fteqw/engine/gl/ltface.c
Spoike 9033f7b237 fixed eztv md4 incompatibility.
reimplemented qtvreverse command.
fixed some stuffcmds being handled by the wrong splitscreen seats (was noticable in TF).
rework smartjump to try to be more predictable...
rework relighting to try to be more robust (and more self-contained).
allow the csqc to actually use VF_PROJECTIONOFFSET.
jump now moves upwards instead of trying to lock on to a nearby player when spectating.
assume 32 fullbright pixels when running with a palette.lmp yet no colormap.lmp (happens with some total conversions).
tweaked scoreboard for fainter backgrounds.
rearranged autoid, to be smaller etc.
hacked around dodgy conchars.lmp - don't treat 128*128 qpics as qpics to work around workarounds for buggy wad tools (with a warning).
fixed missing fullbrights on h2holey models.
avoided warning about mod_h2holey_bugged on dedicated servers.
added net_ice_exchangeprivateips, for people worried about exposing lan IPs when using ICE.
sv_public 2: implemented client support for our webrtc broker in order to use our own ICE implementation without needing to faff around with irc accounts or plugins etc.
TODO: ensure at least one ephemerial udp port when using ice or come up with some better sv_port handling
fixed multiple tls bugs (one could cause server problems). change net_enable_tls to disabled by default anyway (reenable for the server to be able to respond to https/wss/tls schemes again).
don't colourmap when there appears to be a highres diffusemap on q1 models.
imgtool now understands exporting from qpics in wads, as well as just mips.
implemented speed-o-meter in ezhud.
added removeinstant builtin to avoid the half-second rule.


git-svn-id: https://svn.code.sf.net/p/fteqw/code/trunk@5614 fc73d0e0-1445-4013-8a0c-d673dee63da5
2020-02-11 18:06:10 +00:00

1241 lines
28 KiB
C

#include "quakedef.h"
#ifdef RUNTIMELIGHTING
typedef struct mentity_s {
vec3_t origin;
float light;
float angle;
float cone;
int style;
vec3_t colour;
char classname[64];
char target[64];
char targetname[64];
int targetentnum;
} mentity_t;
struct relight_ctx_s
{
unsigned int nummodels;
model_t *models[2048]; //0 is the worldmodel and must be valid
qboolean parsed; //ents have been parsed okay.
qboolean loaded; //needed models are all loaded.
float minlight;
qboolean skiplit; //lux only
qboolean shadows;
mentity_t *entities;
unsigned int num_entities;
unsigned int max_entities;
size_t lightmapsamples;
unsigned long nextface;
};
#define bsptexinfo(i) (*i)
#define dsurfedges lightmodel->surfedges
#define dvertexes lightmodel->vertexes
#define dedges lightmodel->edges
#define texinfo_t mtexinfo_t
#define Q_PI M_PI
#define dfaces lightmodel->surfaces
#define dplanes lightmodel->planes
#define dface_t msurface_t
#define dvertex_t mvertex_t
#define side flags & SURF_PLANEBACK
#define scaledist 1
#define rangescale 0.5
#define extrasamples 1
#define scalecos 0.5
#define bsp_origin vec3_origin
/*
============
CastRay
Returns the distance between the points, or -1 if blocked
=============
*/
static vec_t CastRay (struct relight_ctx_s *ctx, vec3_t p1, vec3_t p2)
{
trace_t trace;
vec3_t move;
if (ctx->shadows)
{
ctx->models[0]->funcs.NativeTrace (ctx->models[0], 0, NULLFRAMESTATE, NULL, p1, p2, vec3_origin, vec3_origin, false, FTECONTENTS_SOLID, &trace);
if (trace.fraction < 1)
return -1;
}
VectorSubtract(p1, p2, move);
return VectorLength(move);
}
static void ParseEpair (mentity_t *mapent, char *key, char *value)
{
double vec[3];
if (!strcmp(key, "classname"))
strcpy(mapent->classname, value);
else if (!strcmp(key, "target"))
strcpy(mapent->target, value);
else if (!strcmp(key, "targetname"))
strcpy(mapent->targetname, value);
else if (!strcmp(key, "light") || !strcmp(key, "_light"))
mapent->light = atoi(value);
else if (!strcmp(key, "style") || !strcmp(key, "_style"))
mapent->style = atoi(value);
else if (!strcmp(key, "angle") || !strcmp(key, "_angle"))
mapent->angle = atof(value);
