Files
test2/source/blender/blenkernel/intern/bvhutils.cc

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

1398 lines
43 KiB
C++
Raw Normal View History

/* SPDX-FileCopyrightText: Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
2011-02-27 20:40:57 +00:00
/** \file
* \ingroup bke
2011-02-27 20:40:57 +00:00
*/
#include <cmath>
#include <cstdio>
#include <cstring>
2018-05-08 10:07:21 +02:00
#include "DNA_mesh_types.h"
#include "DNA_meshdata_types.h"
#include "DNA_pointcloud_types.h"
#include "BLI_bit_vector.hh"
2011-10-22 01:53:35 +00:00
#include "BLI_linklist.h"
#include "BLI_math_geom.h"
#include "BLI_math_vector.h"
Mesh: Remove redundant custom data pointers For copy-on-write, we want to share attribute arrays between meshes where possible. Mutable pointers like `Mesh.mvert` make that difficult by making ownership vague. They also make code more complex by adding redundancy. The simplest solution is just removing them and retrieving layers from `CustomData` as needed. Similar changes have already been applied to curves and point clouds (e9f82d3dc7ee, 410a6efb747f). Removing use of the pointers generally makes code more obvious and more reusable. Mesh data is now accessed with a C++ API (`Mesh::edges()` or `Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`). The CoW changes this commit makes possible are described in T95845 and T95842, and started in D14139 and D14140. The change also simplifies the ongoing mesh struct-of-array refactors from T95965. **RNA/Python Access Performance** Theoretically, accessing mesh elements with the RNA API may become slower, since the layer needs to be found on every random access. However, overhead is already high enough that this doesn't make a noticible differenc, and performance is actually improved in some cases. Random access can be up to 10% faster, but other situations might be a bit slower. Generally using `foreach_get/set` are the best way to improve performance. See the differential revision for more discussion about Python performance. Cycles has been updated to use raw pointers and the internal Blender mesh types, mostly because there is no sense in having this overhead when it's already compiled with Blender. In my tests this roughly halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million face grid). Differential Revision: https://developer.blender.org/D15488
2022-09-05 11:56:34 -05:00
#include "BLI_span.hh"
#include "BLI_task.h"
Make bvhutil safe for multi-threaded usage There were couple of reasons why it wasn't safe for usage from multiple threads. First of all, it was trying to cache BVH in derived mesh, which wasn't safe because multiple threads might have requested BVH tree and simultaneous reading and writing to the cache became a big headache. Solved this with RW lock so now access to BVH cache is safe. Another issue is causes by the fact that it's not guaranteed DM to have vert/edge/face arrays pre-allocated and when one was calling functions like getVertDataArray() array could have been allocated and marked as temporary. This is REALLY bad, because NO ONE is ever allowed to modify data which doesn't belong to him. This lead to situations when multiple threads were using BVH tree and they run into race condition with this temporary allocated arrays. Now bvhtree owns allocated arrays and keeps track of them, so no race condition happens with temporary data stored in the derived mesh. This solved threading issues and likely wouldn't introduce noticeable slowdown. Even when DM was keeping track of this arrays, they were re-allocated on every BVH creation anyway, because those arrays were temporary and were freed with dm->release() call. We might re-consider this a bit and make it so BVH trees are allocated when DM itself is being allocated based on the DAG layout, but that i'd consider an optimization and not something we need to do 1st priority. Fixes crash happening with 05_4g_track.blend from Mango after the threaded object update landed to master.
2014-01-10 17:21:39 +06:00
#include "BLI_threads.h"
#include "BLI_utildefines.h"
#include "BKE_attribute.hh"
#include "BKE_bvhutils.hh"
#include "BKE_editmesh.hh"
#include "BKE_mesh.hh"
#include "BKE_mesh_runtime.hh"
#include "BKE_pointcloud.hh"
#include "MEM_guardedalloc.h"
using blender::BitSpan;
using blender::BitVector;
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
using blender::float3;
using blender::IndexRange;
using blender::int3;
Mesh: Remove redundant custom data pointers For copy-on-write, we want to share attribute arrays between meshes where possible. Mutable pointers like `Mesh.mvert` make that difficult by making ownership vague. They also make code more complex by adding redundancy. The simplest solution is just removing them and retrieving layers from `CustomData` as needed. Similar changes have already been applied to curves and point clouds (e9f82d3dc7ee, 410a6efb747f). Removing use of the pointers generally makes code more obvious and more reusable. Mesh data is now accessed with a C++ API (`Mesh::edges()` or `Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`). The CoW changes this commit makes possible are described in T95845 and T95842, and started in D14139 and D14140. The change also simplifies the ongoing mesh struct-of-array refactors from T95965. **RNA/Python Access Performance** Theoretically, accessing mesh elements with the RNA API may become slower, since the layer needs to be found on every random access. However, overhead is already high enough that this doesn't make a noticible differenc, and performance is actually improved in some cases. Random access can be up to 10% faster, but other situations might be a bit slower. Generally using `foreach_get/set` are the best way to improve performance. See the differential revision for more discussion about Python performance. Cycles has been updated to use raw pointers and the internal Blender mesh types, mostly because there is no sense in having this overhead when it's already compiled with Blender. In my tests this roughly halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million face grid). Differential Revision: https://developer.blender.org/D15488
2022-09-05 11:56:34 -05:00
using blender::Span;
using blender::VArray;
/* -------------------------------------------------------------------- */
/** \name BVHCache
* \{ */
struct BVHCacheItem {
bool is_filled;
BVHTree *tree;
};
struct BVHCache {
BVHCacheItem items[BVHTREE_MAX_ITEM];
ThreadMutex mutex;
};
/**
* Queries a bvhcache for the cache bvhtree of the request type
*
* When the `r_locked` is filled and the tree could not be found the caches mutex will be
* locked. This mutex can be unlocked by calling `bvhcache_unlock`.
*
* When `r_locked` is used the `mesh_eval_mutex` must contain the `MeshRuntime.eval_mutex`.
*/
static bool bvhcache_find(BVHCache **bvh_cache_p,
BVHCacheType type,
BVHTree **r_tree,
bool *r_locked,
std::mutex *mesh_eval_mutex)
{
bool do_lock = r_locked;
if (r_locked) {
*r_locked = false;
}
if (*bvh_cache_p == nullptr) {
if (!do_lock) {
/* Cache does not exist and no lock is requested. */
return false;
}
/* Lazy initialization of the bvh_cache using the `mesh_eval_mutex`. */
std::lock_guard lock{*mesh_eval_mutex};
if (*bvh_cache_p == nullptr) {
*bvh_cache_p = bvhcache_init();
}
}
BVHCache *bvh_cache = *bvh_cache_p;
Make bvhutil safe for multi-threaded usage There were couple of reasons why it wasn't safe for usage from multiple threads. First of all, it was trying to cache BVH in derived mesh, which wasn't safe because multiple threads might have requested BVH tree and simultaneous reading and writing to the cache became a big headache. Solved this with RW lock so now access to BVH cache is safe. Another issue is causes by the fact that it's not guaranteed DM to have vert/edge/face arrays pre-allocated and when one was calling functions like getVertDataArray() array could have been allocated and marked as temporary. This is REALLY bad, because NO ONE is ever allowed to modify data which doesn't belong to him. This lead to situations when multiple threads were using BVH tree and they run into race condition with this temporary allocated arrays. Now bvhtree owns allocated arrays and keeps track of them, so no race condition happens with temporary data stored in the derived mesh. This solved threading issues and likely wouldn't introduce noticeable slowdown. Even when DM was keeping track of this arrays, they were re-allocated on every BVH creation anyway, because those arrays were temporary and were freed with dm->release() call. We might re-consider this a bit and make it so BVH trees are allocated when DM itself is being allocated based on the DAG layout, but that i'd consider an optimization and not something we need to do 1st priority. Fixes crash happening with 05_4g_track.blend from Mango after the threaded object update landed to master.
