jedi-outcast/code/ghoul2/G2_misc.cpp

1301 lines
36 KiB
C++

// leave this as first line for PCH reasons...
//
#include "../server/exe_headers.h"
#ifndef __Q_SHARED_H
#include "../game/q_shared.h"
#endif
#if !defined(TR_LOCAL_H)
#include "../renderer/tr_local.h"
#endif
#include "../renderer/matcomp.h"
#if !defined(G2_H_INC)
#include "G2.h"
#endif
#if !defined (MINIHEAP_H_INC)
#include "../qcommon/miniheap.h"
#endif
#define G2_MODEL_OK(g) ((g)&&(g)->mValid&&(g)->aHeader&&(g)->currentModel&&(g)->animModel)
#include "../server/server.h"
extern mdxaBone_t worldMatrix;
extern mdxaBone_t worldMatrixInv;
const mdxaBone_t &EvalBoneCache(int index,CBoneCache *boneCache);
#pragma warning(disable : 4512) //assignment op could not be genereated
class CTraceSurface
{
public:
int surfaceNum;
surfaceInfo_v &rootSList;
const model_t *currentModel;
const int lod;
vec3_t rayStart;
vec3_t rayEnd;
CCollisionRecord *collRecMap;
const int entNum;
const int modelIndex;
const skin_t *skin;
const shader_t *cust_shader;
size_t *TransformedVertsArray;
const EG2_Collision eG2TraceType;
bool hitOne;
float m_fRadius;
CTraceSurface(
int initsurfaceNum,
surfaceInfo_v &initrootSList,
const model_t *initcurrentModel,
int initlod,
vec3_t initrayStart,
vec3_t initrayEnd,
CCollisionRecord *initcollRecMap,
int initentNum,
int initmodelIndex,
const skin_t *initskin,
const shader_t *initcust_shader,
size_t *initTransformedVertsArray,
const EG2_Collision einitG2TraceType,
float fRadius
):
surfaceNum(initsurfaceNum),
rootSList(initrootSList),
currentModel(initcurrentModel),
lod(initlod),
collRecMap(initcollRecMap),
entNum(initentNum),
modelIndex(initmodelIndex),
skin(initskin),
cust_shader(initcust_shader),
eG2TraceType(einitG2TraceType),
hitOne(false),
TransformedVertsArray(initTransformedVertsArray),
m_fRadius(fRadius)
{
VectorCopy(initrayStart, rayStart);
VectorCopy(initrayEnd, rayEnd);
}
};
// assorted Ghoul 2 functions.
// list all surfaces associated with a model
void G2_List_Model_Surfaces(const char *fileName)
{
int i, x;
model_t *mod_m = R_GetModelByHandle(RE_RegisterModel(fileName));
mdxmSurfHierarchy_t *surf;
surf = (mdxmSurfHierarchy_t *) ( (byte *)mod_m->mdxm + mod_m->mdxm->ofsSurfHierarchy );
mdxmSurface_t *surface = (mdxmSurface_t *)((byte *)mod_m->mdxm + mod_m->mdxm->ofsLODs + sizeof(mdxmLOD_t));
for ( x = 0 ; x < mod_m->mdxm->numSurfaces ; x++)
{
Com_Printf("Surface %i Name %s\n", x, surf->name);
if (r_verbose->value)
{
Com_Printf("Num Descendants %i\n", surf->numChildren);
for (i=0; i<surf->numChildren; i++)
{
Com_Printf("Descendant %i\n", surf->childIndexes[i]);
}
}
// find the next surface
surf = (mdxmSurfHierarchy_t *)( (byte *)surf + (size_t)( &((mdxmSurfHierarchy_t *)0)->childIndexes[ surf->numChildren ] ));
surface =(mdxmSurface_t *)( (byte *)surface + surface->ofsEnd );
}
}
// list all bones associated with a model
void G2_List_Model_Bones(const char *fileName, int frame)
{
int x, i;
mdxaSkel_t *skel;
mdxaSkelOffsets_t *offsets;
model_t *mod_m = R_GetModelByHandle(RE_RegisterModel(fileName));
model_t *mod_a = R_GetModelByHandle(mod_m->mdxm->animIndex);
// mdxaFrame_t *aframe=0;
// int frameSize;
mdxaHeader_t *header = mod_a->mdxa;
// figure out where the offset list is
offsets = (mdxaSkelOffsets_t *)((byte *)header + sizeof(mdxaHeader_t));
// frameSize = (size_t)( &((mdxaFrame_t *)0)->boneIndexes[ header->numBones ] );
// aframe = (mdxaFrame_t *)((byte *)header + header->ofsFrames + (frame * frameSize));
// walk each bone and list it's name
for (x=0; x< mod_a->mdxa->numBones; x++)
{
skel = (mdxaSkel_t *)((byte *)header + sizeof(mdxaHeader_t) + offsets->offsets[x]);
Com_Printf("Bone %i Name %s\n", x, skel->name);
Com_Printf("X pos %f, Y pos %f, Z pos %f\n", skel->BasePoseMat.matrix[0][3], skel->BasePoseMat.matrix[1][3], skel->BasePoseMat.matrix[2][3]);
// if we are in verbose mode give us more details
if (r_verbose->value)
{
Com_Printf("Num Descendants %i\n", skel->numChildren);
for (i=0; i<skel->numChildren; i++)
{
Com_Printf("Num Descendants %i\n", skel->numChildren);
}
}
}
}
/************************************************************************************************
* G2_GetAnimFileName
* obtain the .gla filename for a model
*
* Input
* filename of model
*
* Output
* true if we successfully obtained a filename, false otherwise
*
************************************************************************************************/
qboolean G2_GetAnimFileName(const char *fileName, char **filename)
{
// find the model we want
model_t *mod = R_GetModelByHandle(RE_RegisterModel(fileName));
if (mod && mod->mdxm && (mod->mdxm->animName[0] != 0))
{
*filename = mod->mdxm->animName;
return qtrue;
}
return qfalse;
}
/////////////////////////////////////////////////////////////////////
//
// Code for collision detection for models gameside
//
/////////////////////////////////////////////////////////////////////
int G2_DecideTraceLod(CGhoul2Info &ghoul2, int useLod)
{
int returnLod = useLod;
// if we are overriding the LOD at top level, then we can afford to only check this level of model
if (ghoul2.mLodBias > returnLod)
{
returnLod = ghoul2.mLodBias;
}
assert(G2_MODEL_OK(&ghoul2));
assert(ghoul2.currentModel);
assert(ghoul2.currentModel->mdxm);
//what about r_lodBias?