else if (!strcmp(key, "cone") || !strcmp(key, "_cone"))
mapent->cone = atof(value);
else if (!strcmp(key, "origin"))
{
sscanf (value, "%lf %lf %lf", &vec[0], &vec[1], &vec[2]);
mapent->origin[0]=vec[0];
mapent->origin[1]=vec[1];
mapent->origin[2]=vec[2];
}
else if (!strcmp(key, "colour") || !strcmp(key, "color") || !strcmp(key, "_colour") || !strcmp(key, "_color"))
{
sscanf (value, "%lf %lf %lf", &vec[0], &vec[1], &vec[2]);
mapent->colour[0]=vec[0];
mapent->colour[1]=vec[1];
mapent->colour[2]=vec[2];
}
}
void LightShutdown(struct relight_ctx_s *ctx)
{
Z_Free(ctx->entities);
Z_Free(ctx);
}
struct relight_ctx_s *LightStartup(struct relight_ctx_s *ctx, model_t *model, qboolean shadows, qboolean skiplit)
{
if (!ctx)
{
ctx = Z_Malloc(sizeof(*ctx));
ctx->shadows = shadows;
ctx->skiplit = skiplit;
}
if (ctx->nummodels < countof(ctx->models))
ctx->models[ctx->nummodels++] = model;
return ctx;
}
void LightReloadEntities(struct relight_ctx_s *ctx, const char *entstring, qboolean ignorestyles)
{
#define DEFAULTLIGHTLEVEL 300
mentity_t *mapent;
char key[1024];
char value[1024];
int i;
int switchedstyle=32;
ctx->num_entities = 0;
while(1)
{
entstring = COM_ParseOut(entstring, key, sizeof(key));
if (!entstring || !*key)
break;
if (strcmp(key, "{"))
{ //someone messed up. Stop parsing.
Con_Printf("token wasn't an open brace\n");
break;
}
if (ctx->num_entities == ctx->max_entities)
{
ctx->max_entities = ctx->max_entities + 128;
ctx->entities = BZ_Realloc(ctx->entities, sizeof(*ctx->entities) * ctx->max_entities);
}
mapent = &ctx->entities[ctx->num_entities];
memset(mapent, 0, sizeof(*mapent));
mapent->colour[0] = 0;
mapent->colour[1] = 0;
mapent->colour[2] = 0;
mapent->targetentnum = -1;
while(entstring)
{
entstring = COM_ParseOut(entstring, key, sizeof(key));
if (!strcmp(key, "}"))
break;
entstring = COM_ParseOut(entstring, value, sizeof(value));
ParseEpair(mapent, key, value);
}
if (!mapent->colour[0] && !mapent->colour[1] && !mapent->colour[2])
{
int cont;
vec3_t v;
v[0] = mapent->origin[0];
v[1] = mapent->origin[1];
cont=0;
for (i = 0; i < 256; i+=16)
{
v[2] = mapent->origin[2]-i;
cont = ctx->models[0]->funcs.PointContents (ctx->models[0], NULL, v);
if (cont & (FTECONTENTS_LAVA | FTECONTENTS_SLIME | FTECONTENTS_SOLID))
break;
}
if (cont & FTECONTENTS_LAVA)
{
mapent->colour[0] = 1;
mapent->colour[1] = i/256.0;
mapent->colour[2] = i/256.0;
}
else if (cont & FTECONTENTS_SLIME)
{
mapent->colour[0] = 0.5+0.5*i/256.0;
mapent->colour[1] = 1;
mapent->colour[2] = 0.5+0.5*i/256.0;
}
else
{
if (mapent->style == 9) //hmm..
{
mapent->colour[1] = 1;
}
else
{
if (!strncmp(mapent->classname, "light_torch_small_walltorch", 12))
{
mapent->colour[0] = 1;
mapent->colour[1] = 0.7;
mapent->colour[2] = 0.7;
}
else
{
mapent->colour[0] = 1;
mapent->colour[1] = 1;
if (strncmp(mapent->classname, "light_fluoro", 12))
mapent->colour[2] = 1;
}
}
}
}
if (!mapent->light && !strncmp (mapent->classname, "light", 5))
mapent->light = DEFAULTLIGHTLEVEL;
if (*mapent->targetname && !mapent->style && !strcmp(mapent->classname, "light"))
{
for (i = 0; i <= ctx->num_entities; i++)
{
if (ctx->entities[i].style >= 32 && !strcmp(ctx->entities[i].targetname, mapent->targetname))
{
mapent->style = ctx->entities[i].style;
break;
}
}
if (i == ctx->num_entities)
mapent->style = switchedstyle++;
}
if (ignorestyles)
mapent->style = 0;
ctx->num_entities++;
}
if (ctx->num_entities)
if (ctx->entities[0].light)
ctx->minlight = ctx->entities[0].light;
for (mapent = ctx->entities; mapent < &ctx->entities[ctx->num_entities]; mapent++)
{
if (*mapent->target)
{
for (i = 0; i < ctx->num_entities; i++)
{
if (mapent == &ctx->entities[i])
continue;
if (!strcmp(mapent->target, ctx->entities[i].targetname))
{
mapent->targetentnum = i;
break;
}
}
}
}
}
/*
===============================================================================
SAMPLE POINT DETERMINATION
void SetupBlock (dface_t *f) Returns with surfpt[] set
This is a little tricky because the lightmap covers more area than the face.
