mirror of
https://github.com/ioquake/ioq3.git
synced 2024-11-10 15:21:35 +00:00
1865 lines
50 KiB
C
1865 lines
50 KiB
C
/*
|
|
===========================================================================
|
|
Copyright (C) 1999-2005 Id Software, Inc.
|
|
|
|
This file is part of Quake III Arena source code.
|
|
|
|
Quake III Arena source code is free software; you can redistribute it
|
|
and/or modify it under the terms of the GNU General Public License as
|
|
published by the Free Software Foundation; either version 2 of the License,
|
|
or (at your option) any later version.
|
|
|
|
Quake III Arena source code is distributed in the hope that it will be
|
|
useful, but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
GNU General Public License for more details.
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
along with Quake III Arena source code; if not, write to the Free Software
|
|
Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
|
|
===========================================================================
|
|
*/
|
|
// tr_map.c
|
|
|
|
#include "tr_local.h"
|
|
|
|
/*
|
|
|
|
Loads and prepares a map file for scene rendering.
|
|
|
|
A single entry point:
|
|
|
|
void RE_LoadWorldMap( const char *name );
|
|
|
|
*/
|
|
|
|
static world_t s_worldData;
|
|
static byte *fileBase;
|
|
|
|
int c_subdivisions;
|
|
int c_gridVerts;
|
|
|
|
//===============================================================================
|
|
|
|
static void HSVtoRGB( float h, float s, float v, float rgb[3] )
|
|
{
|
|
int i;
|
|
float f;
|
|
float p, q, t;
|
|
|
|
h *= 5;
|
|
|
|
i = floor( h );
|
|
f = h - i;
|
|
|
|
p = v * ( 1 - s );
|
|
q = v * ( 1 - s * f );
|
|
t = v * ( 1 - s * ( 1 - f ) );
|
|
|
|
switch ( i )
|
|
{
|
|
case 0:
|
|
rgb[0] = v;
|
|
rgb[1] = t;
|
|
rgb[2] = p;
|
|
break;
|
|
case 1:
|
|
rgb[0] = q;
|
|
rgb[1] = v;
|
|
rgb[2] = p;
|
|
break;
|
|
case 2:
|
|
rgb[0] = p;
|
|
rgb[1] = v;
|
|
rgb[2] = t;
|
|
break;
|
|
case 3:
|
|
rgb[0] = p;
|
|
rgb[1] = q;
|
|
rgb[2] = v;
|
|
break;
|
|
case 4:
|
|
rgb[0] = t;
|
|
rgb[1] = p;
|
|
rgb[2] = v;
|
|
break;
|
|
case 5:
|
|
rgb[0] = v;
|
|
rgb[1] = p;
|
|
rgb[2] = q;
|
|
break;
|
|
}
|
|
}
|
|
|
|
/*
|
|
===============
|
|
R_ColorShiftLightingBytes
|
|
|
|
===============
|
|
*/
|
|
static void R_ColorShiftLightingBytes( byte in[4], byte out[4] ) {
|
|
int shift, r, g, b;
|
|
|
|
// shift the color data based on overbright range
|
|
shift = r_mapOverBrightBits->integer - tr.overbrightBits;
|
|
|
|
// shift the data based on overbright range
|
|
r = in[0] << shift;
|
|
g = in[1] << shift;
|
|
b = in[2] << shift;
|
|
|
|
// normalize by color instead of saturating to white
|
|
if ( ( r | g | b ) > 255 ) {
|
|
int max;
|
|
|
|
max = r > g ? r : g;
|
|
max = max > b ? max : b;
|
|
r = r * 255 / max;
|
|
g = g * 255 / max;
|
|
b = b * 255 / max;
|
|
}
|
|
|
|
out[0] = r;
|
|
out[1] = g;
|
|
out[2] = b;
|
|
out[3] = in[3];
|
|
}
|
|
|
|
/*
|
|
===============
|
|
R_LoadLightmaps
|
|
|
|
===============
|
|
*/
|
|
#define LIGHTMAP_SIZE 128
|
|
static void R_LoadLightmaps( lump_t *l ) {
|
|
byte *buf, *buf_p;
|
|
int len;
|
|
byte image[LIGHTMAP_SIZE*LIGHTMAP_SIZE*4];
|
|
int i, j;
|
|
float maxIntensity = 0;
|
|
double sumIntensity = 0;
|
|
|
|
len = l->filelen;
|
|
if ( !len ) {
|
|
return;
|
|
}
|
|
buf = fileBase + l->fileofs;
|
|
|
|
// we are about to upload textures
|
|
R_SyncRenderThread();
|
|
|
|
// create all the lightmaps
|
|
tr.numLightmaps = len / (LIGHTMAP_SIZE * LIGHTMAP_SIZE * 3);
|
|
if ( tr.numLightmaps == 1 ) {
|
|
//FIXME: HACK: maps with only one lightmap turn up fullbright for some reason.
|
|
//this avoids this, but isn't the correct solution.
|
|
tr.numLightmaps++;
|
|
} else if ( tr.numLightmaps >= MAX_LIGHTMAPS ) { // 20051020 misantropia
|
|
ri.Printf( PRINT_WARNING, "WARNING: number of lightmaps > MAX_LIGHTMAPS\n" );
|
|
tr.numLightmaps = MAX_LIGHTMAPS;
|
|
}
|
|
|
|
// if we are in r_vertexLight mode, we don't need the lightmaps at all
|
|
if ( r_vertexLight->integer || glConfig.hardwareType == GLHW_PERMEDIA2 ) {
|
|
return;
|
|
}
|
|
|
|
for ( i = 0 ; i < tr.numLightmaps ; i++ ) {
|
|
// expand the 24 bit on-disk to 32 bit
|
|
buf_p = buf + i * LIGHTMAP_SIZE*LIGHTMAP_SIZE * 3;
|
|
|
|
if ( r_lightmap->integer == 2 )
|
|
{ // color code by intensity as development tool (FIXME: check range)
|
|
for ( j = 0; j < LIGHTMAP_SIZE * LIGHTMAP_SIZE; j++ )
|
|
{
|
|
float r = buf_p[j*3+0];
|
|
float g = buf_p[j*3+1];
|
|
float b = buf_p[j*3+2];
|
|
float intensity;
|
|
float out[3] = {0.0, 0.0, 0.0};
|
|
|
|
intensity = 0.33f * r + 0.685f * g + 0.063f * b;
|
|
|
|
if ( intensity > 255 )
|
|
intensity = 1.0f;
|
|
else
|
|
intensity /= 255.0f;
|
|
|
|
if ( intensity > maxIntensity )
|
|
maxIntensity = intensity;
|
|
|
|
HSVtoRGB( intensity, 1.00, 0.50, out );
|
|
|
|
image[j*4+0] = out[0] * 255;
|
|
image[j*4+1] = out[1] * 255;
|
|
image[j*4+2] = out[2] * 255;
|
|
image[j*4+3] = 255;
|
|
|
|
sumIntensity += intensity;
|
|
}
|
|
} else {
|
|
for ( j = 0 ; j < LIGHTMAP_SIZE * LIGHTMAP_SIZE; j++ ) {
|
|
R_ColorShiftLightingBytes( &buf_p[j*3], &image[j*4] );
|
|
image[j*4+3] = 255;
|
|
}
|
|
}
|
|
tr.lightmaps[i] = R_CreateImage( va("*lightmap%d",i), image,
|
|
LIGHTMAP_SIZE, LIGHTMAP_SIZE, qfalse, qfalse, GL_CLAMP );
|
|
}
|
|
|
|
if ( r_lightmap->integer == 2 ) {
|
|
ri.Printf( PRINT_ALL, "Brightest lightmap value: %d\n", ( int ) ( maxIntensity * 255 ) );
|
|
}
|
|
}
|
|
|
|
|
|
/*
|
|
=================
|
|
RE_SetWorldVisData
|
|
|
|
This is called by the clipmodel subsystem so we can share the 1.8 megs of
|
|
space in big maps...
