This repository has been archived on 2023-10-09. You can view files and clone it, but cannot push or open issues or pull requests.
Files
blender-archive/source/blender/bmesh/tools/bmesh_path.c
Campbell Barton c434782e3a File headers: SPDX License migration
Use a shorter/simpler license convention, stops the header taking so
much space.

Follow the SPDX license specification: https://spdx.org/licenses

- C/C++/objc/objc++
- Python
- Shell Scripts
- CMake, GNUmakefile

While most of the source tree has been included

- `./extern/` was left out.
- `./intern/cycles` & `./intern/atomic` are also excluded because they
  use different header conventions.

doc/license/SPDX-license-identifiers.txt has been added to list SPDX all
used identifiers.

See P2788 for the script that automated these edits.

Reviewed By: brecht, mont29, sergey

Ref D14069
2022-02-11 09:14:36 +11:00

571 lines
18 KiB
C

/* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup bmesh
*
* Find a path between 2 elements.
*
* \note All 3 functions are similar, changes to one most likely apply to another.
*/
#include "MEM_guardedalloc.h"
#include "BLI_heap_simple.h"
#include "BLI_linklist.h"
#include "BLI_math.h"
#include "bmesh.h"
#include "bmesh_path.h" /* own include */
#define COST_INIT_MAX FLT_MAX
/* -------------------------------------------------------------------- */
/** \name Generic Helpers
* \{ */
/**
* Use skip options when we want to start measuring from a boundary.
*/
static float step_cost_3_v3_ex(
const float v1[3], const float v2[3], const float v3[3], bool skip_12, bool skip_23)
{
float d1[3], d2[3];
/* The cost is based on the simple sum of the length of the two edges. */
sub_v3_v3v3(d1, v2, v1);
sub_v3_v3v3(d2, v3, v2);
const float cost_12 = normalize_v3(d1);
const float cost_23 = normalize_v3(d2);
const float cost = ((skip_12 ? 0.0f : cost_12) + (skip_23 ? 0.0f : cost_23));
/* But is biased to give higher values to sharp turns, so that it will take paths with
* fewer "turns" when selecting between equal-weighted paths between the two edges. */
return cost * (1.0f + 0.5f * (2.0f - sqrtf(fabsf(dot_v3v3(d1, d2)))));
}
static float step_cost_3_v3(const float v1[3], const float v2[3], const float v3[3])
{
return step_cost_3_v3_ex(v1, v2, v3, false, false);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name BM_mesh_calc_path_vert
* \{ */
static void verttag_add_adjacent(HeapSimple *heap,
BMVert *v_a,
BMVert **verts_prev,
float *cost,
const struct BMCalcPathParams *params)
{
const int v_a_index = BM_elem_index_get(v_a);
{
BMIter eiter;
BMEdge *e;
/* Loop over faces of face, but do so by first looping over loops. */
BM_ITER_ELEM (e, &eiter, v_a, BM_EDGES_OF_VERT) {
BMVert *v_b = BM_edge_other_vert(e, v_a);
if (!BM_elem_flag_test(v_b, BM_ELEM_TAG)) {
/* We know 'v_b' is not visited, check it out! */
const int v_b_index = BM_elem_index_get(v_b);
const float cost_cut = params->use_topology_distance ? 1.0f : len_v3v3(v_a->co, v_b->co);
const float cost_new = cost[v_a_index] + cost_cut;
if (cost[v_b_index] > cost_new) {
cost[v_b_index] = cost_new;
verts_prev[v_b_index] = v_a;
BLI_heapsimple_insert(heap, cost_new, v_b);
