Files
test2/source/blender/io/collada/GeometryExporter.cpp

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

700 lines
22 KiB
C++
Raw Normal View History

/* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup collada
2011-02-27 20:30:35 +00:00
*/
#include <sstream>
#include "COLLADABUUtils.h"
#include "COLLADASWPrimitves.h"
#include "COLLADASWSource.h"
#include "COLLADASWVertices.h"
#include "GeometryExporter.h"
#include "DNA_meshdata_types.h"
#include "BLI_utildefines.h"
#include "BKE_attribute.hh"
#include "BKE_customdata.h"
#include "BKE_global.h"
#include "BKE_lib_id.h"
#include "BKE_material.h"
#include "BKE_mesh.h"
#include "collada_internal.h"
#include "collada_utils.h"
Mesh: Remove redundant custom data pointers For copy-on-write, we want to share attribute arrays between meshes where possible. Mutable pointers like `Mesh.mvert` make that difficult by making ownership vague. They also make code more complex by adding redundancy. The simplest solution is just removing them and retrieving layers from `CustomData` as needed. Similar changes have already been applied to curves and point clouds (e9f82d3dc7ee, 410a6efb747f). Removing use of the pointers generally makes code more obvious and more reusable. Mesh data is now accessed with a C++ API (`Mesh::edges()` or `Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`). The CoW changes this commit makes possible are described in T95845 and T95842, and started in D14139 and D14140. The change also simplifies the ongoing mesh struct-of-array refactors from T95965. **RNA/Python Access Performance** Theoretically, accessing mesh elements with the RNA API may become slower, since the layer needs to be found on every random access. However, overhead is already high enough that this doesn't make a noticible differenc, and performance is actually improved in some cases. Random access can be up to 10% faster, but other situations might be a bit slower. Generally using `foreach_get/set` are the best way to improve performance. See the differential revision for more discussion about Python performance. Cycles has been updated to use raw pointers and the internal Blender mesh types, mostly because there is no sense in having this overhead when it's already compiled with Blender. In my tests this roughly halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million face grid). Differential Revision: https://developer.blender.org/D15488
2022-09-05 11:56:34 -05:00
using blender::Span;
Partial rewrite of the Collada Module for Blender 2.8 Most important changes are in the Animation exporter and Animation Importer. There is still some cleaning up to be done. But the Exporter/Importer basically work within Blender 2.8 Some details: User Interface: The interface has been reorganized to look more like the FBX interface. New options in user interface: * keep_keyframes: When sampling the distance between 2 keyframes is defined by the sampling rate. Furthermore the keyframes defined in the FCurves are not exported. However when this option is enabled then also the defined keyframes will be added to the exported fcurves * keep_smooth_curves: When sampling we do not use FCurves. So we also have no Curve handles for smooth exporting. However when this option is enabled, Blender does its best to recreate the handles for export. This is a very experimental feature and it is know to break when: - the exported animated objects have parent inverse matrices different from the unit matrix - The exported objects have negative scaling There may be many other situations when this feature breaks. This needs to be further tested. It may be removed later or replaced by something less wonky. BlenderContext: is a new class that contains the bridge to Blender. It contains pointers to the current export/import context plus derived values of Depsgraph, Scene, Main Reporting: I reorganized the output on the Blender Console to become more informative and more readable Preservation of Item names: name attributes are now encoded with XML entities. This makes sure that i can export/import names exactly defined in the tool. This affects material names, bone names and object names. Hierarchy export: * Object and Bone Hierarchies are now exported correctly by taking the Blender parent/child hierarchy into account * Export also not selected intermediate objects Problem: When we export an Object Hierarchy, then we must export all elements of the hierarchy to maintain the transforms. This is especially important when exporting animated objects, because the animation curves are exported as relative curves based on the parent-child hierarchy. If an intermediate animated object is missing then the exported animation breaks. Solution: If the "Selected" Optioon is enabled, then take care to also export all objects which are not selected and hidden, but which are parents of selected objects. Node Based Material Importer (wip): Added basic support for Materials with diffuse color and diffuse textures. More properties (opacity, emission) need changes in the used shader. Note: Materials are all constructed by using the principled BSDF shader. Animation Exporter: * Massive optimization of the Animation Bake tool (Animation Sampler). Instead of sampling each fcurve separately, i