jedioutcast/CODE-mp/renderer/tr_shade_calc.cpp
2013-04-04 13:24:26 -05:00

1307 lines
31 KiB
C++

// tr_shade_calc.c
#include "tr_local.h"
#define WAVEVALUE( table, base, amplitude, phase, freq ) ((base) + table[ myftol( ( ( (phase) + tess.shaderTime * (freq) ) * FUNCTABLE_SIZE ) ) & FUNCTABLE_MASK ] * (amplitude))
static float *TableForFunc( genFunc_t func )
{
switch ( func )
{
case GF_SIN:
return tr.sinTable;
case GF_TRIANGLE:
return tr.triangleTable;
case GF_SQUARE:
return tr.squareTable;
case GF_SAWTOOTH:
return tr.sawToothTable;
case GF_INVERSE_SAWTOOTH:
return tr.inverseSawToothTable;
case GF_NONE:
default:
break;
}
ri.Error( ERR_DROP, "TableForFunc called with invalid function '%d' in shader '%s'\n", func, tess.shader->name );
return NULL;
}
/*
** EvalWaveForm
**
** Evaluates a given waveForm_t, referencing backEnd.refdef.time directly
*/
extern float GetNoiseTime( int t ); //from tr_noise, returns 0 to 2
static float EvalWaveForm( const waveForm_t *wf )
{
float *table;
if ( wf->func == GF_NOISE ) {
return ( wf->base + R_NoiseGet4f( 0, 0, 0, ( backEnd.refdef.floatTime + wf->phase ) * wf->frequency ) * wf->amplitude );
} else if (wf->func == GF_RAND) {
if( GetNoiseTime( backEnd.refdef.time + wf->phase ) <= wf->frequency ) {
return (wf->base + wf->amplitude);
} else {
return wf->base;
}
}
table = TableForFunc( wf->func );
return WAVEVALUE( table, wf->base, wf->amplitude, wf->phase, wf->frequency );
}
static float EvalWaveFormClamped( const waveForm_t *wf )
{
float glow = EvalWaveForm( wf );
if ( glow < 0 )
{
return 0;
}
if ( glow > 1 )
{
return 1;
}
return glow;
}
/*
** RB_CalcStretchTexCoords
*/
void RB_CalcStretchTexCoords( const waveForm_t *wf, float *st )
{
float p;
texModInfo_t tmi;
p = 1.0f / EvalWaveForm( wf );
tmi.matrix[0][0] = p;
tmi.matrix[1][0] = 0;
tmi.translate[0] = 0.5f - 0.5f * p;
tmi.matrix[0][1] = 0;
tmi.matrix[1][1] = p;
tmi.translate[1] = 0.5f - 0.5f * p;
RB_CalcTransformTexCoords( &tmi, st );
}
/*
====================================================================
DEFORMATIONS
====================================================================
*/
/*
========================
RB_CalcDeformVertexes
========================
*/
void RB_CalcDeformVertexes( deformStage_t *ds )
{
int i;
vec3_t offset;
float scale;
float *xyz = ( float * ) tess.xyz;
float *normal = ( float * ) tess.normal;
float *table;
if ( ds->deformationWave.frequency == 0 )
{
scale = EvalWaveForm( &ds->deformationWave );
for ( i = 0; i < tess.numVertexes; i++, xyz += 4, normal += 4 )
{
VectorScale( normal, scale, offset );
xyz[0] += offset[0];
xyz[1] += offset[1];
xyz[2] += offset[2];
}
}
else
{
table = TableForFunc( ds->deformationWave.func );
for ( i = 0; i < tess.numVertexes; i++, xyz += 4, normal += 4 )
{
float off = ( xyz[0] + xyz[1] + xyz[2] ) * ds->deformationSpread;
scale = WAVEVALUE( table, ds->deformationWave.base,
ds->deformationWave.amplitude,
ds->deformationWave.phase + off,
ds->deformationWave.frequency );
VectorScale( normal, scale, offset );
xyz[0] += offset[0];
xyz[1] += offset[1];
xyz[2] += offset[2];
}
}
}
/*
=========================
RB_CalcDeformNormals
Wiggle the normals for wavy environment mapping
=========================
*/
void RB_CalcDeformNormals( deformStage_t *ds ) {
int i;
float scale;
float *xyz = ( float * ) tess.xyz;
float *normal = ( float * ) tess.normal;
for ( i = 0; i < tess.numVertexes; i++, xyz += 4, normal += 4 ) {
scale = 0.98f;
scale = R_NoiseGet4f( xyz[0] * scale, xyz[1] * scale, xyz[2] * scale,
tess.shaderTime * ds->deformationWave.frequency );
