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

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

1276 lines
37 KiB
C++
Raw Normal View History

/* SPDX-FileCopyrightText: 2011-2024 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup bke
2011-02-27 20:40:57 +00:00
*/
#include <algorithm>
#include <climits>
#include <cstdio>
#include <cstring>
#include "CLG_log.h"
#include "DNA_mesh_types.h"
#include "DNA_meshdata_types.h"
#include "DNA_object_types.h"
#include "BLI_map.hh"
#include "BLI_math_base.h"
#include "BLI_math_vector.h"
#include "BLI_ordered_edge.hh"
#include "BLI_sort.hh"
#include "BLI_sys_types.h"
#include "BLI_utildefines.h"
#include "BLI_vector_set.hh"
#include "BKE_attribute.hh"
#include "BKE_customdata.hh"
2024-01-29 18:57:16 -05:00
#include "BKE_deform.hh"
#include "BKE_mesh.hh"
#include "BKE_mesh_runtime.hh"
#include "DEG_depsgraph.hh"
#include "MEM_guardedalloc.h"
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;
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::MutableSpan;
using blender::Span;
/* corner v/e are unsigned, so using max uint_32 value as invalid marker... */
#define INVALID_CORNER_EDGE_MARKER 4294967295u
static CLG_LogRef LOG = {"bke.mesh"};
void strip_loose_faces_corners(Mesh *mesh, blender::BitSpan faces_to_remove);
void mesh_strip_edges(Mesh *mesh);
/* -------------------------------------------------------------------- */
/** \name Internal functions
* \{ */
union EdgeUUID {
uint32_t verts[2];
int64_t edval;
};
struct SortFaceLegacy {
EdgeUUID es[4];
uint index;
};
/* Used to detect faces using exactly the same vertices. */
/* Used to detect corners used by no (disjoint) or more than one (intersect) faces. */
struct SortFace {
int *verts = nullptr;
int numverts = 0;
int corner_start = 0;
uint index = 0;
bool invalid = false;
};
static void edge_store_assign(uint32_t verts[2], const uint32_t v1, const uint32_t v2)
{
if (v1 < v2) {
verts[0] = v1;
verts[1] = v2;
}
else {
verts[0] = v2;
verts[1] = v1;
}
}
static void edge_store_from_mface_quad(EdgeUUID es[4], const MFace *mf)
{
edge_store_assign(es[0].verts, mf->v1, mf->v2);
edge_store_assign(es[1].verts, mf->v2, mf->v3);
edge_store_assign(es[2].verts, mf->v3, mf->v4);
edge_store_assign(es[3].verts, mf->v4, mf->v1);
}
static void edge_store_from_mface_tri(EdgeUUID es[4], const MFace *mf)
{
edge_store_assign(es[0].verts, mf->v1, mf->v2);
edge_store_assign(es[1].verts, mf->v2, mf->v3);
edge_store_assign(es[2].verts, mf->v3, mf->v1);
es[3].verts[0] = es[3].verts[1] = UINT_MAX;
}
static bool search_legacy_face_cmp(const SortFaceLegacy &sfa, const SortFaceLegacy &sfb)
{
if (sfa.es[0].edval != sfb.es[0].edval) {
return sfa.es[0].edval < sfb.es[0].edval;
}
if (sfa.es[1].edval != sfb.es[1].edval) {
return sfa.es[1].edval < sfb.es[1].edval;
}
if (sfa.es[2].edval != sfb.es[2].edval) {
return sfa.es[2].edval < sfb.es[2].edval;
}
return sfa.es[3].edval < sfb.es[3].edval;
}
static bool search_face_cmp(const SortFace &sp1, const SortFace &sp2)
{
/* Reject all invalid faces at end of list! */
if (sp1.invalid || sp2.invalid) {
return sp1.invalid < sp2.invalid;
}
/* Else, sort on first non-equal verts (remember verts of valid faces are sorted). */
const int max_idx = std::min(sp1.numverts, sp2.numverts);
2020-09-22 10:08:02 +02:00
for (int idx = 0; idx < max_idx; idx++) {
const int v1_i = sp1.verts[idx];
const int v2_i = sp2.verts[idx];
2015-11-23 11:27:02 +11:00
if (v1_i != v2_i) {
return v1_i < v2_i;
}
}
return sp1.numverts < sp2.numverts;
}
static bool search_face_corner_cmp(const SortFace &sp1, const SortFace &sp2)
{
/* Reject all invalid faces at end of list! */
if (sp1.invalid || sp2.invalid) {
return sp1.invalid < sp2.invalid;
}
/* Else, sort on corner start. */
return sp1.corner_start < sp2.corner_start;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Mesh Validation
* \{ */
#define PRINT_MSG(...) \
if (do_verbose) { \
CLOG_INFO(&LOG, 1, __VA_ARGS__); \
} \
((void)0)
#define PRINT_ERR(...) \
do { \
is_valid = false; \
if (do_verbose) { \
CLOG_ERROR(&LOG, __VA_ARGS__); \
} \
} while (0)
/* NOLINTNEXTLINE: readability-function-size */
2018-07-13 08:36:10 +02:00
bool BKE_mesh_validate_arrays(Mesh *mesh,
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
float (*vert_positions)[3],
uint verts_num,
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
blender::int2 *edges,
uint edges_num,
MFace *legacy_faces,
uint legacy_faces_num,
const int *corner_verts,
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
int *corner_edges,
uint corners_num,
2024-01-19 12:08:40 -05:00
const int *face_offsets,
uint faces_num,
MDeformVert *dverts, /* assume verts_num length */
2018-07-13 08:36:10 +02:00
const bool do_verbose,
const bool do_fixes,
bool *r_changed)
{
using namespace blender;