now sample all exported fcurves simultaneously. So i avoid many (many!) scene updates during animation export. * Add support for Continuous Acceleration (Fcurve handles) This allows us to create smoother FCurves during importing Collada Animation curves. Possibly this should become an option ionstead of a fixed import feature. * Add support for sampling curves (to bake animations) * The animation sampler now can be used for any animation curve. Before the sampler only looked at curves which are supported by Standard Collada 1.4. However the Collada exporter currently ignores all animation curves which are not covered by the 1.4.1 Collada Standards. There is still some room for improvements here (work in progres) Known issues: * Some exports do currently not work reliably, among those are the camera animations, material animations and light animations those animations will be added back next (work in progres) * Exporting animation curves with keyframes (and tangents) sometimes results in odd curves (when parent inverse matrix is involved) This needs to be checked in more depth (probably it can not be solved). * Export of "all animations in scene" is disabled because the Collada Importer can not handle this reliably at the moment (work in progres). * Support for Animation Clip export Added one extra level to the exported animations such that now all scene animations are enclosed: <Animation name="id_name(ob)_Action"> <Animation>...</Animation> ... </Animation> Animation Importer: * Import of animations for objects with multiple materials When importing multiple materials for one object, the imported material animation curves have all been assigned to the first material in the object. Error handling (wip): The Importer was a bit confused as it sometimes ignored fatal parsing errors and continued to import. I did my best to unconfuse it, but i believe that this needs to be tested more. Refactoring: update : move generation of effect id names into own function update : adjust importer/exporter for no longer supported HEMI lights cleanup: Removed no lopnger existing attribute from the exporter presets cleanup: Removed not needed Context attribute from DocumentExporter fix : Avoid duplicate deletion of temporary items cleanup: fixed indentation and white space issues update : Make BCAnimation class more self contained cleanup: Renamed classes, updated comments for better reading cleanup: Moved static class functions to collada_utils cleanup: Moved typedefs to more intuitive locations cleanup: indentation and class method declarations cleanup: Removed no longer needed methods update : Moved Classes into separate files cleanup: Added comments cleanup: take care of name conventions ... : many more small changes, not helpful to list them all
2018-11-23 15:57:45 +01:00
void GeometryExporter::exportGeom()
{
Partial rewrite of the Collada Module for Blender 2.8 Most important changes are in the Animation exporter and Animation Importer. There is still some cleaning up to be done. But the Exporter/Importer basically work within Blender 2.8 Some details: User Interface: The interface has been reorganized to look more like the FBX interface. New options in user interface: * keep_keyframes: When sampling the distance between 2 keyframes is defined by the sampling rate. Furthermore the keyframes defined in the FCurves are not exported. However when this option is enabled then also the defined keyframes will be added to the exported fcurves * keep_smooth_curves: When sampling we do not use FCurves. So we also have no Curve handles for smooth exporting. However when this option is enabled, Blender does its best to recreate the handles for export. This is a very experimental feature and it is know to break when: - the exported animated objects have parent inverse matrices different from the unit matrix - The exported objects have negative scaling There may be many other situations when this feature breaks. This needs to be further tested. It may be removed later or replaced by something less wonky. BlenderContext: is a new class that contains the bridge to Blender. It contains pointers to the current export/import context plus derived values of Depsgraph, Scene, Main Reporting: I reorganized the output on the Blender Console to become more informative and more readable Preservation of Item names: name attributes are now encoded with XML entities. This makes sure that i can export/import names exactly defined in the tool. This affects material names, bone names and object names. Hierarchy export: * Object and Bone Hierarchies are now exported correctly by taking the Blender parent/child hierarchy into account * Export also not selected intermediate objects Problem: When we export an Object Hierarchy, then we must export all elements of the hierarchy to maintain the transforms. This is especially important when exporting animated objects, because the animation curves are exported as relative curves based on the parent-child hierarchy. If an intermediate animated object is missing then the exported animation breaks. Solution: If the "Selected" Optioon is enabled, then take care to also export all objects which are not selected and hidden, but which are parents of selected objects. Node Based Material