If done in the straightforward fashion, some of the
sample points will be inside walls or on the other side of walls, causing
false shadows and light bleeds.
To solve this, I only consider a sample point valid if a line can be drawn
between it and the exact midpoint of the face. If invalid, it is adjusted
towards the center until it is valid.
(this doesn't completely work)
===============================================================================
*/
#define MAXIMUMEXTENT 128
#define SINGLEMAP (MAXIMUMEXTENT*MAXIMUMEXTENT*4)
typedef struct llightinfo_s
{
struct relight_ctx_s *ctx; //relight context, shared between threads.
vec3_t lightmaps[MAXQ1LIGHTMAPS][SINGLEMAP];
vec3_t lightnorm[MAXQ1LIGHTMAPS][SINGLEMAP];
int numlightstyles;
vec_t *light;
vec_t facedist;
vec3_t facenormal;
int numsurfpt;
vec3_t surfpt[SINGLEMAP];
vec3_t texorg;
vec3_t worldtotex[2]; // s = (world - texorg) . worldtotex[0]
vec3_t textoworld[2]; // world = texorg + s * textoworld[0]
vec_t exactmins[2], exactmaxs[2];
int texmins[2], texsize[2];
int lightstyles[MAXQ1LIGHTMAPS];
} llightinfo_t;
const size_t lightthreadctxsize = sizeof(llightinfo_t);
/*
================
CalcFaceVectors
Fills in texorg, worldtotex. and textoworld
================
*/
static void LightCalcFaceVectors (llightinfo_t *l, vec4_t surf_texplanes[2])
{
int i, j;
vec3_t texnormal;
float distscale;
vec_t dist, len;
// convert from float to vec_t
for (i=0 ; i<2 ; i++)
for (j=0 ; j<3 ; j++)
l->worldtotex[i][j] = surf_texplanes[i][j];
// calculate a normal to the texture axis. points can be moved along this
// without changing their S/T
texnormal[0] = surf_texplanes[1][1]*surf_texplanes[0][2]
- surf_texplanes[1][2]*surf_texplanes[0][1];
texnormal[1] = surf_texplanes[1][2]*surf_texplanes[0][0]
- surf_texplanes[1][0]*surf_texplanes[0][2];
texnormal[2] = surf_texplanes[1][0]*surf_texplanes[0][1]
- surf_texplanes[1][1]*surf_texplanes[0][0];
VectorNormalize (texnormal);
// flip it towards plane normal
distscale = DotProduct (texnormal, l->facenormal);
if (!distscale)
{
VectorCopy(l->facenormal, texnormal);
distscale = 1;
Con_Printf ("Texture axis perpendicular to face\n");
}
if (distscale < 0)
{
distscale = -distscale;
VectorNegate (texnormal, texnormal);
}
// distscale is the ratio of the distance along the texture normal to
// the distance along the plane normal
distscale = 1/distscale;
for (i=0 ; i<2 ; i++)
{
len = VectorLength (l->worldtotex[i]);
dist = DotProduct (l->worldtotex[i], l->facenormal);
dist *= distscale;
VectorMA (l->worldtotex[i], -dist, texnormal, l->textoworld[i]);
VectorScale (l->textoworld[i], (1/len)*(1/len), l->textoworld[i]);
}
// calculate texorg on the texture plane
for (i=0 ; i<3 ; i++)
l->texorg[i] = -surf_texplanes[0][3]* l->textoworld[0][i] - surf_texplanes[1][3] * l->textoworld[1][i];
// project back to the face plane
dist = DotProduct (l->texorg, l->facenormal) - l->facedist - 1;
dist *= distscale;
VectorMA (l->texorg, -dist, texnormal, l->texorg);
}
/*
================
CalcFaceExtents
Fills in s->texmins[] and s->texsize[]
also sets exactmins[] and exactmaxs[]
================
*/
static void LightCalcFaceExtents (model_t *lightmodel, dface_t *s, vec2_t exactmins, vec2_t exactmaxs, int texmins[2], int texsize[2])
{
vec_t mins[2], maxs[2], val;
int i,j, e;
dvertex_t *v;
texinfo_t *tex;
mins[0] = mins[1] = 999999;
maxs[0] = maxs[1] = -999999;
tex = &bsptexinfo(s->texinfo);
for (i=0 ; i<s->numedges ; i++)
{
e = dsurfedges[s->firstedge+i];
if (e >= 0)
v = dvertexes + dedges[e].v[0];
else
v = dvertexes + dedges[-e].v[1];
for (j=0 ; j<2 ; j++)
{
val = v->position[0] * tex->vecs[j][0] +
v->position[1] * tex->vecs[j][1] +
v->position[2] * tex->vecs[j][2] +
tex->vecs[j][3];
if (val < mins[j])
mins[j] = val;
if (val > maxs[j])
maxs[j] = val;
}
}
for (i=0 ; i<2 ; i++)
