#include "tr_local.h"
#include "tr_model.h"
#include "ref_import.h"
#include "tr_shader.h"

#define	LL(x) x=LittleLong(x)


/*
=============================================================

UNCOMPRESSING BONES

=============================================================
*/

#define MC_BITS_X (16)
#define MC_BITS_Y (16)
#define MC_BITS_Z (16)
#define MC_BITS_VECT (16)

#define MC_SCALE_X (1.0f/64)
#define MC_SCALE_Y (1.0f/64)
#define MC_SCALE_Z (1.0f/64)

#define MC_MASK_X ((1<<(MC_BITS_X))-1)
#define MC_MASK_Y ((1<<(MC_BITS_Y))-1)
#define MC_MASK_Z ((1<<(MC_BITS_Z))-1)
#define MC_MASK_VECT ((1<<(MC_BITS_VECT))-1)

#define MC_SCALE_VECT (1.0f/(float)((1<<(MC_BITS_VECT-1))-2))

#define MC_POS_X (0)
#define MC_SHIFT_X (0)

#define MC_POS_Y ((((MC_BITS_X))/8))
#define MC_SHIFT_Y ((((MC_BITS_X)%8)))

#define MC_POS_Z ((((MC_BITS_X+MC_BITS_Y))/8))
#define MC_SHIFT_Z ((((MC_BITS_X+MC_BITS_Y)%8)))

#define MC_POS_V11 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z))/8))
#define MC_SHIFT_V11 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z)%8)))

#define MC_POS_V12 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT))/8))
#define MC_SHIFT_V12 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT)%8)))

#define MC_POS_V13 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*2))/8))
#define MC_SHIFT_V13 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*2)%8)))

#define MC_POS_V21 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*3))/8))
#define MC_SHIFT_V21 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*3)%8)))

#define MC_POS_V22 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*4))/8))
#define MC_SHIFT_V22 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*4)%8)))

#define MC_POS_V23 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*5))/8))
#define MC_SHIFT_V23 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*5)%8)))

#define MC_POS_V31 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*6))/8))
#define MC_SHIFT_V31 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*6)%8)))

#define MC_POS_V32 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*7))/8))
#define MC_SHIFT_V32 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*7)%8)))

#define MC_POS_V33 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*8))/8))
#define MC_SHIFT_V33 ((((MC_BITS_X+MC_BITS_Y+MC_BITS_Z+MC_BITS_VECT*8)%8)))

void MC_UnCompress(float mat[3][4], const unsigned char* comp)
{
	int val;

	val=(int)((unsigned short *)(comp))[0];
	val-=1<<(MC_BITS_X-1);
	mat[0][3]=((float)(val))*MC_SCALE_X;

	val=(int)((unsigned short *)(comp))[1];
	val-=1<<(MC_BITS_Y-1);
	mat[1][3]=((float)(val))*MC_SCALE_Y;

	val=(int)((unsigned short *)(comp))[2];
	val-=1<<(MC_BITS_Z-1);
	mat[2][3]=((float)(val))*MC_SCALE_Z;

	val=(int)((unsigned short *)(comp))[3];
	val-=1<<(MC_BITS_VECT-1);
	mat[0][0]=((float)(val))*MC_SCALE_VECT;

	val=(int)((unsigned short *)(comp))[4];
	val-=1<<(MC_BITS_VECT-1);
	mat[0][1]=((float)(val))*MC_SCALE_VECT;

	val=(int)((unsigned short *)(comp))[5];
	val-=1<<(MC_BITS_VECT-1);
	mat[0][2]=((float)(val))*MC_SCALE_VECT;


	val=(int)((unsigned short *)(comp))[6];
	val-=1<<(MC_BITS_VECT-1);
	mat[1][0]=((float)(val))*MC_SCALE_VECT;

	val=(int)((unsigned short *)(comp))[7];
	val-=1<<(MC_BITS_VECT-1);
	mat[1][1]=((float)(val))*MC_SCALE_VECT;