}
}
}
}
if (params->use_step_face) {
BMIter liter;
BMLoop *l;
/* Loop over faces of face, but do so by first looping over loops. */
BM_ITER_ELEM (l, &liter, v_a, BM_LOOPS_OF_VERT) {
if (l->f->len > 3) {
/* Skip loops on adjacent edges. */
BMLoop *l_iter = l->next->next;
do {
BMVert *v_b = l_iter->v;
if (!BM_elem_flag_test(v_b, BM_ELEM_TAG)) {
/* We know 'v_b' is not visited, check it out! */
const int v_b_index = BM_elem_index_get(v_b);
const float cost_cut = params->use_topology_distance ? 1.0f :
len_v3v3(v_a->co, v_b->co);
const float cost_new = cost[v_a_index] + cost_cut;
if (cost[v_b_index] > cost_new) {
cost[v_b_index] = cost_new;
verts_prev[v_b_index] = v_a;
BLI_heapsimple_insert(heap, cost_new, v_b);
}
}
} while ((l_iter = l_iter->next) != l->prev);
}
}
}
}
LinkNode *BM_mesh_calc_path_vert(BMesh *bm,
BMVert *v_src,
BMVert *v_dst,
const struct BMCalcPathParams *params,
bool (*filter_fn)(BMVert *, void *user_data),
void *user_data)
{
LinkNode *path = NULL;
/* #BM_ELEM_TAG flag is used to store visited edges. */
BMVert *v;
BMIter viter;
HeapSimple *heap;
float *cost;
BMVert **verts_prev;
int i, totvert;
/* NOTE: would pass #BM_EDGE except we are looping over all faces anyway. */
// BM_mesh_elem_index_ensure(bm, BM_VERT /* | BM_EDGE */); // NOT NEEDED FOR FACETAG
BM_ITER_MESH_INDEX (v, &viter, bm, BM_VERTS_OF_MESH, i) {
BM_elem_flag_set(v, BM_ELEM_TAG, !filter_fn(v, user_data));
BM_elem_index_set(v, i); /* set_inline */
}
bm->elem_index_dirty &= ~BM_VERT;
/* Allocate. */
totvert = bm->totvert;
verts_prev = MEM_callocN(sizeof(*verts_prev) * totvert, __func__);
cost = MEM_mallocN(sizeof(*cost) * totvert, __func__);
copy_vn_fl(cost, totvert, COST_INIT_MAX);
/*
* Arrays are now filled as follows:
*
* As the search continues, `verts_prev[n]` will be the previous verts on the shortest
* path found so far to face `n`. #BM_ELEM_TAG is used to tag elements we have visited,
* `cost[n]` will contain the length of the shortest
* path to face n found so far, Finally, heap is a priority heap which is built on the
* the same data as the cost array, but inverted: it is a work-list of faces prioritized
* by the shortest path found so far to the face.
*/
/* Regular dijkstra shortest path, but over faces instead of vertices. */
heap = BLI_heapsimple_new();
BLI_heapsimple_insert(heap, 0.0f, v_src);
cost[BM_elem_index_get(v_src)] = 0.0f;
while (!BLI_heapsimple_is_empty(heap)) {
v = BLI_heapsimple_pop_min(heap);
if (v == v_dst) {
break;
}
if (!BM_elem_flag_test(v, BM_ELEM_TAG)) {
BM_elem_flag_enable(v, BM_ELEM_TAG);
verttag_add_adjacent(heap, v, verts_prev, cost, params);
}
}
if (v == v_dst) {
do {
BLI_linklist_prepend(&path, v);
} while ((v = verts_prev[BM_elem_index_get(v)]));
}
MEM_freeN(verts_prev);
MEM_freeN(cost);
BLI_heapsimple_free(heap, NULL);
return path;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name BM_mesh_calc_path_edge
* \{ */
static float edgetag_cut_cost_vert(BMEdge *e_a, BMEdge *e_b, BMVert *v)
{
BMVert *v1 = BM_edge_other_vert(e_a, v);
BMVert *v2 = BM_edge_other_vert(e_b, v);
return step_cost_3_v3(v1->co, v->co, v2->co);
}
static float edgetag_cut_cost_face(BMEdge *e_a, BMEdge *e_b, BMFace *f)
