1
Fork 0
mirror of https://github.com/redstrate/Novus.git synced 2025-04-26 05:37:46 +00:00
novus/parts/mdl/mdlimport.cpp

248 lines
11 KiB
C++
Raw Normal View History

// SPDX-FileCopyrightText: 2023 Joshua Goins <josh@redstrate.com>
// SPDX-License-Identifier: GPL-3.0-or-later
#include "mdlimport.h"
#include <QDebug>
#include <glm/glm.hpp>
#include "tiny_gltf.h"
void importModel(physis_MDL &existingModel, const QString &filename)
{
tinygltf::Model model;
std::string error, warning;
tinygltf::TinyGLTF loader;
if (!loader.LoadBinaryFromFile(&model, &error, &warning, filename.toStdString())) {
qInfo() << "Error when loading glTF model:" << error;
return;
}
if (!warning.empty()) {
qInfo() << "Warnings when loading glTF model:" << warning;
}
struct ProcessedSubMesh {
uint32_t subMeshIndex = 0;
std::vector<Vertex> vertices;
std::vector<uint32_t> indices;
};
bool duplicateBuffers = false; // I hate this.
// We may be reading the parts of order (0.1, then 1.0, maybe 0.2 and so on) so we have to keep track of our buffers
struct ProcessedPart {
uint32_t partIndex = 0;
int lastPositionViewUsed = -1; // detect duplicate accessor and check their offsets
std::vector<ProcessedSubMesh> subMeshes;
};
std::vector<ProcessedPart> processingParts;
for (const auto &node : model.nodes) {
// Detect if it's a mesh node
if (node.mesh >= 0) {
qInfo() << "Importing" << node.name;
const QStringList parts = QString::fromStdString(node.name).split(QLatin1Char(' '));
const QStringList lodPartNumber = parts[2].split(QLatin1Char('.'));
const int lodNumber = 0;
const int partNumber = lodPartNumber[0].toInt();
const int submeshNumber = lodPartNumber[1].toInt();
qInfo() << "- Part:" << partNumber;
qInfo() << "- Submesh:" << submeshNumber;
if (partNumber >= existingModel.lods[lodNumber].num_parts) {
qInfo() << "- Skipping because of missing part...";
continue;
}
if (submeshNumber >= existingModel.lods[lodNumber].parts[partNumber].num_submeshes) {
qInfo() << "- Skipping because of missing submesh...";
continue;
}
ProcessedPart *processedPart = nullptr;
for (auto &part : processingParts) {
if (part.partIndex == partNumber) {
processedPart = &part;
break;
}
}
if (processedPart == nullptr) {
processedPart = &processingParts.emplace_back();
processedPart->partIndex = partNumber;
}
ProcessedSubMesh &processedSubMesh = processedPart->subMeshes.emplace_back();
processedSubMesh.subMeshIndex = submeshNumber;
auto &mesh = model.meshes[node.mesh];
auto &primitive = mesh.primitives[0];
const auto getAccessor = [&model, &primitive](const std::string &name, const size_t index) -> unsigned char const * {
const auto &positionAccessor = model.accessors[primitive.attributes[name]];
const auto &positionView = model.bufferViews[positionAccessor.bufferView];
const auto &positionBuffer = model.buffers[positionView.buffer];
return (positionBuffer.data.data() + (positionAccessor.ByteStride(positionView) * index) + positionView.byteOffset
+ positionAccessor.byteOffset);
};
// All the accessors are mapped to the same buffer vertex view
const auto &positionAccessor = model.accessors[primitive.attributes["POSITION"]];
const auto &colorAccessor = model.accessors[primitive.attributes["COLOR_0"]];
const auto &indexAccessor = model.accessors[primitive.indices];
const auto &indexView = model.bufferViews[indexAccessor.bufferView];
const auto &indexBuffer = model.buffers[indexView.buffer];
qInfo() << "- Importing mesh of" << positionAccessor.count << "vertices and" << indexAccessor.count << "indices.";
auto indexData = reinterpret_cast<const uint16_t *>(indexBuffer.data.data() + indexView.byteOffset + indexAccessor.byteOffset);
for (size_t k = 0; k < indexAccessor.count; k++) {
processedSubMesh.indices.push_back(indexData[k]);
}
std::vector<Vertex> newVertices;
for (size_t i = 0; i < positionAccessor.count; i++) {
// vertex data
glm::vec3 const *positionData = reinterpret_cast<glm::vec3 const *>(getAccessor("POSITION", i));
glm::vec3 const *normalData = reinterpret_cast<glm::vec3 const *>(getAccessor("NORMAL", i));
glm::vec2 const *uv0Data = reinterpret_cast<glm::vec2 const *>(getAccessor("TEXCOORD_0", i));
glm::vec2 const *uv1Data = reinterpret_cast<glm::vec2 const *>(getAccessor("TEXCOORD_1", i));
glm::vec4 const *weightsData = reinterpret_cast<glm::vec4 const *>(getAccessor("WEIGHTS_0", i));
uint8_t const *jointsData = reinterpret_cast<uint8_t const *>(getAccessor("JOINTS_0", i));
glm::vec4 const *tangent1Data = reinterpret_cast<glm::vec4 const *>(getAccessor("TANGENT", i));
// Replace position data
2023-12-09 21:18:34 -05:00
Vertex vertex{};
vertex.position[0] = positionData->x;
vertex.position[1] = positionData->y;
vertex.position[2] = positionData->z;
vertex.normal[0] = normalData->x;
vertex.normal[1] = normalData->y;
vertex.normal[2] = normalData->z;
vertex.uv0[0] = uv0Data->x;
vertex.uv0[1] = uv0Data->y;
vertex.uv1[0] = uv1Data->x;
vertex.uv1[1] = uv1Data->y;
vertex.bone_weight[0] = weightsData->x;
vertex.bone_weight[1] = weightsData->y;
vertex.bone_weight[2] = weightsData->z;
vertex.bone_weight[3] = weightsData->w;
// calculate binormal, because glTF won't give us those!!
