Task: Separate Finalize into Reduce And Free

In preparation of TBB we need to split the finalize function into reduce
and free. Reduce is used to combine results and free for freeing any
allocated memory.

The reduce function is called to join user data chunk into another, to reduce the
result to the original userdata_chunk memory. These functions should have no side
effects so that they can be run on any thread.
The free functions should free data created during execution (TaskParallelRangeFunc).

Original patch by Brecht van Lommel
{rB61f49db843cf5095203112226ae386f301be1e1a}.

Reviewed By: Brecht van Lommel, Bastien Montagne

Differential Revision: https://developer.blender.org/D7394

5 files changed, 69 insertions, 72 deletions

diff --git a/source/blender/blenkernel/intern/colortools.c b/source/blender/blenkernel/intern/colortools.c
index f82b8b6675c..3da384a2745 100644
--- a/source/blender/blenkernel/intern/colortools.c
+++ b/source/blender/blenkernel/intern/colortools.c
@@ -1383,8 +1383,6 @@ typedef struct ScopesUpdateData {
   struct ColormanageProcessor *cm_processor;
   const unsigned char *display_buffer;
   const int ycc_mode;
-
-  unsigned int *bin_lum, *bin_r, *bin_g, *bin_b, *bin_a;
 } ScopesUpdateData;
 
 typedef struct ScopesUpdateDataChunk {
@@ -1495,23 +1493,24 @@ static void scopes_update_cb(void *__restrict userdata,
   }
 }
 
-static void scopes_update_finalize(void *__restrict userdata, void *__restrict userdata_chunk)
+static void scopes_update_reduce(const void *__restrict UNUSED(userdata),
+                                 void *__restrict chunk_join,
+                                 void *__restrict chunk)
 {
-  const ScopesUpdateData *data = userdata;
-  const ScopesUpdateDataChunk *data_chunk = userdata_chunk;
-
-  unsigned int *bin_lum = data->bin_lum;
-  unsigned int *bin_r = data->bin_r;
-  unsigned int *bin_g = data->bin_g;
-  unsigned int *bin_b = data->bin_b;
-  unsigned int *bin_a = data->bin_a;
+  ScopesUpdateDataChunk *join_chunk = chunk_join;
+  const ScopesUpdateDataChunk *data_chunk = chunk;
+
+  unsigned int *bin_lum = join_chunk->bin_lum;
+  unsigned int *bin_r = join_chunk->bin_r;
+  unsigned int *bin_g = join_chunk->bin_g;
+  unsigned int *bin_b = join_chunk->bin_b;
+  unsigned int *bin_a = join_chunk->bin_a;
   const unsigned int *bin_lum_c = data_chunk->bin_lum;
   const unsigned int *bin_r_c = data_chunk->bin_r;
   const unsigned int *bin_g_c = data_chunk->bin_g;
   const unsigned int *bin_b_c = data_chunk->bin_b;
   const unsigned int *bin_a_c = data_chunk->bin_a;
 
-  float(*minmax)[2] = data->scopes->minmax;
   const float *min = data_chunk->min;
   const float *max = data_chunk->max;
 
@@ -1524,11 +1523,11 @@ static void scopes_update_finalize(void *__restrict userdata, void *__restrict u
   }
 