2014-01-10 17:21:39 +06:00
if (bvh_cache->items[type].is_filled) {
*r_tree = bvh_cache->items[type].tree;
return true;
}
if (do_lock) {
BLI_mutex_lock(&bvh_cache->mutex);
bool in_cache = bvhcache_find(bvh_cache_p, type, r_tree, nullptr, nullptr);
if (in_cache) {
BLI_mutex_unlock(&bvh_cache->mutex);
return in_cache;
}
*r_locked = true;
}
return false;
}
static void bvhcache_unlock(BVHCache *bvh_cache, bool lock_started)
{
if (lock_started) {
BLI_mutex_unlock(&bvh_cache->mutex);
}
}
bool bvhcache_has_tree(const BVHCache *bvh_cache, const BVHTree *tree)
{
if (bvh_cache == nullptr) {
return false;
}
for (int i = 0; i < BVHTREE_MAX_ITEM; i++) {
if (bvh_cache->items[i].tree == tree) {
return true;
}
}
return false;
}
BVHCache *bvhcache_init()
{
BVHCache *cache = MEM_cnew<BVHCache>(__func__);
BLI_mutex_init(&cache->mutex);
return cache;
}
/**
* Inserts a BVHTree of the given type under the cache
* After that the caller no longer needs to worry when to free the BVHTree
* as that will be done when the cache is freed.
*
* A call to this assumes that there was no previous cached tree of the given type
* \warning The #BVHTree can be nullptr.
*/
static void bvhcache_insert(BVHCache *bvh_cache, BVHTree *tree, BVHCacheType type)
{
BVHCacheItem *item = &bvh_cache->items[type];
BLI_assert(!item->is_filled);
item->tree = tree;
item->is_filled = true;
}
void bvhcache_free(BVHCache *bvh_cache)
{
for (int index = 0; index < BVHTREE_MAX_ITEM; index++) {
BVHCacheItem *item = &bvh_cache->items[index];
BLI_bvhtree_free(item->tree);
item->tree = nullptr;
}
BLI_mutex_end(&bvh_cache->mutex);
MEM_freeN(bvh_cache);
}
2021-10-12 17:52:35 +11:00
/**
* BVH-tree balancing inside a mutex lock must be run in isolation. Balancing
* is multithreaded, and we do not want the current thread to start another task
2021-10-12 17:52:35 +11:00
* that may involve acquiring the same mutex lock that it is waiting for.
*/
static void bvhtree_balance_isolated(void *userdata)
{
BLI_bvhtree_balance((BVHTree *)userdata);
}
static void bvhtree_balance(BVHTree *tree, const bool isolate)
{
if (tree) {
if (isolate) {
BLI_task_isolate(bvhtree_balance_isolated, tree);
}
else {
BLI_bvhtree_balance(tree);
}
}
}
/** \} */
2015-07-22 20:35:33 +10:00
/* -------------------------------------------------------------------- */
/** \name Local Callbacks
* \{ */
/* Math stuff for ray casting on mesh faces and for nearest surface */
2015-07-22 20:35:33 +10:00
float bvhtree_ray_tri_intersection(const BVHTreeRay *ray,
const float /*m_dist*/,
2015-07-22 20:35:33 +10:00
const float v0[3],
const float v1[3],
const float v2[3])
{
float dist;
#ifdef USE_KDOPBVH_WATERTIGHT
if (isect_ray_tri_watertight_v3(ray->origin, ray->isect_precalc, v0, v1, v2, &dist, nullptr))
#else
if (isect_ray_tri_epsilon_v3(
ray->origin, ray->direction, v0, v1, v2, &dist, nullptr, FLT_EPSILON))
#endif
{
return dist;
}
return FLT_MAX;
}
float bvhtree_sphereray_tri_intersection(const BVHTreeRay *ray,
2015-07-22 20:35:33 +10:00
float radius,
const float m_dist,
const float v0[3],
const float v1[3],
const float v2[3])
{
2018-06-17 17:05:51 +02:00
float idist;
float p1[3];
float hit_point[3];
2011-11-06 15:17:43 +00:00
madd_v3_v3v3fl(p1, ray->origin, ray->direction, m_dist);
2012-03-07 04:53:43 +00:00
if (isect_sweeping_sphere_tri_v3(ray->origin, p1, radius, v0, v1, v2, &idist, hit_point)) {
return idist * m_dist;
}
return FLT_MAX;
}
/*
* BVH from meshes callbacks
*/
2021-10-12 17:52:35 +11:00
/**
* Callback to BVH-tree nearest point.
* The tree must have been built using #bvhtree_from_mesh_faces.
*
* \param userdata: Must be a #BVHMeshCallbackUserdata built from the same mesh as the tree.
*/
2011-11-06 15:17:43 +00:00
static void mesh_faces_nearest_point(void *userdata,
int index,
const float co[3],
BVHTreeNearest *nearest)
{
2012-05-06 17:22:54 +00:00
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
2015-07-22 17:39:33 +10:00
const MFace *face = data->face + index;
const float *t0, *t1, *t2, *t3;
t0 = data->vert_positions[face->v1];
t1 = data->vert_positions[face->v2];
t2 = data->vert_positions[face->v3];
t3 = face->v4 ? &data->vert_positions[face->v4].x : nullptr;
2012-05-06 17:22:54 +00:00
do {
float nearest_tmp[3], dist_sq;
closest_on_tri_to_point_v3(nearest_tmp, co, t0, t1, t2);
dist_sq = len_squared_v3v3(co, nearest_tmp);
if (dist_sq < nearest->dist_sq) {
nearest->index = index;
nearest->dist_sq = dist_sq;
2011-11-06 15:17:43 +00:00
copy_v3_v3(nearest->co, nearest_tmp);
2012-04-29 15:47:02 +00:00
normal_tri_v3(nearest->no, t0, t1, t2);
}
t1 = t2;
t2 = t3;
t3 = nullptr;
} while (t2);
}
/* copy of function above */
static void mesh_corner_tris_nearest_point(void *userdata,
int index,
const float co[3],
BVHTreeNearest *nearest)
{
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
const int3 &tri = data->corner_tris[index];
const float *vtri_co[3] = {
data->vert_positions[data->corner_verts[tri[0]]],
data->vert_positions[data->corner_verts[tri[1]]],
data->vert_positions[data->corner_verts[tri[2]]],
};
float nearest_tmp[3], dist_sq;
closest_on_tri_to_point_v3(nearest_tmp, co, UNPACK3(vtri_co));
dist_sq = len_squared_v3v3(co, nearest_tmp);
if (dist_sq < nearest->dist_sq) {
nearest->index = index;
nearest->dist_sq = dist_sq;
copy_v3_v3(nearest->co, nearest_tmp);
normal_tri_v3(nearest->no, UNPACK3(vtri_co));
}
}
/* Copy of function above (warning, should de-duplicate with `editmesh_bvh.cc`). */
static void editmesh_corner_tris_nearest_point(void *userdata,
int index,
const float co[3],
BVHTreeNearest *nearest)
{
BMEditMesh *em = static_cast<BMEditMesh *>(userdata);
const BMLoop **ltri = const_cast<const BMLoop **>(em->looptris[index]);
const float *t0, *t1, *t2;
t0 = ltri[0]->v->co;
t1 = ltri[1]->v->co;
t2 = ltri[2]->v->co;
{
float nearest_tmp[3], dist_sq;
closest_on_tri_to_point_v3(nearest_tmp, co, t0, t1, t2);
dist_sq = len_squared_v3v3(co, nearest_tmp);
if (dist_sq < nearest->dist_sq) {
nearest->index = index;
nearest->dist_sq = dist_sq;
copy_v3_v3(nearest->co, nearest_tmp);
normal_tri_v3(nearest->no, t0, t1, t2);
}
}
}
2021-10-12 17:52:35 +11:00
/**
* Callback to BVH-tree ray-cast.
