#include "quakedef.h" /* room for improvement: There is no screen-space culling of lit surfaces. model meshes are interpolated multiple times per frame */ #if defined(RTLIGHTS) && !defined(SERVERONLY) #ifdef VKQUAKE #include "../vk/vkrenderer.h" #endif #include "glquake.h" #include "shader.h" #ifdef D3D9QUAKE #include "shader.h" #if !defined(HMONITOR_DECLARED) && (WINVER < 0x0500) #define HMONITOR_DECLARED DECLARE_HANDLE(HMONITOR); #endif #include extern LPDIRECT3DDEVICE9 pD3DDev9; void D3D9BE_Cull(unsigned int sflags); void D3D9BE_RenderShadowBuffer(unsigned int numverts, IDirect3DVertexBuffer9 *vbuf, unsigned int numindicies, IDirect3DIndexBuffer9 *ibuf); #endif #ifdef D3D11QUAKE void D3D11BE_GenerateShadowBuffer(void **vbuf, vecV_t *verts, int numverts, void **ibuf, index_t *indicies, int numindicies); void D3D11BE_RenderShadowBuffer(unsigned int numverts, void *vbuf, unsigned int numindicies, void *ibuf); void D3D11_DestroyShadowBuffer(void *vbuf, void *ibuf); void D3D11BE_DoneShadows(void); #endif #ifdef VKQUAKE #endif void GLBE_RenderShadowBuffer(unsigned int numverts, int vbo, vecV_t *verts, unsigned numindicies, int ibo, index_t *indicies); static void SHM_Shutdown(void); #define SHADOWMAP_SIZE 512 #define PROJECTION_DISTANCE (float)(dl->radius*2)//0x7fffffff #ifdef BEF_PUSHDEPTH extern qboolean r_pushdepth; #endif texid_t crepuscular_texture_id; fbostate_t crepuscular_fbo; shader_t *crepuscular_shader; cvar_t r_shadow_shadowmapping_nearclip = CVAR("r_shadow_shadowmapping_nearclip", "1"); cvar_t r_shadow_shadowmapping_bias = CVAR("r_shadow_shadowmapping_bias", "0.03"); cvar_t r_shadow_scissor = CVARD("r_shadow_scissor", "1", "constrains stencil shadows to the onscreen box that contains the maxmium extents of the light. This avoids unnecessary work."); cvar_t r_shadow_realtime_world = CVARFD ("r_shadow_realtime_world", "0", CVAR_ARCHIVE, "Enables the use of static/world realtime lights."); cvar_t r_shadow_realtime_world_shadows = CVARF ("r_shadow_realtime_world_shadows", "1", CVAR_ARCHIVE); cvar_t r_shadow_realtime_world_lightmaps = CVARFD ("r_shadow_realtime_world_lightmaps", "0", 0, "Specifies how much of the map's normal lightmap to retain when using world realtime lights. 0 completely replaces lighting."); cvar_t r_shadow_realtime_dlight = CVARFD ("r_shadow_realtime_dlight", "1", CVAR_ARCHIVE, "Enables the use of dynamic realtime lights, allowing explosions to use bumpmaps etc properly."); cvar_t r_shadow_realtime_dlight_shadows = CVARFD ("r_shadow_realtime_dlight_shadows", "1", CVAR_ARCHIVE, "Allows dynamic realtime lights to cast shadows as they move."); cvar_t r_shadow_realtime_dlight_ambient = CVAR ("r_shadow_realtime_dlight_ambient", "0"); cvar_t r_shadow_realtime_dlight_diffuse = CVAR ("r_shadow_realtime_dlight_diffuse", "1"); cvar_t r_shadow_realtime_dlight_specular = CVAR ("r_shadow_realtime_dlight_specular", "4"); //excessive, but noticable. its called stylized, okay? shiesh, some people cvar_t r_shadow_playershadows = CVARD ("r_shadow_playershadows", "1", "Controls the presence of shadows on the local player."); cvar_t r_shadow_shadowmapping = CVARD ("r_shadow_shadowmapping", "1", "Enables soft shadows instead of stencil shadows."); cvar_t r_shadow_shadowmapping_precision = CVARD ("r_shadow_shadowmapping_precision", "1", "Scales