normal[ 0 ] += ds->deformationWave.amplitude * scale;
scale = 0.98f;
scale = R_NoiseGet4f( 100 + xyz[0] * scale, xyz[1] * scale, xyz[2] * scale,
tess.shaderTime * ds->deformationWave.frequency );
normal[ 1 ] += ds->deformationWave.amplitude * scale;
scale = 0.98f;
scale = R_NoiseGet4f( 200 + xyz[0] * scale, xyz[1] * scale, xyz[2] * scale,
tess.shaderTime * ds->deformationWave.frequency );
normal[ 2 ] += ds->deformationWave.amplitude * scale;
VectorNormalizeFast( normal );
}
}
/*
========================
RB_CalcBulgeVertexes
========================
*/
void RB_CalcBulgeVertexes( deformStage_t *ds )
{
//Old bulge code:
/*
int i;
const float *st = ( const float * ) tess.texCoords[0];
float *xyz = ( float * ) tess.xyz;
float *normal = ( float * ) tess.normal;
float now;
now = backEnd.refdef.time * ds->bulgeSpeed * 0.001f;
for ( i = 0; i < tess.numVertexes; i++, xyz += 4, st += 2 * NUM_TEX_COORDS, normal += 4 ) {
int off;
float scale;
off = (float)( FUNCTABLE_SIZE / (M_PI*2) ) * ( st[0] * ds->bulgeWidth + now );
scale = tr.sinTable[ off & FUNCTABLE_MASK ] * ds->bulgeHeight;
xyz[0] += normal[0] * scale;
xyz[1] += normal[1] * scale;
xyz[2] += normal[2] * scale;
}
*/
int i;
float *xyz = ( float * ) tess.xyz;
float *normal = ( float * ) tess.normal;
float scale;
if ( ds->bulgeSpeed == 0.0f && ds->bulgeWidth == 0.0f )
{
// We don't have a speed and width, so just use height to expand uniformly
for ( i = 0; i < tess.numVertexes; i++, xyz += 4, normal += 4 )
{
xyz[0] += normal[0] * ds->bulgeHeight;
xyz[1] += normal[1] * ds->bulgeHeight;
xyz[2] += normal[2] * ds->bulgeHeight;
}
}
else
{
// I guess do some extra dumb stuff..the fact that it uses ST seems bad though because skin pages may be set up in certain ways that can cause
// very noticeable seams on sufaces ( like on the huge ion_cannon ).
const float *st = ( const float * ) tess.texCoords[0];
float now;
int off;
now = backEnd.refdef.time * ds->bulgeSpeed * 0.001f;
for ( i = 0; i < tess.numVertexes; i++, xyz += 4, st += 2 * NUM_TEX_COORDS, normal += 4 )
{
off = (float)( FUNCTABLE_SIZE / (M_PI*2) ) * ( st[0] * ds->bulgeWidth + now );
scale = tr.sinTable[ off & FUNCTABLE_MASK ] * ds->bulgeHeight;
xyz[0] += normal[0] * scale;
xyz[1] += normal[1] * scale;
xyz[2] += normal[2] * scale;
}
}
}
/*
======================
RB_CalcMoveVertexes
A deformation that can move an entire surface along a wave path
======================
*/
void RB_CalcMoveVertexes( deformStage_t *ds ) {
int i;
float *xyz;
float *table;
float scale;
vec3_t offset;
table = TableForFunc( ds->deformationWave.func );
scale = WAVEVALUE( table, ds->deformationWave.base,
ds->deformationWave.amplitude,
ds->deformationWave.phase,
ds->deformationWave.frequency );
VectorScale( ds->moveVector, scale, offset );
xyz = ( float * ) tess.xyz;
for ( i = 0; i < tess.numVertexes; i++, xyz += 4 ) {
VectorAdd( xyz, offset, xyz );
}
}
/*
=============
DeformText
Change a polygon into a bunch of text polygons
=============
*/
void DeformText( const char *text ) {
int i;
vec3_t origin, width, height;
int len;
int ch;
byte color[4];
float bottom, top;
vec3_t mid;
height[0] = 0;
height[1] = 0;
height[2] = -1;
CrossProduct( tess.normal[0], height, width );
// find the midpoint of the box
VectorClear( mid );
bottom = 999999;
top = -999999;
for ( i = 0 ; i < 4 ; i++ ) {
VectorAdd( tess.xyz[i], mid, mid );
if ( tess.xyz[i][2] < bottom ) {
bottom = tess.xyz[i][2];
}
if ( tess.xyz[i][2] > top ) {
top = tess.xyz[i][2];
}
}
VectorScale( mid, 0.25f, origin );
// determine the individual character size
height[0] = 0;
height[1] = 0;
height[2] = ( top - bottom ) * 0.5f;