using namespace blender::bke;
2019-01-03 15:52:07 +11:00
#define REMOVE_EDGE_TAG(_me) \
{ \
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
_me[0] = _me[1]; \
2019-01-03 15:52:07 +11:00
free_flag.edges = do_fixes; \
} \
(void)0
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
#define IS_REMOVED_EDGE(_me) (_me[0] == _me[1])
#define REMOVE_CORNER_TAG(corner) \
2019-01-03 15:52:07 +11:00
{ \
corner_edges[corner] = INVALID_CORNER_EDGE_MARKER; \
free_flag.face_corners = do_fixes; \
2019-01-03 15:52:07 +11:00
} \
(void)0
blender::BitVector<> faces_to_remove(faces_num);
blender::bke::AttributeWriter<int> material_indices =
mesh->attributes_for_write().lookup_for_write<int>("material_index");
blender::MutableVArraySpan<int> material_indices_span(material_indices.varray);
uint i, j;
int *v;
bool is_valid = true;
union {
struct {
int verts : 1;
int verts_weight : 1;
int corners_edge : 1;
};
int as_flag;
} fix_flag;
union {
struct {
int edges : 1;
int faces : 1;
/* This regroups corners and faces! */
int face_corners : 1;
int mselect : 1;
};
int as_flag;
} free_flag;
union {
struct {
int edges : 1;
};
int as_flag;
} recalc_flag;
Map<OrderedEdge, int> edge_hash;
edge_hash.reserve(edges_num);
BLI_assert(!(do_fixes && mesh == nullptr));
fix_flag.as_flag = 0;
free_flag.as_flag = 0;
recalc_flag.as_flag = 0;
PRINT_MSG("verts(%u), edges(%u), corners(%u), faces(%u)",
verts_num,
edges_num,
corners_num,
faces_num);
if (edges_num == 0 && faces_num != 0) {
PRINT_ERR("\tLogical error, %u faces and 0 edges", faces_num);
recalc_flag.edges = do_fixes;
}
for (i = 0; i < verts_num; i++) {
for (j = 0; j < 3; j++) {
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
if (!isfinite(vert_positions[i][j])) {
PRINT_ERR("\tVertex %u: has invalid coordinate", i);
if (do_fixes) {
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
zero_v3(vert_positions[i]);
fix_flag.verts = true;
}
}
}
}
for (i = 0; i < edges_num; i++) {
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
blender::int2 &edge = edges[i];
2014-04-11 11:25:41 +10:00
bool remove = false;
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
if (edge[0] == edge[1]) {
PRINT_ERR("\tEdge %u: has matching verts, both %d", i, edge[0]);
remove = do_fixes;
}
if (edge[0] >= verts_num) {
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
PRINT_ERR("\tEdge %u: v1 index out of range, %d", i, edge[0]);
remove = do_fixes;
}
if (edge[1] >= verts_num) {
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
PRINT_ERR("\tEdge %u: v2 index out of range, %d", i, edge[1]);
remove = do_fixes;
}
if ((edge[0] != edge[1]) && edge_hash.contains(edge)) {
PRINT_ERR("\tEdge %u: is a duplicate of %d", i, edge_hash.lookup(edge));
remove = do_fixes;
}
if (remove == false) {
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
if (edge[0] != edge[1]) {
edge_hash.add(edge, i);
}
}
else {
REMOVE_EDGE_TAG(edge);
}
}
if (legacy_faces && !face_offsets) {
#define REMOVE_FACE_TAG(_mf) \
{ \
_mf->v3 = 0; \
free_flag.faces = do_fixes; \
} \
(void)0
#define CHECK_FACE_VERT_INDEX(a, b) \
if (mf->a == mf->b) { \
PRINT_ERR(" face %u: verts invalid, " STRINGIFY(a) "/" STRINGIFY(b) " both %u", i, mf->a); \
remove = do_fixes; \
} \
(void)0
#define CHECK_FACE_EDGE(a, b) \
if (!edge_hash.contains({mf->a, mf->b})) { \
PRINT_ERR(" face %u: edge " STRINGIFY(a) "/" STRINGIFY(b) " (%u,%u) is missing edge data", \
i, \
mf->a, \
mf->b); \
recalc_flag.edges = do_fixes; \
} \
(void)0
MFace *mf;
const MFace *mf_prev;
Array<SortFaceLegacy> sort_faces(legacy_faces_num);
SortFaceLegacy *sf;
SortFaceLegacy *sf_prev;
uint totsortface = 0;
PRINT_ERR("No faces, only tessellated Faces");
for (i = 0, mf = legacy_faces, sf = sort_faces.data(); i < legacy_faces_num; i++, mf++) {
2014-04-11 11:25:41 +10:00
bool remove = false;
int fidx;
uint fv[4];
fidx = mf->v4 ? 3 : 2;
do {
fv[fidx] = *(&(mf->v1) + fidx);
if (fv[fidx] >= verts_num) {
PRINT_ERR("\tFace %u: 'v%d' index out of range, %u", i, fidx + 1, fv[fidx]);
remove = do_fixes;
}
} while (fidx--);
if (remove == false) {
if (mf->v4) {
CHECK_FACE_VERT_INDEX(v1, v2);
CHECK_FACE_VERT_INDEX(v1, v3);
CHECK_FACE_VERT_INDEX(v1, v4);
CHECK_FACE_VERT_INDEX(v2, v3);
CHECK_FACE_VERT_INDEX(v2, v4);
CHECK_FACE_VERT_INDEX(v3, v4);
}
else {
CHECK_FACE_VERT_INDEX(v1, v2);
CHECK_FACE_VERT_INDEX(v1, v3);
CHECK_FACE_VERT_INDEX(v2, v3);
}
if (remove == false) {
if (edges_num) {
if (mf->v4) {
CHECK_FACE_EDGE(v1, v2);
CHECK_FACE_EDGE(v2, v3);
CHECK_FACE_EDGE(v3, v4);
CHECK_FACE_EDGE(v4, v1);
}
else {
CHECK_FACE_EDGE(v1, v2);
CHECK_FACE_EDGE(v2, v3);
CHECK_FACE_EDGE(v3, v1);
}
}
sf->index = i;
if (mf->v4) {
edge_store_from_mface_quad(sf->es, mf);
std::sort(sf->es, sf->es + 4, [](const EdgeUUID &a, const EdgeUUID &b) {
return a.edval < b.edval;
});
}
else {
edge_store_from_mface_tri(sf->es, mf);