// now ensure that we haven't selected a lod that doesn't exist for this model
if ( returnLod >= ghoul2.currentModel->mdxm->numLODs )
{
returnLod = ghoul2.currentModel->mdxm->numLODs - 1;
}
return returnLod;
}
void R_TransformEachSurface( const mdxmSurface_t *surface, vec3_t scale, CMiniHeap *G2VertSpace, size_t *TransformedVertsArray,CBoneCache *boneCache)
{
int j, k;
mdxmVertex_t *v;
float *TransformedVerts;
//
// deform the vertexes by the lerped bones
//
int *piBoneReferences = (int*) ((byte*)surface + surface->ofsBoneReferences);
// alloc some space for the transformed verts to get put in
TransformedVerts = (float *)G2VertSpace->MiniHeapAlloc(surface->numVerts * 5 * 4);
TransformedVertsArray[surface->thisSurfaceIndex] = (size_t)TransformedVerts;
if (!TransformedVerts)
{
Com_Error(ERR_DROP, "Ran out of transform space for Ghoul2 Models. Adjust MiniHeapSize in SV_SpawnServer.\n");
}
// whip through and actually transform each vertex
const int numVerts = surface->numVerts;
v = (mdxmVertex_t *) ((byte *)surface + surface->ofsVerts);
mdxmVertexTexCoord_t *pTexCoords = (mdxmVertexTexCoord_t *) &v[numVerts];
// optimisation issue
if ((scale[0] != 1.0) || (scale[1] != 1.0) || (scale[2] != 1.0))
{
for ( j = 0; j < numVerts; j++ )
{
vec3_t tempVert, tempNormal;
// mdxmWeight_t *w;
VectorClear( tempVert );
VectorClear( tempNormal );
// w = v->weights;
const int iNumWeights = G2_GetVertWeights( v );
float fTotalWeight = 0.0f;
for ( k = 0 ; k < iNumWeights ; k++ )
{
int iBoneIndex = G2_GetVertBoneIndex( v, k );
float fBoneWeight = G2_GetVertBoneWeight( v, k, fTotalWeight, iNumWeights );
const mdxaBone_t &bone=EvalBoneCache(piBoneReferences[iBoneIndex],boneCache);
tempVert[0] += fBoneWeight * ( DotProduct( bone.matrix[0], v->vertCoords ) + bone.matrix[0][3] );
tempVert[1] += fBoneWeight * ( DotProduct( bone.matrix[1], v->vertCoords ) + bone.matrix[1][3] );
tempVert[2] += fBoneWeight * ( DotProduct( bone.matrix[2], v->vertCoords ) + bone.matrix[2][3] );
tempNormal[0] += fBoneWeight * DotProduct( bone.matrix[0], v->normal );
tempNormal[1] += fBoneWeight * DotProduct( bone.matrix[1], v->normal );
tempNormal[2] += fBoneWeight * DotProduct( bone.matrix[2], v->normal );
}
int pos = j * 5;
// copy tranformed verts into temp space
TransformedVerts[pos++] = tempVert[0] * scale[0];
TransformedVerts[pos++] = tempVert[1] * scale[1];
TransformedVerts[pos++] = tempVert[2] * scale[2];
// we will need the S & T coors too for hitlocation and hitmaterial stuff
TransformedVerts[pos++] = pTexCoords[j].texCoords[0];
TransformedVerts[pos] = pTexCoords[j].texCoords[1];
v++;// = (mdxmVertex_t *)&v->weights[/*v->numWeights*/surface->maxVertBoneWeights];
}
}
else
{
int pos = 0;
for ( j = 0; j < numVerts; j++ )
{
vec3_t tempVert, tempNormal;
// const mdxmWeight_t *w;
VectorClear( tempVert );
VectorClear( tempNormal );
// w = v->weights;
const int iNumWeights = G2_GetVertWeights( v );
float fTotalWeight = 0.0f;
for ( k = 0 ; k < iNumWeights ; k++ )
{
int iBoneIndex = G2_GetVertBoneIndex( v, k );
float fBoneWeight = G2_GetVertBoneWeight( v, k, fTotalWeight, iNumWeights );
const mdxaBone_t &bone=EvalBoneCache(piBoneReferences[iBoneIndex],boneCache);
tempVert[0] += fBoneWeight * ( DotProduct( bone.matrix[0], v->vertCoords ) + bone.matrix[0][3] );
tempVert[1] += fBoneWeight * ( DotProduct( bone.matrix[1], v->vertCoords ) + bone.matrix[1][3] );
tempVert[2] += fBoneWeight * ( DotProduct( bone.matrix[2], v->vertCoords ) + bone.matrix[2][3] );
tempNormal[0] += fBoneWeight * DotProduct( bone.matrix[0], v->normal );
tempNormal[1] += fBoneWeight * DotProduct( bone.matrix[1], v->normal );
tempNormal[2] += fBoneWeight * DotProduct( bone.matrix[2], v->normal );
}
// copy tranformed verts into temp space
TransformedVerts[pos++] = tempVert[0];
TransformedVerts[pos++] = tempVert[1];
TransformedVerts[pos++] = tempVert[2];
// we will need the S & T coors too for hitlocation and hitmaterial stuff
TransformedVerts[pos++] = pTexCoords[j].texCoords[0];
TransformedVerts[pos++] = pTexCoords[j].texCoords[1];