Importer (wip): Added basic support for Materials with diffuse color and diffuse textures. More properties (opacity, emission) need changes in the used shader. Note: Materials are all constructed by using the principled BSDF shader. Animation Exporter: * Massive optimization of the Animation Bake tool (Animation Sampler). Instead of sampling each fcurve separately, i now sample all exported fcurves simultaneously. So i avoid many (many!) scene updates during animation export. * Add support for Continuous Acceleration (Fcurve handles) This allows us to create smoother FCurves during importing Collada Animation curves. Possibly this should become an option ionstead of a fixed import feature. * Add support for sampling curves (to bake animations) * The animation sampler now can be used for any animation curve. Before the sampler only looked at curves which are supported by Standard Collada 1.4. However the Collada exporter currently ignores all animation curves which are not covered by the 1.4.1 Collada Standards. There is still some room for improvements here (work in progres) Known issues: * Some exports do currently not work reliably, among those are the camera animations, material animations and light animations those animations will be added back next (work in progres) * Exporting animation curves with keyframes (and tangents) sometimes results in odd curves (when parent inverse matrix is involved) This needs to be checked in more depth (probably it can not be solved). * Export of "all animations in scene" is disabled because the Collada Importer can not handle this reliably at the moment (work in progres). * Support for Animation Clip export Added one extra level to the exported animations such that now all scene animations are enclosed: <Animation name="id_name(ob)_Action"> <Animation>...</Animation> ... </Animation> Animation Importer: * Import of animations for objects with multiple materials When importing multiple materials for one object, the imported material animation curves have all been assigned to the first material in the object. Error handling (wip): The Importer was a bit confused as it sometimes ignored fatal parsing errors and continued to import. I did my best to unconfuse it, but i believe that this needs to be tested more. Refactoring: update : move generation of effect id names into own function update : adjust importer/exporter for no longer supported HEMI lights cleanup: Removed no lopnger existing attribute from the exporter presets cleanup: Removed not needed Context attribute from DocumentExporter fix : Avoid duplicate deletion of temporary items cleanup: fixed indentation and white space issues update : Make BCAnimation class more self contained cleanup: Renamed classes, updated comments for better reading cleanup: Moved static class functions to collada_utils cleanup: Moved typedefs to more intuitive locations cleanup: indentation and class method declarations cleanup: Removed no longer needed methods update : Moved Classes into separate files cleanup: Added comments cleanup: take care of name conventions ... : many more small changes, not helpful to list them all
2018-11-23 15:57:45 +01:00
Scene *sce = blender_context.get_scene();
openLibrary();
GeometryFunctor gf;
gf.forEachMeshObjectInExportSet<GeometryExporter>(
sce, *this, this->export_settings.get_export_set());
closeLibrary();
}
void GeometryExporter::operator()(Object *ob)
{
bool use_instantiation = this->export_settings.get_use_object_instantiation();
Mesh *me = bc_get_mesh_copy(blender_context,
ob,
this->export_settings.get_export_mesh_type(),
this->export_settings.get_apply_modifiers(),
this->export_settings.get_triangulate());
std::string geom_id = get_geometry_id(ob, use_instantiation);
std::vector<Normal> nor;
std::vector<BCPolygonNormalsIndices> norind;
/* Skip if linked geometry was already exported from another reference */
2012-06-12 22:05:33 +00:00
if (use_instantiation && exportedGeometry.find(geom_id) != exportedGeometry.end()) {
return;
2012-06-12 22:05:33 +00:00
}
std::string geom_name = (use_instantiation) ? id_name(ob->data) : id_name(ob);
Partial rewrite of the Collada Module for Blender 2.8 Most important changes are in the Animation exporter and Animation Importer. There is still some cleaning up to be done. But the Exporter/Importer basically work within Blender 2.8 Some details: User Interface: The interface has been reorganized to look more like the FBX interface. New options in user interface: * keep_keyframes: When sampling the distance between 2 keyframes is defined by the sampling rate. Furthermore the keyframes defined in the FCurves are not exported. However when this option is enabled then also the defined keyframes will be added to the exported fcurves * keep_smooth_curves: When sampling we do not use FCurves. So we also have no Curve handles for smooth exporting. However when this option is enabled, Blender does its best to recreate the handles for export. This is a very experimental feature and it is know to break when: - the exported animated objects have parent inverse matrices different from the unit matrix - The exported objects have negative scaling There may be many other situations when this feature breaks. This needs to be further tested. It may be