	val=(int)((unsigned short *)(comp))[8];
	val-=1<<(MC_BITS_VECT-1);
	mat[1][2]=((float)(val))*MC_SCALE_VECT;


	val=(int)((unsigned short *)(comp))[9];
	val-=1<<(MC_BITS_VECT-1);
	mat[2][0]=((float)(val))*MC_SCALE_VECT;

	val=(int)((unsigned short *)(comp))[10];
	val-=1<<(MC_BITS_VECT-1);
	mat[2][1]=((float)(val))*MC_SCALE_VECT;

	val=(int)((unsigned short *)(comp))[11];
	val-=1<<(MC_BITS_VECT-1);
	mat[2][2]=((float)(val))*MC_SCALE_VECT;
}



qboolean R_LoadMDR( model_t *mod, void *buffer, int filesize, const char *mod_name ) 
{
	int					i, j, k, l;
	mdrHeader_t			*pinmodel, *mdr;
	mdrFrame_t			*frame;
	mdrLOD_t			*lod, *curlod;
	mdrSurface_t			*surf, *cursurf;
	mdrTriangle_t			*tri, *curtri;
	mdrVertex_t			*v, *curv;
	mdrWeight_t			*weight, *curweight;
	mdrTag_t			*tag, *curtag;
	int					size;
	shader_t			*sh;

	pinmodel = (mdrHeader_t *)buffer;

	pinmodel->version = LittleLong(pinmodel->version);
	if (pinmodel->version != MDR_VERSION) 
	{
		ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has wrong version (%i should be %i)\n", mod_name, pinmodel->version, MDR_VERSION);
		return qfalse;
	}

	size = LittleLong(pinmodel->ofsEnd);
	
	if(size > filesize)
	{
		ri.Printf(PRINT_WARNING, "R_LoadMDR: Header of %s is broken. Wrong filesize declared!\n", mod_name);
		return qfalse;
	}
	
	mod->type = MOD_MDR;

	LL(pinmodel->numFrames);
	LL(pinmodel->numBones);
	LL(pinmodel->ofsFrames);

	// This is a model that uses some type of compressed Bones. We don't want to uncompress every bone for each rendered frame
	// over and over again, we'll uncompress it in this function already, so we must adjust the size of the target mdr.
	if(pinmodel->ofsFrames < 0)
	{
		// mdrFrame_t is larger than mdrCompFrame_t:
		size += pinmodel->numFrames * sizeof(frame->name);
		// now add enough space for the uncompressed bones.
		size += pinmodel->numFrames * pinmodel->numBones * ((sizeof(mdrBone_t) - sizeof(mdrCompBone_t)));
	}
	
	// simple bounds check
	if(pinmodel->numBones < 0 ||
		sizeof(*mdr) + pinmodel->numFrames * (sizeof(*frame) + (pinmodel->numBones - 1) * sizeof(*frame->bones)) > size)
	{
		ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has broken structure.\n", mod_name);
		return qfalse;
	}

	mod->dataSize += size;
	mod->modelData = mdr = ri.Hunk_Alloc( size, h_low );

	// Copy all the values over from the file and fix endian issues in the process, if necessary.
	
	mdr->ident = LittleLong(pinmodel->ident);
	mdr->version = pinmodel->version;	// Don't need to swap byte order on this one, we already did above.
	Q_strncpyz(mdr->name, pinmodel->name, sizeof(mdr->name));
	mdr->numFrames = pinmodel->numFrames;
	mdr->numBones = pinmodel->numBones;
	mdr->numLODs = LittleLong(pinmodel->numLODs);
	mdr->numTags = LittleLong(pinmodel->numTags);
	// We don't care about the other offset values, we'll generate them ourselves while loading.

	mod->numLods = mdr->numLODs;

	if ( mdr->numFrames < 1 ) 
	{
		ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has no frames\n", mod_name);
		return qfalse;
	}

