Files
test/intern/cycles/scene/mesh_subdivision.cpp
William Leeson 6c03339e48 Cycles: reduce mesh memory usage by unflattening
To improve mesh upload speeds and reduce the size of the scene data which allows larger scenes to be rendered.

The meshes in Cycles are currently stored as flattened meshes, where each triangle is stored as a set of 3 vertices. Unflattening writes out the vertices in a list according to the index buffer. This uses a lot of memory and for current hardware does not provide a noticeable benefit. This change unflattens the mesh by directly using the meshes vertex and index buffers directly and skips the unflattening. This change allows for larger scenes and also a reduction in the sizes of the meshes. Further it results in a decrease the amount of time it takes to upload the data to a GPU. This is especially important for when multiple GPUs are used in a single machine.

Pull Request #105173
2023-02-27 10:39:19 +01:00

681 lines
20 KiB
C++

/* SPDX-License-Identifier: Apache-2.0
* Copyright 2011-2022 Blender Foundation */
#include "scene/attribute.h"
#include "scene/camera.h"
#include "scene/mesh.h"
#include "subd/patch.h"
#include "subd/patch_table.h"
#include "subd/split.h"
#include "util/algorithm.h"
#include "util/foreach.h"
#include "util/hash.h"
CCL_NAMESPACE_BEGIN
#ifdef WITH_OPENSUBDIV
CCL_NAMESPACE_END
# include <opensubdiv/far/patchMap.h>
# include <opensubdiv/far/patchTableFactory.h>
# include <opensubdiv/far/primvarRefiner.h>
# include <opensubdiv/far/topologyRefinerFactory.h>
/* specializations of TopologyRefinerFactory for ccl::Mesh */
namespace OpenSubdiv {
namespace OPENSUBDIV_VERSION {
namespace Far {
template<>
bool TopologyRefinerFactory<ccl::Mesh>::resizeComponentTopology(TopologyRefiner &refiner,
ccl::Mesh const &mesh)
{
setNumBaseVertices(refiner, mesh.get_verts().size());
setNumBaseFaces(refiner, mesh.get_num_subd_faces());
for (int i = 0; i < mesh.get_num_subd_faces(); i++) {
setNumBaseFaceVertices(refiner, i, mesh.get_subd_num_corners()[i]);
}
return true;
}
template<>
bool TopologyRefinerFactory<ccl::Mesh>::assignComponentTopology(TopologyRefiner &refiner,
ccl::Mesh const &mesh)
{
const ccl::array<int> &subd_face_corners = mesh.get_subd_face_corners();
const ccl::array<int> &subd_start_corner = mesh.get_subd_start_corner();
const ccl::array<int> &subd_num_corners = mesh.get_subd_num_corners();
for (int i = 0; i < mesh.get_num_subd_faces(); i++) {
IndexArray face_verts = getBaseFaceVertices(refiner, i);
int start_corner = subd_start_corner[i];
int *corner = &subd_face_corners[start_corner];
for (int j = 0; j < subd_num_corners[i]; j++, corner++) {
face_verts[j] = *corner;
}
}
return true;
}
template<>
bool TopologyRefinerFactory<ccl::Mesh>::assignComponentTags(TopologyRefiner &refiner,
ccl::Mesh const &mesh)
{
/* Historical maximum crease weight used at Pixar, influencing the maximum in OpenSubDiv. */
static constexpr float CREASE_SCALE = 10.0f;
size_t num_creases = mesh.get_subd_creases_weight().size();
size_t num_vertex_creases = mesh.get_subd_vert_creases().size();
/* The last loop is over the vertices, so early exit to avoid iterating them needlessly. */
if (num_creases == 0 && num_vertex_creases == 0) {
return true;
}
for (int i = 0; i < num_creases; i++) {
ccl::Mesh::SubdEdgeCrease crease = mesh.get_subd_crease(i);
Index edge = findBaseEdge(refiner, crease.v[0], crease.v[1]);
if (edge != INDEX_INVALID) {
setBaseEdgeSharpness(refiner, edge, crease.crease * CREASE_SCALE);
}
}
std::map<int, float> vertex_creases;
