Files
test/source/blender/modifiers/intern/MOD_mask.cc

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

864 lines
30 KiB
C++
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later
* Copyright 2005 Blender Foundation. All rights reserved. */
/** \file
* \ingroup modifiers
*/
#include "MEM_guardedalloc.h"
#include "BLI_utildefines.h"
#include "BLI_ghash.h"
#include "BLI_listbase.h"
#include "BLI_math.h"
#include "BLT_translation.h"
#include "DNA_armature_types.h"
#include "DNA_defaults.h"
#include "DNA_mesh_types.h"
#include "DNA_meshdata_types.h"
#include "DNA_modifier_types.h"
#include "DNA_object_types.h"
#include "DNA_screen_types.h"
#include "BKE_action.h" /* BKE_pose_channel_find_name */
#include "BKE_context.h"
#include "BKE_customdata.h"
#include "BKE_deform.h"
#include "BKE_lib_query.h"
#include "BKE_mesh.h"
#include "BKE_modifier.h"
#include "BKE_screen.h"
#include "UI_interface.h"
#include "UI_resources.h"
#include "RNA_access.h"
#include "RNA_prototypes.h"
#include "DEG_depsgraph_build.h"
#include "DEG_depsgraph_query.h"
#include "MOD_modifiertypes.h"
#include "MOD_ui_common.h"
#include "BLI_array.hh"
#include "BLI_listbase_wrapper.hh"
#include "BLI_vector.hh"
using blender::Array;
using blender::IndexRange;
using blender::ListBaseWrapper;
using blender::MutableSpan;
using blender::Span;
using blender::Vector;
static void initData(ModifierData *md)
{
MaskModifierData *mmd = (MaskModifierData *)md;
BLI_assert(MEMCMP_STRUCT_AFTER_IS_ZERO(mmd, modifier));
MEMCPY_STRUCT_AFTER(mmd, DNA_struct_default_get(MaskModifierData), modifier);
}
static void requiredDataMask(ModifierData * /*md*/, CustomData_MeshMasks *r_cddata_masks)
{
r_cddata_masks->vmask |= CD_MASK_MDEFORMVERT;
}
static void foreachIDLink(ModifierData *md, Object *ob, IDWalkFunc walk, void *userData)
{
MaskModifierData *mmd = reinterpret_cast<MaskModifierData *>(md);
walk(userData, ob, (ID **)&mmd->ob_arm, IDWALK_CB_NOP);
}
static void updateDepsgraph(ModifierData *md, const ModifierUpdateDepsgraphContext *ctx)
{
MaskModifierData *mmd = reinterpret_cast<MaskModifierData *>(md);
if (mmd->ob_arm) {
bArmature *arm = (bArmature *)mmd->ob_arm->data;
/* Tag relationship in depsgraph, but also on the armature. */
/* TODO(sergey): Is it a proper relation here? */
DEG_add_object_relation(ctx->node, mmd->ob_arm, DEG_OB_COMP_TRANSFORM, "Mask Modifier");
arm->flag |= ARM_HAS_VIZ_DEPS;
DEG_add_depends_on_transform_relation(ctx->node, "Mask Modifier");
}
}
/* A vertex will be in the mask if a selected bone influences it more than a certain threshold. */
static void compute_vertex_mask__armature_mode(const MDeformVert *dvert,
Mesh *mesh,
Object *armature_ob,
float threshold,
MutableSpan<bool> r_vertex_mask)
{
/* Element i is true if there is a selected bone that uses vertex group i. */
Vector<bool> selected_bone_uses_group;
LISTBASE_FOREACH (bDeformGroup *, def, &mesh->vertex_group_names) {
bPoseChannel *pchan = BKE_pose_channel_find_name(armature_ob->pose, def->name);
bool bone_for_group_exists = pchan && pchan->bone && (pchan->bone->flag & BONE_SELECTED);
selected_bone_uses_group.append(bone_for_group_exists);
}
Span<bool> use_vertex_group = selected_bone_uses_group;
for (int i : r_vertex_mask.index_range()) {
Span<MDeformWeight> weights(dvert[i].dw, dvert[i].totweight);
r_vertex_mask[i] = false;
