Sculpt dyntopo

* Added a limited "fast draw" mode to pbvh drawing
  that tries to limit data sent to the GPU.
 - Facesets and mask data are disabled.
 - Indexed mode is forced.
 - Does not work (at all) for multires.

* Finally fixed some outstanding bmesh sculpt undo bugs:
 - Forgot to mark pbvh nodes to update their bounds, which
   produced a bug I had thought was caused by something else.
 - Hopefully fixed a long-standing occasional memory corruption
   bug.

13 files changed, 346 insertions, 174 deletions

diff --git a/source/blender/blenkernel/BKE_paint.h b/source/blender/blenkernel/BKE_paint.h
index 7314a2dc8ff..db091a8b878 100644
--- a/source/blender/blenkernel/BKE_paint.h
+++ b/source/blender/blenkernel/BKE_paint.h
@@ -655,6 +655,8 @@ typedef struct SculptSession {
   // id of current stroke, used to detect
   // if vertex original data needs to be updated
   int stroke_id;
+
+  bool fast_draw;  // hides facesets/masks and forces smooth to save GPU bandwidth
 } SculptSession;
 
 void BKE_sculptsession_free(struct Object *ob);
diff --git a/source/blender/blenkernel/BKE_pbvh.h b/source/blender/blenkernel/BKE_pbvh.h
index 97d3c77fb50..2f87895b262 100644
--- a/source/blender/blenkernel/BKE_pbvh.h
+++ b/source/blender/blenkernel/BKE_pbvh.h
@@ -240,21 +240,24 @@ void BKE_pbvh_build_mesh(PBVH *pbvh,
                          struct CustomData *ldata,
                          struct CustomData *pdata,
                          const struct MLoopTri *looptri,
-                         int looptri_num);
+                         int looptri_num,
+                         bool fast_draw);
 void BKE_pbvh_build_grids(PBVH *pbvh,
                           struct CCGElem **grids,
                           int totgrid,
                           struct CCGKey *key,
                           void **gridfaces,
                           struct DMFlagMat *flagmats,
-                          unsigned int **grid_hidden);
+                          unsigned int **grid_hidden,
+                          bool fast_draw);
 void BKE_pbvh_build_bmesh(PBVH *pbvh,
                           struct BMesh *bm,
                           bool smooth_shading,
                           struct BMLog *log,
                           const int cd_vert_node_offset,
                           const int cd_face_node_offset,
-                          const int cd_dyn_vert);
+                          const int cd_dyn_vert,
+                          bool fast_draw);
 void BKE_pbvh_update_offsets(PBVH *pbvh,
                              const int cd_vert_node_offset,
                              const int cd_face_node_offset,
@@ -682,10 +685,10 @@ bool BKE_pbvh_node_vert_update_check_any(PBVH *pbvh, PBVHNode *node);
 // void BKE_pbvh_node_BB_reset(PBVHNode *node);
 // void BKE_pbvh_node_BB_expand(PBVHNode *node, float co[3]);
 
-bool pbvh_has_mask(const PBVH *pbvh);
+bool BKE_pbvh_draw_mask(const PBVH *pbvh);
 void pbvh_show_mask_set(PBVH *pbvh, bool show_mask);
 
-bool pbvh_has_face_sets(PBVH *pbvh);
+bool BKE_pbvh_draw_face_sets(PBVH *pbvh);
 void pbvh_show_face_sets_set(PBVH *pbvh, bool show_face_sets);
 
 /* Parallelization */
@@ -740,6 +743,8 @@ void BKE_pbvh_update_vert_boundary(int cd_dyn_vert, int cd_faceset_offset, struc
 
 #define DYNTOPO_DYNAMIC_TESS
 
+PBVHNode *BKE_pbvh_get_node_leaf_safe(PBVH *pbvh, int i);
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/source/blender/blenkernel/intern/paint.c b/source/blender/blenkernel/intern/paint.c
index 6463fcf8780..2cc5561000b 100644
--- a/source/blender/blenkernel/intern/paint.c
+++ b/source/blender/blenkernel/intern/paint.c
@@ -1701,6 +1701,7 @@ static void sculpt_update_object(Depsgraph *depsgraph,
   }
 
   ss->subdiv_ccg = me_eval->runtime.subdiv_ccg;
+  ss->fast_draw = (scene->toolsettings->sculpt->flags & SCULPT_FAST_DRAW) != 0;
 
   PBVH *pbvh = BKE_sculpt_object_pbvh_ensure(depsgraph, ob);
   BLI_assert(pbvh == ss->pbvh);
@@ -2137,7 +2138,8 @@ static PBVH *build_pbvh_for_dynamic_topology(Object *ob)
                        ob->sculpt->bm_log,
                        ob->sculpt->cd_vert_node_offset,
                        ob->sculpt->cd_face_node_offset,
-                       ob->sculpt->cd_dyn_vert);
+                       ob->sculpt->cd_dyn_vert,
+                       ob->sculpt->fast_draw);
   pbvh_show_mask_set(pbvh, ob->sculpt->show_mask);
   pbvh_show_face_sets_set(pbvh, false);
 
@@ -2167,7 +2169,8 @@ static PBVH *build_pbvh_from_regular_mesh(Object *ob, Mesh *me_eval_deform, bool
                       &me->ldata,
                       &me->pdata,
                       looptri,
-                      looptris_num);
+                      looptris_num,
+                      ob->sculpt->fast_draw);
 
   pbvh_show_mask_set(pbvh, ob->sculpt->show_mask);
   pbvh_show_face_sets_set(pbvh, ob->sculpt->show_face_sets);
@@ -2199,7 +2202,8 @@ static PBVH *build_pbvh_from_ccg(Object *ob, SubdivCCG *subdiv_ccg, bool respect
                        &key,
                        (void **)subdiv_ccg->grid_faces,
                        subdiv_ccg->grid_flag_mats,
-                       subdiv_ccg->grid_hidden);
+                       subdiv_ccg->grid_hidden,
+                       ob->sculpt->fast_draw);
   pbvh_show_mask_set(pbvh, ob->sculpt->show_mask);
   pbvh_show_face_sets_set(pbvh, ob->sculpt->show_face_sets);
   return pbvh;
diff --git a/source/blender/blenkernel/intern/pbvh.c b/source/blender/blenkernel/intern/pbvh.c
index abadab127a8..52058661da6 100644
--- a/source/blender/blenkernel/intern/pbvh.c
+++ b/source/blender/blenkernel/intern/pbvh.c
@@ -591,7 +591,8 @@ void BKE_pbvh_build_mesh(PBVH *pbvh,
                          struct CustomData *ldata,
                          struct CustomData *pdata,
                          const MLoopTri *looptri,
-                         int looptri_num)
+                         int looptri_num,
+                         bool fast_draw)
 {
   BBC *prim_bbc = NULL;
   BB cb;
@@ -639,6 +640,10 @@ void BKE_pbvh_build_mesh(PBVH *pbvh,
     pbvh_build(pbvh, &cb, prim_bbc, looptri_num);
   }
 
+  if (fast_draw) {
+    pbvh->flags |= PBVH_FAST_DRAW;
+  }
+
   MEM_freeN(prim_bbc);
   MEM_freeN(pbvh->vert_bitmap);
 }
@@ -650,7 +655,8 @@ void BKE_pbvh_build_grids(PBVH *pbvh,
                           CCGKey *key,
                           void **gridfaces,
                           DMFlagMat *flagmats,
-                          BLI_bitmap **grid_hidden)
+                          BLI_bitmap **grid_hidden,
+                          bool fast_draw)
 {
   const int gridsize = key->grid_size;
 
@@ -689,6 +695,10 @@ void BKE_pbvh_build_grids(PBVH *pbvh,
     pbvh_build(pbvh, &cb, prim_bbc, totgrid);
   }
 
+  if (fast_draw) {
+    pbvh->flags |= PBVH_FAST_DRAW;
+  }
+
   MEM_freeN(prim_bbc);
 }
 
