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

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

1255 lines
51 KiB
C++
Raw Normal View History

/* SPDX-FileCopyrightText: 2018 Blender Authors
*
* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup bke
*/
#include <mutex>
#include "DNA_key_types.h"
#include "DNA_mesh_types.h"
2022-08-17 18:11:01 -04:00
#include "BLI_array.hh"
#include "BLI_math_vector.h"
#include "BLI_math_vector.hh"
Mesh: Move UV layers to generic attributes Currently the `MLoopUV` struct stores UV coordinates and flags related to editing UV maps in the UV editor. This patch changes the coordinates to use the generic 2D vector type, and moves the flags into three separate boolean attributes. This follows the design in T95965, with the ultimate intention of simplifying code and improving performance. Importantly, the change allows exporters and renderers to use UVs "touched" by geometry nodes, which only creates generic attributes. It also allows geometry nodes to create "proper" UV maps from scratch, though only with the Store Named Attribute node for now. The new design considers any 2D vector attribute on the corner domain to be a UV map. In the future, they might be distinguished from regular 2D vectors with attribute metadata, which may be helpful because they are often interpolated differently. Most of the code changes deal with passing around UV BMesh custom data offsets and tracking the boolean "sublayers". The boolean layers are use the following prefixes for attribute names: vert selection: `.vs.`, edge selection: `.es.`, pinning: `.pn.`. Currently these are short to avoid using up the maximum length of attribute names. To accommodate for these 4 extra characters, the name length limit is enlarged to 68 bytes, while the maximum user settable name length is still 64 bytes. Unfortunately Python/RNA API access to the UV flag data becomes slower. Accessing the boolean layers directly is be better for performance in general. Like the other mesh SoA refactors, backward and forward compatibility aren't affected, and won't be changed until 4.0. We pay for that by making mesh reading and writing more expensive with conversions. Resolves T85962 Differential Revision: https://developer.blender.org/D14365
2023-01-10 00:47:04 -05:00
#include "BLI_math_vector_types.hh"
#include "BLI_task.hh"
#include "BKE_attribute_math.hh"
#include "BKE_customdata.hh"
2024-01-30 14:42:07 -05:00
#include "BKE_key.hh"
#include "BKE_mesh.hh"
#include "BKE_mesh_mapping.hh"
#include "BKE_subdiv.hh"
#include "BKE_subdiv_eval.hh"
#include "BKE_subdiv_foreach.hh"
#include "BKE_subdiv_mesh.hh"
#include "MEM_guardedalloc.h"
namespace blender::bke::subdiv {
/* -------------------------------------------------------------------- */
/** \name Subdivision Context
* \{ */
2022-08-17 18:11:01 -04:00
struct SubdivMeshContext {
const ToMeshSettings *settings;
const Mesh *coarse_mesh;
Span<float3> coarse_positions;
Span<int2> coarse_edges;
OffsetIndices<int> coarse_faces;
Span<int> coarse_corner_verts;
Mesh: Remove redundant custom data pointers For copy-on-write, we want to share attribute arrays between meshes where possible. Mutable pointers like `Mesh.mvert` make that difficult by making ownership vague. They also make code more complex by adding redundancy. The simplest solution is just removing them and retrieving layers from `CustomData` as needed. Similar changes have already been applied to curves and point clouds (e9f82d3dc7ee, 410a6efb747f). Removing use of the pointers generally makes code more obvious and more reusable. Mesh data is now accessed with a C++ API (`Mesh::edges()` or `Mesh::edges_for_write()`), and a C API (`BKE_mesh_edges(mesh)`). The CoW changes this commit makes possible are described in T95845 and T95842, and started in D14139 and D14140. The change also simplifies the ongoing mesh struct-of-array refactors from T95965. **RNA/Python Access Performance** Theoretically, accessing mesh elements with the RNA API may become slower, since the layer needs to be found on every random access. However, overhead is already high enough that this doesn't make a noticible differenc, and performance is actually improved in some cases. Random access can be up to 10% faster, but other situations might be a bit slower. Generally using `foreach_get/set` are the best way to improve performance. See the differential revision for more discussion about Python performance. Cycles has been updated to use raw pointers and the internal Blender mesh types, mostly because there is no sense in having this overhead when it's already compiled with Blender. In my tests this roughly halves the Cycles mesh creation time (0.19s to 0.10s for a 1 million face grid). Differential Revision: https://developer.blender.org/D15488
2022-09-05 11:56:34 -05:00
/**
* Contains all face corner custom data from the original coarse mesh except for the
* ".corner_vert" and ".corner_edge" topology layers. This prevents unnecessary interpolation of
* that data which would just be overwritten anyway.
*/
CustomData coarse_corner_data_interp;
Subdiv *subdiv;
Mesh *subdiv_mesh;
MutableSpan<float3> subdiv_positions;
MutableSpan<int2> subdiv_edges;
MutableSpan<int> subdiv_face_offsets;
/**
* Owning pointers to topology arrays, not added to the result mesh until face corner value
* interpolation finishes.
*/
int *subdiv_corner_verts;
int *subdiv_corner_edges;
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
/* Cached custom data arrays for faster access. */
int *vert_origindex;
int *edge_origindex;
int *loop_origindex;
int *face_origindex;
/* UV layers interpolation. */
int num_uv_layers;
Mesh: Move UV layers to generic attributes Currently the `MLoopUV` struct stores UV coordinates and flags related to editing UV maps in the UV editor. This patch changes the coordinates to use the generic 2D vector type, and moves the flags into three separate boolean attributes. This follows the design in T95965, with the ultimate intention of simplifying code and improving performance. Importantly, the change allows exporters and renderers to use UVs "touched" by geometry nodes, which only creates generic attributes. It also allows geometry nodes to create "proper" UV maps from scratch, though only with the Store Named Attribute node for now. The new design considers any 2D vector attribute on the corner domain to be a UV map. In the future, they might be distinguished from regular 2D vectors with attribute metadata, which may be helpful because they are often interpolated differently. Most of the code changes deal with passing around UV BMesh custom data offsets and tracking the boolean "sublayers". The boolean layers are use the following prefixes for attribute names: vert selection: `.vs.`, edge selection: `.es.`, pinning: `.pn.`. Currently these are short to avoid using up the maximum length of attribute names. To accommodate for these 4 extra characters, the name length limit is enlarged to 68 bytes, while the maximum user settable name length is still 64 bytes. Unfortunately Python/RNA API access to the UV flag data becomes slower. Accessing the boolean layers directly is be better for performance in general. Like the other mesh SoA refactors, backward and forward compatibility aren't affected, and won't be changed until 4.0. We pay for that by making mesh reading and writing more expensive with conversions. Resolves T85962 Differential Revision: https://developer.blender.org/D14365
2023-01-10 00:47:04 -05:00
float2 *uv_layers[MAX_MTFACE];
/* Original coordinates (ORCO) interpolation. */
float (*orco)[3];
float (*cloth_orco)[3];
/* Per-subdivided vertex counter of averaged values. */
int *accumulated_counters;
bool have_displacement;
/* Write optimal display edge tags into a boolean array rather than the final bit vector
* to avoid race conditions when setting bits. */
Array<bool> subdiv_display_edges;
/* Lazily initialize a map from vertices to connected edges. */
Array<int> vert_to_edge_offsets;
Array<int> vert_to_edge_indices;
GroupedSpan<int> vert_to_edge_map;
2022-08-17 18:11:01 -04:00
};
static void subdiv_mesh_ctx_cache_uv_layers(SubdivMeshContext *ctx)
{
Mesh *subdiv_mesh = ctx->subdiv_mesh;
ctx->num_uv_layers = std::min(
CustomData_number_of_layers(&subdiv_mesh->corner_data, CD_PROP_FLOAT2), MAX_MTFACE);
for (int layer_index = 0; layer_index < ctx->num_uv_layers; layer_index++) {
ctx->uv_layers[layer_index] = static_cast<float2 *>(CustomData_get_layer_n_for_write(
