Task: Use TBB as Task Scheduler

This patch enables TBB as the default task scheduler. TBB stands for Threading Building Blocks and is developed by Intel. The library contains several threading patters. This patch maps blenders BLI_task_* function to their counterpart. After this patch we can add more patterns. A promising one is TBB:graph that can be used for depsgraph, draw manager and compositor.

Performance changes depends on the actual hardware. It was tested on different hardwares from laptops to workstations and we didn't detected any downgrade of the performance.
* Linux Xeon E5-2699 v4 got FPS boost from 12 to 17 using Spring's 04_010_A.anim.blend.
* AMD Ryzen Threadripper 2990WX 32-Core Animation playback goes from 9.5-10.5 FPS to 13.0-14.0 FPS on Agent 327 , 10_03_B.anim.blend.

Reviewed By: brecht, sergey

Differential Revision: https://developer.blender.org/D7475

8 files changed, 34 insertions, 221 deletions

diff --git a/source/blender/blenkernel/intern/editmesh_tangent.c b/source/blender/blenkernel/intern/editmesh_tangent.c
index c3ae2a54e13..6fcaf84d4ca 100644
--- a/source/blender/blenkernel/intern/editmesh_tangent.c
+++ b/source/blender/blenkernel/intern/editmesh_tangent.c
@@ -251,9 +251,7 @@ finally:
   pRes[3] = fSign;
 }
 
-static void emDM_calc_loop_tangents_thread(TaskPool *__restrict UNUSED(pool),
-                                           void *taskdata,
-                                           int UNUSED(threadid))
+static void emDM_calc_loop_tangents_thread(TaskPool *__restrict UNUSED(pool), void *taskdata)
 {
   struct SGLSLEditMeshToTangent *mesh2tangent = taskdata;
   /* new computation method */
@@ -362,9 +360,8 @@ void BKE_editmesh_loop_tangent_calc(BMEditMesh *em,
 #endif
     /* Calculation */
     if (em->tottri != 0) {
-      TaskScheduler *scheduler = BLI_task_scheduler_get();
       TaskPool *task_pool;
-      task_pool = BLI_task_pool_create(scheduler, NULL, TASK_PRIORITY_LOW);
+      task_pool = BLI_task_pool_create(NULL, TASK_PRIORITY_LOW);
 
       tangent_mask_curr = 0;
       /* Calculate tangent layers */
diff --git a/source/blender/blenkernel/intern/mesh_evaluate.c b/source/blender/blenkernel/intern/mesh_evaluate.c
index 78991ad063d..f0d19f01aab 100644
--- a/source/blender/blenkernel/intern/mesh_evaluate.c
+++ b/source/blender/blenkernel/intern/mesh_evaluate.c
@@ -1300,7 +1300,7 @@ static void loop_split_worker_do(LoopSplitTaskDataCommon *common_data,
   }
 }
 
-static void loop_split_worker(TaskPool *__restrict pool, void *taskdata, int UNUSED(threadid))
+static void loop_split_worker(TaskPool *__restrict pool, void *taskdata)
 {
   LoopSplitTaskDataCommon *common_data = BLI_task_pool_user_data(pool);
   LoopSplitTaskData *data = taskdata;
@@ -1704,11 +1704,7 @@ void BKE_mesh_normals_loop_split(const MVert *mverts,
     loop_split_generator(NULL, &common_data);
   }
   else {
-    TaskScheduler *task_scheduler;
-    TaskPool *task_pool;
-
-    task_scheduler = BLI_task_scheduler_get();
-    task_pool = BLI_task_pool_create(task_scheduler, &common_data, TASK_PRIORITY_HIGH);
+    TaskPool *task_pool = BLI_task_pool_create(&common_data, TASK_PRIORITY_HIGH);
 
     loop_split_generator(task_pool, &common_data);
 
diff --git a/source/blender/blenkernel/intern/mesh_tangent.c b/source/blender/blenkernel/intern/mesh_tangent.c
index a2a198cdb0d..d6f945cf34f 100644
--- a/source/blender/blenkernel/intern/mesh_tangent.c
+++ b/source/blender/blenkernel/intern/mesh_tangent.c
@@ -452,9 +452,7 @@ finally:
   pRes[3] = fSign;
 }
 
-static void DM_calc_loop_tangents_thread(TaskPool *__restrict UNUSED(pool),
-                                         void *taskdata,
-                                         int UNUSED(threadid))
+static void DM_calc_loop_tangents_thread(TaskPool *__restrict UNUSED(pool), void *taskdata)
 {
   struct SGLSLMeshToTangent *mesh2tangent = taskdata;
   /* new computation method */
@@ -658,9 +656,7 @@ void BKE_mesh_calc_loop_tangent_ex(const MVert *mvert,
 