{
float e_a_cent[3], e_b_cent[3], f_cent[3];
mid_v3_v3v3(e_a_cent, e_a->v1->co, e_a->v1->co);
mid_v3_v3v3(e_b_cent, e_b->v1->co, e_b->v1->co);
BM_face_calc_center_median_weighted(f, f_cent);
return step_cost_3_v3(e_a_cent, e_b_cent, f_cent);
}
static void edgetag_add_adjacent(HeapSimple *heap,
BMEdge *e_a,
BMEdge **edges_prev,
float *cost,
const struct BMCalcPathParams *params)
{
const int e_a_index = BM_elem_index_get(e_a);
/* Unlike vert/face, stepping faces disables scanning connected edges
* and only steps over faces (selecting a ring of edges instead of a loop). */
if (params->use_step_face == false || e_a->l == NULL) {
BMIter viter;
BMVert *v;
BMIter eiter;
BMEdge *e_b;
BM_ITER_ELEM (v, &viter, e_a, BM_VERTS_OF_EDGE) {
/* Don't walk over previous vertex. */
if ((edges_prev[e_a_index]) && (BM_vert_in_edge(edges_prev[e_a_index], v))) {
continue;
}
BM_ITER_ELEM (e_b, &eiter, v, BM_EDGES_OF_VERT) {
if (!BM_elem_flag_test(e_b, BM_ELEM_TAG)) {
/* We know 'e_b' is not visited, check it out! */
const int e_b_index = BM_elem_index_get(e_b);
const float cost_cut = params->use_topology_distance ?
1.0f :
edgetag_cut_cost_vert(e_a, e_b, v);
const float cost_new = cost[e_a_index] + cost_cut;
if (cost[e_b_index] > cost_new) {
cost[e_b_index] = cost_new;
edges_prev[e_b_index] = e_a;
BLI_heapsimple_insert(heap, cost_new, e_b);
}
}
}
}
}
else {
BMLoop *l_first, *l_iter;
l_iter = l_first = e_a->l;
do {
BMLoop *l_cycle_iter, *l_cycle_end;
l_cycle_iter = l_iter->next;
l_cycle_end = l_iter;
/* Good, but we need to allow this otherwise paths may fail to connect at all. */
#if 0
if (l_iter->f->len > 3) {
l_cycle_iter = l_cycle_iter->next;
l_cycle_end = l_cycle_end->prev;
}
#endif
do {
BMEdge *e_b = l_cycle_iter->e;
if (!BM_elem_flag_test(e_b, BM_ELEM_TAG)) {
/* We know 'e_b' is not visited, check it out! */
const int e_b_index = BM_elem_index_get(e_b);
const float cost_cut = params->use_topology_distance ?
1.0f :
edgetag_cut_cost_face(e_a, e_b, l_iter->f);
const float cost_new = cost[e_a_index] + cost_cut;
if (cost[e_b_index] > cost_new) {
cost[e_b_index] = cost_new;
edges_prev[e_b_index] = e_a;
BLI_heapsimple_insert(heap, cost_new, e_b);
}
}
} while ((l_cycle_iter = l_cycle_iter->next) != l_cycle_end);
} while ((l_iter = l_iter->radial_next) != l_first);
}
}
LinkNode *BM_mesh_calc_path_edge(BMesh *bm,
BMEdge *e_src,
BMEdge *e_dst,
const struct BMCalcPathParams *params,
bool (*filter_fn)(BMEdge *, void *user_data),
void *user_data)
{
LinkNode *path = NULL;
/* #BM_ELEM_TAG flag is used to store visited edges. */
BMEdge *e;
BMIter eiter;
HeapSimple *heap;
float *cost;
BMEdge **edges_prev;
int i, totedge;
/* NOTE: would pass #BM_EDGE except we are looping over all edges anyway. */
BM_mesh_elem_index_ensure(bm, BM_VERT /* | BM_EDGE */);
BM_ITER_MESH_INDEX (e, &eiter, bm, BM_EDGES_OF_MESH, i) {
BM_elem_flag_set(e, BM_ELEM_TAG, !filter_fn(e, user_data));
BM_elem_index_set(e, i); /* set_inline */
}
bm->elem_index_dirty &= ~BM_EDGE;
/* Allocate. */
totedge = bm->totedge;
edges_prev = MEM_callocN(sizeof(*edges_prev) * totedge, __func__);
cost = MEM_mallocN(sizeof(*cost) * totedge, __func__);
