Further speedup of new obj exporter.

This change from Aras further parallelizes wihin large meshes (the previous one
just parallelized over objects).

Some stats: on A Windows machine, AMD Ryzen (32 threads):

(one mesh) Monkey subdivided to level 6: 4.9s -> 1.2s (blender 3.1 was 6.3s; 3.0 was 49.4s).
(one mesh) "Rungholt" minecraft level: 8.5s -> 2.9s (3.1 was 10.5s; 3.0 was 73.7s).
(lots of meshes) Blender 3 splash: 6.2s -> 5.2s (3.1 was 48.9s; 3.0 was 392.3s).

On a Linux machine (Threadripper, 48 threads, writing to SSD):
Monkey - 5.08s -> 1.18s (4.2x speedup)
Rungholt - 9.52s -> 3.22s (2.95x speedup)
Blender 3 splash - 5.91s -> 4.61s (1.28x speedup)

For details see patch D14028.
This commit is contained in:
Aras Pranckevicius 2022-02-06 14:28:22 -05:00 committed by Howard Trickey
parent 96cda4da28
commit 9261bc9476
3 changed files with 133 additions and 123 deletions

View File

@ -24,6 +24,7 @@
#include "BKE_blender_version.h"
#include "BLI_path_util.h"
#include "BLI_task.hh"
#include "obj_export_mesh.hh"
#include "obj_export_mtl.hh"
@ -167,108 +168,85 @@ void OBJWriter::write_object_name(FormatHandler<eFileType::OBJ> &fh,
fh.write<eOBJSyntaxElement::object_name>(object_name);
}
/* Split up large meshes into multi-threaded jobs; each job processes
* this amount of items. */
static const int chunk_size = 32768;
static int calc_chunk_count(int count)
{
return (count + chunk_size - 1) / chunk_size;
}
/* Write /tot_count/ items to OBJ file output. Each item is written
* by a /function/ that should be independent from other items.
* If the amount of items is large enough (> chunk_size), then writing
* will be done in parallel, into temporary FormatHandler buffers that
* will be written into the final /fh/ buffer at the end.
*/
template<typename Function>
void obj_parallel_chunked_output(FormatHandler<eFileType::OBJ> &fh,
int tot_count,
const Function &function)
{
if (tot_count <= 0) {
return;
}
/* If we have just one chunk, process it directly into the output
* buffer - avoids all the job scheduling and temporary vector allocation
* overhead. */
const int chunk_count = calc_chunk_count(tot_count);
if (chunk_count == 1) {
for (int i = 0; i < tot_count; i++) {
function(fh, i);
}
return;
}
/* Give each chunk its own temporary output buffer, and process them in parallel. */
std::vector<FormatHandler<eFileType::OBJ>> buffers(chunk_count);
blender::threading::parallel_for(IndexRange(chunk_count), 1, [&](IndexRange range) {
for (const int r : range) {
int i_start = r * chunk_size;
int i_end = std::min(i_start + chunk_size, tot_count);
auto &buf = buffers[r];
for (int i = i_start; i < i_end; i++) {
function(buf, i);
}
}
});
/* Emit all temporary output buffers into the destination buffer. */
for (auto &buf : buffers) {
fh.append_from(buf);
}
}
void OBJWriter::write_vertex_coords(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data) const
{
const int tot_vertices = obj_mesh_data.tot_vertices();
for (int i = 0; i < tot_vertices; i++) {
const int tot_count = obj_mesh_data.tot_vertices();
obj_parallel_chunked_output(fh, tot_count, [&](FormatHandler<eFileType::OBJ> &buf, int i) {
float3 vertex = obj_mesh_data.calc_vertex_coords(i, export_params_.scaling_factor);
fh.write<eOBJSyntaxElement::vertex_coords>(vertex[0], vertex[1], vertex[2]);
}
buf.write<eOBJSyntaxElement::vertex_coords>(vertex[0], vertex[1], vertex[2]);
});
}
void OBJWriter::write_uv_coords(FormatHandler<eFileType::OBJ> &fh, OBJMesh &r_obj_mesh_data) const
{
for (const float2 &uv_vertex : r_obj_mesh_data.get_uv_coords()) {
fh.write<eOBJSyntaxElement::uv_vertex_coords>(uv_vertex[0], uv_vertex[1]);
}
const Vector<float2> &uv_coords = r_obj_mesh_data.get_uv_coords();
const int tot_count = uv_coords.size();
obj_parallel_chunked_output(fh, tot_count, [&](FormatHandler<eFileType::OBJ> &buf, int i) {
const float2 &uv_vertex = uv_coords[i];
buf.write<eOBJSyntaxElement::uv_vertex_coords>(uv_vertex[0], uv_vertex[1]);
});
}
void OBJWriter::write_poly_normals(FormatHandler<eFileType::OBJ> &fh, OBJMesh &obj_mesh_data)
{
/* Poly normals should be calculated earlier via store_normal_coords_and_indices. */
for (const float3 &normal : obj_mesh_data.get_normal_coords()) {
fh.write<eOBJSyntaxElement::normal>(normal[0], normal[1], normal[2]);
}
}
int OBJWriter::write_smooth_group(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data,
const int poly_index,
const int last_poly_smooth_group) const
{
int current_group = SMOOTH_GROUP_DISABLED;
if (!export_params_.export_smooth_groups && obj_mesh_data.is_ith_poly_smooth(poly_index)) {
/* Smooth group calculation is disabled, but polygon is smooth-shaded. */
current_group = SMOOTH_GROUP_DEFAULT;
}
else if (obj_mesh_data.is_ith_poly_smooth(poly_index)) {
/* Smooth group calc is enabled and polygon is smoothshaded, so find the group. */
current_group = obj_mesh_data.ith_smooth_group(poly_index);
}
if (current_group == last_poly_smooth_group) {
/* Group has already been written, even if it is "s 0". */
return current_group;
}
fh.write<eOBJSyntaxElement::smooth_group>(current_group);
return current_group;
}
int16_t OBJWriter::write_poly_material(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data,
const int poly_index,