{
exactmins[i] = mins[i];
exactmaxs[i] = maxs[i];
mins[i] = floor(mins[i]/(1<<s->lmshift));
maxs[i] = ceil(maxs[i]/(1<<s->lmshift));
texmins[i] = mins[i];
texsize[i] = maxs[i] - mins[i];
if (texsize[i] > MAXIMUMEXTENT-1)
{
texsize[i] = MAXIMUMEXTENT-1;
Con_Printf("Bad surface extents");
}
}
}
/*
=================
CalcPoints
For each texture aligned grid point, back project onto the plane
to get the world xyz value of the sample point
=================
*/
static void LightCalcPoints (llightinfo_t *l, float lmscale)
{
int i;
int s, t, j;
int w, h;
vec_t step;
vec_t starts, startt, us, ut;
vec_t *surf;
vec_t mids, midt;
vec3_t facemid, move;
//
// fill in surforg
// the points are biased towards the center of the surface
// to help avoid edge cases just inside walls
//
surf = l->surfpt[0];
mids = (l->exactmaxs[0] + l->exactmins[0])/2;
midt = (l->exactmaxs[1] + l->exactmins[1])/2;
for (j=0 ; j<3 ; j++)
facemid[j] = l->texorg[j] + l->textoworld[0][j]*mids + l->textoworld[1][j]*midt;
if (extrasamples)
{ // extra filtering
h = (l->texsize[1]+1)*2;
w = (l->texsize[0]+1)*2;
starts = (l->texmins[0]-0.5)*lmscale;
startt = (l->texmins[1]-0.5)*lmscale;
step = 0.5 * lmscale;
}
else
{
h = l->texsize[1]+1;
w = l->texsize[0]+1;
starts = l->texmins[0]*lmscale;
startt = l->texmins[1]*lmscale;
step = lmscale;
}
l->numsurfpt = w * h;
for (t=0 ; t<h ; t++)
{
for (s=0 ; s<w ; s++, surf+=3)
{
us = starts + s*step;
ut = startt + t*step;
// if a line can be traced from surf to facemid, the point is good
for (i=0 ; i<6 ; i++)
{
// calculate texture point
for (j=0 ; j<3 ; j++)
surf[j] = l->texorg[j] + l->textoworld[0][j]*us
+ l->textoworld[1][j]*ut;
if (CastRay (l->ctx, facemid, surf) != -1)
break; // got it
if (i & 1)
{
if (us > mids)
{
us -= lmscale*0.5;
if (us < mids)
us = mids;
}
else
{
us += lmscale*0.5;
if (us > mids)
us = mids;
}
}
else
{
if (ut > midt)
{
ut -= lmscale*0.5;
if (ut < midt)
ut = midt;
}
else
{
ut += lmscale*0.5;
if (ut > midt)
ut = midt;
}
}
// move surf 8 pixels towards the center
VectorSubtract (facemid, surf, move);
VectorNormalize (move);
VectorMA (surf, 8, move, surf);
}
}
}
}
/*
===============================================================================
FACE LIGHTING
===============================================================================
*/
/*
================
SingleLightFace
================
*/
static void SingleLightFace (mentity_t *light, llightinfo_t *l)
{
vec_t dist;
vec3_t incoming;
vec_t angle;
vec_t add;
vec_t *surf;
qboolean hit;
int mapnum;
int c;
vec3_t rel;
vec3_t spotvec;
vec_t falloff;
vec3_t *lightsamp;
vec3_t *norms;
VectorSubtract (light->origin, bsp_origin, rel);
dist = scaledist * (DotProduct (rel, l->facenormal) - l->facedist);
// don't bother with lights behind the surface
if (dist <= 0)
return;
// don't bother with light too far away
if (dist > light->light)
return;
if (light->targetentnum>=0)
{
VectorSubtract (l->ctx->entities[light->targetentnum].origin, light->origin, spotvec);
VectorNormalize (spotvec);
if (!light->angle)
falloff = -cos(20*Q_PI/180);
else
falloff = -cos(light->angle/2*Q_PI/180);
}
else
falloff = 0; // shut up compiler warnings
mapnum = 0;
for (mapnum=0 ; mapnum<l->numlightstyles ; mapnum++)
if (l->lightstyles[mapnum] == light->style)
break;
lightsamp = l->lightmaps[mapnum];
norms = l->lightnorm[mapnum];
if (mapnum == l->numlightstyles)
{ // init a new light map
#ifdef UTILITY
if (mapnum == MAXQ1LIGHTMAPS)
{
printf ("WARNING: Too many light styles on a face\n");
return;
}
size = (l->texsize[1]+1)*(l->texsize[0]+1);
for (i=0 ; i<size ; i++)
{
lightsamp[i][0] = 0;
lightsamp[i][1] = 0;
lightsamp[i][2] = 0;
norms[i][0] = 0;
norms[i][1] = 0;
norms[i][2] = 0;
}
#else
return; //can't light a surface with a lightstyle that did not previously exist, due to lightmap space limits.