@@ -1454,20 +1464,11 @@ void pbvh_update_free_all_draw_buffers(PBVH *pbvh, PBVHNode *node)
   }
 }
 
-static void pbvh_update_draw_buffers(PBVH *pbvh, PBVHNode **nodes, int totnode, int update_flag)
+ATTR_NO_OPT static void pbvh_update_draw_buffers(PBVH *pbvh,
+                                                 PBVHNode **nodes,
+                                                 int totnode,
+                                                 int update_flag)
 {
-  if ((update_flag & PBVH_RebuildDrawBuffers) || ELEM(pbvh->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) {
-        pbvh_free_all_draw_buffers(node);
-      }
-      else if ((node->flag & PBVH_UpdateDrawBuffers)) {
-        pbvh_update_free_all_draw_buffers(pbvh, node);
-      }
-    }
-  }
 
   CustomData *vdata;
   CustomData *ldata;
@@ -1486,7 +1487,21 @@ static void pbvh_update_draw_buffers(PBVH *pbvh, PBVHNode **nodes, int totnode,
     ldata = pbvh->ldata;
   }
 
-  GPU_pbvh_update_attribute_names(vdata, ldata, GPU_pbvh_need_full_render_get());
+  GPU_pbvh_update_attribute_names(
+      vdata, ldata, GPU_pbvh_need_full_render_get(), pbvh->flags & PBVH_FAST_DRAW);
+
+  if ((update_flag & PBVH_RebuildDrawBuffers) || ELEM(pbvh->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) {
+        pbvh_free_all_draw_buffers(node);
+      }
+      else if ((node->flag & PBVH_UpdateDrawBuffers)) {
+        pbvh_update_free_all_draw_buffers(pbvh, node);
+      }
+    }
+  }
 
   /* Parallel creation and update of draw buffers. */
   PBVHUpdateData data = {
@@ -3252,8 +3267,12 @@ void pbvh_vertex_iter_init(PBVH *pbvh, PBVHNode *node, PBVHVertexIter *vi, int m
   }
 }
 
-bool pbvh_has_mask(const PBVH *pbvh)
+bool BKE_pbvh_draw_mask(const PBVH *pbvh)
 {
+  if (pbvh->flags & PBVH_FAST_DRAW) {
+    return false;
+  }
+
   switch (pbvh->type) {
     case PBVH_GRIDS:
       return (pbvh->gridkey.has_mask != 0);
@@ -3286,8 +3305,12 @@ SculptFaceRef BKE_pbvh_table_index_to_face(PBVH *pbvh, int idx)
   return BKE_pbvh_make_fref(idx);
 }
 
-bool pbvh_has_face_sets(PBVH *pbvh)
+bool BKE_pbvh_draw_face_sets(PBVH *pbvh)
 {
+  if (pbvh->flags & PBVH_FAST_DRAW) {
+    return false;
+  }
+
   switch (pbvh->type) {
     case PBVH_GRIDS:
       return (pbvh->pdata && CustomData_get_layer(pbvh->pdata, CD_SCULPT_FACE_SETS));
diff --git a/source/blender/blenkernel/intern/pbvh_bmesh.c b/source/blender/blenkernel/intern/pbvh_bmesh.c
index e2f926fa2f8..f8f3beaccaa 100644
--- a/source/blender/blenkernel/intern/pbvh_bmesh.c
+++ b/source/blender/blenkernel/intern/pbvh_bmesh.c
@@ -545,11 +545,13 @@ void bke_pbvh_insert_face(PBVH *pbvh, struct BMFace *f)
 
 static void pbvh_bmesh_regen_node_verts(PBVH *pbvh, PBVHNode *node)
 {
-  int usize = BLI_table_gset_len(node->bm_other_verts);
+  node->flag &= ~PBVH_RebuildNodeVerts;
+
+  int usize = BLI_table_gset_len(node->bm_unique_verts);
   int osize = BLI_table_gset_len(node->bm_other_verts);
 
-  BLI_table_gset_free(node->bm_other_verts, NULL);
   BLI_table_gset_free(node->bm_unique_verts, NULL);
+  BLI_table_gset_free(node->bm_other_verts, NULL);
 
   node->bm_unique_verts = BLI_table_gset_new("bm_unique_verts");
   node->bm_other_verts = BLI_table_gset_new("bm_other_verts");
@@ -557,11 +559,20 @@ static void pbvh_bmesh_regen_node_verts(PBVH *pbvh, PBVHNode *node)
   const int cd_vert_node = pbvh->cd_vert_node_offset;
   const int ni = (int)(node - pbvh->nodes);
 
+  bool update = false;
+
   BMFace *f;
   TGSET_ITER (f, node->bm_faces) {
     BMLoop *l = f->l_first;
     do {
       int ni2 = BM_ELEM_CD_GET_INT(l->v, cd_vert_node);
+
+      if (ni2 == DYNTOPO_NODE_NONE) {
+        BM_ELEM_CD_SET_INT(l->v, cd_vert_node, ni);
+        ni2 = ni;
+        update = true;
+      }
+
       if (ni2 == ni) {
         BLI_table_gset_add(node->bm_unique_verts, l->v);
       }
@@ -573,16 +584,25 @@ static void pbvh_bmesh_regen_node_verts(PBVH *pbvh, PBVHNode *node)
   TGSET_ITER_END;
 
   if (usize != BLI_table_gset_len(node->bm_unique_verts)) {
+    update = true;
     printf("possible pbvh error: bm_unique_verts might have had bad data. old: %d, new: %d\n",
            usize,
            BLI_table_gset_len(node->bm_unique_verts));
   }
 
   if (osize != BLI_table_gset_len(node->bm_other_verts)) {
+    update = true;
     printf("possible pbvh error: bm_other_verts might have had bad data. old: %d, new: %d\n",
            osize,
            BLI_table_gset_len(node->bm_other_verts));
   }
+
+  if (update) {
+    node->flag |= PBVH_UpdateNormals | PBVH_UpdateDrawBuffers | PBVH_RebuildDrawBuffers |
+                  PBVH_UpdateBB;
+    node->flag |= PBVH_UpdateOriginalBB | PBVH_UpdateRedraw | PBVH_UpdateColor | PBVH_UpdateTris |
+                  PBVH_UpdateVisibility;
+  }
 }
 
 void BKE_pbvh_bmesh_mark_node_regen(PBVH *pbvh, PBVHNode *node)
@@ -590,6 +610,18 @@ void BKE_pbvh_bmesh_mark_node_regen(PBVH *pbvh, PBVHNode *node)
   node->flag |= PBVH_RebuildNodeVerts;
 }
 
+PBVHNode *BKE_pbvh_get_node_leaf_safe(PBVH *pbvh, int i)
+{
+  if (i >= 0 && i < pbvh->totnode) {
+    PBVHNode *node = pbvh->nodes + i;
+    if ((node->flag & PBVH_Leaf) && !(node->flag & PBVH_Delete)) {
+      return node;
+    }
+  }
+
+  return NULL;
+}
+
 void BKE_pbvh_bmesh_regen_node_verts(PBVH *pbvh)
 {
   for (int i = 0; i < pbvh->totnode; i++) {
@@ -1226,13 +1258,16 @@ void BKE_pbvh_build_bmesh(PBVH *pbvh,
                           BMLog *log,
                           const int cd_vert_node_offset,
                           const int cd_face_node_offset,
-                          const int cd_dyn_vert)
+                          const int cd_dyn_vert,
+                          bool fast_draw)
 {
   pbvh->cd_vert_node_offset = cd_vert_node_offset;
   pbvh->cd_face_node_offset = cd_face_node_offset;
   pbvh->cd_vert_mask_offset = CustomData_get_offset(&bm->vdata, CD_PAINT_MASK);
   pbvh->cd_dyn_vert = cd_dyn_vert;
 
+  smooth_shading |= fast_draw;
+
   pbvh->bm = bm;
 
