "bf_blenlib")