&subdiv_mesh->corner_data, CD_PROP_FLOAT2, layer_index, subdiv_mesh->corners_num));
}
}
static void subdiv_mesh_ctx_cache_custom_data_layers(SubdivMeshContext *ctx)
{
Mesh *subdiv_mesh = ctx->subdiv_mesh;
ctx->subdiv_positions = subdiv_mesh->vert_positions_for_write();
ctx->subdiv_edges = subdiv_mesh->edges_for_write();
ctx->subdiv_face_offsets = subdiv_mesh->face_offsets_for_write();
/* Pointers to original indices layers. */
ctx->vert_origindex = static_cast<int *>(CustomData_get_layer_for_write(
&subdiv_mesh->vert_data, CD_ORIGINDEX, subdiv_mesh->verts_num));
ctx->edge_origindex = static_cast<int *>(CustomData_get_layer_for_write(
&subdiv_mesh->edge_data, CD_ORIGINDEX, subdiv_mesh->edges_num));
ctx->loop_origindex = static_cast<int *>(CustomData_get_layer_for_write(
&subdiv_mesh->corner_data, CD_ORIGINDEX, subdiv_mesh->corners_num));
ctx->face_origindex = static_cast<int *>(CustomData_get_layer_for_write(
&subdiv_mesh->face_data, CD_ORIGINDEX, subdiv_mesh->faces_num));
/* UV layers interpolation. */
subdiv_mesh_ctx_cache_uv_layers(ctx);
/* Orco interpolation. */
ctx->orco = static_cast<float(*)[3]>(
CustomData_get_layer_for_write(&subdiv_mesh->vert_data, CD_ORCO, subdiv_mesh->verts_num));
ctx->cloth_orco = static_cast<float(*)[3]>(CustomData_get_layer_for_write(
&subdiv_mesh->vert_data, CD_CLOTH_ORCO, subdiv_mesh->verts_num));
}
static void subdiv_mesh_prepare_accumulator(SubdivMeshContext *ctx, int num_vertices)
{
if (!ctx->have_displacement) {
return;
}
2022-08-17 18:11:01 -04:00
ctx->accumulated_counters = static_cast<int *>(
MEM_calloc_arrayN(num_vertices, sizeof(*ctx->accumulated_counters), __func__));
}
static void subdiv_mesh_context_free(SubdivMeshContext *ctx)
{
MEM_SAFE_FREE(ctx->accumulated_counters);
MEM_SAFE_FREE(ctx->subdiv_corner_verts);
MEM_SAFE_FREE(ctx->subdiv_corner_edges);
CustomData_free(&ctx->coarse_corner_data_interp, ctx->coarse_mesh->corners_num);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Loop custom data copy helpers
* \{ */
2022-08-17 18:11:01 -04:00
struct LoopsOfPtex {
/* First loop of the ptex, starts at ptex (0, 0) and goes in u direction. */
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
int first_loop;
/* Last loop of the ptex, starts at ptex (0, 0) and goes in v direction. */
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
int last_loop;
/* For quad coarse faces only. */
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
int second_loop;
int third_loop;
2022-08-17 18:11:01 -04:00
};
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
static void loops_of_ptex_get(LoopsOfPtex *loops_of_ptex,
const IndexRange coarse_face,
const int ptex_of_face_index)
{
const int first_ptex_loop_index = coarse_face.start() + ptex_of_face_index;
/* Loop which look in the (opposite) V direction of the current
* ptex face.
*
* TODO(sergey): Get rid of using module on every iteration. */
const int last_ptex_loop_index = coarse_face.start() +
(ptex_of_face_index + coarse_face.size() - 1) %
coarse_face.size();
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
loops_of_ptex->first_loop = first_ptex_loop_index;
loops_of_ptex->last_loop = last_ptex_loop_index;
if (coarse_face.size() == 4) {
loops_of_ptex->second_loop = loops_of_ptex->first_loop + 1;
loops_of_ptex->third_loop = loops_of_ptex->first_loop + 2;
}
else {
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
loops_of_ptex->second_loop = -1;
loops_of_ptex->third_loop = -1;
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Vertex custom data interpolation helpers
* \{ */
/* TODO(sergey): Somehow de-duplicate with loops storage, without too much
* exception cases all over the code. */
2022-08-17 18:11:01 -04:00
struct VerticesForInterpolation {
/* This field points to a vertex data which is to be used for interpolation.
* The idea is to avoid unnecessary allocations for regular faces, where
2019-06-17 12:51:53 +10:00
* we can simply use corner vertices. */
const CustomData *vertex_data;
/* Vertices data calculated for ptex corners. There are always 4 elements
* in this custom data, aligned the following way:
*
* index 0 -> uv (0, 0)
* index 1 -> uv (0, 1)
* index 2 -> uv (1, 1)
* index 3 -> uv (1, 0)
*
* Is allocated for non-regular faces (triangles and n-gons). */
CustomData vertex_data_storage;
bool vertex_data_storage_allocated;
2019-06-17 12:51:53 +10:00
/* Indices within vertex_data to interpolate for. The indices are aligned
* with uv coordinates in a similar way as indices in corner_data_storage. */
int vertex_indices[4];
2022-08-17 18:11:01 -04:00
};
static void vertex_interpolation_init(const SubdivMeshContext *ctx,
VerticesForInterpolation *vertex_interpolation,
const IndexRange coarse_face)
{
const Mesh *coarse_mesh = ctx->coarse_mesh;
if (coarse_face.size() == 4) {
vertex_interpolation->vertex_data = &coarse_mesh->vert_data;
vertex_interpolation->vertex_indices[0] = ctx->coarse_corner_verts[coarse_face.start() + 0];
vertex_interpolation->vertex_indices[1] = ctx->coarse_corner_verts[coarse_face.start() + 1];
vertex_interpolation->vertex_indices[2] = ctx->coarse_corner_verts[coarse_face.start() + 2];
vertex_interpolation->vertex_indices[3] = ctx->coarse_corner_verts[coarse_face.start() + 3];
vertex_interpolation->vertex_data_storage_allocated = false;
}
else {
vertex_interpolation->vertex_data = &vertex_interpolation->vertex_data_storage;
/* Allocate storage for loops corresponding to ptex corners. */
CustomData_init_layout_from(&ctx->coarse_mesh->vert_data,
&vertex_interpolation->vertex_data_storage,
CD_MASK_EVERYTHING.vmask,
CD_SET_DEFAULT,
4);
/* Initialize indices. */
vertex_interpolation->vertex_indices[0] = 0;
vertex_interpolation->vertex_indices[1] = 1;
vertex_interpolation->vertex_indices[2] = 2;
vertex_interpolation->vertex_indices[3] = 3;
vertex_interpolation->vertex_data_storage_allocated = true;
/* Interpolate center of face right away, it stays unchanged for all
* ptex faces. */
const float weight = 1.0f / float(coarse_face.size());
Array<float, 32> weights(coarse_face.size());
Array<int, 32> indices(coarse_face.size());
for (int i = 0; i < coarse_face.size(); i++) {
weights[i] = weight;
indices[i] = ctx->coarse_corner_verts[coarse_face.start() + i];
}
CustomData_interp(&coarse_mesh->vert_data,
&vertex_interpolation->vertex_data_storage,
2022-08-17 18:11:01 -04:00
indices.data(),
weights.data(),
nullptr,
coarse_face.size(),
2);
}
}
static void vertex_interpolation_from_corner(const SubdivMeshContext *ctx,
VerticesForInterpolation *vertex_interpolation,
const IndexRange coarse_face,
const int corner)
{
if (coarse_face.size() == 4) {
/* Nothing to do, all indices and data is already assigned. */
2018-07-19 16:06:37 +10:00
}
else {
const CustomData *vertex_data = &ctx->coarse_mesh->vert_data;
LoopsOfPtex loops_of_ptex;
loops_of_ptex_get(&loops_of_ptex, coarse_face, corner);
/* PTEX face corner corresponds to a face loop with same index. */
CustomData_copy_data(vertex_data,
&vertex_interpolation->vertex_data_storage,
ctx->coarse_corner_verts[coarse_face.start() + corner],
0,
1);
/* Interpolate remaining ptex face corners, which hits loops
* middle points.
*
* TODO(sergey): Re-use one of interpolation results from previous
* iteration. */
const float weights[2] = {0.5f, 0.5f};
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
const int first_loop_index = loops_of_ptex.first_loop;
const int last_loop_index = loops_of_ptex.last_loop;
const int first_indices[2] = {
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
ctx->coarse_corner_verts[first_loop_index],
ctx->coarse_corner_verts[coarse_face.start() +
(first_loop_index - coarse_face.start() + 1) %
coarse_face.size()]};
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
const int last_indices[2] = {ctx->coarse_corner_verts[first_loop_index],
ctx->coarse_corner_verts[last_loop_index]};
CustomData_interp(vertex_data,
&vertex_interpolation->vertex_data_storage,
first_indices,
weights,
2022-08-17 18:11:01 -04:00
nullptr,
2,
1);
CustomData_interp(vertex_data,
&vertex_interpolation->vertex_data_storage,
last_indices,
weights,
2022-08-17 18:11:01 -04:00
nullptr,
2,
3);
}
}
static void vertex_interpolation_end(VerticesForInterpolation *vertex_interpolation)
{
if (vertex_interpolation->vertex_data_storage_allocated) {
CustomData_free(&vertex_interpolation->vertex_data_storage, 4);
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Loop custom data interpolation helpers
* \{ */
2022-08-17 18:11:01 -04:00
struct LoopsForInterpolation {
/* This field points to a loop data which is to be used for interpolation.