removed later or replaced by something less wonky. BlenderContext: is a new class that contains the bridge to Blender. It contains pointers to the current export/import context plus derived values of Depsgraph, Scene, Main Reporting: I reorganized the output on the Blender Console to become more informative and more readable Preservation of Item names: name attributes are now encoded with XML entities. This makes sure that i can export/import names exactly defined in the tool. This affects material names, bone names and object names. Hierarchy export: * Object and Bone Hierarchies are now exported correctly by taking the Blender parent/child hierarchy into account * Export also not selected intermediate objects Problem: When we export an Object Hierarchy, then we must export all elements of the hierarchy to maintain the transforms. This is especially important when exporting animated objects, because the animation curves are exported as relative curves based on the parent-child hierarchy. If an intermediate animated object is missing then the exported animation breaks. Solution: If the "Selected" Optioon is enabled, then take care to also export all objects which are not selected and hidden, but which are parents of selected objects. Node Based Material Importer (wip): Added basic support for Materials with diffuse color and diffuse textures. More properties (opacity, emission) need changes in the used shader. Note: Materials are all constructed by using the principled BSDF shader. Animation Exporter: * Massive optimization of the Animation Bake tool (Animation Sampler). Instead of sampling each fcurve separately, i now sample all exported fcurves simultaneously. So i avoid many (many!) scene updates during animation export. * Add support for Continuous Acceleration (Fcurve handles) This allows us to create smoother FCurves during importing Collada Animation curves. Possibly this should become an option ionstead of a fixed import feature. * Add support for sampling curves (to bake animations) * The animation sampler now can be used for any animation curve. Before the sampler only looked at curves which are supported by Standard Collada 1.4. However the Collada exporter currently ignores all animation curves which are not covered by the 1.4.1 Collada Standards. There is still some room for improvements here (work in progres) Known issues: * Some exports do currently not work reliably, among those are the camera animations, material animations and light animations those animations will be added back next (work in progres) * Exporting animation curves with keyframes (and tangents) sometimes results in odd curves (when parent inverse matrix is involved) This needs to be checked in more depth (probably it can not be solved). * Export of "all animations in scene" is disabled because the Collada Importer can not handle this reliably at the moment (work in progres). * Support for Animation Clip export Added one extra level to the exported animations such that now all scene animations are enclosed: <Animation name="id_name(ob)_Action"> <Animation>...</Animation> ... </Animation> Animation Importer: * Import of animations for objects with multiple materials When importing multiple materials for one object, the imported material animation curves have all been assigned to the first material in the object. Error handling (wip): The Importer was a bit confused as it sometimes ignored fatal parsing errors and continued to import. I did my best to unconfuse it, but i believe that this needs to be tested more. Refactoring: update : move generation of effect id names into own function update : adjust importer/exporter for no longer supported HEMI lights cleanup: Removed no lopnger existing attribute from the exporter presets cleanup: Removed not needed Context attribute from DocumentExporter fix : Avoid duplicate deletion of temporary items cleanup: fixed indentation and white space issues update : Make BCAnimation class more self contained cleanup: Renamed classes, updated comments for better reading cleanup: Moved static class functions to collada_utils cleanup: Moved typedefs to more intuitive locations cleanup: indentation and class method declarations cleanup: Removed no longer needed methods update : Moved Classes into separate files cleanup: Added comments cleanup: take care of name conventions ... : many more small changes, not helpful to list them all
2018-11-23 15:57:45 +01:00
geom_name = encode_xml(geom_name);
exportedGeometry.insert(geom_id);
bool has_color = bool(CustomData_has_layer(&me->fdata, CD_MCOL));
create_normals(nor, norind, me);
/* openMesh(geoId, geoName, meshId) */
openMesh(geom_id, geom_name);
/* writes <source> for vertex coords */
createVertsSource(geom_id, me);
/* writes <source> for normal coords */
createNormalsSource(geom_id, me, nor);
bool has_uvs = bool(CustomData_has_layer(&me->ldata, CD_MLOOPUV));
/* writes <source> for uv coords if mesh has uv coords */
if (has_uvs) {
createTexcoordsSource(geom_id, me);
}
if (has_color) {
createVertexColorSource(geom_id, me);
}
/* <vertices> */
COLLADASW::Vertices verts(mSW);
verts.setId(getIdBySemantics(geom_id, COLLADASW::InputSemantic::VERTEX));
COLLADASW::InputList &input_list = verts.getInputList();