the shadowmap detail level up or down."); extern cvar_t r_shadow_shadowmapping_nearclip; extern cvar_t r_shadow_shadowmapping_bias; cvar_t r_sun_dir = CVARD ("r_sun_dir", "0.2 0.5 0.8", "Specifies the direction that crepusular rays appear along"); cvar_t r_sun_colour = CVARFD ("r_sun_colour", "0 0 0", CVAR_ARCHIVE, "Specifies the colour of sunlight that appears in the form of crepuscular rays."); static void Sh_DrawEntLighting(dlight_t *light, vec3_t colour); static pvsbuffer_t lvisb, lvisb2; /* called on framebuffer resize. flushes textures so they can be regenerated at the real size */ void Sh_Reset(void) { #ifdef GLQUAKE if (crepuscular_texture_id) { Image_DestroyTexture(crepuscular_texture_id); crepuscular_texture_id = r_nulltex; } GLBE_FBO_Destroy(&crepuscular_fbo); #endif } void Sh_Shutdown(void) { Sh_Reset(); SHM_Shutdown(); } typedef struct { unsigned int count; unsigned int max; texture_t *tex; vbo_t *vbo; mesh_t **s; } shadowmeshbatch_t; typedef struct shadowmesh_s { enum { SMT_STENCILVOLUME, //build edges mesh (and surface list) SMT_SHADOWMAP, //build front faces mesh (and surface list) SMT_ORTHO, //bounded by a box and with a single direction rather than an origin. SMT_SHADOWLESS, //build vis+surface list only SMT_DEFERRED //build vis without caring about any surfaces at all. } type; unsigned int numindicies; unsigned int maxindicies; index_t *indicies; unsigned int numverts; unsigned int maxverts; vecV_t *verts; //we also have a list of all the surfaces that this light lights. unsigned int numbatches; shadowmeshbatch_t *batches; unsigned int leafbytes; unsigned char *litleaves; #ifdef VKQUAKE struct vk_shadowbuffer *vkbuffer; #endif #ifdef GLQUAKE GLuint vebo[2]; #endif #ifdef D3D9QUAKE IDirect3DVertexBuffer9 *d3d9_vbuffer; IDirect3DIndexBuffer9 *d3d9_ibuffer; #endif #ifdef D3D11QUAKE void *d3d11_vbuffer; void *d3d11_ibuffer; #endif } shadowmesh_t; /*state of the current shadow mesh*/ #define inc 128 int sh_shadowframe; static int sh_firstindex; static int sh_vertnum; //vertex number (set to 0 at SH_Begin) static shadowmesh_t *sh_shmesh, sh_tempshmesh; /* functions to add geometry to the shadow mesh */ static void SHM_BeginQuads (void) { sh_firstindex = sh_shmesh->numverts; } static void SHM_End (void) { int i; i = (sh_shmesh->numindicies+(sh_vertnum/4)*6+inc+5)&~(inc-1); //and a bit of padding if (sh_shmesh->maxindicies != i) { sh_shmesh->maxindicies = i; sh_shmesh->indicies = BZ_Realloc(sh_shmesh->indicies, i * sizeof(*sh_shmesh->indicies)); } //add the extra triangles for (i = 0; i < sh_vertnum; i+=4) { sh_shmesh->indicies[sh_shmesh->numindicies++] = sh_firstindex + i+0; sh_shmesh->indicies[sh_shmesh->numindicies++] = sh_firstindex + i+1; sh_shmesh->indicies[sh_shmesh->numindicies++] = sh_firstindex + i+2; sh_shmesh->indicies[sh_shmesh->numindicies++] = sh_firstindex + i+0; sh_shmesh->indicies[sh_shmesh->numindicies++] = sh_firstindex + i+2; sh_shmesh->indicies[sh_shmesh->numindicies++] = sh_firstindex + i+3; } sh_vertnum = 0; } static void SHM_Vertex3fv (const float *v) { int i; //add the verts as we go i = (sh_shmesh->numverts+inc+5)&~(inc-1); //and a bit of padding if (sh_shmesh->maxverts < i) { sh_shmesh->maxverts = i; sh_shmesh->verts = BZ_Realloc(sh_shmesh->verts, i * sizeof(*sh_shmesh->verts)); } sh_shmesh->verts[sh_shmesh->numverts][0] = v[0]; sh_shmesh->verts[sh_shmesh->numverts][1] = v[1]; sh_shmesh->verts[sh_shmesh->numverts][2] = v[2]; sh_vertnum++; sh_shmesh->numverts++; if (sh_vertnum == 4) { SHM_End(); sh_firstindex = sh_shmesh->numverts; } } static void SHM_MeshFrontOnly(int numverts, vecV_t *verts, int numidx, index_t *idx) { int first = sh_shmesh->numverts; int v, i; vecV_t *outv; index_t *outi; /*make sure there's space*/ v = (sh_shmesh->numverts+numverts + inc)&~(inc-1); //and a bit of padding if (sh_shmesh->maxverts < v) { v *= 2; v += 1024; sh_shmesh->maxverts = v; sh_shmesh->verts = BZ_Realloc(sh_shmesh->verts, v * sizeof(*sh_shmesh->verts)); } outv = sh_shmesh->verts + sh_shmesh->numverts; for (v = 0; v < numverts; v++) { VectorCopy(verts[v], outv[v]); } v = (sh_shmesh->numindicies+numidx + inc)&~(inc-1); //and a bit of padding if (sh_shmesh->maxindicies < v) { v *= 2; v += 1024; sh_shmesh->maxindicies = v; sh_shmesh->indicies = BZ_Realloc(sh_shmesh->indicies, v * sizeof(*sh_shmesh->indicies)); } outi = sh_shmesh->indicies + sh_shmesh->numindicies; for (i = 0; i < numidx; i++) { outi[i] = first + idx[i]; } sh_shmesh->numverts += numverts; sh_shmesh->numindicies += numidx; } #if 0 static void SHM_MeshBackOnly(int numverts, vecV_t *verts, int numidx, index_t *idx) { int first = sh_shmesh->numverts; int v, i; vecV_t *outv; index_t *outi; /*make sure there's space*/ v = (sh_shmesh->numverts+numverts + inc)&~(inc-1); //and a bit of padding if (sh_shmesh->maxverts < v) { v += 1024; sh_shmesh->maxverts = v; sh_shmesh->verts = BZ_Realloc(sh_shmesh->verts, v * sizeof(*sh_shmesh->verts)); } outv = sh_shmesh->verts + sh_shmesh->numverts; for (v = 0; v < numverts; v++) { VectorCopy(verts[v], outv[v]); } v = (sh_shmesh->numindicies+numidx + inc)&~(inc-1); //and a bit of padding if (sh_shmesh->maxindicies < v) { v += 1024; sh_shmesh->maxindicies = v; sh_shmesh->indicies = BZ_Realloc(sh_shmesh->indicies, v * sizeof(*sh_shmesh->indicies)); } outi = sh_shmesh->indicies + sh_shmesh->numindicies; for (i = 0; i < numidx; i+=3) { outi[i+0] = first + idx[i+2]; outi[i+1] = first + idx[i+1]; outi[i+2] = first + idx[i+0]; } sh_shmesh->numverts += numverts; sh_shmesh->numindicies += numidx; } #endif static void SHM_TriangleFan(int numverts, vecV_t *verts, vec3_t lightorg, float pd) { int v, i, idxs; float *v1; vec3_t v3; vecV_t *outv; index_t *outi; /*make sure there's space*/ v = (sh_shmesh->numverts+numverts*2 + inc)&~(inc-1); //and a bit of padding if (sh_shmesh->maxverts < v) { v += 1024; sh_shmesh->maxverts = v; sh_shmesh->verts = BZ_Realloc(sh_shmesh->verts, v * sizeof(*sh_shmesh->verts)); } outv = sh_shmesh->verts + sh_shmesh->numverts; for (v = 0; v < numverts; v++) { v1 = verts[v]; VectorCopy(v1, outv[v]); v3[0] = ( v1[0]-lightorg[0] )*pd; v3[1] = ( v1[1]-lightorg[1] )*pd; v3[2] = ( v1[2]-lightorg[2] )*pd; outv[v+numverts][0] = v1[0]+v3[0]; outv[v+numverts][1] = v1[1]+v3[1]; outv[v+numverts][2] = v1[2]+v3[2]; } idxs = (numverts-2)*3; /*now add the verts in a fan*/ v = (sh_shmesh->numindicies+idxs*2+inc)&~(inc-1); //and a bit of padding if (sh_shmesh->maxindicies < v) { v += 1024; sh_shmesh->maxindicies = v; sh_shmesh->indicies = BZ_Realloc(sh_shmesh->indicies, v * sizeof(*sh_shmesh->indicies)); } outi = sh_shmesh->indicies + sh_shmesh->numindicies; for (v = 2, i = 0; v < numverts; v++, i+=3) { outi[i+0] = sh_shmesh->numverts; outi[i+1] = sh_shmesh->numverts+v-1; outi[i+2] = sh_shmesh->numverts+v; outi[i+0+idxs] = sh_shmesh->numverts+numverts+v; outi[i+1+idxs] = sh_shmesh->numverts+numverts+v-1; outi[i+2+idxs] = sh_shmesh->numverts+numverts; } /*we added this many*/ sh_shmesh->numverts += numverts*2; sh_shmesh->numindicies += i*2; } static void SHM_Shadow_Cache_Surface(msurface_t *surf) { int i; i = surf->sbatch->shadowbatch; if (i < 0) return; if (sh_shmesh->batches[i].count == sh_shmesh->batches[i].max) { sh_shmesh->batches[i].max += 64; sh_shmesh->batches[i].s = BZ_Realloc(sh_shmesh->batches[i].s, sizeof(void*)*(sh_shmesh->batches[i].max)); } sh_shmesh->batches[i].s[sh_shmesh->batches[i].count] = surf->mesh; sh_shmesh->batches[i].count++; } static void SHM_Shadow_Cache_Leaf(mleaf_t *leaf) { int i; i = (leaf - cl.worldmodel->leafs)-1; sh_shmesh->litleaves[i>>3] |= 1<<(i&7); } static void SH_FreeShadowMesh_(shadowmesh_t *sm) { unsigned int i; for (i = 0; i < sm->numbatches; i++) Z_Free(sm->batches[i].s); sm->numbatches = 0; Z_Free(sm->batches); sm->batches = NULL; Z_Free(sm->indicies); sm->indicies = NULL; Z_Free(sm->verts); sm->verts = NULL; sm->numindicies = 0; sm->numverts = 0; switch (qrenderer) { case QR_NONE: case QR_SOFTWARE: default: break; #ifdef GLQUAKE case QR_OPENGL: if (qglDeleteBuffersARB) qglDeleteBuffersARB(2, sm->vebo); sm->vebo[0] = 0; sm->vebo[1] = 0; break; #endif #ifdef VKQUAKE case QR_VULKAN: VKBE_DestroyShadowBuffer(sm->vkbuffer); sm->vkbuffer = NULL; break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: if (sm->d3d9_ibuffer) IDirect3DIndexBuffer9_Release(sm->d3d9_ibuffer); sm->d3d9_ibuffer = NULL; if (sm->d3d9_vbuffer) IDirect3DVertexBuffer9_Release(sm->d3d9_vbuffer); sm->d3d9_vbuffer = NULL; break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: D3D11_DestroyShadowBuffer(sm->d3d11_vbuffer, sm->d3d11_ibuffer); sm->d3d11_vbuffer = NULL; sm->d3d11_ibuffer = NULL; break; #endif } } void SH_FreeShadowMesh(shadowmesh_t *sm) { SH_FreeShadowMesh_(sm); Z_Free(sm); } static void SH_CalcShadowBatches(model_t *mod) { int s; batch_t *b; batch_t *l = NULL; int sb; l = NULL; for (s = 0; s < SHADER_SORT_COUNT; s++) { for (b = mod->batches[s]; b; b = b->next) { if (!l || l->vbo != b->vbo || l->texture != b->texture) { b->shadowbatch = mod->numshadowbatches++; l = b; } else b->shadowbatch = l->shadowbatch; } } if (!mod->numshadowbatches) mod->shadowbatches = NULL; else { l = NULL; sb = 0; mod->shadowbatches = BZ_Malloc(sizeof(*mod->shadowbatches)*mod->numshadowbatches); for (s = 0; s < SHADER_SORT_COUNT; s++) { for (b = mod->batches[s]; b; b = b->next) { if (!l || l->vbo != b->vbo || l->texture != b->texture) { mod->shadowbatches[sb].tex = b->texture; mod->shadowbatches[sb].vbo = b->vbo; sb++; l = b; } } } } } static void SHM_BeginShadowMesh(dlight_t *dl, int type) { unsigned int i; unsigned int lb; sh_vertnum = 0; lb = (cl.worldmodel->numclusters+7)/8; if (!dl->die || !dl->key) { sh_shmesh = dl->worldshadowmesh; if (!sh_shmesh || sh_shmesh->leafbytes != lb) { /*this shouldn't happen too often*/ if (sh_shmesh) { //FIXME: if the light is the same light, reuse the memory allocations where possible... SH_FreeShadowMesh(sh_shmesh); } /*Create a new shadowmesh for this light*/ sh_shmesh = Z_Malloc(sizeof(*sh_shmesh) + lb); sh_shmesh->leafbytes = lb; sh_shmesh->litleaves = (unsigned char*)(sh_shmesh+1); dl->worldshadowmesh = sh_shmesh; } memset(sh_shmesh->litleaves, 0, sh_shmesh->leafbytes); dl->rebuildcache = false; } else { sh_shmesh = &sh_tempshmesh; if (sh_shmesh->leafbytes != lb) { /*this happens on map changes*/ sh_shmesh->leafbytes = lb; Z_Free(sh_shmesh->litleaves); sh_shmesh->litleaves = Z_Malloc(lb); } } sh_shmesh->maxverts = 0; sh_shmesh->numverts = 0; sh_shmesh->maxindicies = 0; sh_shmesh->numindicies = 0; sh_shmesh->type = type; if (!cl.worldmodel->numshadowbatches) { SH_CalcShadowBatches(cl.worldmodel); } if (sh_shmesh->numbatches != cl.worldmodel->numshadowbatches) { if (sh_shmesh->batches) { for (i = 0; i < sh_shmesh->numbatches; i++) Z_Free(sh_shmesh->batches[i].s); Z_Free(sh_shmesh->batches); } sh_shmesh->batches = Z_Malloc(sizeof(shadowmeshbatch_t)*cl.worldmodel->numshadowbatches); sh_shmesh->numbatches=cl.worldmodel->numshadowbatches; } for (i = 0; i < sh_shmesh->numbatches; i++) { sh_shmesh->batches[i].count = 0; } } static struct shadowmesh_s *SHM_FinishShadowMesh(dlight_t *dl) { if (sh_shmesh != &sh_tempshmesh || 1) { switch (qrenderer) { case QR_NONE: case QR_SOFTWARE: default: break; #ifdef GLQUAKE case QR_OPENGL: if (!qglGenBuffersARB) return sh_shmesh; if (!sh_shmesh->vebo[0]) qglGenBuffersARB(2, sh_shmesh->vebo); GL_DeselectVAO(); GL_SelectVBO(sh_shmesh->vebo[0]); qglBufferDataARB(GL_ARRAY_BUFFER_ARB, sizeof(*sh_shmesh->verts) * sh_shmesh->numverts, sh_shmesh->verts, GL_STATIC_DRAW_ARB); GL_SelectEBO(sh_shmesh->vebo[1]); qglBufferDataARB(GL_ELEMENT_ARRAY_BUFFER_ARB, sizeof(*sh_shmesh->indicies) * sh_shmesh->numindicies, sh_shmesh->indicies, GL_STATIC_DRAW_ARB); break; #endif #ifdef VKQUAKE case QR_VULKAN: VKBE_DestroyShadowBuffer(sh_shmesh->vkbuffer); sh_shmesh->vkbuffer = VKBE_GenerateShadowBuffer(sh_shmesh->verts, sh_shmesh->numverts, sh_shmesh->indicies, sh_shmesh->numindicies, sh_shmesh == &sh_tempshmesh); break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: if (sh_shmesh->numindicies && sh_shmesh->numverts) { void *map; IDirect3DDevice9_CreateIndexBuffer(pD3DDev9, sizeof(index_t) * sh_shmesh->numindicies, 0, D3DFMT_QINDEX, D3DPOOL_MANAGED, &sh_shmesh->d3d9_ibuffer, NULL); IDirect3DIndexBuffer9_Lock(sh_shmesh->d3d9_ibuffer, 0, sizeof(index_t) * sh_shmesh->numindicies, &map, D3DLOCK_DISCARD); memcpy(map, sh_shmesh->indicies, sizeof(index_t) * sh_shmesh->numindicies); IDirect3DIndexBuffer9_Unlock(sh_shmesh->d3d9_ibuffer); IDirect3DDevice9_CreateVertexBuffer(pD3DDev9, sizeof(vecV_t) * sh_shmesh->numverts, D3DUSAGE_WRITEONLY, 0, D3DPOOL_MANAGED, &sh_shmesh->d3d9_vbuffer, NULL); IDirect3DVertexBuffer9_Lock(sh_shmesh->d3d9_vbuffer, 0, sizeof(vecV_t) * sh_shmesh->numverts, &map, D3DLOCK_DISCARD); memcpy(map, sh_shmesh->verts, sizeof(vecV_t) * sh_shmesh->numverts); IDirect3DVertexBuffer9_Unlock(sh_shmesh->d3d9_vbuffer); } break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: D3D11BE_GenerateShadowBuffer(&sh_shmesh->d3d11_vbuffer, sh_shmesh->verts, sh_shmesh->numverts, &sh_shmesh->d3d11_ibuffer, sh_shmesh->indicies, sh_shmesh->numindicies); break; #endif } Z_Free(sh_shmesh->verts); sh_shmesh->verts = NULL; Z_Free(sh_shmesh->indicies); sh_shmesh->indicies = NULL; } return sh_shmesh; } /*state of the world that is