   BKE_pbvh_bmesh_detail_size_set(pbvh, 0.75f, 0.4f);
@@ -1270,10 +1305,15 @@ void BKE_pbvh_build_bmesh(PBVH *pbvh,
       zero_v4(mv->origcolor);
     }
   }
+
   if (smooth_shading) {
     pbvh->flags |= PBVH_DYNTOPO_SMOOTH_SHADING;
   }
 
+  if (fast_draw) {
+    pbvh->flags |= PBVH_FAST_DRAW;
+  }
+
   /* bounding box array of all faces, no need to recalculate every time */
   BBC *bbc_array = MEM_mallocN(sizeof(BBC) * bm->totface, "BBC");
   BMFace **nodeinfo = MEM_mallocN(sizeof(*nodeinfo) * bm->totface, "nodeinfo");
@@ -1384,13 +1424,16 @@ static void pbvh_free_tribuf(PBVHTriBuf *tribuf)
   MEM_SAFE_FREE(tribuf->verts);
   MEM_SAFE_FREE(tribuf->tris);
   MEM_SAFE_FREE(tribuf->loops);
+  MEM_SAFE_FREE(tribuf->edges);
 
   tribuf->verts = NULL;
   tribuf->tris = NULL;
   tribuf->loops = NULL;
+  tribuf->edges = NULL;
 
   tribuf->verts_size = 0;
   tribuf->tris_size = 0;
+  tribuf->edges_size = 0;
 }
 
 PBVHTriBuf *BKE_pbvh_bmesh_get_tris(PBVH *pbvh, PBVHNode *node)
diff --git a/source/blender/blenkernel/intern/pbvh_intern.h b/source/blender/blenkernel/intern/pbvh_intern.h
index f06c95b1755..efe5c426e91 100644
--- a/source/blender/blenkernel/intern/pbvh_intern.h
+++ b/source/blender/blenkernel/intern/pbvh_intern.h
@@ -129,6 +129,7 @@ struct PBVHNode {
 
 typedef enum {
   PBVH_DYNTOPO_SMOOTH_SHADING = 1,
+  PBVH_FAST_DRAW = 2  // hides facesets/masks and forces smooth to save GPU bandwidth
 } PBVHFlags;
 
 typedef struct PBVHBMeshLog PBVHBMeshLog;
diff --git a/source/blender/draw/engines/overlay/overlay_sculpt.c b/source/blender/draw/engines/overlay/overlay_sculpt.c
index 24c52ec427d..24ff807bba8 100644
--- a/source/blender/draw/engines/overlay/overlay_sculpt.c
+++ b/source/blender/draw/engines/overlay/overlay_sculpt.c
@@ -60,7 +60,7 @@ void OVERLAY_sculpt_cache_populate(OVERLAY_Data *vedata, Object *ob)
     return;
   }
 
-  if (!pbvh_has_mask(pbvh) && !pbvh_has_face_sets(pbvh)) {
+  if (!BKE_pbvh_draw_mask(pbvh) && !BKE_pbvh_draw_face_sets(pbvh)) {
     /* The SculptSession and the PBVH can be created without a Mask data-layer or Face Set
      * data-layer. (masks data-layers are created after using a mask tool), so in these cases there
      * is nothing to draw. */
diff --git a/source/blender/editors/sculpt_paint/paint_mask.c b/source/blender/editors/sculpt_paint/paint_mask.c
index b3fcb9155f3..04db16d4516 100644
--- a/source/blender/editors/sculpt_paint/paint_mask.c
+++ b/source/blender/editors/sculpt_paint/paint_mask.c
@@ -1305,19 +1305,20 @@ static void sculpt_gesture_apply_trim(SculptGestureContext *sgcontext)
 
   MEM_freeN(looptris);
 
-  if (sgcontext->ss && sgcontext->ss->bm) { //rebuild pbvh
+  if (sgcontext->ss && sgcontext->ss->bm) {  // rebuild pbvh
     BKE_pbvh_free(sgcontext->ss->pbvh);
     sgcontext->ss->pbvh = BKE_pbvh_new();
 
     BKE_pbvh_build_bmesh(sgcontext->ss->pbvh,
                          sgcontext->ss->bm,
-                         false,
+                         sgcontext->ss->bm_smooth_shading,
                          sgcontext->ss->bm_log,
                          sgcontext->ss->cd_vert_node_offset,
                          sgcontext->ss->cd_face_node_offset,
-                         sgcontext->ss->cd_dyn_vert);
+                         sgcontext->ss->cd_dyn_vert,
+                         sgcontext->ss->fast_draw);
   }
-  else { //save result to mesh
+  else {  // save result to mesh
     Mesh *result = BKE_mesh_from_bmesh_nomain(bm,
                                               (&(struct BMeshToMeshParams){
                                                   .calc_object_remap = false,
diff --git a/source/blender/editors/sculpt_paint/sculpt_undo.c b/source/blender/editors/sculpt_paint/sculpt_undo.c
index 3e5aa90a6ba..1cc928525ec 100644
--- a/source/blender/editors/sculpt_paint/sculpt_undo.c
+++ b/source/blender/editors/sculpt_paint/sculpt_undo.c
@@ -435,6 +435,8 @@ typedef struct BmeshUndoData {
   bool balance_pbvh;
   int cd_face_node_offset, cd_vert_node_offset;
   int cd_dyn_vert;
+  bool regen_all_unique_verts;
+  bool is_redo;
 } BmeshUndoData;
 
 static void bmesh_undo_on_vert_kill(BMVert *v, void *userdata)
@@ -448,29 +450,17 @@ static void bmesh_undo_on_vert_kill(BMVert *v, void *userdata)
     data->do_full_recalc = true;
     return;
   }
-  else {
-    int ni = BM_ELEM_CD_GET_INT(v, data->cd_vert_node_offset);
-
-    /*
-    regenerate bm_unique_verts, which can end up with
-    freed verts for some reason.  I've run this through
-    ASAN and fixed one likely cause, but it still happens.
-    - joeedh
-    */
-    if (ni >= 0) {
-      PBVHNode *node = BKE_pbvh_get_node(data->pbvh, ni);
-      BKE_pbvh_bmesh_mark_node_regen(data->pbvh, node);
-    }
-  }
 
   BKE_pbvh_bmesh_remove_vertex(data->pbvh, v, false);
   data->balance_pbvh = true;
 }
+
 static void bmesh_undo_on_vert_add(BMVert *v, void *userdata)
 {
   BmeshUndoData *data = (BmeshUndoData *)userdata;
   BM_ELEM_CD_SET_INT(v, data->cd_vert_node_offset, -1);
   // data->do_full_recalc = true;
+
   data->balance_pbvh = true;
 
   MDynTopoVert *mv = BKE_PBVH_DYNVERT(data->cd_dyn_vert, v);
@@ -482,19 +472,13 @@ static void bmesh_undo_on_face_kill(BMFace *f, void *userdata)
   BmeshUndoData *data = (BmeshUndoData *)userdata;
   int ni = BM_ELEM_CD_GET_INT(f, data->cd_face_node_offset);
 
-  /*
-  regenerate bm_unique_verts, which can end up with
-  freed verts for some reason.  I've run this through
-  ASAN and fixed one likely cause, but it still happens.
-  - joeedh
-  */
+  BKE_pbvh_bmesh_remove_face(data->pbvh, f, false);
+
   if (ni >= 0) {
     PBVHNode *node = BKE_pbvh_get_node(data->pbvh, ni);
     BKE_pbvh_bmesh_mark_node_regen(data->pbvh, node);
   }
 
-  BKE_pbvh_bmesh_remove_face(data->pbvh, f, false);
-
   // data->do_full_recalc = true;
   data->balance_pbvh = true;
 }
@@ -521,22 +505,36 @@ static void bmesh_undo_full_mesh(void *userdata)
   data->do_full_recalc = true;
 }
 