for (size_t i = 0; i < num_vertex_creases; ++i) {
const int vertex_idx = mesh.get_subd_vert_creases()[i];
const float weight = mesh.get_subd_vert_creases_weight()[i];
vertex_creases[vertex_idx] = weight * CREASE_SCALE;
}
for (int i = 0; i < mesh.get_verts().size(); i++) {
float sharpness = 0.0f;
std::map<int, float>::const_iterator iter = vertex_creases.find(i);
if (iter != vertex_creases.end()) {
sharpness = iter->second;
}
ConstIndexArray vert_edges = getBaseVertexEdges(refiner, i);
if (vert_edges.size() == 2) {
const float sharpness0 = refiner.getLevel(0).getEdgeSharpness(vert_edges[0]);
const float sharpness1 = refiner.getLevel(0).getEdgeSharpness(vert_edges[1]);
sharpness += ccl::min(sharpness0, sharpness1);
sharpness = ccl::min(sharpness, CREASE_SCALE);
}
if (sharpness != 0.0f) {
setBaseVertexSharpness(refiner, i, sharpness);
}
}
return true;
}
template<>
bool TopologyRefinerFactory<ccl::Mesh>::assignFaceVaryingTopology(TopologyRefiner & /*refiner*/,
ccl::Mesh const & /*mesh*/)
{
return true;
}
template<>
void TopologyRefinerFactory<ccl::Mesh>::reportInvalidTopology(TopologyError /*err_code*/,
char const * /*msg*/,
ccl::Mesh const & /*mesh*/)
{
}
} /* namespace Far */
} /* namespace OPENSUBDIV_VERSION */
} /* namespace OpenSubdiv */
CCL_NAMESPACE_BEGIN
using namespace OpenSubdiv;
/* struct that implements OpenSubdiv's vertex interface */
template<typename T> struct OsdValue {
T value;
OsdValue()
{
}
void Clear(void * = 0)
{
memset(&value, 0, sizeof(T));
}
void AddWithWeight(OsdValue<T> const &src, float weight)
{
value += src.value * weight;
}
};
template<> void OsdValue<uchar4>::AddWithWeight(OsdValue<uchar4> const &src, float weight)
{
for (int i = 0; i < 4; i++) {
value[i] += (uchar)(src.value[i] * weight);
}
}
/* class for holding OpenSubdiv data used during tessellation */
class OsdData {
Mesh *mesh;
vector<OsdValue<float3>> verts;
Far::TopologyRefiner *refiner;
Far::PatchTable *patch_table;
Far::PatchMap *patch_map;
public:
OsdData() : mesh(NULL), refiner(NULL), patch_table(NULL), patch_map(NULL)
{
}
~OsdData()
{
delete refiner;
delete patch_table;
delete patch_map;
}
void build_from_mesh(Mesh *mesh_)
{
mesh = mesh_;
/* type and options */
Sdc::SchemeType type = Sdc::SCHEME_CATMARK;
Sdc::Options options;
options.SetVtxBoundaryInterpolation(Sdc::Options::VTX_BOUNDARY_EDGE_ONLY);
/* create refiner */
refiner = Far::TopologyRefinerFactory<Mesh>::Create(
*mesh, Far::TopologyRefinerFactory<Mesh>::Options(type, options));
/* adaptive refinement */
int max_isolation = calculate_max_isolation();
refiner->RefineAdaptive(Far::TopologyRefiner::AdaptiveOptions(max_isolation));
/* create patch table */
Far::PatchTableFactory::Options patch_options;
patch_options.endCapType = Far::PatchTableFactory::Options::ENDCAP_GREGORY_BASIS;
patch_table = Far::PatchTableFactory::Create(*refiner, patch_options);
/* interpolate verts */
int num_refiner_verts = refiner->GetNumVerticesTotal();
int num_local_points = patch_table->GetNumLocalPoints();
verts.resize(num_refiner_verts + num_local_points);
for (int i = 0; i < mesh->get_verts().size(); i++) {
verts[i].value = mesh->get_verts()[i];
}
OsdValue<float3> *src = verts.data();
for (int i = 0; i < refiner->GetMaxLevel(); i++) {
OsdValue<float3> *dest = src + refiner->GetLevel(i).GetNumVertices();
Far::PrimvarRefiner(*refiner).Interpolate(i + 1, src, dest);
src = dest;
}
if (num_local_points) {
patch_table->ComputeLocalPointValues(&verts[0], &verts[num_refiner_verts]);
}
/* create patch map */
patch_map = new Far::PatchMap(*patch_table);
}