|
|
=================
|
|
*/
|
|
void RE_SetWorldVisData( const byte *vis ) {
|
|
tr.externalVisData = vis;
|
|
}
|
|
|
|
|
|
/*
|
|
=================
|
|
R_LoadVisibility
|
|
=================
|
|
*/
|
|
static void R_LoadVisibility( lump_t *l ) {
|
|
int len;
|
|
byte *buf;
|
|
|
|
len = ( s_worldData.numClusters + 63 ) & ~63;
|
|
s_worldData.novis = ri.Hunk_Alloc( len, h_low );
|
|
Com_Memset( s_worldData.novis, 0xff, len );
|
|
|
|
len = l->filelen;
|
|
if ( !len ) {
|
|
return;
|
|
}
|
|
buf = fileBase + l->fileofs;
|
|
|
|
s_worldData.numClusters = LittleLong( ((int *)buf)[0] );
|
|
s_worldData.clusterBytes = LittleLong( ((int *)buf)[1] );
|
|
|
|
// CM_Load should have given us the vis data to share, so
|
|
// we don't need to allocate another copy
|
|
if ( tr.externalVisData ) {
|
|
s_worldData.vis = tr.externalVisData;
|
|
} else {
|
|
byte *dest;
|
|
|
|
dest = ri.Hunk_Alloc( len - 8, h_low );
|
|
Com_Memcpy( dest, buf + 8, len - 8 );
|
|
s_worldData.vis = dest;
|
|
}
|
|
}
|
|
|
|
//===============================================================================
|
|
|
|
|
|
/*
|
|
===============
|
|
ShaderForShaderNum
|
|
===============
|
|
*/
|
|
static shader_t *ShaderForShaderNum( int shaderNum, int lightmapNum ) {
|
|
shader_t *shader;
|
|
dshader_t *dsh;
|
|
|
|
shaderNum = LittleLong( shaderNum );
|
|
if ( shaderNum < 0 || shaderNum >= s_worldData.numShaders ) {
|
|
ri.Error( ERR_DROP, "ShaderForShaderNum: bad num %i", shaderNum );
|
|
}
|
|
dsh = &s_worldData.shaders[ shaderNum ];
|
|
|
|
if ( r_vertexLight->integer || glConfig.hardwareType == GLHW_PERMEDIA2 ) {
|
|
lightmapNum = LIGHTMAP_BY_VERTEX;
|
|
}
|
|
|
|
if ( r_fullbright->integer ) {
|
|
lightmapNum = LIGHTMAP_WHITEIMAGE;
|
|
}
|
|
|
|
shader = R_FindShader( dsh->shader, lightmapNum, qtrue );
|
|
|
|
// if the shader had errors, just use default shader
|
|
if ( shader->defaultShader ) {
|
|
return tr.defaultShader;
|
|
}
|
|
|
|
return shader;
|
|
}
|
|
|
|
/*
|
|
===============
|
|
ParseFace
|
|
===============
|
|
*/
|
|
static void ParseFace( dsurface_t *ds, drawVert_t *verts, msurface_t *surf, int *indexes ) {
|
|
int i, j;
|
|
srfSurfaceFace_t *cv;
|
|
int numPoints, numIndexes;
|
|
int lightmapNum;
|
|
int sfaceSize, ofsIndexes;
|
|
|
|
lightmapNum = LittleLong( ds->lightmapNum );
|
|
|
|
// get fog volume
|
|
surf->fogIndex = LittleLong( ds->fogNum ) + 1;
|
|
|
|
// get shader value
|
|
surf->shader = ShaderForShaderNum( ds->shaderNum, lightmapNum );
|
|
if ( r_singleShader->integer && !surf->shader->isSky ) {
|
|
surf->shader = tr.defaultShader;
|
|
}
|
|
|
|
numPoints = LittleLong( ds->numVerts );
|
|
if (numPoints > MAX_FACE_POINTS) {
|
|
ri.Printf( PRINT_WARNING, "WARNING: MAX_FACE_POINTS exceeded: %i\n", numPoints);
|
|
numPoints = MAX_FACE_POINTS;
|
|
surf->shader = tr.defaultShader;
|
|
}
|
|
|
|
numIndexes = LittleLong( ds->numIndexes );
|
|
|
|
// create the srfSurfaceFace_t
|
|
sfaceSize = ( size_t ) &((srfSurfaceFace_t *)0)->points[numPoints];
|
|
ofsIndexes = sfaceSize;
|
|
sfaceSize += sizeof( int ) * numIndexes;
|
|
|
|
cv = ri.Hunk_Alloc( sfaceSize, h_low );
|
|
cv->surfaceType = SF_FACE;
|
|
cv->numPoints = numPoints;
|
|
cv->numIndices = numIndexes;
|
|
cv->ofsIndices = ofsIndexes;
|
|
|
|
verts += LittleLong( ds->firstVert );
|
|
for ( i = 0 ; i < numPoints ; i++ ) {
|
|
for ( j = 0 ; j < 3 ; j++ ) {
|
|
cv->points[i][j] = LittleFloat( verts[i].xyz[j] );
|
|
}
|
|
for ( j = 0 ; j < 2 ; j++ ) {
|
|
cv->points[i][3+j] = LittleFloat( verts[i].st[j] );
|
|
cv->points[i][5+j] = LittleFloat( verts[i].lightmap[j] );
|
|
}
|
|
R_ColorShiftLightingBytes( verts[i].color, (byte *)&cv->points[i][7] );
|
|
}
|
|
|
|
indexes += LittleLong( ds->firstIndex );
|
|
for ( i = 0 ; i < numIndexes ; i++ ) {
|
|
((int *)((byte *)cv + cv->ofsIndices ))[i] = LittleLong( indexes[ i ] );
|
|
}
|
|
|
|
// take the plane information from the lightmap vector
|
|
for ( i = 0 ; i < 3 ; i++ ) {
|
|
cv->plane.normal[i] = LittleFloat( ds->lightmapVecs[2][i] );
|
|
}
|
|
cv->plane.dist = DotProduct( cv->points[0], cv->plane.normal );
|
|
SetPlaneSignbits( &cv->plane );
|
|
cv->plane.type = PlaneTypeForNormal( cv->plane.normal );
|
|
|
|
surf->data = (surfaceType_t *)cv;
|
|
}
|
|
|
|
|
|
/*
|
|
===============
|
|
ParseMesh
|
|
===============
|
|
*/
|
|
static void ParseMesh ( dsurface_t *ds, drawVert_t *verts, msurface_t *surf ) {
|
|
srfGridMesh_t *grid;
|
|
int i, j;
|
|
int width, height, numPoints;
|
|
drawVert_t points[MAX_PATCH_SIZE*MAX_PATCH_SIZE];
|
|
int lightmapNum;
|
|
vec3_t bounds[2];
|
|
vec3_t tmpVec;
|
|
static surfaceType_t skipData = SF_SKIP;
|
|
|
|
lightmapNum = LittleLong( ds->lightmapNum );
|
|
|
|
// get fog volume
|
|
surf->fogIndex = LittleLong( ds->fogNum ) + 1;
|
|
|
|
// get shader value
|
|
surf->shader = ShaderForShaderNum( ds->shaderNum, lightmapNum );
|
|
if ( r_singleShader->integer && !surf->shader->isSky ) {
|
|
surf->shader = tr.defaultShader;
|
|
}
|
|
|
|
// we may have a nodraw surface, because they might still need to
|
|
// be around for movement clipping
|
|
if ( s_worldData.shaders[ LittleLong( ds->shaderNum ) ].surfaceFlags & SURF_NODRAW ) {
|
|
surf->data = &skipData;
|
|
return;
|
|
}
|
|
|
|
width = LittleLong( ds->patchWidth );
|
|
height = LittleLong( ds->patchHeight );
|
|
|
|
verts += LittleLong( ds->firstVert );
|
|
numPoints = width * height;
|
|
for ( i = 0 ; i < numPoints ; i++ ) {
|
|
for ( j = 0 ; j < 3 ; j++ ) {
|
|
