LineArt: Try using sort/parallel_sort but crasheslineart-object-load

5 files changed, 79 insertions, 34 deletions

diff --git a/source/blender/gpencil_modifiers/CMakeLists.txt b/source/blender/gpencil_modifiers/CMakeLists.txt
index 6108629183c..3b6f00b56d8 100644
--- a/source/blender/gpencil_modifiers/CMakeLists.txt
+++ b/source/blender/gpencil_modifiers/CMakeLists.txt
@@ -68,12 +68,29 @@ set(SRC
   intern/lineart/lineart_cpu.c
   intern/lineart/lineart_ops.c
   intern/lineart/lineart_util.c
+  intern/lineart/lineart_cpp_bridge.cpp
 
   intern/lineart/MOD_lineart.h
   intern/lineart/lineart_intern.h
 
 )
 
+if(WITH_TBB)
+add_definitions(-DWITH_TBB)
+if(WIN32)
+  # TBB includes Windows.h which will define min/max macros
+  # that will collide with the stl versions.
+  add_definitions(-DNOMINMAX)
+endif()
+list(APPEND INC_SYS
+  ${TBB_INCLUDE_DIRS}
+)
+
+list(APPEND LIB
+  ${TBB_LIBRARIES}
+)
+endif()
+
 set(LIB
 )
 
diff --git a/source/blender/gpencil_modifiers/intern/lineart/MOD_lineart.h b/source/blender/gpencil_modifiers/intern/lineart/MOD_lineart.h
index 830066ac0fb..a30ec4490cf 100644
--- a/source/blender/gpencil_modifiers/intern/lineart/MOD_lineart.h
+++ b/source/blender/gpencil_modifiers/intern/lineart/MOD_lineart.h
@@ -197,6 +197,12 @@ typedef struct LineartChainRegisterEntry {
   char is_left;
 } LineartChainRegisterEntry;
 
+typedef struct LineartAdjacentItem {
+  unsigned int v1;
+  unsigned int v2;
+  unsigned int e;
+} LineartAdjacentItem;
+
 enum eLineArtTileRecursiveLimit {
   /* If tile gets this small, it's already much smaller than a pixel. No need to continue
    * splitting. */
diff --git a/source/blender/gpencil_modifiers/intern/lineart/lineart_cpp_bridge.cpp b/source/blender/gpencil_modifiers/intern/lineart/lineart_cpp_bridge.cpp
new file mode 100644
index 00000000000..2a630e69665
--- /dev/null
+++ b/source/blender/gpencil_modifiers/intern/lineart/lineart_cpp_bridge.cpp
@@ -0,0 +1,28 @@
+#include "BLI_sort.hh"
+#include "BLI_vector.hh"
+#include "MOD_lineart.h"
+#include "lineart_intern.h"
+
+static int cmp_adjacent_items(const LineartAdjacentItem &p1, const LineartAdjacentItem &p2)
+{
+  int a = (int)p1.v1 - (int)p2.v1;
+  int b = (int)p1.v2 - (int)p2.v2;
+  return a ? a : b;
+}
+
+void lineart_sort_adjacent_items(LineartAdjacentItem *ai, int length)
+{
+  blender::Vector<LineartAdjacentItem> _ai;
+  _ai.reserve(length);
+  for (int i = 0; i < length; i++) {
+    _ai[i] = ai[i];
+    printf("(%d %d %d)", _ai[i].v1, _ai[i].v2, _ai[i].e);
+  }
+  _ai.resize(length);
+  printf("\n");
+  std::sort(ai, ai + length, cmp_adjacent_items);
+  for (int i = 0; i < length; i++) {
+    ai[i] = _ai[i];
+    printf("(%d %d %d)", _ai[i].v1, _ai[i].v2, _ai[i].e);
+  }
+}
diff --git a/source/blender/gpencil_modifiers/intern/lineart/lineart_cpu.c b/source/blender/gpencil_modifiers/intern/lineart/lineart_cpu.c
index 88ac80e153f..b6639f80f63 100644
--- a/source/blender/gpencil_modifiers/intern/lineart/lineart_cpu.c
+++ b/source/blender/gpencil_modifiers/intern/lineart/lineart_cpu.c
@@ -1452,12 +1452,6 @@ static void lineart_vert_transform_me(
   mul_v4_m4v3_db(vt->fbcoord, mvp_mat, co);
 }
 
-typedef struct LineartAdjacentItem {
-  unsigned int v1;
-  unsigned int v2;
-  unsigned int e;
-} LineartAdjacentItem;
-
 typedef struct LineartEdgeNeighbor {
   int e;
   short flags;
@@ -1517,7 +1511,6 @@ typedef struct EdgeFeatData {
   LineartTriangle *tri_array;
   LineartVert *v_array;
   float crease_threshold;
-  float **poly_normals;
   bool use_auto_smooth;
   bool use_freestyle_face;
   int freestyle_face_index;
@@ -1537,8 +1530,9 @@ static void feat_data_sum_reduce(const void *__restrict UNUSED(userdata),
   feat_chunk_join->feat_edges += feat_chunk->feat_edges;
 }
 