COLLADASW::Input input(COLLADASW::InputSemantic::POSITION,
getUrlBySemantics(geom_id, COLLADASW::InputSemantic::POSITION));
input_list.push_back(input);
verts.add();
createLooseEdgeList(ob, me, geom_id);
/* Only create Polylists if number of faces > 0 */
if (me->totface > 0) {
/* XXX slow */
if (ob->totcol) {
for (int a = 0; a < ob->totcol; a++) {
create_mesh_primitive_list(a, has_uvs, has_color, ob, me, geom_id, norind);
}
}
else {
create_mesh_primitive_list(0, has_uvs, has_color, ob, me, geom_id, norind);
}
}
closeMesh();
closeGeometry();
if (this->export_settings.get_include_shapekeys()) {
2018-10-11 08:49:28 +11:00
Key *key = BKE_key_from_object(ob);
2013-02-10 17:06:05 +00:00
if (key) {
blender::MutableSpan<MVert> verts = me->verts_for_write();
2018-10-11 08:49:28 +11:00
KeyBlock *kb = (KeyBlock *)key->block.first;
/* skip the basis */
kb = kb->next;
for (; kb; kb = kb->next) {
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
BKE_keyblock_convert_to_mesh(kb, verts.data(), me->totvert);
export_key_mesh(ob, me, kb);
}
}
}
BKE_id_free(nullptr, me);
}
void GeometryExporter::export_key_mesh(Object *ob, Mesh *me, KeyBlock *kb)
2013-02-10 17:06:05 +00:00
{
std::string geom_id = get_geometry_id(ob, false) + "_morph_" + translate_id(kb->name);
std::vector<Normal> nor;
std::vector<BCPolygonNormalsIndices> norind;
if (exportedGeometry.find(geom_id) != exportedGeometry.end()) {
return;
}
std::string geom_name = kb->name;
exportedGeometry.insert(geom_id);
bool has_color = bool(CustomData_has_layer(&me->fdata, CD_MCOL));
create_normals(nor, norind, me);
// openMesh(geoId, geoName, meshId)
openMesh(geom_id, geom_name);
/* writes <source> for vertex coords */
createVertsSource(geom_id, me);
/* writes <source> for normal coords */
createNormalsSource(geom_id, me, nor);
bool has_uvs = bool(CustomData_has_layer(&me->ldata, CD_MLOOPUV));
/* writes <source> for uv coords if mesh has uv coords */
if (has_uvs) {
createTexcoordsSource(geom_id, me);
}
if (has_color) {
createVertexColorSource(geom_id, me);
}
/* <vertices> */
COLLADASW::Vertices verts(mSW);
verts.setId(getIdBySemantics(geom_id, COLLADASW::InputSemantic::VERTEX));
COLLADASW::InputList &input_list = verts.getInputList();
COLLADASW::Input input(COLLADASW::InputSemantic::POSITION,
getUrlBySemantics(geom_id, COLLADASW::InputSemantic::POSITION));
input_list.push_back(input);
verts.add();
// createLooseEdgeList(ob, me, geom_id, norind);
/* XXX slow */
if (ob->totcol) {
for (int a = 0; a < ob->totcol; a++) {
create_mesh_primitive_list(a, has_uvs, has_color, ob, me, geom_id, norind);
}
}
else {
create_mesh_primitive_list(0, has_uvs, has_color, ob, me, geom_id, norind);
}
closeMesh();
closeGeometry();
}
void GeometryExporter::createLooseEdgeList(Object *ob, Mesh *me, std::string &geom_id)
{
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 = me->edges();
int totedges = me->totedge;
int edges_in_linelist = 0;
std::vector<uint> edge_list;
int index;
/* Find all loose edges in Mesh
* and save vertex indices in edge_list */
for (index = 0; index < totedges; 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 *edge = &edges[index];
if (edge->flag & ME_LOOSEEDGE) {
edges_in_linelist += 1;
edge_list.push_back(edge->v1);
edge_list.push_back(edge->v2);
}
}
if (edges_in_linelist > 0) {
/* Create the list of loose edges */
COLLADASW::Lines lines(mSW);
lines.setCount(edges_in_linelist);
COLLADASW::InputList &til = lines.getInputList();
/* creates <input> in <lines> for vertices */
COLLADASW::Input input1(COLLADASW::InputSemantic::VERTEX,
getUrlBySemantics(geom_id, COLLADASW::InputSemantic::VERTEX),
0);
til.push_back(input1);
lines.prepareToAppendValues();
for (index = 0; index < edges_in_linelist; index++) {
2012-06-12 22:05:33 +00:00
lines.appendValues(edge_list[2 * index + 1]);
lines.appendValues(edge_list[2 * index]);
}
lines.finish();
}
}
static void prepareToAppendValues(bool is_triangulated,
COLLADASW::PrimitivesBase &primitive_list,
std::vector<ulong> &vcount_list)
{
/* performs the actual writing */
if (is_triangulated) {
((COLLADASW::Triangles &)primitive_list).prepareToAppendValues();
}
else {
/* sets <vcount> */
primitive_list.setVCountList(vcount_list);
((COLLADASW::Polylist &)primitive_list).prepareToAppendValues();
}
}
static void finish_and_delete_primitive_List(bool is_triangulated,
COLLADASW::PrimitivesBase *primitive_list)
{
if (is_triangulated) {
((COLLADASW::Triangles *)primitive_list)->finish();
}
else {
((COLLADASW::Polylist *)primitive_list)->finish();
}
delete primitive_list;
}
static COLLADASW::PrimitivesBase *create_primitive_list(bool is_triangulated,
COLLADASW::StreamWriter *mSW)
{
COLLADASW::PrimitivesBase *primitive_list;
if (is_triangulated) {
primitive_list = new COLLADASW::Triangles(mSW);
}
else {
primitive_list = new COLLADASW::Polylist(mSW);
}
return primitive_list;
}
static bool collect_vertex_counts_per_poly(Mesh *me,
int material_index,
std::vector<ulong> &vcount_list)
{
const Span<MPoly> polys = me->polys();
const blender::bke::AttributeAccessor attributes = me->attributes();
const blender::VArray<int> material_indices = attributes.lookup_or_default<int>(
"material_index", ATTR_DOMAIN_FACE, 0);
bool is_triangulated = true;
/* Expecting that the material index is always 0 if the mesh has no materials assigned */