v++;// = (mdxmVertex_t *)&v->weights[/*v->numWeights*/surface->maxVertBoneWeights];
}
}
}
void G2_TransformSurfaces(int surfaceNum, surfaceInfo_v &rootSList,
CBoneCache *boneCache, const model_t *currentModel, int lod, vec3_t scale, CMiniHeap *G2VertSpace, size_t *TransformedVertArray, bool secondTimeAround)
{
int i;
assert(currentModel);
assert(currentModel->mdxm);
// back track and get the surfinfo struct for this surface
const mdxmSurface_t *surface = (mdxmSurface_t *)G2_FindSurface(currentModel, surfaceNum, lod);
const mdxmHierarchyOffsets_t *surfIndexes = (mdxmHierarchyOffsets_t *)((byte *)currentModel->mdxm + sizeof(mdxmHeader_t));
const mdxmSurfHierarchy_t *surfInfo = (mdxmSurfHierarchy_t *)((byte *)surfIndexes + surfIndexes->offsets[surface->thisSurfaceIndex]);
// see if we have an override surface in the surface list
const surfaceInfo_t *surfOverride = G2_FindOverrideSurface(surfaceNum, rootSList);
// really, we should use the default flags for this surface unless it's been overriden
int offFlags = surfInfo->flags;
if (surfOverride)
{
offFlags = surfOverride->offFlags;
}
// if this surface is not off, add it to the shader render list
if (!offFlags)
{
R_TransformEachSurface(surface, scale, G2VertSpace, TransformedVertArray, boneCache);
}
// if we are turning off all descendants, then stop this recursion now
if (offFlags & G2SURFACEFLAG_NODESCENDANTS)
{
return;
}
// now recursively call for the children
for (i=0; i< surfInfo->numChildren; i++)
{
G2_TransformSurfaces(surfInfo->childIndexes[i], rootSList, boneCache, currentModel, lod, scale, G2VertSpace, TransformedVertArray, secondTimeAround);
}
}
// main calling point for the model transform for collision detection. At this point all of the skeleton has been transformed.
void G2_TransformModel(CGhoul2Info_v &ghoul2, const int frameNum, vec3_t scale, CMiniHeap *G2VertSpace, int useLod)
{
int i, lod;
vec3_t correctScale;
VectorCopy(scale, correctScale);
// check for scales of 0 - that's the default I believe
if (!scale[0])
{
correctScale[0] = 1.0;
}
if (!scale[1])
{
correctScale[1] = 1.0;
}
if (!scale[2])
{
correctScale[2] = 1.0;
}
// walk each possible model for this entity and try rendering it out
for (i=0; i<ghoul2.size(); i++)
{
CGhoul2Info &g=ghoul2[i];
// don't bother with models that we don't care about.
if (!g.mValid)
{
continue;
}
assert(g.mBoneCache);
assert(G2_MODEL_OK(&g));
// stop us building this model more than once per frame
g.mMeshFrameNum = frameNum;
// decide the LOD
lod = G2_DecideTraceLod(g, useLod);
// give us space for the transformed vertex array to be put in
ghoul2[i].mTransformedVertsArray = (size_t*)G2VertSpace->MiniHeapAlloc(g.currentModel->mdxm->numSurfaces * 4);
if (!g.mTransformedVertsArray)
{
Com_Error(ERR_DROP, "Ran out of transform space for Ghoul2 Models. Adjust MiniHeapSize in SV_SpawnServer.\n");
}
memset(g.mTransformedVertsArray, 0,(g.currentModel->mdxm->numSurfaces * 4));
G2_FindOverrideSurface(-1,g.mSlist); //reset the quick surface override lookup;
// recursively call the model surface transform
G2_TransformSurfaces(g.mSurfaceRoot, g.mSlist, g.mBoneCache, g.currentModel, lod, correctScale, G2VertSpace, g.mTransformedVertsArray, false);
}
}
// work out how much space a triangle takes
static float G2_AreaOfTri(const vec3_t A, const vec3_t B, const vec3_t C)
{
vec3_t cross, ab, cb;
VectorSubtract(A, B, ab);
VectorSubtract(C, B, cb);
CrossProduct(ab, cb, cross);
return VectorLength(cross);
}
// actually determine the S and T of the coordinate we hit in a given poly
static void G2_BuildHitPointST( const vec3_t A, const float SA, const float TA,
const vec3_t B, const float SB, const float TB,
const vec3_t C, const float SC, const float TC,
const vec3_t P, float *s, float *t,float &bary_i,float &bary_j)
{
float areaABC = G2_AreaOfTri(A, B, C);
float i = G2_AreaOfTri(P, B, C) / areaABC;
bary_i=i;
float j = G2_AreaOfTri(A, P, C) / areaABC;
bary_j=j;
float k = G2_AreaOfTri(A, B, P) / areaABC;
*s = SA * i + SB * j + SC * k;
*t = TA * i + TB * j + TC * k;
*s=fmod(*s, 1);
if (*s< 0)
{
*s+= 1.0;
}