const glm::vec3 normal = glm::vec3(vertex.normal[0], vertex.normal[1], vertex.normal[2]);
const glm::vec4 tangent = *tangent1Data;
const glm::vec3 bitangent = glm::cross(normal, glm::vec3(tangent));
const float handedness = tangent.w;
// In a cruel twist of fate, Tangent1 is actually the **BINORMAL** and not the tangent data. Square Enix is AMAZING.
vertex.bitangent[0] = bitangent.x * handedness;
vertex.bitangent[1] = bitangent.y * handedness;
vertex.bitangent[2] = bitangent.z * handedness;
vertex.bitangent[3] = handedness;
if (colorAccessor.componentType == TINYGLTF_COMPONENT_TYPE_UNSIGNED_SHORT) {
unsigned short const *colorData = reinterpret_cast<unsigned short const *>(getAccessor("COLOR_0", i));
vertex.color[0] = static_cast<float>(*colorData) / std::numeric_limits<unsigned short>::max();
vertex.color[1] = static_cast<float>(*(colorData + 1)) / std::numeric_limits<unsigned short>::max();
vertex.color[2] = static_cast<float>(*(colorData + 2)) / std::numeric_limits<unsigned short>::max();
vertex.color[3] = static_cast<float>(*(colorData + 3)) / std::numeric_limits<unsigned short>::max();
} else {
glm::vec4 const *colorData = reinterpret_cast<glm::vec4 const *>(getAccessor("COLOR_0", i));
vertex.color[0] = colorData->x;
vertex.color[1] = colorData->y;
vertex.color[2] = colorData->z;
vertex.color[3] = colorData->w;
}
// We need to ensure the bones are mapped correctly
// When exporting from modeling software, it's possible it sorted the nodes (Blender does this)
for (int i = 0; i < 4; i++) {
int originalBoneId = *(jointsData + i);
auto joints = model.skins[0].joints;
int realBoneId = 0;
for (int j = 0; j < existingModel.num_affected_bones; j++) {
if (strcmp(existingModel.affected_bone_names[j], model.nodes[joints[originalBoneId]].name.c_str()) == 0) {
realBoneId = j;
break;
}
}
vertex.bone_id[i] = realBoneId;
}
newVertices.push_back(vertex);
}
// don't add duplicate vertex data!!
if (processedPart->lastPositionViewUsed != positionAccessor.bufferView) {
processedPart->lastPositionViewUsed = positionAccessor.bufferView;
processedSubMesh.vertices = newVertices;
} else {
duplicateBuffers = true;
}
}
}
size_t index_offset = 0;
for (auto &part : processingParts) {
std::vector<Vertex> combinedVertices;
std::vector<uint16_t> combinedIndices;
std::vector<SubMesh> newSubmeshes;
size_t vertex_offset = 0;
// Turn 0.3, 0.2, 0.1 into 0.1, 0.2, 0.3 so they're all in the combined vertex list correctly
std::sort(part.subMeshes.begin(), part.subMeshes.end(), [](const ProcessedSubMesh &a, const ProcessedSubMesh &b) {
return a.subMeshIndex < b.subMeshIndex;
});
for (auto &submesh : part.subMeshes) {
std::copy(submesh.vertices.cbegin(), submesh.vertices.cend(), std::back_inserter(combinedVertices));
for (unsigned int indice : submesh.indices) {
// if the buffers are duplicate and shared (like when exporting from Novus)
// then we don't need to add vertex offset, they are already done
if (duplicateBuffers) {
combinedIndices.push_back(indice);
} else {
combinedIndices.push_back(indice + vertex_offset);
}
}
newSubmeshes.push_back({.index_count = static_cast<uint32_t>(submesh.indices.size()), .index_offset = static_cast<uint32_t>(index_offset)});
index_offset += submesh.indices.size();
vertex_offset += submesh.vertices.size();
}
physis_mdl_replace_vertices(&existingModel,
0,
part.partIndex,
combinedVertices.size(),
combinedVertices.data(),
combinedIndices.size(),
combinedIndices.data(),
newSubmeshes.size(),
newSubmeshes.data());
}
qInfo() << "Successfully imported model!";
}