Mesh Batch Cache: Move index buffer range hack to be more local

- Use Extract naming convention to name extract functions that fill vbo/ibo
- Separate extract functions into separate file (for clarity)
- Make simpler iter loops to avoid as much overhead as possible
- Separate loose elements looping functions to avoid iteration complexity
  (unfortunately this makes the code more verbose).
- Some iter functions are threadable and tagged as such.
- Add multithreaded iteration for extract functions that supports them.

source/blender/blenkernel/BKE_mesh.h

@@ -480,6 +480,7 @@ void BKE_mesh_calc_poly_center(const struct MPoly *mpoly,
 float BKE_mesh_calc_poly_area(const struct MPoly *mpoly,
                               const struct MLoop *loopstart,
                               const struct MVert *mvarray);
+float BKE_mesh_calc_poly_uv_area(const struct MPoly *mpoly, const struct MLoopUV *uv_array);
 void BKE_mesh_calc_poly_angles(const struct MPoly *mpoly,
                                const struct MLoop *loopstart,
                                const struct MVert *mvarray,

source/blender/blenkernel/intern/mesh_evaluate.c

@@ -2378,6 +2378,24 @@ float BKE_mesh_calc_poly_area(const MPoly *mpoly, const MLoop *loopstart, const
   }
 }
 
+float BKE_mesh_calc_poly_uv_area(const MPoly *mpoly, const MLoopUV *uv_array)
+{
+
+  int i, l_iter = mpoly->loopstart;
+  float area;
+  float(*vertexcos)[2] = BLI_array_alloca(vertexcos, (size_t)mpoly->totloop);
+
+  /* pack vertex cos into an array for area_poly_v2 */
+  for (i = 0; i < mpoly->totloop; i++, l_iter++) {
+    copy_v2_v2(vertexcos[i], uv_array[l_iter].uv);
+  }
+
+  /* finally calculate the area */
+  area = area_poly_v2((const float(*)[2])vertexcos, (unsigned int)mpoly->totloop);
+
+  return area;
+}
+
 /**
  * Calculate the volume and volume-weighted centroid of the volume
  * formed by the polygon and the origin.

source/blender/draw/CMakeLists.txt

@@ -52,6 +52,7 @@ set(SRC
   intern/draw_anim_viz.c
   intern/draw_armature.c
   intern/draw_cache.c
+  intern/draw_cache_extract_mesh.c
   intern/draw_cache_impl_curve.c
   intern/draw_cache_impl_displist.c
   intern/draw_cache_impl_lattice.c
@@ -130,6 +131,7 @@ set(SRC
   DRW_engine.h
   intern/DRW_render.h
   intern/draw_cache.h
+  intern/draw_cache_extract.h
   intern/draw_cache_impl.h
   intern/draw_cache_inline.h
   intern/draw_common.h

source/blender/draw/intern/draw_cache.c

@@ -4033,7 +4033,7 @@ void drw_batch_cache_validate(Object *ob)
 void drw_batch_cache_generate_requested(Object *ob)
 {
   const DRWContextState *draw_ctx = DRW_context_state_get();
-  const ToolSettings *ts = draw_ctx->scene->toolsettings;
+  const Scene *scene = draw_ctx->scene;
   const enum eContextObjectMode mode = CTX_data_mode_enum_ex(
       draw_ctx->object_edit, draw_ctx->obact, draw_ctx->object_mode);
   const bool is_paint_mode = ELEM(
@@ -4047,13 +4047,13 @@ void drw_batch_cache_generate_requested(Object *ob)
   struct Mesh *mesh_eval = ob->runtime.mesh_eval;
   switch (ob->type) {
     case OB_MESH:
-      DRW_mesh_batch_cache_create_requested(ob, (Mesh *)ob->data, ts, is_paint_mode, use_hide);
+      DRW_mesh_batch_cache_create_requested(ob, (Mesh *)ob->data, scene, is_paint_mode, use_hide);
       break;
     case OB_CURVE:
     case OB_FONT:
     case OB_SURF:
       if (mesh_eval) {
-        DRW_mesh_batch_cache_create_requested(ob, mesh_eval, ts, is_paint_mode, use_hide);
+        DRW_mesh_batch_cache_create_requested(ob, mesh_eval, scene, is_paint_mode, use_hide);
       }
       DRW_curve_batch_cache_create_requested(ob);
       break;