* The idea is to avoid unnecessary allocations for regular faces, where
2019-06-17 12:51:53 +10:00
* we can simply interpolate corner vertices. */
const CustomData *corner_data;
/* Loops data calculated for ptex corners. There are always 4 elements
* in this custom data, aligned the following way:
*
* index 0 -> uv (0, 0)
* index 1 -> uv (0, 1)
* index 2 -> uv (1, 1)
* index 3 -> uv (1, 0)
*
* Is allocated for non-regular faces (triangles and n-gons). */
CustomData corner_data_storage;
bool corner_data_storage_allocated;
/* Indices within corner_data to interpolate for. The indices are aligned with
* uv coordinates in a similar way as indices in corner_data_storage. */
int loop_indices[4];
2022-08-17 18:11:01 -04:00
};
static void loop_interpolation_init(const SubdivMeshContext *ctx,
LoopsForInterpolation *loop_interpolation,
const IndexRange coarse_face)
{
if (coarse_face.size() == 4) {
loop_interpolation->corner_data = &ctx->coarse_corner_data_interp;
loop_interpolation->loop_indices[0] = coarse_face.start() + 0;
loop_interpolation->loop_indices[1] = coarse_face.start() + 1;
loop_interpolation->loop_indices[2] = coarse_face.start() + 2;
loop_interpolation->loop_indices[3] = coarse_face.start() + 3;
loop_interpolation->corner_data_storage_allocated = false;
}
else {
loop_interpolation->corner_data = &loop_interpolation->corner_data_storage;
/* Allocate storage for loops corresponding to ptex corners. */
CustomData_init_layout_from(&ctx->coarse_corner_data_interp,
&loop_interpolation->corner_data_storage,
CD_MASK_EVERYTHING.lmask,
CD_SET_DEFAULT,
4);
/* Initialize indices. */
loop_interpolation->loop_indices[0] = 0;
loop_interpolation->loop_indices[1] = 1;
loop_interpolation->loop_indices[2] = 2;
loop_interpolation->loop_indices[3] = 3;
loop_interpolation->corner_data_storage_allocated = true;
/* Interpolate center of face right away, it stays unchanged for all
* ptex faces. */
const float weight = 1.0f / float(coarse_face.size());
Array<float, 32> weights(coarse_face.size());
Array<int, 32> indices(coarse_face.size());
for (int i = 0; i < coarse_face.size(); i++) {
weights[i] = weight;
indices[i] = coarse_face.start() + i;
}
CustomData_interp(&ctx->coarse_corner_data_interp,
&loop_interpolation->corner_data_storage,
2022-08-17 18:11:01 -04:00
indices.data(),
weights.data(),
nullptr,
coarse_face.size(),
2);
}
}
static void loop_interpolation_from_corner(const SubdivMeshContext *ctx,
LoopsForInterpolation *loop_interpolation,
const IndexRange coarse_face,
const int corner)
{
if (coarse_face.size() == 4) {
/* Nothing to do, all indices and data is already assigned. */
2018-07-19 16:06:37 +10:00
}
else {
const CustomData *corner_data = &ctx->coarse_corner_data_interp;
LoopsOfPtex loops_of_ptex;
loops_of_ptex_get(&loops_of_ptex, coarse_face, corner);
/* PTEX face corner corresponds to a face loop with same index. */
CustomData_free_elem(&loop_interpolation->corner_data_storage, 0, 1);
CustomData_copy_data(
corner_data, &loop_interpolation->corner_data_storage, coarse_face.start() + corner, 0, 1);
/* Interpolate remaining ptex face corners, which hits loops
* middle points.
*
* TODO(sergey): Re-use one of interpolation results from previous
* iteration. */
const float weights[2] = {0.5f, 0.5f};
const int base_loop_index = coarse_face.start();
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
const int first_loop_index = loops_of_ptex.first_loop;
const int second_loop_index = base_loop_index +
(first_loop_index - base_loop_index + 1) % coarse_face.size();
const int first_indices[2] = {first_loop_index, second_loop_index};
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
const int last_indices[2] = {loops_of_ptex.last_loop, loops_of_ptex.first_loop};
CustomData_interp(corner_data,
&loop_interpolation->corner_data_storage,
first_indices,
weights,
nullptr,
2,
1);
CustomData_interp(corner_data,
&loop_interpolation->corner_data_storage,
last_indices,
weights,
nullptr,
2,
3);
}
}
static void loop_interpolation_end(LoopsForInterpolation *loop_interpolation)
{
if (loop_interpolation->corner_data_storage_allocated) {
CustomData_free(&loop_interpolation->corner_data_storage, 4);
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name TLS
* \{ */
2022-08-17 18:11:01 -04:00
struct SubdivMeshTLS {
bool vertex_interpolation_initialized;
VerticesForInterpolation vertex_interpolation;
int vertex_interpolation_coarse_face_index;
int vertex_interpolation_coarse_corner;
bool loop_interpolation_initialized;
LoopsForInterpolation loop_interpolation;
int loop_interpolation_coarse_face_index;
int loop_interpolation_coarse_corner;
2022-08-17 18:11:01 -04:00
};
static void subdiv_mesh_tls_free(void *tls_v)
{
2022-08-17 18:11:01 -04:00
SubdivMeshTLS *tls = static_cast<SubdivMeshTLS *>(tls_v);
if (tls->vertex_interpolation_initialized) {
vertex_interpolation_end(&tls->vertex_interpolation);
}
if (tls->loop_interpolation_initialized) {
loop_interpolation_end(&tls->loop_interpolation);
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Evaluation helper functions
* \{ */
static void subdiv_vertex_orco_evaluate(const SubdivMeshContext *ctx,
const int ptex_face_index,
const float u,
const float v,
const int subdiv_vertex_index)
{
if (ctx->orco || ctx->cloth_orco) {
float vertex_data[6];
eval_vertex_data(ctx->subdiv, ptex_face_index, u, v, vertex_data);
if (ctx->orco) {
copy_v3_v3(ctx->orco[subdiv_vertex_index], vertex_data);
if (ctx->cloth_orco) {
copy_v3_v3(ctx->cloth_orco[subdiv_vertex_index], vertex_data + 3);
}
}
else if (ctx->cloth_orco) {
copy_v3_v3(ctx->cloth_orco[subdiv_vertex_index], vertex_data);
}
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Accumulation helpers
* \{ */
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
static void subdiv_accumulate_vertex_displacement(SubdivMeshContext *ctx,
const int ptex_face_index,
const float u,
const float v,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
const int subdiv_vertex_index)
{
/* Accumulate displacement. */
Subdiv *subdiv = ctx->subdiv;
float dummy_P[3], dPdu[3], dPdv[3], D[3];
eval_limit_point_and_derivatives(subdiv, ptex_face_index, u, v, dummy_P, dPdu, dPdv);
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