*t=fmod(*t, 1);
if (*t< 0)
{
*t+= 1.0;
}
}
// routine that works out given a ray whether or not it hits a poly
qboolean G2_SegmentTriangleTest( const vec3_t start, const vec3_t end,
const vec3_t A, const vec3_t B, const vec3_t C,
qboolean backFaces,qboolean frontFaces,vec3_t returnedPoint,vec3_t returnedNormal, float *denom)
{
static const float tiny=1E-10f;
vec3_t returnedNormalT;
vec3_t edgeAC;
VectorSubtract(C, A, edgeAC);
VectorSubtract(B, A, returnedNormalT);
CrossProduct(returnedNormalT, edgeAC, returnedNormal);
vec3_t ray;
VectorSubtract(end, start, ray);
*denom=DotProduct(ray, returnedNormal);
if (fabs(*denom)<tiny|| // triangle parallel to ray
(!backFaces && *denom>0)|| // not accepting back faces
(!frontFaces && *denom<0)) //not accepting front faces
{
return qfalse;
}
vec3_t toPlane;
VectorSubtract(A, start, toPlane);
float t=DotProduct(toPlane, returnedNormal)/ *denom;
if (t<0.0f||t>1.0f)
{
return qfalse; // off segment
}
VectorScale(ray, t, ray);
VectorAdd(ray, start, returnedPoint);
vec3_t edgePA;
VectorSubtract(A, returnedPoint, edgePA);
vec3_t edgePB;
VectorSubtract(B, returnedPoint, edgePB);
vec3_t edgePC;
VectorSubtract(C, returnedPoint, edgePC);
vec3_t temp;
CrossProduct(edgePA, edgePB, temp);
if (DotProduct(temp, returnedNormal)<0.0f)
{
return qfalse; // off triangle
}
CrossProduct(edgePC, edgePA, temp);
if (DotProduct(temp,returnedNormal)<0.0f)
{
return qfalse; // off triangle
}
CrossProduct(edgePB, edgePC, temp);
if (DotProduct(temp, returnedNormal)<0.0f)
{
return qfalse; // off triangle
}
return qtrue;
}
// now we're at poly level, check each model space transformed poly against the model world transfomed ray
static bool G2_TracePolys(const mdxmSurface_t *surface, const mdxmSurfHierarchy_t *surfInfo, CTraceSurface &TS)
{
int j, numTris;
// whip through and actually transform each vertex
const mdxmTriangle_t *tris = (mdxmTriangle_t *) ((byte *)surface + surface->ofsTriangles);
const float *verts = (float *)TS.TransformedVertsArray[surface->thisSurfaceIndex];
numTris = surface->numTriangles;
for ( j = 0; j < numTris; j++ )
{
float face;
vec3_t hitPoint, normal;
// determine actual coords for this triangle
const float *point1 = &verts[(tris[j].indexes[0] * 5)];
const float *point2 = &verts[(tris[j].indexes[1] * 5)];
const float *point3 = &verts[(tris[j].indexes[2] * 5)];
// did we hit it?
if (G2_SegmentTriangleTest(TS.rayStart, TS.rayEnd, point1, point2, point3, qtrue, qtrue, hitPoint, normal, &face))
{ // find space in the collision records for this record
int i;
for (i=0; i<MAX_G2_COLLISIONS;i++)
{
if (TS.collRecMap[i].mEntityNum == -1)
{
CCollisionRecord &newCol = TS.collRecMap[i];
vec3_t distVect;
float x_pos = 0, y_pos = 0;
newCol.mPolyIndex = j;
newCol.mEntityNum = TS.entNum;
newCol.mSurfaceIndex = surface->thisSurfaceIndex;
newCol.mModelIndex = TS.modelIndex;
if (face>0)
{
newCol.mFlags = G2_FRONTFACE;
}
else
{
newCol.mFlags = G2_BACKFACE;
}
VectorSubtract(hitPoint, TS.rayStart, distVect);
newCol.mDistance = VectorLength(distVect);
// put the hit point back into world space
TransformAndTranslatePoint(hitPoint, newCol.mCollisionPosition, &worldMatrix);
// transform normal (but don't translate) into world angles
TransformPoint(normal, newCol.mCollisionNormal, &worldMatrix);
VectorNormalize(newCol.mCollisionNormal);
newCol.mMaterial = newCol.mLocation = 0;
// Determine our location within the texture, and barycentric coordinates
G2_BuildHitPointST(point1, point1[3], point1[4],
point2, point2[3], point2[4],
point3, point3[3], point3[4],
hitPoint, &x_pos, &y_pos,newCol.mBarycentricI,newCol.mBarycentricJ);
/*
const shader_t *shader = 0;
// now, we know what surface this hit belongs to, we need to go get the shader handle so we can get the correct hit location and hit material info
if ( cust_shader )
{
shader = cust_shader;
}
else if ( skin )
{
int j;
// match the surface name to something in the skin file
shader = tr.defaultShader;
for ( j = 0 ; j < skin->numSurfaces ; j++ )
{
// the names have both been lowercased
if ( !strcmp( skin->surfaces[j]->name, surfInfo->name ) )