	/* The first frame will be put into the first free space after the header */
	frame = (mdrFrame_t *)(mdr + 1);
	mdr->ofsFrames = (int)((byte *) frame - (byte *) mdr);
		
	if (pinmodel->ofsFrames < 0)
	{
		mdrCompFrame_t *cframe;
				
		// compressed model...				
		cframe = (mdrCompFrame_t *)((byte *) pinmodel - pinmodel->ofsFrames);
		
		for(i = 0; i < mdr->numFrames; i++)
		{
			for(j = 0; j < 3; j++)
			{
				frame->bounds[0][j] = LittleFloat(cframe->bounds[0][j]);
				frame->bounds[1][j] = LittleFloat(cframe->bounds[1][j]);
				frame->localOrigin[j] = LittleFloat(cframe->localOrigin[j]);
			}

			frame->radius = LittleFloat(cframe->radius);
			frame->name[0] = '\0';	// No name supplied in the compressed version.
			
			for(j = 0; j < mdr->numBones; j++)
			{
#if defined( Q3_BIG_ENDIAN )				
				for(k = 0; k < (sizeof(cframe->bones[j].Comp) / 2); k++)
				{
					// Do swapping for the uncompressing functions. They seem to use shorts
					// values only, so I assume this will work. Never tested it on other
					// platforms, though.
					
					((unsigned short *)(cframe->bones[j].Comp))[k] =
						LittleShort( ((unsigned short *)(cframe->bones[j].Comp))[k] );
				}
#endif				
				/* Now do the actual uncompressing */
				MC_UnCompress(frame->bones[j].matrix, cframe->bones[j].Comp);
			}
			
			// Next Frame...
			// cframe = (mdrCompFrame_t *) &cframe->bones[j];
			// frame = (mdrFrame_t *) &frame->bones[j];
            // this suppress GCC strict-aliasing warning 
            {
			    // Next Frame...
                mdrCompBone_t *p = &(cframe->bones[j]);
			    cframe = (mdrCompFrame_t *) p;
            }

            {
                mdrBone_t *p = &frame->bones[j];
                frame = (mdrFrame_t *) p;
            }
		}
	}
	else
	{
		mdrFrame_t *curframe;
		
		// uncompressed model...
		//
    
		curframe = (mdrFrame_t *)((byte *) pinmodel + pinmodel->ofsFrames);
		
		// swap all the frames
		for ( i = 0 ; i < mdr->numFrames ; i++) 
		{
			for(j = 0; j < 3; j++)
			{
				frame->bounds[0][j] = LittleFloat(curframe->bounds[0][j]);
				frame->bounds[1][j] = LittleFloat(curframe->bounds[1][j]);
				frame->localOrigin[j] = LittleFloat(curframe->localOrigin[j]);
			}
			
			frame->radius = LittleFloat(curframe->radius);
			Q_strncpyz(frame->name, curframe->name, sizeof(frame->name));
			
            // suppress GCC strict-aliasing warning 
#if defined( Q3_BIG_ENDIAN )
            for (j = 0; j < (int) (mdr->numBones * sizeof(mdrBone_t) / 4); j++) 
			{
			    ((float *)frame->bones)[j]=FloatSwap(&((float *)curframe->bones)[j]);
            }
#else
            for (j = 0; j < mdr->numBones; j++) 
			{
                frame->bones[j] = curframe->bones[j];
            }
#endif		
			
			//curframe = (mdrFrame_t *) &curframe->bones[mdr->numBones];
			//frame = (mdrFrame_t *) &frame->bones[mdr->numBones];
            // suppress GCC strict-aliasing warning 
            {
			    mdrBone_t* p = &curframe->bones[mdr->numBones];
                curframe = (mdrFrame_t *) p;
            }

            {
                mdrBone_t* p = &frame->bones[mdr->numBones];
                frame = (mdrFrame_t *) p;
            }