VectorScale( width, height[2] * -0.75f, width );
// determine the starting position
len = strlen( text );
VectorMA( origin, (len-1), width, origin );
// clear the shader indexes
tess.numIndexes = 0;
tess.numVertexes = 0;
color[0] = color[1] = color[2] = color[3] = 255;
// draw each character
for ( i = 0 ; i < len ; i++ ) {
ch = text[i];
ch &= 255;
if ( ch != ' ' ) {
int row, col;
float frow, fcol, size;
row = ch>>4;
col = ch&15;
frow = row*0.0625f;
fcol = col*0.0625f;
size = 0.0625f;
RB_AddQuadStampExt( origin, width, height, color, fcol, frow, fcol + size, frow + size );
}
VectorMA( origin, -2, width, origin );
}
}
/*
==================
GlobalVectorToLocal
==================
*/
static void GlobalVectorToLocal( const vec3_t in, vec3_t out ) {
out[0] = DotProduct( in, backEnd.ori.axis[0] );
out[1] = DotProduct( in, backEnd.ori.axis[1] );
out[2] = DotProduct( in, backEnd.ori.axis[2] );
}
/*
=====================
AutospriteDeform
Assuming all the triangles for this shader are independant
quads, rebuild them as forward facing sprites
=====================
*/
static void AutospriteDeform( void ) {
int i;
int oldVerts;
float *xyz;
vec3_t mid, delta;
float radius;
vec3_t left, up;
vec3_t leftDir, upDir;
if ( tess.numVertexes & 3 ) {
ri.Printf( PRINT_WARNING, "Autosprite shader %s had odd vertex count", tess.shader->name );
}
if ( tess.numIndexes != ( tess.numVertexes >> 2 ) * 6 ) {
ri.Printf( PRINT_WARNING, "Autosprite shader %s had odd index count", tess.shader->name );
}
oldVerts = tess.numVertexes;
tess.numVertexes = 0;
tess.numIndexes = 0;
if ( backEnd.currentEntity != &tr.worldEntity ) {
GlobalVectorToLocal( backEnd.viewParms.ori.axis[1], leftDir );
GlobalVectorToLocal( backEnd.viewParms.ori.axis[2], upDir );
} else {
VectorCopy( backEnd.viewParms.ori.axis[1], leftDir );
VectorCopy( backEnd.viewParms.ori.axis[2], upDir );
}
for ( i = 0 ; i < oldVerts ; i+=4 ) {
// find the midpoint
xyz = tess.xyz[i];
mid[0] = 0.25f * (xyz[0] + xyz[4] + xyz[8] + xyz[12]);
mid[1] = 0.25f * (xyz[1] + xyz[5] + xyz[9] + xyz[13]);
mid[2] = 0.25f * (xyz[2] + xyz[6] + xyz[10] + xyz[14]);
VectorSubtract( xyz, mid, delta );
radius = VectorLength( delta ) * 0.707f; // / sqrt(2)
VectorScale( leftDir, radius, left );
VectorScale( upDir, radius, up );
if ( backEnd.viewParms.isMirror ) {
VectorSubtract( vec3_origin, left, left );
}
// compensate for scale in the axes if necessary
if ( backEnd.currentEntity->e.nonNormalizedAxes ) {
float axisLength;
axisLength = VectorLength( backEnd.currentEntity->e.axis[0] );
if ( !axisLength ) {
axisLength = 0;
} else {
axisLength = 1.0f / axisLength;
}
VectorScale(left, axisLength, left);
VectorScale(up, axisLength, up);
}
RB_AddQuadStamp( mid, left, up, tess.vertexColors[i] );
}
}
/*
=====================
Autosprite2Deform
Autosprite2 will pivot a rectangular quad along the center of its long axis
=====================
*/
int edgeVerts[6][2] = {
{ 0, 1 },
{ 0, 2 },
{ 0, 3 },
{ 1, 2 },
{ 1, 3 },
{ 2, 3 }
};
static void Autosprite2Deform( void ) {
int i, j, k;
int indexes;
float *xyz;
vec3_t forward;
if ( tess.numVertexes & 3 ) {
ri.Printf( PRINT_WARNING, "Autosprite2 shader %s had odd vertex count", tess.shader->name );
}
if ( tess.numIndexes != ( tess.numVertexes >> 2 ) * 6 ) {
ri.Printf( PRINT_WARNING, "Autosprite2 shader %s had odd index count", tess.shader->name );
}
if ( backEnd.currentEntity != &tr.worldEntity ) {
GlobalVectorToLocal( backEnd.viewParms.ori.axis[0], forward );
} else {
VectorCopy( backEnd.viewParms.ori.axis[0], forward );
}
// this is a lot of work for two triangles...
