Sculpt: multithread GPU draw buffer filling for workbench

This improves performance of some sculpt tools, particularly those that modify
many vertices like filter and mask tools, or use brushes with large radius.

For mask expand it can make updates up to 2x faster on heavy meshes, but for
most tools it's more on the order of 1-1.1x. There are bigger bottlenecks to
solve, like normal updates.

Ref T70295

Differential Revision: https://developer.blender.org/D5926

1 files changed, 147 insertions, 100 deletions

diff --git a/source/blender/blenkernel/intern/pbvh.c b/source/blender/blenkernel/intern/pbvh.c
index bfc513454b4..50ea4fb0b36 100644
--- a/source/blender/blenkernel/intern/pbvh.c
+++ b/source/blender/blenkernel/intern/pbvh.c
@@ -988,6 +988,7 @@ typedef struct PBVHUpdateData {
 
   float (*vnors)[3];
   int flag;
+  bool show_vcol;
 } PBVHUpdateData;
 
 static void pbvh_update_normals_accum_task_cb(void *__restrict userdata,
@@ -1148,71 +1149,105 @@ static int pbvh_get_buffers_update_flags(PBVH *bvh, bool show_vcol)
   return update_flags;
 }
 
-static void pbvh_update_draw_buffers(PBVH *bvh, PBVHNode **nodes, int totnode, bool show_vcol)
-{
-  /* can't be done in parallel with OpenGL */
-  for (int n = 0; n < totnode; n++) {
-    PBVHNode *node = nodes[n];
-
-    if (node->flag & PBVH_RebuildDrawBuffers) {
-      GPU_pbvh_buffers_free(node->draw_buffers);
-      switch (bvh->type) {
-        case PBVH_GRIDS:
-          node->draw_buffers = GPU_pbvh_grid_buffers_build(node->totprim, bvh->grid_hidden);
-          break;
-        case PBVH_FACES:
-          node->draw_buffers = GPU_pbvh_mesh_buffers_build(node->face_vert_indices,
-                                                           bvh->mpoly,
-                                                           bvh->mloop,
-                                                           bvh->looptri,
-                                                           bvh->verts,
-                                                           node->prim_indices,
-                                                           node->totprim);
-          break;
-        case PBVH_BMESH:
-          node->draw_buffers = GPU_pbvh_bmesh_buffers_build(bvh->flags &
-                                                            PBVH_DYNTOPO_SMOOTH_SHADING);
-          break;
-      }
-
-      node->flag &= ~PBVH_RebuildDrawBuffers;
-    }
+static void pbvh_update_draw_buffer_cb(void *__restrict userdata,
+                                       const int n,
+                                       const TaskParallelTLS *__restrict UNUSED(tls))
+{
+  /* Create and update draw buffers. The functions called here must not
+   * do any OpenGL calls. Flags are not cleared immediately, that happens
+   * after GPU_pbvh_buffer_flush() which does the final OpenGL calls. */
+  PBVHUpdateData *data = userdata;
+  PBVH *bvh = data->bvh;
+  PBVHNode *node = data->nodes[n];
 