still to compile*/ static struct { short count; short count2; int next; int prev; } *edge; static int firstedge; static int maxedge; static void SHM_RecursiveWorldNodeQ1_r (dlight_t *dl, mnode_t *node) { int c, side; mplane_t *plane; msurface_t *surf, **mark; mleaf_t *pleaf; double dot; int v; float l, maxdist; int j, s, t; vec3_t impact; if (node->shadowframe != sh_shadowframe) return; if (node->contents == Q1CONTENTS_SOLID) return; // solid //if light areabox is outside node, ignore node + children for (c = 0; c < 3; c++) { if (dl->origin[c] + dl->radius < node->minmaxs[c]) return; if (dl->origin[c] - dl->radius > node->minmaxs[3+c]) return; } // if a leaf node, draw stuff if (node->contents < 0) { pleaf = (mleaf_t *)node; SHM_Shadow_Cache_Leaf(pleaf); if (sh_shmesh->type == SMT_DEFERRED) //such rtlights don't need ANY surface info, just a tight pvs return; mark = pleaf->firstmarksurface; c = pleaf->nummarksurfaces; if (c) { do { (*mark++)->shadowframe = sh_shadowframe; } while (--c); } return; } // node is just a decision point, so go down the apropriate sides // find which side of the node we are on plane = node->plane; switch (plane->type) { case PLANE_X: dot = dl->origin[0] - plane->dist; break; case PLANE_Y: dot = dl->origin[1] - plane->dist; break; case PLANE_Z: dot = dl->origin[2] - plane->dist; break; default: dot = DotProduct (dl->origin, plane->normal) - plane->dist; break; } if (dot >= 0) side = 0; else side = 1; // recurse down the children, front side first SHM_RecursiveWorldNodeQ1_r (dl, node->children[side]); // draw stuff c = node->numsurfaces; if (c) { surf = cl.worldmodel->surfaces + node->firstsurface; { maxdist = dl->radius*dl->radius; for ( ; c ; c--, surf++) { if (surf->shadowframe != sh_shadowframe) continue; // if ((dot < 0) ^ !!(surf->flags & SURF_PLANEBACK)) // continue; // wrong side // if (surf->flags & SURF_PLANEBACK) // continue; if (surf->flags & (SURF_DRAWALPHA | SURF_DRAWTILED)) { // no shadows continue; } //is the light on the right side? if (surf->flags & SURF_PLANEBACK) {//inverted normal. if (-DotProduct(surf->plane->normal, dl->origin)+surf->plane->dist >= dl->radius) continue; } else { if (DotProduct(surf->plane->normal, dl->origin)-surf->plane->dist >= dl->radius) continue; } //Yeah, you can blame LordHavoc for this alternate code here. for (j=0 ; j<3 ; j++) impact[j] = dl->origin[j] - surf->plane->normal[j]*dot; // clamp center of light to corner and check brightness l = DotProduct (impact, surf->texinfo->vecs[0]) + surf->texinfo->vecs[0][3] - surf->texturemins[0]; s = l+0.5;if (s < 0) s = 0;else if (s > surf->extents[0]) s = surf->extents[0]; s = (l - s)*surf->texinfo->vecscale[0]; l = DotProduct (impact, surf->texinfo->vecs[1]) + surf->texinfo->vecs[1][3] - surf->texturemins[1]; t = l+0.5;if (t < 0) t = 0;else if (t > surf->extents[1]) t = surf->extents[1]; t = (l - t)*surf->texinfo->vecscale[1]; // compare to minimum light if ((s*s+t*t+dot*dot) < maxdist) { SHM_Shadow_Cache_Surface(surf); if (sh_shmesh->type == SMT_SHADOWMAP) { SHM_MeshFrontOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); continue; } if (sh_shmesh->type != SMT_STENCILVOLUME) continue; //build a list of the edges that are to be drawn. for (v = 0; v < surf->numedges; v++) { int e, delta; e = cl.worldmodel->surfedges[surf->firstedge+v]; //negative