   for (int c = 3; c--;) {
-    if (min[c] < minmax[c][0]) {
-      minmax[c][0] = min[c];
+    if (min[c] < join_chunk->min[c]) {
+      join_chunk->min[c] = min[c];
     }
-    if (max[c] > minmax[c][1]) {
-      minmax[c][1] = max[c];
+    if (max[c] > join_chunk->max[c]) {
+      join_chunk->max[c] = max[c];
     }
   }
 }
@@ -1542,7 +1541,6 @@ void BKE_scopes_update(Scopes *scopes,
   unsigned int nl, na, nr, ng, nb;
   double divl, diva, divr, divg, divb;
   const unsigned char *display_buffer = NULL;
-  uint bin_lum[256] = {0}, bin_r[256] = {0}, bin_g[256] = {0}, bin_b[256] = {0}, bin_a[256] = {0};
   int ycc_mode = -1;
   void *cache_handle = NULL;
   struct ColormanageProcessor *cm_processor = NULL;
@@ -1638,11 +1636,6 @@ void BKE_scopes_update(Scopes *scopes,
       .cm_processor = cm_processor,
       .display_buffer = display_buffer,
       .ycc_mode = ycc_mode,
-      .bin_lum = bin_lum,
-      .bin_r = bin_r,
-      .bin_g = bin_g,
-      .bin_b = bin_b,
-      .bin_a = bin_a,
   };
   ScopesUpdateDataChunk data_chunk = {{0}};
   INIT_MINMAX(data_chunk.min, data_chunk.max);
@@ -1652,26 +1645,26 @@ void BKE_scopes_update(Scopes *scopes,
   settings.use_threading = (ibuf->y > 256);
   settings.userdata_chunk = &data_chunk;
   settings.userdata_chunk_size = sizeof(data_chunk);
-  settings.func_finalize = scopes_update_finalize;
+  settings.func_reduce = scopes_update_reduce;
   BLI_task_parallel_range(0, ibuf->y, &data, scopes_update_cb, &settings);
 
   /* convert hist data to float (proportional to max count) */
   nl = na = nr = nb = ng = 0;
   for (a = 0; a < 256; a++) {
-    if (bin_lum[a] > nl) {
-      nl = bin_lum[a];
+    if (data_chunk.bin_lum[a] > nl) {
+      nl = data_chunk.bin_lum[a];
     }
-    if (bin_r[a] > nr) {
-      nr = bin_r[a];
+    if (data_chunk.bin_r[a] > nr) {
+      nr = data_chunk.bin_r[a];
     }
-    if (bin_g[a] > ng) {
-      ng = bin_g[a];
+    if (data_chunk.bin_g[a] > ng) {
+      ng = data_chunk.bin_g[a];
     }
-    if (bin_b[a] > nb) {
-      nb = bin_b[a];
+    if (data_chunk.bin_b[a] > nb) {
+      nb = data_chunk.bin_b[a];
     }
-    if (bin_a[a] > na) {
-      na = bin_a[a];
+    if (data_chunk.bin_a[a] > na) {
+      na = data_chunk.bin_a[a];
     }
   }
   divl = nl ? 1.0 / (double)nl : 1.0;
@@ -1681,11 +1674,11 @@ void BKE_scopes_update(Scopes *scopes,
   divb = nb ? 1.0 / (double)nb : 1.0;
 
   for (a = 0; a < 256; a++) {
-    scopes->hist.data_luma[a] = bin_lum[a] * divl;
-    scopes->hist.data_r[a] = bin_r[a] * divr;
-    scopes->hist.data_g[a] = bin_g[a] * divg;
-    scopes->hist.data_b[a] = bin_b[a] * divb;
-    scopes->hist.data_a[a] = bin_a[a] * diva;
+    scopes->hist.data_luma[a] = data_chunk.bin_lum[a] * divl;
+    scopes->hist.data_r[a] = data_chunk.bin_r[a] * divr;
+    scopes->hist.data_g[a] = data_chunk.bin_g[a] * divg;
+    scopes->hist.data_b[a] = data_chunk.bin_b[a] * divb;
+    scopes->hist.data_a[a] = data_chunk.bin_a[a] * diva;
   }
 
   if (cm_processor) {
diff --git a/source/blender/blenkernel/intern/dynamicpaint.c b/source/blender/blenkernel/intern/dynamicpaint.c
index 4c78c88d168..3a8b846a41d 100644
--- a/source/blender/blenkernel/intern/dynamicpaint.c
+++ b/source/blender/blenkernel/intern/dynamicpaint.c
@@ -653,15 +653,15 @@ static void grid_bound_insert_cb_ex(void *__restrict userdata,
   boundInsert(grid_bound, bData->realCoord[bData->s_pos[i]].v);
 }
 