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 : polys.index_range()) {
if (material_indices[i] == material_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 MPoly &poly = polys[i];
const int vertex_count = poly.totloop;
vcount_list.push_back(vertex_count);
if (vertex_count != 3) {
is_triangulated = false;
}
}
}
return is_triangulated;
}
std::string GeometryExporter::makeVertexColorSourceId(std::string &geom_id, char *layer_name)
{
std::string result = getIdBySemantics(geom_id, COLLADASW::InputSemantic::COLOR) + "-" +
layer_name;
return result;
}
void GeometryExporter::create_mesh_primitive_list(short material_index,
bool has_uvs,
bool has_color,
Object *ob,
Mesh *me,
std::string &geom_id,
std::vector<BCPolygonNormalsIndices> &norind)
{
const Span<MPoly> polys = me->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 = me->loops();
std::vector<ulong> vcount_list;
bool is_triangulated = collect_vertex_counts_per_poly(me, material_index, vcount_list);
int polygon_count = vcount_list.size();
/* no faces using this material */
if (polygon_count == 0) {
fprintf(
stderr, "%s: material with index %d is not used.\n", id_name(ob).c_str(), material_index);
return;
}
Material *ma = ob->totcol ? BKE_object_material_get(ob, material_index + 1) : nullptr;
COLLADASW::PrimitivesBase *primitive_list = create_primitive_list(is_triangulated, mSW);
/* sets count attribute in <polylist> */
primitive_list->setCount(polygon_count);
/* sets material name */
if (ma) {
std::string material_id = get_material_id(ma);
std::ostringstream ostr;
ostr << translate_id(material_id);
primitive_list->setMaterial(ostr.str());
}
COLLADASW::Input vertex_input(COLLADASW::InputSemantic::VERTEX,
getUrlBySemantics(geom_id, COLLADASW::InputSemantic::VERTEX),
0);
COLLADASW::Input normals_input(COLLADASW::InputSemantic::NORMAL,
getUrlBySemantics(geom_id, COLLADASW::InputSemantic::NORMAL),
1);
COLLADASW::InputList &til = primitive_list->getInputList();
til.push_back(vertex_input);
til.push_back(normals_input);
/* if mesh has uv coords writes <input> for TEXCOORD */
int num_layers = CustomData_number_of_layers(&me->ldata, CD_MLOOPUV);
int active_uv_index = CustomData_get_active_layer_index(&me->ldata, CD_MLOOPUV);
for (int i = 0; i < num_layers; i++) {
int layer_index = CustomData_get_layer_index_n(&me->ldata, CD_MLOOPUV, i);
if (!this->export_settings.get_active_uv_only() || layer_index == active_uv_index) {
// char *name = CustomData_get_layer_name(&me->ldata, CD_MLOOPUV, i);
COLLADASW::Input texcoord_input(
COLLADASW::InputSemantic::TEXCOORD,
makeUrl(makeTexcoordSourceId(geom_id, i, this->export_settings.get_active_uv_only())),
2, /* this is only until we have optimized UV sets */
(this->export_settings.get_active_uv_only()) ? 0 : layer_index - 1 /* set (0,1,2,...) */
);
til.push_back(texcoord_input);
}
}
int totlayer_mcol = CustomData_number_of_layers(&me->ldata, CD_PROP_BYTE_COLOR);
if (totlayer_mcol > 0) {
int map_index = 0;
for (int a = 0; a < totlayer_mcol; a++) {
char *layer_name = bc_CustomData_get_layer_name(&me->ldata, CD_PROP_BYTE_COLOR, a);
COLLADASW::Input input4(COLLADASW::InputSemantic::COLOR,
makeUrl(makeVertexColorSourceId(geom_id, layer_name)),
(has_uvs) ? 3 : 2, /* all color layers have same index order */
map_index /* set number equals color map index */
);
til.push_back(input4);
map_index++;
}
}
/* performs the actual writing */
prepareToAppendValues(is_triangulated, *primitive_list, vcount_list);
const blender::bke::AttributeAccessor attributes = me->attributes();
const blender::VArray<int> material_indices = attributes.lookup_or_default<int>(
"material_index", ATTR_DOMAIN_FACE, 0);
/* <p> */
int texindex = 0;
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 : polys.index_range()) {
const MPoly *p = &polys[i];
int loop_count = p->totloop;
if (material_indices[i] == material_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 MLoop *l = &loops[p->loopstart];
BCPolygonNormalsIndices normal_indices = norind[i];
for (int j = 0; j < loop_count; j++) {
primitive_list->appendValues(l[j].v);
primitive_list->appendValues(normal_indices[j]);
if (has_uvs) {
primitive_list->appendValues(texindex + j);
}
if (has_color) {
primitive_list->appendValues(texindex + j);
}
}
}
texindex += loop_count;
}
finish_and_delete_primitive_List(is_triangulated, primitive_list);
}
void GeometryExporter::createVertsSource(std::string geom_id, Mesh *me)
{
const Span<MVert> verts = me->verts();
COLLADASW::FloatSourceF source(mSW);
source.setId(getIdBySemantics(geom_id, COLLADASW::InputSemantic::POSITION));
source.setArrayId(getIdBySemantics(geom_id, COLLADASW::InputSemantic::POSITION) +
2012-06-12 22:05:33 +00:00
ARRAY_ID_SUFFIX);
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
source.setAccessorCount(verts.size());
source.setAccessorStride(3);
COLLADASW::SourceBase::ParameterNameList &param = source.getParameterNameList();
param.push_back("X");
param.push_back("Y");
param.push_back("Z");
2012-10-04 13:26:15 +00:00
/* main function, it creates <source id = "">, <float_array id = ""
* count = ""> */
source.prepareToAppendValues();
/* appends data to <float_array> */
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 : verts.index_range()) {
Vector co;
if (export_settings.get_apply_global_orientation()) {