std::sort(sf->es, sf->es + 3, [](const EdgeUUID &a, const EdgeUUID &b) {
return a.edval < b.edval;
});
}
totsortface++;
sf++;
}
}
if (remove) {
REMOVE_FACE_TAG(mf);
}
}
blender::parallel_sort(sort_faces.begin(), sort_faces.end(), search_legacy_face_cmp);
sf = sort_faces.data();
sf_prev = sf;
sf++;
for (i = 1; i < totsortface; i++, sf++) {
2014-04-11 11:25:41 +10:00
bool remove = false;
/* on a valid mesh, code below will never run */
if (memcmp(sf->es, sf_prev->es, sizeof(sf_prev->es)) == 0) {
mf = legacy_faces + sf->index;
if (do_verbose) {
mf_prev = legacy_faces + sf_prev->index;
if (mf->v4) {
PRINT_ERR("\tFace %u & %u: are duplicates (%u,%u,%u,%u) (%u,%u,%u,%u)",
sf->index,
sf_prev->index,
mf->v1,
mf->v2,
mf->v3,
mf->v4,
mf_prev->v1,
mf_prev->v2,
mf_prev->v3,
mf_prev->v4);
}
else {
PRINT_ERR("\tFace %u & %u: are duplicates (%u,%u,%u) (%u,%u,%u)",
sf->index,
sf_prev->index,
mf->v1,
mf->v2,
mf->v3,
mf_prev->v1,
mf_prev->v2,
mf_prev->v3);
}
}
remove = do_fixes;
}
else {
sf_prev = sf;
}
if (remove) {
REMOVE_FACE_TAG(mf);
}
}
#undef REMOVE_FACE_TAG
#undef CHECK_FACE_VERT_INDEX
#undef CHECK_FACE_EDGE
}
/* Checking corners and faces is a bit tricky, as they are quite intricate...
*
* Faces must have:
* - a valid corner_start value.
* - a valid corners_num value (>= 3 and corner_start+corners_num < mesh.corners_num).
*
* corners must have:
* - a valid v value.
* - a valid e value (corresponding to the edge it defines with the next corner in face).
*
* Also, corners not used by faces can be discarded.
* And "intersecting" corners (i.e. corners used by more than one face) are invalid,
* so be sure to leave at most one face per corner!
*/
{
BitVector<> vert_tag(mesh->verts_num);
Array<SortFace> sort_faces(faces_num);
Array<int> sort_face_verts(faces_num == 0 ? 0 : face_offsets[faces_num]);
int64_t sort_face_verts_offset = 0;
for (const int64_t i : blender::IndexRange(faces_num)) {
SortFace *sp = &sort_faces[i];
const int face_start = face_offsets[i];
const int face_size = face_offsets[i + 1] - face_start;
sp->index = i;
/* Material index, isolated from other tests here. While large indices are clamped,
* negative indices aren't supported by drawing, exporters etc.
* To check the indices are in range, use #BKE_mesh_validate_material_indices */
if (material_indices && material_indices_span[i] < 0) {
PRINT_ERR("\tFace %u has invalid material (%d)", sp->index, material_indices_span[i]);
if (do_fixes) {
material_indices_span[i] = 0;
}
}
if (face_start < 0 || face_size < 3) {
/* Invalid corner data. */
PRINT_ERR("\tFace %u is invalid (corner_start: %d, corners_num: %d)",
sp->index,
face_start,
face_size);
sp->invalid = true;
}
else if (face_start + face_size > corners_num) {
/* Invalid corner data. */
PRINT_ERR(
"\tFace %u uses corners out of range "
"(corner_start: %d, corner_end: %d, max number of corners: %u)",
sp->index,
face_start,
face_start + face_size - 1,
corners_num - 1);
sp->invalid = true;
}
else {
/* Face itself is valid, for now. */
int v1, v2; /* v1 is prev corner vert idx, v2 is current corner one. */
sp->invalid = false;
sp->verts = v = sort_face_verts.data() + sort_face_verts_offset;
sort_face_verts_offset += face_size;
sp->numverts = face_size;
sp->corner_start = face_start;
/* Ideally we would only have to do that once on all vertices
* before we start checking each face, but several faces can use same vert,
* so we have to ensure here all verts of current face are cleared. */
for (j = 0; j < face_size; j++) {
const int vert = corner_verts[sp->corner_start + j];
if (vert < verts_num) {
vert_tag[vert].reset();
}
}
/* Test all face's corners' vert idx. */
for (j = 0; j < face_size; j++, v++) {
const int vert = corner_verts[sp->corner_start + j];
if (vert >= verts_num) {
/* Invalid vert idx. */
PRINT_ERR("\tCorner %u has invalid vert reference (%d)", sp->corner_start + j, vert);
sp->invalid = true;
}
else if (vert_tag[vert].test()) {
PRINT_ERR("\tFace %u has duplicated vert reference at corner (%u)", uint(i), j);
sp->invalid = true;
}
2014-08-19 21:09:10 +10:00
else {
vert_tag[vert].set();
2014-08-19 21:09:10 +10:00
}
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
*v = vert;
}
if (sp->invalid) {
sp++;
continue;
}
/* Test all face's corners. */
for (j = 0; j < face_size; j++) {
const int corner = sp->corner_start + j;
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 int vert = corner_verts[corner];
const int edge_i = corner_edges[corner];
v1 = vert;
v2 = corner_verts[sp->corner_start + (j + 1) % face_size];
if (!edge_hash.contains({v1, v2})) {
/* Edge not existing. */
PRINT_ERR("\tFace %u needs missing edge (%d, %d)", sp->index, v1, v2);
if (do_fixes) {
recalc_flag.edges = true;
}
else {
sp->invalid = true;
}
}
else if (edge_i >= edges_num) {
/* Invalid edge idx.