/* NOTE: The subdivided mesh is allocated in this module, and its vertices are kept at zero
* locations as a default calloc(). */
eval_displacement(subdiv, ptex_face_index, u, v, dPdu, dPdv, D);
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
ctx->subdiv_positions[subdiv_vertex_index] += D;
if (ctx->accumulated_counters) {
++ctx->accumulated_counters[subdiv_vertex_index];
}
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Callbacks
* \{ */
static bool subdiv_mesh_topology_info(const ForeachContext *foreach_context,
const int num_vertices,
const int num_edges,
const int num_loops,
const int num_faces,
const int * /*subdiv_face_offset*/)
{
2022-01-06 13:54:52 +11:00
/* Multi-resolution grid data will be applied or become invalid after subdivision,
* so don't try to preserve it and use memory. Crease values should also not be interpolated. */
CustomData_MeshMasks mask = CD_MASK_EVERYTHING;
mask.lmask &= ~CD_MASK_MULTIRES_GRIDS;
2022-08-17 18:11:01 -04:00
SubdivMeshContext *subdiv_context = static_cast<SubdivMeshContext *>(foreach_context->user_data);
const Mesh &coarse_mesh = *subdiv_context->coarse_mesh;
subdiv_context->subdiv_mesh = bke::mesh_new_no_attributes(
num_vertices, num_edges, num_faces, num_loops);
Mesh &subdiv_mesh = *subdiv_context->subdiv_mesh;
BKE_mesh_copy_parameters_for_eval(subdiv_context->subdiv_mesh, &coarse_mesh);
CustomData_free(&subdiv_mesh.vert_data, 0);
CustomData_init_layout_from(
&coarse_mesh.vert_data, &subdiv_mesh.vert_data, mask.vmask, CD_SET_DEFAULT, num_vertices);
CustomData_free(&subdiv_mesh.edge_data, 0);
CustomData_init_layout_from(
&coarse_mesh.edge_data, &subdiv_mesh.edge_data, mask.emask, CD_SET_DEFAULT, num_edges);
CustomData_free(&subdiv_mesh.face_data, 0);
CustomData_init_layout_from(
&coarse_mesh.face_data, &subdiv_mesh.face_data, mask.pmask, CD_SET_DEFAULT, num_faces);
if (num_faces != 0) {
subdiv_mesh.face_offsets_for_write().last() = num_loops;
}
/* Create corner data for interpolation without topology attributes. */
CustomData_init_from(&coarse_mesh.corner_data,
&subdiv_context->coarse_corner_data_interp,
mask.lmask,
coarse_mesh.corners_num);
CustomData_free_layer_named(
&subdiv_context->coarse_corner_data_interp, ".corner_vert", coarse_mesh.corners_num);
CustomData_free_layer_named(
&subdiv_context->coarse_corner_data_interp, ".corner_edge", coarse_mesh.corners_num);
CustomData_free(&subdiv_mesh.corner_data, 0);
CustomData_init_layout_from(&subdiv_context->coarse_corner_data_interp,
&subdiv_mesh.corner_data,
mask.lmask,
CD_SET_DEFAULT,
num_loops);
/* Allocate corner topology arrays which are added to the result at the end. */
subdiv_context->subdiv_corner_verts = static_cast<int *>(
MEM_malloc_arrayN(num_loops, sizeof(int), __func__));
subdiv_context->subdiv_corner_edges = static_cast<int *>(
MEM_malloc_arrayN(num_loops, sizeof(int), __func__));
subdiv_mesh_ctx_cache_custom_data_layers(subdiv_context);
subdiv_mesh_prepare_accumulator(subdiv_context, num_vertices);
subdiv_mesh.runtime->subsurf_face_dot_tags.clear();
subdiv_mesh.runtime->subsurf_face_dot_tags.resize(num_vertices);
if (subdiv_context->settings->use_optimal_display) {
subdiv_context->subdiv_display_edges = Array<bool>(num_edges, false);
}
return true;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Vertex subdivision process
* \{ */
static void subdiv_vertex_data_copy(const SubdivMeshContext *ctx,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
const int coarse_vertex_index,
const int subdiv_vertex_index)
{
const Mesh *coarse_mesh = ctx->coarse_mesh;
CustomData_copy_data(&coarse_mesh->vert_data,
&ctx->subdiv_mesh->vert_data,
coarse_vertex_index,
subdiv_vertex_index,
1);
}
static void subdiv_vertex_data_interpolate(const SubdivMeshContext *ctx,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
const int subdiv_vertex_index,
const VerticesForInterpolation *vertex_interpolation,
const float u,
const float v)
{
const float weights[4] = {(1.0f - u) * (1.0f - v), u * (1.0f - v), u * v, (1.0f - u) * v};
CustomData_interp(vertex_interpolation->vertex_data,
&ctx->subdiv_mesh->vert_data,
vertex_interpolation->vertex_indices,
weights,
2022-08-17 18:11:01 -04:00
nullptr,
4,
subdiv_vertex_index);
2022-08-17 18:11:01 -04:00
if (ctx->vert_origindex != nullptr) {
ctx->vert_origindex[subdiv_vertex_index] = ORIGINDEX_NONE;
}
}
static void evaluate_vertex_and_apply_displacement_copy(const SubdivMeshContext *ctx,
const int ptex_face_index,
const float u,
const float v,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
const int coarse_vertex_index,
const int subdiv_vertex_index)
{
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
float3 &subdiv_position = ctx->subdiv_positions[subdiv_vertex_index];
/* Displacement is accumulated in subdiv vertex position.
2019-02-04 01:23:48 +01:00
* Needs to be backed up before copying data from original vertex. */
float D[3] = {0.0f, 0.0f, 0.0f};
if (ctx->have_displacement) {
const float inv_num_accumulated = 1.0f / ctx->accumulated_counters[subdiv_vertex_index];
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
copy_v3_v3(D, subdiv_position);
mul_v3_fl(D, inv_num_accumulated);
}
/* Copy custom data and evaluate position. */
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
subdiv_vertex_data_copy(ctx, coarse_vertex_index, subdiv_vertex_index);
eval_limit_point(ctx->subdiv, ptex_face_index, u, v, subdiv_position);
/* Apply displacement. */
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
subdiv_position += D;
/* Evaluate undeformed texture coordinate. */
subdiv_vertex_orco_evaluate(ctx, ptex_face_index, u, v, subdiv_vertex_index);
2022-08-18 08:46:24 +10:00
/* Remove face-dot flag. This can happen if there is more than one subsurf modifier. */
ctx->subdiv_mesh->runtime->subsurf_face_dot_tags[subdiv_vertex_index].reset();
}
static void evaluate_vertex_and_apply_displacement_interpolate(
const SubdivMeshContext *ctx,
const int ptex_face_index,
const float u,
const float v,
VerticesForInterpolation *vertex_interpolation,
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
const int subdiv_vertex_index)
{
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
float3 &subdiv_position = ctx->subdiv_positions[subdiv_vertex_index];
/* Displacement is accumulated in subdiv vertex position.