{
shader = skin->surfaces[j]->shader;
break;
}
}
}
else
{
shader = R_GetShaderByHandle( surfInfo->shaderIndex );
}
// do we even care to decide what the hit or location area's are? If we don't have them in the shader there is little point
if ((shader->hitLocation) || (shader->hitMaterial))
{
// ok, we have a floating point position. - determine location in data we need to look at
if (shader->hitLocation)
{
newCol.mLocation = *(hitMatReg[shader->hitLocation].loc +
((int)(y_pos * hitMatReg[shader->hitLocation].height) * hitMatReg[shader->hitLocation].width) +
((int)(x_pos * hitMatReg[shader->hitLocation].width)));
Com_Printf("G2_TracePolys hit location: %d\n", newCol.mLocation);
}
if (shader->hitMaterial)
{
newCol.mMaterial = *(hitMatReg[shader->hitMaterial].loc +
((int)(y_pos * hitMatReg[shader->hitMaterial].height) * hitMatReg[shader->hitMaterial].width) +
((int)(x_pos * hitMatReg[shader->hitMaterial].width)));
}
}
*/
// exit now if we should
if (TS.eG2TraceType == G2_RETURNONHIT)
{
TS.hitOne = true;
return true;
}
break;
}
}
if (i==MAX_G2_COLLISIONS)
{
assert(i!=MAX_G2_COLLISIONS); // run out of collision record space - will probalbly never happen
TS.hitOne = true; //force stop recursion
return true; // return true to avoid wasting further time, but no hit will result without a record
}
}
}
return false;
}
struct SVertexTemp
{
int flags;
// int touch;
// int newindex;
// float tex[2];
SVertexTemp()
{
// touch=0;
}
};
#define MAX_GORE_VERTS (3000)
static SVertexTemp GoreVerts[MAX_GORE_VERTS];
// now we're at poly level, check each model space transformed poly against the model world transfomed ray
static bool G2_RadiusTracePolys(
const mdxmSurface_t *surface,
CTraceSurface &TS
)
{
int j;
vec3_t basis1;
vec3_t basis2;
vec3_t taxis;
vec3_t saxis;
basis2[0]=0.0f;
basis2[1]=0.0f;
basis2[2]=1.0f;
vec3_t v3RayDir;
VectorSubtract(TS.rayEnd, TS.rayStart, v3RayDir);
CrossProduct(v3RayDir,basis2,basis1);
if (DotProduct(basis1,basis1)<.1f)
{
basis2[0]=0.0f;
basis2[1]=1.0f;
basis2[2]=0.0f;
CrossProduct(v3RayDir,basis2,basis1);
}
CrossProduct(v3RayDir,basis1,basis2);
// Give me a shot direction not a bunch of zeros :) -Gil
// assert(DotProduct(basis1,basis1)>.0001f);
// assert(DotProduct(basis2,basis2)>.0001f);
VectorNormalize(basis1);
VectorNormalize(basis2);
const float c=cos(0);//theta
const float s=sin(0);//theta
VectorScale(basis1, 0.5f * c / TS.m_fRadius,taxis);
VectorMA(taxis, 0.5f * s / TS.m_fRadius,basis2,taxis);
VectorScale(basis1,-0.5f * s /TS.m_fRadius,saxis);
VectorMA( saxis, 0.5f * c /TS.m_fRadius,basis2,saxis);
const float * const verts = (float *)TS.TransformedVertsArray[surface->thisSurfaceIndex];
const int numVerts = surface->numVerts;
int flags=63;
//rayDir/=lengthSquared(raydir);
const float f = VectorLengthSquared(v3RayDir);
v3RayDir[0]/=f;
v3RayDir[1]/=f;
v3RayDir[2]/=f;
for ( j = 0; j < numVerts; j++ )
{
const int pos=j*5;
vec3_t delta;
delta[0]=verts[pos+0]-TS.rayStart[0];
delta[1]=verts[pos+1]-TS.rayStart[1];
delta[2]=verts[pos+2]-TS.rayStart[2];
const float s=DotProduct(delta,saxis)+0.5f;
const float t=DotProduct(delta,taxis)+0.5f;
const float u=DotProduct(delta,v3RayDir);
int vflags=0;
if (s>0)
{
vflags|=1;
}
if (s<1)
{
vflags|=2;
}
if (t>0)
{
vflags|=4;
}
if (t<1)
{
vflags|=8;
}
if (u>0)
{
vflags|=16;
}
if (u<1)
{
vflags|=32;
}
vflags=(~vflags);
flags&=vflags;
GoreVerts[j].flags=vflags;
}
if (flags)
{
return false; // completely off the gore splotch (so presumably hit nothing? -Ste)
}
const int numTris = surface->numTriangles;
const mdxmTriangle_t * const tris = (mdxmTriangle_t *) ((byte *)surface + surface->ofsTriangles);
for ( j = 0; j < numTris; j++ )
{
assert(tris[j].indexes[0]>=0&&tris[j].indexes[0]<numVerts);
assert(tris[j].indexes[1]>=0&&tris[j].indexes[1]<numVerts);
assert(tris[j].indexes[2]>=0&&tris[j].indexes[2]<numVerts);
flags=63&
GoreVerts[tris[j].indexes[0]].flags&
GoreVerts[tris[j].indexes[1]].flags&
GoreVerts[tris[j].indexes[2]].flags;
if (flags)
{
continue;
}
else
{
// we hit a triangle, so init a collision record...