source/blender/draw/intern/draw_cache_extract.h

@@ -0,0 +1,244 @@
+/*
+ * 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.
+ *
+ * Copyright 2019, Blender Foundation.
+ */
+
+/** \file
+ * \ingroup draw
+ */
+
+#ifndef __DRAW_CACHE_EXTRACT_MESH_H__
+#define __DRAW_CACHE_EXTRACT_MESH_H__
+
+/* Vertex Group Selection and display options */
+typedef struct DRW_MeshWeightState {
+  int defgroup_active;
+  int defgroup_len;
+
+  short flags;
+  char alert_mode;
+
+  /* Set of all selected bones for Multipaint. */
+  bool *defgroup_sel; /* [defgroup_len] */
+  int defgroup_sel_count;
+} DRW_MeshWeightState;
+
+/* DRW_MeshWeightState.flags */
+enum {
+  DRW_MESH_WEIGHT_STATE_MULTIPAINT = (1 << 0),
+  DRW_MESH_WEIGHT_STATE_AUTO_NORMALIZE = (1 << 1),
+};
+
+typedef struct DRW_MeshCDMask {
+  uint32_t uv : 8;
+  uint32_t tan : 8;
+  uint32_t vcol : 8;
+  uint32_t orco : 1;
+  uint32_t tan_orco : 1;
+} DRW_MeshCDMask;
+
+typedef enum eMRIterType {
+  MR_ITER_LOOPTRI = 1 << 0,
+  MR_ITER_LOOP = 1 << 1,
+  MR_ITER_LEDGE = 1 << 2,
+  MR_ITER_LVERT = 1 << 3,
+} eMRIterType;
+
+typedef enum eMRDataType {
+  MR_DATA_POLY_NOR = 1 << 1,
+  MR_DATA_LOOP_NOR = 1 << 2,
+  MR_DATA_LOOPTRI = 1 << 3,
+} eMRDataType;
+
+typedef enum eMRExtractType {
+  MR_EXTRACT_BMESH,
+  MR_EXTRACT_MAPPED,
+  MR_EXTRACT_MESH,
+} eMRExtractType;
+
+BLI_INLINE int mesh_render_mat_len_get(Mesh *me)
+{
+  return MAX2(1, me->totcol);
+}
+
+typedef struct MeshBufferCache {
+  /* Every VBO below contains at least enough
+   * data for every loops in the mesh (except fdots).
+   * For some VBOs, it extends to (in this exact order) :
+   * loops + loose_edges*2 + loose_verts */
+  struct {
+    GPUVertBuf *pos_nor;  /* extend */
+    GPUVertBuf *lnor;     /* extend */
+    GPUVertBuf *edge_fac; /* extend */
+    GPUVertBuf *weights;  /* extend */
+    GPUVertBuf *uv_tan;
+    GPUVertBuf *vcol;
+    GPUVertBuf *orco;
+    /* Only for edit mode. */
+    GPUVertBuf *edit_data; /* extend */
+    GPUVertBuf *edituv_data;
+    GPUVertBuf *stretch_area;
+    GPUVertBuf *stretch_angle;
+    GPUVertBuf *mesh_analysis;
+    GPUVertBuf *fdots_pos;
+    GPUVertBuf *fdots_nor;
+    GPUVertBuf *fdots_uv;
+    // GPUVertBuf *fdots_edit_data; /* inside fdots_nor for now. */
+    GPUVertBuf *fdots_edituv_data;
+    /* Selection */
+    GPUVertBuf *vert_idx; /* extend */
+    GPUVertBuf *edge_idx; /* extend */
+    GPUVertBuf *poly_idx;
+    GPUVertBuf *fdot_idx;
+  } vbo;
+  /* Index Buffers:
+   * Only need to be updated when topology changes. */
+  struct {
+    /* Indices to vloops. */
+    GPUIndexBuf *tris;  /* Ordered per material. */
+    GPUIndexBuf *lines; /* Loose edges last. */
+    GPUIndexBuf *points;
+    GPUIndexBuf *fdots;
+    /* 3D overlays. */
+    GPUIndexBuf *lines_paint_mask; /* no loose edges. */
+    GPUIndexBuf *lines_adjacency;
+    /* Uv overlays. (visibility can differ from 3D view) */
+    GPUIndexBuf *edituv_tris;
+    GPUIndexBuf *edituv_lines;
+    GPUIndexBuf *edituv_points;
+    GPUIndexBuf *edituv_fdots;
+  } ibo;
+} MeshBufferCache;
+
+typedef enum DRWBatchFlag {
+  MBC_SURFACE = (1 << 0),
+  MBC_SURFACE_WEIGHTS = (1 << 1),
+  MBC_EDIT_TRIANGLES = (1 << 2),
+  MBC_EDIT_VERTICES = (1 << 3),
+  MBC_EDIT_EDGES = (1 << 4),
+  MBC_EDIT_VNOR = (1 << 5),
+  MBC_EDIT_LNOR = (1 << 6),
+  MBC_EDIT_FACEDOTS = (1 << 7),
+  MBC_EDIT_MESH_ANALYSIS = (1 << 8),
+  MBC_EDITUV_FACES_STRECH_AREA = (1 << 9),
+  MBC_EDITUV_FACES_STRECH_ANGLE = (1 << 10),
+  MBC_EDITUV_FACES = (1 << 11),
+  MBC_EDITUV_EDGES = (1 << 12),
+  MBC_EDITUV_VERTS = (1 << 13),
+  MBC_EDITUV_FACEDOTS = (1 << 14),
+  MBC_EDIT_SELECTION_VERTS = (1 << 15),
+  MBC_EDIT_SELECTION_EDGES = (1 << 16),
+  MBC_EDIT_SELECTION_FACES = (1 << 17),
+  MBC_EDIT_SELECTION_FACEDOTS = (1 << 18),
+  MBC_ALL_VERTS = (1 << 19),
+  MBC_ALL_EDGES = (1 << 20),
+  MBC_LOOSE_EDGES = (1 << 21),
+  MBC_EDGE_DETECTION = (1 << 22),
+  MBC_WIRE_EDGES = (1 << 23),
+  MBC_WIRE_LOOPS = (1 << 24),
+  MBC_WIRE_LOOPS_UVS = (1 << 25),
+  MBC_SURF_PER_MAT = (1 << 26),
+} DRWBatchFlag;
+
+#define MBC_EDITUV \
+  (MBC_EDITUV_FACES_STRECH_AREA | MBC_EDITUV_FACES_STRECH_ANGLE | MBC_EDITUV_FACES | \
+   MBC_EDITUV_EDGES | MBC_EDITUV_VERTS | MBC_EDITUV_FACEDOTS | MBC_WIRE_LOOPS_UVS)
+
+#define FOREACH_MESH_BUFFER_CACHE(batch_cache, mbc) \
+  for (MeshBufferCache *mbc = &batch_cache->final; \
+       mbc == &batch_cache->final || mbc == &batch_cache->cage; \
+       mbc = (mbc == &batch_cache->final) ? &batch_cache->cage : NULL)
+
+typedef struct MeshBatchCache {
+  MeshBufferCache final, cage;
+
+  struct {
+    /* Surfaces / Render */
+    GPUBatch *surface;
+    GPUBatch *surface_weights;
+    /* Edit mode */
+    GPUBatch *edit_triangles;
+    GPUBatch *edit_vertices;
+    GPUBatch *edit_edges;
+    GPUBatch *edit_vnor;
+    GPUBatch *edit_lnor;
+    GPUBatch *edit_fdots;
+    GPUBatch *edit_mesh_analysis;
+    /* Edit UVs */
+    GPUBatch *edituv_faces_strech_area;
+    GPUBatch *edituv_faces_strech_angle;
+    GPUBatch *edituv_faces;
+    GPUBatch *edituv_edges;
+    GPUBatch *edituv_verts;
+    GPUBatch *edituv_fdots;
+    /* Edit selection */
+    GPUBatch *edit_selection_verts;
+    GPUBatch *edit_selection_edges;
+    GPUBatch *edit_selection_faces;
+    GPUBatch *edit_selection_fdots;
+    /* Common display / Other */
+    GPUBatch *all_verts;
+    GPUBatch *all_edges;
+    GPUBatch *loose_edges;
+    GPUBatch *edge_detection;
+    GPUBatch *wire_edges;     /* Individual edges with face normals. */
+    GPUBatch *wire_loops;     /* Loops around faces. no edges between selected faces */
+    GPUBatch *wire_loops_uvs; /* Same as wire_loops but only has uvs. */
+  } batch;
+
+  GPUBatch **surface_per_mat;
+
+  /* arrays of bool uniform names (and value) that will be use to
+   * set srgb conversion for auto attributes.*/
+  char *auto_layer_names;
+  int *auto_layer_is_srgb;
+  int auto_layer_len;
+
+  DRWBatchFlag batch_requested;
+  DRWBatchFlag batch_ready;
+
+  /* settings to determine if cache is invalid */
+  int edge_len;
+  int tri_len;
+  int poly_len;
+  int vert_len;
+  int mat_len;
+  bool is_dirty; /* Instantly invalidates cache, skipping mesh check */
+  bool is_editmode;
+  bool is_uvsyncsel;
+
+  struct DRW_MeshWeightState weight_state;
+
+  DRW_MeshCDMask cd_used, cd_needed, cd_used_over_time;
+
+  int lastmatch;
+
+  /* Valid only if edge_detection is up to date. */
+  bool is_manifold;
+
+  bool no_loose_wire;
+} MeshBatchCache;
+
+void mesh_buffer_cache_create_requested(MeshBatchCache *cache,
+                                        MeshBufferCache mbc,
+                                        Mesh *me,
+                                        const bool do_final,
+                                        const bool use_subsurf_fdots,
+                                        const DRW_MeshCDMask *cd_layer_used,
+                                        const ToolSettings *ts,
+                                        const bool use_hide);
+
+#endif /* __DRAW_CACHE_EXTRACT_MESH_H__ */

source/blender/draw/intern/draw_cache_impl.h

@@ -119,7 +119,7 @@ struct GPUBatch *DRW_lattice_batch_cache_get_edit_verts(struct Lattice *lt);
 /* Mesh */
 void DRW_mesh_batch_cache_create_requested(struct Object *ob,
                                            struct Mesh *me,
-                                           const struct ToolSettings *ts,
+                                           const struct Scene *scene,
                                            const bool is_paint_mode,
                                            const bool use_hide);
 
@@ -143,6 +143,7 @@ struct GPUBatch *DRW_mesh_batch_cache_get_surface_weights(struct Mesh *me);
 struct GPUBatch *DRW_mesh_batch_cache_get_edit_triangles(struct Mesh *me);
 struct GPUBatch *DRW_mesh_batch_cache_get_edit_vertices(struct Mesh *me);
 struct GPUBatch *DRW_mesh_batch_cache_get_edit_edges(struct Mesh *me);
+struct GPUBatch *DRW_mesh_batch_cache_get_edit_vnors(struct Mesh *me);
 struct GPUBatch *DRW_mesh_batch_cache_get_edit_lnors(struct Mesh *me);
 struct GPUBatch *DRW_mesh_batch_cache_get_edit_facedots(struct Mesh *me);
 /* edit-mesh selection */

source/blender/draw/intern/draw_manager.c

@@ -2772,7 +2772,7 @@ void DRW_draw_select_id_object(Scene *scene,
         if (use_faceselect && draw_facedot) {
           geom_facedots = DRW_mesh_batch_cache_get_facedots_with_select_id(me);
         }
-        DRW_mesh_batch_cache_create_requested(ob, me, NULL, false, true);
+        DRW_mesh_batch_cache_create_requested(ob, me, scene, false, true);
 
         draw_mesh_face(geom_faces, initial_offset, use_faceselect, world_clip_planes);
 
@@ -2824,7 +2824,7 @@ void DRW_draw_select_id_object(Scene *scene,
             ((ob->mode & OB_MODE_WEIGHT_PAINT) || (ob->mode & OB_MODE_VERTEX_PAINT))) {
 
           GPUBatch *geom_verts = DRW_mesh_batch_cache_get_verts_with_select_id(me_eval);
-          DRW_mesh_batch_cache_create_requested(ob, me_eval, NULL, false, true);
+          DRW_mesh_batch_cache_create_requested(ob, me_eval, scene, false, true);
 