		}
	}
	
	// frame should now point to the first free address after all frames.
	lod = (mdrLOD_t *) frame;
	mdr->ofsLODs = (int) ((byte *) lod - (byte *)mdr);
	
	curlod = (mdrLOD_t *)((byte *) pinmodel + LittleLong(pinmodel->ofsLODs));
		
	// swap all the LOD's
	for ( l = 0 ; l < mdr->numLODs ; l++)
	{
		// simple bounds check
		if((byte *) (lod + 1) > (byte *) mdr + size)
		{
			ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has broken structure.\n", mod_name);
			return qfalse;
		}

		lod->numSurfaces = LittleLong(curlod->numSurfaces);
		
		// swap all the surfaces
		surf = (mdrSurface_t *) (lod + 1);
		lod->ofsSurfaces = (int)((byte *) surf - (byte *) lod);
		cursurf = (mdrSurface_t *) ((byte *)curlod + LittleLong(curlod->ofsSurfaces));
		
		for ( i = 0 ; i < lod->numSurfaces ; i++)
		{
			// simple bounds check
			if((byte *) (surf + 1) > (byte *) mdr + size)
			{
				ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has broken structure.\n", mod_name);
				return qfalse;
			}

			// first do some copying stuff
			
			surf->ident = SF_MDR;
			Q_strncpyz(surf->name, cursurf->name, sizeof(surf->name));
			Q_strncpyz(surf->shader, cursurf->shader, sizeof(surf->shader));
			
			surf->ofsHeader = (byte *) mdr - (byte *) surf;
			
			surf->numVerts = LittleLong(cursurf->numVerts);
			surf->numTriangles = LittleLong(cursurf->numTriangles);
			// numBoneReferences and BoneReferences generally seem to be unused
			
			// now do the checks that may fail.
			if ( surf->numVerts >= SHADER_MAX_VERTEXES ) 
			{
				ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has more than %i verts on %s (%i).\n",
					  mod_name, SHADER_MAX_VERTEXES - 1, surf->name[0] ? surf->name : "a surface",
					  surf->numVerts );
				return qfalse;
			}
			if ( surf->numTriangles*3 >= SHADER_MAX_INDEXES ) 
			{
				ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has more than %i triangles on %s (%i).\n",
					  mod_name, ( SHADER_MAX_INDEXES / 3 ) - 1, surf->name[0] ? surf->name : "a surface",
					  surf->numTriangles );
				return qfalse;
			}
			// lowercase the surface name so skin compares are faster
			Q_strlwr( surf->name );

			// register the shaders
			sh = R_FindShader(surf->shader, LIGHTMAP_NONE, qtrue);
			if ( sh->defaultShader ) {
				surf->shaderIndex = 0;
			} else {
				surf->shaderIndex = sh->index;
			}
			
			// now copy the vertexes.
			v = (mdrVertex_t *) (surf + 1);
			surf->ofsVerts = (int)((byte *) v - (byte *) surf);
			curv = (mdrVertex_t *) ((byte *)cursurf + LittleLong(cursurf->ofsVerts));
			
			for(j = 0; j < surf->numVerts; j++)
			{
				LL(curv->numWeights);
			
				// simple bounds check
				if(curv->numWeights < 0 || (byte *) (v + 1) + (curv->numWeights - 1) * sizeof(*weight) > (byte *) mdr + size)
				{
					ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has broken structure.\n", mod_name);
					return qfalse;
				}

				v->normal[0] = LittleFloat(curv->normal[0]);
				v->normal[1] = LittleFloat(curv->normal[1]);
				v->normal[2] = LittleFloat(curv->normal[2]);
				
				v->texCoords[0] = LittleFloat(curv->texCoords[0]);
				v->texCoords[1] = LittleFloat(curv->texCoords[1]);
				
				v->numWeights = curv->numWeights;
				weight = &v->weights[0];
				curweight = &curv->weights[0];
				