//
int i;
for (i=0; i<MAX_G2_COLLISIONS;i++)
{
if (TS.collRecMap[i].mEntityNum == -1)
{
CCollisionRecord &newCol = TS.collRecMap[i];
newCol.mPolyIndex = j;
newCol.mEntityNum = TS.entNum;
newCol.mSurfaceIndex = surface->thisSurfaceIndex;
newCol.mModelIndex = TS.modelIndex;
// if (face>0)
// {
newCol.mFlags = G2_FRONTFACE;
// }
// else
// {
// newCol.mFlags = G2_BACKFACE;
// }
//get normal from triangle
const float *A = &verts[(tris[j].indexes[0] * 5)];
const float *B = &verts[(tris[j].indexes[1] * 5)];
const float *C = &verts[(tris[j].indexes[2] * 5)];
vec3_t normal;
vec3_t edgeAC, edgeBA;
VectorSubtract(C, A, edgeAC);
VectorSubtract(B, A, edgeBA);
CrossProduct(edgeBA, edgeAC, normal);
// transform normal (but don't translate) into world angles
TransformPoint(normal, newCol.mCollisionNormal, &worldMatrix);
VectorNormalize(newCol.mCollisionNormal);
newCol.mMaterial = newCol.mLocation = 0;
// exit now if we should
if (TS.eG2TraceType == G2_RETURNONHIT)
{
TS.hitOne = true;
return true;
}
//i don't know the hitPoint, but let's just assume it's the first vert for now...
const float *hitPoint = A;
vec3_t distVect;
VectorSubtract(hitPoint, TS.rayStart, distVect);
newCol.mDistance = VectorLength(distVect);
// put the hit point back into world space
TransformAndTranslatePoint(hitPoint, newCol.mCollisionPosition, &worldMatrix);
newCol.mBarycentricI = newCol.mBarycentricJ = 0.0f;
break;
}
}
if (i==MAX_G2_COLLISIONS)
{
//assert(i!=MAX_G2_COLLISIONS); // run out of collision record space - happens OFTEN
TS.hitOne = true; //force stop recursion
return true; // return true to avoid wasting further time, but no hit will result without a record
}
}
}
return false;
}
// look at a surface and then do the trace on each poly
static void G2_TraceSurfaces(CTraceSurface &TS)
{
int i;
// back track and get the surfinfo struct for this surface
assert(TS.currentModel);
assert(TS.currentModel->mdxm);
const mdxmSurface_t *surface = (mdxmSurface_t *)G2_FindSurface(TS.currentModel, TS.surfaceNum, TS.lod);
const mdxmHierarchyOffsets_t *surfIndexes = (mdxmHierarchyOffsets_t *)((byte *)TS.currentModel->mdxm + sizeof(mdxmHeader_t));
const mdxmSurfHierarchy_t *surfInfo = (mdxmSurfHierarchy_t *)((byte *)surfIndexes + surfIndexes->offsets[surface->thisSurfaceIndex]);
// see if we have an override surface in the surface list
const surfaceInfo_t *surfOverride = G2_FindOverrideSurface(TS.surfaceNum, TS.rootSList);
// don't allow recursion if we've already hit a polygon
if (TS.hitOne)
{
return;
}
// really, we should use the default flags for this surface unless it's been overriden
int offFlags = surfInfo->flags;
// set the off flags if we have some
if (surfOverride)
{
offFlags = surfOverride->offFlags;
}
// if this surface is not off, try to hit it
if (!offFlags)
{
if (!(fabs(TS.m_fRadius) < 0.1)) // if not a point-trace
{
// .. then use radius check
//
if (G2_RadiusTracePolys(surface, // const mdxmSurface_t *surface,
TS
)
&& (TS.eG2TraceType == G2_RETURNONHIT)
)
{
TS.hitOne = true;
return;
}
}
else
{
// go away and trace the polys in this surface
if (G2_TracePolys(surface, surfInfo, TS)
&& (TS.eG2TraceType == G2_RETURNONHIT)
)
{
// ok, we hit one, *and* we want to return instantly because the returnOnHit is set
// so indicate we've hit one, so other surfaces don't get hit and return
TS.hitOne = true;
return;
}
}
}
// if we are turning off all descendants, then stop this recursion now
if (offFlags & G2SURFACEFLAG_NODESCENDANTS)
{
return;
}
// now recursively call for the children
for (i=0; i< surfInfo->numChildren && !TS.hitOne; i++)
{
TS.surfaceNum = surfInfo->childIndexes[i];
G2_TraceSurfaces(TS);
}
}
void G2_TraceModels(CGhoul2Info_v &ghoul2, vec3_t rayStart, vec3_t rayEnd, CCollisionRecord *collRecMap, int entNum, EG2_Collision eG2TraceType, int useLod, float fRadius)
{
int i, lod;
skin_t *skin;
shader_t *cust_shader;
// walk each possible model for this entity and try tracing against it
for (i=0; i<ghoul2.size(); i++)
{
// don't bother with models that we don't care about.
if (!ghoul2[i].mValid)
{
continue;
}
assert(G2_MODEL_OK(&ghoul2[i]));
// do we really want to collide with this object?
