add mesh loader
parent
ad298cd84f
commit
f38dd8e6e2
|
@ -115,6 +115,12 @@ aiMatrix4x4 Read<aiMatrix4x4>(IOStream * stream)
|
|||
return m;
|
||||
}
|
||||
|
||||
template <typename T> void ReadBounds( IOStream * stream, T* p, unsigned int n )
|
||||
{
|
||||
// not sure what to do here, the data isn't really useful.
|
||||
stream->Seek( sizeof(T) * n, aiOrigin_CUR );
|
||||
}
|
||||
|
||||
void AssbinImporter::ReadBinaryNode( IOStream * stream, aiNode** node )
|
||||
{
|
||||
ai_assert( Read<uint32_t>(stream) == ASSBIN_CHUNK_AINODE);
|
||||
|
@ -145,6 +151,153 @@ void AssbinImporter::ReadBinaryNode( IOStream * stream, aiNode** node )
|
|||
|
||||
}
|
||||
|
||||
// -----------------------------------------------------------------------------------
|
||||
void AssbinImporter::ReadBinaryBone( IOStream * stream, aiBone* b )
|
||||
{
|
||||
ai_assert( Read<uint32_t>(stream) == ASSBIN_CHUNK_AIBONE );
|
||||
uint32_t size = Read<uint32_t>(stream);
|
||||
|
||||
b->mName = Read<aiString>(stream);
|
||||
b->mNumWeights = Read<unsigned int>(stream);
|
||||
b->mOffsetMatrix = Read<aiMatrix4x4>(stream);
|
||||
|
||||
// for the moment we write dumb min/max values for the bones, too.
|
||||
// maybe I'll add a better, hash-like solution later
|
||||
if (shortened)
|
||||
{
|
||||
ReadBounds(stream,b->mWeights,b->mNumWeights);
|
||||
} // else write as usual
|
||||
else
|
||||
{
|
||||
b->mWeights = new aiVertexWeight[b->mNumWeights];
|
||||
stream->Read(b->mWeights,1,b->mNumWeights*sizeof(aiVertexWeight));
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void AssbinImporter::ReadBinaryMesh( IOStream * stream, aiMesh* mesh )
|
||||
{
|
||||
ai_assert( Read<uint32_t>(stream) == ASSBIN_CHUNK_AIMESH);
|
||||
uint32_t size = Read<uint32_t>(stream);
|
||||
|
||||
mesh->mPrimitiveTypes = Read<unsigned int>(stream);
|
||||
mesh->mNumVertices = Read<unsigned int>(stream);
|
||||
mesh->mNumFaces = Read<unsigned int>(stream);
|
||||
mesh->mNumBones = Read<unsigned int>(stream);
|
||||
mesh->mMaterialIndex = Read<unsigned int>(stream);
|
||||
|
||||
// first of all, write bits for all existent vertex components
|
||||
unsigned int c = Read<unsigned int>(stream);
|
||||
|
||||
if (c & ASSBIN_MESH_HAS_POSITIONS)
|
||||
{
|
||||
if (shortened) {
|
||||
ReadBounds(stream,mesh->mVertices,mesh->mNumVertices);
|
||||
} // else write as usual
|
||||
else
|
||||
{
|
||||
mesh->mVertices = new aiVector3D[mesh->mNumVertices];
|
||||
stream->Read(mesh->mVertices,1,12*mesh->mNumVertices);
|
||||
}
|
||||
}
|
||||
if (c & ASSBIN_MESH_HAS_NORMALS)
|
||||
{
|
||||
if (shortened) {
|
||||
ReadBounds(stream,mesh->mNormals,mesh->mNumVertices);
|
||||
} // else write as usual
|
||||
else
|
||||
{
|
||||
mesh->mNormals = new aiVector3D[mesh->mNumVertices];
|
||||
stream->Read(mesh->mNormals,1,12*mesh->mNumVertices);
|
||||
}
|
||||
}
|
||||
if (c & ASSBIN_MESH_HAS_TANGENTS_AND_BITANGENTS)
|
||||
{
|
||||
if (shortened) {
|
||||
ReadBounds(stream,mesh->mTangents,mesh->mNumVertices);
|
||||
ReadBounds(stream,mesh->mBitangents,mesh->mNumVertices);
|
||||
} // else write as usual
|
||||
else {
|
||||
mesh->mTangents = new aiVector3D[mesh->mNumVertices];
|
||||
stream->Read(mesh->mTangents,1,12*mesh->mNumVertices);
|
||||
mesh->mBitangents = new aiVector3D[mesh->mNumVertices];
|
||||
stream->Read(mesh->mBitangents,1,12*mesh->mNumVertices);
|
||||
}
|
||||
}
|
||||
for (unsigned int n = 0; n < AI_MAX_NUMBER_OF_COLOR_SETS;++n)
|
||||
{
|
||||
if (!(c & ASSBIN_MESH_HAS_COLOR(n)))
|
||||
break;
|
||||
|
||||
if (shortened)
|
||||
{
|
||||