/* check the groups that vertex is assigned to, and see if it was any use */
for (const MDeformWeight &dw : weights) {
if (use_vertex_group.get(dw.def_nr, false)) {
if (dw.weight > threshold) {
r_vertex_mask[i] = true;
break;
}
}
}
}
}
/* A vertex will be in the mask if the vertex group influences it more than a certain threshold. */
static void compute_vertex_mask__vertex_group_mode(const MDeformVert *dvert,
int defgrp_index,
float threshold,
MutableSpan<bool> r_vertex_mask)
{
for (int i : r_vertex_mask.index_range()) {
const bool found = BKE_defvert_find_weight(&dvert[i], defgrp_index) > threshold;
r_vertex_mask[i] = found;
}
}
static void invert_boolean_array(MutableSpan<bool> array)
{
for (bool &value : array) {
value = !value;
}
}
static void compute_masked_verts(Span<bool> vertex_mask,
MutableSpan<int> r_vertex_map,
uint *r_verts_masked_num)
{
BLI_assert(vertex_mask.size() == r_vertex_map.size());
uint verts_masked_num = 0;
for (uint i_src : r_vertex_map.index_range()) {
if (vertex_mask[i_src]) {
r_vertex_map[i_src] = verts_masked_num;
verts_masked_num++;
}
else {
r_vertex_map[i_src] = -1;
}
}
*r_verts_masked_num = verts_masked_num;
}
static void computed_masked_edges(const Mesh *mesh,
Span<bool> vertex_mask,
MutableSpan<int> r_edge_map,
uint *r_edges_masked_num)
{
BLI_assert(mesh->totedge == r_edge_map.size());
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
const Span<MEdge> edges = mesh->edges();
uint edges_masked_num = 0;
for (int i : IndexRange(mesh->totedge)) {
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
const MEdge &edge = edges[i];
/* only add if both verts will be in new mesh */
if (vertex_mask[edge.v1] && vertex_mask[edge.v2]) {
r_edge_map[i] = edges_masked_num;
edges_masked_num++;
}
else {
r_edge_map[i] = -1;
}
}
*r_edges_masked_num = edges_masked_num;
}
static void computed_masked_edges_smooth(const Mesh *mesh,
Span<bool> vertex_mask,
MutableSpan<int> r_edge_map,
uint *r_edges_masked_num,
uint *r_verts_add_num)
{
BLI_assert(mesh->totedge == r_edge_map.size());
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
const Span<MEdge> edges = mesh->edges();
uint edges_masked_num = 0;
uint verts_add_num = 0;
for (int i : IndexRange(mesh->totedge)) {
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
const MEdge &edge = edges[i];
/* only add if both verts will be in new mesh */
bool v1 = vertex_mask[edge.v1];
bool v2 = vertex_mask[edge.v2];
if (v1 && v2) {
r_edge_map[i] = edges_masked_num;
edges_masked_num++;
}
else if (v1 != v2) {
r_edge_map[i] = -2;
verts_add_num++;
}
else {
r_edge_map[i] = -1;
}
}
edges_masked_num += verts_add_num;
*r_edges_masked_num = edges_masked_num;
*r_verts_add_num = verts_add_num;
}
static void computed_masked_polys(const Mesh *mesh,
Span<bool> vertex_mask,
Vector<int> &r_masked_poly_indices,
Vector<int> &r_loop_starts,
uint *r_polys_masked_num,
uint *r_loops_masked_num)
{
BLI_assert(mesh->totvert == vertex_mask.size());
const Span<MPoly> polys = mesh->polys();
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
const Span<MLoop> loops = mesh->loops();
r_masked_poly_indices.reserve(mesh->totpoly);
r_loop_starts.reserve(mesh->totpoly);
uint loops_masked_num = 0;
for (int i : IndexRange(mesh->totpoly)) {
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
const MPoly &poly_src = polys[i];