-static void bmesh_undo_on_vert_change(BMVert *v, void *userdata, void *old_customdata)
+ATTR_NO_OPT static void bmesh_undo_on_vert_change(BMVert *v, void *userdata, void *old_customdata)
 {
   BmeshUndoData *data = (BmeshUndoData *)userdata;
 
   if (!old_customdata) {
     BM_ELEM_CD_SET_INT(v, data->cd_vert_node_offset, -1);
+    data->regen_all_unique_verts = true;
     return;
   }
 
-  BM_ELEM_CD_SET_INT(v, data->cd_vert_node_offset, -1);
+  BMElem h;
+  h.head.data = old_customdata;
+
+  int ni = BM_ELEM_CD_GET_INT(&h, data->cd_vert_node_offset);
+
+  // attempt to find old node
+  PBVHNode *node = BKE_pbvh_get_node_leaf_safe(data->pbvh, ni);
+  if (node) {
+    // BKE_pbvh_bmesh_mark_node_regen(data->pbvh, node);
+    BKE_pbvh_node_mark_update(node);
+    // BM_ELEM_CD_SET_INT(v, data->cd_vert_node_offset, -1);
+  }
+  else {
+    BM_ELEM_CD_SET_INT(v, data->cd_vert_node_offset, -1);
+    data->regen_all_unique_verts = true;
+  }
+
   return;
   // preserve pbvh node references
 
-  BMVert h;
-  h.head.data = old_customdata;
-
   int oldnode_i = BM_ELEM_CD_GET_INT(&h, data->cd_vert_node_offset);
 
   BM_ELEM_CD_SET_INT(v, data->cd_vert_node_offset, oldnode_i);
@@ -563,7 +561,9 @@ static void sculpt_undo_bmesh_restore_generic(SculptUndoNode *unode, Object *ob,
                         false,
                         ss->cd_face_node_offset,
                         ss->cd_vert_node_offset,
-                        ss->cd_dyn_vert};
+                        ss->cd_dyn_vert,
+                        false,
+                        !unode->applied};
 
   BMLogCallbacks callbacks = {bmesh_undo_on_vert_add,
                               bmesh_undo_on_vert_kill,
@@ -589,11 +589,17 @@ static void sculpt_undo_bmesh_restore_generic(SculptUndoNode *unode, Object *ob,
     int totnode;
     PBVHNode **nodes;
 
+    BKE_pbvh_search_gather(ss->pbvh, NULL, NULL, &nodes, &totnode);
+
+    if (data.regen_all_unique_verts) {
+      for (int i = 0; i < totnode; i++) {
+        BKE_pbvh_bmesh_mark_node_regen(ss->pbvh, nodes[i]);
+      }
+    }
+
     BKE_pbvh_bmesh_regen_node_verts(ss->pbvh);
     pbvh_bmesh_check_nodes(ss->pbvh);
 
-    BKE_pbvh_search_gather(ss->pbvh, NULL, NULL, &nodes, &totnode);
-
     TaskParallelSettings settings;
     BKE_pbvh_parallel_range_settings(&settings, true, totnode);
     BLI_task_parallel_range(
@@ -608,6 +614,8 @@ static void sculpt_undo_bmesh_restore_generic(SculptUndoNode *unode, Object *ob,
     if (data.balance_pbvh) {
       BKE_pbvh_bmesh_after_stroke(ss->pbvh);
     }
+
+    BKE_pbvh_update_bounds(ss->pbvh, PBVH_UpdateBB | PBVH_UpdateOriginalBB | PBVH_UpdateRedraw);
   }
   else {
     SCULPT_pbvh_clear(ob);
diff --git a/source/blender/gpu/GPU_buffers.h b/source/blender/gpu/GPU_buffers.h
index cf0ba4b16bf..e2ccf8fe941 100644
--- a/source/blender/gpu/GPU_buffers.h
+++ b/source/blender/gpu/GPU_buffers.h
@@ -87,9 +87,11 @@ void GPU_pbvh_mesh_buffers_update(GPU_PBVH_Buffers *buffers,
                                   const struct MPropCol *vtcol,
                                   const int update_flags);
 
-void GPU_pbvh_update_attribute_names(struct CustomData *vdata,
-                                     struct CustomData *ldata,
-                                     bool need_full_render);
+void GPU_pbvh_update_attribute_names(
+    struct CustomData *vdata,
+    struct CustomData *ldata,
+    bool need_full_render,
+    bool fast_mode);  // fast mode renders without vcol, uv, facesets, even mask, etc
 
 void GPU_pbvh_bmesh_buffers_update(GPU_PBVH_Buffers *buffers,
                                    struct BMesh *bm,
diff --git a/source/blender/gpu/intern/gpu_buffers.c b/source/blender/gpu/intern/gpu_buffers.c
index d6968dbc158..67e52b6c964 100644
--- a/source/blender/gpu/intern/gpu_buffers.c
+++ b/source/blender/gpu/intern/gpu_buffers.c
@@ -61,6 +61,8 @@
 
 #include "bmesh.h"
 
+//#define GPU_PERF_TEST  // will disable vcol, uvs, mask, fset colors, etc
+
 /* XXX: the rest of the code in this file is used for optimized PBVH
  * drawing and doesn't interact at all with the buffer code above */
 
@@ -116,6 +118,8 @@ struct GPU_PBVH_Buffers {
    * smooth-shaded or all faces are flat-shaded */
   bool smooth;
 
+  void *last_tribuf_tris;  // used to detect if we can reuse index buffers
+
   bool show_overlay;
 #ifdef QUANTIZED_PERF_TEST
   float matrix[4][4];
@@ -157,6 +161,7 @@ static struct {
 
   bool active_vcol_only;
   bool need_full_render;
+  bool fast_mode;
 } g_vbo_id = {{0}};
 
 #ifdef NEW_ATTR_SYSTEM
@@ -355,7 +360,7 @@ static void free_cd_layers(CDAttrLayers *cdattr)
 
 void gpu_pbvh_init()
 {
-  GPU_pbvh_update_attribute_names(NULL, NULL, false);
+  GPU_pbvh_update_attribute_names(NULL, NULL, false, false);
 }
 
 void gpu_pbvh_exit()
@@ -462,9 +467,11 @@ void GPU_pbvh_mesh_buffers_update(GPU_PBVH_Buffers *buffers,
                                   const MPropCol *vtcol,
                                   const int update_flags)
 {
-  const bool show_mask = vmask && (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0;
+  const bool show_mask = vmask && (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0 &&
+                         !g_vbo_id.fast_mode;
   const bool show_face_sets = sculpt_face_sets &&
-                              (update_flags & GPU_PBVH_BUFFERS_SHOW_SCULPT_FACE_SETS) != 0;
+                              (update_flags & GPU_PBVH_BUFFERS_SHOW_SCULPT_FACE_SETS) != 0 &&
+                              !g_vbo_id.fast_mode;
   const bool show_vcol = (vcol || (vtcol && U.experimental.use_sculpt_vertex_colors)) &&
                          (update_flags & GPU_PBVH_BUFFERS_SHOW_VCOL) != 0;
   bool empty_mask = true;
@@ -566,8 +573,11 @@ void GPU_pbvh_mesh_buffers_update(GPU_PBVH_Buffers *buffers,
               memcpy(GPU_vertbuf_raw_step(&col_step), scol, sizeof(scol));
             }
           }
-          /* Face Sets. */
-          memcpy(GPU_vertbuf_raw_step(&fset_step), face_set_color, sizeof(uchar[3]));
+
+          if (!g_vbo_id.fast_mode) {
+            /* Face Sets. */
+            memcpy(GPU_vertbuf_raw_step(&fset_step), face_set_color, sizeof(uchar[3]));
+          }
         }
       }
     }
@@ -580,7 +590,7 @@ void GPU_pbvh_mesh_buffers_update(GPU_PBVH_Buffers *buffers,
   const MPoly *mp = &buffers->mpoly[lt->poly];
   buffers->material_index = mp->mat_nr;
 
-  buffers->show_overlay = !empty_mask || !default_face_set;
+  buffers->show_overlay = (!empty_mask || !default_face_set) && !g_vbo_id.fast_mode;
   buffers->mvert = mvert;
 }
 