#endif
}
//
// check it for real
//
hit = false;
surf = l->surfpt[0];
for (c=0 ; c<l->numsurfpt ; c++, surf+=3)
{
dist = CastRay(l->ctx, light->origin, surf)*scaledist;
if (dist < 0)
continue; // light doesn't reach
VectorSubtract (light->origin, surf, incoming);
VectorNormalize (incoming);
if (light->targetentnum >= 0)
{ // spotlight cutoff
if (DotProduct (spotvec, incoming) > falloff)
continue;
}
angle = DotProduct (incoming, l->facenormal);
angle = (1.0-scalecos) + scalecos*angle;
add = light->light - dist;
add *= angle;
if (add < 0)
continue;
lightsamp[c][0] += add*light->colour[0];
lightsamp[c][1] += add*light->colour[1];
lightsamp[c][2] += add*light->colour[2];
norms[c][0] += add * incoming[0];
norms[c][1] += add * incoming[1];
norms[c][2] += add * incoming[2];
if (add > 1) // ignore real tiny lights
hit = true;
}
if (mapnum == l->numlightstyles && hit)
{
l->lightstyles[mapnum] = light->style;
l->numlightstyles++; // the style has some real data now
}
}
/*
============
FixMinlight
============
*/
static void FixMinlight (llightinfo_t *l)
{
int i, j;
float minlight = l->ctx->minlight;
// if minlight is set, there must be a style 0 light map
if (!minlight)
return;
for (i=0 ; i< l->numlightstyles ; i++)
{
if (l->lightstyles[i] == 0)
break;
}
if (i == l->numlightstyles)
{
if (l->numlightstyles == MAXQ1LIGHTMAPS)
return; // oh well..
for (j=0 ; j<l->numsurfpt ; j++)
{
l->lightmaps[i][j][0] = minlight;
l->lightmaps[i][j][1] = minlight;
l->lightmaps[i][j][2] = minlight;
}
l->lightstyles[i] = 0;
l->numlightstyles++;
}
else
{
for (j=0 ; j<l->numsurfpt ; j++)
{
if ( l->lightmaps[i][j][0] < minlight)
l->lightmaps[i][j][0] = minlight;
if ( l->lightmaps[i][j][1] < minlight)
l->lightmaps[i][j][1] = minlight;
if ( l->lightmaps[i][j][2] < minlight)
l->lightmaps[i][j][2] = minlight;
}
}
}
static unsigned int PackE5BRG9(vec3_t rgb)
{ //5 bits exponent, 3*9 bits of mantissa. no sign bit.
int e = 0;
float m = max(max(rgb[0], rgb[1]), rgb[2]);
float scale;
unsigned int hdr;
if (m >= 0.5)
{ //positive exponent
while (m >= (1<<(e)) && e < 30-15) //don't do nans.
e++;
}
else
{ //negative exponent...
while (m < 1/(1<<-e) && e > -15) //don't do denormals.