* We already know from previous text that a valid edge exists, use it (if allowed)! */
if (do_fixes) {
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
int prev_e = edge_i;
corner_edges[corner] = edge_hash.lookup({v1, v2});
fix_flag.corners_edge = true;
PRINT_ERR("\tCorner %d has invalid edge reference (%d), fixed using edge %d",
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
corner,
prev_e,
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
corner_edges[corner]);
}
else {
PRINT_ERR("\tCorner %d has invalid edge reference (%d)", corner, edge_i);
sp->invalid = true;
}
}
else {
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
const blender::int2 &edge = edges[edge_i];
if (IS_REMOVED_EDGE(edge) ||
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
!((edge[0] == v1 && edge[1] == v2) || (edge[0] == v2 && edge[1] == v1)))
{
/* The pointed edge is invalid (tagged as removed, or vert idx mismatch),
* and we already know from previous test that a valid one exists,
* use it (if allowed)! */
if (do_fixes) {
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
int prev_e = edge_i;
corner_edges[corner] = edge_hash.lookup({v1, v2});
fix_flag.corners_edge = true;
PRINT_ERR(
"\tFace %u has invalid edge reference (%d, is_removed: %d), fixed using edge "
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
"%d",
sp->index,
prev_e,
IS_REMOVED_EDGE(edge),
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
corner_edges[corner]);
}
else {
PRINT_ERR("\tFace %u has invalid edge reference (%d)", sp->index, edge_i);
sp->invalid = true;
}
}
}
}
if (!sp->invalid) {
/* Needed for checking faces using same verts below. */
std::sort(sp->verts, sp->verts + sp->numverts);
}
}
sp++;
}
BLI_assert(sort_face_verts_offset <= sort_face_verts.size());
vert_tag.clear_and_shrink();
/* Second check pass, testing faces using the same verts. */
blender::parallel_sort(sort_faces.begin(), sort_faces.end(), search_face_cmp);
SortFace *sp, *prev_sp;
sp = prev_sp = sort_faces.data();
sp++;
for (i = 1; i < faces_num; i++, sp++) {
int p1_nv = sp->numverts, p2_nv = prev_sp->numverts;
const int *p1_v = sp->verts, *p2_v = prev_sp->verts;
if (sp->invalid) {
/* Break, because all known invalid faces have been put at the end by the sort above. */
break;
}
/* Test same faces. */
if ((p1_nv == p2_nv) && (memcmp(p1_v, p2_v, p1_nv * sizeof(*p1_v)) == 0)) {
if (do_verbose) {
/* TODO: convert list to string */
PRINT_ERR("\tFaces %u and %u use same vertices (%d", prev_sp->index, sp->index, *p1_v);
for (j = 1; j < p1_nv; j++) {
PRINT_ERR(", %d", p1_v[j]);
}
PRINT_ERR("), considering face %u as invalid.", sp->index);
}
else {
is_valid = false;
}
sp->invalid = true;
}
else {
prev_sp = sp;
}
}
/* Third check pass, testing corners used by none or more than one face. */
blender::parallel_sort(sort_faces.begin(), sort_faces.end(), search_face_corner_cmp);
sp = sort_faces.data();
prev_sp = nullptr;
int prev_end = 0;
for (i = 0; i < faces_num; i++, sp++) {
/* We don't need the verts anymore, and avoid us another corner! */
sp->verts = nullptr;
/* Note above prev_sp: in following code, we make sure it is always valid face (or nullptr).
*/
if (sp->invalid) {
if (do_fixes) {
faces_to_remove[sp->index].set();
free_flag.face_corners = do_fixes;
/* DO NOT REMOVE ITS corners!!!
* As already invalid faces are at the end of the SortFace list, the corners they
* were the only users have already been tagged as "to remove" during previous
* iterations, and we don't want to remove some corners that may be used by
* another valid face! */
}
}
/* Test corners users. */
else {
/* Unused corners. */
if (prev_end < sp->corner_start) {
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
int corner;
for (j = prev_end, corner = prev_end; j < sp->corner_start; j++, corner++) {
PRINT_ERR("\tCorner %u is unused.", j);
if (do_fixes) {
REMOVE_CORNER_TAG(corner);
}
}
prev_end = sp->corner_start + sp->numverts;
prev_sp = sp;
}
/* Multi-used corners. */
else if (prev_end > sp->corner_start) {
PRINT_ERR(
"\tFaces %u and %u share corners from %d to %d, considering face %u as invalid.",
prev_sp->index,
sp->index,
sp->corner_start,
prev_end,
sp->index);
if (do_fixes) {
faces_to_remove[sp->index].set();
free_flag.face_corners = do_fixes;
/* DO NOT REMOVE ITS corners!!!
* They might be used by some next, valid face!
