Files
test/source/blender/geometry/intern/mesh_to_volume.cc

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

229 lines
8.7 KiB
C++
Raw Normal View History

/* SPDX-FileCopyrightText: 2023 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
#include "BLI_math_matrix.hh"
#include "BLI_task.hh"
#include "BKE_volume.hh"
#include "BKE_volume_grid.hh"
#include "BKE_volume_openvdb.hh"
#include "GEO_mesh_to_volume.hh"
#ifdef WITH_OPENVDB
# include <algorithm>
# include <openvdb/openvdb.h>
# include <openvdb/tools/GridTransformer.h>
# include <openvdb/tools/LevelSetUtil.h>
# include <openvdb/tools/VolumeToMesh.h>
namespace blender::geometry {
/* This class follows the MeshDataAdapter interface from openvdb. */
class OpenVDBMeshAdapter {
private:
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
Span<float3> positions_;
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
Span<int> corner_verts_;
Span<int3> corner_tris_;
float4x4 transform_;
public:
OpenVDBMeshAdapter(const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const float4x4 &transform);
size_t polygonCount() const;
size_t pointCount() const;
size_t vertexCount(size_t /*polygon_index*/) const;
void getIndexSpacePoint(size_t polygon_index, size_t vertex_index, openvdb::Vec3d &pos) const;
};
OpenVDBMeshAdapter::OpenVDBMeshAdapter(const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const float4x4 &transform)
: positions_(positions),
corner_verts_(corner_verts),
corner_tris_(corner_tris),
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
transform_(transform)
{
}
size_t OpenVDBMeshAdapter::polygonCount() const
{
return size_t(corner_tris_.size());
}
size_t OpenVDBMeshAdapter::pointCount() const
{
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
return size_t(positions_.size());
}
size_t OpenVDBMeshAdapter::vertexCount(size_t /*polygon_index*/) const
{
/* All polygons are triangles. */
return 3;
}
void OpenVDBMeshAdapter::getIndexSpacePoint(size_t polygon_index,
size_t vertex_index,
openvdb::Vec3d &pos) const
{
const int3 &tri = corner_tris_[polygon_index];
const float3 transformed_co = math::transform_point(
transform_, positions_[corner_verts_[tri[vertex_index]]]);
pos = &transformed_co.x;
}
float volume_compute_voxel_size(const Depsgraph *depsgraph,
const FunctionRef<Bounds<float3>()> bounds_fn,
const MeshToVolumeResolution res,
const float exterior_band_width,
const float4x4 &transform)
{
const float volume_simplify = BKE_volume_simplify_factor(depsgraph);
if (volume_simplify == 0.0f) {
return 0.0f;
}
if (res.mode == MESH_TO_VOLUME_RESOLUTION_MODE_VOXEL_SIZE) {
return res.settings.voxel_size / volume_simplify;
}
if (res.settings.voxel_amount <= 0) {
return 0;
}
const Bounds<float3> bounds = bounds_fn();
/* Compute the diagonal of the bounding box. This is used because
* it will always be bigger than the widest side of the mesh. */
const float diagonal = math::distance(math::transform_point(transform, bounds.min),
math::transform_point(transform, bounds.max));
/* To get the approximate size per voxel, first subtract the exterior band from the requested
* voxel amount, then divide the diagonal with this value if it's bigger than 1. */
const float voxel_size =
(diagonal / std::max(1.0f, float(res.settings.voxel_amount) - 2.0f * exterior_band_width));
/* Return the simplified voxel size. */
return voxel_size / volume_simplify;
}
static openvdb::FloatGrid::Ptr mesh_to_density_grid_impl(
const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const float4x4 &mesh_to_volume_space_transform,
const float voxel_size,
const float interior_band_width,
const float density)
{
if (voxel_size < 1e-5f) {
return nullptr;
}
float4x4 mesh_to_index_space_transform = math::from_scale<float4x4>(float3(1.0f / voxel_size));
mesh_to_index_space_transform *= mesh_to_volume_space_transform;
/* Better align generated grid with the source mesh. */
mesh_to_index_space_transform.location() -= 0.5f;
OpenVDBMeshAdapter mesh_adapter{
positions, corner_verts, corner_tris, mesh_to_index_space_transform};
const float interior = std::max(1.0f, interior_band_width / voxel_size);
openvdb::math::Transform::Ptr transform = openvdb::math::Transform::createLinearTransform(
voxel_size);
openvdb::FloatGrid::Ptr new_grid = openvdb::tools::meshToVolume<openvdb::FloatGrid>(
mesh_adapter, *transform, 1.0f, interior);
openvdb::tools::sdfToFogVolume(*new_grid);
if (density != 1.0f) {