if (ghoul2[i].mFlags & GHOUL2_NOCOLLIDE)
{
continue;
}
if (ghoul2[i].mCustomShader)
{
cust_shader = R_GetShaderByHandle(ghoul2[i].mCustomShader );
}
else
{
cust_shader = NULL;
}
// figure out the custom skin thing
if ( ghoul2[i].mSkin > 0 && ghoul2[i].mSkin < tr.numSkins )
{
skin = R_GetSkinByHandle( ghoul2[i].mSkin );
}
else
{
skin = NULL;
}
lod = G2_DecideTraceLod(ghoul2[i],useLod);
//reset the quick surface override lookup
G2_FindOverrideSurface(-1, ghoul2[i].mSlist);
CTraceSurface TS(ghoul2[i].mSurfaceRoot, ghoul2[i].mSlist, ghoul2[i].currentModel, lod, rayStart, rayEnd, collRecMap, entNum, i, skin, cust_shader, ghoul2[i].mTransformedVertsArray, eG2TraceType, fRadius);
// start the surface recursion loop
G2_TraceSurfaces(TS);
// if we've hit one surface on one model, don't bother doing the rest
if (TS.hitOne)
{
break;
}
}
}
void TransformPoint (const vec3_t in, vec3_t out, mdxaBone_t *mat) {
for (int i=0;i<3;i++)
{
out[i]= in[0]*mat->matrix[i][0] + in[1]*mat->matrix[i][1] + in[2]*mat->matrix[i][2];
}
}
void TransformAndTranslatePoint (const vec3_t in, vec3_t out, mdxaBone_t *mat) {
for (int i=0;i<3;i++)
{
out[i]= in[0]*mat->matrix[i][0] + in[1]*mat->matrix[i][1] + in[2]*mat->matrix[i][2] + mat->matrix[i][3];
}
}
// create a matrix using a set of angles
void Create_Matrix(const float *angle, mdxaBone_t *matrix)
{
vec3_t axis[3];
// convert angles to axis
AnglesToAxis( angle, axis );
matrix->matrix[0][0] = axis[0][0];
matrix->matrix[1][0] = axis[0][1];
matrix->matrix[2][0] = axis[0][2];
matrix->matrix[0][1] = axis[1][0];
matrix->matrix[1][1] = axis[1][1];
matrix->matrix[2][1] = axis[1][2];
matrix->matrix[0][2] = axis[2][0];
matrix->matrix[1][2] = axis[2][1];
matrix->matrix[2][2] = axis[2][2];
matrix->matrix[0][3] = 0;
matrix->matrix[1][3] = 0;
matrix->matrix[2][3] = 0;
}
// given a matrix, generate the inverse of that matrix
void Inverse_Matrix(mdxaBone_t *src, mdxaBone_t *dest)
{
int i, j;
for (i = 0; i < 3; i++)
{
for (j = 0; j < 3; j++)
{
dest->matrix[i][j]=src->matrix[j][i];
}
}
for (i = 0; i < 3; i++)
{
dest->matrix[i][3]=0;
for (j = 0; j < 3; j++)
{
dest->matrix[i][3]-=dest->matrix[i][j]*src->matrix[j][3];
}
}
}
// generate the world matrix for a given set of angles and origin - called from lots of places
void G2_GenerateWorldMatrix(const vec3_t angles, const vec3_t origin)
{
Create_Matrix(angles, &worldMatrix);
worldMatrix.matrix[0][3] = origin[0];
worldMatrix.matrix[1][3] = origin[1];
worldMatrix.matrix[2][3] = origin[2];
Inverse_Matrix(&worldMatrix, &worldMatrixInv);
}
// go away and determine what the pointer for a specific surface definition within the model definition is
void *G2_FindSurface(const model_s *mod, int index, int lod)
{
assert(mod);
assert(mod->mdxm);
// point at first lod list
byte *current = (byte*)((size_t)mod->mdxm + (size_t)mod->mdxm->ofsLODs);
int i;
//walk the lods
assert(lod>=0&&lod<mod->mdxm->numLODs);
for (i=0; i<lod; i++)
{
mdxmLOD_t *lodData = (mdxmLOD_t *)current;
current += lodData->ofsEnd;
}
// avoid the lod pointer data structure
current += sizeof(mdxmLOD_t);
mdxmLODSurfOffset_t *indexes = (mdxmLODSurfOffset_t *)current;
// we are now looking at the offset array
assert(index>=0&&index<mod->mdxm->numSurfaces);
current += indexes->offsets[index];
return (void *)current;
}
#define SURFACE_SAVE_BLOCK_SIZE sizeof(surfaceInfo_t)
#define BOLT_SAVE_BLOCK_SIZE sizeof(boltInfo_t)
#define BONE_SAVE_BLOCK_SIZE sizeof(boneInfo_t)
qboolean G2_SaveGhoul2Models(CGhoul2Info_v &ghoul2, char **buffer, int *size)
{
// is there anything to save?