points[i].xyz[j] = LittleFloat( verts[i].xyz[j] );
|
|
points[i].normal[j] = LittleFloat( verts[i].normal[j] );
|
|
}
|
|
for ( j = 0 ; j < 2 ; j++ ) {
|
|
points[i].st[j] = LittleFloat( verts[i].st[j] );
|
|
points[i].lightmap[j] = LittleFloat( verts[i].lightmap[j] );
|
|
}
|
|
R_ColorShiftLightingBytes( verts[i].color, points[i].color );
|
|
}
|
|
|
|
// pre-tesseleate
|
|
grid = R_SubdividePatchToGrid( width, height, points );
|
|
surf->data = (surfaceType_t *)grid;
|
|
|
|
// copy the level of detail origin, which is the center
|
|
// of the group of all curves that must subdivide the same
|
|
// to avoid cracking
|
|
for ( i = 0 ; i < 3 ; i++ ) {
|
|
bounds[0][i] = LittleFloat( ds->lightmapVecs[0][i] );
|
|
bounds[1][i] = LittleFloat( ds->lightmapVecs[1][i] );
|
|
}
|
|
VectorAdd( bounds[0], bounds[1], bounds[1] );
|
|
VectorScale( bounds[1], 0.5f, grid->lodOrigin );
|
|
VectorSubtract( bounds[0], grid->lodOrigin, tmpVec );
|
|
grid->lodRadius = VectorLength( tmpVec );
|
|
}
|
|
|
|
/*
|
|
===============
|
|
ParseTriSurf
|
|
===============
|
|
*/
|
|
static void ParseTriSurf( dsurface_t *ds, drawVert_t *verts, msurface_t *surf, int *indexes ) {
|
|
srfTriangles_t *tri;
|
|
int i, j;
|
|
int numVerts, numIndexes;
|
|
|
|
// get fog volume
|
|
surf->fogIndex = LittleLong( ds->fogNum ) + 1;
|
|
|
|
// get shader
|
|
surf->shader = ShaderForShaderNum( ds->shaderNum, LIGHTMAP_BY_VERTEX );
|
|
if ( r_singleShader->integer && !surf->shader->isSky ) {
|
|
surf->shader = tr.defaultShader;
|
|
}
|
|
|
|
numVerts = LittleLong( ds->numVerts );
|
|
numIndexes = LittleLong( ds->numIndexes );
|
|
|
|
tri = ri.Hunk_Alloc( sizeof( *tri ) + numVerts * sizeof( tri->verts[0] )
|
|
+ numIndexes * sizeof( tri->indexes[0] ), h_low );
|
|
tri->surfaceType = SF_TRIANGLES;
|
|
tri->numVerts = numVerts;
|
|
tri->numIndexes = numIndexes;
|
|
tri->verts = (drawVert_t *)(tri + 1);
|
|
tri->indexes = (int *)(tri->verts + tri->numVerts );
|
|
|
|
surf->data = (surfaceType_t *)tri;
|
|
|
|
// copy vertexes
|
|
ClearBounds( tri->bounds[0], tri->bounds[1] );
|
|
verts += LittleLong( ds->firstVert );
|
|
for ( i = 0 ; i < numVerts ; i++ ) {
|
|
for ( j = 0 ; j < 3 ; j++ ) {
|
|
tri->verts[i].xyz[j] = LittleFloat( verts[i].xyz[j] );
|
|
tri->verts[i].normal[j] = LittleFloat( verts[i].normal[j] );
|
|
}
|
|
AddPointToBounds( tri->verts[i].xyz, tri->bounds[0], tri->bounds[1] );
|
|
for ( j = 0 ; j < 2 ; j++ ) {
|
|
tri->verts[i].st[j] = LittleFloat( verts[i].st[j] );
|
|
tri->verts[i].lightmap[j] = LittleFloat( verts[i].lightmap[j] );
|
|
}
|
|
|
|
R_ColorShiftLightingBytes( verts[i].color, tri->verts[i].color );
|
|
}
|
|
|
|
// copy indexes
|
|
indexes += LittleLong( ds->firstIndex );
|
|
for ( i = 0 ; i < numIndexes ; i++ ) {
|
|
tri->indexes[i] = LittleLong( indexes[i] );
|
|
if ( tri->indexes[i] < 0 || tri->indexes[i] >= numVerts ) {
|
|
ri.Error( ERR_DROP, "Bad index in triangle surface" );
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
===============
|
|
ParseFlare
|
|
===============
|
|
*/
|
|
static void ParseFlare( dsurface_t *ds, drawVert_t *verts, msurface_t *surf, int *indexes ) {
|
|
srfFlare_t *flare;
|
|
int i;
|
|
|
|
// get fog volume
|
|
surf->fogIndex = LittleLong( ds->fogNum ) + 1;
|
|
|
|
// get shader
|
|
surf->shader = ShaderForShaderNum( ds->shaderNum, LIGHTMAP_BY_VERTEX );
|
|
if ( r_singleShader->integer && !surf->shader->isSky ) {
|
|
surf->shader = tr.defaultShader;
|
|
}
|
|
|
|
flare = ri.Hunk_Alloc( sizeof( *flare ), h_low );
|
|
flare->surfaceType = SF_FLARE;
|
|
|
|
surf->data = (surfaceType_t *)flare;
|
|
|
|
for ( i = 0 ; i < 3 ; i++ ) {
|
|
flare->origin[i] = LittleFloat( ds->lightmapOrigin[i] );
|
|
flare->color[i] = LittleFloat( ds->lightmapVecs[0][i] );
|
|
flare->normal[i] = LittleFloat( ds->lightmapVecs[2][i] );
|
|
}
|
|
}
|
|
|
|
|
|
/*
|
|
=================
|
|
R_MergedWidthPoints
|
|
|
|
returns true if there are grid points merged on a width edge
|
|
=================
|
|
*/
|
|
int R_MergedWidthPoints(srfGridMesh_t *grid, int offset) {
|
|
int i, j;
|
|
|
|
for (i = 1; i < grid->width-1; i++) {
|
|
for (j = i + 1; j < grid->width-1; j++) {
|
|
if ( fabs(grid->verts[i + offset].xyz[0] - grid->verts[j + offset].xyz[0]) > .1) continue;
|
|
if ( fabs(grid->verts[i + offset].xyz[1] - grid->verts[j + offset].xyz[1]) > .1) continue;
|
|
if ( fabs(grid->verts[i + offset].xyz[2] - grid->verts[j + offset].xyz[2]) > .1) continue;
|
|
return qtrue;
|
|
}
|
|
}
|
|
return qfalse;
|
|
}
|
|
|
|
/*
|
|
=================
|
|
R_MergedHeightPoints
|
|
|
|
returns true if there are grid points merged on a height edge
|
|
=================
|
|
*/
|
|
int R_MergedHeightPoints(srfGridMesh_t *grid, int offset) {
|
|
int i, j;
|
|
|
|
for (i = 1; i < grid->height-1; i++) {
|
|
for (j = i + 1; j < grid->height-1; j++) {
|
|
if ( fabs(grid->verts[grid->width * i + offset].xyz[0] - grid->verts[grid->width * j + offset].xyz[0]) > .1) continue;
|
|
if ( fabs(grid->verts[grid->width * i + offset].xyz[1] - grid->verts[grid->width * j + offset].xyz[1]) > .1) continue;
|
|
if ( fabs(grid->verts[grid->width * i + offset].xyz[2] - grid->verts[grid->width * j + offset].xyz[2]) > .1) continue;
|
|
return qtrue;
|
|
}
|
|
}
|
|
return qfalse;
|
|
}
|
|
|
|
/*
|
|
=================
|
|
R_FixSharedVertexLodError_r
|
|
|
|
NOTE: never sync LoD through grid edges with merged points!
|
|
|
|
FIXME: write generalized version that also avoids cracks between a patch and one that meets half way?