bool all_verts_in_mask = true;
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
Span<MLoop> loops_src = loops.slice(poly_src.loopstart, poly_src.totloop);
for (const MLoop &loop : loops_src) {
if (!vertex_mask[loop.v]) {
all_verts_in_mask = false;
break;
}
}
if (all_verts_in_mask) {
r_masked_poly_indices.append_unchecked(i);
r_loop_starts.append_unchecked(loops_masked_num);
loops_masked_num += poly_src.totloop;
}
}
*r_polys_masked_num = r_masked_poly_indices.size();
*r_loops_masked_num = loops_masked_num;
}
static void compute_interpolated_polys(const Mesh *mesh,
Span<bool> vertex_mask,
uint verts_add_num,
uint loops_masked_num,
Vector<int> &r_masked_poly_indices,
Vector<int> &r_loop_starts,
uint *r_edges_add_num,
uint *r_polys_add_num,
uint *r_loops_add_num)
{
BLI_assert(mesh->totvert == vertex_mask.size());
/* Can't really know ahead of time how much space to use exactly. Estimate limit instead. */
/* NOTE: this reserve can only lift the capacity if there are ngons, which get split. */
r_masked_poly_indices.reserve(r_masked_poly_indices.size() + verts_add_num);
r_loop_starts.reserve(r_loop_starts.size() + verts_add_num);
const Span<MPoly> polys = mesh->polys();
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
const Span<MLoop> loops = mesh->loops();
uint edges_add_num = 0;
uint polys_add_num = 0;
uint loops_add_num = 0;
for (int i : IndexRange(mesh->totpoly)) {
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
const MPoly &poly_src = polys[i];
int in_count = 0;
int start = -1;
int dst_totloop = -1;
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
const Span<MLoop> loops_src = loops.slice(poly_src.loopstart, poly_src.totloop);
for (const int j : loops_src.index_range()) {
const MLoop &loop = loops_src[j];
if (vertex_mask[loop.v]) {
in_count++;
}
else if (start == -1) {
start = j;
}
}
if (0 < in_count && in_count < poly_src.totloop) {
/* Ring search starting at a vertex which is not included in the mask. */
const MLoop *last_loop = &loops_src[start];
bool v_loop_in_mask_last = vertex_mask[last_loop->v];
for (const int j : loops_src.index_range()) {
const MLoop &loop = loops_src[(start + 1 + j) % poly_src.totloop];
const bool v_loop_in_mask = vertex_mask[loop.v];
if (v_loop_in_mask && !v_loop_in_mask_last) {
dst_totloop = 3;
}
else if (!v_loop_in_mask && v_loop_in_mask_last) {
BLI_assert(dst_totloop > 2);
r_masked_poly_indices.append(i);
r_loop_starts.append(loops_masked_num + loops_add_num);
loops_add_num += dst_totloop;
polys_add_num++;
edges_add_num++;
dst_totloop = -1;
}
else if (v_loop_in_mask && v_loop_in_mask_last) {
BLI_assert(dst_totloop > 2);
dst_totloop++;
}
last_loop = &loop;
v_loop_in_mask_last = v_loop_in_mask;
}
}
}
*r_edges_add_num = edges_add_num;
*r_polys_add_num = polys_add_num;
*r_loops_add_num = loops_add_num;
}
static void copy_masked_verts_to_new_mesh(const Mesh &src_mesh,
Mesh &dst_mesh,
Span<int> vertex_map)
{
BLI_assert(src_mesh.totvert == vertex_map.size());
const Span<MVert> src_verts = src_mesh.verts();
MutableSpan<MVert> dst_verts = dst_mesh.verts_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
for (const int i_src : vertex_map.index_range()) {
const int i_dst = vertex_map[i_src];
if (i_dst == -1) {
continue;
}
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
const MVert &v_src = src_verts[i_src];
MVert &v_dst = dst_verts[i_dst];