* The tree must have been built using bvhtree_from_mesh_faces.
*
* \param userdata: Must be a #BVHMeshCallbackUserdata built from the same mesh as the tree.
*/
static void mesh_faces_spherecast(void *userdata,
int index,
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
2012-05-06 17:22:54 +00:00
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
2015-07-22 17:39:33 +10:00
const MFace *face = &data->face[index];
const float *t0, *t1, *t2, *t3;
t0 = data->vert_positions[face->v1];
t1 = data->vert_positions[face->v2];
t2 = data->vert_positions[face->v3];
t3 = face->v4 ? &data->vert_positions[face->v4].x : nullptr;
2012-05-06 17:22:54 +00:00
do {
float dist;
if (ray->radius == 0.0f) {
dist = bvhtree_ray_tri_intersection(ray, hit->dist, t0, t1, t2);
}
else {
dist = bvhtree_sphereray_tri_intersection(ray, ray->radius, hit->dist, t0, t1, t2);
}
2012-03-07 04:53:43 +00:00
if (dist >= 0 && dist < hit->dist) {
hit->index = index;
hit->dist = dist;
2011-11-06 15:17:43 +00:00
madd_v3_v3v3fl(hit->co, ray->origin, ray->direction, dist);
2012-04-29 15:47:02 +00:00
normal_tri_v3(hit->no, t0, t1, t2);
}
t1 = t2;
t2 = t3;
t3 = nullptr;
} while (t2);
}
/* copy of function above */
static void mesh_corner_tris_spherecast(void *userdata,
int index,
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
const Span<float3> positions = data->vert_positions;
const int3 &tri = data->corner_tris[index];
const float *vtri_co[3] = {
positions[data->corner_verts[tri[0]]],
positions[data->corner_verts[tri[1]]],
positions[data->corner_verts[tri[2]]],
};
float dist;
if (ray->radius == 0.0f) {
dist = bvhtree_ray_tri_intersection(ray, hit->dist, UNPACK3(vtri_co));
}
else {
dist = bvhtree_sphereray_tri_intersection(ray, ray->radius, hit->dist, UNPACK3(vtri_co));
}
if (dist >= 0 && dist < hit->dist) {
hit->index = index;
hit->dist = dist;
madd_v3_v3v3fl(hit->co, ray->origin, ray->direction, dist);
normal_tri_v3(hit->no, UNPACK3(vtri_co));
}
}
/* Copy of function above (warning, should de-duplicate with `editmesh_bvh.cc`). */
static void editmesh_corner_tris_spherecast(void *userdata,
int index,
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
BMEditMesh *em = static_cast<BMEditMesh *>(userdata);
const BMLoop **ltri = const_cast<const BMLoop **>(em->looptris[index]);
const float *t0, *t1, *t2;
t0 = ltri[0]->v->co;
t1 = ltri[1]->v->co;
t2 = ltri[2]->v->co;
{
float dist;
if (ray->radius == 0.0f) {
dist = bvhtree_ray_tri_intersection(ray, hit->dist, t0, t1, t2);
}
else {
dist = bvhtree_sphereray_tri_intersection(ray, ray->radius, hit->dist, t0, t1, t2);
}
if (dist >= 0 && dist < hit->dist) {
hit->index = index;
hit->dist = dist;
madd_v3_v3v3fl(hit->co, ray->origin, ray->direction, dist);
normal_tri_v3(hit->no, t0, t1, t2);
}
}
}
2021-10-12 17:52:35 +11:00
/**
* Callback to BVH-tree nearest point.
* The tree must have been built using #bvhtree_from_mesh_edges.
*
* \param userdata: Must be a #BVHMeshCallbackUserdata built from the same mesh as the tree.
*/
2011-11-06 15:17:43 +00:00
static void mesh_edges_nearest_point(void *userdata,
int index,
const float co[3],
BVHTreeNearest *nearest)
{
2012-05-06 17:22:54 +00:00
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
const Span<float3> positions = data->vert_positions;
const blender::int2 edge = data->edges[index];
float nearest_tmp[3], dist_sq;
const float *t0, *t1;
Mesh: Move edges to a generic attribute Implements #95966, as the final step of #95965. This commit changes the storage of mesh edge vertex indices from the `MEdge` type to the generic `int2` attribute type. This follows the general design for geometry and the attribute system, where the data storage type and the usage semantics are separated. The main benefit of the change is reduced memory usage-- the requirements of storing mesh edges is reduced by 1/3. For example, this saves 8MB on a 1 million vertex grid. This also gives performance benefits to any memory-bound mesh processing algorithm that uses edges. Another benefit is that all of the edge's vertex indices are contiguous. In a few cases, it's helpful to process all of them as `Span<int>` rather than `Span<int2>`. Similarly, the type is more likely to match a generic format used by a library, or code that shouldn't know about specific Blender `Mesh` types. Various Notes: - The `.edge_verts` name is used to reflect a mapping between domains, similar to `.corner_verts`, etc. The period means that it the data shouldn't change arbitrarily by the user or procedural operations. - `edge[0]` is now used instead of `edge.v1` - Signed integers are used instead of unsigned to reduce the mixing of signed-ness, which can be error prone. - All of the previously used core mesh data types (`MVert`, `MEdge`, `MLoop`, `MPoly` are now deprecated. Only generic types are used). - The `vec2i` DNA type is used in the few C files where necessary. Pull Request: https://projects.blender.org/blender/blender/pulls/106638
2023-04-17 13:47:41 +02:00
t0 = positions[edge[0]];
t1 = positions[edge[1]];
2011-11-06 15:17:43 +00:00
closest_to_line_segment_v3(nearest_tmp, co, t0, t1);
dist_sq = len_squared_v3v3(nearest_tmp, co);
if (dist_sq < nearest->dist_sq) {
nearest->index = index;
nearest->dist_sq = dist_sq;
2011-11-06 15:17:43 +00:00
copy_v3_v3(nearest->co, nearest_tmp);
sub_v3_v3v3(nearest->no, t0, t1);
normalize_v3(nearest->no);
}
}
2021-07-02 12:11:54 +10:00
/* Helper, does all the point-sphere-cast work actually. */
static void mesh_verts_spherecast_do(int index,
const float v[3],
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
float dist;
const float *r1;
float r2[3], i1[3];
r1 = ray->origin;
add_v3_v3v3(r2, r1, ray->direction);
closest_to_line_segment_v3(i1, v, r1, r2);
/* No hit if closest point is 'behind' the origin of the ray, or too far away from it. */
if ((dot_v3v3v3(r1, i1, r2) >= 0.0f) && ((dist = len_v3v3(r1, i1)) < hit->dist)) {
hit->index = index;
hit->dist = dist;
copy_v3_v3(hit->co, i1);
}
}
Make bvhutil safe for multi-threaded usage There were couple of reasons why it wasn't safe for usage from multiple threads. First of all, it was trying to cache BVH in derived mesh, which wasn't safe because multiple threads might have requested BVH tree and simultaneous reading and writing to the cache became a big headache. Solved this with RW lock so now access to BVH cache is safe. Another issue is causes by the fact that it's not guaranteed DM to have vert/edge/face arrays pre-allocated and when one was calling functions like getVertDataArray() array could have been allocated and marked as temporary. This is REALLY bad, because NO ONE is ever allowed to modify data which doesn't belong to him. This lead to situations when multiple threads were using BVH tree and they run into race condition with this temporary allocated arrays. Now bvhtree owns allocated arrays and keeps track of them, so no race condition happens with temporary data stored in the derived mesh. This solved threading issues and likely wouldn't introduce noticeable slowdown. Even when DM was keeping track of this arrays, they were re-allocated on every BVH creation anyway, because those arrays were temporary and were freed with dm->release() call. We might re-consider this a bit and make it so BVH trees are allocated when DM itself is being allocated based on the DAG layout, but that i'd consider an optimization and not something we need to do 1st priority. Fixes crash happening with 05_4g_track.blend from Mango after the threaded object update landed to master.