           /* Only draw faces to mask out verts, we don't want their selection ID's. */
           draw_mesh_face(geom_faces, 0, false, world_clip_planes);
@@ -2835,7 +2835,7 @@ void DRW_draw_select_id_object(Scene *scene,
         }
         else {
           const bool use_hide = (me_orig->editflag & ME_EDIT_PAINT_FACE_SEL);
-          DRW_mesh_batch_cache_create_requested(ob, me_eval, NULL, false, use_hide);
+          DRW_mesh_batch_cache_create_requested(ob, me_eval, scene, false, use_hide);
 
           draw_mesh_face(geom_faces, initial_offset, true, world_clip_planes);
 

source/blender/draw/modes/edit_mesh_mode.c

@@ -727,7 +727,7 @@ static void EDIT_MESH_cache_populate(void *vedata, Object *ob)
       }
 
       if (vnormals_do) {
-        geom = DRW_mesh_batch_cache_get_edit_vertices(ob->data);
+        geom = DRW_mesh_batch_cache_get_edit_vnors(ob->data);
         DRW_shgroup_call_no_cull(g_data->vnormals_shgrp, geom, ob);
       }
       if (lnormals_do) {

source/blender/editors/sculpt_paint/sculpt_uv.c

@@ -488,6 +488,7 @@ static bool uv_edge_compare(const void *a, const void *b)
 
 static UvSculptData *uv_sculpt_stroke_init(bContext *C, wmOperator *op, const wmEvent *event)
 {
+  Depsgraph *depsgraph = CTX_data_depsgraph(C);
   Scene *scene = CTX_data_scene(C);
   Object *obedit = CTX_data_edit_object(C);
   ToolSettings *ts = scene->toolsettings;
@@ -554,7 +555,7 @@ static UvSculptData *uv_sculpt_stroke_init(bContext *C, wmOperator *op, const wm
       UvElement *element;
       UvNearestHit hit = UV_NEAREST_HIT_INIT;
       Image *ima = CTX_data_edit_image(C);
-      uv_find_nearest_vert(scene, ima, obedit, co, 0.0f, &hit);
+      uv_find_nearest_vert(depsgraph, scene, ima, obedit, co, 0.0f, &hit);
 
       element = BM_uv_element_get(data->elementMap, hit.efa, hit.l);
       island_index = element->island;

source/blender/editors/uvedit/uvedit_draw.c

@@ -163,13 +163,13 @@ void ED_image_draw_cursor(ARegion *ar, const float cursor[2])
 
 static void uvedit_get_batches(Object *ob,
                                SpaceImage *sima,
-                               const ToolSettings *ts,
+                               const Scene *scene,
                                GPUBatch **faces,
                                GPUBatch **edges,
                                GPUBatch **verts,
                                GPUBatch **facedots)
 {
-  int drawfaces = draw_uvs_face_check(ts);
+  int drawfaces = draw_uvs_face_check(scene->toolsettings);
   const bool draw_stretch = (sima->flag & SI_DRAW_STRETCH) != 0;
   const bool draw_faces = (sima->flag & SI_NO_DRAWFACES) == 0;
 
@@ -197,7 +197,7 @@ static void uvedit_get_batches(Object *ob,
     *faces = NULL;
   }
 
-  DRW_mesh_batch_cache_create_requested(ob, ob->data, ts, false, false);
+  DRW_mesh_batch_cache_create_requested(ob, ob->data, scene, false, false);
 }
 
 static void draw_uvs_shadow(SpaceImage *UNUSED(sima),
@@ -212,7 +212,7 @@ static void draw_uvs_shadow(SpaceImage *UNUSED(sima),
 
   DRW_mesh_batch_cache_validate(me);
   GPUBatch *edges = DRW_mesh_batch_cache_get_uv_edges(me);
-  DRW_mesh_batch_cache_create_requested(eval_ob, me, scene->toolsettings, false, false);
+  DRW_mesh_batch_cache_create_requested(eval_ob, me, scene, false, false);
 
   if (edges) {
     GPU_batch_program_set_builtin(edges, GPU_SHADER_2D_UV_UNIFORM_COLOR);
@@ -235,7 +235,7 @@ static void draw_uvs_texpaint(Scene *scene, Object *ob, Depsgraph *depsgraph)
 
   DRW_mesh_batch_cache_validate(me);
   GPUBatch *geom = DRW_mesh_batch_cache_get_uv_edges(me);
-  DRW_mesh_batch_cache_create_requested(eval_ob, me, scene->toolsettings, false, false);
+  DRW_mesh_batch_cache_create_requested(eval_ob, me, scene, false, false);
 
   GPU_batch_program_set_builtin(geom, GPU_SHADER_2D_UV_UNIFORM_COLOR);
   GPU_batch_uniform_4fv(geom, "color", col);
@@ -300,7 +300,7 @@ static void draw_uvs(SpaceImage *sima, Scene *scene, Object *obedit, Depsgraph *
     }
   }
 
-  uvedit_get_batches(eval_ob, sima, ts, &faces, &edges, &verts, &facedots);
+  uvedit_get_batches(eval_ob, sima, scene, &faces, &edges, &verts, &facedots);
 
   bool interpedges;
   bool draw_stretch = (sima->flag & SI_DRAW_STRETCH) != 0;

source/blender/editors/uvedit/uvedit_intern.h

@@ -57,13 +57,15 @@ typedef struct UvNearestHit {
     .dist_sq = FLT_MAX, \
   }
 
-bool uv_find_nearest_vert(struct Scene *scene,
+bool uv_find_nearest_vert(struct Depsgraph *depsgraph,
+                          struct Scene *scene,
                           struct Image *ima,
                           struct Object *obedit,
                           const float co[2],
                           const float penalty_dist,
                           struct UvNearestHit *hit_final);
-bool uv_find_nearest_vert_multi(struct Scene *scene,
+bool uv_find_nearest_vert_multi(struct Depsgraph *depsgraph,
+                                struct Scene *scene,
                                 struct Image *ima,
                                 struct Object **objects,
                                 const uint objects_len,
@@ -71,24 +73,28 @@ bool uv_find_nearest_vert_multi(struct Scene *scene,
                                 const float penalty_dist,
                                 struct UvNearestHit *hit_final);
 
-bool uv_find_nearest_edge(struct Scene *scene,
+bool uv_find_nearest_edge(struct Depsgraph *depsgraph,
+                          struct Scene *scene,
                           struct Image *ima,
                           struct Object *obedit,
                           const float co[2],
                           struct UvNearestHit *hit_final);
-bool uv_find_nearest_edge_multi(struct Scene *scene,
+bool uv_find_nearest_edge_multi(struct Depsgraph *depsgraph,
+                                struct Scene *scene,
                                 struct Image *ima,
                                 struct Object **objects,
                                 const uint objects_len,
                                 const float co[2],
                                 struct UvNearestHit *hit_final);
 
-bool uv_find_nearest_face(struct Scene *scene,
+bool uv_find_nearest_face(struct Depsgraph *depsgraph,
+                          struct Scene *scene,
                           struct Image *ima,
                           struct Object *obedit,
                           const float co[2],
                           struct UvNearestHit *hit_final);
-bool uv_find_nearest_face_multi(struct Scene *scene,
+bool uv_find_nearest_face_multi(struct Depsgraph *depsgraph,
+                                struct Scene *scene,
                                 struct Image *ima,
                                 struct Object **objects,
                                 const uint objects_len,

source/blender/editors/uvedit/uvedit_ops.c

@@ -56,6 +56,7 @@
 #include "BKE_material.h"
 #include "BKE_mesh.h"
 #include "BKE_mesh_mapping.h"
+#include "BKE_modifier.h"
 #include "BKE_node.h"
 #include "BKE_report.h"
 #include "BKE_scene.h"
@@ -606,6 +607,18 @@ void uv_poly_center(BMFace *f, float r_cent[2], const int cd_loop_uv_offset)
   mul_v2_fl(r_cent, 1.0f / (float)f->len);
 }
 
+static void uv_mpoly_center(const MPoly *mpoly, const MLoopUV *mloopuv, float r_cent[2])
+{
+  zero_v2(r_cent);
+
+  for (int i = 0; i < mpoly->totloop; i++) {
+    const MLoopUV *luv = &mloopuv[mpoly->loopstart + i];
+    add_v2_v2(r_cent, luv->uv);
+  }
+
+  mul_v2_fl(r_cent, 1.0f / (float)mpoly->totloop);
+}
+
 void uv_poly_copy_aspect(float uv_orig[][2], float uv[][2], float aspx, float aspy, int len)
 {
   int i;
@@ -753,8 +766,12 @@ bool ED_uvedit_center(Scene *scene, Image *ima, Object *obedit, float cent[2], c
 /** \name Find Nearest Elements
  * \{ */
 
-bool uv_find_nearest_edge(
-    Scene *scene, Image *ima, Object *obedit, const float co[2], UvNearestHit *hit)
+bool uv_find_nearest_edge(Depsgraph *depsgraph,
+                          Scene *scene,
+                          Image *ima,
+                          Object *obedit,
+                          const float co[2],
+                          UvNearestHit *hit)
 {
   BMEditMesh *em = BKE_editmesh_from_object(obedit);
   BMFace *efa;
@@ -766,35 +783,85 @@ bool uv_find_nearest_edge(
 
   const int cd_loop_uv_offset = CustomData_get_offset(&em->bm->ldata, CD_MLOOPUV);
 
-  BM_mesh_elem_index_ensure(em->bm, BM_VERT);
+  /* (fclem) Is it garanteed that obedit_eval will be updated? */
+  Object *obedit_eval = DEG_get_evaluated_object(depsgraph, obedit);
+  BMEditMesh *em_eval = BKE_editmesh_from_object(obedit_eval);
 