copy_vn_fl(cost, totedge, COST_INIT_MAX);
/*
* Arrays are now filled as follows:
*
* As the search continues, `edges_prev[n]` will be the previous edge on the shortest
* path found so far to edge `n`. #BM_ELEM_TAG is used to tag elements we have visited,
* `cost[n]` will contain the length of the shortest
* path to edge n found so far, Finally, heap is a priority heap which is built on the
* the same data as the cost array, but inverted: it is a work-list of edges prioritized
* by the shortest path found so far to the edge.
*/
/* Regular dijkstra shortest path, but over edges instead of vertices. */
heap = BLI_heapsimple_new();
BLI_heapsimple_insert(heap, 0.0f, e_src);
cost[BM_elem_index_get(e_src)] = 0.0f;
while (!BLI_heapsimple_is_empty(heap)) {
e = BLI_heapsimple_pop_min(heap);
if (e == e_dst) {
break;
}
if (!BM_elem_flag_test(e, BM_ELEM_TAG)) {
BM_elem_flag_enable(e, BM_ELEM_TAG);
edgetag_add_adjacent(heap, e, edges_prev, cost, params);
}
}
if (e == e_dst) {
do {
BLI_linklist_prepend(&path, e);
} while ((e = edges_prev[BM_elem_index_get(e)]));
}
MEM_freeN(edges_prev);
MEM_freeN(cost);
BLI_heapsimple_free(heap, NULL);
return path;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name BM_mesh_calc_path_face
* \{ */
static float facetag_cut_cost_edge(BMFace *f_a,
BMFace *f_b,
BMEdge *e,
const void *const f_endpoints[2])
{
float f_a_cent[3];
float f_b_cent[3];
float e_cent[3];
BM_face_calc_center_median_weighted(f_a, f_a_cent);
BM_face_calc_center_median_weighted(f_b, f_b_cent);
#if 0
mid_v3_v3v3(e_cent, e->v1->co, e->v2->co);
#else
/* For triangle fans it gives better results to pick a point on the edge. */
{
float ix_e[3], ix_f[3];
isect_line_line_v3(e->v1->co, e->v2->co, f_a_cent, f_b_cent, ix_e, ix_f);
const float factor = line_point_factor_v3(ix_e, e->v1->co, e->v2->co);
if (factor < 0.0f) {
copy_v3_v3(e_cent, e->v1->co);
}
else if (factor > 1.0f) {
copy_v3_v3(e_cent, e->v2->co);
}
else {
copy_v3_v3(e_cent, ix_e);
}
}
#endif
return step_cost_3_v3_ex(
f_a_cent, e_cent, f_b_cent, (f_a == f_endpoints[0]), (f_b == f_endpoints[1]));
}
static float facetag_cut_cost_vert(BMFace *f_a,
BMFace *f_b,
BMVert *v,
const void *const f_endpoints[2])
{
float f_a_cent[3];
float f_b_cent[3];
BM_face_calc_center_median_weighted(f_a, f_a_cent);
BM_face_calc_center_median_weighted(f_b, f_b_cent);
return step_cost_3_v3_ex(
f_a_cent, v->co, f_b_cent, (f_a == f_endpoints[0]), (f_b == f_endpoints[1]));
}
static void facetag_add_adjacent(HeapSimple *heap,
BMFace *f_a,
BMFace **faces_prev,
float *cost,
const void *const f_endpoints[2],
const struct BMCalcPathParams *params)
{
const int f_a_index = BM_elem_index_get(f_a);
/* Loop over faces of face, but do so by first looping over loops. */
{
BMIter liter;
BMLoop *l_a;
BM_ITER_ELEM (l_a, &liter, f_a, BM_LOOPS_OF_FACE) {
BMLoop *l_first, *l_iter;
l_iter = l_first = l_a;
do {
BMFace *f_b = l_iter->f;
if (!BM_elem_flag_test(f_b, BM_ELEM_TAG)) {
/* We know 'f_b' is not visited, check it out! */
const int f_b_index = BM_elem_index_get(f_b);
const float cost_cut = params->use_topology_distance ?
1.0f :
facetag_cut_cost_edge(f_a, f_b, l_iter->e, f_endpoints);