v_dst = v_src;
CustomData_copy_data(&src_mesh.vdata, &dst_mesh.vdata, i_src, i_dst, 1);
}
}
static float get_interp_factor_from_vgroup(
const MDeformVert *dvert, int defgrp_index, float threshold, uint v1, uint v2)
{
/* NOTE: this calculation is done twice for every vertex,
* instead of storing it the first time and then reusing it. */
float value1 = BKE_defvert_find_weight(&dvert[v1], defgrp_index);
float value2 = BKE_defvert_find_weight(&dvert[v2], defgrp_index);
return (threshold - value1) / (value2 - value1);
}
static void add_interp_verts_copy_edges_to_new_mesh(const Mesh &src_mesh,
Mesh &dst_mesh,
Span<bool> vertex_mask,
Span<int> vertex_map,
const MDeformVert *dvert,
int defgrp_index,
float threshold,
uint edges_masked_num,
uint verts_add_num,
MutableSpan<int> r_edge_map)
{
BLI_assert(src_mesh.totvert == vertex_mask.size());
BLI_assert(src_mesh.totedge == r_edge_map.size());
const Span<MVert> src_verts = src_mesh.verts();
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
const Span<MEdge> src_edges = src_mesh.edges();
MutableSpan<MVert> dst_verts = dst_mesh.verts_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
MutableSpan<MEdge> dst_edges = dst_mesh.edges_for_write();
uint vert_index = dst_mesh.totvert - verts_add_num;
uint edge_index = edges_masked_num - verts_add_num;
for (int i_src : IndexRange(src_mesh.totedge)) {
if (r_edge_map[i_src] != -1) {
int i_dst = r_edge_map[i_src];
if (i_dst == -2) {
i_dst = edge_index;
}
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
const MEdge &e_src = src_edges[i_src];
MEdge &e_dst = dst_edges[i_dst];
CustomData_copy_data(&src_mesh.edata, &dst_mesh.edata, i_src, i_dst, 1);
e_dst = e_src;
e_dst.v1 = vertex_map[e_src.v1];
e_dst.v2 = vertex_map[e_src.v2];
}
if (r_edge_map[i_src] == -2) {
const int i_dst = edge_index++;
r_edge_map[i_src] = i_dst;
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
const MEdge &e_src = src_edges[i_src];
/* Cut destination edge and make v1 the new vertex. */
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
MEdge &e_dst = dst_edges[i_dst];
if (!vertex_mask[e_src.v1]) {
e_dst.v1 = vert_index;
}
else {
BLI_assert(!vertex_mask[e_src.v2]);
e_dst.v2 = e_dst.v1;
e_dst.v1 = vert_index;
}
/* Create the new vertex. */
float fac = get_interp_factor_from_vgroup(
dvert, defgrp_index, threshold, e_src.v1, e_src.v2);
float weights[2] = {1.0f - fac, fac};
CustomData_interp(
2021-08-27 17:29:05 +10:00
&src_mesh.vdata, &dst_mesh.vdata, (int *)&e_src.v1, weights, nullptr, 2, vert_index);
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
MVert &v = dst_verts[vert_index];
const MVert &v1 = src_verts[e_src.v1];
const MVert &v2 = src_verts[e_src.v2];
interp_v3_v3v3(v.co, v1.co, v2.co, fac);
vert_index++;
}
}
BLI_assert(vert_index == dst_mesh.totvert);
BLI_assert(edge_index == edges_masked_num);
}
static void copy_masked_edges_to_new_mesh(const Mesh &src_mesh,
Mesh &dst_mesh,
Span<int> vertex_map,
Span<int> edge_map)
{
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
const Span<MEdge> src_edges = src_mesh.edges();
MutableSpan<MEdge> dst_edges = dst_mesh.edges_for_write();
BLI_assert(src_mesh.totvert == vertex_map.size());
BLI_assert(src_mesh.totedge == edge_map.size());
for (const int i_src : IndexRange(src_mesh.totedge)) {
const int i_dst = edge_map[i_src];