-  BM_ITER_MESH (efa, &iter, em->bm, BM_FACES_OF_MESH) {
-    if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
-      continue;
+  if (!em_eval->mesh_eval_cage || em_eval->mesh_eval_cage->runtime.is_original) {
+    BM_mesh_elem_index_ensure(em->bm, BM_VERT);
+
+    BM_ITER_MESH (efa, &iter, em->bm, BM_FACES_OF_MESH) {
+      if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+        continue;
+      }
+      BM_ITER_ELEM_INDEX (l, &liter, efa, BM_LOOPS_OF_FACE, i) {
+        luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset);
+        luv_next = BM_ELEM_CD_GET_VOID_P(l->next, cd_loop_uv_offset);
+
+        const float dist_test_sq = dist_squared_to_line_segment_v2(co, luv->uv, luv_next->uv);
+
+        if (dist_test_sq < hit->dist_sq) {
+          hit->efa = efa;
+
+          hit->l = l;
+          hit->luv = luv;
+          hit->luv_next = luv_next;
+          hit->lindex = i;
+
+          hit->dist_sq = dist_test_sq;
+          found = true;
+        }
+      }
     }
-    BM_ITER_ELEM_INDEX (l, &liter, efa, BM_LOOPS_OF_FACE, i) {
-      luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset);
-      luv_next = BM_ELEM_CD_GET_VOID_P(l->next, cd_loop_uv_offset);
+  }
+  else {
+    Mesh *me = em_eval->mesh_eval_cage;
 
-      const float dist_test_sq = dist_squared_to_line_segment_v2(co, luv->uv, luv_next->uv);
+    BM_mesh_elem_table_ensure(em->bm, BM_FACE | BM_EDGE);
 
-      if (dist_test_sq < hit->dist_sq) {
-        hit->efa = efa;
+    int *p_origindex = CustomData_get_layer(&me->pdata, CD_ORIGINDEX);
+    int *e_origindex = CustomData_get_layer(&me->edata, CD_ORIGINDEX);
 
-        hit->l = l;
-        hit->luv = luv;
-        hit->luv_next = luv_next;
-        hit->lindex = i;
+    MLoopUV *mluv = CustomData_get_layer(&me->ldata, CD_MLOOPUV);
 
-        hit->dist_sq = dist_test_sq;
-        found = true;
+    MPoly *mpoly = me->mpoly;
+    for (int p = 0; p < me->totpoly; p++, mpoly++) {
+      if (p_origindex[p] == ORIGINDEX_NONE) {
+        continue;
+      }
+      efa = BM_face_at_index(em->bm, p_origindex[p]);
+      if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+        continue;
+      }
+      for (i = 0; i < mpoly->totloop; i++) {
+        MLoop *mloop = &me->mloop[mpoly->loopstart + i];
+        if (e_origindex[mloop->e] == ORIGINDEX_NONE) {
+          continue;
+        }
+        luv = &mluv[mpoly->loopstart + i];
+        luv_next = &mluv[mpoly->loopstart + ((i + 1) % mpoly->totloop)];
+        const float dist_test_sq = dist_squared_to_line_segment_v2(co, luv->uv, luv_next->uv);
+
+        if (dist_test_sq < hit->dist_sq) {
+          BMEdge *eed = BM_edge_at_index(em->bm, e_origindex[mloop->e]);
+          hit->efa = efa;
+
+          hit->l = BM_face_edge_share_loop(efa, eed);
+          hit->luv = luv;
+          hit->luv_next = luv_next;
+          hit->lindex = i;
+
+          hit->dist_sq = dist_test_sq;
+          found = true;
+        }
       }
     }
   }
   return found;
 }
 
-bool uv_find_nearest_edge_multi(Scene *scene,
+bool uv_find_nearest_edge_multi(Depsgraph *depsgraph,
+                                Scene *scene,
                                 Image *ima,
                                 Object **objects,
                                 const uint objects_len,
@@ -804,7 +871,7 @@ bool uv_find_nearest_edge_multi(Scene *scene,
   bool found = false;
   for (uint ob_index = 0; ob_index < objects_len; ob_index++) {
     Object *obedit = objects[ob_index];
-    if (uv_find_nearest_edge(scene, ima, obedit, co, hit_final)) {
+    if (uv_find_nearest_edge(depsgraph, scene, ima, obedit, co, hit_final)) {
       hit_final->ob = obedit;
       found = true;
     }
@@ -812,8 +879,12 @@ bool uv_find_nearest_edge_multi(Scene *scene,
   return found;
 }
 
-bool uv_find_nearest_face(
-    Scene *scene, Image *ima, Object *obedit, const float co[2], UvNearestHit *hit_final)
+bool uv_find_nearest_face(Depsgraph *depsgraph,
+                          Scene *scene,
+                          Image *ima,
+                          Object *obedit,
+                          const float co[2],
+                          UvNearestHit *hit_final)
 {
   BMEditMesh *em = BKE_editmesh_from_object(obedit);
   bool found = false;
@@ -823,28 +894,94 @@ bool uv_find_nearest_face(
   /* this will fill in hit.vert1 and hit.vert2 */
   float dist_sq_init = hit_final->dist_sq;
   UvNearestHit hit = *hit_final;
-  if (uv_find_nearest_edge(scene, ima, obedit, co, &hit)) {
+  if (uv_find_nearest_edge(depsgraph, scene, ima, obedit, co, &hit)) {
     hit.dist_sq = dist_sq_init;
     hit.l = NULL;
     hit.luv = hit.luv_next = NULL;
 
-    BMIter iter;
+    /* (fclem) Is it garanteed that obedit_eval will be updated? */
+    Object *obedit_eval = DEG_get_evaluated_object(depsgraph, obedit);
+    BMEditMesh *em_eval = BKE_editmesh_from_object(obedit_eval);
     BMFace *efa;
 
-    BM_ITER_MESH (efa, &iter, em->bm, BM_FACES_OF_MESH) {
-      if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
-        continue;
+    if (!em_eval->mesh_eval_cage || em_eval->mesh_eval_cage->runtime.is_original) {
+      BMIter iter;
+
+      BM_ITER_MESH (efa, &iter, em->bm, BM_FACES_OF_MESH) {
+        if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+          continue;
+        }
+
+        float cent[2];
+        uv_poly_center(efa, cent, cd_loop_uv_offset);
+
+        const float dist_test_sq = len_squared_v2v2(co, cent);
+
+        if (dist_test_sq < hit.dist_sq) {
+          hit.efa = efa;
+          hit.dist_sq = dist_test_sq;
+          found = true;
+        }
       }
+    }
+    else {
+      Mesh *me = em_eval->mesh_eval_cage;
 
-      float cent[2];
-      uv_poly_center(efa, cent, cd_loop_uv_offset);
+      BM_mesh_elem_table_ensure(em->bm, BM_FACE);
 
-      const float dist_test_sq = len_squared_v2v2(co, cent);
+      int *p_origindex = CustomData_get_layer(&me->pdata, CD_ORIGINDEX);
 
-      if (dist_test_sq < hit.dist_sq) {
-        hit.efa = efa;
-        hit.dist_sq = dist_test_sq;
-        found = true;
+      MLoopUV *mluv = CustomData_get_layer(&me->ldata, CD_MLOOPUV);
+
+      if (modifiers_usesSubsurfFacedots(scene, obedit)) {
+        const MPoly *mpoly = me->mpoly;
+        for (int p = 0; p < me->totpoly; p++, mpoly++) {
+          if (p_origindex[p] == ORIGINDEX_NONE) {
+            continue;
+          }
+          efa = BM_face_at_index(em->bm, p_origindex[p]);
+          if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+            continue;
+          }
+          for (int i = 0; i < mpoly->totloop; i++) {
+            MLoop *mloop = &me->mloop[mpoly->loopstart + i];
+            MVert *mvert = &me->mvert[mloop->v];
+            if ((mvert->flag & ME_VERT_FACEDOT) == 0) {
+              continue;
+            }
+
+            const float dist_test_sq = len_squared_v2v2(co, mluv[mpoly->loopstart + i].uv);
+
+            if (dist_test_sq < hit.dist_sq) {
+              hit.efa = efa;
+              hit.dist_sq = dist_test_sq;
+              found = true;
+            }
+          }
+        }
+      }
+      else {
+        const MPoly *mpoly = me->mpoly;
+        for (int p = 0; p < me->totpoly; p++, mpoly++) {
+          if (p_origindex[p] == ORIGINDEX_NONE) {
+            continue;
+          }
+          efa = BM_face_at_index(em->bm, p_origindex[p]);
+          if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+            continue;
+          }
+
+          float cent[2];
+          uv_mpoly_center(mpoly, mluv, cent);
+
+          const float dist_test_sq = len_squared_v2v2(co, cent);
+
+          if (dist_test_sq < hit.dist_sq) {
+            hit.efa = efa;
+            hit.dist_sq = dist_test_sq;
+            found = true;
+          }
+        }
       }
     }
   }
@@ -854,7 +991,8 @@ bool uv_find_nearest_face(
   return found;
 }
 