void subdivide_attribute(Attribute &attr)
{
Far::PrimvarRefiner primvar_refiner(*refiner);
if (attr.element == ATTR_ELEMENT_VERTEX) {
int num_refiner_verts = refiner->GetNumVerticesTotal();
int num_local_points = patch_table->GetNumLocalPoints();
attr.resize(num_refiner_verts + num_local_points);
attr.flags |= ATTR_FINAL_SIZE;
char *src = attr.buffer.data();
for (int i = 0; i < refiner->GetMaxLevel(); i++) {
char *dest = src + refiner->GetLevel(i).GetNumVertices() * attr.data_sizeof();
if (attr.same_storage(attr.type, TypeDesc::TypeFloat)) {
primvar_refiner.Interpolate(i + 1, (OsdValue<float> *)src, (OsdValue<float> *&)dest);
}
else if (attr.same_storage(attr.type, TypeFloat2)) {
primvar_refiner.Interpolate(i + 1, (OsdValue<float2> *)src, (OsdValue<float2> *&)dest);
// float3 is not interchangeable with float4 and so needs to be handled
// separately
}
else if (attr.same_storage(attr.type, TypeFloat4)) {
primvar_refiner.Interpolate(i + 1, (OsdValue<float4> *)src, (OsdValue<float4> *&)dest);
}
else {
primvar_refiner.Interpolate(i + 1, (OsdValue<float3> *)src, (OsdValue<float3> *&)dest);
}
src = dest;
}
if (num_local_points) {
if (attr.same_storage(attr.type, TypeDesc::TypeFloat)) {
patch_table->ComputeLocalPointValues(
(OsdValue<float> *)&attr.buffer[0],
(OsdValue<float> *)&attr.buffer[num_refiner_verts * attr.data_sizeof()]);
}
else if (attr.same_storage(attr.type, TypeFloat2)) {
patch_table->ComputeLocalPointValues(
(OsdValue<float2> *)&attr.buffer[0],
(OsdValue<float2> *)&attr.buffer[num_refiner_verts * attr.data_sizeof()]);
}
else if (attr.same_storage(attr.type, TypeFloat4)) {
// float3 is not interchangeable with float4 and so needs to be handled
// separately
patch_table->ComputeLocalPointValues(
(OsdValue<float4> *)&attr.buffer[0],
(OsdValue<float4> *)&attr.buffer[num_refiner_verts * attr.data_sizeof()]);
}
else {
// float3 is not interchangeable with float4 and so needs to be handled
// separately
patch_table->ComputeLocalPointValues(
(OsdValue<float3> *)&attr.buffer[0],
(OsdValue<float3> *)&attr.buffer[num_refiner_verts * attr.data_sizeof()]);
}
}
}
else if (attr.element == ATTR_ELEMENT_CORNER || attr.element == ATTR_ELEMENT_CORNER_BYTE) {
// TODO(mai): fvar interpolation
}
}
int calculate_max_isolation()
{
/* loop over all edges to find longest in screen space */
const Far::TopologyLevel &level = refiner->GetLevel(0);
const SubdParams *subd_params = mesh->get_subd_params();
Transform objecttoworld = subd_params->objecttoworld;
Camera *cam = subd_params->camera;
float longest_edge = 0.0f;
for (size_t i = 0; i < level.GetNumEdges(); i++) {
Far::ConstIndexArray verts = level.GetEdgeVertices(i);
float3 a = mesh->get_verts()[verts[0]];
float3 b = mesh->get_verts()[verts[1]];
float edge_len;
if (cam) {
a = transform_point(&objecttoworld, a);
b = transform_point(&objecttoworld, b);
edge_len = len(a - b) / cam->world_to_raster_size((a + b) * 0.5f);
}
else {
edge_len = len(a - b);
}
longest_edge = max(longest_edge, edge_len);
}
/* calculate isolation level */
int isolation = (int)(log2f(max(longest_edge / subd_params->dicing_rate, 1.0f)) + 1.0f);
return min(isolation, 10);
}
friend struct OsdPatch;
friend class Mesh;
};
/* ccl::Patch implementation that uses OpenSubdiv for eval */
struct OsdPatch : Patch {
OsdData *osd_data;
OsdPatch()
{
}
OsdPatch(OsdData *data) : osd_data(data)
{
}
void eval(float3 *P, float3 *dPdu, float3 *dPdv, float3 *N, float u, float v)
{
const Far::PatchTable::PatchHandle *handle = osd_data->patch_map->FindPatch(
patch_index, (double)u, (double)v);