2019-02-04 01:23:48 +01:00
* Needs to be backed up before copying data from original vertex. */
float D[3] = {0.0f, 0.0f, 0.0f};
if (ctx->have_displacement) {
const float inv_num_accumulated = 1.0f / ctx->accumulated_counters[subdiv_vertex_index];
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
copy_v3_v3(D, subdiv_position);
mul_v3_fl(D, inv_num_accumulated);
}
/* Interpolate custom data and evaluate position. */
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
subdiv_vertex_data_interpolate(ctx, subdiv_vertex_index, vertex_interpolation, u, v);
eval_limit_point(ctx->subdiv, ptex_face_index, u, v, subdiv_position);
/* Apply displacement. */
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
add_v3_v3(subdiv_position, D);
/* Evaluate undeformed texture coordinate. */
subdiv_vertex_orco_evaluate(ctx, ptex_face_index, u, v, subdiv_vertex_index);
}
static void subdiv_mesh_vertex_displacement_every_corner_or_edge(
const ForeachContext *foreach_context,
void * /*tls*/,
const int ptex_face_index,
const float u,
const float v,
const int subdiv_vertex_index)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
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
subdiv_accumulate_vertex_displacement(ctx, ptex_face_index, u, v, subdiv_vertex_index);
}
static void subdiv_mesh_vertex_displacement_every_corner(const ForeachContext *foreach_context,
void *tls,
const int ptex_face_index,
const float u,
const float v,
const int /*coarse_vertex_index*/,
const int /*coarse_face_index*/,
const int /*coarse_corner*/,
const int subdiv_vertex_index)
{
subdiv_mesh_vertex_displacement_every_corner_or_edge(
foreach_context, tls, ptex_face_index, u, v, subdiv_vertex_index);
}
static void subdiv_mesh_vertex_displacement_every_edge(const ForeachContext *foreach_context,
void *tls,
const int ptex_face_index,
const float u,
const float v,
const int /*coarse_edge_index*/,
const int /*coarse_face_index*/,
const int /*coarse_corner*/,
const int subdiv_vertex_index)
{
subdiv_mesh_vertex_displacement_every_corner_or_edge(
foreach_context, tls, ptex_face_index, u, v, subdiv_vertex_index);
}
static void subdiv_mesh_vertex_corner(const ForeachContext *foreach_context,
void * /*tls*/,
const int ptex_face_index,
const float u,
const float v,
const int coarse_vertex_index,
const int /*coarse_face_index*/,
const int /*coarse_corner*/,
const int subdiv_vertex_index)
{
BLI_assert(coarse_vertex_index != ORIGINDEX_NONE);
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
evaluate_vertex_and_apply_displacement_copy(
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
ctx, ptex_face_index, u, v, coarse_vertex_index, subdiv_vertex_index);
}
static void subdiv_mesh_ensure_vertex_interpolation(SubdivMeshContext *ctx,
SubdivMeshTLS *tls,
const int coarse_face_index,
const int coarse_corner)
{
const IndexRange coarse_face = ctx->coarse_faces[coarse_face_index];
/* Check whether we've moved to another corner or face. */
if (tls->vertex_interpolation_initialized) {
if (tls->vertex_interpolation_coarse_face_index != coarse_face_index ||
tls->vertex_interpolation_coarse_corner != coarse_corner)
{
vertex_interpolation_end(&tls->vertex_interpolation);
tls->vertex_interpolation_initialized = false;
}
}
/* Initialize the interpolation. */
if (!tls->vertex_interpolation_initialized) {
vertex_interpolation_init(ctx, &tls->vertex_interpolation, coarse_face);
}
/* Update it for a new corner if needed. */
if (!tls->vertex_interpolation_initialized ||
2018-08-24 10:26:59 +10:00
tls->vertex_interpolation_coarse_corner != coarse_corner)
{
vertex_interpolation_from_corner(ctx, &tls->vertex_interpolation, coarse_face, coarse_corner);
2018-08-24 10:26:59 +10:00
}
/* Store settings used for the current state of interpolator. */
tls->vertex_interpolation_initialized = true;
tls->vertex_interpolation_coarse_face_index = coarse_face_index;
tls->vertex_interpolation_coarse_corner = coarse_corner;
}
static void subdiv_mesh_vertex_edge(const ForeachContext *foreach_context,
void *tls_v,
const int ptex_face_index,
const float u,
const float v,
const int /*coarse_edge_index*/,
const int coarse_face_index,
const int coarse_corner,
const int subdiv_vertex_index)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
SubdivMeshTLS *tls = static_cast<SubdivMeshTLS *>(tls_v);
subdiv_mesh_ensure_vertex_interpolation(ctx, tls, coarse_face_index, coarse_corner);
evaluate_vertex_and_apply_displacement_interpolate(
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
ctx, ptex_face_index, u, v, &tls->vertex_interpolation, subdiv_vertex_index);
}
static bool subdiv_mesh_is_center_vertex(const IndexRange coarse_face,
Mesh: Replace MPoly struct with offset indices Implements #95967. Currently the `MPoly` struct is 12 bytes, and stores the index of a face's first corner and the number of corners/verts/edges. Polygons and corners are always created in order by Blender, meaning each face's corners will be after the previous face's corners. We can take advantage of this fact and eliminate the redundancy in mesh face storage by only storing a single integer corner offset for each face. The size of the face is then encoded by the offset of the next face. The size of a single integer is 4 bytes, so this reduces memory usage by 3 times. The same method is used for `CurvesGeometry`, so Blender already has an abstraction to simplify using these offsets called `OffsetIndices`. This class is used to easily retrieve a range of corner indices for each face. This also gives the opportunity for sharing some logic with curves. Another benefit of the change is that the offsets and sizes stored in `MPoly` can no longer disagree with each other. Storing faces in the order of their corners can simplify some code too. Face/polygon variables now use the `IndexRange` type, which comes with quite a few utilities that can simplify code. Some: - The offset integer array has to be one longer than the face count to avoid a branch for every face, which means the data is no longer part of the mesh's `CustomData`. - We lose the ability to "reference" an original mesh's offset array until more reusable CoW from #104478 is committed. That will be added in a separate commit. - Since they aren't part of `CustomData`, poly offsets often have to be copied manually. - To simplify using `OffsetIndices` in many places, some functions and structs in headers were moved to only compile in C++. - All meshes created by Blender use the same order for faces and face corners, but just in case, meshes with mismatched order are fixed by versioning code. - `MeshPolygon.totloop` is no longer editable in RNA. This API break is necessary here unfortunately. It should be worth it in 3.6, since that's the best way to allow loading meshes from 4.0, which is important for an LTS version. Pull Request: https://projects.blender.org/blender/blender/pulls/105938
2023-04-04 20:39:28 +02:00
const float u,
const float v)
{
if (coarse_face.size() == 4) {
if (u == 0.5f && v == 0.5f) {
return true;
}
}
else {
if (u == 1.0f && v == 1.0f) {
return true;
}
}
return false;
}
static void subdiv_mesh_tag_center_vertex(const IndexRange coarse_face,
const int subdiv_vertex_index,
const float u,
const float v,
Mesh *subdiv_mesh)
{
if (subdiv_mesh_is_center_vertex(coarse_face, u, v)) {
subdiv_mesh->runtime->subsurf_face_dot_tags[subdiv_vertex_index].set();
}
}
static void subdiv_mesh_vertex_inner(const ForeachContext *foreach_context,
void *tls_v,
const int ptex_face_index,
const float u,
const float v,
const int coarse_face_index,
const int coarse_corner,
const int subdiv_vertex_index)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
SubdivMeshTLS *tls = static_cast<SubdivMeshTLS *>(tls_v);
Subdiv *subdiv = ctx->subdiv;
const IndexRange coarse_face = ctx->coarse_faces[coarse_face_index];
Mesh *subdiv_mesh = ctx->subdiv_mesh;
Mesh: Move positions to a generic attribute **Changes** As described in T93602, this patch removes all use of the `MVert` struct, replacing it with a generic named attribute with the name `"position"`, consistent with other geometry types. Variable names have been changed from `verts` to `positions`, to align with the attribute name and the more generic design (positions are not vertices, they are just an attribute stored on the point domain). This change is made possible by previous commits that moved all other data out of `MVert` to runtime data or other generic attributes. What remains is mostly a simple type change. Though, the type still shows up 859 times, so the patch is quite large. One compromise is that now `CD_MASK_BAREMESH` now contains `CD_PROP_FLOAT3`. With the general move towards generic attributes over custom data types, we are removing use of these type masks anyway. **Benefits** The most obvious benefit is reduced memory usage and the benefits that brings in memory-bound situations. `float3` is only 3 bytes, in comparison to `MVert` which was 4. When there are millions of vertices this starts to matter more. The other benefits come from using a more generic type. Instead of writing algorithms specifically for `MVert`, code can just use arrays of vectors. This will allow eliminating many temporary arrays or wrappers used to extract positions. Many possible improvements aren't implemented in this patch, though I did switch simplify or remove the process of creating temporary position arrays in a few places. The design clarity that "positions are just another attribute" brings allows removing explicit copying of vertices in some procedural operations-- they are just processed like most other attributes. **Performance** This touches so many areas that it's hard to benchmark exhaustively, but I observed some areas as examples. * The mesh line node with 4 million count was 1.5x (8ms to 12ms) faster. * The Spring splash screen went from ~4.3 to ~4.5 fps. * The subdivision surface modifier/node was slightly faster RNA access through Python may be slightly slower, since now we need a name lookup instead of just a custom data type lookup for each index. **Future Improvements** * Remove uses of "vert_coords" functions: * `BKE_mesh_vert_coords_alloc` * `BKE_mesh_vert_coords_get` * `BKE_mesh_vert_coords_apply{_with_mat4}` * Remove more hidden copying of positions * General simplification now possible in many areas * Convert more code to C++ to use `float3` instead of `float[3]` * Currently `reinterpret_cast` is used for those C-API functions Differential Revision: https://developer.blender.org/D15982
2023-01-10 00:10:43 -05:00
float3 &subdiv_position = ctx->subdiv_positions[subdiv_vertex_index];
subdiv_mesh_ensure_vertex_interpolation(ctx, tls, coarse_face_index, coarse_corner);
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
subdiv_vertex_data_interpolate(ctx, subdiv_vertex_index, &tls->vertex_interpolation, u, v);
eval_final_point(subdiv, ptex_face_index, u, v, subdiv_position);
subdiv_mesh_tag_center_vertex(coarse_face, subdiv_vertex_index, u, v, subdiv_mesh);
subdiv_vertex_orco_evaluate(ctx, ptex_face_index, u, v, subdiv_vertex_index);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Edge subdivision process