				// Now copy all the weights
				for(k = 0; k < v->numWeights; k++)
				{
					weight->boneIndex = LittleLong(curweight->boneIndex);
					weight->boneWeight = LittleFloat(curweight->boneWeight);
					
					weight->offset[0] = LittleFloat(curweight->offset[0]);
					weight->offset[1] = LittleFloat(curweight->offset[1]);
					weight->offset[2] = LittleFloat(curweight->offset[2]);
					
					weight++;
					curweight++;
				}
				
				v = (mdrVertex_t *) weight;
				curv = (mdrVertex_t *) curweight;
			}
						
			// we know the offset to the triangles now:
			tri = (mdrTriangle_t *) v;
			surf->ofsTriangles = (int)((byte *) tri - (byte *) surf);
			curtri = (mdrTriangle_t *)((byte *) cursurf + LittleLong(cursurf->ofsTriangles));
			
			// simple bounds check
			if(surf->numTriangles < 0 || (byte *) (tri + surf->numTriangles) > (byte *) mdr + size)
			{
				ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has broken structure.\n", mod_name);
				return qfalse;
			}

			for(j = 0; j < surf->numTriangles; j++)
			{
				tri->indexes[0] = LittleLong(curtri->indexes[0]);
				tri->indexes[1] = LittleLong(curtri->indexes[1]);
				tri->indexes[2] = LittleLong(curtri->indexes[2]);
				
				tri++;
				curtri++;
			}
			
			// tri now points to the end of the surface.
			surf->ofsEnd = (byte *) tri - (byte *) surf;
			surf = (mdrSurface_t *) tri;

			// find the next surface.
			cursurf = (mdrSurface_t *) ((byte *) cursurf + LittleLong(cursurf->ofsEnd));
		}

		// surf points to the next lod now.
		lod->ofsEnd = (int)((byte *) surf - (byte *) lod);
		lod = (mdrLOD_t *) surf;

		// find the next LOD.
		curlod = (mdrLOD_t *)((byte *) curlod + LittleLong(curlod->ofsEnd));
	}
	
	// lod points to the first tag now, so update the offset too.
	tag = (mdrTag_t *) lod;
	mdr->ofsTags = (int)((byte *) tag - (byte *) mdr);
	curtag = (mdrTag_t *) ((byte *)pinmodel + LittleLong(pinmodel->ofsTags));

	// simple bounds check
	if(mdr->numTags < 0 || (byte *) (tag + mdr->numTags) > (byte *) mdr + size)
	{
		ri.Printf(PRINT_WARNING, "R_LoadMDR: %s has broken structure.\n", mod_name);
		return qfalse;
	}
	
	for (i = 0 ; i < mdr->numTags ; i++)
	{
		tag->boneIndex = LittleLong(curtag->boneIndex);
		Q_strncpyz(tag->name, curtag->name, sizeof(tag->name));
		
		tag++;
		curtag++;
	}
	
	// And finally we know the real offset to the end.
	mdr->ofsEnd = (int)((byte *) tag - (byte *) mdr);

	// phew! we're done.
	
	return qtrue;
}


qhandle_t R_RegisterMDR(const char *name, model_t *mod)
{
	
	char* buf;

	qboolean loaded = qfalse;
	int filesize = ri.FS_ReadFile(name, (void**)&buf);
	if(!buf)
	{
		mod->type = MOD_BAD;
		return 0;
	}
	

#if defined( Q3_BIG_ENDIAN )		
	int ident = LittleLong(*(int *)buf);
#else
	int ident = *(int *)buf;
#endif	

	
	if(ident == MDR_IDENT)
		loaded = R_LoadMDR(mod, buf, filesize, name);

	ri.FS_FreeFile(buf);
	
	if(!loaded)
	{
		ri.Printf(PRINT_WARNING,"R_RegisterMDR: couldn't load mdr file %s\n", name);
		mod->type = MOD_BAD;
		return 0;
	}
	
	return mod->index;
}