2021-10-20 10:45:59 +11:00
if (ELEM(i_dst, -1, -2)) {
continue;
}
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
const MEdge &e_src = src_edges[i_src];
MEdge &e_dst = dst_edges[i_dst];
CustomData_copy_data(&src_mesh.edata, &dst_mesh.edata, i_src, i_dst, 1);
e_dst = e_src;
e_dst.v1 = vertex_map[e_src.v1];
e_dst.v2 = vertex_map[e_src.v2];
}
}
static void copy_masked_polys_to_new_mesh(const Mesh &src_mesh,
Mesh &dst_mesh,
Span<int> vertex_map,
Span<int> edge_map,
Span<int> masked_poly_indices,
Span<int> new_loop_starts,
int polys_masked_num)
{
const Span<MPoly> src_polys = src_mesh.polys();
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
const Span<MLoop> src_loops = src_mesh.loops();
MutableSpan<MPoly> dst_polys = dst_mesh.polys_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
MutableSpan<MLoop> dst_loops = dst_mesh.loops_for_write();
for (const int i_dst : IndexRange(polys_masked_num)) {
const int i_src = masked_poly_indices[i_dst];
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
const MPoly &mp_src = src_polys[i_src];
MPoly &mp_dst = dst_polys[i_dst];
const int i_ml_src = mp_src.loopstart;
const int i_ml_dst = new_loop_starts[i_dst];
CustomData_copy_data(&src_mesh.pdata, &dst_mesh.pdata, i_src, i_dst, 1);
CustomData_copy_data(&src_mesh.ldata, &dst_mesh.ldata, i_ml_src, i_ml_dst, mp_src.totloop);
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
const MLoop *ml_src = src_loops.data() + i_ml_src;
MLoop *ml_dst = dst_loops.data() + i_ml_dst;
mp_dst = mp_src;
mp_dst.loopstart = i_ml_dst;
for (int i : IndexRange(mp_src.totloop)) {
ml_dst[i].v = vertex_map[ml_src[i].v];
ml_dst[i].e = edge_map[ml_src[i].e];
}
}
}
static void add_interpolated_polys_to_new_mesh(const Mesh &src_mesh,
Mesh &dst_mesh,
Span<bool> vertex_mask,
Span<int> vertex_map,
Span<int> edge_map,
const MDeformVert *dvert,
int defgrp_index,
float threshold,
Span<int> masked_poly_indices,
Span<int> new_loop_starts,
int polys_masked_num,
int edges_add_num)
{
const Span<MPoly> src_polys = src_mesh.polys();
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
const Span<MLoop> src_loops = src_mesh.loops();
MutableSpan<MEdge> dst_edges = dst_mesh.edges_for_write();
MutableSpan<MPoly> dst_polys = dst_mesh.polys_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
MutableSpan<MLoop> dst_loops = dst_mesh.loops_for_write();
int edge_index = dst_mesh.totedge - edges_add_num;
int sub_poly_index = 0;
int last_i_src = -1;
for (const int i_dst :
IndexRange(polys_masked_num, masked_poly_indices.size() - polys_masked_num)) {
const int i_src = masked_poly_indices[i_dst];
if (i_src == last_i_src) {
sub_poly_index++;
}
else {
sub_poly_index = 0;
last_i_src = i_src;
}
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
const MPoly &mp_src = src_polys[i_src];
MPoly &mp_dst = dst_polys[i_dst];
const int i_ml_src = mp_src.loopstart;
int i_ml_dst = new_loop_starts[i_dst];
const int mp_totloop = (i_dst + 1 < new_loop_starts.size() ? new_loop_starts[i_dst + 1] :
dst_mesh.totloop) -
i_ml_dst;
CustomData_copy_data(&src_mesh.pdata, &dst_mesh.pdata, i_src, i_dst, 1);
mp_dst = mp_src;
mp_dst.loopstart = i_ml_dst;
mp_dst.totloop = mp_totloop;
/* Ring search starting at a vertex which is not included in the mask. */
int start = -sub_poly_index - 1;
bool skip = false;
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