assert(handle);
float p_weights[20], du_weights[20], dv_weights[20];
osd_data->patch_table->EvaluateBasis(*handle, u, v, p_weights, du_weights, dv_weights);
Far::ConstIndexArray cv = osd_data->patch_table->GetPatchVertices(*handle);
float3 du, dv;
if (P)
*P = zero_float3();
du = zero_float3();
dv = zero_float3();
for (int i = 0; i < cv.size(); i++) {
float3 p = osd_data->verts[cv[i]].value;
if (P)
*P += p * p_weights[i];
du += p * du_weights[i];
dv += p * dv_weights[i];
}
if (dPdu)
*dPdu = du;
if (dPdv)
*dPdv = dv;
if (N) {
*N = cross(du, dv);
float t = len(*N);
*N = (t != 0.0f) ? *N / t : make_float3(0.0f, 0.0f, 1.0f);
}
}
};
#endif
void Mesh::tessellate(DiagSplit *split)
{
/* reset the number of subdivision vertices, in case the Mesh was not cleared
* between calls or data updates */
num_subd_verts = 0;
#ifdef WITH_OPENSUBDIV
OsdData osd_data;
bool need_packed_patch_table = false;
if (subdivision_type == SUBDIVISION_CATMULL_CLARK) {
if (get_num_subd_faces()) {
osd_data.build_from_mesh(this);
}
}
else
#endif
{
/* force linear subdivision if OpenSubdiv is unavailable to avoid
* falling into catmull-clark code paths by accident
*/
subdivision_type = SUBDIVISION_LINEAR;
/* force disable attribute subdivision for same reason as above */
foreach (Attribute &attr, subd_attributes.attributes) {
attr.flags &= ~ATTR_SUBDIVIDED;
}
}
int num_faces = get_num_subd_faces();
Attribute *attr_vN = subd_attributes.find(ATTR_STD_VERTEX_NORMAL);
float3 *vN = (attr_vN) ? attr_vN->data_float3() : NULL;
/* count patches */
int num_patches = 0;
for (int f = 0; f < num_faces; f++) {
SubdFace face = get_subd_face(f);
if (face.is_quad()) {
num_patches++;
}
else {
num_patches += face.num_corners;
}
}
/* build patches from faces */
#ifdef WITH_OPENSUBDIV
if (subdivision_type == SUBDIVISION_CATMULL_CLARK) {
vector<OsdPatch> osd_patches(num_patches, &osd_data);
OsdPatch *patch = osd_patches.data();
for (int f = 0; f < num_faces; f++) {
SubdFace face = get_subd_face(f);
if (face.is_quad()) {
patch->patch_index = face.ptex_offset;
patch->from_ngon = false;
patch->shader = face.shader;
patch++;
}
else {
for (int corner = 0; corner < face.num_corners; corner++) {
patch->patch_index = face.ptex_offset + corner;
patch->from_ngon = true;
patch->shader = face.shader;
patch++;
}
}
}
/* split patches */
split->split_patches(osd_patches.data(), sizeof(OsdPatch));
}
else
#endif
{
vector<LinearQuadPatch> linear_patches(num_patches);
LinearQuadPatch *patch = linear_patches.data();
for (int f = 0; f < num_faces; f++) {
SubdFace face = get_subd_face(f);
if (face.is_quad()) {
float3 *hull = patch->hull;
float3 *normals = patch->normals;
patch->patch_index = face.ptex_offset;
patch->from_ngon = false;
for (int i = 0; i < 4; i++) {
hull[i] = verts[subd_face_corners[face.start_corner + i]];
}
if (face.smooth) {
for (int i = 0; i < 4; i++) {
normals[i] = vN[subd_face_corners[face.start_corner + i]];
}
}
else {
float3 N = face.normal(this);
for (int i = 0; i < 4; i++) {
normals[i] = N;
}
}
swap(hull[2], hull[3]);
swap(normals[2], normals[3]);
patch->shader = face.shader;
patch++;
}
else {
/* ngon */
float3 center_vert = zero_float3();
float3 center_normal = zero_float3();
float inv_num_corners = 1.0f / float(face.num_corners);
for (int corner = 0; corner < face.num_corners; corner++) {
center_vert += verts[subd_face_corners[face.start_corner + corner]] * inv_num_corners;
center_normal += vN[subd_face_corners[face.start_corner + corner]] * inv_num_corners;
}
for (int corner = 0; corner < face.num_corners; corner++) {