@@ -601,7 +611,7 @@ GPU_PBVH_Buffers *GPU_pbvh_mesh_buffers_build(const MPoly *mpoly,
   buffers = MEM_callocN(sizeof(GPU_PBVH_Buffers), "GPU_Buffers");
 
   /* smooth or flat for all */
-  buffers->smooth = mpoly[looptri[face_indices[0]].poly].flag & ME_SMOOTH;
+  buffers->smooth = (mpoly[looptri[face_indices[0]].poly].flag & ME_SMOOTH) || g_vbo_id.fast_mode;
 
   buffers->show_overlay = false;
 
@@ -679,13 +689,13 @@ GPU_PBVH_Buffers *GPU_pbvh_mesh_buffers_build(const MPoly *mpoly,
 /** \name Grid PBVH
  * \{ */
 
-static void gpu_pbvh_grid_fill_index_buffers(GPU_PBVH_Buffers *buffers,
-                                             SubdivCCG *UNUSED(subdiv_ccg),
-                                             const int *UNUSED(face_sets),
-                                             const int *grid_indices,
-                                             uint visible_quad_len,
-                                             int totgrid,
-                                             int gridsize)
+ATTR_NO_OPT static void gpu_pbvh_grid_fill_index_buffers(GPU_PBVH_Buffers *buffers,
+                                                         SubdivCCG *UNUSED(subdiv_ccg),
+                                                         const int *UNUSED(face_sets),
+                                                         const int *grid_indices,
+                                                         uint visible_quad_len,
+                                                         int totgrid,
+                                                         int gridsize)
 {
   GPUIndexBufBuilder elb, elb_lines;
   GPUIndexBufBuilder elb_fast, elb_lines_fast;
@@ -695,7 +705,7 @@ static void gpu_pbvh_grid_fill_index_buffers(GPU_PBVH_Buffers *buffers,
   GPU_indexbuf_init(&elb_lines, GPU_PRIM_LINES, 2 * totgrid * gridsize * (gridsize - 1), INT_MAX);
   GPU_indexbuf_init(&elb_lines_fast, GPU_PRIM_LINES, 4 * totgrid, INT_MAX);
 
-  if (buffers->smooth) {
+  if (buffers->smooth || g_vbo_id.fast_mode) {
     uint offset = 0;
     const uint grid_vert_len = gridsize * gridsize;
     for (int i = 0; i < totgrid; i++, offset += grid_vert_len) {
@@ -815,7 +825,7 @@ void GPU_pbvh_grid_buffers_update_free(GPU_PBVH_Buffers *buffers,
                                        const struct DMFlagMat *grid_flag_mats,
                                        const int *grid_indices)
 {
-  const bool smooth = grid_flag_mats[grid_indices[0]].flag & ME_SMOOTH;
+  const bool smooth = (grid_flag_mats[grid_indices[0]].flag & ME_SMOOTH) || g_vbo_id.fast_mode;
 
   if (buffers->smooth != smooth) {
     buffers->smooth = smooth;
@@ -832,22 +842,23 @@ void GPU_pbvh_grid_buffers_update_free(GPU_PBVH_Buffers *buffers,
 }
 
 /* Threaded - do not call any functions that use OpenGL calls! */
-void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
-                                  SubdivCCG *subdiv_ccg,
-                                  CCGElem **grids,
-                                  const struct DMFlagMat *grid_flag_mats,
-                                  int *grid_indices,
-                                  int totgrid,
-                                  const int *sculpt_face_sets,
-                                  const int face_sets_color_seed,
-                                  const int face_sets_color_default,
-                                  const struct CCGKey *key,
-                                  const int update_flags)
+ATTR_NO_OPT void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
+                                              SubdivCCG *subdiv_ccg,
+                                              CCGElem **grids,
+                                              const struct DMFlagMat *grid_flag_mats,
+                                              int *grid_indices,
+                                              int totgrid,
+                                              const int *sculpt_face_sets,
+                                              const int face_sets_color_seed,
+                                              const int face_sets_color_default,
+                                              const struct CCGKey *key,
+                                              const int update_flags)
 {
-  const bool show_mask = (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0;
+  const bool show_mask = (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0 && !g_vbo_id.fast_mode;
   const bool show_vcol = (update_flags & GPU_PBVH_BUFFERS_SHOW_VCOL) != 0;
   const bool show_face_sets = sculpt_face_sets &&
-                              (update_flags & GPU_PBVH_BUFFERS_SHOW_SCULPT_FACE_SETS) != 0;
+                              (update_flags & GPU_PBVH_BUFFERS_SHOW_SCULPT_FACE_SETS) != 0 &&
+                              !g_vbo_id.fast_mode;
   bool empty_mask = true;
   bool default_face_set = true;
 
@@ -856,7 +867,13 @@ void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
   /* Build VBO */
   const int has_mask = key->has_mask;
 
-  buffers->smooth = grid_flag_mats[grid_indices[0]].flag & ME_SMOOTH;
+  bool smooth = (grid_flag_mats[grid_indices[0]].flag & ME_SMOOTH) || g_vbo_id.fast_mode;
+  ;
+  if (smooth != buffers->smooth) {
+    GPU_pbvh_grid_buffers_update_free(buffers, grid_flag_mats, grid_indices);
+  }
+
+  buffers->smooth = smooth;
 
   uint vert_per_grid = (buffers->smooth) ? key->grid_area : (square_i(key->grid_size - 1) * 4);
   uint vert_count = totgrid * vert_per_grid;
@@ -906,7 +923,7 @@ void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
         }
       }
 
-      if (buffers->smooth) {
+      if (buffers->smooth || g_vbo_id.fast_mode) {
         for (y = 0; y < key->grid_size; y++) {
           for (x = 0; x < key->grid_size; x++) {
             CCGElem *elem = CCG_grid_elem(key, grid, x, y);
@@ -929,7 +946,9 @@ void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
               GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index, &vcol);
             }
 
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, vbo_index, &face_set_color);
+            if (!g_vbo_id.fast_mode) {
+              GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, vbo_index, &face_set_color);
+            }
 
             vbo_index += 1;
           }
@@ -980,15 +999,23 @@ void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
             }
 
             const ushort vcol[4] = {USHRT_MAX, USHRT_MAX, USHRT_MAX, USHRT_MAX};
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 0, &vcol);
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 1, &vcol);
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 2, &vcol);
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 3, &vcol);
+            if (1) {  // g_vbo_id.totcol > 0 || !g_vbo_id.fast_mode) {
+              GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 0, &vcol);
+              GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 1, &vcol);
+              GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 2, &vcol);
+              GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.col[0], vbo_index + 3, &vcol);
+            }
 
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, vbo_index + 0, &face_set_color);
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, vbo_index + 1, &face_set_color);
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, vbo_index + 2, &face_set_color);
-            GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, vbo_index + 3, &face_set_color);
+            if (!g_vbo_id.fast_mode) {
+              GPU_vertbuf_attr_set(
+                  buffers->vert_buf, g_vbo_id.fset, vbo_index + 0, &face_set_color);
+              GPU_vertbuf_attr_set(
+                  buffers->vert_buf, g_vbo_id.fset, vbo_index + 1, &face_set_color);
+              GPU_vertbuf_attr_set(
+                  buffers->vert_buf, g_vbo_id.fset, vbo_index + 2, &face_set_color);
+              GPU_vertbuf_attr_set(
+                  buffers->vert_buf, g_vbo_id.fset, vbo_index + 3, &face_set_color);
+            }
 
             vbo_index += 4;
           }
@@ -1008,7 +1035,7 @@ void GPU_pbvh_grid_buffers_update(GPU_PBVH_Buffers *buffers,
   buffers->totgrid = totgrid;
   buffers->grid_flag_mats = grid_flag_mats;
   buffers->gridkey = *key;
-  buffers->show_overlay = !empty_mask || !default_face_set;
+  buffers->show_overlay = (!empty_mask || !default_face_set) && !g_vbo_id.fast_mode;
 }
 