Span<MLoop> loops_src(&src_loops[i_ml_src], mp_src.totloop);
for (const int j : loops_src.index_range()) {
if (!vertex_mask[loops_src[j].v]) {
if (start == -1) {
start = j;
break;
}
2021-08-27 17:29:05 +10:00
if (!skip) {
skip = true;
}
}
else if (skip) {
skip = false;
start++;
}
}
BLI_assert(start >= 0);
BLI_assert(edge_index < dst_mesh.totedge);
const MLoop *last_loop = &loops_src[start];
bool v_loop_in_mask_last = vertex_mask[last_loop->v];
int last_index = start;
for (const int j : loops_src.index_range()) {
const int index = (start + 1 + j) % mp_src.totloop;
const MLoop &loop = loops_src[index];
const bool v_loop_in_mask = vertex_mask[loop.v];
if (v_loop_in_mask && !v_loop_in_mask_last) {
/* Start new cut. */
float fac = get_interp_factor_from_vgroup(
dvert, defgrp_index, threshold, last_loop->v, loop.v);
float weights[2] = {1.0f - fac, fac};
int indices[2] = {i_ml_src + last_index, i_ml_src + index};
2021-08-27 17:29:05 +10:00
CustomData_interp(
&src_mesh.ldata, &dst_mesh.ldata, indices, weights, nullptr, 2, i_ml_dst);
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
MLoop &cut_dst_loop = dst_loops[i_ml_dst];
cut_dst_loop.e = edge_map[last_loop->e];
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
cut_dst_loop.v = dst_edges[cut_dst_loop.e].v1;
i_ml_dst++;
CustomData_copy_data(&src_mesh.ldata, &dst_mesh.ldata, i_ml_src + index, i_ml_dst, 1);
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
MLoop &next_dst_loop = dst_loops[i_ml_dst];
next_dst_loop.v = vertex_map[loop.v];
next_dst_loop.e = edge_map[loop.e];
i_ml_dst++;
}
else if (!v_loop_in_mask && v_loop_in_mask_last) {
BLI_assert(i_ml_dst != mp_dst.loopstart);
/* End active cut. */
float fac = get_interp_factor_from_vgroup(
dvert, defgrp_index, threshold, last_loop->v, loop.v);
float weights[2] = {1.0f - fac, fac};
int indices[2] = {i_ml_src + last_index, i_ml_src + index};
2021-08-27 17:29:05 +10:00
CustomData_interp(
&src_mesh.ldata, &dst_mesh.ldata, indices, weights, nullptr, 2, i_ml_dst);
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
MLoop &cut_dst_loop = dst_loops[i_ml_dst];
cut_dst_loop.e = edge_index;
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
cut_dst_loop.v = dst_edges[edge_map[last_loop->e]].v1;
i_ml_dst++;
/* Create closing edge. */
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
MEdge &cut_edge = dst_edges[edge_index];
cut_edge.v1 = dst_loops[mp_dst.loopstart].v;
cut_edge.v2 = cut_dst_loop.v;
BLI_assert(cut_edge.v1 != cut_edge.v2);
cut_edge.flag = ME_EDGEDRAW | ME_EDGERENDER;
edge_index++;
/* Only handle one of the cuts per iteration. */
break;
}
else if (v_loop_in_mask && v_loop_in_mask_last) {
BLI_assert(i_ml_dst != mp_dst.loopstart);
/* Extend active poly. */
CustomData_copy_data(&src_mesh.ldata, &dst_mesh.ldata, i_ml_src + index, i_ml_dst, 1);
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
MLoop &dst_loop = dst_loops[i_ml_dst];
dst_loop.v = vertex_map[loop.v];
dst_loop.e = edge_map[loop.e];
i_ml_dst++;
}
last_loop = &loop;
last_index = index;
v_loop_in_mask_last = v_loop_in_mask;
}
BLI_assert(mp_dst.loopstart + mp_dst.totloop == i_ml_dst);
}
BLI_assert(edge_index == dst_mesh.totedge);
}
/* Components of the algorithm:
* 1. Figure out which vertices should be present in the output mesh.
* 2. Find edges and polygons only using those vertices.