     /* Calculation */
     if (looptri_len != 0) {
-      TaskScheduler *scheduler = BLI_task_scheduler_get();
-      TaskPool *task_pool;
-      task_pool = BLI_task_pool_create(scheduler, NULL, TASK_PRIORITY_LOW);
+      TaskPool *task_pool = BLI_task_pool_create(NULL, TASK_PRIORITY_LOW);
 
       tangent_mask_curr = 0;
       /* Calculate tangent layers */
diff --git a/source/blender/blenkernel/intern/ocean.c b/source/blender/blenkernel/intern/ocean.c
index 4835a4c6a62..8957628c76a 100644
--- a/source/blender/blenkernel/intern/ocean.c
+++ b/source/blender/blenkernel/intern/ocean.c
@@ -453,9 +453,7 @@ static void ocean_compute_htilda(void *__restrict userdata,
   }
 }
 
-static void ocean_compute_displacement_y(TaskPool *__restrict pool,
-                                         void *UNUSED(taskdata),
-                                         int UNUSED(threadid))
+static void ocean_compute_displacement_y(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -463,9 +461,7 @@ static void ocean_compute_displacement_y(TaskPool *__restrict pool,
   fftw_execute(o->_disp_y_plan);
 }
 
-static void ocean_compute_displacement_x(TaskPool *__restrict pool,
-                                         void *UNUSED(taskdata),
-                                         int UNUSED(threadid))
+static void ocean_compute_displacement_x(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -494,9 +490,7 @@ static void ocean_compute_displacement_x(TaskPool *__restrict pool,
   fftw_execute(o->_disp_x_plan);
 }
 
-static void ocean_compute_displacement_z(TaskPool *__restrict pool,
-                                         void *UNUSED(taskdata),
-                                         int UNUSED(threadid))
+static void ocean_compute_displacement_z(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -525,9 +519,7 @@ static void ocean_compute_displacement_z(TaskPool *__restrict pool,
   fftw_execute(o->_disp_z_plan);
 }
 
-static void ocean_compute_jacobian_jxx(TaskPool *__restrict pool,
-                                       void *UNUSED(taskdata),
-                                       int UNUSED(threadid))
+static void ocean_compute_jacobian_jxx(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -560,9 +552,7 @@ static void ocean_compute_jacobian_jxx(TaskPool *__restrict pool,
   }
 }
 
-static void ocean_compute_jacobian_jzz(TaskPool *__restrict pool,
-                                       void *UNUSED(taskdata),
-                                       int UNUSED(threadid))
+static void ocean_compute_jacobian_jzz(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -595,9 +585,7 @@ static void ocean_compute_jacobian_jzz(TaskPool *__restrict pool,
   }
 }
 
-static void ocean_compute_jacobian_jxz(TaskPool *__restrict pool,
-                                       void *UNUSED(taskdata),
-                                       int UNUSED(threadid))
+static void ocean_compute_jacobian_jxz(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -624,9 +612,7 @@ static void ocean_compute_jacobian_jxz(TaskPool *__restrict pool,
   fftw_execute(o->_Jxz_plan);
 }
 
-static void ocean_compute_normal_x(TaskPool *__restrict pool,
-                                   void *UNUSED(taskdata),
-                                   int UNUSED(threadid))
+static void ocean_compute_normal_x(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -645,9 +631,7 @@ static void ocean_compute_normal_x(TaskPool *__restrict pool,
   fftw_execute(o->_N_x_plan);
 }
 
-static void ocean_compute_normal_z(TaskPool *__restrict pool,
-                                   void *UNUSED(taskdata),
-                                   int UNUSED(threadid))
+static void ocean_compute_normal_z(TaskPool *__restrict pool, void *UNUSED(taskdata))
 {
   OceanSimulateData *osd = BLI_task_pool_user_data(pool);
   const Ocean *o = osd->o;
@@ -668,7 +652,6 @@ static void ocean_compute_normal_z(TaskPool *__restrict pool,
 
 void BKE_ocean_simulate(struct Ocean *o, float t, float scale, float chop_amount)
 {
-  TaskScheduler *scheduler = BLI_task_scheduler_get();
   TaskPool *pool;
 
   OceanSimulateData osd;
@@ -680,7 +663,7 @@ void BKE_ocean_simulate(struct Ocean *o, float t, float scale, float chop_amount
   osd.scale = scale;
   osd.chop_amount = chop_amount;
 