-bool uv_find_nearest_face_multi(Scene *scene,
+bool uv_find_nearest_face_multi(Depsgraph *depsgraph,
+                                Scene *scene,
                                 Image *ima,
                                 Object **objects,
                                 const uint objects_len,
@@ -864,7 +1002,7 @@ bool uv_find_nearest_face_multi(Scene *scene,
   bool found = false;
   for (uint ob_index = 0; ob_index < objects_len; ob_index++) {
     Object *obedit = objects[ob_index];
-    if (uv_find_nearest_face(scene, ima, obedit, co, hit_final)) {
+    if (uv_find_nearest_face(depsgraph, scene, ima, obedit, co, hit_final)) {
       hit_final->ob = obedit;
       found = true;
     }
@@ -882,7 +1020,8 @@ static bool uv_nearest_between(const BMLoop *l, const float co[2], const int cd_
           (line_point_side_v2(uv_next, uv_curr, co) <= 0.0f));
 }
 
-bool uv_find_nearest_vert(Scene *scene,
+bool uv_find_nearest_vert(Depsgraph *depsgraph,
+                          Scene *scene,
                           Image *ima,
                           Object *obedit,
                           float const co[2],
@@ -894,7 +1033,7 @@ bool uv_find_nearest_vert(Scene *scene,
   /* this will fill in hit.vert1 and hit.vert2 */
   float dist_sq_init = hit_final->dist_sq;
   UvNearestHit hit = *hit_final;
-  if (uv_find_nearest_edge(scene, ima, obedit, co, &hit)) {
+  if (uv_find_nearest_edge(depsgraph, scene, ima, obedit, co, &hit)) {
     hit.dist_sq = dist_sq_init;
 
     hit.l = NULL;
@@ -902,46 +1041,112 @@ bool uv_find_nearest_vert(Scene *scene,
 
     BMEditMesh *em = BKE_editmesh_from_object(obedit);
     BMFace *efa;
-    BMIter iter;
 
-    BM_mesh_elem_index_ensure(em->bm, BM_VERT);
+    /* (fclem) Is it garanteed that obedit_eval will be updated? */
+    Object *obedit_eval = DEG_get_evaluated_object(depsgraph, obedit);
+    BMEditMesh *em_eval = BKE_editmesh_from_object(obedit_eval);
 
     const int cd_loop_uv_offset = CustomData_get_offset(&em->bm->ldata, CD_MLOOPUV);
 
-    BM_ITER_MESH (efa, &iter, em->bm, BM_FACES_OF_MESH) {
-      if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
-        continue;
-      }
+    if (!em_eval->mesh_eval_cage || em_eval->mesh_eval_cage->runtime.is_original) {
+      BMIter iter;
 
-      BMIter liter;
-      BMLoop *l;
-      int i;
-      BM_ITER_ELEM_INDEX (l, &liter, efa, BM_LOOPS_OF_FACE, i) {
-        float dist_test_sq;
-        MLoopUV *luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset);
-        if (penalty_dist != 0.0f && uvedit_uv_select_test(scene, l, cd_loop_uv_offset)) {
-          dist_test_sq = len_v2v2(co, luv->uv) + penalty_dist;
-          dist_test_sq = SQUARE(dist_test_sq);
-        }
-        else {
-          dist_test_sq = len_squared_v2v2(co, luv->uv);
+      BM_mesh_elem_index_ensure(em->bm, BM_VERT);
+
+      BM_ITER_MESH (efa, &iter, em->bm, BM_FACES_OF_MESH) {
+        if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+          continue;
         }
 
-        if (dist_test_sq <= hit.dist_sq) {
-          if (dist_test_sq == hit.dist_sq) {
-            if (!uv_nearest_between(l, co, cd_loop_uv_offset)) {
-              continue;
-            }
+        BMIter liter;
+        BMLoop *l;
+        int i;
+        BM_ITER_ELEM_INDEX (l, &liter, efa, BM_LOOPS_OF_FACE, i) {
+          float dist_test_sq;
+          MLoopUV *luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset);
+          if (penalty_dist != 0.0f && uvedit_uv_select_test(scene, l, cd_loop_uv_offset)) {
+            dist_test_sq = len_v2v2(co, luv->uv) + penalty_dist;
+            dist_test_sq = SQUARE(dist_test_sq);
+          }
+          else {
+            dist_test_sq = len_squared_v2v2(co, luv->uv);
           }
 
-          hit.dist_sq = dist_test_sq;
+          if (dist_test_sq <= hit.dist_sq) {
+            if (dist_test_sq == hit.dist_sq) {
+              if (!uv_nearest_between(l, co, cd_loop_uv_offset)) {
+                continue;
+              }
+            }
 
-          hit.l = l;
-          hit.luv = luv;
-          hit.luv_next = BM_ELEM_CD_GET_VOID_P(l->next, cd_loop_uv_offset);
-          hit.efa = efa;
-          hit.lindex = i;
-          found = true;
+            hit.dist_sq = dist_test_sq;
+
+            hit.l = l;
+            hit.luv = luv;
+            hit.luv_next = BM_ELEM_CD_GET_VOID_P(l->next, cd_loop_uv_offset);
+            hit.efa = efa;
+            hit.lindex = i;
+            found = true;
+          }
+        }
+      }
+    }
+    else {
+      Mesh *me = em_eval->mesh_eval_cage;
+
+      BM_mesh_elem_table_ensure(em->bm, BM_FACE | BM_EDGE);
+
+      int *p_origindex = CustomData_get_layer(&me->pdata, CD_ORIGINDEX);
+      int *e_origindex = CustomData_get_layer(&me->edata, CD_ORIGINDEX);
+
+      MLoopUV *mluv = CustomData_get_layer(&me->ldata, CD_MLOOPUV);
+
+      MPoly *mpoly = me->mpoly;
+      for (int p = 0; p < me->totpoly; p++, mpoly++) {
+        if (p_origindex[p] == ORIGINDEX_NONE) {
+          continue;
+        }
+        efa = BM_face_at_index(em->bm, p_origindex[p]);
+        if (!uvedit_face_visible_test(scene, obedit, ima, efa)) {
+          continue;
+        }
+        for (int i = 0; i < mpoly->totloop; i++) {
+          MLoop *mloop = &me->mloop[mpoly->loopstart + i];
+          if (e_origindex[mloop->e] == ORIGINDEX_NONE) {
+            continue;
+          }
+          MLoopUV *luv = &mluv[mpoly->loopstart + i];
+          MLoopUV *luv_next = &mluv[mpoly->loopstart + ((i + 1) % mpoly->totloop)];
+
+          BMEdge *eed = BM_edge_at_index(em->bm, e_origindex[mloop->e]);
+          BMLoop *l = BM_face_edge_share_loop(efa, eed);
+
+          float dist_test_sq;
+          if (penalty_dist != 0.0f && uvedit_uv_select_test(scene, l, cd_loop_uv_offset)) {
+            dist_test_sq = len_v2v2(co, luv->uv) + penalty_dist;
+            dist_test_sq = SQUARE(dist_test_sq);
+          }
+          else {
+            dist_test_sq = len_squared_v2v2(co, luv->uv);
+          }
+
+          if (dist_test_sq <= hit.dist_sq) {
+            /* FIXME this is testing against original coord, not deformed cage. */
+            if (dist_test_sq == hit.dist_sq) {
+              if (!uv_nearest_between(l, co, cd_loop_uv_offset)) {
+                continue;
+              }
+            }
+
+            hit.dist_sq = dist_test_sq;
+
+            hit.l = l;
+            hit.luv = luv;
+            hit.luv_next = luv_next;
+            hit.efa = efa;
+            hit.lindex = i;
+            found = true;
+          }
         }
       }
     }
@@ -954,7 +1159,8 @@ bool uv_find_nearest_vert(Scene *scene,
   return found;
 }
 
-bool uv_find_nearest_vert_multi(Scene *scene,
+bool uv_find_nearest_vert_multi(Depsgraph *depsgraph,
+                                Scene *scene,
                                 Image *ima,
                                 Object **objects,
                                 const uint objects_len,
@@ -965,7 +1171,7 @@ bool uv_find_nearest_vert_multi(Scene *scene,
   bool found = false;
   for (uint ob_index = 0; ob_index < objects_len; ob_index++) {
     Object *obedit = objects[ob_index];
-    if (uv_find_nearest_vert(scene, ima, obedit, co, penalty_dist, hit_final)) {
+    if (uv_find_nearest_vert(depsgraph, scene, ima, obedit, co, penalty_dist, hit_final)) {
       hit_final->ob = obedit;
       found = true;
     }
@@ -1274,9 +1480,9 @@ static void uv_select_linked_multi(Scene *scene,
 