// we could precalculate a lot of it is an issue, but it would mess up
// the shader abstraction
for ( i = 0, indexes = 0 ; i < tess.numVertexes ; i+=4, indexes+=6 ) {
float lengths[2];
int nums[2];
vec3_t mid[2];
vec3_t major, minor;
float *v1, *v2;
// find the midpoint
xyz = tess.xyz[i];
// identify the two shortest edges
nums[0] = nums[1] = 0;
lengths[0] = lengths[1] = 999999;
for ( j = 0 ; j < 6 ; j++ ) {
float l;
vec3_t temp;
v1 = xyz + 4 * edgeVerts[j][0];
v2 = xyz + 4 * edgeVerts[j][1];
VectorSubtract( v1, v2, temp );
l = DotProduct( temp, temp );
if ( l < lengths[0] ) {
nums[1] = nums[0];
lengths[1] = lengths[0];
nums[0] = j;
lengths[0] = l;
} else if ( l < lengths[1] ) {
nums[1] = j;
lengths[1] = l;
}
}
for ( j = 0 ; j < 2 ; j++ ) {
v1 = xyz + 4 * edgeVerts[nums[j]][0];
v2 = xyz + 4 * edgeVerts[nums[j]][1];
mid[j][0] = 0.5f * (v1[0] + v2[0]);
mid[j][1] = 0.5f * (v1[1] + v2[1]);
mid[j][2] = 0.5f * (v1[2] + v2[2]);
}
// find the vector of the major axis
VectorSubtract( mid[1], mid[0], major );
// cross this with the view direction to get minor axis
CrossProduct( major, forward, minor );
VectorNormalize( minor );
// re-project the points
for ( j = 0 ; j < 2 ; j++ ) {
float l;
v1 = xyz + 4 * edgeVerts[nums[j]][0];
v2 = xyz + 4 * edgeVerts[nums[j]][1];
l = 0.5 * sqrt( lengths[j] );
// we need to see which direction this edge
// is used to determine direction of projection
for ( k = 0 ; k < 5 ; k++ ) {
if ( tess.indexes[ indexes + k ] == i + edgeVerts[nums[j]][0]
&& tess.indexes[ indexes + k + 1 ] == i + edgeVerts[nums[j]][1] ) {
break;
}
}
if ( k == 5 ) {
VectorMA( mid[j], l, minor, v1 );
VectorMA( mid[j], -l, minor, v2 );
} else {
VectorMA( mid[j], -l, minor, v1 );
VectorMA( mid[j], l, minor, v2 );
}
}
}
}
/*
=====================
RB_DeformTessGeometry
=====================
*/
void RB_DeformTessGeometry( void ) {
int i;
deformStage_t *ds;
for ( i = 0 ; i < tess.shader->numDeforms ; i++ ) {
ds = &tess.shader->deforms[ i ];
switch ( ds->deformation ) {
case DEFORM_NONE:
break;
case DEFORM_NORMALS:
RB_CalcDeformNormals( ds );
break;
case DEFORM_WAVE:
RB_CalcDeformVertexes( ds );
break;
case DEFORM_BULGE:
RB_CalcBulgeVertexes( ds );
break;
case DEFORM_MOVE:
RB_CalcMoveVertexes( ds );
break;
case DEFORM_PROJECTION_SHADOW:
RB_ProjectionShadowDeform();
break;
case DEFORM_AUTOSPRITE:
AutospriteDeform();
break;
case DEFORM_AUTOSPRITE2:
Autosprite2Deform();
break;
case DEFORM_TEXT0:
case DEFORM_TEXT1:
case DEFORM_TEXT2:
case DEFORM_TEXT3:
case DEFORM_TEXT4:
case DEFORM_TEXT5:
case DEFORM_TEXT6:
case DEFORM_TEXT7:
DeformText( backEnd.refdef.text[ds->deformation - DEFORM_TEXT0] );
break;
}
}
}
/*
====================================================================
COLORS
====================================================================
*/
/*
** RB_CalcColorFromEntity
*/
void RB_CalcColorFromEntity( unsigned char *dstColors )
{
int i;
int *pColors = ( int * ) dstColors;
int c;
if ( !backEnd.currentEntity )
return;
c = * ( int * ) backEnd.currentEntity->e.shaderRGBA;
for ( i = 0; i < tess.numVertexes; i++, pColors++ )
{
*pColors = c;
}
}
/*
** RB_CalcColorFromOneMinusEntity
*/
void RB_CalcColorFromOneMinusEntity( unsigned char *dstColors )
{
int i;
int *pColors = ( int * ) dstColors;
unsigned char invModulate[3];
int c;
if ( !backEnd.currentEntity )
return;
invModulate[0] = 255 - backEnd.currentEntity->e.shaderRGBA[0];
invModulate[1] = 255 - backEnd.currentEntity->e.shaderRGBA[1];
invModulate[2] = 255 - backEnd.currentEntity->e.shaderRGBA[2];