const int16_t last_poly_mat_nr,
std::function<const char *(int)> matname_fn) const
{
if (!export_params_.export_materials || obj_mesh_data.tot_materials() <= 0) {
return last_poly_mat_nr;
}
const int16_t current_mat_nr = obj_mesh_data.ith_poly_matnr(poly_index);
/* Whenever a polygon with a new material is encountered, write its material
* and/or group, otherwise pass. */
if (last_poly_mat_nr == current_mat_nr) {
return current_mat_nr;
}
if (current_mat_nr == NOT_FOUND) {
fh.write<eOBJSyntaxElement::poly_usemtl>(MATERIAL_GROUP_DISABLED);
return current_mat_nr;
}
if (export_params_.export_object_groups) {
write_object_group(fh, obj_mesh_data);
}
const char *mat_name = matname_fn(current_mat_nr);
if (!mat_name) {
mat_name = MATERIAL_GROUP_DISABLED;
}
fh.write<eOBJSyntaxElement::poly_usemtl>(mat_name);
return current_mat_nr;
}
int16_t OBJWriter::write_vertex_group(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data,
const int poly_index,
const int16_t last_poly_vertex_group) const
{
if (!export_params_.export_vertex_groups) {
return last_poly_vertex_group;
}
const int16_t current_group = obj_mesh_data.get_poly_deform_group_index(poly_index);
if (current_group == last_poly_vertex_group) {
/* No vertex group found in this polygon, just like in the last iteration. */
return current_group;
}
if (current_group == NOT_FOUND) {
fh.write<eOBJSyntaxElement::object_group>(DEFORM_GROUP_DISABLED);
}
else {
fh.write<eOBJSyntaxElement::object_group>(
obj_mesh_data.get_poly_deform_group_name(current_group));
}
return current_group;
const Vector<float3> &normal_coords = obj_mesh_data.get_normal_coords();
const int tot_count = normal_coords.size();
obj_parallel_chunked_output(fh, tot_count, [&](FormatHandler<eFileType::OBJ> &buf, int i) {
const float3 &normal = normal_coords[i];
buf.write<eOBJSyntaxElement::normal>(normal[0], normal[1], normal[2]);
});
}
OBJWriter::func_vert_uv_normal_indices OBJWriter::get_poly_element_writer(
@ -290,30 +268,78 @@ OBJWriter::func_vert_uv_normal_indices OBJWriter::get_poly_element_writer(
return &OBJWriter::write_vert_indices;
}
static int get_smooth_group(const OBJMesh &mesh, const OBJExportParams &params, int poly_idx)
{
if (poly_idx < 0) {
return NEGATIVE_INIT;
}
int group = SMOOTH_GROUP_DISABLED;
if (mesh.is_ith_poly_smooth(poly_idx)) {
group = !params.export_smooth_groups ? SMOOTH_GROUP_DEFAULT : mesh.ith_smooth_group(poly_idx);
}
return group;
}
void OBJWriter::write_poly_elements(FormatHandler<eFileType::OBJ> &fh,
const IndexOffsets &offsets,
const OBJMesh &obj_mesh_data,
std::function<const char *(int)> matname_fn)
{
int last_poly_smooth_group = NEGATIVE_INIT;
int16_t last_poly_vertex_group = NEGATIVE_INIT;
int16_t last_poly_mat_nr = NEGATIVE_INIT;
const func_vert_uv_normal_indices poly_element_writer = get_poly_element_writer(
obj_mesh_data.tot_uv_vertices());
const int tot_polygons = obj_mesh_data.tot_polygons();
for (int i = 0; i < tot_polygons; i++) {
obj_parallel_chunked_output(fh, tot_polygons, [&](FormatHandler<eFileType::OBJ> &buf, int i) {
Vector<int> poly_vertex_indices = obj_mesh_data.calc_poly_vertex_indices(i);
Span<int> poly_uv_indices = obj_mesh_data.calc_poly_uv_indices(i);
Vector<int> poly_normal_indices = obj_mesh_data.calc_poly_normal_indices(i);
last_poly_smooth_group = write_smooth_group(fh, obj_mesh_data, i, last_poly_smooth_group);
last_poly_vertex_group = write_vertex_group(fh, obj_mesh_data, i, last_poly_vertex_group);
last_poly_mat_nr = write_poly_material(fh, obj_mesh_data, i, last_poly_mat_nr, matname_fn);
/* Write smoothing group if different from previous. */
{
const int prev_group = get_smooth_group(obj_mesh_data, export_params_, i - 1);
const int group = get_smooth_group(obj_mesh_data, export_params_, i);
if (group != prev_group) {
buf.write<eOBJSyntaxElement::smooth_group>(group);
}
}
/* Write vertex group if different from previous. */
if (export_params_.export_vertex_groups) {
const int16_t prev_group = i == 0 ? NEGATIVE_INIT :
obj_mesh_data.get_poly_deform_group_index(i - 1);
const int16_t group = obj_mesh_data.get_poly_deform_group_index(i);
if (group != prev_group) {
buf.write<eOBJSyntaxElement::object_group>(
group == NOT_FOUND ? DEFORM_GROUP_DISABLED :
obj_mesh_data.get_poly_deform_group_name(group));
}
}
/* Write material name and material group if different from previous. */
if (export_params_.export_materials && obj_mesh_data.tot_materials() > 0) {
const int16_t prev_mat = i == 0 ? NEGATIVE_INIT : obj_mesh_data.ith_poly_matnr(i - 1);
const int16_t mat = obj_mesh_data.ith_poly_matnr(i);
if (mat != prev_mat) {
if (mat == NOT_FOUND) {
buf.write<eOBJSyntaxElement::poly_usemtl>(MATERIAL_GROUP_DISABLED);
}
else {
if (export_params_.export_object_groups) {
write_object_group(buf, obj_mesh_data);
}
const char *mat_name = matname_fn(mat);
if (!mat_name) {
mat_name = MATERIAL_GROUP_DISABLED;
}
buf.write<eOBJSyntaxElement::poly_usemtl>(mat_name);
}
}
}
/* Write polygon elements. */
(this->*poly_element_writer)(
fh, offsets, poly_vertex_indices, poly_uv_indices, poly_normal_indices);
}
buf, offsets, poly_vertex_indices, poly_uv_indices, poly_normal_indices);
});
}
void OBJWriter::write_edges_indices(FormatHandler<eFileType::OBJ> &fh,