* \{ */
2018-07-19 16:06:37 +10:00
static void subdiv_copy_edge_data(SubdivMeshContext *ctx,
const int subdiv_edge_index,
const int coarse_edge_index)
{
if (coarse_edge_index == ORIGINDEX_NONE) {
2022-08-17 18:11:01 -04:00
if (ctx->edge_origindex != nullptr) {
2018-07-18 17:49:12 +02:00
ctx->edge_origindex[subdiv_edge_index] = ORIGINDEX_NONE;
}
return;
}
CustomData_copy_data(&ctx->coarse_mesh->edge_data,
&ctx->subdiv_mesh->edge_data,
coarse_edge_index,
subdiv_edge_index,
1);
if (ctx->settings->use_optimal_display) {
ctx->subdiv_display_edges[subdiv_edge_index] = true;
}
}
static void subdiv_mesh_edge(const ForeachContext *foreach_context,
void * /*tls*/,
const int coarse_edge_index,
const int subdiv_edge_index,
const bool /*is_loose*/,
const int subdiv_v1,
const int subdiv_v2)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
subdiv_copy_edge_data(ctx, subdiv_edge_index, coarse_edge_index);
Mesh: Move edges to a generic attribute Implements #95966, as the final step of #95965. This commit changes the storage of mesh edge vertex indices from the `MEdge` type to the generic `int2` attribute type. This follows the general design for geometry and the attribute system, where the data storage type and the usage semantics are separated. The main benefit of the change is reduced memory usage-- the requirements of storing mesh edges is reduced by 1/3. For example, this saves 8MB on a 1 million vertex grid. This also gives performance benefits to any memory-bound mesh processing algorithm that uses edges. Another benefit is that all of the edge's vertex indices are contiguous. In a few cases, it's helpful to process all of them as `Span<int>` rather than `Span<int2>`. Similarly, the type is more likely to match a generic format used by a library, or code that shouldn't know about specific Blender `Mesh` types. Various Notes: - The `.edge_verts` name is used to reflect a mapping between domains, similar to `.corner_verts`, etc. The period means that it the data shouldn't change arbitrarily by the user or procedural operations. - `edge[0]` is now used instead of `edge.v1` - Signed integers are used instead of unsigned to reduce the mixing of signed-ness, which can be error prone. - All of the previously used core mesh data types (`MVert`, `MEdge`, `MLoop`, `MPoly` are now deprecated. Only generic types are used). - The `vec2i` DNA type is used in the few C files where necessary. Pull Request: https://projects.blender.org/blender/blender/pulls/106638
2023-04-17 13:47:41 +02:00
ctx->subdiv_edges[subdiv_edge_index][0] = subdiv_v1;
ctx->subdiv_edges[subdiv_edge_index][1] = subdiv_v2;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Loops creation/interpolation
* \{ */
static void subdiv_interpolate_corner_data(const SubdivMeshContext *ctx,
const int subdiv_loop_index,
const LoopsForInterpolation *loop_interpolation,
const float u,
const float v)
{
const float weights[4] = {(1.0f - u) * (1.0f - v), u * (1.0f - v), u * v, (1.0f - u) * v};
CustomData_interp(loop_interpolation->corner_data,
&ctx->subdiv_mesh->corner_data,
loop_interpolation->loop_indices,
weights,
2022-08-17 18:11:01 -04:00
nullptr,
4,
subdiv_loop_index);
/* TODO(sergey): Set ORIGINDEX. */
}
static void subdiv_eval_uv_layer(SubdivMeshContext *ctx,
Mesh: Replace MLoop struct with generic attributes Implements #102359. Split the `MLoop` struct into two separate integer arrays called `corner_verts` and `corner_edges`, referring to the vertex each corner is attached to and the next edge around the face at each corner. These arrays can be sliced to give access to the edges or vertices in a face. Then they are often referred to as "poly_verts" or "poly_edges". The main benefits are halving the necessary memory bandwidth when only one array is used and simplifications from using regular integer indices instead of a special-purpose struct. The commit also starts a renaming from "loop" to "corner" in mesh code. Like the other mesh struct of array refactors, forward compatibility is kept by writing files with the older format. This will be done until 4.0 to ease the transition process. Looking at a small portion of the patch should give a good impression for the rest of the changes. I tried to make the changes as small as possible so it's easy to tell the correctness from the diff. Though I found Blender developers have been very inventive over the last decade when finding different ways to loop over the corners in a face. For performance, nearly every piece of code that deals with `Mesh` is slightly impacted. Any algorithm that is memory bottle-necked should see an improvement. For example, here is a comparison of interpolating a vertex float attribute to face corners (Ryzen 3700x): **Before** (Average: 3.7 ms, Min: 3.4 ms) ``` threading::parallel_for(loops.index_range(), 4096, [&](IndexRange range) { for (const int64_t i : range) { dst[i] = src[loops[i].v]; } }); ``` **After** (Average: 2.9 ms, Min: 2.6 ms) ``` array_utils::gather(src, corner_verts, dst); ``` That's an improvement of 28% to the average timings, and it's also a simplification, since an index-based routine can be used instead. For more examples using the new arrays, see the design task. Pull Request: https://projects.blender.org/blender/blender/pulls/104424
2023-03-20 15:55:13 +01:00
const int corner_index,
const int ptex_face_index,
const float u,
const float v)
{
if (ctx->num_uv_layers == 0) {
return;
}
Subdiv *subdiv = ctx->subdiv;
for (int layer_index = 0; layer_index < ctx->num_uv_layers; layer_index++) {
eval_face_varying(
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
subdiv, layer_index, ptex_face_index, u, v, ctx->uv_layers[layer_index][corner_index]);
}
}
static void subdiv_mesh_ensure_loop_interpolation(SubdivMeshContext *ctx,
SubdivMeshTLS *tls,
const int coarse_face_index,
const int coarse_corner)
{
const IndexRange coarse_face = ctx->coarse_faces[coarse_face_index];
/* Check whether we've moved to another corner or face. */
if (tls->loop_interpolation_initialized) {
if (tls->loop_interpolation_coarse_face_index != coarse_face_index ||
tls->loop_interpolation_coarse_corner != coarse_corner)
{
loop_interpolation_end(&tls->loop_interpolation);
tls->loop_interpolation_initialized = false;
}
}
/* Initialize the interpolation. */
if (!tls->loop_interpolation_initialized) {
loop_interpolation_init(ctx, &tls->loop_interpolation, coarse_face);
}
/* Update it for a new corner if needed. */
if (!tls->loop_interpolation_initialized ||
tls->loop_interpolation_coarse_corner != coarse_corner)
{
loop_interpolation_from_corner(ctx, &tls->loop_interpolation, coarse_face, coarse_corner);
2018-08-24 10:26:59 +10:00
}
/* Store settings used for the current state of interpolator. */
tls->loop_interpolation_initialized = true;
tls->loop_interpolation_coarse_face_index = coarse_face_index;
tls->loop_interpolation_coarse_corner = coarse_corner;
}
static void subdiv_mesh_loop(const ForeachContext *foreach_context,
void *tls_v,
const int ptex_face_index,
const float u,
const float v,
const int /*coarse_loop_index*/,
const int coarse_face_index,
2018-08-24 10:26:59 +10:00
const int coarse_corner,
const int subdiv_loop_index,
const int subdiv_vertex_index,
const int subdiv_edge_index)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
SubdivMeshTLS *tls = static_cast<SubdivMeshTLS *>(tls_v);
subdiv_mesh_ensure_loop_interpolation(ctx, tls, coarse_face_index, coarse_corner);
subdiv_interpolate_corner_data(ctx, subdiv_loop_index, &tls->loop_interpolation, u, v);
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
subdiv_eval_uv_layer(ctx, subdiv_loop_index, ptex_face_index, u, v);
ctx->subdiv_corner_verts[subdiv_loop_index] = subdiv_vertex_index;
ctx->subdiv_corner_edges[subdiv_loop_index] = subdiv_edge_index;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Polygons subdivision process
* \{ */
static void subdiv_mesh_face(const ForeachContext *foreach_context,
void * /*tls*/,
const int coarse_face_index,
const int subdiv_face_index,
const int start_loop_index,
Mesh: Replace MPoly struct with offset indices Implements #95967. Currently the `MPoly` struct is 12 bytes, and stores the index of a face's first corner and the number of corners/verts/edges. Polygons and corners are always created in order by Blender, meaning each face's corners will be after the previous face's corners. We can take advantage of this fact and eliminate the redundancy in mesh face storage by only storing a single integer corner offset for each face. The size of the face is then encoded by the offset of the next face. The size of a single integer is 4 bytes, so this reduces memory usage by 3 times. The same method is used for `CurvesGeometry`, so Blender already has an abstraction to simplify using these offsets called `OffsetIndices`. This class is used to easily retrieve a range of corner indices for each face. This also gives the opportunity for sharing some logic with curves. Another benefit of the change is that the offsets and sizes stored in `MPoly` can no longer disagree with each other. Storing faces in the order of their corners can simplify some code too. Face/polygon variables now use the `IndexRange` type, which comes with quite a few utilities that can simplify code. Some: - The offset integer array has to be one longer than the face count to avoid a branch for every face, which means the data is no longer part of the mesh's `CustomData`. - We lose the ability to "reference" an original mesh's offset array until more reusable CoW from #104478 is committed. That will be added in a separate commit. - Since they aren't part of `CustomData`, poly offsets often have to be copied manually. - To simplify using `OffsetIndices` in many places, some functions and structs in headers were moved to only compile in C++. - All meshes created by Blender use the same order for faces and face corners, but just in case, meshes with mismatched order are fixed by versioning code. - `MeshPolygon.totloop` is no longer editable in RNA. This API break is necessary here unfortunately. It should be worth it in 3.6, since that's the best way to allow loading meshes from 4.0, which is important for an LTS version. Pull Request: https://projects.blender.org/blender/blender/pulls/105938
2023-04-04 20:39:28 +02:00
const int /*num_loops*/)
{
BLI_assert(coarse_face_index != ORIGINDEX_NONE);
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
CustomData_copy_data(&ctx->coarse_mesh->face_data,
&ctx->subdiv_mesh->face_data,
coarse_face_index,
subdiv_face_index,
1);
ctx->subdiv_face_offsets[subdiv_face_index] = start_loop_index;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Loose elements subdivision process
* \{ */
static void subdiv_mesh_vertex_loose(const ForeachContext *foreach_context,
void * /*tls*/,
const int coarse_vertex_index,
const int subdiv_vertex_index)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
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
subdiv_vertex_data_copy(ctx, coarse_vertex_index, subdiv_vertex_index);
}
/* Get neighbor edges of the given one.