|
|
=================
|
|
*/
|
|
void R_FixSharedVertexLodError_r( int start, srfGridMesh_t *grid1 ) {
|
|
int j, k, l, m, n, offset1, offset2, touch;
|
|
srfGridMesh_t *grid2;
|
|
|
|
for ( j = start; j < s_worldData.numsurfaces; j++ ) {
|
|
//
|
|
grid2 = (srfGridMesh_t *) s_worldData.surfaces[j].data;
|
|
// if this surface is not a grid
|
|
if ( grid2->surfaceType != SF_GRID ) continue;
|
|
// if the LOD errors are already fixed for this patch
|
|
if ( grid2->lodFixed == 2 ) continue;
|
|
// grids in the same LOD group should have the exact same lod radius
|
|
if ( grid1->lodRadius != grid2->lodRadius ) continue;
|
|
// grids in the same LOD group should have the exact same lod origin
|
|
if ( grid1->lodOrigin[0] != grid2->lodOrigin[0] ) continue;
|
|
if ( grid1->lodOrigin[1] != grid2->lodOrigin[1] ) continue;
|
|
if ( grid1->lodOrigin[2] != grid2->lodOrigin[2] ) continue;
|
|
//
|
|
touch = qfalse;
|
|
for (n = 0; n < 2; n++) {
|
|
//
|
|
if (n) offset1 = (grid1->height-1) * grid1->width;
|
|
else offset1 = 0;
|
|
if (R_MergedWidthPoints(grid1, offset1)) continue;
|
|
for (k = 1; k < grid1->width-1; k++) {
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (m) offset2 = (grid2->height-1) * grid2->width;
|
|
else offset2 = 0;
|
|
if (R_MergedWidthPoints(grid2, offset2)) continue;
|
|
for ( l = 1; l < grid2->width-1; l++) {
|
|
//
|
|
if ( fabs(grid1->verts[k + offset1].xyz[0] - grid2->verts[l + offset2].xyz[0]) > .1) continue;
|
|
if ( fabs(grid1->verts[k + offset1].xyz[1] - grid2->verts[l + offset2].xyz[1]) > .1) continue;
|
|
if ( fabs(grid1->verts[k + offset1].xyz[2] - grid2->verts[l + offset2].xyz[2]) > .1) continue;
|
|
// ok the points are equal and should have the same lod error
|
|
grid2->widthLodError[l] = grid1->widthLodError[k];
|
|
touch = qtrue;
|
|
}
|
|
}
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (m) offset2 = grid2->width-1;
|
|
else offset2 = 0;
|
|
if (R_MergedHeightPoints(grid2, offset2)) continue;
|
|
for ( l = 1; l < grid2->height-1; l++) {
|
|
//
|
|
if ( fabs(grid1->verts[k + offset1].xyz[0] - grid2->verts[grid2->width * l + offset2].xyz[0]) > .1) continue;
|
|
if ( fabs(grid1->verts[k + offset1].xyz[1] - grid2->verts[grid2->width * l + offset2].xyz[1]) > .1) continue;
|
|
if ( fabs(grid1->verts[k + offset1].xyz[2] - grid2->verts[grid2->width * l + offset2].xyz[2]) > .1) continue;
|
|
// ok the points are equal and should have the same lod error
|
|
grid2->heightLodError[l] = grid1->widthLodError[k];
|
|
touch = qtrue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
for (n = 0; n < 2; n++) {
|
|
//
|
|
if (n) offset1 = grid1->width-1;
|
|
else offset1 = 0;
|
|
if (R_MergedHeightPoints(grid1, offset1)) continue;
|
|
for (k = 1; k < grid1->height-1; k++) {
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (m) offset2 = (grid2->height-1) * grid2->width;
|
|
else offset2 = 0;
|
|
if (R_MergedWidthPoints(grid2, offset2)) continue;
|
|
for ( l = 1; l < grid2->width-1; l++) {
|
|
//
|
|
if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[0] - grid2->verts[l + offset2].xyz[0]) > .1) continue;
|
|
if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[1] - grid2->verts[l + offset2].xyz[1]) > .1) continue;
|
|
if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[2] - grid2->verts[l + offset2].xyz[2]) > .1) continue;
|
|
// ok the points are equal and should have the same lod error
|
|
grid2->widthLodError[l] = grid1->heightLodError[k];
|
|
touch = qtrue;
|
|
}
|
|
}
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (m) offset2 = grid2->width-1;
|
|
else offset2 = 0;
|
|
if (R_MergedHeightPoints(grid2, offset2)) continue;
|
|
for ( l = 1; l < grid2->height-1; l++) {
|
|
//
|
|
if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[0] - grid2->verts[grid2->width * l + offset2].xyz[0]) > .1) continue;
|
|
if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[1] - grid2->verts[grid2->width * l + offset2].xyz[1]) > .1) continue;
|
|
if ( fabs(grid1->verts[grid1->width * k + offset1].xyz[2] - grid2->verts[grid2->width * l + offset2].xyz[2]) > .1) continue;
|
|
// ok the points are equal and should have the same lod error
|
|
grid2->heightLodError[l] = grid1->heightLodError[k];
|
|
touch = qtrue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
if (touch) {
|
|
grid2->lodFixed = 2;
|
|
R_FixSharedVertexLodError_r ( start, grid2 );
|
|
//NOTE: this would be correct but makes things really slow
|
|
//grid2->lodFixed = 1;
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
=================
|
|
R_FixSharedVertexLodError
|
|
|
|
This function assumes that all patches in one group are nicely stitched together for the highest LoD.
|
|
If this is not the case this function will still do its job but won't fix the highest LoD cracks.
|
|
=================
|
|
*/
|
|
void R_FixSharedVertexLodError( void ) {
|
|
int i;
|
|
srfGridMesh_t *grid1;
|
|
|
|
for ( i = 0; i < s_worldData.numsurfaces; i++ ) {
|
|
//
|
|
grid1 = (srfGridMesh_t *) s_worldData.surfaces[i].data;
|
|
// if this surface is not a grid
|
|
if ( grid1->surfaceType != SF_GRID )
|
|
continue;
|
|
//
|
|
if ( grid1->lodFixed )
|
|
continue;
|
|
//
|
|
grid1->lodFixed = 2;
|
|
// recursively fix other patches in the same LOD group
|
|
R_FixSharedVertexLodError_r( i + 1, grid1);
|
|
}
|
|
}
|
|
|
|
|
|
/*
|
|
===============
|
|
R_StitchPatches
|
|
===============
|
|
*/
|
|
int R_StitchPatches( int grid1num, int grid2num ) {
|
|
float *v1, *v2;
|
|
srfGridMesh_t *grid1, *grid2;
|
|
int k, l, m, n, offset1, offset2, row, column;
|
|
|
|
grid1 = (srfGridMesh_t *) s_worldData.surfaces[grid1num].data;
|
|
grid2 = (srfGridMesh_t *) s_worldData.surfaces[grid2num].data;
|
|
for (n = 0; n < 2; n++) {
|
|
//
|
|
if (n) offset1 = (grid1->height-1) * grid1->width;
|
|
else offset1 = 0;
|
|
if (R_MergedWidthPoints(grid1, offset1))
|
|
continue;
|
|
for (k = 0; k < grid1->width-2; k += 2) {
|
|
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if ( grid2->width >= MAX_GRID_SIZE )
|
|
break;
|
|
if (m) offset2 = (grid2->height-1) * grid2->width;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->width-1; l++) {
|
|
//
|
|
v1 = grid1->verts[k + offset1].xyz;
|
|
v2 = grid2->verts[l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[k + 2 + offset1].xyz;
|
|
v2 = grid2->verts[l + 1 + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[l + offset2].xyz;
|
|
v2 = grid2->verts[l + 1 + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert column into grid2 right after after column l
|
|
if (m) row = grid2->height-1;
|
|
else row = 0;
|
|
grid2 = R_GridInsertColumn( grid2, l+1, row,
|
|
grid1->verts[k + 1 + offset1].xyz, grid1->widthLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (grid2->height >= MAX_GRID_SIZE)
|
|
break;
|
|