openvdb::tools::foreach(new_grid->beginValueOn(),
[&](const openvdb::FloatGrid::ValueOnIter &iter) {
iter.modifyValue([&](float &value) { value *= density; });
});
}
return new_grid;
}
bke::VolumeGrid<float> mesh_to_density_grid(const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const float voxel_size,
const float interior_band_width,
const float density)
{
openvdb::FloatGrid::Ptr grid = mesh_to_density_grid_impl(positions,
corner_verts,
corner_tris,
float4x4::identity(),
voxel_size,
interior_band_width,
density);
if (!grid) {
return {};
}
return bke::VolumeGrid<float>(std::move(grid));
}
bke::VolumeGrid<float> mesh_to_sdf_grid(const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
const float voxel_size,
const float half_band_width)
{
if (voxel_size <= 0.0f || half_band_width <= 0.0f) {
return {};
}
std::vector<openvdb::Vec3s> points(positions.size());
std::vector<openvdb::Vec3I> triangles(corner_tris.size());
threading::parallel_for(positions.index_range(), 2048, [&](const IndexRange range) {
for (const int i : range) {
const float3 &co = positions[i];
points[i] = openvdb::Vec3s(co.x, co.y, co.z) - 0.5f * voxel_size;
}
});
threading::parallel_for(corner_tris.index_range(), 2048, [&](const IndexRange range) {
for (const int i : range) {
const int3 &tri = corner_tris[i];
triangles[i] = openvdb::Vec3I(
corner_verts[tri[0]], corner_verts[tri[1]], corner_verts[tri[2]]);
}
});
openvdb::math::Transform::Ptr transform = openvdb::math::Transform::createLinearTransform(
voxel_size);
openvdb::FloatGrid::Ptr new_grid = openvdb::tools::meshToLevelSet<openvdb::FloatGrid>(
*transform, points, triangles, half_band_width);
return bke::VolumeGrid<float>(std::move(new_grid));
}
Volumes: refactor volume grid storage This refactors how volume grids are stored with the following new goals in mind: * Get a **stand-alone volume grid** data structure that can be used by geometry nodes. Previously, the `VolumeGrid` data structure was tightly coupled with the `Volume` data block. * Support **implicit sharing of grids and trees**. Previously, it was possible to share data when multiple `Volume` data blocks loaded grids from the same `.vdb` files but this was not flexible enough. * Get a safe API for **lazy-loading and unloading** of grids without requiring explicit calls to some "load" function all the time. * Get a safe API for **caching grids from files** that is not coupled to the `Volume` data block. * Get a **tiered API** for different levels of `openvdb` involvement: * No `OpenVDB`: Since `WITH_OPENVDB` is optional, it's helpful to have parts of the API that still work in this case. This makes it possible to write high level code for volumes that does not require `#ifdef WITH_OPENVDB` checks everywhere. This is in `BKE_volume_grid_fwd.hh`. * Shallow `OpenVDB`: Code using this API requires `WITH_OPENVDB` checks. However, care is taken to not include the expensive parts of `OpenVDB` and to use forward declarations as much as possible. This is in `BKE_volume_grid.hh` and uses `openvdb_fwd.hh`. * "Full" `OpenVDB`: This API requires more heavy `OpenVDB` includes. Fortunately, it turned out to be not necessary for the common API. So this is only used for task specific APIs. At the core of the new API is the `VolumeGridData` type. It's a wrapper around an `openvdb::Grid` and adds some features on top like implicit sharing, lazy-loading and unloading. Then there are `GVolumeGrid` and `VolumeGrid` which are containers for a volume grid. Semantically, each `VolumeGrid` has its own independent grid, but this is cheap due to implicit sharing. At highest level we currently have the `Volume` data-block which contains a list of `VolumeGrid`. ```mermaid flowchart LR Volume --> VolumeGrid --> VolumeGridData --> openvdb::Grid ``` The loading of `.vdb` files is abstracted away behind the volume file cache API. This API makes it easy to load and reuse entire files and individual grids from disk. It also supports caching simplify levels for grids on disk. An important new concept are the "tree access tokens". Whenever some code wants to work with an openvdb tree, it has to retrieve an access token from the corresponding `VolumeGridData`. This access token has to be kept alive for as long as the code works with the grid data. The same token is valid for read and write access. The purpose of these access tokens is to make it possible to detect when some code is currently working with the openvdb tree. This allows freeing it if it's possible to reload it later on (e.g. from disk). It's possible to free a tree that is referenced by multiple owners, but only no one is actively working with. In some sense, this is similar to the existing `ImageUser` concept. The most important new files to read are `BKE_volume_grid.hh` and `BKE_volume_grid_file_cache.hh`. Most other changes are updates to existing code to use the new API. Pull Request: https://projects.blender.org/blender/blender/pulls/116315