invModulate[3] = 255 - backEnd.currentEntity->e.shaderRGBA[3]; // this trashes alpha, but the AGEN block fixes it
c = * ( int * ) invModulate;
for ( i = 0; i < tess.numVertexes; i++, pColors++ )
{
*pColors = * ( int * ) invModulate;
}
}
/*
** RB_CalcAlphaFromEntity
*/
void RB_CalcAlphaFromEntity( unsigned char *dstColors )
{
int i;
if ( !backEnd.currentEntity )
return;
dstColors += 3;
for ( i = 0; i < tess.numVertexes; i++, dstColors += 4 )
{
*dstColors = backEnd.currentEntity->e.shaderRGBA[3];
}
}
/*
** RB_CalcAlphaFromOneMinusEntity
*/
void RB_CalcAlphaFromOneMinusEntity( unsigned char *dstColors )
{
int i;
if ( !backEnd.currentEntity )
return;
dstColors += 3;
for ( i = 0; i < tess.numVertexes; i++, dstColors += 4 )
{
*dstColors = 0xff - backEnd.currentEntity->e.shaderRGBA[3];
}
}
/*
** RB_CalcWaveColor
*/
void RB_CalcWaveColor( const waveForm_t *wf, unsigned char *dstColors )
{
int i;
int v;
float glow;
int *colors = ( int * ) dstColors;
byte color[4];
if ( wf->func == GF_NOISE ) {
glow = wf->base + R_NoiseGet4f( 0, 0, 0, ( tess.shaderTime + wf->phase ) * wf->frequency ) * wf->amplitude;
} else {
glow = EvalWaveForm( wf ) * tr.identityLight;
}
if ( glow < 0 ) {
glow = 0;
}
else if ( glow > 1 ) {
glow = 1;
}
v = myftol( 255 * glow );
color[0] = color[1] = color[2] = v;
color[3] = 255;
v = *(int *)color;
for ( i = 0; i < tess.numVertexes; i++, colors++ ) {
*colors = v;
}
}
/*
** RB_CalcWaveAlpha
*/
void RB_CalcWaveAlpha( const waveForm_t *wf, unsigned char *dstColors )
{
int i;
int v;
float glow;
glow = EvalWaveFormClamped( wf );
v = 255 * glow;
for ( i = 0; i < tess.numVertexes; i++, dstColors += 4 )
{
dstColors[3] = v;
}
}
/*
** RB_CalcModulateColorsByFog
*/
void RB_CalcModulateColorsByFog( unsigned char *colors ) {
int i;
float texCoords[SHADER_MAX_VERTEXES][2];
// calculate texcoords so we can derive density
// this is not wasted, because it would only have
// been previously called if the surface was opaque
RB_CalcFogTexCoords( texCoords[0] );
for ( i = 0; i < tess.numVertexes; i++, colors += 4 ) {
float f = 1.0 - R_FogFactor( texCoords[i][0], texCoords[i][1] );
colors[0] *= f;
colors[1] *= f;
colors[2] *= f;
}
}
/*
** RB_CalcModulateAlphasByFog
*/
void RB_CalcModulateAlphasByFog( unsigned char *colors ) {
int i;
float texCoords[SHADER_MAX_VERTEXES][2];
// calculate texcoords so we can derive density
// this is not wasted, because it would only have
// been previously called if the surface was opaque
RB_CalcFogTexCoords( texCoords[0] );
for ( i = 0; i < tess.numVertexes; i++, colors += 4 ) {
float f = 1.0 - R_FogFactor( texCoords[i][0], texCoords[i][1] );
colors[3] *= f;
}
}
/*
** RB_CalcModulateRGBAsByFog
*/
void RB_CalcModulateRGBAsByFog( unsigned char *colors ) {
int i;
float texCoords[SHADER_MAX_VERTEXES][2];
// calculate texcoords so we can derive density
// this is not wasted, because it would only have
// been previously called if the surface was opaque
RB_CalcFogTexCoords( texCoords[0] );
for ( i = 0; i < tess.numVertexes; i++, colors += 4 ) {
float f = 1.0 - R_FogFactor( texCoords[i][0], texCoords[i][1] );
colors[0] *= f;
colors[1] *= f;
colors[2] *= f;
colors[3] *= f;
}
}
/*
====================================================================
TEX COORDS
====================================================================
*/
/*
========================
RB_CalcFogTexCoords
To do the clipped fog plane really correctly, we should use
projected textures, but I don't trust the drivers and it
doesn't fit our shader data.