2014-01-10 17:21:39 +06:00
static void editmesh_verts_spherecast(void *userdata,
int index,
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
BMEditMesh *em = static_cast<BMEditMesh *>(userdata);
BMVert *eve = BM_vert_at_index(em->bm, index);
mesh_verts_spherecast_do(index, eve->co, ray, hit);
}
2021-10-12 17:52:35 +11:00
/**
* Callback to BVH-tree ray-cast.
* The tree must have been built using bvhtree_from_mesh_verts.
*
* \param userdata: Must be a #BVHMeshCallbackUserdata built from the same mesh as the tree.
*/
static void mesh_verts_spherecast(void *userdata,
int index,
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
const float *v = data->vert_positions[index];
Make bvhutil safe for multi-threaded usage There were couple of reasons why it wasn't safe for usage from multiple threads. First of all, it was trying to cache BVH in derived mesh, which wasn't safe because multiple threads might have requested BVH tree and simultaneous reading and writing to the cache became a big headache. Solved this with RW lock so now access to BVH cache is safe. Another issue is causes by the fact that it's not guaranteed DM to have vert/edge/face arrays pre-allocated and when one was calling functions like getVertDataArray() array could have been allocated and marked as temporary. This is REALLY bad, because NO ONE is ever allowed to modify data which doesn't belong to him. This lead to situations when multiple threads were using BVH tree and they run into race condition with this temporary allocated arrays. Now bvhtree owns allocated arrays and keeps track of them, so no race condition happens with temporary data stored in the derived mesh. This solved threading issues and likely wouldn't introduce noticeable slowdown. Even when DM was keeping track of this arrays, they were re-allocated on every BVH creation anyway, because those arrays were temporary and were freed with dm->release() call. We might re-consider this a bit and make it so BVH trees are allocated when DM itself is being allocated based on the DAG layout, but that i'd consider an optimization and not something we need to do 1st priority. Fixes crash happening with 05_4g_track.blend from Mango after the threaded object update landed to master.
2014-01-10 17:21:39 +06:00
mesh_verts_spherecast_do(index, v, ray, hit);
}
2021-10-12 17:52:35 +11:00
/**
* Callback to BVH-tree ray-cast.
* The tree must have been built using bvhtree_from_mesh_edges.
*
* \param userdata: Must be a #BVHMeshCallbackUserdata built from the same mesh as the tree.
*/
static void mesh_edges_spherecast(void *userdata,
int index,
const BVHTreeRay *ray,
BVHTreeRayHit *hit)
{
const BVHTreeFromMesh *data = (BVHTreeFromMesh *)userdata;
const Span<float3> positions = data->vert_positions;
const blender::int2 edge = data->edges[index];
const float radius_sq = square_f(ray->radius);
float dist;
const float *v1, *v2, *r1;
float r2[3], i1[3], i2[3];
Mesh: Move edges to a generic attribute Implements #95966, as the final step of #95965. This commit changes the storage of mesh edge vertex indices from the `MEdge` type to the generic `int2` attribute type. This follows the general design for geometry and the attribute system, where the data storage type and the usage semantics are separated. The main benefit of the change is reduced memory usage-- the requirements of storing mesh edges is reduced by 1/3. For example, this saves 8MB on a 1 million vertex grid. This also gives performance benefits to any memory-bound mesh processing algorithm that uses edges. Another benefit is that all of the edge's vertex indices are contiguous. In a few cases, it's helpful to process all of them as `Span<int>` rather than `Span<int2>`. Similarly, the type is more likely to match a generic format used by a library, or code that shouldn't know about specific Blender `Mesh` types. Various Notes: - The `.edge_verts` name is used to reflect a mapping between domains, similar to `.corner_verts`, etc. The period means that it the data shouldn't change arbitrarily by the user or procedural operations. - `edge[0]` is now used instead of `edge.v1` - Signed integers are used instead of unsigned to reduce the mixing of signed-ness, which can be error prone. - All of the previously used core mesh data types (`MVert`, `MEdge`, `MLoop`, `MPoly` are now deprecated. Only generic types are used). - The `vec2i` DNA type is used in the few C files where necessary. Pull Request: https://projects.blender.org/blender/blender/pulls/106638
2023-04-17 13:47:41 +02:00
v1 = positions[edge[0]];
v2 = positions[edge[1]];
/* In case we get a zero-length edge, handle it as a point! */
if (equals_v3v3(v1, v2)) {
mesh_verts_spherecast_do(index, v1, ray, hit);
return;
}
r1 = ray->origin;
add_v3_v3v3(r2, r1, ray->direction);
if (isect_line_line_v3(v1, v2, r1, r2, i1, i2)) {
/* No hit if intersection point is 'behind' the origin of the ray, or too far away from it. */
if ((dot_v3v3v3(r1, i2, r2) >= 0.0f) && ((dist = len_v3v3(r1, i2)) < hit->dist)) {
const float e_fac = line_point_factor_v3(i1, v1, v2);
if (e_fac < 0.0f) {
copy_v3_v3(i1, v1);
}
else if (e_fac > 1.0f) {
copy_v3_v3(i1, v2);
}
/* Ensure ray is really close enough from edge! */
if (len_squared_v3v3(i1, i2) <= radius_sq) {
hit->index = index;
hit->dist = dist;
copy_v3_v3(hit->co, i2);
}
}
}
}
2015-07-22 20:35:33 +10:00
/** \} */
/*
* BVH builders
*/
/* -------------------------------------------------------------------- */
/** \name Common Utils
* \{ */
static void bvhtree_from_mesh_setup_data(BVHTree *tree,
const BVHCacheType bvh_cache_type,
const Span<float3> positions,
const Span<blender::int2> edges,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const MFace *face,
BVHTreeFromMesh *r_data)
{
*r_data = {};
r_data->tree = tree;
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
r_data->vert_positions = positions;
r_data->edges = edges;
r_data->face = face;
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
r_data->corner_verts = corner_verts;
r_data->corner_tris = corner_tris;
switch (bvh_cache_type) {
case BVHTREE_FROM_VERTS:
case BVHTREE_FROM_LOOSEVERTS:
/* a nullptr nearest callback works fine