-static void grid_bound_insert_finalize(void *__restrict userdata, void *__restrict userdata_chunk)
+static void grid_bound_insert_reduce(const void *__restrict UNUSED(userdata),
+                                     void *__restrict chunk_join,
+                                     void *__restrict chunk)
 {
-  PaintBakeData *bData = userdata;
-  VolumeGrid *grid = bData->grid;
-
-  Bounds3D *grid_bound = userdata_chunk;
+  Bounds3D *join = chunk_join;
+  Bounds3D *grid_bound = chunk;
 
-  boundInsert(&grid->grid_bounds, grid_bound->min);
-  boundInsert(&grid->grid_bounds, grid_bound->max);
+  boundInsert(join, grid_bound->min);
+  boundInsert(join, grid_bound->max);
 }
 
 static void grid_cell_points_cb_ex(void *__restrict userdata,
@@ -685,17 +685,20 @@ static void grid_cell_points_cb_ex(void *__restrict userdata,
   s_num[temp_t_index[i]]++;
 }
 
-static void grid_cell_points_finalize(void *__restrict userdata, void *__restrict userdata_chunk)
+static void grid_cell_points_reduce(const void *__restrict userdata,
+                                    void *__restrict chunk_join,
+                                    void *__restrict chunk)
 {
-  PaintBakeData *bData = userdata;
-  VolumeGrid *grid = bData->grid;
+  const PaintBakeData *bData = userdata;
+  const VolumeGrid *grid = bData->grid;
   const int grid_cells = grid->dim[0] * grid->dim[1] * grid->dim[2];
 
-  int *s_num = userdata_chunk;
+  int *join_s_num = chunk_join;
+  int *s_num = chunk;
 
   /* calculate grid indexes */
   for (int i = 0; i < grid_cells; i++) {
-    grid->s_num[i] += s_num[i];
+    join_s_num[i] += s_num[i];
   }
 }
 
@@ -753,7 +756,7 @@ static void surfaceGenerateGrid(struct DynamicPaintSurface *surface)
       settings.use_threading = (sData->total_points > 1000);
       settings.userdata_chunk = &grid->grid_bounds;
       settings.userdata_chunk_size = sizeof(grid->grid_bounds);
-      settings.func_finalize = grid_bound_insert_finalize;
+      settings.func_reduce = grid_bound_insert_reduce;
       BLI_task_parallel_range(0, sData->total_points, bData, grid_bound_insert_cb_ex, &settings);
     }
     /* get dimensions */
@@ -814,7 +817,7 @@ static void surfaceGenerateGrid(struct DynamicPaintSurface *surface)
         settings.use_threading = (sData->total_points > 1000);
         settings.userdata_chunk = grid->s_num;
         settings.userdata_chunk_size = sizeof(*grid->s_num) * grid_cells;
-        settings.func_finalize = grid_cell_points_finalize;
+        settings.func_reduce = grid_cell_points_reduce;
         BLI_task_parallel_range(0, sData->total_points, bData, grid_cell_points_cb_ex, &settings);
       }
 
@@ -4880,7 +4883,7 @@ static void dynamicPaint_prepareAdjacencyData(DynamicPaintSurface *surface, cons
       0, sData->total_points, sData, dynamic_paint_prepare_adjacency_cb, &settings);
 
   /* calculate average values (single thread).
-   * Note: tried to put this in threaded callback (using _finalize feature),
+   * Note: tried to put this in threaded callback (using _reduce feature),
    * but gave ~30% slower result! */
   bData->average_dist = 0.0;
   for (index = 0; index < sData->total_points; index++) {
diff --git a/source/blender/blenkernel/intern/particle_system.c b/source/blender/blenkernel/intern/particle_system.c
index 5ef2f7aeeff..14b1ef7b87f 100644
--- a/source/blender/blenkernel/intern/particle_system.c
+++ b/source/blender/blenkernel/intern/particle_system.c
@@ -3692,10 +3692,11 @@ typedef struct DynamicStepSolverTaskData {
   SpinLock spin;
 } DynamicStepSolverTaskData;
 