* Just not updating prev_end/prev_sp vars is enough to ensure the corners
* effectively no more needed will be marked as "to be removed"! */
}
}
else {
prev_end = sp->corner_start + sp->numverts;
prev_sp = sp;
}
}
}
/* We may have some remaining unused corners to get rid of! */
if (prev_end < corners_num) {
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
int corner;
for (j = prev_end, corner = prev_end; j < corners_num; j++, corner++) {
PRINT_ERR("\tCorner %u is unused.", j);
if (do_fixes) {
REMOVE_CORNER_TAG(corner);
}
}
}
}
/* fix deform verts */
if (dverts) {
MDeformVert *dv;
for (i = 0, dv = dverts; i < verts_num; i++, dv++) {
MDeformWeight *dw;
for (j = 0, dw = dv->dw; j < dv->totweight; j++, dw++) {
/* NOTE: greater than max defgroups is accounted for in our code, but not < 0. */
if (!isfinite(dw->weight)) {
PRINT_ERR("\tVertex deform %u, group %u has weight: %f", i, dw->def_nr, dw->weight);
if (do_fixes) {
dw->weight = 0.0f;
fix_flag.verts_weight = true;
}
}
else if (dw->weight < 0.0f || dw->weight > 1.0f) {
PRINT_ERR("\tVertex deform %u, group %u has weight: %f", i, dw->def_nr, dw->weight);
if (do_fixes) {
CLAMP(dw->weight, 0.0f, 1.0f);
fix_flag.verts_weight = true;
}
}
/* Not technically incorrect since this is unsigned, however,
* a value over INT_MAX is almost certainly caused by wrapping an uint. */
if (dw->def_nr >= INT_MAX) {
PRINT_ERR("\tVertex deform %u, has invalid group %u", i, dw->def_nr);
if (do_fixes) {
BKE_defvert_remove_group(dv, dw);
fix_flag.verts_weight = true;
if (dv->dw) {
/* re-allocated, the new values compensate for stepping
* within the for corner and may not be valid */
j--;
dw = dv->dw + j;
}
else { /* all freed */
break;
}
}
}
}
}
}
#undef REMOVE_EDGE_TAG
#undef IS_REMOVED_EDGE
#undef REMOVE_CORNER_TAG
#undef REMOVE_FACE_TAG
if (mesh) {
if (free_flag.faces) {
BKE_mesh_strip_loose_faces(mesh);
}
if (free_flag.face_corners) {
strip_loose_faces_corners(mesh, faces_to_remove);
}
if (free_flag.edges) {
mesh_strip_edges(mesh);
}
if (recalc_flag.edges) {
mesh_calc_edges(*mesh, true, false);
}
}
if (mesh && mesh->mselect) {
MSelect *msel;
for (i = 0, msel = mesh->mselect; i < mesh->totselect; i++, msel++) {
int tot_elem = 0;
if (msel->index < 0) {
PRINT_ERR(
"\tMesh select element %u type %d index is negative, "
"resetting selection stack.\n",
i,
msel->type);
free_flag.mselect = do_fixes;
break;
}
switch (msel->type) {
case ME_VSEL:
tot_elem = mesh->verts_num;
break;
case ME_ESEL:
tot_elem = mesh->edges_num;
break;
case ME_FSEL:
tot_elem = mesh->faces_num;
break;
}
if (msel->index > tot_elem) {
PRINT_ERR(
"\tMesh select element %u type %d index %d is larger than data array size %d, "
"resetting selection stack.\n",
i,
msel->type,
msel->index,
tot_elem);
free_flag.mselect = do_fixes;
break;
}
}
if (free_flag.mselect) {
MEM_freeN(mesh->mselect);
mesh->mselect = nullptr;
mesh->totselect = 0;
}
}
material_indices_span.save();
material_indices.finish();
PRINT_MSG("%s: finished\n\n", __func__);
*r_changed = (fix_flag.as_flag || free_flag.as_flag || recalc_flag.as_flag);
BLI_assert((*r_changed == false) || (do_fixes == true));
return is_valid;
}
2018-07-13 08:36:10 +02:00
static bool mesh_validate_customdata(CustomData *data,
eCustomDataMask mask,
const uint totitems,
2018-07-13 08:36:10 +02:00
const bool do_verbose,
const bool do_fixes,
bool *r_change)
{
bool is_valid = true;
bool has_fixes = false;
int i = 0;
PRINT_MSG("%s: Checking %d CD layers...\n", __func__, data->totlayer);
/* Set dummy values so the layer-type is always initialized on first access. */
int layer_num = -1;
int layer_num_type = -1;
while (i < data->totlayer) {
CustomDataLayer *layer = &data->layers[i];
const eCustomDataType type = eCustomDataType(layer->type);
bool ok = true;
/* Count layers when the type changes. */
if (layer_num_type != type) {
layer_num = CustomData_number_of_layers(data, type);
layer_num_type = type;
}
/* Validate active index, for a time this could be set to a negative value, see: #105860. */
int *active_index_array[] = {
&layer->active,
&layer->active_rnd,
&layer->active_clone,
&layer->active_mask,
};
for (int *active_index : Span(active_index_array, ARRAY_SIZE(active_index_array))) {
if (*active_index < 0) {
PRINT_ERR("\tCustomDataLayer type %d has a negative active index (%d)\n",
layer->type,
*active_index);
if (do_fixes) {
*active_index = 0;
has_fixes = true;
}
}
else {
if (*active_index >= layer_num) {
PRINT_ERR("\tCustomDataLayer type %d has an out of bounds active index (%d >= %d)\n",
layer->type,
*active_index,
layer_num);
if (do_fixes) {
BLI_assert(layer_num > 0);
*active_index = layer_num - 1;
has_fixes = true;
}
}
}
}
if (CustomData_layertype_is_singleton(type)) {
if (layer_num > 1) {
PRINT_ERR("\tCustomDataLayer type %d is a singleton, found %d in Mesh structure\n",
type,
layer_num);
ok = false;
}
}
if (mask != 0) {
eCustomDataMask layer_typemask = CD_TYPE_AS_MASK(type);
if ((layer_typemask & mask) == 0) {
PRINT_ERR("\tCustomDataLayer type %d which isn't in the mask\n", type);
ok = false;
}
}
if (ok == false) {
if (do_fixes) {
CustomData_free_layer(data, type, 0, i);
has_fixes = true;
}
}
if (ok) {
if (CustomData_layer_validate(layer, totitems, do_fixes)) {
PRINT_ERR("\tCustomDataLayer type %d has some invalid data\n", type);
has_fixes = do_fixes;
}
i++;
}
}
PRINT_MSG("%s: Finished (is_valid=%d)\n\n", __func__, int(!has_fixes));
*r_change = has_fixes;
return is_valid;
}
bool BKE_mesh_validate_all_customdata(CustomData *vert_data,
const uint verts_num,
CustomData *edge_data,
const uint edges_num,
CustomData *corner_data,
const uint corners_num,
CustomData *face_data,
const uint faces_num,
2018-07-13 08:36:10 +02:00
const bool check_meshmask,
const bool do_verbose,
const bool do_fixes,