     BM_mesh_elem_table_ensure(em->bm, BM_FACE); /* we can use this too */
 
-    /* Note, we had 'use winding' so we don't consider overlapping islands as connected, see T44320
-     * this made *every* projection split the island into front/back islands.
-     * Keep 'use_winding' to false, see: T50970.
+    /* Note, we had 'use winding' so we don't consider overlapping islands as connected, see
+     * T44320 this made *every* projection split the island into front/back islands. Keep
+     * 'use_winding' to false, see: T50970.
      *
      * Better solve this by having a delimit option for select-linked operator,
      * keeping island-select working as is. */
@@ -2513,12 +2719,12 @@ static int uv_mouse_select_multi(bContext *C,
   /* find nearest element */
   if (loop) {
     /* find edge */
-    found_item = uv_find_nearest_edge_multi(scene, ima, objects, objects_len, co, &hit);
+    found_item = uv_find_nearest_edge_multi(depsgraph, scene, ima, objects, objects_len, co, &hit);
   }
   else if (selectmode == UV_SELECT_VERTEX) {
     /* find vertex */
     found_item = uv_find_nearest_vert_multi(
-        scene, ima, objects, objects_len, co, penalty_dist, &hit);
+        depsgraph, scene, ima, objects, objects_len, co, penalty_dist, &hit);
     found_item = found_item && (!deselect_all || hit.dist_sq < penalty_dist);
 
     if (found_item) {
@@ -2535,7 +2741,7 @@ static int uv_mouse_select_multi(bContext *C,
   }
   else if (selectmode == UV_SELECT_EDGE) {
     /* find edge */
-    found_item = uv_find_nearest_edge_multi(scene, ima, objects, objects_len, co, &hit);
+    found_item = uv_find_nearest_edge_multi(depsgraph, scene, ima, objects, objects_len, co, &hit);
     found_item = found_item && (!deselect_all || hit.dist_sq < penalty_dist);
 
     if (found_item) {
@@ -2554,7 +2760,7 @@ static int uv_mouse_select_multi(bContext *C,
   }
   else if (selectmode == UV_SELECT_FACE) {
     /* find face */
-    found_item = uv_find_nearest_face_multi(scene, ima, objects, objects_len, co, &hit);
+    found_item = uv_find_nearest_face_multi(depsgraph, scene, ima, objects, objects_len, co, &hit);
     found_item = found_item && (!deselect_all || hit.dist_sq < penalty_dist);
 
     if (found_item) {
@@ -2578,7 +2784,7 @@ static int uv_mouse_select_multi(bContext *C,
     }
   }
   else if (selectmode == UV_SELECT_ISLAND) {
-    found_item = uv_find_nearest_edge_multi(scene, ima, objects, objects_len, co, &hit);
+    found_item = uv_find_nearest_edge_multi(depsgraph, scene, ima, objects, objects_len, co, &hit);
     found_item = found_item && (!deselect_all || hit.dist_sq < penalty_dist);
   }
 
@@ -2892,6 +3098,7 @@ static void UV_OT_select_loop(wmOperatorType *ot)
 
 static int uv_select_linked_internal(bContext *C, wmOperator *op, const wmEvent *event, bool pick)
 {
+  Depsgraph *depsgraph = CTX_data_depsgraph(C);
   SpaceImage *sima = CTX_wm_space_image(C);
   Scene *scene = CTX_data_scene(C);
   ToolSettings *ts = scene->toolsettings;
@@ -2936,7 +3143,7 @@ static int uv_select_linked_internal(bContext *C, wmOperator *op, const wmEvent
       RNA_float_get_array(op->ptr, "location", co);
     }
 
-    if (!uv_find_nearest_edge_multi(scene, ima, objects, objects_len, co, &hit)) {
+    if (!uv_find_nearest_edge_multi(depsgraph, scene, ima, objects, objects_len, co, &hit)) {
       MEM_freeN(objects);
       return OPERATOR_CANCELLED;
     }

source/blender/editors/uvedit/uvedit_smart_stitch.c

@@ -2550,11 +2550,13 @@ static StitchState *stitch_select(bContext *C,
   UvNearestHit hit = UV_NEAREST_HIT_INIT;
   ARegion *ar = CTX_wm_region(C);
   Image *ima = CTX_data_edit_image(C);
+  Depsgraph *depsgraph = CTX_data_depsgraph(C);
 
   UI_view2d_region_to_view(&ar->v2d, event->mval[0], event->mval[1], &co[0], &co[1]);
 
   if (ssc->mode == STITCH_VERT) {
-    if (uv_find_nearest_vert_multi(scene, ima, ssc->objects, ssc->objects_len, co, 0.0f, &hit)) {
+    if (uv_find_nearest_vert_multi(
+            depsgraph, scene, ima, ssc->objects, ssc->objects_len, co, 0.0f, &hit)) {
       /* Add vertex to selection, deselect all common uv's of vert other than selected and
        * update the preview. This behavior was decided so that you can do stuff like deselect
        * the opposite stitchable vertex and the initial still gets deselected */
@@ -2575,7 +2577,8 @@ static StitchState *stitch_select(bContext *C,
       return state;
     }
   }
-  else if (uv_find_nearest_edge_multi(scene, ima, ssc->objects, ssc->objects_len, co, &hit)) {
+  else if (uv_find_nearest_edge_multi(
+               depsgraph, scene, ima, ssc->objects, ssc->objects_len, co, &hit)) {
     /* find StitchState from hit->ob */
     StitchState *state = NULL;
     for (uint ob_index = 0; ob_index < ssc->objects_len; ob_index++) {

source/blender/gpu/GPU_batch.h

@@ -40,7 +40,7 @@ typedef enum {
   GPU_BATCH_READY_TO_DRAW,
 } GPUBatchPhase;
 
-#define GPU_BATCH_VBO_MAX_LEN 4
+#define GPU_BATCH_VBO_MAX_LEN 5
 #define GPU_BATCH_VAO_STATIC_LEN 3
 #define GPU_BATCH_VAO_DYN_ALLOC_COUNT 16
 
@@ -114,6 +114,7 @@ void GPU_batch_vao_cache_clear(GPUBatch *);
 void GPU_batch_callback_free_set(GPUBatch *, void (*callback)(GPUBatch *, void *), void *);
 
 void GPU_batch_instbuf_set(GPUBatch *, GPUVertBuf *, bool own_vbo); /* Instancing */
+void GPU_batch_elembuf_set(GPUBatch *batch, GPUIndexBuf *elem, bool own_ibo);
 
 int GPU_batch_vertbuf_add_ex(GPUBatch *, GPUVertBuf *, bool own_vbo);
 

source/blender/gpu/GPU_element.h

@@ -36,14 +36,19 @@ typedef enum {
 } GPUIndexBufType;
 
 typedef struct GPUIndexBuf {
+  uint index_start;
   uint index_len;
+  bool is_subrange;
 #if GPU_TRACK_INDEX_RANGE
   GPUIndexBufType index_type;
   uint32_t gl_index_type;
   uint base_index;
 #endif
   uint32_t ibo_id; /* 0 indicates not yet sent to VRAM */
-  void *data;      /* non-NULL indicates not yet sent to VRAM */
+  union {
+    void *data;              /* non-NULL indicates not yet sent to VRAM */
+    struct GPUIndexBuf *src; /* if is_subrange is true, this is the source buffer. */
+  };
 } GPUIndexBuf;
 
 void GPU_indexbuf_use(GPUIndexBuf *);
@@ -71,9 +76,21 @@ void GPU_indexbuf_add_line_verts(GPUIndexBufBuilder *, uint v1, uint v2);
 void GPU_indexbuf_add_tri_verts(GPUIndexBufBuilder *, uint v1, uint v2, uint v3);
 void GPU_indexbuf_add_line_adj_verts(GPUIndexBufBuilder *, uint v1, uint v2, uint v3, uint v4);
 
+void GPU_indexbuf_set_point_vert(GPUIndexBufBuilder *builder, uint elem, uint v1);
+void GPU_indexbuf_set_line_verts(GPUIndexBufBuilder *builder, uint elem, uint v1, uint v2);
+void GPU_indexbuf_set_tri_verts(GPUIndexBufBuilder *builder, uint elem, uint v1, uint v2, uint v3);
+
+/* Skip primitive rendering at the given index. */
+void GPU_indexbuf_set_point_restart(GPUIndexBufBuilder *builder, uint elem);
+void GPU_indexbuf_set_line_restart(GPUIndexBufBuilder *builder, uint elem);
+void GPU_indexbuf_set_tri_restart(GPUIndexBufBuilder *builder, uint elem);
+
 GPUIndexBuf *GPU_indexbuf_build(GPUIndexBufBuilder *);
 void GPU_indexbuf_build_in_place(GPUIndexBufBuilder *, GPUIndexBuf *);
 