const float cost_new = cost[f_a_index] + cost_cut;
if (cost[f_b_index] > cost_new) {
cost[f_b_index] = cost_new;
faces_prev[f_b_index] = f_a;
BLI_heapsimple_insert(heap, cost_new, f_b);
}
}
} while ((l_iter = l_iter->radial_next) != l_first);
}
}
if (params->use_step_face) {
BMIter liter;
BMLoop *l_a;
BM_ITER_ELEM (l_a, &liter, f_a, BM_LOOPS_OF_FACE) {
BMIter litersub;
BMLoop *l_b;
BM_ITER_ELEM (l_b, &litersub, l_a->v, BM_LOOPS_OF_VERT) {
if ((l_a != l_b) && !BM_loop_share_edge_check(l_a, l_b)) {
BMFace *f_b = l_b->f;
if (!BM_elem_flag_test(f_b, BM_ELEM_TAG)) {
/* We know 'f_b' is not visited, check it out! */
const int f_b_index = BM_elem_index_get(f_b);
const float cost_cut = params->use_topology_distance ?
1.0f :
facetag_cut_cost_vert(f_a, f_b, l_a->v, f_endpoints);
const float cost_new = cost[f_a_index] + cost_cut;
if (cost[f_b_index] > cost_new) {
cost[f_b_index] = cost_new;
faces_prev[f_b_index] = f_a;
BLI_heapsimple_insert(heap, cost_new, f_b);
}
}
}
}
}
}
}
LinkNode *BM_mesh_calc_path_face(BMesh *bm,
BMFace *f_src,
BMFace *f_dst,
const struct BMCalcPathParams *params,
bool (*filter_fn)(BMFace *, void *user_data),
void *user_data)
{
LinkNode *path = NULL;
/* #BM_ELEM_TAG flag is used to store visited edges. */
BMFace *f;
BMIter fiter;
HeapSimple *heap;
float *cost;
BMFace **faces_prev;
int i, totface;
/* Start measuring face path at the face edges, ignoring their centers. */
const void *const f_endpoints[2] = {f_src, f_dst};
/* NOTE: would pass #BM_EDGE except we are looping over all faces anyway. */
// BM_mesh_elem_index_ensure(bm, BM_VERT /* | BM_EDGE */); // NOT NEEDED FOR FACETAG
BM_ITER_MESH_INDEX (f, &fiter, bm, BM_FACES_OF_MESH, i) {
BM_elem_flag_set(f, BM_ELEM_TAG, !filter_fn(f, user_data));
BM_elem_index_set(f, i); /* set_inline */
}
bm->elem_index_dirty &= ~BM_FACE;
/* Allocate. */
totface = bm->totface;
faces_prev = MEM_callocN(sizeof(*faces_prev) * totface, __func__);
cost = MEM_mallocN(sizeof(*cost) * totface, __func__);
copy_vn_fl(cost, totface, COST_INIT_MAX);
/*
* Arrays are now filled as follows:
*
* As the search continues, `faces_prev[n]` will be the previous face on the shortest
* path found so far to face `n`. #BM_ELEM_TAG is used to tag elements we have visited,
* `cost[n]` will contain the length of the shortest
* path to face n found so far, Finally, heap is a priority heap which is built on the
* the same data as the cost array, but inverted: it is a work-list of faces prioritized
* by the shortest path found so far to the face.
*/
/* Regular dijkstra shortest path, but over faces instead of vertices. */
heap = BLI_heapsimple_new();
BLI_heapsimple_insert(heap, 0.0f, f_src);
cost[BM_elem_index_get(f_src)] = 0.0f;
while (!BLI_heapsimple_is_empty(heap)) {
f = BLI_heapsimple_pop_min(heap);
if (f == f_dst) {
break;
}
if (!BM_elem_flag_test(f, BM_ELEM_TAG)) {
BM_elem_flag_enable(f, BM_ELEM_TAG);
facetag_add_adjacent(heap, f, faces_prev, cost, f_endpoints, params);
}
}
if (f == f_dst) {
do {
BLI_linklist_prepend(&path, f);
} while ((f = faces_prev[BM_elem_index_get(f)]));
}
MEM_freeN(faces_prev);
MEM_freeN(cost);
BLI_heapsimple_free(heap, NULL);
return path;
}
/** \} */