-    if (node->flag & PBVH_UpdateDrawBuffers) {
-      const int update_flags = pbvh_get_buffers_update_flags(bvh, show_vcol);
-      switch (bvh->type) {
-        case PBVH_GRIDS:
-          GPU_pbvh_grid_buffers_update(node->draw_buffers,
-                                       bvh->grids,
-                                       bvh->grid_flag_mats,
-                                       node->prim_indices,
-                                       node->totprim,
-                                       &bvh->gridkey,
-                                       update_flags);
-          break;
-        case PBVH_FACES:
-          GPU_pbvh_mesh_buffers_update(node->draw_buffers,
-                                       bvh->verts,
-                                       node->vert_indices,
-                                       node->uniq_verts + node->face_verts,
-                                       CustomData_get_layer(bvh->vdata, CD_PAINT_MASK),
-                                       CustomData_get_layer(bvh->ldata, CD_MLOOPCOL),
-                                       node->face_vert_indices,
-                                       update_flags);
-          break;
-        case PBVH_BMESH:
-          GPU_pbvh_bmesh_buffers_update(node->draw_buffers,
-                                        bvh->bm,
-                                        node->bm_faces,
-                                        node->bm_unique_verts,
-                                        node->bm_other_verts,
-                                        update_flags);
-          break;
+  if (node->flag & PBVH_RebuildDrawBuffers) {
+    switch (bvh->type) {
+      case PBVH_GRIDS:
+        node->draw_buffers = GPU_pbvh_grid_buffers_build(node->totprim, bvh->grid_hidden);
+        break;
+      case PBVH_FACES:
+        node->draw_buffers = GPU_pbvh_mesh_buffers_build(node->face_vert_indices,
+                                                         bvh->mpoly,
+                                                         bvh->mloop,
+                                                         bvh->looptri,
+                                                         bvh->verts,
+                                                         node->prim_indices,
+                                                         node->totprim);
+        break;
+      case PBVH_BMESH:
+        node->draw_buffers = GPU_pbvh_bmesh_buffers_build(bvh->flags &
+                                                          PBVH_DYNTOPO_SMOOTH_SHADING);
+        break;
+    }
+  }
+
+  if (node->flag & PBVH_UpdateDrawBuffers) {
+    const int update_flags = pbvh_get_buffers_update_flags(bvh, data->show_vcol);
+    switch (bvh->type) {
+      case PBVH_GRIDS:
+        GPU_pbvh_grid_buffers_update(node->draw_buffers,
+                                     bvh->grids,
+                                     bvh->grid_flag_mats,
+                                     node->prim_indices,
+                                     node->totprim,
+                                     &bvh->gridkey,
+                                     update_flags);
+        break;
+      case PBVH_FACES:
+        GPU_pbvh_mesh_buffers_update(node->draw_buffers,
+                                     bvh->verts,
+                                     node->vert_indices,
+                                     node->uniq_verts + node->face_verts,
+                                     CustomData_get_layer(bvh->vdata, CD_PAINT_MASK),
+                                     CustomData_get_layer(bvh->ldata, CD_MLOOPCOL),
+                                     node->face_vert_indices,
+                                     update_flags);
+        break;
+      case PBVH_BMESH:
+        GPU_pbvh_bmesh_buffers_update(node->draw_buffers,
+                                      bvh->bm,
+                                      node->bm_faces,
+                                      node->bm_unique_verts,
+                                      node->bm_other_verts,
+                                      update_flags);
+        break;
+    }
+  }
+}
+
+static void pbvh_update_draw_buffers(
+    PBVH *bvh, PBVHNode **nodes, int totnode, bool show_vcol, int update_flag)
+{
+  if ((update_flag & PBVH_RebuildDrawBuffers) || ELEM(bvh->type, PBVH_GRIDS, PBVH_BMESH)) {
+    /* Free buffers uses OpenGL, so not in parallel. */
+    for (int n = 0; n < totnode; n++) {
+      PBVHNode *node = nodes[n];
+      if (node->flag & PBVH_RebuildDrawBuffers) {
+        GPU_pbvh_buffers_free(node->draw_buffers);
+        node->draw_buffers = NULL;
+      }
+      else if ((node->flag & PBVH_UpdateDrawBuffers) && node->draw_buffers) {
+        if (bvh->type == PBVH_GRIDS) {
+          GPU_pbvh_grid_buffers_update_free(
+              node->draw_buffers, bvh->grid_flag_mats, node->prim_indices);
+        }
+        else if (bvh->type == PBVH_BMESH) {
+          GPU_pbvh_bmesh_buffers_update_free(node->draw_buffers);
+        }
       }
-
-      node->flag &= ~PBVH_UpdateDrawBuffers;
     }
   }
+
+  /* Parallel creation and update of draw buffers. */
+  PBVHUpdateData data = {
+      .bvh = bvh,
+      .nodes = nodes,
+      .show_vcol = show_vcol,
+  };
+
+  TaskParallelSettings settings;
+  BKE_pbvh_parallel_range_settings(&settings, true, totnode);
+  BLI_task_parallel_range(0, totnode, &data, pbvh_update_draw_buffer_cb, &settings);
 }
 
 static int pbvh_flush_bb(PBVH *bvh, PBVHNode *node, int flag)
@@ -1568,8 +1603,8 @@ void BKE_pbvh_node_get_bm_orco_data(PBVHNode *node,
 
 /**
  * \note doing a full search on all vertices here seems expensive,
- * however this is important to avoid having to recalculate boundbox & sync the buffers to the GPU
- * (which is far more expensive!) See: T47232.
+ * however this is important to avoid having to recalculate boundbox & sync the buffers to the
+ * GPU (which is far more expensive!) See: T47232.
  */
 bool BKE_pbvh_node_vert_update_check_any(PBVH *bvh, PBVHNode *node)
 {
@@ -2216,20 +2251,6 @@ bool BKE_pbvh_node_frustum_exclude_AABB(PBVHNode *node, void *data)
   return test_frustum_aabb(bb_min, bb_max, data) != ISECT_INSIDE;
 }
 