2023-12-20 15:32:52 +01:00
bke::VolumeGridData *fog_volume_grid_add_from_mesh(Volume *volume,
const StringRefNull name,
const Span<float3> positions,
const Span<int> corner_verts,
const Span<int3> corner_tris,
Volumes: refactor volume grid storage This refactors how volume grids are stored with the following new goals in mind: * Get a **stand-alone volume grid** data structure that can be used by geometry nodes. Previously, the `VolumeGrid` data structure was tightly coupled with the `Volume` data block. * Support **implicit sharing of grids and trees**. Previously, it was possible to share data when multiple `Volume` data blocks loaded grids from the same `.vdb` files but this was not flexible enough. * Get a safe API for **lazy-loading and unloading** of grids without requiring explicit calls to some "load" function all the time. * Get a safe API for **caching grids from files** that is not coupled to the `Volume` data block. * Get a **tiered API** for different levels of `openvdb` involvement: * No `OpenVDB`: Since `WITH_OPENVDB` is optional, it's helpful to have parts of the API that still work in this case. This makes it possible to write high level code for volumes that does not require `#ifdef WITH_OPENVDB` checks everywhere. This is in `BKE_volume_grid_fwd.hh`. * Shallow `OpenVDB`: Code using this API requires `WITH_OPENVDB` checks. However, care is taken to not include the expensive parts of `OpenVDB` and to use forward declarations as much as possible. This is in `BKE_volume_grid.hh` and uses `openvdb_fwd.hh`. * "Full" `OpenVDB`: This API requires more heavy `OpenVDB` includes. Fortunately, it turned out to be not necessary for the common API. So this is only used for task specific APIs. At the core of the new API is the `VolumeGridData` type. It's a wrapper around an `openvdb::Grid` and adds some features on top like implicit sharing, lazy-loading and unloading. Then there are `GVolumeGrid` and `VolumeGrid` which are containers for a volume grid. Semantically, each `VolumeGrid` has its own independent grid, but this is cheap due to implicit sharing. At highest level we currently have the `Volume` data-block which contains a list of `VolumeGrid`. ```mermaid flowchart LR Volume --> VolumeGrid --> VolumeGridData --> openvdb::Grid ``` The loading of `.vdb` files is abstracted away behind the volume file cache API. This API makes it easy to load and reuse entire files and individual grids from disk. It also supports caching simplify levels for grids on disk. An important new concept are the "tree access tokens". Whenever some code wants to work with an openvdb tree, it has to retrieve an access token from the corresponding `VolumeGridData`. This access token has to be kept alive for as long as the code works with the grid data. The same token is valid for read and write access. The purpose of these access tokens is to make it possible to detect when some code is currently working with the openvdb tree. This allows freeing it if it's possible to reload it later on (e.g. from disk). It's possible to free a tree that is referenced by multiple owners, but only no one is actively working with. In some sense, this is similar to the existing `ImageUser` concept. The most important new files to read are `BKE_volume_grid.hh` and `BKE_volume_grid_file_cache.hh`. Most other changes are updates to existing code to use the new API. Pull Request: https://projects.blender.org/blender/blender/pulls/116315
2023-12-20 15:32:52 +01:00
const float4x4 &mesh_to_volume_space_transform,
const float voxel_size,
const float interior_band_width,
const float density)
{
openvdb::FloatGrid::Ptr mesh_grid = mesh_to_density_grid_impl(positions,
corner_verts,
corner_tris,
mesh_to_volume_space_transform,
voxel_size,
interior_band_width,
density);
return mesh_grid ? BKE_volume_grid_add_vdb(*volume, name, std::move(mesh_grid)) : nullptr;
}
} // namespace blender::geometry
#endif