bool *r_change)
{
bool is_valid = true;
bool is_change_v, is_change_e, is_change_l, is_change_p;
CustomData_MeshMasks mask = {0};
if (check_meshmask) {
mask = CD_MASK_MESH;
}
is_valid &= mesh_validate_customdata(
vert_data, mask.vmask, verts_num, do_verbose, do_fixes, &is_change_v);
is_valid &= mesh_validate_customdata(
edge_data, mask.emask, edges_num, do_verbose, do_fixes, &is_change_e);
is_valid &= mesh_validate_customdata(
corner_data, mask.lmask, corners_num, do_verbose, do_fixes, &is_change_l);
is_valid &= mesh_validate_customdata(
face_data, mask.pmask, faces_num, do_verbose, do_fixes, &is_change_p);
const int uv_maps_num = CustomData_number_of_layers(corner_data, CD_PROP_FLOAT2);
if (uv_maps_num > MAX_MTFACE) {
PRINT_ERR(
"\tMore UV layers than %d allowed, %d last ones won't be available for render, shaders, "
"etc.\n",
MAX_MTFACE,
uv_maps_num - MAX_MTFACE);
}
/* check indices of clone/stencil */
if (do_fixes && CustomData_get_clone_layer(corner_data, CD_PROP_FLOAT2) >= uv_maps_num) {
CustomData_set_layer_clone(corner_data, CD_PROP_FLOAT2, 0);
is_change_l = true;
}
if (do_fixes && CustomData_get_stencil_layer(corner_data, CD_PROP_FLOAT2) >= uv_maps_num) {
CustomData_set_layer_stencil(corner_data, CD_PROP_FLOAT2, 0);
is_change_l = true;
}
*r_change = (is_change_v || is_change_e || is_change_l || is_change_p);
return is_valid;
}
bool BKE_mesh_validate(Mesh *mesh, const bool do_verbose, const bool cddata_check_mask)
{
bool changed;
if (do_verbose) {
CLOG_INFO(&LOG, 0, "MESH: %s", mesh->id.name + 2);
}
BKE_mesh_validate_all_customdata(&mesh->vert_data,
mesh->verts_num,
&mesh->edge_data,
mesh->edges_num,
&mesh->corner_data,
mesh->corners_num,
&mesh->face_data,
mesh->faces_num,
cddata_check_mask,
do_verbose,
true,
&changed);
MutableSpan<float3> positions = mesh->vert_positions_for_write();
MutableSpan<blender::int2> edges = mesh->edges_for_write();
Span<int> face_offsets = mesh->face_offsets();
Span<int> corner_verts = mesh->corner_verts();
MutableSpan<int> corner_edges = mesh->corner_edges_for_write();
MDeformVert *dverts = static_cast<MDeformVert *>(
CustomData_get_layer_for_write(&mesh->vert_data, CD_MDEFORMVERT, mesh->verts_num));
BKE_mesh_validate_arrays(
mesh,
reinterpret_cast<float(*)[3]>(positions.data()),
positions.size(),
edges.data(),
edges.size(),
(MFace *)CustomData_get_layer_for_write(&mesh->fdata_legacy, CD_MFACE, mesh->totface_legacy),
mesh->totface_legacy,
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
corner_verts.data(),
corner_edges.data(),
corner_verts.size(),
face_offsets.data(),
mesh->faces_num,
dverts,
do_verbose,
true,
&changed);
if (changed) {
BKE_mesh_runtime_clear_cache(mesh);
DEG_id_tag_update(&mesh->id, ID_RECALC_GEOMETRY_ALL_MODES);
return true;
}
return false;
}
bool BKE_mesh_is_valid(Mesh *mesh)
{
const bool do_verbose = true;
const bool do_fixes = false;
bool is_valid = true;
bool changed = true;
is_valid &= BKE_mesh_validate_all_customdata(
&mesh->vert_data,
mesh->verts_num,
&mesh->edge_data,
mesh->edges_num,
&mesh->corner_data,
mesh->corners_num,
&mesh->face_data,
mesh->faces_num,
false, /* setting mask here isn't useful, gives false positives */
do_verbose,
do_fixes,
&changed);
MutableSpan<float3> positions = mesh->vert_positions_for_write();
MutableSpan<blender::int2> edges = mesh->edges_for_write();
Span<int> face_offsets = mesh->face_offsets();
Span<int> corner_verts = mesh->corner_verts();
MutableSpan<int> corner_edges = mesh->corner_edges_for_write();
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
MDeformVert *dverts = static_cast<MDeformVert *>(
CustomData_get_layer_for_write(&mesh->vert_data, CD_MDEFORMVERT, mesh->verts_num));
is_valid &= BKE_mesh_validate_arrays(
mesh,
reinterpret_cast<float(*)[3]>(positions.data()),
positions.size(),
edges.data(),
edges.size(),
(MFace *)CustomData_get_layer_for_write(&mesh->fdata_legacy, CD_MFACE, mesh->totface_legacy),
mesh->totface_legacy,
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
corner_verts.data(),
corner_edges.data(),
corner_verts.size(),
face_offsets.data(),
mesh->faces_num,
dverts,
do_verbose,
do_fixes,
&changed);
BLI_assert(changed == false);
return is_valid;
}
bool BKE_mesh_validate_material_indices(Mesh *mesh)
{
const int mat_nr_max = max_ii(0, mesh->totcol - 1);
bool is_valid = true;
blender::bke::AttributeWriter<int> material_indices =
mesh->attributes_for_write().lookup_for_write<int>("material_index");
blender::MutableVArraySpan<int> material_indices_span(material_indices.varray);
for (const int i : material_indices_span.index_range()) {
if (material_indices_span[i] < 0 || material_indices_span[i] > mat_nr_max) {
material_indices_span[i] = 0;
is_valid = false;
}
}
material_indices_span.save();
material_indices.finish();
if (!is_valid) {
DEG_id_tag_update(&mesh->id, ID_RECALC_GEOMETRY_ALL_MODES);
return true;
}
return false;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Mesh Stripping (removing invalid data)
* \{ */
void strip_loose_faces_corners(Mesh *mesh, blender::BitSpan faces_to_remove)
{
MutableSpan<int> face_offsets = mesh->face_offsets_for_write();
MutableSpan<int> corner_edges = mesh->corner_edges_for_write();
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
int a, b;
/* New corners idx! */
int *new_idx = (int *)MEM_mallocN(sizeof(int) * mesh->corners_num, __func__);
for (a = b = 0; a < mesh->faces_num; a++) {
bool invalid = false;
int start = face_offsets[a];
int size = face_offsets[a + 1] - start;