* remember the min distance to point is the same as the min distance to BV of point */
r_data->nearest_callback = nullptr;
r_data->raycast_callback = mesh_verts_spherecast;
break;
case BVHTREE_FROM_EDGES:
case BVHTREE_FROM_LOOSEEDGES:
r_data->nearest_callback = mesh_edges_nearest_point;
r_data->raycast_callback = mesh_edges_spherecast;
break;
case BVHTREE_FROM_FACES:
r_data->nearest_callback = mesh_faces_nearest_point;
r_data->raycast_callback = mesh_faces_spherecast;
break;
case BVHTREE_FROM_CORNER_TRIS:
case BVHTREE_FROM_CORNER_TRIS_NO_HIDDEN:
r_data->nearest_callback = mesh_corner_tris_nearest_point;
r_data->raycast_callback = mesh_corner_tris_spherecast;
break;
case BVHTREE_FROM_EM_LOOSEVERTS:
case BVHTREE_FROM_EM_EDGES:
case BVHTREE_FROM_EM_LOOPTRIS:
case BVHTREE_MAX_ITEM:
BLI_assert(false);
break;
}
}
static void bvhtree_from_editmesh_setup_data(BVHTree *tree,
const BVHCacheType bvh_cache_type,
BVHTreeFromEditMesh *r_data)
{
memset(r_data, 0, sizeof(*r_data));
r_data->tree = tree;
switch (bvh_cache_type) {
case BVHTREE_FROM_EM_LOOSEVERTS:
r_data->nearest_callback = nullptr;
r_data->raycast_callback = editmesh_verts_spherecast;
break;
case BVHTREE_FROM_EM_EDGES:
r_data->nearest_callback = nullptr; /* TODO */
r_data->raycast_callback = nullptr; /* TODO */
break;
case BVHTREE_FROM_EM_LOOPTRIS:
r_data->nearest_callback = editmesh_corner_tris_nearest_point;
r_data->raycast_callback = editmesh_corner_tris_spherecast;
break;
case BVHTREE_FROM_VERTS:
case BVHTREE_FROM_LOOSEVERTS:
case BVHTREE_FROM_EDGES:
case BVHTREE_FROM_LOOSEEDGES:
case BVHTREE_FROM_FACES:
case BVHTREE_FROM_CORNER_TRIS:
case BVHTREE_FROM_CORNER_TRIS_NO_HIDDEN:
case BVHTREE_MAX_ITEM:
BLI_assert(false);
break;
}
}
2023-06-16 12:20:40 +10:00
static BVHTree *bvhtree_new_common(
float epsilon, int tree_type, int axis, int elems_num, int &elems_num_active)
{
if (elems_num_active != -1) {
BLI_assert(IN_RANGE_INCL(elems_num_active, 0, elems_num));
}
else {
elems_num_active = elems_num;
}
if (elems_num_active == 0) {
return nullptr;
}
return BLI_bvhtree_new(elems_num_active, epsilon, tree_type, axis);
}
/** \} */
2015-07-22 20:35:33 +10:00
/* -------------------------------------------------------------------- */
/** \name Vertex Builder
* \{ */
static BVHTree *bvhtree_from_editmesh_verts_create_tree(float epsilon,
2015-07-22 20:35:33 +10:00
int tree_type,
int axis,
BMEditMesh *em,
const BitSpan verts_mask,
int verts_num_active)
{
const int verts_num = em->bm->totvert;
2023-06-16 12:20:40 +10:00
BVHTree *tree = bvhtree_new_common(epsilon, tree_type, axis, verts_num, verts_num_active);
if (!tree) {
return nullptr;
}
BM_mesh_elem_table_ensure(em->bm, BM_VERT);
for (int i = 0; i < verts_num; i++) {
if (!verts_mask.is_empty() && !verts_mask[i]) {
continue;
}
BMVert *eve = BM_vert_at_index(em->bm, i);
BLI_bvhtree_insert(tree, i, eve->co, 1);
}
BLI_assert(BLI_bvhtree_get_len(tree) == verts_num_active);
return tree;
}
static BVHTree *bvhtree_from_mesh_verts_create_tree(float epsilon,
int tree_type,
int axis,
const Span<float3> positions,
const BitSpan verts_mask,
int verts_num_active)
{
BVHTree *tree = bvhtree_new_common(epsilon, tree_type, axis, positions.size(), verts_num_active);
if (!tree) {
return nullptr;
}
for (const int i : positions.index_range()) {
if (!verts_mask.is_empty() && !verts_mask[i]) {
continue;
}
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
BLI_bvhtree_insert(tree, i, positions[i], 1);
}
BLI_assert(BLI_bvhtree_get_len(tree) == verts_num_active);
return tree;
}
BVHTree *bvhtree_from_editmesh_verts_ex(BVHTreeFromEditMesh *data,
BMEditMesh *em,
const BitSpan verts_mask,
int verts_num_active,
float epsilon,
int tree_type,
int axis)
{
BVHTree *tree = bvhtree_from_editmesh_verts_create_tree(
epsilon, tree_type, axis, em, verts_mask, verts_num_active);
bvhtree_balance(tree, false);
if (data) {
bvhtree_from_editmesh_setup_data(tree, BVHTREE_FROM_EM_LOOSEVERTS, data);
}
return tree;
}
BVHTree *bvhtree_from_editmesh_verts(
BVHTreeFromEditMesh *data, BMEditMesh *em, float epsilon, int tree_type, int axis)
{
return bvhtree_from_editmesh_verts_ex(data, em, {}, -1, epsilon, tree_type, axis);
}
2015-07-22 20:35:33 +10:00
BVHTree *bvhtree_from_mesh_verts_ex(BVHTreeFromMesh *data,
const Span<float3> vert_positions,
const BitSpan verts_mask,
int verts_num_active,
2015-07-22 20:35:33 +10:00
float epsilon,
int tree_type,
int axis)
{
BVHTree *tree = bvhtree_from_mesh_verts_create_tree(
epsilon, tree_type, axis, vert_positions, verts_mask, verts_num_active);
bvhtree_balance(tree, false);
if (data) {
/* Setup BVHTreeFromMesh */
bvhtree_from_mesh_setup_data(tree, BVHTREE_FROM_VERTS, vert_positions, {}, {}, {}, {}, data);
}
Make bvhutil safe for multi-threaded usage There were couple of reasons why it wasn't safe for usage from multiple threads. First of all, it was trying to cache BVH in derived mesh, which wasn't safe because multiple threads might have requested BVH tree and simultaneous reading and writing to the cache became a big headache. Solved this with RW lock so now access to BVH cache is safe. Another issue is causes by the fact that it's not guaranteed DM to have vert/edge/face arrays pre-allocated and when one was calling functions like getVertDataArray() array could have been allocated and marked as temporary. This is REALLY bad, because NO ONE is ever allowed to modify data which doesn't belong to him. This lead to situations when multiple threads were using BVH tree and they run into race condition with this temporary allocated arrays. Now bvhtree owns allocated arrays and keeps track of them, so no race condition happens with temporary data stored in the derived mesh. This solved threading issues and likely wouldn't introduce noticeable slowdown. Even when DM was keeping track of this arrays, they were re-allocated on every BVH creation anyway, because those arrays were temporary and were freed with dm->release() call. We might re-consider this a bit and make it so BVH trees are allocated when DM itself is being allocated based on the DAG layout, but that i'd consider an optimization and not something we need to do 1st priority. Fixes crash happening with 05_4g_track.blend from Mango after the threaded object update landed to master.