View File

@ -102,30 +102,6 @@ class OBJWriter : NonMovable, NonCopyable {
* \note Normal indices ares stored here, but written with polygons later.
*/
void write_poly_normals(FormatHandler<eFileType::OBJ> &fh, OBJMesh &obj_mesh_data);
/**
* Write smooth group if polygon at the given index is shaded smooth else "s 0"
*/
int write_smooth_group(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data,
int poly_index,
int last_poly_smooth_group) const;
/**
* Write material name and material group of a polygon in the .OBJ file.
* \return #mat_nr of the polygon at the given index.
* \note It doesn't write to the material library.
*/
int16_t write_poly_material(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data,
int poly_index,
int16_t last_poly_mat_nr,
std::function<const char *(int)> matname_fn) const;
/**
* Write the name of the deform group of a polygon.
*/
int16_t write_vertex_group(FormatHandler<eFileType::OBJ> &fh,
const OBJMesh &obj_mesh_data,
int poly_index,
int16_t last_poly_vertex_group) const;
/**
* Write polygon elements with at least vertex indices, and conditionally with UV vertex
* indices and polygon normal indices. Also write groups: smooth, vertex, material.

View File

@ -313,6 +313,14 @@ class FormatHandler : NonCopyable, NonMovable {
return blocks_.size();
}
void append_from(FormatHandler<filetype, buffer_chunk_size, write_local_buffer_size> &v)
{
blocks_.insert(blocks_.end(),
std::make_move_iterator(v.blocks_.begin()),
std::make_move_iterator(v.blocks_.end()));
v.blocks_.clear();
}
/**
* Example invocation: `writer->write<eMTLSyntaxElement::newmtl>("foo")`.
*