if (m) offset2 = grid2->width-1;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->height-1; l++) {
|
|
//
|
|
v1 = grid1->verts[k + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[k + 2 + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert row into grid2 right after after row l
|
|
if (m) column = grid2->width-1;
|
|
else column = 0;
|
|
grid2 = R_GridInsertRow( grid2, l+1, column,
|
|
grid1->verts[k + 1 + offset1].xyz, grid1->widthLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
for (n = 0; n < 2; n++) {
|
|
//
|
|
if (n) offset1 = grid1->width-1;
|
|
else offset1 = 0;
|
|
if (R_MergedHeightPoints(grid1, offset1))
|
|
continue;
|
|
for (k = 0; k < grid1->height-2; k += 2) {
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if ( grid2->width >= MAX_GRID_SIZE )
|
|
break;
|
|
if (m) offset2 = (grid2->height-1) * grid2->width;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->width-1; l++) {
|
|
//
|
|
v1 = grid1->verts[grid1->width * k + offset1].xyz;
|
|
v2 = grid2->verts[l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[grid1->width * (k + 2) + offset1].xyz;
|
|
v2 = grid2->verts[l + 1 + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[l + offset2].xyz;
|
|
v2 = grid2->verts[(l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert column into grid2 right after after column l
|
|
if (m) row = grid2->height-1;
|
|
else row = 0;
|
|
grid2 = R_GridInsertColumn( grid2, l+1, row,
|
|
grid1->verts[grid1->width * (k + 1) + offset1].xyz, grid1->heightLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (grid2->height >= MAX_GRID_SIZE)
|
|
break;
|
|
if (m) offset2 = grid2->width-1;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->height-1; l++) {
|
|
//
|
|
v1 = grid1->verts[grid1->width * k + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[grid1->width * (k + 2) + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert row into grid2 right after after row l
|
|
if (m) column = grid2->width-1;
|
|
else column = 0;
|
|
grid2 = R_GridInsertRow( grid2, l+1, column,
|
|
grid1->verts[grid1->width * (k + 1) + offset1].xyz, grid1->heightLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
for (n = 0; n < 2; n++) {
|
|
//
|
|
if (n) offset1 = (grid1->height-1) * grid1->width;
|
|
else offset1 = 0;
|
|
if (R_MergedWidthPoints(grid1, offset1))
|
|
continue;
|
|
for (k = grid1->width-1; k > 1; k -= 2) {
|
|
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if ( grid2->width >= MAX_GRID_SIZE )
|
|
break;
|
|
if (m) offset2 = (grid2->height-1) * grid2->width;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->width-1; l++) {
|
|
//
|
|
v1 = grid1->verts[k + offset1].xyz;
|
|
v2 = grid2->verts[l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[k - 2 + offset1].xyz;
|
|
v2 = grid2->verts[l + 1 + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[l + offset2].xyz;
|
|
v2 = grid2->verts[(l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert column into grid2 right after after column l
|
|
if (m) row = grid2->height-1;
|
|
else row = 0;
|
|
grid2 = R_GridInsertColumn( grid2, l+1, row,
|
|
grid1->verts[k - 1 + offset1].xyz, grid1->widthLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (grid2->height >= MAX_GRID_SIZE)
|
|
break;
|
|
if (m) offset2 = grid2->width-1;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->height-1; l++) {
|
|
//
|
|
v1 = grid1->verts[k + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[k - 2 + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert row into grid2 right after after row l
|
|
if (m) column = grid2->width-1;
|
|
else column = 0;
|
|
grid2 = R_GridInsertRow( grid2, l+1, column,
|
|
grid1->verts[k - 1 + offset1].xyz, grid1->widthLodError[k+1]);
|
|
if (!grid2)
|
|
break;
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
for (n = 0; n < 2; n++) {
|
|
//
|
|
if (n) offset1 = grid1->width-1;
|
|
else offset1 = 0;
|
|
if (R_MergedHeightPoints(grid1, offset1))
|
|
continue;
|
|
for (k = grid1->height-1; k > 1; k -= 2) {
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if ( grid2->width >= MAX_GRID_SIZE )
|
|
break;
|
|
if (m) offset2 = (grid2->height-1) * grid2->width;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->width-1; l++) {
|
|
//
|
|
v1 = grid1->verts[grid1->width * k + offset1].xyz;
|
|
v2 = grid2->verts[l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[grid1->width * (k - 2) + offset1].xyz;
|
|
v2 = grid2->verts[l + 1 + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[l + offset2].xyz;
|
|
v2 = grid2->verts[(l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert column into grid2 right after after column l
|
|
if (m) row = grid2->height-1;
|
|
else row = 0;
|
|
grid2 = R_GridInsertColumn( grid2, l+1, row,
|
|
grid1->verts[grid1->width * (k - 1) + offset1].xyz, grid1->heightLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
for (m = 0; m < 2; m++) {
|
|
|
|
if (grid2->height >= MAX_GRID_SIZE)
|
|
break;
|
|
if (m) offset2 = grid2->width-1;
|
|
else offset2 = 0;
|
|
for ( l = 0; l < grid2->height-1; l++) {
|
|
//
|
|
v1 = grid1->verts[grid1->width * k + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
|
|
v1 = grid1->verts[grid1->width * (k - 2) + offset1].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) > .1)
|
|
continue;
|
|
if ( fabs(v1[1] - v2[1]) > .1)
|
|
continue;
|
|
if ( fabs(v1[2] - v2[2]) > .1)
|
|
continue;
|
|
//
|
|
v1 = grid2->verts[grid2->width * l + offset2].xyz;
|
|
v2 = grid2->verts[grid2->width * (l + 1) + offset2].xyz;
|
|
if ( fabs(v1[0] - v2[0]) < .01 &&
|
|
fabs(v1[1] - v2[1]) < .01 &&
|
|
fabs(v1[2] - v2[2]) < .01)
|
|
continue;
|
|
//
|
|
//ri.Printf( PRINT_ALL, "found highest LoD crack between two patches\n" );
|
|
// insert row into grid2 right after after row l
|
|
if (m) column = grid2->width-1;
|
|
else column = 0;
|
|
grid2 = R_GridInsertRow( grid2, l+1, column,
|
|
grid1->verts[grid1->width * (k - 1) + offset1].xyz, grid1->heightLodError[k+1]);
|
|
grid2->lodStitched = qfalse;
|
|
s_worldData.surfaces[grid2num].data = (void *) grid2;
|
|
return qtrue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
return qfalse;
|
|
}
|
|
|
|
/*
|
|
===============
|
|
R_TryStitchPatch
|
|
|
|
This function will try to stitch patches in the same LoD group together for the highest LoD.
|
|
|
|
Only single missing vertice cracks will be fixed.
|
|
|
|
Vertices will be joined at the patch side a crack is first found, at the other side
|
|
of the patch (on the same row or column) the vertices will not be joined and cracks
|
|
might still appear at that side.
|
|
===============
|
|
*/
|
|
int R_TryStitchingPatch( int grid1num ) {
|
|
int j, numstitches;
|
|
srfGridMesh_t *grid1, *grid2;
|
|
|
|
numstitches = 0;
|
|
grid1 = (srfGridMesh_t *) s_worldData.surfaces[grid1num].data;
|