========================
*/
void RB_CalcFogTexCoords( float *st ) {
int i;
float *v;
float s, t;
float eyeT;
qboolean eyeOutside;
fog_t *fog;
vec3_t local;
vec4_t fogDistanceVector, fogDepthVector;
fog = tr.world->fogs + tess.fogNum;
// all fogging distance is based on world Z units
VectorSubtract( backEnd.ori.origin, backEnd.viewParms.ori.origin, local );
fogDistanceVector[0] = -backEnd.ori.modelMatrix[2];
fogDistanceVector[1] = -backEnd.ori.modelMatrix[6];
fogDistanceVector[2] = -backEnd.ori.modelMatrix[10];
fogDistanceVector[3] = DotProduct( local, backEnd.viewParms.ori.axis[0] );
// scale the fog vectors based on the fog's thickness
fogDistanceVector[0] *= fog->tcScale;
fogDistanceVector[1] *= fog->tcScale;
fogDistanceVector[2] *= fog->tcScale;
fogDistanceVector[3] *= fog->tcScale;
// rotate the gradient vector for this orientation
if ( fog->hasSurface ) {
fogDepthVector[0] = fog->surface[0] * backEnd.ori.axis[0][0] +
fog->surface[1] * backEnd.ori.axis[0][1] + fog->surface[2] * backEnd.ori.axis[0][2];
fogDepthVector[1] = fog->surface[0] * backEnd.ori.axis[1][0] +
fog->surface[1] * backEnd.ori.axis[1][1] + fog->surface[2] * backEnd.ori.axis[1][2];
fogDepthVector[2] = fog->surface[0] * backEnd.ori.axis[2][0] +
fog->surface[1] * backEnd.ori.axis[2][1] + fog->surface[2] * backEnd.ori.axis[2][2];
fogDepthVector[3] = -fog->surface[3] + DotProduct( backEnd.ori.origin, fog->surface );
eyeT = DotProduct( backEnd.ori.viewOrigin, fogDepthVector ) + fogDepthVector[3];
} else {
eyeT = 1; // non-surface fog always has eye inside
}
// see if the viewpoint is outside
// this is needed for clipping distance even for constant fog
if ( eyeT < 0 ) {
eyeOutside = qtrue;
} else {
eyeOutside = qfalse;
}
fogDistanceVector[3] += 1.0/512;
// calculate density for each point
for (i = 0, v = tess.xyz[0] ; i < tess.numVertexes ; i++, v += 4) {
// calculate the length in fog
s = DotProduct( v, fogDistanceVector ) + fogDistanceVector[3];
t = DotProduct( v, fogDepthVector ) + fogDepthVector[3];
// partially clipped fogs use the T axis
if ( eyeOutside ) {
if ( t < 1.0 ) {
t = 1.0/32; // point is outside, so no fogging
} else {
t = 1.0/32 + 30.0/32 * t / ( t - eyeT ); // cut the distance at the fog plane
}
} else {
if ( t < 0 ) {
t = 1.0/32; // point is outside, so no fogging
} else {
t = 31.0/32;
}
}
st[0] = s;
st[1] = t;
st += 2;
}
}
/*
** RB_CalcEnvironmentTexCoords
*/
void RB_CalcEnvironmentTexCoords( float *st )
{
int i;
float *v, *normal;
vec3_t viewer, reflected;
float d;
v = tess.xyz[0];
normal = tess.normal[0];
for (i = 0 ; i < tess.numVertexes ; i++, v += 4, normal += 4, st += 2 )
{
VectorSubtract (backEnd.ori.viewOrigin, v, viewer);
VectorNormalizeFast (viewer);
d = DotProduct (normal, viewer);
reflected[0] = normal[0]*2*d - viewer[0];
reflected[1] = normal[1]*2*d - viewer[1];
reflected[2] = normal[2]*2*d - viewer[2];
st[0] = 0.5 + reflected[1] * 0.5;
st[1] = 0.5 - reflected[2] * 0.5;
}
}
/*
** RB_CalcTurbulentTexCoords
*/
void RB_CalcTurbulentTexCoords( const waveForm_t *wf, float *st )
{
int i;
float now;
now = ( wf->phase + tess.shaderTime * wf->frequency );
for ( i = 0; i < tess.numVertexes; i++, st += 2 )
{
float s = st[0];
float t = st[1];
st[0] = s + tr.sinTable[ ( ( int ) ( ( ( tess.xyz[i][0] + tess.xyz[i][2] )* 1.0/128 * 0.125 + now ) * FUNCTABLE_SIZE ) ) & ( FUNCTABLE_MASK ) ] * wf->amplitude;
st[1] = t + tr.sinTable[ ( ( int ) ( ( tess.xyz[i][1] * 1.0/128 * 0.125 + now ) * FUNCTABLE_SIZE ) ) & ( FUNCTABLE_MASK ) ] * wf->amplitude;
}
}
/*
** RB_CalcScaleTexCoords
*/
void RB_CalcScaleTexCoords( const float scale[2], float *st )
{
int i;
for ( i = 0; i < tess.numVertexes; i++, st += 2 )