 /* Threaded - do not call any functions that use OpenGL calls! */
@@ -1061,9 +1088,11 @@ static void gpu_bmesh_vert_to_buffer_copy(BMVert *v,
   GPU_vertbuf_attr_set(vert_buf, g_vbo_id.pos, v_index, v->co);
 
   short no_short[3];
+
   normal_float_to_short_v3(no_short, fno ? fno : v->no);
   GPU_vertbuf_attr_set(vert_buf, g_vbo_id.nor, v_index, no_short);
 
+#ifndef GPU_PERF_TEST
   if (show_mask) {
     float effective_mask = fmask ? *fmask : BM_ELEM_CD_GET_FLOAT(v, cd_vert_mask_offset);
 
@@ -1078,7 +1107,7 @@ static void gpu_bmesh_vert_to_buffer_copy(BMVert *v,
     *empty_mask = *empty_mask && (cmask == 0);
   }
 
-#ifndef NEW_ATTR_SYSTEM
+#  ifndef NEW_ATTR_SYSTEM
   if (show_vcol && totvcol > 0) {
     for (int i = 0; i < totvcol; i++) {
       ushort vcol[4] = {USHRT_MAX, USHRT_MAX, USHRT_MAX, USHRT_MAX};
@@ -1094,20 +1123,23 @@ static void gpu_bmesh_vert_to_buffer_copy(BMVert *v,
       GPU_vertbuf_attr_set(vert_buf, g_vbo_id.col[i], v_index, vcol);
     }
   }
-  else if (show_vcol) {  // ensure first vcol attribute is not zero
+  else if (show_vcol && !g_vbo_id.fast_mode) {  // ensure first vcol attribute is not zero
     const ushort vcol[4] = {USHRT_MAX, USHRT_MAX, USHRT_MAX, USHRT_MAX};
     GPU_vertbuf_attr_set(vert_buf, g_vbo_id.col[0], v_index, vcol);
   }
-#else
+#  else
   if (show_vcol && totvcol == 0) {  // ensure first vcol attribute is not zero
     const ushort vcol[4] = {USHRT_MAX, USHRT_MAX, USHRT_MAX, USHRT_MAX};
     GPU_vertbuf_attr_set(vert_buf, g_vbo_id.col[0], v_index, vcol);
   }
-#endif
+#  endif
 
-  /* Add default face sets color to avoid artifacts. */
-  const uchar face_set[3] = {UCHAR_MAX, UCHAR_MAX, UCHAR_MAX};
-  GPU_vertbuf_attr_set(vert_buf, g_vbo_id.fset, v_index, &face_set);
+  if (!g_vbo_id.fast_mode) {
+    /* Add default face sets color to avoid artifacts. */
+    const uchar face_set[3] = {UCHAR_MAX, UCHAR_MAX, UCHAR_MAX};
+    GPU_vertbuf_attr_set(vert_buf, g_vbo_id.fset, v_index, &face_set);
+  }
+#endif
 }
 
 /* Return the total number of vertices that don't have BM_ELEM_HIDDEN set */
@@ -1151,17 +1183,13 @@ static int gpu_bmesh_face_visible_count(TableGSet *bm_faces, int mat_nr)
 
 void GPU_pbvh_bmesh_buffers_update_free(GPU_PBVH_Buffers *buffers)
 {
-  if (buffers->smooth) {
-    /* Smooth needs to recreate index buffer, so we have to invalidate the batch. */
-    GPU_BATCH_DISCARD_SAFE(buffers->triangles);
-    GPU_BATCH_DISCARD_SAFE(buffers->lines);
-    GPU_INDEXBUF_DISCARD_SAFE(buffers->index_lines_buf);
-    GPU_INDEXBUF_DISCARD_SAFE(buffers->index_buf);
-  }
-  else {
-    GPU_BATCH_DISCARD_SAFE(buffers->lines);
-    GPU_INDEXBUF_DISCARD_SAFE(buffers->index_lines_buf);
+  if (buffers->last_tribuf_tris) {
+    // bmesh indexed drawing frees buffers by itself
+    return;
   }
+
+  GPU_BATCH_DISCARD_SAFE(buffers->lines);
+  GPU_INDEXBUF_DISCARD_SAFE(buffers->index_lines_buf);
 }
 
 static int gpu_pbvh_bmesh_make_vcol_offs(CustomData *vdata,
@@ -1232,12 +1260,17 @@ bool GPU_pbvh_need_full_render_get()
   return g_vbo_id.need_full_render;
 }
 
-void GPU_pbvh_update_attribute_names(CustomData *vdata, CustomData *ldata, bool need_full_render)
+void GPU_pbvh_update_attribute_names(CustomData *vdata,
+                                     CustomData *ldata,
+                                     bool need_full_render,
+                                     bool fast_mode)
 {
   const bool active_only = !need_full_render;
 
   GPU_vertformat_clear(&g_vbo_id.format);
 
+  g_vbo_id.fast_mode = fast_mode;
+
   // g_vbo_id.loop_attrs = build_cd_layers(vdata, )
   /* Initialize vertex buffer (match 'VertexBufferFormat'). */
   if (g_vbo_id.format.attr_len == 0) {
@@ -1259,8 +1292,12 @@ void GPU_pbvh_update_attribute_names(CustomData *vdata, CustomData *ldata, bool
     return;
 #endif
 
-    g_vbo_id.msk = GPU_vertformat_attr_add(
-        &g_vbo_id.format, "msk", GPU_COMP_U8, 1, GPU_FETCH_INT_TO_FLOAT_UNIT);
+#ifndef GPU_PERF_TEST
+    if (!fast_mode) {
+      g_vbo_id.msk = GPU_vertformat_attr_add(
+          &g_vbo_id.format, "msk", GPU_COMP_U8, 1, GPU_FETCH_INT_TO_FLOAT_UNIT);
+    }
+#endif
 
     g_vbo_id.totcol = 0;
 
@@ -1302,7 +1339,7 @@ void GPU_pbvh_update_attribute_names(CustomData *vdata, CustomData *ldata, bool
     g_vbo_id.loop_attrs_len = BLI_array_len(lattr);
 #endif
 
-#ifndef NEW_ATTR_SYSTEM
+#if !defined(NEW_ATTR_SYSTEM) && !defined(GPU_PERF_TEST)
     if (vdata && CustomData_has_layer(vdata, CD_PROP_COLOR)) {
       const int act = CustomData_get_active_layer_index(vdata, CD_PROP_COLOR);
       int ci = 0;
@@ -1331,14 +1368,14 @@ void GPU_pbvh_update_attribute_names(CustomData *vdata, CustomData *ldata, bool
     }
 
     // ensure at least one vertex color layer
-    if (g_vbo_id.totcol == 0) {
+    if (!fast_mode && g_vbo_id.totcol == 0) {
       g_vbo_id.col[0] = GPU_vertformat_attr_add(
           &g_vbo_id.format, "c", GPU_COMP_U16, 4, GPU_FETCH_INT_TO_FLOAT_UNIT);
       g_vbo_id.totcol = 1;
 
       GPU_vertformat_alias_add(&g_vbo_id.format, "ac");
     }
-#else
+#elif !defined(GPU_PERF_TEST)
     // ensure at least one vertex color layer
     if (!vdata || !CustomData_has_layer(vdata, CD_PROP_COLOR)) {
       g_vbo_id.col[0] = GPU_vertformat_attr_add(
@@ -1349,20 +1386,25 @@ void GPU_pbvh_update_attribute_names(CustomData *vdata, CustomData *ldata, bool
     }
 
 #endif
-    g_vbo_id.fset = GPU_vertformat_attr_add(
-        &g_vbo_id.format, "fset", GPU_COMP_U8, 3, GPU_FETCH_INT_TO_FLOAT_UNIT);
 