e--;
}
scale = pow(2, e-9);
hdr = ((e+15)<<27);
hdr |= bound(0, (int)(rgb[0]/scale + 0.5), 0x1ff)<<0;
hdr |= bound(0, (int)(rgb[1]/scale + 0.5), 0x1ff)<<9;
hdr |= bound(0, (int)(rgb[2]/scale + 0.5), 0x1ff)<<18;
return hdr;
}
/*
============
LightFace
============
*/
void LightPlane (struct relight_ctx_s *ctx, struct llightinfo_s *l, lightstyleindex_t surf_styles[MAXQ1LIGHTMAPS], unsigned int *surf_expsamples, qbyte *surf_rgbsamples, qbyte *surf_deluxesamples, vec4_t surf_plane, vec4_t surf_texplanes[2], vec2_t exactmins, vec2_t exactmaxs, int texmins[2], int texsize[2], float lmscale)
{
int s, t;
int i,c,ch;
vec_t total, mean;
int size;
int lightmapwidth;
#ifdef UTILITY
int lightmapsize;
byte *out;
#endif
unsigned int *expout;
qbyte *rgbout;
qbyte *dulout;
vec3_t *light, *norm;
vec3_t wnorm, temp, svector, tvector;
int w;
//
// some surfaces don't need lightmaps
//
if (!surf_rgbsamples && !surf_expsamples)
return;
// memset (l, 0, sizeof(*l));
l->ctx = ctx;
//
// rotate plane
//
VectorCopy (surf_plane, l->facenormal);
l->facedist = surf_plane[3];
LightCalcFaceVectors (l, surf_texplanes);
Vector2Copy(exactmins, l->exactmins);
Vector2Copy(exactmaxs, l->exactmaxs);
Vector2Copy(texmins, l->texmins);
Vector2Copy(texsize, l->texsize);
LightCalcPoints (l, lmscale);
lightmapwidth = l->texsize[0]+1;
size = lightmapwidth*(l->texsize[1]+1);
if (size > SINGLEMAP)
Host_Error ("Bad lightmap size");
i = 0;
#ifndef UTILITY
for (; surf_styles[i] != INVALID_LIGHTSTYLE && i < MAXQ1LIGHTMAPS; i++)
{
l->lightstyles[i] = surf_styles[i];
memset(&l->lightmaps[i], 0, sizeof(l->lightmaps[i][0])*l->numsurfpt);
memset(&l->lightnorm[i], 0, sizeof(l->lightnorm[i][0])*l->numsurfpt);
}
#endif
l->numlightstyles = i;
for ( ; i<MAXQ1LIGHTMAPS ; i++)
l->lightstyles[i] = INVALID_LIGHTSTYLE;
//
// cast all lights
//
for (i=0 ; i<ctx->num_entities ; i++)
{
if (ctx->entities[i].light)
SingleLightFace (&ctx->entities[i], l);
}
FixMinlight (l);
if (!l->numlightstyles)
{ // no light hitting it
#ifdef UTILITY
f->lightofs = -1;
#endif
return;
}
//
// save out the values
//
for (i=0 ; i <MAXQ1LIGHTMAPS ; i++)
surf_styles[i] = l->lightstyles[i];
#ifdef UTILITY
lightmapsize = size*l->numlightstyles;
if (runningrgblightdatabase)
{
out = GetFakeFileSpace(&f->lightofs, lightmapsize);
expout = NULL;
rgbout = runningrgblightdatabase + f->lightofs*3;
dulout = runninglightnormbase + f->lightofs*3;
}
else
{
out = GetFileSpace (&f->lightofs, lightmapsize);
expout = NULL;
rgbout = GetRGBFileSpace (f->lightofs, lightmapsize);
dulout = GetNormFileSpace (f->lightofs, lightmapsize);
}
#else
if (!ctx->skiplit)
{
expout = surf_expsamples;
rgbout = surf_rgbsamples;
}
else
{
expout = NULL;
rgbout = NULL;
}
if (l->ctx->models[0]->deluxdata)
{
dulout = surf_deluxesamples;
VectorCopy(surf_texplanes[0], svector);
VectorNegate(surf_texplanes[1], tvector);
VectorNormalize(svector);
VectorNormalize(tvector);
}
else
dulout = NULL;
#endif
// extra filtering
// h = (l.texsize[1]+1)*2;
w = l->texsize[0]+1;
if (extrasamples)
w *= 2;
for (i=0 ; i< l->numlightstyles ; i++)
{
if (l->lightstyles[i] == 0xff)
Host_Error ("Wrote empty lightmap");
light = l->lightmaps[i];
norm = l->lightnorm[i];
c = 0;
for (t=0 ; t<=l->texsize[1] ; t++)
{
for (s=0 ; s<=l->texsize[0] ; s++, c++)
{
mean = 0;
for (ch = 0; ch < 3; ch++)
{
if (extrasamples)
{ // filtered sample
total = light[t*2*w+s*2][ch] + light[t*2*w+s*2+1][ch]
+ light[(t*2+1)*w+s*2][ch] + light[(t*2+1)*w+s*2+1][ch];
total *= 0.25;
wnorm[ch] = norm[t*2*w+s*2][ch] + norm[t*2*w+s*2+1][ch]
+ norm[(t*2+1)*w+s*2][ch] + norm[(t*2+1)*w+s*2+1][ch];
}
else
{
total = light[c][ch];
wnorm[ch] = norm[c][ch];
}
total *= rangescale; // scale before clamping
temp[ch] = total/0x80; // quake bsps store logical light values between 0 and 2 for overbrights. normalise it appropriately.