2014-01-10 17:21:39 +06:00
return tree;
}
2015-07-22 20:35:33 +10:00
/** \} */
/* -------------------------------------------------------------------- */
/** \name Edge Builder
* \{ */
static BVHTree *bvhtree_from_editmesh_edges_create_tree(float epsilon,
int tree_type,
int axis,
BMEditMesh *em,
const BitSpan edges_mask,
int edges_num_active)
{
const int edges_num = em->bm->totedge;
2023-06-16 12:20:40 +10:00
BVHTree *tree = bvhtree_new_common(epsilon, tree_type, axis, edges_num, edges_num_active);
if (!tree) {
return nullptr;
}
BM_mesh_elem_table_ensure(em->bm, BM_EDGE);
int i;
BMIter iter;
BMEdge *eed;
BM_ITER_MESH_INDEX (eed, &iter, em->bm, BM_EDGES_OF_MESH, i) {
if (!edges_mask.is_empty() && !edges_mask[i]) {
continue;
}
float co[2][3];
copy_v3_v3(co[0], eed->v1->co);
copy_v3_v3(co[1], eed->v2->co);
BLI_bvhtree_insert(tree, i, co[0], 2);
}
BLI_assert(BLI_bvhtree_get_len(tree) == edges_num_active);
return tree;
}
static BVHTree *bvhtree_from_mesh_edges_create_tree(const Span<float3> positions,
const blender::Span<blender::int2> edges,
const BitSpan edges_mask,
int edges_num_active,
float epsilon,
int tree_type,
int axis)
{
2023-06-16 12:20:40 +10:00
BVHTree *tree = bvhtree_new_common(epsilon, tree_type, axis, edges.size(), edges_num_active);
if (!tree) {
return nullptr;
}
for (const int i : edges.index_range()) {
if (!edges_mask.is_empty() && !edges_mask[i]) {
continue;
}
float co[2][3];
copy_v3_v3(co[0], positions[edges[i][0]]);
copy_v3_v3(co[1], positions[edges[i][1]]);
BLI_bvhtree_insert(tree, i, co[0], 2);
}
return tree;
}
BVHTree *bvhtree_from_editmesh_edges_ex(BVHTreeFromEditMesh *data,
BMEditMesh *em,
const BitSpan edges_mask,
int edges_num_active,
float epsilon,
int tree_type,
int axis)
{
BVHTree *tree = bvhtree_from_editmesh_edges_create_tree(
epsilon, tree_type, axis, em, edges_mask, edges_num_active);
bvhtree_balance(tree, false);
if (data) {
bvhtree_from_editmesh_setup_data(tree, BVHTREE_FROM_EM_EDGES, data);
}
return tree;
}
BVHTree *bvhtree_from_editmesh_edges(
BVHTreeFromEditMesh *data, BMEditMesh *em, float epsilon, int tree_type, int axis)
{
return bvhtree_from_editmesh_edges_ex(data, em, {}, -1, epsilon, tree_type, axis);
}
BVHTree *bvhtree_from_mesh_edges_ex(BVHTreeFromMesh *data,
const Span<float3> vert_positions,
const Span<blender::int2> edges,
const BitSpan edges_mask,
int edges_num_active,
float epsilon,
int tree_type,
int axis)
{
BVHTree *tree = bvhtree_from_mesh_edges_create_tree(
vert_positions, edges, edges_mask, edges_num_active, epsilon, tree_type, axis);
bvhtree_balance(tree, false);
if (data) {
/* Setup BVHTreeFromMesh */
bvhtree_from_mesh_setup_data(
tree, BVHTREE_FROM_EDGES, vert_positions, edges, {}, {}, {}, data);
}
return tree;
}
2015-07-22 20:35:33 +10:00
/** \} */
/* -------------------------------------------------------------------- */
/** \name Tessellated Face Builder
* \{ */
static BVHTree *bvhtree_from_mesh_faces_create_tree(float epsilon,
int tree_type,
int axis,
const Span<float3> positions,
const MFace *face,
const int faces_num,
const BitSpan faces_mask,
int faces_num_active)
{
2023-06-16 12:20:40 +10:00
BVHTree *tree = bvhtree_new_common(epsilon, tree_type, axis, faces_num, faces_num_active);
if (!tree) {
return nullptr;
}
if (!positions.is_empty() && face) {
for (int i = 0; i < faces_num; i++) {
float co[4][3];
if (!faces_mask.is_empty() && !faces_mask[i]) {
continue;
}
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
copy_v3_v3(co[0], positions[face[i].v1]);
copy_v3_v3(co[1], positions[face[i].v2]);
copy_v3_v3(co[2], positions[face[i].v3]);
if (face[i].v4) {
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
copy_v3_v3(co[3], positions[face[i].v4]);
}
BLI_bvhtree_insert(tree, i, co[0], face[i].v4 ? 4 : 3);
}
}
BLI_assert(BLI_bvhtree_get_len(tree) == faces_num_active);
return tree;
}
2015-07-22 20:35:33 +10:00
/** \} */
/* -------------------------------------------------------------------- */
/** \name corner_tri Face Builder
* \{ */
static BVHTree *bvhtree_from_editmesh_corner_tris_create_tree(float epsilon,
int tree_type,
int axis,
BMEditMesh *em,
const BitSpan corner_tris_mask,
int corner_tris_num_active)
{
const int corner_tris_num = em->tottri;
BVHTree *tree = bvhtree_new_common(
epsilon, tree_type, axis, corner_tris_num, corner_tris_num_active);
if (!tree) {
return nullptr;
}
const BMLoop *(*corner_tris)[3] = const_cast<const BMLoop *(*)[3]>(em->looptris);
/* Insert BMesh-tessellation triangles into the BVH-tree, unless they are hidden
* and/or selected. Even if the faces themselves are not selected for the snapped
* transform, having a vertex selected means the face (and thus it's tessellated
* triangles) will be moving and will not be a good snap targets. */
for (int i = 0; i < corner_tris_num; i++) {
const BMLoop **ltri = corner_tris[i];
bool insert = !corner_tris_mask.is_empty() ? corner_tris_mask[i] : true;
if (insert) {
/* No reason found to block hit-testing the triangle for snap, so insert it now. */
float co[3][3];
copy_v3_v3(co[0], ltri[0]->v->co);
copy_v3_v3(co[1], ltri[1]->v->co);
copy_v3_v3(co[2], ltri[2]->v->co);
BLI_bvhtree_insert(tree, i, co[0], 3);
}
}
BLI_assert(BLI_bvhtree_get_len(tree) == corner_tris_num_active);
return tree;
}
static BVHTree *bvhtree_from_mesh_corner_tris_create_tree(float epsilon,
int tree_type,
int axis,
const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const BitSpan corner_tris_mask,
int corner_tris_num_active)
{
if (positions.is_empty()) {
return nullptr;
}
2023-06-16 12:20:40 +10:00
BVHTree *tree = bvhtree_new_common(
epsilon, tree_type, axis, corner_tris.size(), corner_tris_num_active);
if (!tree) {
return nullptr;
}
for (const int i : corner_tris.index_range()) {
float co[3][3];
if (!corner_tris_mask.is_empty() && !corner_tris_mask[i]) {
continue;
}
copy_v3_v3(co[0], positions[corner_verts[corner_tris[i][0]]]);
copy_v3_v3(co[1], positions[corner_verts[corner_tris[i][1]]]);
copy_v3_v3(co[2], positions[corner_verts[corner_tris[i][2]]]);
BLI_bvhtree_insert(tree, i, co[0], 3);
}
BLI_assert(BLI_bvhtree_get_len(tree) == corner_tris_num_active);
return tree;
}
BVHTree *bvhtree_from_editmesh_looptris_ex(BVHTreeFromEditMesh *data,
BMEditMesh *em,
const BitSpan corner_tris_mask,
int corner_tris_num_active,
float epsilon,
int tree_type,
int axis)
{
/* BMESH specific check that we have tessfaces,
* we _could_ tessellate here but rather not - campbell */
BVHTree *tree = bvhtree_from_editmesh_corner_tris_create_tree(
epsilon, tree_type, axis, em, corner_tris_mask, corner_tris_num_active);
bvhtree_balance(tree, false);
if (data) {
bvhtree_from_editmesh_setup_data(tree, BVHTREE_FROM_EM_LOOPTRIS, data);
}
return tree;
}
BVHTree *bvhtree_from_editmesh_corner_tris(
BVHTreeFromEditMesh *data, BMEditMesh *em, float epsilon, int tree_type, int axis)
{
return bvhtree_from_editmesh_looptris_ex(data, em, {}, -1, epsilon, tree_type, axis);
}
BVHTree *bvhtree_from_mesh_corner_tris_ex(BVHTreeFromMesh *data,
const Span<float3> vert_positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const BitSpan corner_tris_mask,
int corner_tris_num_active,
float epsilon,
int tree_type,
int axis)
{
BVHTree *tree = bvhtree_from_mesh_corner_tris_create_tree(epsilon,
tree_type,
axis,
vert_positions,
corner_verts,
corner_tris,
corner_tris_mask,
corner_tris_num_active);
bvhtree_balance(tree, false);
if (data) {
/* Setup BVHTreeFromMesh */
bvhtree_from_mesh_setup_data(tree,
BVHTREE_FROM_CORNER_TRIS,
vert_positions,
{},
corner_verts,