-    g_vbo_id.uv = GPU_vertformat_attr_add(
-        &g_vbo_id.format, "uvs", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
-    GPU_vertformat_alias_add(&g_vbo_id.format, "u");
+#ifndef GPU_PERF_TEST
+    if (!fast_mode) {
+      g_vbo_id.fset = GPU_vertformat_attr_add(
+          &g_vbo_id.format, "fset", GPU_COMP_U8, 3, GPU_FETCH_INT_TO_FLOAT_UNIT);
+    }
 
     if (ldata && CustomData_has_layer(ldata, CD_MLOOPUV)) {
+      g_vbo_id.uv = GPU_vertformat_attr_add(
+          &g_vbo_id.format, "uvs", GPU_COMP_F32, 2, GPU_FETCH_FLOAT);
+      GPU_vertformat_alias_add(&g_vbo_id.format, "u");
+
       const int cd_uv_index = CustomData_get_layer_index(ldata, CD_MLOOPUV);
       CustomDataLayer *cl = ldata->layers + cd_uv_index;
       cl += cl->active;
 
       DRW_make_cdlayer_attr_aliases(&g_vbo_id.format, "u", ldata, cl);
     }
+#endif
   }
 }
 
@@ -1568,29 +1610,37 @@ static void GPU_pbvh_bmesh_buffers_update_flat_vcol(GPU_PBVH_Buffers *buffers,
   /* Get material index from the last face we iterated on. */
   buffers->material_index = mat_nr;
 
-  buffers->show_overlay = !empty_mask || !default_face_set;
+  buffers->show_overlay = (!empty_mask || !default_face_set) && !g_vbo_id.fast_mode;
 
   gpu_pbvh_batch_init(buffers, GPU_PRIM_TRIS);
 }
 
-static void GPU_pbvh_bmesh_buffers_update_indexed(GPU_PBVH_Buffers *buffers,
-                                                  BMesh *bm,
-                                                  TableGSet *bm_faces,
-                                                  TableGSet *bm_unique_verts,
-                                                  TableGSet *bm_other_verts,
-                                                  PBVHTriBuf *tribuf,
-                                                  const int update_flags,
-                                                  const int cd_vert_node_offset,
-                                                  int face_sets_color_seed,
-                                                  int face_sets_color_default,
-                                                  bool flat_vcol,
-                                                  short mat_nr)
+ATTR_NO_OPT static void GPU_pbvh_bmesh_buffers_update_indexed(GPU_PBVH_Buffers *buffers,
+                                                              BMesh *bm,
+                                                              TableGSet *bm_faces,
+                                                              TableGSet *bm_unique_verts,
+                                                              TableGSet *bm_other_verts,
+                                                              PBVHTriBuf *tribuf,
+                                                              const int update_flags,
+                                                              const int cd_vert_node_offset,
+                                                              int face_sets_color_seed,
+                                                              int face_sets_color_default,
+                                                              bool flat_vcol,
+                                                              short mat_nr)
 {
 
   bool active_vcol_only = g_vbo_id.active_vcol_only;
 
-  const bool show_mask = (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0;
-  const bool show_vcol = (update_flags & GPU_PBVH_BUFFERS_SHOW_VCOL) != 0;
+  const bool show_mask = (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0 && !g_vbo_id.fast_mode;
+  bool show_vcol = (update_flags & GPU_PBVH_BUFFERS_SHOW_VCOL) != 0;
+
+  if (g_vbo_id.totcol == 0 && g_vbo_id.fast_mode) {
+    show_vcol = false;
+  }
+
+  bool need_indexed = buffers->last_tribuf_tris != tribuf->tris;
+
+  buffers->last_tribuf_tris = tribuf->tris;
 
   int tottri, totvert;
   bool empty_mask = true;
@@ -1629,11 +1679,6 @@ static void GPU_pbvh_bmesh_buffers_update_indexed(GPU_PBVH_Buffers *buffers,
     return;
   }
 
-  /* Fill the vertex and triangle buffer in one pass over faces. */
-  GPUIndexBufBuilder elb, elb_lines;
-  GPU_indexbuf_init(&elb, GPU_PRIM_TRIS, tottri, totvert);
-  GPU_indexbuf_init(&elb_lines, GPU_PRIM_LINES, tottri * 3, totvert);
-
 #ifdef QUANTIZED_PERF_TEST
   GPUVertBuf *vert_buf = buffers->vert_buf;
 
@@ -1712,6 +1757,24 @@ static void GPU_pbvh_bmesh_buffers_update_indexed(GPU_PBVH_Buffers *buffers,
 #endif
   }
 
+  if (!need_indexed) {
+    buffers->material_index = mat_nr;
+    buffers->show_overlay = (!empty_mask || !default_face_set) && !g_vbo_id.fast_mode;
+
+    gpu_pbvh_batch_init(buffers, GPU_PRIM_TRIS);
+    return;
+  }
+
+  GPU_BATCH_DISCARD_SAFE(buffers->triangles);
+  GPU_BATCH_DISCARD_SAFE(buffers->lines);
+  GPU_INDEXBUF_DISCARD_SAFE(buffers->index_lines_buf);
+  GPU_INDEXBUF_DISCARD_SAFE(buffers->index_buf);
+
+  /* Fill the vertex and triangle buffer in one pass over faces. */
+  GPUIndexBufBuilder elb, elb_lines;
+  GPU_indexbuf_init(&elb, GPU_PRIM_TRIS, tottri, totvert);
+  GPU_indexbuf_init(&elb_lines, GPU_PRIM_LINES, tottri * 3, totvert);
+
   for (int i = 0; i < tribuf->tottri; i++) {
     PBVHTri *tri = tribuf->tris + i;
 
@@ -1735,7 +1798,7 @@ static void GPU_pbvh_bmesh_buffers_update_indexed(GPU_PBVH_Buffers *buffers,
   buffers->index_lines_buf = GPU_indexbuf_build(&elb_lines);
 
   buffers->material_index = mat_nr;
-  buffers->show_overlay = !empty_mask || !default_face_set;
+  buffers->show_overlay = (!empty_mask || !default_face_set) && !g_vbo_id.fast_mode;
 
   gpu_pbvh_batch_init(buffers, GPU_PRIM_TRIS);
 
@@ -1779,9 +1842,10 @@ void GPU_pbvh_bmesh_buffers_update(GPU_PBVH_Buffers *buffers,
   }
 
   const bool have_uv = CustomData_has_layer(&bm->ldata, CD_MLOOPUV);
-  const bool show_mask = (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0;
+  const bool show_mask = (update_flags & GPU_PBVH_BUFFERS_SHOW_MASK) != 0 && !g_vbo_id.fast_mode;
   const bool show_face_sets = CustomData_has_layer(&bm->pdata, CD_SCULPT_FACE_SETS) &&
-                              (update_flags & GPU_PBVH_BUFFERS_SHOW_SCULPT_FACE_SETS) != 0;
+                              (update_flags & GPU_PBVH_BUFFERS_SHOW_SCULPT_FACE_SETS) != 0 &&
+                              !g_vbo_id.fast_mode;
 
   int tottri, totvert;
   bool empty_mask = true;
@@ -1810,6 +1874,8 @@ void GPU_pbvh_bmesh_buffers_update(GPU_PBVH_Buffers *buffers,
     return;
   }
 
+  buffers->last_tribuf_tris = NULL;
+
   /* TODO, make mask layer optional for bmesh buffer */
   const int cd_vert_mask_offset = CustomData_get_offset(&bm->vdata, CD_PAINT_MASK);
   const int cd_mcol_offset = CustomData_get_offset(&bm->ldata, CD_MLOOPCOL);
@@ -1903,7 +1969,7 @@ void GPU_pbvh_bmesh_buffers_update(GPU_PBVH_Buffers *buffers,
                                     &empty_mask,
                                     NULL,
                                     0);
-
+#  ifndef GPU_PERF_TEST
       if (cd_vcol_count >= 0) {
         for (int k = 0; k < cd_vcol_count; k++) {
           MPropCol *mp = BM_ELEM_CD_GET_VOID_P(l[j]->v, cd_vcols[k]);
@@ -1938,8 +2004,10 @@ void GPU_pbvh_bmesh_buffers_update(GPU_PBVH_Buffers *buffers,
         GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.uv, v_index, mu->uv);
       }
 