|
for ( j = 0; j < s_worldData.numsurfaces; j++ ) {
|
|
//
|
|
grid2 = (srfGridMesh_t *) s_worldData.surfaces[j].data;
|
|
// if this surface is not a grid
|
|
if ( grid2->surfaceType != SF_GRID ) continue;
|
|
// grids in the same LOD group should have the exact same lod radius
|
|
if ( grid1->lodRadius != grid2->lodRadius ) continue;
|
|
// grids in the same LOD group should have the exact same lod origin
|
|
if ( grid1->lodOrigin[0] != grid2->lodOrigin[0] ) continue;
|
|
if ( grid1->lodOrigin[1] != grid2->lodOrigin[1] ) continue;
|
|
if ( grid1->lodOrigin[2] != grid2->lodOrigin[2] ) continue;
|
|
//
|
|
while (R_StitchPatches(grid1num, j))
|
|
{
|
|
numstitches++;
|
|
}
|
|
}
|
|
return numstitches;
|
|
}
|
|
|
|
/*
|
|
===============
|
|
R_StitchAllPatches
|
|
===============
|
|
*/
|
|
void R_StitchAllPatches( void ) {
|
|
int i, stitched, numstitches;
|
|
srfGridMesh_t *grid1;
|
|
|
|
numstitches = 0;
|
|
do
|
|
{
|
|
stitched = qfalse;
|
|
for ( i = 0; i < s_worldData.numsurfaces; i++ ) {
|
|
//
|
|
grid1 = (srfGridMesh_t *) s_worldData.surfaces[i].data;
|
|
// if this surface is not a grid
|
|
if ( grid1->surfaceType != SF_GRID )
|
|
continue;
|
|
//
|
|
if ( grid1->lodStitched )
|
|
continue;
|
|
//
|
|
grid1->lodStitched = qtrue;
|
|
stitched = qtrue;
|
|
//
|
|
numstitches += R_TryStitchingPatch( i );
|
|
}
|
|
}
|
|
while (stitched);
|
|
ri.Printf( PRINT_ALL, "stitched %d LoD cracks\n", numstitches );
|
|
}
|
|
|
|
/*
|
|
===============
|
|
R_MovePatchSurfacesToHunk
|
|
===============
|
|
*/
|
|
void R_MovePatchSurfacesToHunk(void) {
|
|
int i, size;
|
|
srfGridMesh_t *grid, *hunkgrid;
|
|
|
|
for ( i = 0; i < s_worldData.numsurfaces; i++ ) {
|
|
//
|
|
grid = (srfGridMesh_t *) s_worldData.surfaces[i].data;
|
|
// if this surface is not a grid
|
|
if ( grid->surfaceType != SF_GRID )
|
|
continue;
|
|
//
|
|
size = (grid->width * grid->height - 1) * sizeof( drawVert_t ) + sizeof( *grid );
|
|
hunkgrid = ri.Hunk_Alloc( size, h_low );
|
|
Com_Memcpy(hunkgrid, grid, size);
|
|
|
|
hunkgrid->widthLodError = ri.Hunk_Alloc( grid->width * 4, h_low );
|
|
Com_Memcpy( hunkgrid->widthLodError, grid->widthLodError, grid->width * 4 );
|
|
|
|
hunkgrid->heightLodError = ri.Hunk_Alloc( grid->height * 4, h_low );
|
|
Com_Memcpy( hunkgrid->heightLodError, grid->heightLodError, grid->height * 4 );
|
|
|
|
R_FreeSurfaceGridMesh( grid );
|
|
|
|
s_worldData.surfaces[i].data = (void *) hunkgrid;
|
|
}
|
|
}
|
|
|
|
/*
|
|
===============
|
|
R_LoadSurfaces
|
|
===============
|
|
*/
|
|
static void R_LoadSurfaces( lump_t *surfs, lump_t *verts, lump_t *indexLump ) {
|
|
dsurface_t *in;
|
|
msurface_t *out;
|
|
drawVert_t *dv;
|
|
int *indexes;
|
|
int count;
|
|
int numFaces, numMeshes, numTriSurfs, numFlares;
|
|
int i;
|
|
|
|
numFaces = 0;
|
|
numMeshes = 0;
|
|
numTriSurfs = 0;
|
|
numFlares = 0;
|
|
|
|
in = (void *)(fileBase + surfs->fileofs);
|
|
if (surfs->filelen % sizeof(*in))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
count = surfs->filelen / sizeof(*in);
|
|
|
|
dv = (void *)(fileBase + verts->fileofs);
|
|
if (verts->filelen % sizeof(*dv))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
|
|
indexes = (void *)(fileBase + indexLump->fileofs);
|
|
if ( indexLump->filelen % sizeof(*indexes))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
|
|
out = ri.Hunk_Alloc ( count * sizeof(*out), h_low );
|
|
|
|
s_worldData.surfaces = out;
|
|
s_worldData.numsurfaces = count;
|
|
|
|
for ( i = 0 ; i < count ; i++, in++, out++ ) {
|
|
switch ( LittleLong( in->surfaceType ) ) {
|
|
case MST_PATCH:
|
|
ParseMesh ( in, dv, out );
|
|
numMeshes++;
|
|
break;
|
|
case MST_TRIANGLE_SOUP:
|
|
ParseTriSurf( in, dv, out, indexes );
|
|
numTriSurfs++;
|
|
break;
|
|
case MST_PLANAR:
|
|
ParseFace( in, dv, out, indexes );
|
|
numFaces++;
|
|
break;
|
|
case MST_FLARE:
|
|
ParseFlare( in, dv, out, indexes );
|
|
numFlares++;
|
|
break;
|
|
default:
|
|
ri.Error( ERR_DROP, "Bad surfaceType" );
|
|
}
|
|
}
|
|
|
|
#ifdef PATCH_STITCHING
|
|
R_StitchAllPatches();
|
|
#endif
|
|
|
|
R_FixSharedVertexLodError();
|
|
|
|
#ifdef PATCH_STITCHING
|
|
R_MovePatchSurfacesToHunk();
|
|
#endif
|
|
|
|
ri.Printf( PRINT_ALL, "...loaded %d faces, %i meshes, %i trisurfs, %i flares\n",
|
|
numFaces, numMeshes, numTriSurfs, numFlares );
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
=================
|
|
R_LoadSubmodels
|
|
=================
|
|
*/
|
|
static void R_LoadSubmodels( lump_t *l ) {
|
|
dmodel_t *in;
|
|
bmodel_t *out;
|
|
int i, j, count;
|
|
|
|
in = (void *)(fileBase + l->fileofs);
|
|
if (l->filelen % sizeof(*in))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
count = l->filelen / sizeof(*in);
|
|
|
|
s_worldData.bmodels = out = ri.Hunk_Alloc( count * sizeof(*out), h_low );
|
|
|
|
for ( i=0 ; i<count ; i++, in++, out++ ) {
|
|
model_t *model;
|
|
|
|
model = R_AllocModel();
|
|
|
|
assert( model != NULL ); // this should never happen
|
|
|
|
model->type = MOD_BRUSH;
|
|
model->bmodel = out;
|
|
Com_sprintf( model->name, sizeof( model->name ), "*%d", i );
|
|
|
|
for (j=0 ; j<3 ; j++) {
|
|
out->bounds[0][j] = LittleFloat (in->mins[j]);
|
|
out->bounds[1][j] = LittleFloat (in->maxs[j]);
|
|
}
|
|
|
|
out->firstSurface = s_worldData.surfaces + LittleLong( in->firstSurface );
|
|
out->numSurfaces = LittleLong( in->numSurfaces );
|
|
}
|
|
}
|
|
|
|
|
|
|
|
//==================================================================
|
|
|
|
/*
|
|
=================
|
|
R_SetParent
|
|
=================
|
|
*/
|
|
static void R_SetParent (mnode_t *node, mnode_t *parent)
|
|
{
|
|
node->parent = parent;
|
|
if (node->contents != -1)
|
|
return;
|
|
R_SetParent (node->children[0], node);
|
|
R_SetParent (node->children[1], node);
|
|
}
|
|
|
|
/*
|
|
=================
|
|
R_LoadNodesAndLeafs
|
|
=================
|
|
*/
|
|
static void R_LoadNodesAndLeafs (lump_t *nodeLump, lump_t *leafLump) {
|
|
int i, j, p;
|
|
dnode_t *in;
|
|
dleaf_t *inLeaf;
|
|
mnode_t *out;
|
|
int numNodes, numLeafs;
|
|
|
|
in = (void *)(fileBase + nodeLump->fileofs);
|
|
if (nodeLump->filelen % sizeof(dnode_t) ||
|
|
leafLump->filelen % sizeof(dleaf_t) ) {
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
}
|
|
numNodes = nodeLump->filelen / sizeof(dnode_t);
|
|
numLeafs = leafLump->filelen / sizeof(dleaf_t);
|
|
|
|
out = ri.Hunk_Alloc ( (numNodes + numLeafs) * sizeof(*out), h_low);
|
|
|
|
s_worldData.nodes = out;
|
|
s_worldData.numnodes = numNodes + numLeafs;
|
|
s_worldData.numDecisionNodes = numNodes;
|
|
|
|
// load nodes
|
|
for ( i=0 ; i<numNodes; i++, in++, out++)
|
|
{
|
|
for (j=0 ; j<3 ; j++)
|
|
{
|
|
out->mins[j] = LittleLong (in->mins[j]);
|
|
out->maxs[j] = LittleLong (in->maxs[j]);
|
|
}
|
|
|
|
p = LittleLong(in->planeNum);
|
|