+/* Create a subrange of an existing indexbuffer. */
+GPUIndexBuf *GPU_indexbuf_create_subrange(GPUIndexBuf *ibo, uint start, uint length);
+
 void GPU_indexbuf_discard(GPUIndexBuf *);
 
 int GPU_indexbuf_primitive_len(GPUPrimType prim_type);

source/blender/gpu/GPU_vertex_format.h

@@ -31,7 +31,7 @@
 #include "BLI_assert.h"
 
 #define GPU_VERT_ATTR_MAX_LEN 16
-#define GPU_VERT_ATTR_MAX_NAMES 5
+#define GPU_VERT_ATTR_MAX_NAMES 6
 #define GPU_VERT_ATTR_NAME_AVERAGE_LEN 11
 #define GPU_VERT_ATTR_NAMES_BUF_LEN ((GPU_VERT_ATTR_NAME_AVERAGE_LEN + 1) * GPU_VERT_ATTR_MAX_LEN)
 
@@ -88,6 +88,8 @@ typedef struct GPUVertFormat {
   uint packed : 1;
   /** Current offset in names[]. */
   uint name_offset : 8;
+  /** Store each attrib in one contiguous buffer region. */
+  uint deinterleaved : 1;
 
   GPUVertAttr attrs[GPU_VERT_ATTR_MAX_LEN];
   char names[GPU_VERT_ATTR_NAMES_BUF_LEN];
@@ -104,6 +106,8 @@ uint GPU_vertformat_attr_add(
     GPUVertFormat *, const char *name, GPUVertCompType, uint comp_len, GPUVertFetchMode);
 void GPU_vertformat_alias_add(GPUVertFormat *, const char *alias);
 
+void GPU_vertformat_deinterleave(GPUVertFormat *format);
+
 int GPU_vertformat_attr_id_get(const GPUVertFormat *, const char *name);
 
 BLI_INLINE const char *GPU_vertformat_attr_name_get(const GPUVertFormat *format,
@@ -122,7 +126,59 @@ typedef struct GPUPackedNormal {
   int w : 2; /* 0 by default, can manually set to { -2, -1, 0, 1 } */
 } GPUPackedNormal;
 
-GPUPackedNormal GPU_normal_convert_i10_v3(const float data[3]);
-GPUPackedNormal GPU_normal_convert_i10_s3(const short data[3]);
+/* OpenGL ES packs in a different order as desktop GL but component conversion is the same.
+ * Of the code here, only struct GPUPackedNormal needs to change. */
+
+#define SIGNED_INT_10_MAX 511
+#define SIGNED_INT_10_MIN -512
+
+BLI_INLINE int clampi(int x, int min_allowed, int max_allowed)
+{
+#if TRUST_NO_ONE
+  assert(min_allowed <= max_allowed);
+#endif
+  if (x < min_allowed) {
+    return min_allowed;
+  }
+  else if (x > max_allowed) {
+    return max_allowed;
+  }
+  else {
+    return x;
+  }
+}
+
+BLI_INLINE int quantize(float x)
+{
+  int qx = x * 511.0f;
+  return clampi(qx, SIGNED_INT_10_MIN, SIGNED_INT_10_MAX);
+}
+
+BLI_INLINE int convert_i16(short x)
+{
+  /* 16-bit signed --> 10-bit signed */
+  /* TODO: round? */
+  return x >> 6;
+}
+
+BLI_INLINE GPUPackedNormal GPU_normal_convert_i10_v3(const float data[3])
+{
+  GPUPackedNormal n = {
+      .x = quantize(data[0]),
+      .y = quantize(data[1]),
+      .z = quantize(data[2]),
+  };
+  return n;
+}
+
+BLI_INLINE GPUPackedNormal GPU_normal_convert_i10_s3(const short data[3])
+{
+  GPUPackedNormal n = {
+      .x = convert_i16(data[0]),
+      .y = convert_i16(data[1]),
+      .z = convert_i16(data[2]),
+  };
+  return n;
+}
 
 #endif /* __GPU_VERTEX_FORMAT_H__ */

source/blender/gpu/intern/gpu_batch.c

@@ -181,6 +181,25 @@ void GPU_batch_instbuf_set(GPUBatch *batch, GPUVertBuf *inst, bool own_vbo)
   }
 }
 
+void GPU_batch_elembuf_set(GPUBatch *batch, GPUIndexBuf *elem, bool own_ibo)
+{
+  BLI_assert(elem != NULL);
+  /* redo the bindings */
+  GPU_batch_vao_cache_clear(batch);
+
+  if (batch->elem != NULL && (batch->owns_flag & GPU_BATCH_OWNS_INDEX)) {
+    GPU_indexbuf_discard(batch->elem);
+  }
+  batch->elem = elem;
+
+  if (own_ibo) {
+    batch->owns_flag |= GPU_BATCH_OWNS_INDEX;
+  }
+  else {
+    batch->owns_flag &= ~GPU_BATCH_OWNS_INDEX;
+  }
+}
+
 /* Returns the index of verts in the batch. */
 int GPU_batch_vertbuf_add_ex(GPUBatch *batch, GPUVertBuf *verts, bool own_vbo)
 {
@@ -361,13 +380,23 @@ static void create_bindings(GPUVertBuf *verts,
   const GPUVertFormat *format = &verts->format;
 
   const uint attr_len = format->attr_len;
-  const uint stride = format->stride;
+  uint stride = format->stride;
+  uint offset = 0;
 
   GPU_vertbuf_use(verts);
 
   for (uint a_idx = 0; a_idx < attr_len; ++a_idx) {
     const GPUVertAttr *a = &format->attrs[a_idx];
-    const GLvoid *pointer = (const GLubyte *)0 + a->offset + v_first * stride;
+
+    if (format->deinterleaved) {
+      offset += ((a_idx == 0) ? 0 : format->attrs[a_idx - 1].sz) * verts->vertex_len;
+      stride = a->sz;
+    }
+    else {
+      offset = a->offset;
+    }
+
+    const GLvoid *pointer = (const GLubyte *)0 + offset + v_first * stride;
 
     for (uint n_idx = 0; n_idx < a->name_len; ++n_idx) {
       const char *name = GPU_vertformat_attr_name_get(format, a, n_idx);
@@ -418,8 +447,11 @@ static void create_bindings(GPUVertBuf *verts,
 
 static void batch_update_program_bindings(GPUBatch *batch, uint v_first)
 {
-  for (int v = 0; v < GPU_BATCH_VBO_MAX_LEN && batch->verts[v] != NULL; ++v) {
-    create_bindings(batch->verts[v], batch->interface, (batch->inst) ? 0 : v_first, false);
+  /* Reverse order so first vbos have more prevalence (in term of attrib override). */
+  for (int v = GPU_BATCH_VBO_MAX_LEN - 1; v > -1; --v) {
+    if (batch->verts[v] != NULL) {
+      create_bindings(batch->verts[v], batch->interface, (batch->inst) ? 0 : v_first, false);
+    }
   }
   if (batch->inst) {
     create_bindings(batch->inst, batch->interface, v_first, true);
@@ -549,10 +581,10 @@ static void *elem_offset(const GPUIndexBuf *el, int v_first)
 {
 #if GPU_TRACK_INDEX_RANGE
   if (el->index_type == GPU_INDEX_U16) {
-    return (GLushort *)0 + v_first;
+    return (GLushort *)0 + v_first + el->index_start;
   }
 #endif
-  return (GLuint *)0 + v_first;
+  return (GLuint *)0 + v_first + el->index_start;
 }
 
 /* Use when drawing with GPU_batch_draw_advanced */

source/blender/gpu/intern/gpu_element.c

@@ -162,6 +162,100 @@ void GPU_indexbuf_add_line_adj_verts(
   GPU_indexbuf_add_generic_vert(builder, v4);
 }
 