corner_tris,
nullptr,
data);
}
return tree;
}
static BitVector<> corner_tris_no_hidden_map_get(const blender::OffsetIndices<int> faces,
const VArray<bool> &hide_poly,
const int corner_tris_len,
int *r_corner_tris_active_len)
{
if (hide_poly.is_single() && !hide_poly.get_internal_single()) {
return {};
}
BitVector<> corner_tris_mask(corner_tris_len);
int corner_tris_no_hidden_len = 0;
int tri_index = 0;
for (const int64_t i : faces.index_range()) {
const int triangles_num = blender::bke::mesh::face_triangles_num(faces[i].size());
if (hide_poly[i]) {
tri_index += triangles_num;
}
else {
for (const int i : IndexRange(triangles_num)) {
2022-11-15 16:41:50 +11:00
UNUSED_VARS(i);
corner_tris_mask[tri_index].set();
tri_index++;
corner_tris_no_hidden_len++;
}
}
}
*r_corner_tris_active_len = corner_tris_no_hidden_len;
return corner_tris_mask;
}
BVHTree *BKE_bvhtree_from_mesh_get(BVHTreeFromMesh *data,
const Mesh *mesh,
2020-05-15 13:54:22 +02:00
const BVHCacheType bvh_cache_type,
const int tree_type)
{
using namespace blender;
using namespace blender::bke;
BVHCache **bvh_cache_p = (BVHCache **)&mesh->runtime->bvh_cache;
Span<int3> corner_tris;
if (ELEM(bvh_cache_type, BVHTREE_FROM_CORNER_TRIS, BVHTREE_FROM_CORNER_TRIS_NO_HIDDEN)) {
corner_tris = mesh->corner_tris();
}
const Span<float3> positions = mesh->vert_positions();
const Span<int2> edges = mesh->edges();
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
const Span<int> corner_verts = mesh->corner_verts();
/* Setup BVHTreeFromMesh */
bvhtree_from_mesh_setup_data(nullptr,
bvh_cache_type,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
positions,
edges,
corner_verts,
corner_tris,
(const MFace *)CustomData_get_layer(&mesh->fdata_legacy, CD_MFACE),
data);
bool lock_started = false;
data->cached = bvhcache_find(
bvh_cache_p, bvh_cache_type, &data->tree, &lock_started, &mesh->runtime->eval_mutex);
if (data->cached) {
BLI_assert(lock_started == false);
/* NOTE: #data->tree can be nullptr. */
return data->tree;
}
/* Create BVHTree. */
switch (bvh_cache_type) {
case BVHTREE_FROM_LOOSEVERTS: {
const LooseVertCache &loose_verts = mesh->loose_verts();
data->tree = bvhtree_from_mesh_verts_create_tree(
0.0f, tree_type, 6, positions, loose_verts.is_loose_bits, loose_verts.count);
break;
}
case BVHTREE_FROM_VERTS: {
data->tree = bvhtree_from_mesh_verts_create_tree(0.0f, tree_type, 6, positions, {}, -1);
break;
}
case BVHTREE_FROM_LOOSEEDGES: {
const LooseEdgeCache &loose_edges = mesh->loose_edges();
data->tree = bvhtree_from_mesh_edges_create_tree(
positions, edges, loose_edges.is_loose_bits, loose_edges.count, 0.0f, tree_type, 6);
break;
}
case BVHTREE_FROM_EDGES: {
data->tree = bvhtree_from_mesh_edges_create_tree(
positions, edges, {}, -1, 0.0f, tree_type, 6);
break;
}
case BVHTREE_FROM_FACES: {
BLI_assert(!(mesh->totface_legacy == 0 && mesh->faces_num != 0));
data->tree = bvhtree_from_mesh_faces_create_tree(
Mesh: Remove redundant custom data pointers For copy-on-write, we want to share attribute arrays between meshes where possible. Mutable pointers like `Mesh.mvert` make that difficult by making ownership vague. They also make code more complex by adding redundancy. The simplest solution is just removing them and retrieving layers from `CustomData` as needed. Similar changes have already been applied to curves and point clouds (e9f82d3dc7ee, 410a6efb747f). Removing use of the pointers generally makes code more obvious and more reusable. Mesh data is now accessed with a C++ API (`Mesh::edges()` or `Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`). The CoW changes this commit makes possible are described in T95845 and T95842, and started in D14139 and D14140. The change also simplifies the ongoing mesh struct-of-array refactors from T95965. **RNA/Python Access Performance** Theoretically, accessing mesh elements with the RNA API may become slower, since the layer needs to be found on every random access. However, overhead is already high enough that this doesn't make a noticible differenc, and performance is actually improved in some cases. Random access can be up to 10% faster, but other situations might be a bit slower. Generally using `foreach_get/set` are the best way to improve performance. See the differential revision for more discussion about Python performance. Cycles has been updated to use raw pointers and the internal Blender mesh types, mostly because there is no sense in having this overhead when it's already compiled with Blender. In my tests this roughly halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million face grid). Differential Revision: https://developer.blender.org/D15488
2022-09-05 11:56:34 -05:00
0.0f,
tree_type,
6,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
positions,
(const MFace *)CustomData_get_layer(&mesh->fdata_legacy, CD_MFACE),
mesh->totface_legacy,
{},
Mesh: Remove redundant custom data pointers For copy-on-write, we want to share attribute arrays between meshes where possible. Mutable pointers like `Mesh.mvert` make that difficult by making ownership vague. They also make code more complex by adding redundancy. The simplest solution is just removing them and retrieving layers from `CustomData` as needed. Similar changes have already been applied to curves and point clouds (e9f82d3dc7ee, 410a6efb747f). Removing use of the pointers generally makes code more obvious and more reusable. Mesh data is now accessed with a C++ API (`Mesh::edges()` or `Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`). The CoW changes this commit makes possible are described in T95845 and T95842, and started in D14139 and D14140. The change also simplifies the ongoing mesh struct-of-array refactors from T95965. **RNA/Python Access Performance** Theoretically, accessing mesh elements with the RNA API may become slower, since the layer needs to be found on every random access. However, overhead is already high enough that this doesn't make a noticible differenc, and performance is actually improved in some cases. Random access can be up to 10% faster, but other situations might be a bit slower. Generally using `foreach_get/set` are the best way to improve performance. See the differential revision for more discussion about Python performance. Cycles has been updated to use raw pointers and the internal Blender mesh types, mostly because there is no sense in having this overhead when it's already compiled with Blender. In my tests this roughly halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million face grid). Differential Revision: https://developer.blender.org/D15488
2022-09-05 11:56:34 -05:00
-1);
break;
}
case BVHTREE_FROM_CORNER_TRIS_NO_HIDDEN: {
AttributeAccessor attributes = mesh->attributes();
int mask_bits_act_len = -1;
const BitVector<> mask = corner_tris_no_hidden_map_get(
mesh->faces(),
*attributes.lookup_or_default(".hide_poly", AttrDomain::Face, false),
corner_tris.size(),
&mask_bits_act_len);
data->tree = bvhtree_from_mesh_corner_tris_create_tree(
0.0f, tree_type, 6, positions, corner_verts, corner_tris, mask, mask_bits_act_len);
break;
}
case BVHTREE_FROM_CORNER_TRIS: {
data->tree = bvhtree_from_mesh_corner_tris_create_tree(
0.0f, tree_type, 6, positions, corner_verts, corner_tris, {}, -1);
break;
}
case BVHTREE_FROM_EM_LOOSEVERTS:
case BVHTREE_FROM_EM_EDGES:
case BVHTREE_FROM_EM_LOOPTRIS:
case BVHTREE_MAX_ITEM:
BLI_assert_unreachable();
break;
}
bvhtree_balance(data->tree, lock_started);
/* Save on cache for later use */