-static void dynamics_step_finalize_sphdata(void *__restrict UNUSED(userdata),
-                                           void *__restrict tls_userdata_chunk)
+static void dynamics_step_sphdata_reduce(const void *__restrict UNUSED(userdata),
+                                         void *__restrict UNUSED(join_v),
+                                         void *__restrict chunk_v)
 {
-  SPHData *sphdata = tls_userdata_chunk;
+  SPHData *sphdata = chunk_v;
 
   psys_sph_flush_springs(sphdata);
 }
@@ -3986,7 +3987,7 @@ static void dynamics_step(ParticleSimulationData *sim, float cfra)
         settings.use_threading = (psys->totpart > 100);
         settings.userdata_chunk = &sphdata;
         settings.userdata_chunk_size = sizeof(sphdata);
-        settings.func_finalize = dynamics_step_finalize_sphdata;
+        settings.func_reduce = dynamics_step_sphdata_reduce;
         BLI_task_parallel_range(
             0, psys->totpart, &task_data, dynamics_step_sph_ddr_task_cb_ex, &settings);
 
@@ -4018,7 +4019,7 @@ static void dynamics_step(ParticleSimulationData *sim, float cfra)
           settings.use_threading = (psys->totpart > 100);
           settings.userdata_chunk = &sphdata;
           settings.userdata_chunk_size = sizeof(sphdata);
-          settings.func_finalize = dynamics_step_finalize_sphdata;
+          settings.func_reduce = dynamics_step_sphdata_reduce;
           BLI_task_parallel_range(0,
                                   psys->totpart,
                                   &task_data,
@@ -4033,7 +4034,7 @@ static void dynamics_step(ParticleSimulationData *sim, float cfra)
           settings.use_threading = (psys->totpart > 100);
           settings.userdata_chunk = &sphdata;
           settings.userdata_chunk_size = sizeof(sphdata);
-          settings.func_finalize = dynamics_step_finalize_sphdata;
+          settings.func_reduce = dynamics_step_sphdata_reduce;
           BLI_task_parallel_range(0,
                                   psys->totpart,
                                   &task_data,
@@ -4189,7 +4190,7 @@ static void particles_fluid_step(ParticleSimulationData *sim,
       ParticleSettings *part = psys->part;
       ParticleData *pa = NULL;
 
-      int p, totpart, tottypepart = 0;
+      int p, totpart = 0, tottypepart = 0;
       int flagActivePart, activeParts = 0;
       float posX, posY, posZ, velX, velY, velZ;
       float resX, resY, resZ;
diff --git a/source/blender/blenkernel/intern/subdiv_ccg.c b/source/blender/blenkernel/intern/subdiv_ccg.c
index 521aeb60e66..d99c41eaa3e 100644
--- a/source/blender/blenkernel/intern/subdiv_ccg.c
+++ b/source/blender/blenkernel/intern/subdiv_ccg.c
@@ -770,8 +770,8 @@ static void subdiv_ccg_recalc_inner_normal_task(void *__restrict userdata_v,
   subdiv_ccg_average_inner_face_normals(data->subdiv_ccg, data->key, tls, grid_index);
 }
 
-static void subdiv_ccg_recalc_inner_normal_finalize(void *__restrict UNUSED(userdata),
-                                                    void *__restrict tls_v)
+static void subdiv_ccg_recalc_inner_normal_free(const void *__restrict UNUSED(userdata),
+                                                void *__restrict tls_v)
 {
   RecalcInnerNormalsTLSData *tls = tls_v;
   MEM_SAFE_FREE(tls->face_normals);
@@ -791,7 +791,7 @@ static void subdiv_ccg_recalc_inner_grid_normals(SubdivCCG *subdiv_ccg)
   BLI_parallel_range_settings_defaults(&parallel_range_settings);
   parallel_range_settings.userdata_chunk = &tls_data;
   parallel_range_settings.userdata_chunk_size = sizeof(tls_data);
-  parallel_range_settings.func_finalize = subdiv_ccg_recalc_inner_normal_finalize;
+  parallel_range_settings.func_free = subdiv_ccg_recalc_inner_normal_free;
   BLI_task_parallel_range(0,
                           subdiv_ccg->num_grids,
                           &data,
@@ -834,8 +834,8 @@ static void subdiv_ccg_recalc_modified_inner_normal_task(void *__restrict userda
   subdiv_ccg_average_inner_face_grids(subdiv_ccg, key, face);
 }
 