* - neighbors[0] is an edge adjacent to edge->v1.
* - neighbors[1] is an edge adjacent to edge->v2. */
static std::array<std::optional<int2>, 2> find_edge_neighbors(
const Span<int2> coarse_edges, const GroupedSpan<int> vert_to_edge_map, const int edge_index)
{
/* Vertices which has more than one neighbor are considered infinitely
* sharp. This is also how topology factory treats vertices of a surface
* which are adjacent to a loose edge. */
const auto neighbor_edge_if_single = [&](const int vert) -> std::optional<int2> {
const Span<int> neighbors = vert_to_edge_map[vert];
if (neighbors.size() != 2) {
return std::nullopt;
}
return neighbors[0] == edge_index ? coarse_edges[neighbors[1]] : coarse_edges[neighbors[0]];
};
const int2 edge = coarse_edges[edge_index];
return {neighbor_edge_if_single(edge[0]), neighbor_edge_if_single(edge[1])};
}
static std::array<float3, 4> find_loose_edge_interpolation_positions(
const Span<float3> coarse_positions,
const int2 &coarse_edge,
const std::array<std::optional<int2>, 2> &neighbors)
{
std::array<float3, 4> result;
/* Middle points corresponds to the edge. */
result[1] = coarse_positions[coarse_edge[0]];
result[2] = coarse_positions[coarse_edge[1]];
/* Start point, duplicate from edge start if no neighbor. */
if (const std::optional<int2> &other = neighbors[0]) {
result[0] = coarse_positions[mesh::edge_other_vert(*other, coarse_edge[0])];
}
else {
result[0] = result[1] * 2.0f - result[2];
}
/* End point, duplicate from edge end if no neighbor. */
if (const std::optional<int2> &other = neighbors[1]) {
result[3] = coarse_positions[mesh::edge_other_vert(*other, coarse_edge[1])];
}
else {
result[3] = result[2] * 2.0f - result[1];
}
return result;
}
float3 mesh_interpolate_position_on_edge(const Span<float3> coarse_positions,
const Span<int2> coarse_edges,
const GroupedSpan<int> vert_to_edge_map,
const int coarse_edge_index,
const bool is_simple,
const float u)
{
const int2 edge = coarse_edges[coarse_edge_index];
if (is_simple) {
return math::interpolate(coarse_positions[edge[0]], coarse_positions[edge[1]], u);
}
/* Find neighbors of the coarse edge. */
const std::array<std::optional<int2>, 2> neighbors = find_edge_neighbors(
coarse_edges, vert_to_edge_map, coarse_edge_index);
const std::array<float3, 4> points = find_loose_edge_interpolation_positions(
coarse_positions, edge, neighbors);
float4 weights;
key_curve_position_weights(u, weights, KEY_BSPLINE);
return bke::attribute_math::mix4(weights, points[0], points[1], points[2], points[3]);
}
static void subdiv_mesh_vertex_of_loose_edge_interpolate(SubdivMeshContext *ctx,
Mesh: Move edges to a generic attribute Implements #95966, as the final step of #95965. This commit changes the storage of mesh edge vertex indices from the `MEdge` type to the generic `int2` attribute type. This follows the general design for geometry and the attribute system, where the data storage type and the usage semantics are separated. The main benefit of the change is reduced memory usage-- the requirements of storing mesh edges is reduced by 1/3. For example, this saves 8MB on a 1 million vertex grid. This also gives performance benefits to any memory-bound mesh processing algorithm that uses edges. Another benefit is that all of the edge's vertex indices are contiguous. In a few cases, it's helpful to process all of them as `Span<int>` rather than `Span<int2>`. Similarly, the type is more likely to match a generic format used by a library, or code that shouldn't know about specific Blender `Mesh` types. Various Notes: - The `.edge_verts` name is used to reflect a mapping between domains, similar to `.corner_verts`, etc. The period means that it the data shouldn't change arbitrarily by the user or procedural operations. - `edge[0]` is now used instead of `edge.v1` - Signed integers are used instead of unsigned to reduce the mixing of signed-ness, which can be error prone. - All of the previously used core mesh data types (`MVert`, `MEdge`, `MLoop`, `MPoly` are now deprecated. Only generic types are used). - The `vec2i` DNA type is used in the few C files where necessary. Pull Request: https://projects.blender.org/blender/blender/pulls/106638
2023-04-17 13:47:41 +02:00
const int2 &coarse_edge,
const float u,
const int subdiv_vertex_index)
{
const Mesh *coarse_mesh = ctx->coarse_mesh;
Mesh *subdiv_mesh = ctx->subdiv_mesh;
/* This is never used for end-points (which are copied from the original). */
BLI_assert(u > 0.0f);
BLI_assert(u < 1.0f);
const float interpolation_weights[2] = {1.0f - u, u};
Mesh: Move edges to a generic attribute Implements #95966, as the final step of #95965. This commit changes the storage of mesh edge vertex indices from the `MEdge` type to the generic `int2` attribute type. This follows the general design for geometry and the attribute system, where the data storage type and the usage semantics are separated. The main benefit of the change is reduced memory usage-- the requirements of storing mesh edges is reduced by 1/3. For example, this saves 8MB on a 1 million vertex grid. This also gives performance benefits to any memory-bound mesh processing algorithm that uses edges. Another benefit is that all of the edge's vertex indices are contiguous. In a few cases, it's helpful to process all of them as `Span<int>` rather than `Span<int2>`. Similarly, the type is more likely to match a generic format used by a library, or code that shouldn't know about specific Blender `Mesh` types. Various Notes: - The `.edge_verts` name is used to reflect a mapping between domains, similar to `.corner_verts`, etc. The period means that it the data shouldn't change arbitrarily by the user or procedural operations. - `edge[0]` is now used instead of `edge.v1` - Signed integers are used instead of unsigned to reduce the mixing of signed-ness, which can be error prone. - All of the previously used core mesh data types (`MVert`, `MEdge`, `MLoop`, `MPoly` are now deprecated. Only generic types are used). - The `vec2i` DNA type is used in the few C files where necessary. Pull Request: https://projects.blender.org/blender/blender/pulls/106638
2023-04-17 13:47:41 +02:00
const int coarse_vertex_indices[2] = {coarse_edge[0], coarse_edge[1]};
CustomData_interp(&coarse_mesh->vert_data,
&subdiv_mesh->vert_data,
coarse_vertex_indices,
interpolation_weights,
2022-08-17 18:11:01 -04:00
nullptr,
2,
subdiv_vertex_index);
2022-08-17 18:11:01 -04:00
if (ctx->vert_origindex != nullptr) {
ctx->vert_origindex[subdiv_vertex_index] = ORIGINDEX_NONE;
}
}
static void subdiv_mesh_vertex_of_loose_edge(const ForeachContext *foreach_context,
void * /*tls*/,
const int coarse_edge_index,
const float u,
const int subdiv_vertex_index)
{
2022-08-17 18:11:01 -04:00
SubdivMeshContext *ctx = static_cast<SubdivMeshContext *>(foreach_context->user_data);
Mesh: Move edges to a generic attribute Implements #95966, as the final step of #95965. This commit changes the storage of mesh edge vertex indices from the `MEdge` type to the generic `int2` attribute type. This follows the general design for geometry and the attribute system, where the data storage type and the usage semantics are separated. The main benefit of the change is reduced memory usage-- the requirements of storing mesh edges is reduced by 1/3. For example, this saves 8MB on a 1 million vertex grid. This also gives performance benefits to any memory-bound mesh processing algorithm that uses edges. Another benefit is that all of the edge's vertex indices are contiguous. In a few cases, it's helpful to process all of them as `Span<int>` rather than `Span<int2>`. Similarly, the type is more likely to match a generic format used by a library, or code that shouldn't know about specific Blender `Mesh` types. Various Notes: - The `.edge_verts` name is used to reflect a mapping between domains, similar to `.corner_verts`, etc. The period means that it the data shouldn't change arbitrarily by the user or procedural operations. - `edge[0]` is now used instead of `edge.v1` - Signed integers are used instead of unsigned to reduce the mixing of signed-ness, which can be error prone. - All of the previously used core mesh data types (`MVert`, `MEdge`, `MLoop`, `MPoly` are now deprecated. Only generic types are used). - The `vec2i` DNA type is used in the few C files where necessary. Pull Request: https://projects.blender.org/blender/blender/pulls/106638
2023-04-17 13:47:41 +02:00
const int2 &coarse_edge = ctx->coarse_edges[coarse_edge_index];
const bool is_simple = ctx->subdiv->settings.is_simple;
/* Interpolate custom data when not an end point.