+void GPU_indexbuf_set_point_vert(GPUIndexBufBuilder *builder, uint elem, uint v1)
+{
+  BLI_assert(builder->prim_type == GPU_PRIM_POINTS);
+  BLI_assert(elem < builder->max_index_len);
+  builder->data[elem++] = v1;
+  if (builder->index_len < elem) {
+    builder->index_len = elem;
+  }
+}
+
+void GPU_indexbuf_set_line_verts(GPUIndexBufBuilder *builder, uint elem, uint v1, uint v2)
+{
+  BLI_assert(builder->prim_type == GPU_PRIM_LINES);
+  BLI_assert(v1 != v2);
+  BLI_assert(v1 <= builder->max_allowed_index);
+  BLI_assert(v2 <= builder->max_allowed_index);
+  BLI_assert((elem + 1) * 2 <= builder->max_index_len);
+  uint idx = elem * 2;
+  builder->data[idx++] = v1;
+  builder->data[idx++] = v2;
+  if (builder->index_len < idx) {
+    builder->index_len = idx;
+  }
+}
+
+void GPU_indexbuf_set_tri_verts(GPUIndexBufBuilder *builder, uint elem, uint v1, uint v2, uint v3)
+{
+  BLI_assert(builder->prim_type == GPU_PRIM_TRIS);
+  BLI_assert(v1 != v2 && v2 != v3 && v3 != v1);
+  BLI_assert(v1 <= builder->max_allowed_index);
+  BLI_assert(v2 <= builder->max_allowed_index);
+  BLI_assert(v3 <= builder->max_allowed_index);
+  BLI_assert((elem + 1) * 3 <= builder->max_index_len);
+  uint idx = elem * 3;
+  builder->data[idx++] = v1;
+  builder->data[idx++] = v2;
+  builder->data[idx++] = v3;
+  if (builder->index_len < idx) {
+    builder->index_len = idx;
+  }
+}
+
+void GPU_indexbuf_set_point_restart(GPUIndexBufBuilder *builder, uint elem)
+{
+  BLI_assert(builder->prim_type == GPU_PRIM_POINTS);
+  BLI_assert(elem < builder->max_index_len);
+  builder->data[elem++] = RESTART_INDEX;
+  if (builder->index_len < elem) {
+    builder->index_len = elem;
+  }
+}
+
+void GPU_indexbuf_set_line_restart(GPUIndexBufBuilder *builder, uint elem)
+{
+  BLI_assert(builder->prim_type == GPU_PRIM_LINES);
+  BLI_assert((elem + 1) * 2 <= builder->max_index_len);
+  uint idx = elem * 2;
+  builder->data[idx++] = RESTART_INDEX;
+  builder->data[idx++] = RESTART_INDEX;
+  if (builder->index_len < idx) {
+    builder->index_len = idx;
+  }
+}
+
+void GPU_indexbuf_set_tri_restart(GPUIndexBufBuilder *builder, uint elem)
+{
+  BLI_assert(builder->prim_type == GPU_PRIM_TRIS);
+  BLI_assert((elem + 1) * 3 <= builder->max_index_len);
+  uint idx = elem * 3;
+  builder->data[idx++] = RESTART_INDEX;
+  builder->data[idx++] = RESTART_INDEX;
+  builder->data[idx++] = RESTART_INDEX;
+  if (builder->index_len < idx) {
+    builder->index_len = idx;
+  }
+}
+
+GPUIndexBuf *GPU_indexbuf_create_subrange(GPUIndexBuf *elem_src, uint start, uint length)
+{
+  GPUIndexBuf *elem = MEM_callocN(sizeof(GPUIndexBuf), "GPUIndexBuf");
+  BLI_assert(elem_src && !elem_src->is_subrange);
+  BLI_assert(start + length <= elem_src->index_len);
+#if GPU_TRACK_INDEX_RANGE
+  elem->index_type = elem_src->index_type;
+  elem->gl_index_type = elem_src->gl_index_type;
+  elem->base_index = elem_src->base_index;
+#endif
+  elem->is_subrange = true;
+  elem->src = elem_src;
+  elem->index_start = start;
+  elem->index_len = length;
+  return elem;
+}
+
 #if GPU_TRACK_INDEX_RANGE
 /* Everything remains 32 bit while building to keep things simple.
  * Find min/max after, then convert to smallest index type possible. */
@@ -271,6 +365,10 @@ static void indexbuf_upload_data(GPUIndexBuf *elem)
 
 void GPU_indexbuf_use(GPUIndexBuf *elem)
 {
+  if (elem->is_subrange) {
+    GPU_indexbuf_use(elem->src);
+    return;
+  }
   if (elem->ibo_id == 0) {
     elem->ibo_id = GPU_buf_alloc();
   }
@@ -285,7 +383,7 @@ void GPU_indexbuf_discard(GPUIndexBuf *elem)
   if (elem->ibo_id) {
     GPU_buf_free(elem->ibo_id);
   }
-  if (elem->data) {
+  if (!elem->is_subrange && elem->data) {
     MEM_freeN(elem->data);
   }
   MEM_freeN(elem);

source/blender/gpu/intern/gpu_vertex_format.c

@@ -216,6 +216,29 @@ int GPU_vertformat_attr_id_get(const GPUVertFormat *format, const char *name)
   return -1;
 }
 
+/* Make attribute layout non-interleaved.
+ * Warning! This does not change data layout!
+ * Use direct buffer access to fill the data.
+ * This is for advanced usage.
+ *
+ * Deinterleaved data means all attrib data for each attrib
+ * is stored continuously like this :
+ * 000011112222
+ * instead of :
+ * 012012012012
+ *
+ * Note this is per attrib deinterleaving, NOT per component.
+ *  */
+void GPU_vertformat_deinterleave(GPUVertFormat *format)
+{
+  /* Ideally we should change the stride and offset here. This would allow
+   * us to use GPU_vertbuf_attr_set / GPU_vertbuf_attr_fill. But since
+   * we use only 11 bits for attr->offset this limits the size of the
+   * buffer considerably. So instead we do the conversion when creating
+   * bindings in create_bindings(). */
+  format->deinterleaved = true;
+}
+
 uint padding(uint offset, uint alignment)
 {
   const uint mod = offset % alignment;
@@ -389,58 +412,3 @@ void GPU_vertformat_from_interface(GPUVertFormat *format, const GPUShaderInterfa
     }
   }
 }
-
-/* OpenGL ES packs in a different order as desktop GL but component conversion is the same.
- * Of the code here, only struct GPUPackedNormal needs to change. */
-
-#define SIGNED_INT_10_MAX 511
-#define SIGNED_INT_10_MIN -512
-
-static int clampi(int x, int min_allowed, int max_allowed)
-{
-#if TRUST_NO_ONE
-  assert(min_allowed <= max_allowed);
-#endif
-  if (x < min_allowed) {
-    return min_allowed;
-  }
-  else if (x > max_allowed) {
-    return max_allowed;
-  }
-  else {
-    return x;
-  }
-}
-
-static int quantize(float x)
-{
-  int qx = x * 511.0f;
-  return clampi(qx, SIGNED_INT_10_MIN, SIGNED_INT_10_MAX);
-}
-
-static int convert_i16(short x)
-{
-  /* 16-bit signed --> 10-bit signed */
-  /* TODO: round? */
-  return x >> 6;
-}
-
-GPUPackedNormal GPU_normal_convert_i10_v3(const float data[3])
-{
-  GPUPackedNormal n = {
-      .x = quantize(data[0]),
-      .y = quantize(data[1]),
-      .z = quantize(data[2]),
-  };
-  return n;
-}
-
-GPUPackedNormal GPU_normal_convert_i10_s3(const short data[3])
-{
-  GPUPackedNormal n = {
-      .x = convert_i16(data[0]),
-      .y = convert_i16(data[1]),
-      .z = convert_i16(data[2]),
-  };
-  return n;
-}

source/blender/gpu/intern/gpu_vertex_format_private.h

@@ -27,6 +27,7 @@
 #define __GPU_VERTEX_FORMAT_PRIVATE_H__
 
 void VertexFormat_pack(GPUVertFormat *format);
+void VertexFormat_deinterleave(GPUVertFormat *format, uint vertex_len);
 uint padding(uint offset, uint alignment);
 uint vertex_buffer_size(const GPUVertFormat *format, uint vertex_len);
 

source/blender/gpu/shaders/gpu_shader_2D_edituvs_stretch_vert.glsl

@@ -8,7 +8,7 @@ in vec2 pos;
 in float stretch;
 #else
 
-in vec4 uv_adj;
+in vec2 uv_angles;
 in float angle;
 #endif
 
@@ -52,6 +52,11 @@ vec3 weight_to_rgb(float weight)
 
 #define M_PI 3.1415926535897932
 
+vec2 angle_to_v2(float angle)
+{
+  return vec2(cos(angle), sin(angle));
+}
+
 /* Adapted from BLI_math_vector.h */
 float angle_normalized_v2v2(vec2 v1, vec2 v2)
 {
@@ -69,7 +74,9 @@ void main()
   gl_Position = ModelViewProjectionMatrix * vec4(pos, 0.0, 1.0);
 
 #ifdef STRETCH_ANGLE
-  float uv_angle = angle_normalized_v2v2(uv_adj.xy, uv_adj.zw) / M_PI;
+  vec2 v1 = angle_to_v2(uv_angles.x * M_PI);
+  vec2 v2 = angle_to_v2(uv_angles.y * M_PI);
+  float uv_angle = angle_normalized_v2v2(v1, v2) / M_PI;
   float stretch = 1.0 - abs(uv_angle - angle);
   stretch = stretch;
   stretch = 1.0 - stretch * stretch;