out->plane = s_worldData.planes + p;
|
|
|
|
out->contents = CONTENTS_NODE; // differentiate from leafs
|
|
|
|
for (j=0 ; j<2 ; j++)
|
|
{
|
|
p = LittleLong (in->children[j]);
|
|
if (p >= 0)
|
|
out->children[j] = s_worldData.nodes + p;
|
|
else
|
|
out->children[j] = s_worldData.nodes + numNodes + (-1 - p);
|
|
}
|
|
}
|
|
|
|
// load leafs
|
|
inLeaf = (void *)(fileBase + leafLump->fileofs);
|
|
for ( i=0 ; i<numLeafs ; i++, inLeaf++, out++)
|
|
{
|
|
for (j=0 ; j<3 ; j++)
|
|
{
|
|
out->mins[j] = LittleLong (inLeaf->mins[j]);
|
|
out->maxs[j] = LittleLong (inLeaf->maxs[j]);
|
|
}
|
|
|
|
out->cluster = LittleLong(inLeaf->cluster);
|
|
out->area = LittleLong(inLeaf->area);
|
|
|
|
if ( out->cluster >= s_worldData.numClusters ) {
|
|
s_worldData.numClusters = out->cluster + 1;
|
|
}
|
|
|
|
out->firstmarksurface = s_worldData.marksurfaces +
|
|
LittleLong(inLeaf->firstLeafSurface);
|
|
out->nummarksurfaces = LittleLong(inLeaf->numLeafSurfaces);
|
|
}
|
|
|
|
// chain decendants
|
|
R_SetParent (s_worldData.nodes, NULL);
|
|
}
|
|
|
|
//=============================================================================
|
|
|
|
/*
|
|
=================
|
|
R_LoadShaders
|
|
=================
|
|
*/
|
|
static void R_LoadShaders( lump_t *l ) {
|
|
int i, count;
|
|
dshader_t *in, *out;
|
|
|
|
in = (void *)(fileBase + l->fileofs);
|
|
if (l->filelen % sizeof(*in))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
count = l->filelen / sizeof(*in);
|
|
out = ri.Hunk_Alloc ( count*sizeof(*out), h_low );
|
|
|
|
s_worldData.shaders = out;
|
|
s_worldData.numShaders = count;
|
|
|
|
Com_Memcpy( out, in, count*sizeof(*out) );
|
|
|
|
for ( i=0 ; i<count ; i++ ) {
|
|
out[i].surfaceFlags = LittleLong( out[i].surfaceFlags );
|
|
out[i].contentFlags = LittleLong( out[i].contentFlags );
|
|
}
|
|
}
|
|
|
|
|
|
/*
|
|
=================
|
|
R_LoadMarksurfaces
|
|
=================
|
|
*/
|
|
static void R_LoadMarksurfaces (lump_t *l)
|
|
{
|
|
int i, j, count;
|
|
int *in;
|
|
msurface_t **out;
|
|
|
|
in = (void *)(fileBase + l->fileofs);
|
|
if (l->filelen % sizeof(*in))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
count = l->filelen / sizeof(*in);
|
|
out = ri.Hunk_Alloc ( count*sizeof(*out), h_low);
|
|
|
|
s_worldData.marksurfaces = out;
|
|
s_worldData.nummarksurfaces = count;
|
|
|
|
for ( i=0 ; i<count ; i++)
|
|
{
|
|
j = LittleLong(in[i]);
|
|
out[i] = s_worldData.surfaces + j;
|
|
}
|
|
}
|
|
|
|
|
|
/*
|
|
=================
|
|
R_LoadPlanes
|
|
=================
|
|
*/
|
|
static void R_LoadPlanes( lump_t *l ) {
|
|
int i, j;
|
|
cplane_t *out;
|
|
dplane_t *in;
|
|
int count;
|
|
int bits;
|
|
|
|
in = (void *)(fileBase + l->fileofs);
|
|
if (l->filelen % sizeof(*in))
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
count = l->filelen / sizeof(*in);
|
|
out = ri.Hunk_Alloc ( count*2*sizeof(*out), h_low);
|
|
|
|
s_worldData.planes = out;
|
|
s_worldData.numplanes = count;
|
|
|
|
for ( i=0 ; i<count ; i++, in++, out++) {
|
|
bits = 0;
|
|
for (j=0 ; j<3 ; j++) {
|
|
out->normal[j] = LittleFloat (in->normal[j]);
|
|
if (out->normal[j] < 0) {
|
|
bits |= 1<<j;
|
|
}
|
|
}
|
|
|
|
out->dist = LittleFloat (in->dist);
|
|
out->type = PlaneTypeForNormal( out->normal );
|
|
out->signbits = bits;
|
|
}
|
|
}
|
|
|
|
/*
|
|
=================
|
|
R_LoadFogs
|
|
|
|
=================
|
|
*/
|
|
static void R_LoadFogs( lump_t *l, lump_t *brushesLump, lump_t *sidesLump ) {
|
|
int i;
|
|
fog_t *out;
|
|
dfog_t *fogs;
|
|
dbrush_t *brushes, *brush;
|
|
dbrushside_t *sides;
|
|
int count, brushesCount, sidesCount;
|
|
int sideNum;
|
|
int planeNum;
|
|
shader_t *shader;
|
|
float d;
|
|
int firstSide;
|
|
|
|
fogs = (void *)(fileBase + l->fileofs);
|
|
if (l->filelen % sizeof(*fogs)) {
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
}
|
|
count = l->filelen / sizeof(*fogs);
|
|
|
|
// create fog strucutres for them
|
|
s_worldData.numfogs = count + 1;
|
|
s_worldData.fogs = ri.Hunk_Alloc ( s_worldData.numfogs*sizeof(*out), h_low);
|
|
out = s_worldData.fogs + 1;
|
|
|
|
if ( !count ) {
|
|
return;
|
|
}
|
|
|
|
brushes = (void *)(fileBase + brushesLump->fileofs);
|
|
if (brushesLump->filelen % sizeof(*brushes)) {
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
}
|
|
brushesCount = brushesLump->filelen / sizeof(*brushes);
|
|
|
|
sides = (void *)(fileBase + sidesLump->fileofs);
|
|
if (sidesLump->filelen % sizeof(*sides)) {
|
|
ri.Error (ERR_DROP, "LoadMap: funny lump size in %s",s_worldData.name);
|
|
}
|
|
sidesCount = sidesLump->filelen / sizeof(*sides);
|
|
|
|
for ( i=0 ; i<count ; i++, fogs++) {
|
|
out->originalBrushNumber = LittleLong( fogs->brushNum );
|
|
|
|
if ( (unsigned)out->originalBrushNumber >= brushesCount ) {
|
|
ri.Error( ERR_DROP, "fog brushNumber out of range" );
|
|
}
|
|
brush = brushes + out->originalBrushNumber;
|
|
|
|
firstSide = LittleLong( brush->firstSide );
|
|
|
|
if ( (unsigned)firstSide > sidesCount - 6 ) {
|
|
ri.Error( ERR_DROP, "fog brush sideNumber out of range" );
|
|
}
|
|
|
|
// brushes are always sorted with the axial sides first
|
|
sideNum = firstSide + 0;
|
|
planeNum = LittleLong( sides[ sideNum ].planeNum );
|
|
out->bounds[0][0] = -s_worldData.planes[ planeNum ].dist;
|
|
|
|
sideNum = firstSide + 1;
|
|
planeNum = LittleLong( sides[ sideNum ].planeNum );
|
|
out->bounds[1][0] = s_worldData.planes[ planeNum ].dist;
|
|
|
|
sideNum = firstSide + 2;
|
|
planeNum = LittleLong( sides[ sideNum ].planeNum );
|
|
out->bounds[0][1] = -s_worldData.planes[ planeNum ].dist;
|
|
|
|
sideNum = firstSide + 3;
|
|
planeNum = LittleLong( sides[ sideNum ].planeNum );
|
|
out->bounds[1][1] = s_worldData.planes[ planeNum ].dist;
|
|
|
|
sideNum = firstSide + 4;
|
|
planeNum = LittleLong( sides[ sideNum ].planeNum );
|
|
out->bounds[0][2] = -s_worldData.planes[ planeNum ].dist;
|
|
|
|
sideNum = firstSide + 5;
|
|
planeNum = LittleLong( sides[ sideNum ].planeNum );
|
|
out->bounds[1][2] = s_worldData.planes[ planeNum ].dist;
|
|
|
|
// get information from the shader for fog parameters
|
|
shader = R_FindShader( fogs->shader, LIGHTMAP_NONE, qtrue );
|
|
|
|
out->parms = shader->fogParms;
|
|
|
|
out->colorInt = ColorBytes4 ( shader->fogParms.color[0] * tr.identityLight,
|
|
shader->fogParms.color[1] * tr.identityLight,
|
|
shader->fogParms.color[2] * tr.identityLight, 1.0 );
|
|
|
|
d = shader->fogParms.depthForOpaque < 1 ? 1 : shader->fogParms.depthForOpaque;
|
|
out->tcScale = 1.0f / ( d * 8 );
|
|
|
|
// set the gradient vector
|
|
sideNum = LittleLong( fogs->visibleSide );
|
|
|
|
if ( sideNum == -1 ) {
|
|
out->hasSurface = qfalse;
|
|
} else {
|
|
out->hasSurface = qtrue;
|
|
planeNum = LittleLong( sides[ firstSide + sideNum ].planeNum );
|
|