-      GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, v_index, face_set_color);
-
+      if (show_face_sets) {
+        GPU_vertbuf_attr_set(buffers->vert_buf, g_vbo_id.fset, v_index, face_set_color);
+      }
+#  endif
       v_index++;
     }
   }
@@ -2068,7 +2136,7 @@ void GPU_pbvh_bmesh_buffers_update(GPU_PBVH_Buffers *buffers,
 
   /* Get material index from the last face we iterated on. */
   buffers->material_index = mat_nr;
-  buffers->show_overlay = !empty_mask || !default_face_set;
+  buffers->show_overlay = (!empty_mask || !default_face_set) && !g_vbo_id.fast_mode;
 
   gpu_pbvh_batch_init(buffers, GPU_PRIM_TRIS);
 }
@@ -2084,8 +2152,8 @@ GPU_PBVH_Buffers *GPU_pbvh_bmesh_buffers_build(bool smooth_shading)
 
   buffers = MEM_callocN(sizeof(GPU_PBVH_Buffers), "GPU_Buffers");
   buffers->use_bmesh = true;
-  buffers->smooth = smooth_shading;
-  buffers->show_overlay = true;
+  buffers->smooth = smooth_shading || g_vbo_id.fast_mode;
+  buffers->show_overlay = (!g_vbo_id.fast_mode) && !g_vbo_id.fast_mode;
 
   return buffers;
 }
@@ -2101,7 +2169,7 @@ GPUBatch *GPU_pbvh_buffers_batch_get(GPU_PBVH_Buffers *buffers, bool fast, bool
 
 bool GPU_pbvh_buffers_has_overlays(GPU_PBVH_Buffers *buffers)
 {
-  return buffers->show_overlay;
+  return buffers->show_overlay && !g_vbo_id.fast_mode;
 }
 
 short GPU_pbvh_buffers_material_index_get(GPU_PBVH_Buffers *buffers)
diff --git a/source/blender/makesdna/DNA_scene_types.h b/source/blender/makesdna/DNA_scene_types.h
index 05a40f46977..3f3bc876126 100644
--- a/source/blender/makesdna/DNA_scene_types.h
+++ b/source/blender/makesdna/DNA_scene_types.h
@@ -484,7 +484,7 @@ typedef struct ImageFormatData {
 #define R_IMF_IMTYPE_INVALID 255
 
 /** #ImageFormatData.flag */
-#define R_IMF_FLAG_ZBUF (1 << 0)        /* was R_OPENEXR_ZBUF */
+#define R_IMF_FLAG_ZBUF (1 << 0) /* was R_OPENEXR_ZBUF */
 #define R_IMF_FLAG_PREVIEW_JPG (1 << 1) /* was R_PREVIEW_JPG */
 
 /* Return values from #BKE_imtype_valid_depths, note this is depths per channel. */
@@ -526,8 +526,8 @@ typedef enum eImageFormatDepth {
 
 /** #ImageFormatData.jp2_flag */
 #define R_IMF_JP2_FLAG_YCC (1 << 0) /* when disabled use RGB */ /* was R_JPEG2K_YCC */
-#define R_IMF_JP2_FLAG_CINE_PRESET (1 << 1)                     /* was R_JPEG2K_CINE_PRESET */
-#define R_IMF_JP2_FLAG_CINE_48 (1 << 2)                         /* was R_JPEG2K_CINE_48FPS */
+#define R_IMF_JP2_FLAG_CINE_PRESET (1 << 1) /* was R_JPEG2K_CINE_PRESET */
+#define R_IMF_JP2_FLAG_CINE_48 (1 << 2) /* was R_JPEG2K_CINE_48FPS */
 
 /** #ImageFormatData.jp2_codec */
 #define R_IMF_JP2_CODEC_JP2 0
@@ -1842,12 +1842,12 @@ typedef struct Scene {
 
 #define R_MODE_UNUSED_20 (1 << 20) /* cleared */
 #define R_MODE_UNUSED_21 (1 << 21) /* cleared */
-#define R_NO_OVERWRITE (1 << 22)   /* skip existing files */
-#define R_TOUCH (1 << 23)          /* touch files before rendering */
+#define R_NO_OVERWRITE (1 << 22) /* skip existing files */
+#define R_TOUCH (1 << 23) /* touch files before rendering */
 #define R_SIMPLIFY (1 << 24)
-#define R_EDGE_FRS (1 << 25)        /* R_EDGE reserved for Freestyle */
+#define R_EDGE_FRS (1 << 25) /* R_EDGE reserved for Freestyle */
 #define R_PERSISTENT_DATA (1 << 26) /* keep data around for re-render */
-#define R_MODE_UNUSED_27 (1 << 27)  /* cleared */
+#define R_MODE_UNUSED_27 (1 << 27) /* cleared */
 
 /** #RenderData.seq_flag */
 enum {
@@ -2237,8 +2237,11 @@ typedef enum eSculptFlags {
 
   /* Don't display face sets in viewport. */
   SCULPT_HIDE_FACE_SETS = (1 << 17),
-  SCULPT_DYNTOPO_FLAT_VCOL_SHADING = (1<<18),
+  SCULPT_DYNTOPO_FLAT_VCOL_SHADING = (1 << 18),
   SCULPT_DYNTOPO_CLEANUP = (1 << 19),
+
+  // hides facesets/masks and forces indexed mode to save GPU bandwidth
+  SCULPT_FAST_DRAW = (1 << 20),
 } eSculptFlags;
 
 /** #ImagePaintSettings.mode */
diff --git a/source/blender/makesrna/intern/rna_sculpt_paint.c b/source/blender/makesrna/intern/rna_sculpt_paint.c
index 579dcf6aeb8..3d5e5ed8896 100644
--- a/source/blender/makesrna/intern/rna_sculpt_paint.c
+++ b/source/blender/makesrna/intern/rna_sculpt_paint.c
@@ -382,6 +382,7 @@ static void rna_Sculpt_update(bContext *C, PointerRNA *UNUSED(ptr))
 
       ob->sculpt->bm_smooth_shading = ((scene->toolsettings->sculpt->flags &
                                         SCULPT_DYNTOPO_SMOOTH_SHADING) != 0);
+      ob->sculpt->fast_draw = ((scene->toolsettings->sculpt->flags & SCULPT_FAST_DRAW) != 0);
     }
 
     DEG_id_tag_update(&ob->id, ID_RECALC_GEOMETRY);
@@ -831,6 +832,17 @@ static void rna_def_sculpt(BlenderRNA *brna)
   RNA_def_property_flag(prop, PROP_CONTEXT_UPDATE);
   RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, "rna_Sculpt_update");
 
+  prop = RNA_def_property(srna, "use_fast_draw", PROP_BOOLEAN, PROP_NONE);
+  RNA_def_property_boolean_sdna(prop, NULL, "flags", SCULPT_FAST_DRAW);
+  RNA_def_property_ui_text(prop,
+                           "Fast Draw Mode",
+                           "Forces smooth shading"
+                           "and disables drawing of masks and face sets"
+                           "to speed up drawing. Useful for posing"
+                           "high-poly meshes.");
+  RNA_def_property_flag(prop, PROP_CONTEXT_UPDATE);
+  RNA_def_property_update(prop, NC_OBJECT | ND_DRAW, "rna_Sculpt_update");
+
   prop = RNA_def_property(srna, "use_dyntopo_cleanup", PROP_BOOLEAN, PROP_NONE);
   RNA_def_property_boolean_sdna(prop, NULL, "flags", SCULPT_DYNTOPO_CLEANUP);
   RNA_def_property_ui_text(prop, "Cleanup", "Removes verts surrounded by only 3 or 4 edges");