#ifndef UTILITY
// if (total > *rgbout) //sorry - for qw
// total = *rgbout;
#endif
if (total < 0)
total = 0;
if (total > 0xff)
total = 0xff;
if (rgbout)
*rgbout++ = total;
mean += total;
}
if (expout)
*expout++ = PackE5BRG9(temp);
#ifdef UTILITY
*out++ = mean/3;
#endif
if (dulout)
{
temp[0] = DotProduct(wnorm, svector);
temp[1] = DotProduct(wnorm, tvector);
temp[2] = DotProduct(wnorm, l->facenormal);
if (!temp[0] && !temp[1] && !temp[2])
VectorSet(temp, 0, 0, 1);
else
VectorNormalize(temp);
*dulout++ = (temp[0]+1)*127;
*dulout++ = (temp[1]+1)*127;
*dulout++ = (temp[2]+1)*127;
}
}
}
}
}
static void LightFace (struct relight_ctx_s *ctx, struct llightinfo_s *threadctx, int facenum)
{
dface_t *f = ctx->models[0]->surfaces + facenum;
vec4_t plane;
vec2_t exactmins;
vec2_t exactmaxs;
int texmins[2], texsize[2];
VectorCopy (f->plane->normal, plane);
plane[3] = f->plane->dist;
if (f->flags & SURF_PLANEBACK)
{
VectorNegate (plane, plane);
plane[3] = -plane[3];
}
//no lighting on these.
if (f->texinfo->flags & TEX_SPECIAL)
return;
LightCalcFaceExtents(ctx->models[0], f, exactmins, exactmaxs, texmins, texsize);
if (ctx->models[0]->lightmaps.fmt == LM_E5BGR9)
LightPlane(ctx, threadctx, f->styles, (unsigned int*)f->samples, NULL, 3*(f->samples - ctx->models[0]->lightdata)/4 + ctx->models[0]->deluxdata, plane, f->texinfo->vecs, exactmins, exactmaxs, texmins, texsize, 1<<f->lmshift);
else
LightPlane(ctx, threadctx, f->styles, NULL, f->samples, f->samples - ctx->models[0]->lightdata + ctx->models[0]->deluxdata, plane, f->texinfo->vecs, exactmins, exactmaxs, texmins, texsize, 1<<f->lmshift);
}
#if defined(MULTITHREAD)
#ifdef _WIN32
#include <windows.h>
#elif defined(__linux__)
#include <unistd.h>
#endif
static void *relightthread[8];
static unsigned int relightthreads;
static volatile qboolean wantrelight;
static struct relight_ctx_s *lightcontext;
static int RelightThread(void *ctx)
{
int surf;
struct relight_ctx_s *lightcontext = ctx;
model_t *lightmodel = lightcontext->models[0];
void *threadctx = malloc(lightthreadctxsize);
while (wantrelight)
{
#ifdef _WIN32
surf = InterlockedIncrement(&lightcontext->nextface);
#elif defined(__GNUC__)
surf = __sync_add_and_fetch(&lightcontext->nextface, 1);
#else
surf = relitsurface++;
#endif
if (surf >= lightmodel->numsurfaces)
break;
LightFace(lightcontext, threadctx, surf);
lightmodel->surfaces[surf].cached_dlight = -1; //invalidate it (slightly racey buy w/e
}
free(threadctx);
return 0;
}
#else
static void *lightmainthreadctx;
#endif
void RelightTerminate(model_t *mod)
{
model_t *lightmodel;
size_t u;
if (!lightcontext)
return;
//if one of the models we're using is being purged then we have to abort the relight to avoid caching partial results (especially if its the model we're actually relighting)
if (mod)
{
for (u = 0; u < lightcontext->nummodels; u++)
if (lightcontext->models[u] == mod)
break;
}
else
u = 0;
if (u < lightcontext->nummodels)
{
lightmodel = lightcontext->models[0];
#ifdef MULTITHREAD
wantrelight = false;
if (relightthreads)
{
int i;
wantrelight = false;
for (i = 0; i < relightthreads; i++)
{
Sys_WaitOnThread(relightthread[i]);
relightthread[i] = NULL;
}
relightthreads = 0;
}
#else
free(lightmainthreadctx);
lightmainthreadctx = NULL;
#endif
if (lightcontext->nextface >= lightmodel->numsurfaces)
{
vfsfile_t *f;
char filename[MAX_QPATH];
if (lightmodel->deluxdata)
{
COM_StripExtension(lightmodel->name, filename, sizeof(filename));
COM_DefaultExtension(filename, ".lux", sizeof(filename));
f = FS_OpenVFS(filename, "wb", FS_GAME);
if (f)
{
VFS_WRITE(f, "QLIT\1\0\0\0", 8);
VFS_WRITE(f, lightmodel->deluxdata, lightcontext->lightmapsamples*3);
VFS_CLOSE(f);
}
else
Con_Printf("Unable to write \"%s\"\n", filename);
}
if (!lightcontext->skiplit) //the user might already have a lit file (don't overwrite it).