bc_add_global_transform(co, verts[i].co, export_settings.get_global_transform());
}
else {
copy_v3_v3(co, verts[i].co);
}
source.appendValues(co[0], co[1], co[2]);
}
source.finish();
}
void GeometryExporter::createVertexColorSource(std::string geom_id, Mesh *me)
{
/* Find number of vertex color layers */
int totlayer_mcol = CustomData_number_of_layers(&me->ldata, CD_PROP_BYTE_COLOR);
if (totlayer_mcol == 0) {
return;
}
int map_index = 0;
for (int a = 0; a < totlayer_mcol; a++) {
map_index++;
const MLoopCol *mloopcol = (const MLoopCol *)CustomData_get_layer_n(
&me->ldata, CD_PROP_BYTE_COLOR, a);
COLLADASW::FloatSourceF source(mSW);
char *layer_name = bc_CustomData_get_layer_name(&me->ldata, CD_PROP_BYTE_COLOR, a);
std::string layer_id = makeVertexColorSourceId(geom_id, layer_name);
source.setId(layer_id);
source.setNodeName(layer_name);
source.setArrayId(layer_id + ARRAY_ID_SUFFIX);
source.setAccessorCount(me->totloop);
2017-09-29 18:12:14 +10:00
source.setAccessorStride(4);
COLLADASW::SourceBase::ParameterNameList &param = source.getParameterNameList();
param.push_back("R");
param.push_back("G");
param.push_back("B");
2017-09-29 18:12:14 +10:00
param.push_back("A");
source.prepareToAppendValues();
const Span<MPoly> polys = me->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
for (const int i : polys.index_range()) {
const MPoly &poly = polys[i];
const MLoopCol *mlc = mloopcol + poly.loopstart;
for (int j = 0; j < poly.totloop; j++, mlc++) {
source.appendValues(mlc->r / 255.0f, mlc->g / 255.0f, mlc->b / 255.0f, mlc->a / 255.0f);
}
}
source.finish();
}
}
std::string GeometryExporter::makeTexcoordSourceId(std::string &geom_id,
int layer_index,
bool is_single_layer)
{
char suffix[20];
if (is_single_layer) {
suffix[0] = '\0';
}
else {
sprintf(suffix, "-%d", layer_index);
}
return getIdBySemantics(geom_id, COLLADASW::InputSemantic::TEXCOORD) + suffix;
}
void GeometryExporter::createTexcoordsSource(std::string geom_id, Mesh *me)
{
int totuv = me->totloop;
const Span<MPoly> polys = me->polys();
int num_layers = CustomData_number_of_layers(&me->ldata, CD_MLOOPUV);
/* write <source> for each layer
* each <source> will get id like meshName + "map-channel-1" */
int active_uv_index = CustomData_get_active_layer_index(&me->ldata, CD_MLOOPUV);
for (int a = 0; a < num_layers; a++) {
int layer_index = CustomData_get_layer_index_n(&me->ldata, CD_MLOOPUV, a);
if (!this->export_settings.get_active_uv_only() || layer_index == active_uv_index) {
MLoopUV *mloops = (MLoopUV *)CustomData_get_layer_n(&me->ldata, CD_MLOOPUV, a);
COLLADASW::FloatSourceF source(mSW);
std::string layer_id = makeTexcoordSourceId(
geom_id, a, this->export_settings.get_active_uv_only());
source.setId(layer_id);
source.setArrayId(layer_id + ARRAY_ID_SUFFIX);
source.setAccessorCount(totuv);
source.setAccessorStride(2);
COLLADASW::SourceBase::ParameterNameList &param = source.getParameterNameList();
param.push_back("S");
param.push_back("T");
source.prepareToAppendValues();
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 : polys.index_range()) {
const MPoly *mpoly = &polys[i];
MLoopUV *mloop = mloops + mpoly->loopstart;
for (int j = 0; j < mpoly->totloop; j++) {
source.appendValues(mloop[j].uv[0], mloop[j].uv[1]);
}
}
source.finish();
}
}
}
bool operator<(const Normal &a, const Normal &b)
{
/* Only needed to sort normal vectors and find() them later in a map. */
return a.x < b.x || (a.x == b.x && (a.y < b.y || (a.y == b.y && a.z < b.z)));
}
void GeometryExporter::createNormalsSource(std::string geom_id, Mesh *me, std::vector<Normal> &nor)
{
#if 0
int totverts = dm->getNumVerts(dm);
MVert *verts = dm->getVertArray(dm);
#endif
COLLADASW::FloatSourceF source(mSW);
source.setId(getIdBySemantics(geom_id, COLLADASW::InputSemantic::NORMAL));
source.setArrayId(getIdBySemantics(geom_id, COLLADASW::InputSemantic::NORMAL) + ARRAY_ID_SUFFIX);
source.setAccessorCount(ulong(nor.size()));
source.setAccessorStride(3);
COLLADASW::SourceBase::ParameterNameList &param = source.getParameterNameList();
param.push_back("X");
param.push_back("Y");
param.push_back("Z");
source.prepareToAppendValues();
std::vector<Normal>::iterator it;
for (it = nor.begin(); it != nor.end(); it++) {
Normal &n = *it;
Vector no{n.x, n.y, n.z};
if (export_settings.get_apply_global_orientation()) {
bc_add_global_transform(no, export_settings.get_global_transform());
}
source.appendValues(no[0], no[1], no[2]);
}
source.finish();
}
void GeometryExporter::create_normals(std::vector<Normal> &normals,
std::vector<BCPolygonNormalsIndices> &polygons_normals,
Mesh *me)
{
std::map<Normal, uint> shared_normal_indices;
int last_normal_index = -1;
const Span<MVert> verts = me->verts();