-  pool = BLI_task_pool_create(scheduler, &osd, TASK_PRIORITY_HIGH);
+  pool = BLI_task_pool_create(&osd, TASK_PRIORITY_HIGH);
 
   BLI_rw_mutex_lock(&o->oceanmutex, THREAD_LOCK_WRITE);
 
diff --git a/source/blender/blenkernel/intern/particle.c b/source/blender/blenkernel/intern/particle.c
index 0e35fa5d19f..48e0a493ee0 100644
--- a/source/blender/blenkernel/intern/particle.c
+++ b/source/blender/blenkernel/intern/particle.c
@@ -2787,9 +2787,7 @@ static void psys_thread_create_path(ParticleTask *task,
   }
 }
 
-static void exec_child_path_cache(TaskPool *__restrict UNUSED(pool),
-                                  void *taskdata,
-                                  int UNUSED(threadid))
+static void exec_child_path_cache(TaskPool *__restrict UNUSED(pool), void *taskdata)
 {
   ParticleTask *task = taskdata;
   ParticleThreadContext *ctx = task->ctx;
@@ -2810,7 +2808,6 @@ void psys_cache_child_paths(ParticleSimulationData *sim,
                             const bool editupdate,
                             const bool use_render_params)
 {
-  TaskScheduler *task_scheduler;
   TaskPool *task_pool;
   ParticleThreadContext ctx;
   ParticleTask *tasks_parent, *tasks_child;
@@ -2826,8 +2823,7 @@ void psys_cache_child_paths(ParticleSimulationData *sim,
     return;
   }
 
-  task_scheduler = BLI_task_scheduler_get();
-  task_pool = BLI_task_pool_create(task_scheduler, &ctx, TASK_PRIORITY_LOW);
+  task_pool = BLI_task_pool_create(&ctx, TASK_PRIORITY_LOW);
   totchild = ctx.totchild;
   totparent = ctx.totparent;
 
@@ -3377,7 +3373,6 @@ void psys_cache_edit_paths(Depsgraph *depsgraph,
 
   TaskParallelSettings settings;
   BLI_parallel_range_settings_defaults(&settings);
-  settings.scheduling_mode = TASK_SCHEDULING_DYNAMIC;
   BLI_task_parallel_range(0, edit->totpoint, &iter_data, psys_cache_edit_paths_iter, &settings);
 
   edit->totcached = totpart;
diff --git a/source/blender/blenkernel/intern/particle_distribute.c b/source/blender/blenkernel/intern/particle_distribute.c
index d91e27a92d8..7b9b2484dbe 100644
--- a/source/blender/blenkernel/intern/particle_distribute.c
+++ b/source/blender/blenkernel/intern/particle_distribute.c
@@ -773,9 +773,7 @@ static void distribute_children_exec(ParticleTask *thread, ChildParticle *cpa, i
   }
 }
 
-static void exec_distribute_parent(TaskPool *__restrict UNUSED(pool),
-                                   void *taskdata,
-                                   int UNUSED(threadid))
+static void exec_distribute_parent(TaskPool *__restrict UNUSED(pool), void *taskdata)
 {
   ParticleTask *task = taskdata;
   ParticleSystem *psys = task->ctx->sim.psys;
@@ -804,9 +802,7 @@ static void exec_distribute_parent(TaskPool *__restrict UNUSED(pool),
   }
 }
 
-static void exec_distribute_child(TaskPool *__restrict UNUSED(pool),
-                                  void *taskdata,
-                                  int UNUSED(threadid))
+static void exec_distribute_child(TaskPool *__restrict UNUSED(pool), void *taskdata)
 {
   ParticleTask *task = taskdata;
   ParticleSystem *psys = task->ctx->sim.psys;
@@ -1324,7 +1320,6 @@ static void psys_task_init_distribute(ParticleTask *task, ParticleSimulationData
 
 static void distribute_particles_on_dm(ParticleSimulationData *sim, int from)
 {
-  TaskScheduler *task_scheduler;
   TaskPool *task_pool;
   ParticleThreadContext ctx;
   ParticleTask *tasks;
@@ -1336,8 +1331,7 @@ static void distribute_particles_on_dm(ParticleSimulationData *sim, int from)
     return;
   }
 