{
st[0] *= scale[0];
st[1] *= scale[1];
}
}
/*
** RB_CalcScrollTexCoords
*/
void RB_CalcScrollTexCoords( const float scrollSpeed[2], float *st )
{
int i;
float timeScale = tess.shaderTime;
float adjustedScrollS, adjustedScrollT;
adjustedScrollS = scrollSpeed[0] * timeScale;
adjustedScrollT = scrollSpeed[1] * timeScale;
// clamp so coordinates don't continuously get larger, causing problems
// with hardware limits
adjustedScrollS = adjustedScrollS - floor( adjustedScrollS );
adjustedScrollT = adjustedScrollT - floor( adjustedScrollT );
for ( i = 0; i < tess.numVertexes; i++, st += 2 )
{
st[0] += adjustedScrollS;
st[1] += adjustedScrollT;
}
}
/*
** RB_CalcTransformTexCoords
*/
void RB_CalcTransformTexCoords( const texModInfo_t *tmi, float *st )
{
int i;
for ( i = 0; i < tess.numVertexes; i++, st += 2 )
{
float s = st[0];
float t = st[1];
st[0] = s * tmi->matrix[0][0] + t * tmi->matrix[1][0] + tmi->translate[0];
st[1] = s * tmi->matrix[0][1] + t * tmi->matrix[1][1] + tmi->translate[1];
}
}
/*
** RB_CalcRotateTexCoords
*/
void RB_CalcRotateTexCoords( float degsPerSecond, float *st )
{
float timeScale = tess.shaderTime;
float degs;
int index;
float sinValue, cosValue;
texModInfo_t tmi;
degs = -degsPerSecond * timeScale;
index = degs * ( FUNCTABLE_SIZE / 360.0f );
sinValue = tr.sinTable[ index & FUNCTABLE_MASK ];
cosValue = tr.sinTable[ ( index + FUNCTABLE_SIZE / 4 ) & FUNCTABLE_MASK ];
tmi.matrix[0][0] = cosValue;
tmi.matrix[1][0] = -sinValue;
tmi.translate[0] = 0.5 - 0.5 * cosValue + 0.5 * sinValue;
tmi.matrix[0][1] = sinValue;
tmi.matrix[1][1] = cosValue;
tmi.translate[1] = 0.5 - 0.5 * sinValue - 0.5 * cosValue;
RB_CalcTransformTexCoords( &tmi, st );
}
#if id386 && !( (defined __linux__ || defined __FreeBSD__ ) && (defined __i386__ ) ) // rb010123
#pragma warning (disable: 4035)//no return value
long myftol( float f ) {
static int tmp;
__asm fld f
__asm fistp tmp
__asm mov eax, tmp
}
#pragma warning (default: 4035)
#endif
/*
** RB_CalcSpecularAlpha
**
** Calculates specular coefficient and places it in the alpha channel
*/
vec3_t lightOrigin = { -960, 1980, 96 }; // FIXME: track dynamically
void RB_CalcSpecularAlpha( unsigned char *alphas ) {
int i;
float *v, *normal;
vec3_t viewer, reflected;
float l, d;
int b;
vec3_t lightDir;
int numVertexes;
v = tess.xyz[0];
normal = tess.normal[0];
alphas += 3;
numVertexes = tess.numVertexes;
for (i = 0 ; i < numVertexes ; i++, v += 4, normal += 4, alphas += 4) {
float ilength;
if (backEnd.currentEntity &&
(backEnd.currentEntity->e.hModel||backEnd.currentEntity->e.ghoul2) ) //this is a model so we can use world lights instead fake light
{
VectorCopy (backEnd.currentEntity->lightDir, lightDir);
} else {
VectorSubtract( lightOrigin, v, lightDir );
VectorNormalizeFast( lightDir );
}
// calculate the specular color
d = 2 * DotProduct (normal, lightDir);
// we don't optimize for the d < 0 case since this tends to
// cause visual artifacts such as faceted "snapping"
reflected[0] = normal[0]*d - lightDir[0];
reflected[1] = normal[1]*d - lightDir[1];
reflected[2] = normal[2]*d - lightDir[2];
VectorSubtract (backEnd.ori.viewOrigin, v, viewer);
ilength = Q_rsqrt( DotProduct( viewer, viewer ) );
l = DotProduct (reflected, viewer);
l *= ilength;
if (l < 0) {
b = 0;
} else {
l = l*l;
l = l*l;
b = l * 255;
if (b > 255) {
b = 255;
}
}
*alphas = b;
}
}
/*
** RB_CalcDiffuseColor
**
** The basic vertex lighting calc
*/
void RB_CalcDiffuseColor( unsigned char *colors )
{
int i, j;
float *v, *normal;
float incoming;
trRefEntity_t *ent;
int ambientLightInt;
vec3_t ambientLight;
vec3_t lightDir;
vec3_t directedLight;
int numVertexes;