{
COM_StripExtension(lightmodel->name, filename, sizeof(filename));
COM_DefaultExtension(filename, ".lit", sizeof(filename));
f = FS_OpenVFS(filename, "wb", FS_GAME);
if (f)
{
if (lightmodel->lightmaps.fmt == LM_E5BGR9)
{
VFS_WRITE(f, "QLIT\x01\0\x01\0", 8);
VFS_WRITE(f, lightmodel->lightdata, lightcontext->lightmapsamples*4);
}
else
{
VFS_WRITE(f, "QLIT\1\0\0\0", 8);
VFS_WRITE(f, lightmodel->lightdata, lightcontext->lightmapsamples*3);
}
VFS_CLOSE(f);
}
else
Con_Printf("Unable to write \"%s\"\n", filename);
}
}
else
Con_Printf("Relighting aborted before completion\n");
LightShutdown(lightcontext);
lightcontext = NULL;
}
}
qboolean RelightSetup (model_t *model, size_t lightsamples, qboolean generatelit)
{
Sys_LockMutex(com_resourcemutex); //models can be loaded on different threads, so no race conditions please.
if (!lightcontext)
{
lightcontext = LightStartup(NULL, model, true, !generatelit);
lightcontext->lightmapsamples = lightsamples;
Sys_UnlockMutex(com_resourcemutex);
return true;
}
Sys_UnlockMutex(com_resourcemutex);
return false;
}
const char *RelightGetProgress(float *progress)
{
char filename[MAX_QPATH];
if (!lightcontext)
return NULL;
*progress = (lightcontext->nextface*100.0f) / lightcontext->models[0]->numsurfaces;
COM_StripExtension(lightcontext->models[0]->name, filename, sizeof(filename));
COM_DefaultExtension(filename, lightcontext->skiplit?".lux":".lit", sizeof(filename));
return va("%s", filename);
}
void RelightThink (void)
{
if (lightcontext)
{
model_t *lightmodel = lightcontext->models[0];
if (!lightcontext->loaded)
{ //make sure everything finished loading properly before we start poking things.
size_t u;
if (!lightcontext->parsed)
{
if (lightcontext->models[0]->loadstate != MLS_LOADED)
return; //not ready yet...
LightReloadEntities(lightcontext, Mod_GetEntitiesString(lightmodel), false);
lightcontext->parsed = true;
}
for (u = 0; u < lightcontext->nummodels; u++)
if (lightcontext->models[u]->loadstate != MLS_LOADED)
return;
lightcontext->loaded = true;
}
#ifdef MULTITHREAD
if (!relightthreads)
{
int i;
#if defined(_WIN32) && !defined(WINRT)
HANDLE me = GetCurrentProcess();
DWORD_PTR proc, sys;
/*count cpus*/
GetProcessAffinityMask(me, &proc, &sys);
relightthreads = 0;
for (i = 0; i < sizeof(proc)*8; i++)
if (proc & ((size_t)1u<<i))
relightthreads++;
/*subtract 1*/
if (relightthreads <= 1)
relightthreads = 1;
else
relightthreads--;
#elif defined(__GNUC__)
#ifdef __linux__
relightthreads = sysconf(_SC_NPROCESSORS_ONLN)-1;
if (relightthreads < 1)
relightthreads = 1;
#else
relightthreads = 2; //erm, lets hope...
#endif
#else
/*can't do atomics*/
relightthreads = 1;
#endif
if (relightthreads > sizeof(relightthread)/sizeof(relightthread[0]))
relightthreads = sizeof(relightthread)/sizeof(relightthread[0]);
wantrelight = true;
for (i = 0; i < relightthreads; i++)
relightthread[i] = Sys_CreateThread("relight", RelightThread, lightcontext, THREADP_NORMAL, 0);
}
if (lightcontext->nextface < lightmodel->numsurfaces)
{
return;
}
#else
if (!lightmainthreadctx)
lightmainthreadctx = malloc(lightthreadctxsize);
LightFace(lightcontext, lightmainthreadctx, relitsurface);
lightmodel->surfaces[relitsurface].cached_dlight = -1;
relitsurface++;
#endif
if (lightcontext->nextface >= lightmodel->numsurfaces)
{
Con_Printf("Finished lighting %s\n", lightmodel->name);
RelightTerminate(lightmodel);
}
}
}
#endif