if (!ghoul2.IsValid()||!ghoul2.size())
{
*buffer = (char *)Z_Malloc(4, TAG_GHOUL2, qfalse);
int *tempBuffer = (int *)*buffer;
*tempBuffer = 0;
*size = 4;
return qtrue;
}
// yeah, lets get busy
*size = 0;
// this one isn't a define since I couldn't work out how to figure it out at compile time
int ghoul2BlockSize = (size_t)&ghoul2[0].mTransformedVertsArray - (size_t)&ghoul2[0].mModelindex;
// add in count for number of ghoul2 models
*size += 4;
// start out working out the total size of the buffer we need to allocate
int i;
for (i=0; i<ghoul2.size();i++)
{
*size += ghoul2BlockSize;
// add in count for number of surfaces
*size += 4;
*size += (ghoul2[i].mSlist.size() * SURFACE_SAVE_BLOCK_SIZE);
// add in count for number of bones
*size += 4;
*size += (ghoul2[i].mBlist.size() * BONE_SAVE_BLOCK_SIZE);
// add in count for number of bolts
*size += 4;
*size += (ghoul2[i].mBltlist.size() * BOLT_SAVE_BLOCK_SIZE);
}
// ok, we should know how much space we need now
*buffer = (char*)Z_Malloc(*size, TAG_GHOUL2, qfalse);
// now lets start putting the data we care about into the buffer
char *tempBuffer = *buffer;
// save out how many ghoul2 models we have
*(int *)tempBuffer = ghoul2.size();
tempBuffer +=4;
for (i=0; i<ghoul2.size();i++)
{
// first save out the ghoul2 details themselves
// OutputDebugString(va("G2_SaveGhoul2Models(): ghoul2[%d].mModelindex = %d\n",i,ghoul2[i].mModelindex));
memcpy(tempBuffer, &ghoul2[i].mModelindex, ghoul2BlockSize);
tempBuffer += ghoul2BlockSize;
// save out how many surfaces we have
*(int*)tempBuffer = ghoul2[i].mSlist.size();
tempBuffer +=4;
// now save the all the surface list info
int x;
for (x=0; x<ghoul2[i].mSlist.size(); x++)
{
memcpy(tempBuffer, &ghoul2[i].mSlist[x], SURFACE_SAVE_BLOCK_SIZE);
tempBuffer += SURFACE_SAVE_BLOCK_SIZE;
}
// save out how many bones we have
*(int*)tempBuffer = ghoul2[i].mBlist.size();
tempBuffer +=4;
// now save the all the bone list info
for (x=0; x<ghoul2[i].mBlist.size(); x++)
{
memcpy(tempBuffer, &ghoul2[i].mBlist[x], BONE_SAVE_BLOCK_SIZE);
tempBuffer += BONE_SAVE_BLOCK_SIZE;
}
// save out how many bolts we have
*(int*)tempBuffer = ghoul2[i].mBltlist.size();
tempBuffer +=4;
// lastly save the all the bolt list info
for (x=0; x<ghoul2[i].mBltlist.size(); x++)
{
memcpy(tempBuffer, &ghoul2[i].mBltlist[x], BOLT_SAVE_BLOCK_SIZE);
tempBuffer += BOLT_SAVE_BLOCK_SIZE;
}
}
return qtrue;
}
// have to free space malloced in the save system here because the game DLL can't.
void G2_FreeSaveBuffer(char *buffer)
{
Z_Free(buffer);
}
int G2_FindConfigStringSpace(char *name, int start, int max)
{
char s[MAX_STRING_CHARS];
int i;
for ( i=1 ; i<max ; i++ )
{
SV_GetConfigstring( start + i, s, sizeof( s ) );
if ( !s[0] )
{
break;
}
if ( !Q_stricmp( s, name ) )
{
return i;
}
}
SV_SetConfigstring(start + i, name);
return i;
}
void G2_LoadGhoul2Model(CGhoul2Info_v &ghoul2, char *buffer)
{
// first thing, lets see how many ghoul2 models we have, and resize our buffers accordingly
int newSize = *(int*)buffer;
ghoul2.resize(newSize);
buffer += 4;
// did we actually resize to a value?
if (!newSize)
{
// no, ok, well, done then.
return;
}
// this one isn't a define since I couldn't work out how to figure it out at compile time
int ghoul2BlockSize = (size_t)&ghoul2[0].mTransformedVertsArray - (size_t)&ghoul2[0].mModelindex;
// now we have enough instances, lets go through each one and load up the relevant details
for (int i=0; i<ghoul2.size(); i++)
{
ghoul2[i].mSkelFrameNum = 0;
ghoul2[i].mModelindex=-1;
ghoul2[i].mFileName[0]=0;
ghoul2[i].mValid=false;
// load the ghoul2 info from the buffer
memcpy(&ghoul2[i].mModelindex, buffer, ghoul2BlockSize);
buffer +=ghoul2BlockSize;
if (ghoul2[i].mModelindex!=-1&&ghoul2[i].mFileName[0])
{
ghoul2[i].mModelindex = i;
G2_SetupModelPointers(&ghoul2[i]);
}
// give us enough surfaces to load up the data
ghoul2[i].mSlist.resize(*(int*)buffer);
buffer +=4;
// now load all the surfaces
int x;
for (x=0; x<ghoul2[i].mSlist.size(); x++)
{
memcpy(&ghoul2[i].mSlist[x], buffer, SURFACE_SAVE_BLOCK_SIZE);
buffer += SURFACE_SAVE_BLOCK_SIZE;
}
// give us enough bones to load up the data
ghoul2[i].mBlist.resize(*(int*)buffer);
buffer +=4;
// now load all the bones
for (x=0; x<ghoul2[i].mBlist.size(); x++)
{
memcpy(&ghoul2[i].mBlist[x], buffer, BONE_SAVE_BLOCK_SIZE);
buffer += BONE_SAVE_BLOCK_SIZE;
}
// give us enough bolts to load up the data
ghoul2[i].mBltlist.resize(*(int*)buffer);
buffer +=4;
// now load all the bolts
for (x=0; x<ghoul2[i].mBltlist.size(); x++)
{
memcpy(&ghoul2[i].mBltlist[x], buffer, BOLT_SAVE_BLOCK_SIZE);
buffer += BOLT_SAVE_BLOCK_SIZE;
}
}
}