Mesh: Replace MPoly struct with offset indices Implements #95967. Currently the `MPoly` struct is 12 bytes, and stores the index of a face's first corner and the number of corners/verts/edges. Polygons and corners are always created in order by Blender, meaning each face's corners will be after the previous face's corners. We can take advantage of this fact and eliminate the redundancy in mesh face storage by only storing a single integer corner offset for each face. The size of the face is then encoded by the offset of the next face. The size of a single integer is 4 bytes, so this reduces memory usage by 3 times. The same method is used for `CurvesGeometry`, so Blender already has an abstraction to simplify using these offsets called `OffsetIndices`. This class is used to easily retrieve a range of corner indices for each face. This also gives the opportunity for sharing some logic with curves. Another benefit of the change is that the offsets and sizes stored in `MPoly` can no longer disagree with each other. Storing faces in the order of their corners can simplify some code too. Face/polygon variables now use the `IndexRange` type, which comes with quite a few utilities that can simplify code. Some: - The offset integer array has to be one longer than the face count to avoid a branch for every face, which means the data is no longer part of the mesh's `CustomData`. - We lose the ability to "reference" an original mesh's offset array until more reusable CoW from #104478 is committed. That will be added in a separate commit. - Since they aren't part of `CustomData`, poly offsets often have to be copied manually. - To simplify using `OffsetIndices` in many places, some functions and structs in headers were moved to only compile in C++. - All meshes created by Blender use the same order for faces and face corners, but just in case, meshes with mismatched order are fixed by versioning code. - `MeshPolygon.totloop` is no longer editable in RNA. This API break is necessary here unfortunately. It should be worth it in 3.6, since that's the best way to allow loading meshes from 4.0, which is important for an LTS version. Pull Request: https://projects.blender.org/blender/blender/pulls/105938
2023-04-04 20:39:28 +02:00
int stop = start + size;
if (faces_to_remove[a]) {
Mesh: Replace MPoly struct with offset indices Implements #95967. Currently the `MPoly` struct is 12 bytes, and stores the index of a face's first corner and the number of corners/verts/edges. Polygons and corners are always created in order by Blender, meaning each face's corners will be after the previous face's corners. We can take advantage of this fact and eliminate the redundancy in mesh face storage by only storing a single integer corner offset for each face. The size of the face is then encoded by the offset of the next face. The size of a single integer is 4 bytes, so this reduces memory usage by 3 times. The same method is used for `CurvesGeometry`, so Blender already has an abstraction to simplify using these offsets called `OffsetIndices`. This class is used to easily retrieve a range of corner indices for each face. This also gives the opportunity for sharing some logic with curves. Another benefit of the change is that the offsets and sizes stored in `MPoly` can no longer disagree with each other. Storing faces in the order of their corners can simplify some code too. Face/polygon variables now use the `IndexRange` type, which comes with quite a few utilities that can simplify code. Some: - The offset integer array has to be one longer than the face count to avoid a branch for every face, which means the data is no longer part of the mesh's `CustomData`. - We lose the ability to "reference" an original mesh's offset array until more reusable CoW from #104478 is committed. That will be added in a separate commit. - Since they aren't part of `CustomData`, poly offsets often have to be copied manually. - To simplify using `OffsetIndices` in many places, some functions and structs in headers were moved to only compile in C++. - All meshes created by Blender use the same order for faces and face corners, but just in case, meshes with mismatched order are fixed by versioning code. - `MeshPolygon.totloop` is no longer editable in RNA. This API break is necessary here unfortunately. It should be worth it in 3.6, since that's the best way to allow loading meshes from 4.0, which is important for an LTS version. Pull Request: https://projects.blender.org/blender/blender/pulls/105938
2023-04-04 20:39:28 +02:00
invalid = true;
}
else if (stop > mesh->corners_num || stop < start || size < 0) {
invalid = true;
}
else {
/* If one of the face's corners is invalid, the whole face is invalid! */
if (corner_edges.slice(start, size).as_span().contains(INVALID_CORNER_EDGE_MARKER)) {
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
invalid = true;
}
}
Mesh: Replace MPoly struct with offset indices Implements #95967. Currently the `MPoly` struct is 12 bytes, and stores the index of a face's first corner and the number of corners/verts/edges. Polygons and corners are always created in order by Blender, meaning each face's corners will be after the previous face's corners. We can take advantage of this fact and eliminate the redundancy in mesh face storage by only storing a single integer corner offset for each face. The size of the face is then encoded by the offset of the next face. The size of a single integer is 4 bytes, so this reduces memory usage by 3 times. The same method is used for `CurvesGeometry`, so Blender already has an abstraction to simplify using these offsets called `OffsetIndices`. This class is used to easily retrieve a range of corner indices for each face. This also gives the opportunity for sharing some logic with curves. Another benefit of the change is that the offsets and sizes stored in `MPoly` can no longer disagree with each other. Storing faces in the order of their corners can simplify some code too. Face/polygon