* 3. Create a new mesh that only uses the found vertices, edges and polygons.
*/
static Mesh *modifyMesh(ModifierData *md, const ModifierEvalContext * /*ctx*/, Mesh *mesh)
{
MaskModifierData *mmd = reinterpret_cast<MaskModifierData *>(md);
const bool invert_mask = mmd->flag & MOD_MASK_INV;
const bool use_interpolation = mmd->mode == MOD_MASK_MODE_VGROUP &&
(mmd->flag & MOD_MASK_SMOOTH);
/* Return empty or input mesh when there are no vertex groups. */
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
const Span<MDeformVert> dverts = mesh->deform_verts();
if (dverts.is_empty()) {
return invert_mask ? mesh : BKE_mesh_new_nomain_from_template(mesh, 0, 0, 0, 0, 0);
}
/* Quick test to see if we can return early. */
if (!ELEM(mmd->mode, MOD_MASK_MODE_ARM, MOD_MASK_MODE_VGROUP) || (mesh->totvert == 0) ||
BLI_listbase_is_empty(&mesh->vertex_group_names)) {
return mesh;
}
int defgrp_index = -1;
Array<bool> vertex_mask;
if (mmd->mode == MOD_MASK_MODE_ARM) {
Object *armature_ob = mmd->ob_arm;
/* Return input mesh if there is no armature with bones. */
2022-01-03 13:52:39 -06:00
if (ELEM(nullptr, armature_ob, armature_ob->pose)) {
return mesh;
}
vertex_mask = Array<bool>(mesh->totvert);
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
compute_vertex_mask__armature_mode(
dverts.data(), mesh, armature_ob, mmd->threshold, vertex_mask);
}
else {
BLI_assert(mmd->mode == MOD_MASK_MODE_VGROUP);
defgrp_index = BKE_id_defgroup_name_index(&mesh->id, mmd->vgroup);
/* Return input mesh if the vertex group does not exist. */
if (defgrp_index == -1) {
return mesh;
}
vertex_mask = Array<bool>(mesh->totvert);
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
compute_vertex_mask__vertex_group_mode(
dverts.data(), defgrp_index, mmd->threshold, vertex_mask);
}
if (invert_mask) {
invert_boolean_array(vertex_mask);
}
Array<int> vertex_map(mesh->totvert);
uint verts_masked_num;
compute_masked_verts(vertex_mask, vertex_map, &verts_masked_num);
Array<int> edge_map(mesh->totedge);
uint edges_masked_num;
uint verts_add_num;
if (use_interpolation) {
computed_masked_edges_smooth(mesh, vertex_mask, edge_map, &edges_masked_num, &verts_add_num);
}
else {
computed_masked_edges(mesh, vertex_mask, edge_map, &edges_masked_num);
verts_add_num = 0;
}
Vector<int> masked_poly_indices;
Vector<int> new_loop_starts;
uint polys_masked_num;
uint loops_masked_num;
computed_masked_polys(mesh,
vertex_mask,
masked_poly_indices,
new_loop_starts,
&polys_masked_num,
&loops_masked_num);
uint edges_add_num = 0;
uint polys_add_num = 0;
uint loops_add_num = 0;
if (use_interpolation) {
compute_interpolated_polys(mesh,
vertex_mask,
verts_add_num,
loops_masked_num,
masked_poly_indices,
new_loop_starts,
&edges_add_num,
&polys_add_num,
&loops_add_num);
}
Mesh *result = BKE_mesh_new_nomain_from_template(mesh,
verts_masked_num + verts_add_num,
edges_masked_num + edges_add_num,
0,
loops_masked_num + loops_add_num,
polys_masked_num + polys_add_num);
copy_masked_verts_to_new_mesh(*mesh, *result, vertex_map);
if (use_interpolation) {
add_interp_verts_copy_edges_to_new_mesh(*mesh,
*result,
vertex_mask,
vertex_map,
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
dverts.data(),
defgrp_index,
mmd->threshold,
edges_masked_num,
verts_add_num,
edge_map);
}
else {
copy_masked_edges_to_new_mesh(*mesh, *result, vertex_map, edge_map);
}
copy_masked_polys_to_new_mesh(*mesh,
*result,
vertex_map,
edge_map,
masked_poly_indices,
new_loop_starts,
polys_masked_num);
if (use_interpolation) {
add_interpolated_polys_to_new_mesh(*mesh,
*result,
vertex_mask,
vertex_map,
edge_map,
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
dverts.data(),
defgrp_index,
mmd->threshold,
masked_poly_indices,
new_loop_starts,
polys_masked_num,
edges_add_num);
}
BKE_mesh_calc_edges_loose(result);
return result;
}
static bool isDisabled(const struct Scene * /*scene*/, ModifierData *md, bool /*useRenderParams*/)
{
MaskModifierData *mmd = reinterpret_cast<MaskModifierData *>(md);
/* The object type check is only needed here in case we have a placeholder
* object assigned (because the library containing the armature is missing).