// printf("BVHTree built and saved on cache\n");
BLI_assert(data->cached == false);
data->cached = true;
bvhcache_insert(*bvh_cache_p, data->tree, bvh_cache_type);
bvhcache_unlock(*bvh_cache_p, lock_started);
#ifndef NDEBUG
if (data->tree != nullptr) {
if (BLI_bvhtree_get_tree_type(data->tree) != tree_type) {
printf("tree_type %d obtained instead of %d\n",
BLI_bvhtree_get_tree_type(data->tree),
tree_type);
}
}
#endif
Refactor: Move normals out of MVert, lazy calculation As described in T91186, this commit moves mesh vertex normals into a contiguous array of float vectors in a custom data layer, how face normals are currently stored. The main interface is documented in `BKE_mesh.h`. Vertex and face normals are now calculated on-demand and cached, retrieved with an "ensure" function. Since the logical state of a mesh is now "has normals when necessary", they can be retrieved from a `const` mesh. The goal is to use on-demand calculation for all derived data, but leave room for eager calculation for performance purposes (modifier evaluation is threaded, but viewport data generation is not). **Benefits** This moves us closer to a SoA approach rather than the current AoS paradigm. Accessing a contiguous `float3` is much more efficient than retrieving data from a larger struct. The memory requirements for accessing only normals or vertex locations are smaller, and at the cost of more memory usage for just normals, they now don't have to be converted between float and short, which also simplifies code In the future, the remaining items can be removed from `MVert`, leaving only `float3`, which has similar benefits (see T93602). Removing the combination of derived and original data makes it conceptually simpler to only calculate normals when necessary. This is especially important now that we have more opportunities for temporary meshes in geometry nodes. **Performance** In addition to the theoretical future performance improvements by making `MVert == float3`, I've done some basic performance testing on this patch directly. The data is fairly rough, but it gives an idea about where things stand generally. - Mesh line primitive 4m Verts: 1.16x faster (36 -> 31 ms), showing that accessing just `MVert` is now more efficient. - Spring Splash Screen: 1.03-1.06 -> 1.06-1.11 FPS, a very slight change that at least shows there is no regression. - Sprite Fright Snail Smoosh: 3.30-3.40 -> 3.42-3.50 FPS, a small but observable speedup. - Set Position Node with Scaled Normal: 1.36x faster (53 -> 39 ms), shows that using normals in geometry nodes is faster. - Normal Calculation 1.6m Vert Cube: 1.19x faster (25 -> 21 ms), shows that calculating normals is slightly faster now. - File Size of 1.6m Vert Cube: 1.03x smaller (214.7 -> 208.4 MB), Normals are not saved in files, which can help with large meshes. As for memory usage, it may be slightly more in some cases, but I didn't observe any difference in the production files I tested. **Tests** Some modifiers and cycles test results need to be updated with this commit, for two reasons: - The subdivision surface modifier is not responsible for calculating normals anymore. In master, the modifier creates different normals than the result of the `Mesh` normal calculation, so this is a bug fix. - There are small differences in the results of some modifiers that use normals because they are not converted to and from `short` anymore. **Future improvements** - Remove `ModifierTypeInfo::dependsOnNormals`. Code in each modifier already retrieves normals if they are needed anyway. - Copy normals as part of a better CoW system for attributes. - Make more areas use lazy instead of eager normal calculation. - Remove `BKE_mesh_normals_tag_dirty` in more places since that is now the default state of a new mesh. - Possibly apply a similar change to derived face corner normals. Differential Revision: https://developer.blender.org/D12770
2022-01-13 14:37:58 -06:00
return data->tree;
}
static BitVector<> bmverts_loose_map_get(BMesh *bm, int *r_bmvert_active_len)
{
BitVector<> bmvert_mask(bm->totvert);
int i, bmvert_loose_len = 0;
BMIter iter;
BMVert *v;
BM_ITER_MESH_INDEX (v, &iter, bm, BM_VERTS_OF_MESH, i) {
if (v->e == nullptr) {
bmvert_mask[i].set();
bmvert_loose_len++;
}
}
*r_bmvert_active_len = bmvert_loose_len;
return bmvert_mask;
}
BVHTree *BKE_bvhtree_from_editmesh_get(BVHTreeFromEditMesh *data,
BMEditMesh *em,
const int tree_type,
2020-05-15 13:54:22 +02:00
const BVHCacheType bvh_cache_type,
BVHCache **bvh_cache_p,
std::mutex *mesh_eval_mutex)
{
bool lock_started = false;
bvhtree_from_editmesh_setup_data(nullptr, bvh_cache_type, data);
if (bvh_cache_p) {
data->cached = bvhcache_find(
bvh_cache_p, bvh_cache_type, &data->tree, &lock_started, mesh_eval_mutex);
if (data->cached) {
BLI_assert(lock_started == false);
return data->tree;
}
}
switch (bvh_cache_type) {
case BVHTREE_FROM_EM_LOOSEVERTS: {
int mask_bits_act_len = -1;
const BitVector<> mask = bmverts_loose_map_get(em->bm, &mask_bits_act_len);
data->tree = bvhtree_from_editmesh_verts_create_tree(
0.0f, tree_type, 6, em, mask, mask_bits_act_len);
break;
}
case BVHTREE_FROM_EM_EDGES:
data->tree = bvhtree_from_editmesh_edges_create_tree(0.0f, tree_type, 6, em, {}, -1);
break;
case BVHTREE_FROM_EM_LOOPTRIS:
data->tree = bvhtree_from_editmesh_corner_tris_create_tree(0.0f, tree_type, 6, em, {}, -1);
break;
case BVHTREE_FROM_VERTS:
case BVHTREE_FROM_EDGES:
case BVHTREE_FROM_FACES:
case BVHTREE_FROM_CORNER_TRIS:
case BVHTREE_FROM_CORNER_TRIS_NO_HIDDEN:
case BVHTREE_FROM_LOOSEVERTS:
case BVHTREE_FROM_LOOSEEDGES:
case BVHTREE_MAX_ITEM:
BLI_assert(false);
break;
}
bvhtree_balance(data->tree, lock_started);
if (bvh_cache_p) {
/* Save on cache for later use */
// printf("BVHTree built and saved on cache\n");
BLI_assert(data->cached == false);
data->cached = true;
bvhcache_insert(*bvh_cache_p, data->tree, bvh_cache_type);
bvhcache_unlock(*bvh_cache_p, lock_started);
}
#ifndef NDEBUG
if (data->tree != nullptr) {
if (BLI_bvhtree_get_tree_type(data->tree) != tree_type) {
printf("tree_type %d obtained instead of %d\n",
BLI_bvhtree_get_tree_type(data->tree),
tree_type);
}
}
#endif
return data->tree;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Free Functions
* \{ */
void free_bvhtree_from_editmesh(BVHTreeFromEditMesh *data)
{
if (data->tree) {
if (!data->cached) {
BLI_bvhtree_free(data->tree);
}
memset(data, 0, sizeof(*data));
}
}
void free_bvhtree_from_mesh(BVHTreeFromMesh *data)
{
if (data->tree && !data->cached) {
BLI_bvhtree_free(data->tree);
}
*data = {};
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Point Cloud BVH Building
* \{ */
[[nodiscard]] BVHTree *BKE_bvhtree_from_pointcloud_get(BVHTreeFromPointCloud *data,
const PointCloud *pointcloud,
const int tree_type)
{
int tot_point = pointcloud->totpoint;
2023-06-16 12:20:40 +10:00
BVHTree *tree = bvhtree_new_common(0.0f, tree_type, 6, tot_point, tot_point);
if (!tree) {
return nullptr;
}
const Span<float3> positions = pointcloud->positions();
for (const int i : positions.index_range()) {
BLI_bvhtree_insert(tree, i, positions[i], 1);
}
BLI_assert(BLI_bvhtree_get_len(tree) == tot_point);
bvhtree_balance(tree, false);
data->coords = (const float(*)[3])positions.data();
data->tree = tree;
data->nearest_callback = nullptr;
return tree;
}
void free_bvhtree_from_pointcloud(BVHTreeFromPointCloud *data)
{
if (data->tree) {
BLI_bvhtree_free(data->tree);
}
memset(data, 0, sizeof(*data));
}
/** \} */