-__attribute__((optimize("O0"))) static void lineart_identify_mlooptri_feature_edges(
-    void *__restrict userdata, const int i, const TaskParallelTLS *__restrict tls)
+static void lineart_identify_mlooptri_feature_edges(void *__restrict userdata,
+                                                    const int i,
+                                                    const TaskParallelTLS *__restrict tls)
 {
   EdgeFeatData *e_feat_data = (EdgeFeatData *)userdata;
   EdgeFeatReduceData *reduce_data = (EdgeFeatReduceData *)tls->userdata_chunk;
@@ -1698,18 +1692,16 @@ static uint16_t lineart_identify_medge_feature_edges(
   return edge_flag_result;
 }
 
-__attribute__((optimize("O0"))) static uint16_t lineart_identify_feature_line_me(
-    LineartRenderBuffer *rb,
-    int eindex,
-    LineartTriangle *rt_array,
-    LineartVert *rv_array,
-    float crease_threshold,
-    bool use_auto_smooth,
-    bool use_freestyle_edge,
-    bool use_freestyle_face,
-    Mesh *me,
-    LineartEdgeNeighbor *en,
-    float (*normals)[3])
+static uint16_t lineart_identify_feature_line_me(LineartRenderBuffer *rb,
+                                                 int eindex,
+                                                 LineartTriangle *rt_array,
+                                                 LineartVert *rv_array,
+                                                 float crease_threshold,
+                                                 bool use_auto_smooth,
+                                                 bool use_freestyle_edge,
+                                                 bool use_freestyle_face,
+                                                 Mesh *me,
+                                                 LineartEdgeNeighbor *en)
 {
 
   MPoly *ll = NULL, *lr = NULL;
@@ -1980,14 +1972,6 @@ static void lineart_load_tri_task(void *__restrict userdata,
   tri->intersecting_verts = (void *)&tri_task_data->tri_adj[i];
 }
 
-static int cmp_adjacent_items(const void *ps1, const void *ps2)
-{
-  LineartAdjacentItem *p1 = (LineartAdjacentItem *)ps1;
-  LineartAdjacentItem *p2 = (LineartAdjacentItem *)ps2;
-  int a = (int)p1->v1 - (int)p2->v1;
-  int b = (int)p1->v2 - (int)p2->v2;
-  return a ? a : b;
-}
 static LineartEdgeNeighbor *lineart_build_edge_neighbor(Mesh *me, int total_edges)
 {
   /* Because the mesh is traingulated, so me->totedge should be reliable? */
@@ -2011,7 +1995,8 @@ static LineartEdgeNeighbor *lineart_build_edge_neighbor(Mesh *me, int total_edge
     en[i].v2 = ai[i].v2;
   }
 
-  qsort(ai, total_edges, sizeof(LineartAdjacentItem), cmp_adjacent_items);
+  lineart_sort_adjacent_items(ai, total_edges);
+  // qsort(ai, total_edges, sizeof(LineartAdjacentItem), cmp_adjacent_items);
 
   if (0) {
     printf("Edge Adjacent tuples ");
@@ -2032,8 +2017,8 @@ static LineartEdgeNeighbor *lineart_build_edge_neighbor(Mesh *me, int total_edge
   return en;
 }
 
-__attribute__((optimize("O0"))) static void lineart_geometry_object_load_no_bmesh(
-    LineartObjectInfo *ob_info, LineartRenderBuffer *re_buf)
+static void lineart_geometry_object_load_no_bmesh(LineartObjectInfo *ob_info,
+                                                  LineartRenderBuffer *re_buf)
 {
   LineartElementLinkNode *elem_link_node;
   LineartVert *la_v_arr;
@@ -2322,7 +2307,6 @@ static void lineart_geometry_object_load_edge_neighbor(LineartObjectInfo *obi,
   Mesh *me = obi->original_me;
 
   BKE_mesh_runtime_looptri_ensure(me);
-  float(*poly_normals)[3] = BKE_mesh_poly_normals_ensure(me);
   const int tot_tri = me->runtime.looptris.len;
 
   unsigned int total_edges = me->runtime.looptris.len * 3;
@@ -2459,7 +2443,7 @@ static void lineart_geometry_object_load_edge_neighbor(LineartObjectInfo *obi,
 
     /* Because e->head.hflag is char, so line type flags should not exceed positive 7 bits. */
     uint16_t eflag = lineart_identify_feature_line_me(
-        rb, i, ort, orv, use_crease, use_auto_smooth, false, false, me, en, poly_normals);
+        rb, i, ort, orv, use_crease, use_auto_smooth, false, false, me, en);
     if (eflag) {
       /* Only allocate for feature lines (instead of all lines) to save memory.
        * If allow duplicated edges, one edge gets added multiple times if it has multiple types.
diff --git a/source/blender/gpencil_modifiers/intern/lineart/lineart_intern.h b/source/blender/gpencil_modifiers/intern/lineart/lineart_intern.h
index 508061d5d98..43da4bd56dd 100644
--- a/source/blender/gpencil_modifiers/intern/lineart/lineart_intern.h
+++ b/source/blender/gpencil_modifiers/intern/lineart/lineart_intern.h
@@ -121,3 +121,13 @@ void lineart_count_and_print_render_buffer_memory(struct LineartRenderBuffer *rb
 /* Initial bounding area row/column count, setting 4 is the simplest way algorithm could function
  * efficiently. */
 #define LRT_BA_ROWS 4
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void lineart_sort_adjacent_items(LineartAdjacentItem *ai, int length);
+
+#ifdef __cplusplus
+}
+#endif