variables now use the `IndexRange` type, which comes with quite a few utilities that can simplify code. Some: - The offset integer array has to be one longer than the face count to avoid a branch for every face, which means the data is no longer part of the mesh's `CustomData`. - We lose the ability to "reference" an original mesh's offset array until more reusable CoW from #104478 is committed. That will be added in a separate commit. - Since they aren't part of `CustomData`, poly offsets often have to be copied manually. - To simplify using `OffsetIndices` in many places, some functions and structs in headers were moved to only compile in C++. - All meshes created by Blender use the same order for faces and face corners, but just in case, meshes with mismatched order are fixed by versioning code. - `MeshPolygon.totloop` is no longer editable in RNA. This API break is necessary here unfortunately. It should be worth it in 3.6, since that's the best way to allow loading meshes from 4.0, which is important for an LTS version. Pull Request: https://projects.blender.org/blender/blender/pulls/105938
2023-04-04 20:39:28 +02:00
if (size >= 3 && !invalid) {
if (a != b) {
face_offsets[b] = face_offsets[a];
CustomData_copy_data(&mesh->face_data, &mesh->face_data, a, b, 1);
}
b++;
}
}
if (a != b) {
CustomData_free_elem(&mesh->face_data, b, a - b);
mesh->faces_num = b;
}
/* And now, get rid of invalid corners. */
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
int corner = 0;
for (a = b = 0; a < mesh->corners_num; a++, corner++) {
if (corner_edges[corner] != INVALID_CORNER_EDGE_MARKER) {
if (a != b) {
CustomData_copy_data(&mesh->corner_data, &mesh->corner_data, a, b, 1);
}
new_idx[a] = b;
b++;
}
else {
/* XXX Theoretically, we should be able to not do this, as no remaining face
* should use any stripped corner. But for security's sake... */
new_idx[a] = -a;
}
}
if (a != b) {
CustomData_free_elem(&mesh->corner_data, b, a - b);
mesh->corners_num = b;
}
face_offsets[mesh->faces_num] = mesh->corners_num;
Mesh: Replace MPoly struct with offset indices Implements #95967. Currently the `MPoly` struct is 12 bytes, and stores the index of a face's first corner and the number of corners/verts/edges. Polygons and corners are always created in order by Blender, meaning each face's corners will be after the previous face's corners. We can take advantage of this fact and eliminate the redundancy in mesh face storage by only storing a single integer corner offset for each face. The size of the face is then encoded by the offset of the next face. The size of a single integer is 4 bytes, so this reduces memory usage by 3 times. The same method is used for `CurvesGeometry`, so Blender already has an abstraction to simplify using these offsets called `OffsetIndices`. This class is used to easily retrieve a range of corner indices for each face. This also gives the opportunity for sharing some logic with curves. Another benefit of the change is that the offsets and sizes stored in `MPoly` can no longer disagree with each other. Storing faces in the order of their corners can simplify some code too. Face/polygon variables now use the `IndexRange` type, which comes with quite a few utilities that can simplify code. Some: - The offset integer array has to be one longer than the face count to avoid a branch for every face, which means the data is no longer part of the mesh's `CustomData`. - We lose the ability to "reference" an original mesh's offset array until more reusable CoW from #104478 is committed. That will be added in a separate commit. - Since they aren't part of `CustomData`, poly offsets often have to be copied manually. - To simplify using `OffsetIndices` in many places, some functions and structs in headers were moved to only compile in C++. - All meshes created by Blender use the same order for faces and face corners, but just in case, meshes with mismatched order are fixed by versioning code. - `MeshPolygon.totloop` is no longer editable in RNA. This API break is necessary here unfortunately. It should be worth it in 3.6, since that's the best way to allow loading meshes from 4.0, which is important for an LTS version. Pull Request: https://projects.blender.org/blender/blender/pulls/105938
2023-04-04 20:39:28 +02:00
/* And now, update faces' start corner index. */
/* NOTE: At this point, there should never be any face using a stripped corner! */
for (const int i : blender::IndexRange(mesh->faces_num)) {
face_offsets[i] = new_idx[face_offsets[i]];
BLI_assert(face_offsets[i] >= 0);
}
MEM_freeN(new_idx);
}
void mesh_strip_edges(Mesh *mesh)
{
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
blender::int2 *e;
int a, b;
uint *new_idx = (uint *)MEM_mallocN(sizeof(int) * mesh->edges_num, __func__);
MutableSpan<blender::int2> edges = mesh->edges_for_write();
for (a = b = 0, e = edges.data(); a < mesh->edges_num; a++, e++) {
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
if ((*e)[0] != (*e)[1]) {
if (a != b) {
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
memcpy(&edges[b], e, sizeof(edges[b]));
CustomData_copy_data(&mesh->edge_data, &mesh->edge_data, a, b, 1);
}
new_idx[a] = b;
b++;
}
else {
new_idx[a] = INVALID_CORNER_EDGE_MARKER;
}
}
if (a != b) {
CustomData_free_elem(&mesh->edge_data, b, a - b);
mesh->edges_num = b;
}
/* And now, update corners' edge indices. */
/* XXX We hope no corner was pointing to a stripped edge!
* Else, its e will be set to INVALID_CORNER_EDGE_MARKER :/ */
MutableSpan<int> corner_edges = mesh->corner_edges_for_write();
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
for (const int i : corner_edges.index_range()) {
corner_edges[i] = new_idx[corner_edges[i]];
}
MEM_freeN(new_idx);
}
/** \} */