Refactor: Move normals out of MVert, lazy calculation As described in T91186, this commit moves mesh vertex normals into a contiguous array of float vectors in a custom data layer, how face normals are currently stored. The main interface is documented in `BKE_mesh.h`. Vertex and face normals are now calculated on-demand and cached, retrieved with an "ensure" function. Since the logical state of a mesh is now "has normals when necessary", they can be retrieved from a `const` mesh. The goal is to use on-demand calculation for all derived data, but leave room for eager calculation for performance purposes (modifier evaluation is threaded, but viewport data generation is not). **Benefits** This moves us closer to a SoA approach rather than the current AoS paradigm. Accessing a contiguous `float3` is much more efficient than retrieving data from a larger struct. The memory requirements for accessing only normals or vertex locations are smaller, and at the cost of more memory usage for just normals, they now don't have to be converted between float and short, which also simplifies code In the future, the remaining items can be removed from `MVert`, leaving only `float3`, which has similar benefits (see T93602). Removing the combination of derived and original data makes it conceptually simpler to only calculate normals when necessary. This is especially important now that we have more opportunities for temporary meshes in geometry nodes. **Performance** In addition to the theoretical future performance improvements by making `MVert == float3`, I've done some basic performance testing on this patch directly. The data is fairly rough, but it gives an idea about where things stand generally. - Mesh line primitive 4m Verts: 1.16x faster (36 -> 31 ms), showing that accessing just `MVert` is now more efficient. - Spring Splash Screen: 1.03-1.06 -> 1.06-1.11 FPS, a very slight change that at least shows there is no regression. - Sprite Fright Snail Smoosh: 3.30-3.40 -> 3.42-3.50 FPS, a small but observable speedup. - Set Position Node with Scaled Normal: 1.36x faster (53 -> 39 ms), shows that using normals in geometry nodes is faster. - Normal Calculation 1.6m Vert Cube: 1.19x faster (25 -> 21 ms), shows that calculating normals is slightly faster now. - File Size of 1.6m Vert Cube: 1.03x smaller (214.7 -> 208.4 MB), Normals are not saved in files, which can help with large meshes. As for memory usage, it may be slightly more in some cases, but I didn't observe any difference in the production files I tested. **Tests** Some modifiers and cycles test results need to be updated with this commit, for two reasons: - The subdivision surface modifier is not responsible for calculating normals anymore. In master, the modifier creates different normals than the result of the `Mesh` normal calculation, so this is a bug fix. - There are small differences in the results of some modifiers that use normals because they are not converted to and from `short` anymore. **Future improvements** - Remove `ModifierTypeInfo::dependsOnNormals`. Code in each modifier already retrieves normals if they are needed anyway. - Copy normals as part of a better CoW system for attributes. - Make more areas use lazy instead of eager normal calculation. - Remove `BKE_mesh_normals_tag_dirty` in more places since that is now the default state of a new mesh. - Possibly apply a similar change to derived face corner normals. Differential Revision: https://developer.blender.org/D12770
2022-01-13 14:37:58 -06:00
const float(*vert_normals)[3] = BKE_mesh_vertex_normals_ensure(me);
const Span<MPoly> polys = me->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 = me->loops();
const float(*lnors)[3] = nullptr;
bool use_custom_normals = false;
BKE_mesh_calc_normals_split(me);
if (CustomData_has_layer(&me->ldata, CD_NORMAL)) {
lnors = (float(*)[3])CustomData_get_layer(&me->ldata, CD_NORMAL);
use_custom_normals = 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
for (const int poly_index : polys.index_range()) {
const MPoly *mpoly = &polys[poly_index];
bool use_vertex_normals = use_custom_normals || mpoly->flag & ME_SMOOTH;
if (!use_vertex_normals) {
/* For flat faces use face normal as vertex normal: */
float vector[3];
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
BKE_mesh_calc_poly_normal(mpoly, &loops[mpoly->loopstart], verts.data(), vector);
Normal n = {vector[0], vector[1], vector[2]};
normals.push_back(n);
last_normal_index++;
}
BCPolygonNormalsIndices poly_indices;
for (int loop_index = 0; loop_index < mpoly->totloop; loop_index++) {
uint loop_idx = mpoly->loopstart + loop_index;
if (use_vertex_normals) {
float normalized[3];
if (use_custom_normals) {
normalize_v3_v3(normalized, lnors[loop_idx]);
}
else {
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
copy_v3_v3(normalized, vert_normals[loops[loop_index].v]);
normalize_v3(normalized);
}
Normal n = {normalized[0], normalized[1], normalized[2]};
if (shared_normal_indices.find(n) != shared_normal_indices.end()) {
poly_indices.add_index(shared_normal_indices[n]);
}
else {
last_normal_index++;
poly_indices.add_index(last_normal_index);
shared_normal_indices[n] = last_normal_index;
normals.push_back(n);
}
}
else {
poly_indices.add_index(last_normal_index);
}
}
polygons_normals.push_back(poly_indices);
}
}
std::string GeometryExporter::getIdBySemantics(std::string geom_id,
COLLADASW::InputSemantic::Semantics type,
std::string other_suffix)
{
return geom_id + getSuffixBySemantic(type) + other_suffix;
}
COLLADASW::URI GeometryExporter::getUrlBySemantics(std::string geom_id,
COLLADASW::InputSemantic::Semantics type,
std::string other_suffix)
{
std::string id(getIdBySemantics(geom_id, type, other_suffix));
return COLLADASW::URI(COLLADABU::Utils::EMPTY_STRING, id);
}
COLLADASW::URI GeometryExporter::makeUrl(std::string id)
{
return COLLADASW::URI(COLLADABU::Utils::EMPTY_STRING, id);
}