-  task_scheduler = BLI_task_scheduler_get();
-  task_pool = BLI_task_pool_create(task_scheduler, &ctx, TASK_PRIORITY_LOW);
+  task_pool = BLI_task_pool_create(&ctx, TASK_PRIORITY_LOW);
 
   totpart = (from == PART_FROM_CHILD ? sim->psys->totchild : sim->psys->totpart);
   psys_tasks_create(&ctx, 0, totpart, &tasks, &numtasks);
diff --git a/source/blender/blenkernel/intern/pbvh.c b/source/blender/blenkernel/intern/pbvh.c
index e6d672ad9d9..aa541b3d715 100644
--- a/source/blender/blenkernel/intern/pbvh.c
+++ b/source/blender/blenkernel/intern/pbvh.c
@@ -1115,11 +1115,11 @@ static void pbvh_faces_update_normals(PBVH *bvh, PBVHNode **nodes, int totnode)
       .vnors = vnors,
   };
 
-  PBVHParallelSettings settings;
+  TaskParallelSettings settings;
   BKE_pbvh_parallel_range_settings(&settings, true, totnode);
 
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_normals_accum_task_cb, &settings);
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_normals_store_task_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_normals_accum_task_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_normals_store_task_cb, &settings);
 
   MEM_freeN(vnors);
 }
@@ -1169,9 +1169,9 @@ static void pbvh_update_mask_redraw(PBVH *bvh, PBVHNode **nodes, int totnode, in
       .flag = flag,
   };
 
-  PBVHParallelSettings settings;
+  TaskParallelSettings settings;
   BKE_pbvh_parallel_range_settings(&settings, true, totnode);
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_mask_redraw_task_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_mask_redraw_task_cb, &settings);
 }
 
 static void pbvh_update_visibility_redraw_task_cb(void *__restrict userdata,
@@ -1207,9 +1207,9 @@ static void pbvh_update_visibility_redraw(PBVH *bvh, PBVHNode **nodes, int totno
       .flag = flag,
   };
 
-  PBVHParallelSettings settings;
+  TaskParallelSettings settings;
   BKE_pbvh_parallel_range_settings(&settings, true, totnode);
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_visibility_redraw_task_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_visibility_redraw_task_cb, &settings);
 }
 
 static void pbvh_update_BB_redraw_task_cb(void *__restrict userdata,
@@ -1245,9 +1245,9 @@ void pbvh_update_BB_redraw(PBVH *bvh, PBVHNode **nodes, int totnode, int flag)
       .flag = flag,
   };
 
-  PBVHParallelSettings settings;
+  TaskParallelSettings settings;
   BKE_pbvh_parallel_range_settings(&settings, true, totnode);
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_BB_redraw_task_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_BB_redraw_task_cb, &settings);
 }
 
 static int pbvh_get_buffers_update_flags(PBVH *bvh, bool show_vcol)
@@ -1365,9 +1365,9 @@ static void pbvh_update_draw_buffers(
       .show_vcol = show_vcol,
   };
 
-  PBVHParallelSettings settings;
+  TaskParallelSettings settings;
   BKE_pbvh_parallel_range_settings(&settings, true, totnode);
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_draw_buffer_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_draw_buffer_cb, &settings);
 }
 
 static int pbvh_flush_bb(PBVH *bvh, PBVHNode *node, int flag)
@@ -1558,9 +1558,9 @@ static void pbvh_update_visibility(PBVH *bvh, PBVHNode **nodes, int totnode)
       .nodes = nodes,
   };
 
-  PBVHParallelSettings settings;
+  TaskParallelSettings settings;
   BKE_pbvh_parallel_range_settings(&settings, true, totnode);
-  BKE_pbvh_parallel_range(0, totnode, &data, pbvh_update_visibility_task_cb, &settings);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_visibility_task_cb, &settings);
 }
 
 void BKE_pbvh_update_visibility(PBVH *bvh)
@@ -2994,7 +2994,7 @@ void BKE_pbvh_get_frustum_planes(PBVH *bvh, PBVHFrustumPlanes *planes)
   }
 }
 