*
* In other cases it should be impossible to have a type mismatch.
*/
return mmd->ob_arm && mmd->ob_arm->type != OB_ARMATURE;
}
static void panel_draw(const bContext * /*C*/, Panel *panel)
{
uiLayout *sub, *row;
uiLayout *layout = panel->layout;
PointerRNA ob_ptr;
PointerRNA *ptr = modifier_panel_get_property_pointers(panel, &ob_ptr);
int mode = RNA_enum_get(ptr, "mode");
uiItemR(layout, ptr, "mode", UI_ITEM_R_EXPAND, nullptr, ICON_NONE);
uiLayoutSetPropSep(layout, true);
if (mode == MOD_MASK_MODE_ARM) {
row = uiLayoutRow(layout, true);
uiItemR(row, ptr, "armature", 0, nullptr, ICON_NONE);
sub = uiLayoutRow(row, true);
uiLayoutSetPropDecorate(sub, false);
uiItemR(sub, ptr, "invert_vertex_group", 0, "", ICON_ARROW_LEFTRIGHT);
}
else if (mode == MOD_MASK_MODE_VGROUP) {
modifier_vgroup_ui(layout, ptr, &ob_ptr, "vertex_group", "invert_vertex_group", nullptr);
uiItemR(layout, ptr, "use_smooth", 0, nullptr, ICON_NONE);
}
uiItemR(layout, ptr, "threshold", 0, nullptr, ICON_NONE);
modifier_panel_end(layout, ptr);
}
static void panelRegister(ARegionType *region_type)
{
modifier_panel_register(region_type, eModifierType_Mask, panel_draw);
}
ModifierTypeInfo modifierType_Mask = {
/* name */ N_("Mask"),
/* structName */ "MaskModifierData",
/* structSize */ sizeof(MaskModifierData),
/* srna */ &RNA_MaskModifier,
/* type */ eModifierTypeType_Nonconstructive,
/* flags */
(ModifierTypeFlag)(eModifierTypeFlag_AcceptsMesh | eModifierTypeFlag_SupportsMapping |
eModifierTypeFlag_SupportsEditmode),
/* icon */ ICON_MOD_MASK,
/* copyData */ BKE_modifier_copydata_generic,
/* deformVerts */ nullptr,
/* deformMatrices */ nullptr,
/* deformVertsEM */ nullptr,
/* deformMatricesEM */ nullptr,
/* modifyMesh */ modifyMesh,
/* modifyGeometrySet */ nullptr,
/* initData */ initData,
/* requiredDataMask */ requiredDataMask,
/* freeData */ nullptr,
/* isDisabled */ isDisabled,
/* updateDepsgraph */ updateDepsgraph,
/* dependsOnTime */ nullptr,
/* dependsOnNormals */ nullptr,
/* foreachIDLink */ foreachIDLink,
/* foreachTexLink */ nullptr,
/* freeRuntimeData */ nullptr,
/* panelRegister */ panelRegister,
/* blendWrite */ nullptr,
/* blendRead */ nullptr,
};