ReadBounds(stream,mesh->mColors[n],mesh->mNumVertices);
|
||||
} // else write as usual
|
||||
else
|
||||
{
|
||||
mesh->mColors[n] = new aiColor4D[mesh->mNumVertices];
|
||||
stream->Read(mesh->mColors[n],16*mesh->mNumVertices,1);
|
||||
}
|
||||
}
|
||||
for (unsigned int n = 0; n < AI_MAX_NUMBER_OF_TEXTURECOORDS;++n)
|
||||
{
|
||||
if (!(c & ASSBIN_MESH_HAS_TEXCOORD(n)))
|
||||
break;
|
||||
|
||||
// write number of UV components
|
||||
mesh->mNumUVComponents[n] = Read<unsigned int>(stream);
|
||||
|
||||
if (shortened) {
|
||||
ReadBounds(stream,mesh->mTextureCoords[n],mesh->mNumVertices);
|
||||
} // else write as usual
|
||||
else
|
||||
{
|
||||
mesh->mTextureCoords[n] = new aiVector3D[mesh->mNumVertices];
|
||||
stream->Read(mesh->mTextureCoords[n],12*mesh->mNumVertices,1);
|
||||
}
|
||||
}
|
||||
|
||||
// write faces. There are no floating-point calculations involved
|
||||
// in these, so we can write a simple hash over the face data
|
||||
// to the dump file. We generate a single 32 Bit hash for 512 faces
|
||||
// using Assimp's standard hashing function.
|
||||
if (shortened) {
|
||||
Read<unsigned int>(stream);
|
||||
}
|
||||
else // else write as usual
|
||||
{
|
||||
// if there are less than 2^16 vertices, we can simply use 16 bit integers ...
|
||||
mesh->mFaces = new aiFace[mesh->mNumFaces];
|
||||
for (unsigned int i = 0; i < mesh->mNumFaces;++i) {
|
||||
aiFace& f = mesh->mFaces[i];
|
||||
|
||||
BOOST_STATIC_ASSERT(AI_MAX_FACE_INDICES <= 0xffff);
|
||||
f.mNumIndices = Read<uint16_t>(stream);
|
||||
f.mIndices = new unsigned int[f.mNumIndices];
|
||||
|
||||
for (unsigned int a = 0; a < f.mNumIndices;++a) {
|
||||
if (mesh->mNumVertices < (1u<<16))
|
||||
{
|
||||
f.mIndices[a] = Read<uint16_t>(stream);
|
||||
}
|
||||
else
|
||||
{
|
||||
f.mIndices[a] = Read<unsigned int>(stream);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// write bones
|
||||
if (mesh->mNumBones) {
|
||||
mesh->mBones = new C_STRUCT aiBone*[mesh->mNumBones];
|
||||
for (unsigned int a = 0; a < mesh->mNumBones;++a) {
|
||||
mesh->mBones[a] = new aiBone();
|
||||
ReadBinaryBone(stream,mesh->mBones[a]);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void AssbinImporter::ReadBinaryScene( IOStream * stream, aiScene* scene )
|
||||
{
|
||||
ai_assert( Read<uint32_t>(stream) == ASSBIN_CHUNK_AISCENE);
|
||||
|
@ -162,13 +315,17 @@ void AssbinImporter::ReadBinaryScene( IOStream * stream, aiScene* scene )
|
|||
scene->mRootNode = new aiNode[1];
|
||||
ReadBinaryNode( stream, &scene->mRootNode );
|
||||
|
||||
/*
|
||||
// Read all meshes
|
||||
for (unsigned int i = 0; i < scene->mNumMeshes;++i) {
|
||||
const aiMesh* mesh = scene->mMeshes[i];
|
||||
ReadBinaryMesh( stream,mesh);
|
||||
}
|
||||
if (scene->mNumMeshes)
|
||||
{
|
||||
scene->mMeshes = new aiMesh*[scene->mNumMeshes];
|
||||
for (unsigned int i = 0; i < scene->mNumMeshes;++i) {
|
||||
scene->mMeshes[i] = new aiMesh();
|
||||
ReadBinaryMesh( stream,scene->mMeshes[i]);
|
||||
}
|
||||
}
|
||||
|
||||
/*
|
||||
// Read materials
|
||||
for (unsigned int i = 0; i< scene->mNumMaterials; ++i) {
|
||||
const aiMaterial* mat = scene->mMaterials[i];
|
||||
|
|
|
@ -76,6 +76,8 @@ public:
|
|||
);
|
||||
void ReadBinaryScene( IOStream * stream, aiScene* pScene );
|
||||
void ReadBinaryNode( IOStream * stream, aiNode** mRootNode );
|
||||
void ReadBinaryMesh( IOStream * stream, aiMesh* mesh );
|
||||
void ReadBinaryBone( IOStream * stream, aiBone* bone );
|
||||
};
|
||||
|
||||
} // end of namespace Assimp
|
||||
|
|
Loading…
Reference in New Issue