-static void subdiv_ccg_recalc_modified_inner_normal_finalize(void *__restrict UNUSED(userdata),
-                                                             void *__restrict tls_v)
+static void subdiv_ccg_recalc_modified_inner_normal_free(const void *__restrict UNUSED(userdata),
+                                                         void *__restrict tls_v)
 {
   RecalcInnerNormalsTLSData *tls = tls_v;
   MEM_SAFE_FREE(tls->face_normals);
@@ -857,7 +857,7 @@ static void subdiv_ccg_recalc_modified_inner_grid_normals(SubdivCCG *subdiv_ccg,
   BLI_parallel_range_settings_defaults(&parallel_range_settings);
   parallel_range_settings.userdata_chunk = &tls_data;
   parallel_range_settings.userdata_chunk_size = sizeof(tls_data);
-  parallel_range_settings.func_finalize = subdiv_ccg_recalc_modified_inner_normal_finalize;
+  parallel_range_settings.func_free = subdiv_ccg_recalc_modified_inner_normal_free;
   BLI_task_parallel_range(0,
                           num_effected_faces,
                           &data,
@@ -1077,8 +1077,8 @@ static void subdiv_ccg_average_grids_boundaries_task(void *__restrict userdata_v
   subdiv_ccg_average_grids_boundary(subdiv_ccg, key, adjacent_edge, tls);
 }
 
-static void subdiv_ccg_average_grids_boundaries_finalize(void *__restrict UNUSED(userdata),
-                                                         void *__restrict tls_v)
+static void subdiv_ccg_average_grids_boundaries_free(const void *__restrict UNUSED(userdata),
+                                                     void *__restrict tls_v)
 {
   AverageGridsBoundariesTLSData *tls = tls_v;
   MEM_SAFE_FREE(tls->accumulators);
@@ -1136,7 +1136,7 @@ static void subdiv_ccg_average_all_boundaries(SubdivCCG *subdiv_ccg, CCGKey *key
   AverageGridsBoundariesTLSData tls_data = {NULL};
   parallel_range_settings.userdata_chunk = &tls_data;
   parallel_range_settings.userdata_chunk_size = sizeof(tls_data);
-  parallel_range_settings.func_finalize = subdiv_ccg_average_grids_boundaries_finalize;
+  parallel_range_settings.func_free = subdiv_ccg_average_grids_boundaries_free;
   BLI_task_parallel_range(0,
                           subdiv_ccg->num_adjacent_edges,
                           &boundaries_data,
diff --git a/source/blender/blenkernel/intern/subdiv_foreach.c b/source/blender/blenkernel/intern/subdiv_foreach.c
index b31fb2c9312..0884f40952f 100644
--- a/source/blender/blenkernel/intern/subdiv_foreach.c
+++ b/source/blender/blenkernel/intern/subdiv_foreach.c
@@ -1838,9 +1838,9 @@ static void subdiv_foreach_boundary_edges_task(void *__restrict userdata,
   subdiv_foreach_boundary_edges(ctx, tls->userdata_chunk, edge_index);
 }
 
-static void subdiv_foreach_finalize(void *__restrict userdata, void *__restrict userdata_chunk)
+static void subdiv_foreach_free(const void *__restrict userdata, void *__restrict userdata_chunk)
 {
-  SubdivForeachTaskContext *ctx = userdata;
+  const SubdivForeachTaskContext *ctx = userdata;
   ctx->foreach_context->user_data_tls_free(userdata_chunk);
 }
 
@@ -1873,7 +1873,7 @@ bool BKE_subdiv_foreach_subdiv_geometry(Subdiv *subdiv,
   parallel_range_settings.userdata_chunk_size = context->user_data_tls_size;
   parallel_range_settings.min_iter_per_thread = 1;
   if (context->user_data_tls_free != NULL) {
-    parallel_range_settings.func_finalize = subdiv_foreach_finalize;
+    parallel_range_settings.func_free = subdiv_foreach_free;
   }
 
   /* TODO(sergey): Possible optimization is to have a single pool and push all