Files
test/source/blender/blenkernel/intern/mesh_merge_customdata.cc
Hans Goudey 05952aa94d 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 (e9f82d3dc7, 410a6efb74). 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

143 lines
4.5 KiB
C++

/* SPDX-License-Identifier: GPL-2.0-or-later */
/** \file
* \ingroup bke
*/
#include "MEM_guardedalloc.h"
#include "DNA_mesh_types.h"
#include "DNA_meshdata_types.h"
#include "BLI_math.h"
#include "BLI_task.hh"
#include "BLI_utildefines.h"
#include "BKE_customdata.h"
#include "BKE_mesh.h"
#include "BKE_mesh_mapping.h"
#include "BLI_memarena.h"
#include "BLI_strict_flags.h"
using namespace blender;
enum {
CMP_CLOSE = 0,
CMP_EQUAL = 1,
CMP_APART = 2,
};
static int compare_v2_classify(const float uv_a[2], const float uv_b[2])
{
if (uv_a[0] == uv_b[0] && uv_a[1] == uv_b[1]) {
return CMP_EQUAL;
}
/* NOTE(@campbellbarton): that the ULP value is the primary value used to compare relative
* values as the absolute value doesn't account for float precision at difference scales.
* - For subdivision-surface ULP of 3 is sufficient,
* although this value is extremely small.
* - For bevel the ULP of 12 is sufficient to merge UV's that appear to be connected
* with bevel on Suzanne beveled 15% with 6 segments.
*
* These values could be tweaked but should be kept on the small side to prevent
* unintentional joining of intentionally dis-connected UV's.
*
* Before v2.91 the threshold was either (`1e-4` or `0.05 / image_size` for selection picking).
* So picking used a threshold of `1e-4` for a 500x500 image and `1e-5` for a 5000x5000 image.
* Given this value worked reasonably well for a long time, the absolute difference should
* never exceed `1e-4` (#STD_UV_CONNECT_LIMIT which is still used in a few areas). */
const float diff_abs = 1e-12f;
const int diff_ulp = 12;
if (compare_ff_relative(uv_a[0], uv_b[0], diff_abs, diff_ulp) &&
compare_ff_relative(uv_a[1], uv_b[1], diff_abs, diff_ulp)) {
return CMP_CLOSE;
}
return CMP_APART;
}
static void merge_uvs_for_vertex(const Span<int> loops_for_vert, Span<MLoopUV *> mloopuv_layers)
{
if (loops_for_vert.size() <= 1) {
return;
}
/* Manipulate a copy of the loop indices, de-duplicating UV's per layer. */
Vector<int, 32> loops_merge;
loops_merge.reserve(loops_for_vert.size());
for (MLoopUV *mloopuv : mloopuv_layers) {
BLI_assert(loops_merge.is_empty());
loops_merge.extend_unchecked(loops_for_vert);
while (loops_merge.size() > 1) {
uint i_last = (uint)loops_merge.size() - 1;
const float *uv_src = mloopuv[loops_merge[0]].uv;
for (uint i = 1; i <= i_last;) {
float *uv_dst = mloopuv[loops_merge[i]].uv;
switch (compare_v2_classify(uv_src, uv_dst)) {
case CMP_CLOSE: {
uv_dst[0] = uv_src[0];
uv_dst[1] = uv_src[1];
ATTR_FALLTHROUGH;
}
case CMP_EQUAL: {
loops_merge[i] = loops_merge[i_last--];
break;
}
case CMP_APART: {
/* Doesn't match, check the next UV. */
i++;
break;
}
default: {
BLI_assert_unreachable();
}
}
}
/* Finished de-duplicating with the first index, throw it away. */
loops_merge[0] = loops_merge[i_last];
loops_merge.resize(i_last);
}
loops_merge.clear();
}
}
void BKE_mesh_merge_customdata_for_apply_modifier(Mesh *me)
{
if (me->totloop == 0) {
return;
}
const int mloopuv_layers_num = CustomData_number_of_layers(&me->ldata, CD_MLOOPUV);
if (mloopuv_layers_num == 0) {
return;
}
int *vert_map_mem;
struct MeshElemMap *vert_to_loop;
BKE_mesh_vert_loop_map_create(&vert_to_loop,
&vert_map_mem,
BKE_mesh_polygons(me),
BKE_mesh_loops(me),
me->totvert,
me->totpoly,
me->totloop);
Vector<MLoopUV *> mloopuv_layers;
mloopuv_layers.reserve(mloopuv_layers_num);
for (int a = 0; a < mloopuv_layers_num; a++) {
MLoopUV *mloopuv = static_cast<MLoopUV *>(CustomData_get_layer_n(&me->ldata, CD_MLOOPUV, a));
mloopuv_layers.append_unchecked(mloopuv);
}
Span<MLoopUV *> mloopuv_layers_as_span = mloopuv_layers.as_span();
threading::parallel_for(IndexRange(me->totvert), 1024, [&](IndexRange range) {
for (const int64_t v_index : range) {
MeshElemMap &loops_for_vert = vert_to_loop[v_index];
Span<int> loops_for_vert_span(loops_for_vert.indices, loops_for_vert.count);
merge_uvs_for_vertex(loops_for_vert_span, mloopuv_layers_as_span);
}
});
MEM_freeN(vert_to_loop);
MEM_freeN(vert_map_mem);
}