* This data has already been copied from the original vertex by #subdiv_mesh_vertex_loose. */
2022-01-07 15:30:14 +11:00
if (!ELEM(u, 0.0, 1.0)) {
subdiv_mesh_vertex_of_loose_edge_interpolate(ctx, coarse_edge, u, subdiv_vertex_index);
}
/* Interpolate coordinate. */
ctx->subdiv_positions[subdiv_vertex_index] = mesh_interpolate_position_on_edge(
ctx->coarse_positions,
ctx->coarse_edges,
ctx->vert_to_edge_map,
coarse_edge_index,
is_simple,
u);
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Initialization
* \{ */
static void setup_foreach_callbacks(const SubdivMeshContext *subdiv_context,
ForeachContext *foreach_context)
{
memset(foreach_context, 0, sizeof(*foreach_context));
/* General information. */
foreach_context->topology_info = subdiv_mesh_topology_info;
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
/* Every boundary geometry. Used for displacement averaging. */
if (subdiv_context->have_displacement) {
foreach_context->vertex_every_corner = subdiv_mesh_vertex_displacement_every_corner;
foreach_context->vertex_every_edge = subdiv_mesh_vertex_displacement_every_edge;
}
foreach_context->vertex_corner = subdiv_mesh_vertex_corner;
foreach_context->vertex_edge = subdiv_mesh_vertex_edge;
foreach_context->vertex_inner = subdiv_mesh_vertex_inner;
foreach_context->edge = subdiv_mesh_edge;
foreach_context->loop = subdiv_mesh_loop;
foreach_context->poly = subdiv_mesh_face;
foreach_context->vertex_loose = subdiv_mesh_vertex_loose;
foreach_context->vertex_of_loose_edge = subdiv_mesh_vertex_of_loose_edge;
foreach_context->user_data_tls_free = subdiv_mesh_tls_free;
}
/** \} */
/* -------------------------------------------------------------------- */
/** \name Public entry point
* \{ */
Mesh *subdiv_to_mesh(Subdiv *subdiv, const ToMeshSettings *settings, const Mesh *coarse_mesh)
{
stats_begin(&subdiv->stats, SUBDIV_STATS_SUBDIV_TO_MESH);
/* Make sure evaluator is up to date with possible new topology, and that
2019-11-25 00:55:11 +11:00
* it is refined for the new positions of coarse vertices. */
if (!eval_begin_from_mesh(subdiv, coarse_mesh, {}, SUBDIV_EVALUATOR_TYPE_CPU, nullptr)) {
/* This could happen in two situations:
* - OpenSubdiv is disabled.
* - Something totally bad happened, and OpenSubdiv rejected our
* topology.
* In either way, we can't safely continue. */
if (coarse_mesh->faces_num) {
stats_end(&subdiv->stats, SUBDIV_STATS_SUBDIV_TO_MESH);
2022-08-17 18:11:01 -04:00
return nullptr;
}
}
/* Initialize subdivision mesh creation context. */
SubdivMeshContext subdiv_context{};
subdiv_context.settings = settings;
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
subdiv_context.coarse_mesh = coarse_mesh;
subdiv_context.coarse_positions = coarse_mesh->vert_positions();
subdiv_context.coarse_edges = coarse_mesh->edges();
subdiv_context.coarse_faces = coarse_mesh->faces();
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
subdiv_context.coarse_corner_verts = coarse_mesh->corner_verts();
if (coarse_mesh->loose_edges().count > 0) {
subdiv_context.vert_to_edge_map = mesh::build_vert_to_edge_map(
subdiv_context.coarse_edges,
coarse_mesh->verts_num,
subdiv_context.vert_to_edge_offsets,
subdiv_context.vert_to_edge_indices);
}
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
subdiv_context.subdiv = subdiv;
2022-08-17 18:11:01 -04:00
subdiv_context.have_displacement = (subdiv->displacement_evaluator != nullptr);
/* Multi-threaded traversal/evaluation. */
stats_begin(&subdiv->stats, SUBDIV_STATS_SUBDIV_TO_MESH_GEOMETRY);
ForeachContext foreach_context;
setup_foreach_callbacks(&subdiv_context, &foreach_context);
SubdivMeshTLS tls{};
foreach_context.user_data = &subdiv_context;
foreach_context.user_data_tls_size = sizeof(SubdivMeshTLS);
foreach_context.user_data_tls = &tls;
foreach_subdiv_geometry(subdiv, &foreach_context, settings, coarse_mesh);
stats_end(&subdiv->stats, SUBDIV_STATS_SUBDIV_TO_MESH_GEOMETRY);
Mesh *result = subdiv_context.subdiv_mesh;
CustomData_add_layer_named_with_data(&result->corner_data,
CD_PROP_INT32,
subdiv_context.subdiv_corner_verts,
result->corners_num,
".corner_vert",
nullptr);
subdiv_context.subdiv_corner_verts = nullptr;
CustomData_add_layer_named_with_data(&result->corner_data,
CD_PROP_INT32,
subdiv_context.subdiv_corner_edges,
result->corners_num,
".corner_edge",
nullptr);
subdiv_context.subdiv_corner_edges = nullptr;
/* NOTE: Using normals from the limit surface gives different results than Blender's vertex
* normal calculation. Since vertex normals are supposed to be a consistent cache, don't bother
* calculating them here. The work may have been pointless anyway if the mesh is deformed or
* changed afterwards. */
/* Move the optimal display edge array to the final bit vector. */
if (!subdiv_context.subdiv_display_edges.is_empty()) {
result->runtime->subsurf_optimal_display_edges = BitVector<>(
subdiv_context.subdiv_display_edges);
}
if (coarse_mesh->verts_no_face().count == 0) {
result->tag_loose_verts_none();
}
if (coarse_mesh->loose_edges().count == 0) {
result->tag_loose_edges_none();
}
result->tag_overlapping_none();
if (subdiv->settings.is_simple) {
/* In simple subdivision, min and max positions are not changed, avoid recomputing bounds. */
result->runtime->bounds_cache = coarse_mesh->runtime->bounds_cache;
}
// BKE_mesh_validate(result, true, true);
stats_end(&subdiv->stats, SUBDIV_STATS_SUBDIV_TO_MESH);
subdiv_mesh_context_free(&subdiv_context);
return result;
}
/** \} */
} // namespace blender::bke::subdiv