-typedef struct PBVHNodeDrawCallbackData {
-  void (*draw_fn)(void *user_data, GPU_PBVH_Buffers *buffers);
-  void *user_data;
-} PBVHNodeDrawCallbackData;
-
-static void pbvh_node_draw_cb(PBVHNode *node, void *data_v)
-{
-  PBVHNodeDrawCallbackData *data = data_v;
-
-  if (!(node->flag & PBVH_FullyHidden)) {
-    data->draw_fn(data->user_data, node->draw_buffers);
-  }
-}
-
 void BKE_pbvh_update_normals(PBVH *bvh, struct SubdivCCG *subdiv_ccg)
 {
   /* Update normals */
@@ -2260,41 +2281,67 @@ void BKE_pbvh_update_normals(PBVH *bvh, struct SubdivCCG *subdiv_ccg)
   }
 }
 
-void BKE_pbvh_update_draw_buffers(PBVH *bvh, bool show_vcol)
+/**
+ * PBVH drawing, updating draw buffers as needed and culling any nodes outside
+ * the specified frustum.
+ */
+typedef struct PBVHDrawSearchData {
+  PBVHFrustumPlanes *frustum;
+} PBVHDrawSearchData;
+
+static bool pbvh_draw_search_cb(PBVHNode *node, void *data_v)
+{
+  PBVHDrawSearchData *data = data_v;
+  if (data->frustum && !BKE_pbvh_node_frustum_contain_AABB(node, data->frustum)) {
+    return false;
+  }
+
+  return true;
+}
+
+void BKE_pbvh_draw_cb(PBVH *bvh,
+                      bool show_vcol,
+                      PBVHFrustumPlanes *frustum,
+                      void (*draw_fn)(void *user_data, GPU_PBVH_Buffers *buffers),
+                      void *user_data)
 {
-  /* Update GPU buffers */
   PBVHNode **nodes;
   int totnode;
 
-  BKE_pbvh_search_gather(
-      bvh, update_search_cb, POINTER_FROM_INT(PBVH_UpdateDrawBuffers), &nodes, &totnode);
+  /* Update all draw buffers. */
+  const int update_flag = PBVH_RebuildDrawBuffers | PBVH_UpdateDrawBuffers;
+  BKE_pbvh_search_gather(bvh, update_search_cb, POINTER_FROM_INT(update_flag), &nodes, &totnode);
 
-  pbvh_update_draw_buffers(bvh, nodes, totnode, show_vcol);
+  if (totnode) {
+    pbvh_update_draw_buffers(bvh, nodes, totnode, show_vcol, update_flag);
+  }
 
   if (nodes) {
     MEM_freeN(nodes);
   }
-}
 
-/**
- * Version of #BKE_pbvh_draw that runs a callback.
- */
-void BKE_pbvh_draw_cb(PBVH *bvh,
-                      PBVHFrustumPlanes *frustum,
-                      void (*draw_fn)(void *user_data, GPU_PBVH_Buffers *buffers),
-                      void *user_data)
-{
-  PBVHNodeDrawCallbackData draw_data = {
-      .draw_fn = draw_fn,
-      .user_data = user_data,
-  };
+  /* Gather visible nodes. */
+  PBVHDrawSearchData data = {.frustum = frustum};
+  BKE_pbvh_search_gather(bvh, pbvh_draw_search_cb, &data, &nodes, &totnode);
+
+  /* Draw. */
+  for (int a = 0; a < totnode; a++) {
+    PBVHNode *node = nodes[a];
 
-  if (frustum) {
-    BKE_pbvh_search_callback(
-        bvh, BKE_pbvh_node_frustum_contain_AABB, frustum, pbvh_node_draw_cb, &draw_data);
+    if (node->flag & PBVH_UpdateDrawBuffers) {
+      /* Flush buffers uses OpenGL, so not in parallel. */
+      GPU_pbvh_buffers_update_flush(node->draw_buffers);
+    }
+
+    node->flag &= ~(PBVH_RebuildDrawBuffers | PBVH_UpdateDrawBuffers);
+
+    if (!(node->flag & PBVH_FullyHidden)) {
+      draw_fn(user_data, node->draw_buffers);
+    }
   }
-  else {
-    BKE_pbvh_search_callback(bvh, NULL, NULL, pbvh_node_draw_cb, &draw_data);
+
+  if (nodes) {
+    MEM_freeN(nodes);
   }
 }