VectorSubtract( vec3_origin, s_worldData.planes[ planeNum ].normal, out->surface );
|
|
out->surface[3] = -s_worldData.planes[ planeNum ].dist;
|
|
}
|
|
|
|
out++;
|
|
}
|
|
|
|
}
|
|
|
|
|
|
/*
|
|
================
|
|
R_LoadLightGrid
|
|
|
|
================
|
|
*/
|
|
void R_LoadLightGrid( lump_t *l ) {
|
|
int i;
|
|
vec3_t maxs;
|
|
int numGridPoints;
|
|
world_t *w;
|
|
float *wMins, *wMaxs;
|
|
|
|
w = &s_worldData;
|
|
|
|
w->lightGridInverseSize[0] = 1.0f / w->lightGridSize[0];
|
|
w->lightGridInverseSize[1] = 1.0f / w->lightGridSize[1];
|
|
w->lightGridInverseSize[2] = 1.0f / w->lightGridSize[2];
|
|
|
|
wMins = w->bmodels[0].bounds[0];
|
|
wMaxs = w->bmodels[0].bounds[1];
|
|
|
|
for ( i = 0 ; i < 3 ; i++ ) {
|
|
w->lightGridOrigin[i] = w->lightGridSize[i] * ceil( wMins[i] / w->lightGridSize[i] );
|
|
maxs[i] = w->lightGridSize[i] * floor( wMaxs[i] / w->lightGridSize[i] );
|
|
w->lightGridBounds[i] = (maxs[i] - w->lightGridOrigin[i])/w->lightGridSize[i] + 1;
|
|
}
|
|
|
|
numGridPoints = w->lightGridBounds[0] * w->lightGridBounds[1] * w->lightGridBounds[2];
|
|
|
|
if ( l->filelen != numGridPoints * 8 ) {
|
|
ri.Printf( PRINT_WARNING, "WARNING: light grid mismatch\n" );
|
|
w->lightGridData = NULL;
|
|
return;
|
|
}
|
|
|
|
w->lightGridData = ri.Hunk_Alloc( l->filelen, h_low );
|
|
Com_Memcpy( w->lightGridData, (void *)(fileBase + l->fileofs), l->filelen );
|
|
|
|
// deal with overbright bits
|
|
for ( i = 0 ; i < numGridPoints ; i++ ) {
|
|
R_ColorShiftLightingBytes( &w->lightGridData[i*8], &w->lightGridData[i*8] );
|
|
R_ColorShiftLightingBytes( &w->lightGridData[i*8+3], &w->lightGridData[i*8+3] );
|
|
}
|
|
}
|
|
|
|
/*
|
|
================
|
|
R_LoadEntities
|
|
================
|
|
*/
|
|
void R_LoadEntities( lump_t *l ) {
|
|
char *p, *token, *s;
|
|
char keyname[MAX_TOKEN_CHARS];
|
|
char value[MAX_TOKEN_CHARS];
|
|
world_t *w;
|
|
|
|
w = &s_worldData;
|
|
w->lightGridSize[0] = 64;
|
|
w->lightGridSize[1] = 64;
|
|
w->lightGridSize[2] = 128;
|
|
|
|
p = (char *)(fileBase + l->fileofs);
|
|
|
|
// store for reference by the cgame
|
|
w->entityString = ri.Hunk_Alloc( l->filelen + 1, h_low );
|
|
strcpy( w->entityString, p );
|
|
w->entityParsePoint = w->entityString;
|
|
|
|
token = COM_ParseExt( &p, qtrue );
|
|
if (!*token || *token != '{') {
|
|
return;
|
|
}
|
|
|
|
// only parse the world spawn
|
|
while ( 1 ) {
|
|
// parse key
|
|
token = COM_ParseExt( &p, qtrue );
|
|
|
|
if ( !*token || *token == '}' ) {
|
|
break;
|
|
}
|
|
Q_strncpyz(keyname, token, sizeof(keyname));
|
|
|
|
// parse value
|
|
token = COM_ParseExt( &p, qtrue );
|
|
|
|
if ( !*token || *token == '}' ) {
|
|
break;
|
|
}
|
|
Q_strncpyz(value, token, sizeof(value));
|
|
|
|
// check for remapping of shaders for vertex lighting
|
|
s = "vertexremapshader";
|
|
if (!Q_strncmp(keyname, s, strlen(s)) ) {
|
|
s = strchr(value, ';');
|
|
if (!s) {
|
|
ri.Printf( PRINT_WARNING, "WARNING: no semi colon in vertexshaderremap '%s'\n", value );
|
|
break;
|
|
}
|
|
*s++ = 0;
|
|
if (r_vertexLight->integer) {
|
|
R_RemapShader(value, s, "0");
|
|
}
|
|
continue;
|
|
}
|
|
// check for remapping of shaders
|
|
s = "remapshader";
|
|
if (!Q_strncmp(keyname, s, strlen(s)) ) {
|
|
s = strchr(value, ';');
|
|
if (!s) {
|
|
ri.Printf( PRINT_WARNING, "WARNING: no semi colon in shaderremap '%s'\n", value );
|
|
break;
|
|
}
|
|
*s++ = 0;
|
|
R_RemapShader(value, s, "0");
|
|
continue;
|
|
}
|
|
// check for a different grid size
|
|
if (!Q_stricmp(keyname, "gridsize")) {
|
|
sscanf(value, "%f %f %f", &w->lightGridSize[0], &w->lightGridSize[1], &w->lightGridSize[2] );
|
|
continue;
|
|
}
|
|
}
|
|
}
|
|
|
|
/*
|
|
=================
|
|
R_GetEntityToken
|
|
=================
|
|
*/
|
|
qboolean R_GetEntityToken( char *buffer, int size ) {
|
|
const char *s;
|
|
|
|
s = COM_Parse( &s_worldData.entityParsePoint );
|
|
Q_strncpyz( buffer, s, size );
|
|
if ( !s_worldData.entityParsePoint || !s[0] ) {
|
|
s_worldData.entityParsePoint = s_worldData.entityString;
|
|
return qfalse;
|
|
} else {
|
|
return qtrue;
|
|
}
|
|
}
|
|
|
|
/*
|
|
=================
|
|
RE_LoadWorldMap
|
|
|
|
Called directly from cgame
|
|
=================
|
|
*/
|
|
void RE_LoadWorldMap( const char *name ) {
|
|
int i;
|
|
dheader_t *header;
|
|
byte *buffer;
|
|
byte *startMarker;
|
|
|
|
if ( tr.worldMapLoaded ) {
|
|
ri.Error( ERR_DROP, "ERROR: attempted to redundantly load world map\n" );
|
|
}
|
|
|
|
// set default sun direction to be used if it isn't
|
|
// overridden by a shader
|
|
tr.sunDirection[0] = 0.45f;
|
|
tr.sunDirection[1] = 0.3f;
|
|
tr.sunDirection[2] = 0.9f;
|
|
|
|
VectorNormalize( tr.sunDirection );
|
|
|
|
tr.worldMapLoaded = qtrue;
|
|
|
|
// load it
|
|
ri.FS_ReadFile( name, (void **)&buffer );
|
|
if ( !buffer ) {
|
|
ri.Error (ERR_DROP, "RE_LoadWorldMap: %s not found", name);
|
|
}
|
|
|
|
// clear tr.world so if the level fails to load, the next
|
|
// try will not look at the partially loaded version
|
|
tr.world = NULL;
|
|
|
|
Com_Memset( &s_worldData, 0, sizeof( s_worldData ) );
|
|
Q_strncpyz( s_worldData.name, name, sizeof( s_worldData.name ) );
|
|
|
|
Q_strncpyz( s_worldData.baseName, COM_SkipPath( s_worldData.name ), sizeof( s_worldData.name ) );
|
|
COM_StripExtension(s_worldData.baseName, s_worldData.baseName, sizeof(s_worldData.baseName));
|
|
|
|
startMarker = ri.Hunk_Alloc(0, h_low);
|
|
c_gridVerts = 0;
|
|
|
|
header = (dheader_t *)buffer;
|
|
fileBase = (byte *)header;
|
|
|
|
i = LittleLong (header->version);
|
|
if ( i != BSP_VERSION ) {
|
|
ri.Error (ERR_DROP, "RE_LoadWorldMap: %s has wrong version number (%i should be %i)",
|
|
name, i, BSP_VERSION);
|
|
}
|
|
|
|
// swap all the lumps
|
|
for (i=0 ; i<sizeof(dheader_t)/4 ; i++) {
|
|
((int *)header)[i] = LittleLong ( ((int *)header)[i]);
|
|
}
|
|
|
|
// load into heap
|
|
R_LoadShaders( &header->lumps[LUMP_SHADERS] );
|
|
R_LoadLightmaps( &header->lumps[LUMP_LIGHTMAPS] );
|
|
R_LoadPlanes (&header->lumps[LUMP_PLANES]);
|
|
R_LoadFogs( &header->lumps[LUMP_FOGS], &header->lumps[LUMP_BRUSHES], &header->lumps[LUMP_BRUSHSIDES] );
|
|
R_LoadSurfaces( &header->lumps[LUMP_SURFACES], &header->lumps[LUMP_DRAWVERTS], &header->lumps[LUMP_DRAWINDEXES] );
|
|
R_LoadMarksurfaces (&header->lumps[LUMP_LEAFSURFACES]);
|
|
R_LoadNodesAndLeafs (&header->lumps[LUMP_NODES], &header->lumps[LUMP_LEAFS]);
|
|
R_LoadSubmodels (&header->lumps[LUMP_MODELS]);
|
|
R_LoadVisibility( &header->lumps[LUMP_VISIBILITY] );
|
|
R_LoadEntities( &header->lumps[LUMP_ENTITIES] );
|
|
R_LoadLightGrid( &header->lumps[LUMP_LIGHTGRID] );
|
|
|
|
s_worldData.dataSize = (byte *)ri.Hunk_Alloc(0, h_low) - startMarker;
|
|
|
|
// only set tr.world now that we know the entire level has loaded properly
|
|
tr.world = &s_worldData;
|
|
|
|
ri.FS_FreeFile( buffer );
|
|
}
|
|
|