ent = backEnd.currentEntity;
ambientLightInt = ent->ambientLightInt;
VectorCopy( ent->ambientLight, ambientLight );
VectorCopy( ent->directedLight, directedLight );
VectorCopy( ent->lightDir, lightDir );
v = tess.xyz[0];
normal = tess.normal[0];
numVertexes = tess.numVertexes;
for (i = 0 ; i < numVertexes ; i++, v += 4, normal += 4) {
incoming = DotProduct (normal, lightDir);
if ( incoming <= 0 ) {
*(int *)&colors[i*4] = ambientLightInt;
continue;
}
j = myftol( ambientLight[0] + incoming * directedLight[0] );
if ( j > 255 ) {
j = 255;
}
colors[i*4+0] = j;
j = myftol( ambientLight[1] + incoming * directedLight[1] );
if ( j > 255 ) {
j = 255;
}
colors[i*4+1] = j;
j = myftol( ambientLight[2] + incoming * directedLight[2] );
if ( j > 255 ) {
j = 255;
}
colors[i*4+2] = j;
colors[i*4+3] = 255;
}
}
//---------------------------------------------------------
void RB_CalcDisintegrateColors( unsigned char *colors )
{
int i, numVertexes;
float dis, threshold;
float *v;
vec3_t temp;
refEntity_t *ent;
ent = &backEnd.currentEntity->e;
v = tess.xyz[0];
// calculate the burn threshold at the given time, anything that passes the threshold will get burnt
threshold = (backEnd.refdef.time - ent->endTime) * 0.045f; // endTime is really the start time, maybe I should just use a completely meaningless substitute?
numVertexes = tess.numVertexes;
if ( ent->renderfx & RF_DISINTEGRATE1 )
{
// this handles the blacken and fading out of the regular player model
for ( i = 0 ; i < numVertexes ; i++, v += 4 )
{
VectorSubtract( backEnd.currentEntity->e.oldorigin, v, temp );
dis = VectorLengthSquared( temp );
if ( dis < threshold * threshold )
{
// completely disintegrated
colors[i*4+3] = 0x00;
}
else if ( dis < threshold * threshold + 60 )
{
// blacken before fading out
colors[i*4+0] = 0x0;
colors[i*4+1] = 0x0;
colors[i*4+2] = 0x0;
colors[i*4+3] = 0xff;
}
else if ( dis < threshold * threshold + 150 )
{
// darken more
colors[i*4+0] = 0x6f;
colors[i*4+1] = 0x6f;
colors[i*4+2] = 0x6f;
colors[i*4+3] = 0xff;
}
else if ( dis < threshold * threshold + 180 )
{
// darken at edge of burn
colors[i*4+0] = 0xaf;
colors[i*4+1] = 0xaf;
colors[i*4+2] = 0xaf;
colors[i*4+3] = 0xff;
}
else
{
// not burning at all yet
colors[i*4+0] = 0xff;
colors[i*4+1] = 0xff;
colors[i*4+2] = 0xff;
colors[i*4+3] = 0xff;
}
}
}
else if ( ent->renderfx & RF_DISINTEGRATE2 )
{
// this handles the glowing, burning bit that scales away from the model
for ( i = 0 ; i < numVertexes ; i++, v += 4 )
{
VectorSubtract( backEnd.currentEntity->e.oldorigin, v, temp );
dis = VectorLengthSquared( temp );
if ( dis < threshold * threshold )
{
// done burning
colors[i*4+0] = 0x00;
colors[i*4+1] = 0x00;
colors[i*4+2] = 0x00;
colors[i*4+3] = 0x00;
}
else
{
// still full burn
colors[i*4+0] = 0xff;
colors[i*4+1] = 0xff;
colors[i*4+2] = 0xff;
colors[i*4+3] = 0xff;
}
}
}
}
//---------------------------------------------------------
void RB_CalcDisintegrateVertDeform( void )
{
float *xyz = ( float * ) tess.xyz;
float *normal = ( float * ) tess.normal;
float scale;
vec3_t temp;
if ( backEnd.currentEntity->e.renderfx & RF_DISINTEGRATE2 )
{
float threshold = (backEnd.refdef.time - backEnd.currentEntity->e.endTime) * 0.045f;
for ( int i = 0; i < tess.numVertexes; i++, xyz += 4, normal += 4 )
{
VectorSubtract( backEnd.currentEntity->e.oldorigin, xyz, temp );
scale = VectorLengthSquared( temp );
if ( scale < threshold * threshold )
{
xyz[0] += normal[0] * 2.0f;
xyz[1] += normal[1] * 2.0f;
xyz[2] += normal[2] * 0.5f;
}
else if ( scale < threshold * threshold + 50 )
{
xyz[0] += normal[0] * 1.0f;
xyz[1] += normal[1] * 1.0f;
// xyz[2] += normal[2] * 1;
}
}
}
}