-void BKE_pbvh_parallel_range_settings(PBVHParallelSettings *settings,
+void BKE_pbvh_parallel_range_settings(TaskParallelSettings *settings,
                                       bool use_threading,
                                       int totnode)
 {
diff --git a/source/blender/blenkernel/intern/pbvh_parallel.cc b/source/blender/blenkernel/intern/pbvh_parallel.cc
deleted file mode 100644
index 2534fdea3ee..00000000000
--- a/source/blender/blenkernel/intern/pbvh_parallel.cc
+++ /dev/null
@@ -1,148 +0,0 @@
-/*
- * This program is free software; you can redistribute it and/or
- * modify it under the terms of the GNU General Public License
- * as published by the Free Software Foundation; either version 2
- * of the License, or (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software Foundation,
- * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
- */
-
-#include "MEM_guardedalloc.h"
-
-#include "BLI_task.h"
-#include "BLI_threads.h"
-
-#include "BKE_pbvh.h"
-
-#include "atomic_ops.h"
-
-#ifdef WITH_TBB
-
-/* Quiet top level deprecation message, unrelated to API usage here. */
-#  define TBB_SUPPRESS_DEPRECATED_MESSAGES 1
-
-#  include <tbb/tbb.h>
-
-/* Functor for running TBB parallel_for and parallel_reduce. */
-struct PBVHTask {
-  PBVHParallelRangeFunc func;
-  void *userdata;
-  const PBVHParallelSettings *settings;
-
-  void *userdata_chunk;
-
-  /* Root constructor. */
-  PBVHTask(PBVHParallelRangeFunc func, void *userdata, const PBVHParallelSettings *settings)
-      : func(func), userdata(userdata), settings(settings)
-  {
-    init_chunk(settings->userdata_chunk);
-  }
-
-  /* Copy constructor. */
-  PBVHTask(const PBVHTask &other)
-      : func(other.func), userdata(other.userdata), settings(other.settings)
-  {
-    init_chunk(other.userdata_chunk);
-  }
-
-  /* Splitting constructor for parallel reduce. */
-  PBVHTask(PBVHTask &other, tbb::split)
-      : func(other.func), userdata(other.userdata), settings(other.settings)
-  {
-    init_chunk(settings->userdata_chunk);
-  }
-
-  ~PBVHTask()
-  {
-    MEM_SAFE_FREE(userdata_chunk);
-  }
-
-  void init_chunk(void *from_chunk)
-  {
-    if (from_chunk) {
-      userdata_chunk = MEM_mallocN(settings->userdata_chunk_size, "PBVHTask");
-      memcpy(userdata_chunk, from_chunk, settings->userdata_chunk_size);
-    }
-    else {
-      userdata_chunk = NULL;
-    }
-  }
-
-  void operator()(const tbb::blocked_range<int> &r) const
-  {
-    TaskParallelTLS tls;
-    tls.thread_id = get_thread_id();
-    tls.userdata_chunk = userdata_chunk;
-    for (int i = r.begin(); i != r.end(); ++i) {
-      func(userdata, i, &tls);
-    }
-  }
-
-  void join(const PBVHTask &other)
-  {
-    settings->func_reduce(userdata, userdata_chunk, other.userdata_chunk);
-  }
-
-  int get_thread_id() const
-  {
-    /* Get a unique thread ID for texture nodes. In the future we should get rid
-     * of the thread ID and change texture evaluation to not require per-thread
-     * storage that can't be efficiently allocated on the stack. */
-    static tbb::enumerable_thread_specific<int> pbvh_thread_id(-1);
-    static int pbvh_thread_id_counter = 0;
-
-    int &thread_id = pbvh_thread_id.local();
-    if (thread_id == -1) {
-      thread_id = atomic_fetch_and_add_int32(&pbvh_thread_id_counter, 1);
-      if (thread_id >= BLENDER_MAX_THREADS) {
-        BLI_assert(!"Maximum number of threads exceeded for sculpting");
-        thread_id = thread_id % BLENDER_MAX_THREADS;
-      }
-    }
-    return thread_id;
-  }
-};
-
-#endif
-
-void BKE_pbvh_parallel_range(const int start,
-                             const int stop,
-                             void *userdata,
-                             PBVHParallelRangeFunc func,
-                             const struct PBVHParallelSettings *settings)
-{
-#ifdef WITH_TBB
-  /* Multithreading. */
-  if (settings->use_threading) {
-    PBVHTask task(func, userdata, settings);
-
-    if (settings->func_reduce) {
-      parallel_reduce(tbb::blocked_range<int>(start, stop), task);
-      if (settings->userdata_chunk) {
-        memcpy(settings->userdata_chunk, task.userdata_chunk, settings->userdata_chunk_size);
-      }
-    }
-    else {
-      parallel_for(tbb::blocked_range<int>(start, stop), task);
-    }
-
-    return;
-  }
-#endif
-
-  /* Single threaded. Nothing to reduce as everything is accumulated into the
-   * main userdata chunk directly. */
-  TaskParallelTLS tls;
-  tls.thread_id = 0;
-  tls.userdata_chunk = settings->userdata_chunk;
-  for (int i = start; i < stop; i++) {
-    func(userdata, i, &tls);
-  }
-}