float3 *hull = patch->hull;
float3 *normals = patch->normals;
patch->patch_index = face.ptex_offset + corner;
patch->from_ngon = true;
patch->shader = face.shader;
hull[0] =
verts[subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)]];
hull[1] =
verts[subd_face_corners[face.start_corner + mod(corner + 1, face.num_corners)]];
hull[2] =
verts[subd_face_corners[face.start_corner + mod(corner - 1, face.num_corners)]];
hull[3] = center_vert;
hull[1] = (hull[1] + hull[0]) * 0.5;
hull[2] = (hull[2] + hull[0]) * 0.5;
if (face.smooth) {
normals[0] =
vN[subd_face_corners[face.start_corner + mod(corner + 0, face.num_corners)]];
normals[1] =
vN[subd_face_corners[face.start_corner + mod(corner + 1, face.num_corners)]];
normals[2] =
vN[subd_face_corners[face.start_corner + mod(corner - 1, face.num_corners)]];
normals[3] = center_normal;
normals[1] = (normals[1] + normals[0]) * 0.5;
normals[2] = (normals[2] + normals[0]) * 0.5;
}
else {
float3 N = face.normal(this);
for (int i = 0; i < 4; i++) {
normals[i] = N;
}
}
patch++;
}
}
}
/* split patches */
split->split_patches(linear_patches.data(), sizeof(LinearQuadPatch));
}
/* interpolate center points for attributes */
foreach (Attribute &attr, subd_attributes.attributes) {
#ifdef WITH_OPENSUBDIV
if (subdivision_type == SUBDIVISION_CATMULL_CLARK && attr.flags & ATTR_SUBDIVIDED) {
if (attr.element == ATTR_ELEMENT_CORNER || attr.element == ATTR_ELEMENT_CORNER_BYTE) {
/* keep subdivision for corner attributes disabled for now */
attr.flags &= ~ATTR_SUBDIVIDED;
}
else if (get_num_subd_faces()) {
osd_data.subdivide_attribute(attr);
need_packed_patch_table = true;
continue;
}
}
#endif
char *data = attr.data();
size_t stride = attr.data_sizeof();
int ngons = 0;
switch (attr.element) {
case ATTR_ELEMENT_VERTEX: {
for (int f = 0; f < num_faces; f++) {
SubdFace face = get_subd_face(f);
if (!face.is_quad()) {
char *center = data + (verts.size() - num_subd_verts + ngons) * stride;
attr.zero_data(center);
float inv_num_corners = 1.0f / float(face.num_corners);
for (int corner = 0; corner < face.num_corners; corner++) {
attr.add_with_weight(center,
data + subd_face_corners[face.start_corner + corner] * stride,
inv_num_corners);
}
ngons++;
}
}
} break;
case ATTR_ELEMENT_VERTEX_MOTION: {
// TODO(mai): implement
} break;
case ATTR_ELEMENT_CORNER: {
for (int f = 0; f < num_faces; f++) {
SubdFace face = get_subd_face(f);
if (!face.is_quad()) {
char *center = data + (subd_face_corners.size() + ngons) * stride;
attr.zero_data(center);
float inv_num_corners = 1.0f / float(face.num_corners);
for (int corner = 0; corner < face.num_corners; corner++) {
attr.add_with_weight(
center, data + (face.start_corner + corner) * stride, inv_num_corners);
}
ngons++;
}
}
} break;
case ATTR_ELEMENT_CORNER_BYTE: {
for (int f = 0; f < num_faces; f++) {
SubdFace face = get_subd_face(f);
if (!face.is_quad()) {
uchar *center = (uchar *)data + (subd_face_corners.size() + ngons) * stride;
float inv_num_corners = 1.0f / float(face.num_corners);
float4 val = zero_float4();
for (int corner = 0; corner < face.num_corners; corner++) {
for (int i = 0; i < 4; i++) {
val[i] += float(*(data + (face.start_corner + corner) * stride + i)) *
inv_num_corners;
}
}
for (int i = 0; i < 4; i++) {
center[i] = uchar(min(max(val[i], 0.0f), 255.0f));
}
ngons++;
}
}
} break;
default:
break;
}
}
#ifdef WITH_OPENSUBDIV
/* pack patch tables */
if (need_packed_patch_table) {
delete patch_table;
patch_table = new PackedPatchTable;
patch_table->pack(osd_data.patch_table);
}
#endif
}
CCL_NAMESPACE_END