edge means backwards edge. if (e < 0) { e=-e; delta = -1; } else { delta = 1; } if (!edge[e].count) { if (firstedge) edge[firstedge].prev = e; edge[e].next = firstedge; edge[e].prev = 0; firstedge = e; edge[e].count = delta; } else { edge[e].count += delta; if (!edge[e].count) //unlink { if (edge[e].next) { edge[edge[e].next].prev = edge[e].prev; } if (edge[e].prev) edge[edge[e].prev].next = edge[e].next; else firstedge = edge[e].next; } } } SHM_TriangleFan(surf->mesh->numvertexes, surf->mesh->xyz_array, dl->origin, PROJECTION_DISTANCE); } } } } // recurse down the back side SHM_RecursiveWorldNodeQ1_r (dl, node->children[!side]); } void CategorizePlane ( mplane_t *plane ); static void SHM_OrthoWorldLeafsQ1 (dlight_t *dl) { int c, i; msurface_t *surf, **mark; mleaf_t *pleaf, *plastleaf; float dot; mplane_t orthoplanes[5]; sh_shadowframe++; VectorCopy(dl->axis[0], orthoplanes[0].normal); VectorNegate(dl->axis[0], orthoplanes[1].normal); VectorCopy(dl->axis[1], orthoplanes[2].normal); VectorNegate(dl->axis[1], orthoplanes[3].normal); VectorNegate(dl->axis[0], orthoplanes[4].normal); for (i = 0; i < countof(orthoplanes); i++) { orthoplanes[i].dist = DotProduct(dl->origin, orthoplanes[i].normal) - dl->radius; CategorizePlane(&orthoplanes[i]); } for (pleaf = cl.worldmodel->leafs+1, plastleaf = cl.worldmodel->leafs+cl.worldmodel->submodels[0].visleafs; pleaf <= plastleaf; pleaf++) { for (i = 0; i < countof(orthoplanes); i++) if (BOX_ON_PLANE_SIDE (pleaf->minmaxs, pleaf->minmaxs+3, &orthoplanes[i]) == 2) goto next; SHM_Shadow_Cache_Leaf(pleaf); mark = pleaf->firstmarksurface; c = pleaf->nummarksurfaces; while (c --> 0) { surf = *mark++; if (surf->flags & (SURF_DRAWALPHA | SURF_DRAWTILED | SURF_DRAWSKY)) continue; if (surf->shadowframe != sh_shadowframe) { surf->shadowframe = sh_shadowframe; dot = DotProduct(surf->plane->normal, dl->axis[0]); if (surf->flags & SURF_PLANEBACK) dot = -dot; if (dot < 0) { SHM_Shadow_Cache_Surface(surf); } // else // SHM_MeshBackOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); SHM_MeshFrontOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); } } next:; } } static void SHM_OrthoWorldLeafsQ3 (dlight_t *dl) { int c, i; msurface_t *surf, **mark; mleaf_t *pleaf, *plastleaf; mplane_t orthoplanes[5]; sh_shadowframe++; VectorCopy(dl->axis[0], orthoplanes[0].normal); VectorNegate(dl->axis[0], orthoplanes[1].normal); VectorCopy(dl->axis[1], orthoplanes[2].normal); VectorNegate(dl->axis[1], orthoplanes[3].normal); VectorNegate(dl->axis[0], orthoplanes[4].normal); for (i = 0; i < countof(orthoplanes); i++) { orthoplanes[i].dist = DotProduct(dl->origin, orthoplanes[i].normal) - dl->radius; CategorizePlane(&orthoplanes[i]); } for (pleaf = cl.worldmodel->leafs+1, plastleaf = cl.worldmodel->leafs+cl.worldmodel->numleafs; pleaf <= plastleaf; pleaf++) { for (i = 0; i < countof(orthoplanes); i++) if (BOX_ON_PLANE_SIDE (pleaf->minmaxs, pleaf->minmaxs+3, &orthoplanes[i]) == 2) goto next; SHM_Shadow_Cache_Leaf(pleaf); mark = pleaf->firstmarksurface; c = pleaf->nummarksurfaces; while (c --> 0) { surf = *mark++; if (surf->flags & (SURF_DRAWALPHA | SURF_DRAWTILED | SURF_DRAWSKY)) continue; if (surf->shadowframe != sh_shadowframe) { surf->shadowframe = sh_shadowframe; // if (dot < 0) { SHM_Shadow_Cache_Surface(surf); } // else // SHM_MeshBackOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); SHM_MeshFrontOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); } } next:; } } #ifdef Q2BSPS static void SHM_RecursiveWorldNodeQ2_r (dlight_t *dl, mnode_t *node) { int c, side; mplane_t *plane; msurface_t *surf, **mark; mleaf_t *pleaf; double dot; int v; float l, maxdist; int j, s, t; vec3_t impact; if (node->shadowframe != sh_shadowframe) return; if (node->contents == Q2CONTENTS_SOLID) return; // solid //if light areabox is outside node, ignore node + children for (c = 0; c < 3; c++) { if (dl->origin[c] + dl->radius < node->minmaxs[c]) return; if (dl->origin[c] - dl->radius > node->minmaxs[3+c]) return; } // if a leaf node, draw stuff if (node->contents != -1) { pleaf = (mleaf_t *)node; if (pleaf->cluster >= 0) sh_shmesh->litleaves[pleaf->cluster>>3] |= 1<<(pleaf->cluster&7); mark = pleaf->firstmarksurface; c = pleaf->nummarksurfaces; if (c) { do { (*mark++)->shadowframe = sh_shadowframe; } while (--c); } return; } // node is just a decision point, so go down the apropriate sides // find which side of the node we are on plane = node->plane; switch (plane->type) { case PLANE_X: dot = dl->origin[0] - plane->dist; break; case PLANE_Y: dot = dl->origin[1] - plane->dist; break; case PLANE_Z: dot = dl->origin[2] - plane->dist; break; default: dot = DotProduct (dl->origin, plane->normal) - plane->dist; break; } if (dot >= 0) side = 0; else side = 1; // recurse down the children, front side first SHM_RecursiveWorldNodeQ2_r (dl, node->children[side]); // draw stuff c = node->numsurfaces; if (c) { surf = cl.worldmodel->surfaces + node->firstsurface; { maxdist = dl->radius*dl->radius; for ( ; c ; c--, surf++) { if (surf->shadowframe != sh_shadowframe) continue; // if ((dot < 0) ^ !!(surf->flags & SURF_PLANEBACK)) // continue; // wrong side // if (surf->flags & SURF_PLANEBACK) // continue; if (surf->flags & (SURF_DRAWALPHA | SURF_DRAWTILED)) { // no shadows continue; } //is the light on the right side? if (surf->flags & SURF_PLANEBACK) {//inverted normal. if (-DotProduct(surf->plane->normal, dl->origin)+surf->plane->dist >= dl->radius) continue; } else { if (DotProduct(surf->plane->normal, dl->origin)-surf->plane->dist >= dl->radius) continue; } //Yeah, you can blame LordHavoc for this alternate code here. for (j=0 ; j<3 ; j++) impact[j] = dl->origin[j] - surf->plane->normal[j]*dot; // clamp center of light to corner and check brightness l = DotProduct (impact, surf->texinfo->vecs[0]) + surf->texinfo->vecs[0][3] - surf->texturemins[0]; s = l;if (s < 0) s = 0;else if (s > surf->extents[0]) s = surf->extents[0]; s = (l - s)*surf->texinfo->vecscale[0]; l = DotProduct (impact, surf->texinfo->vecs[1]) + surf->texinfo->vecs[1][3] - surf->texturemins[1]; t = l;if (t < 0) t = 0;else if (t > surf->extents[1]) t = surf->extents[1]; t = (l - t)*surf->texinfo->vecscale[1]; // compare to minimum light if ((s*s+t*t+dot*dot) < maxdist) { SHM_Shadow_Cache_Surface(surf); if (sh_shmesh->type == SMT_SHADOWMAP) { SHM_MeshFrontOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); continue; } if (sh_shmesh->type != SMT_STENCILVOLUME) continue; //build a list of the edges that are to be drawn. for (v = 0; v < surf->numedges; v++) { int e, delta; e = cl.worldmodel->surfedges[surf->firstedge+v]; //negative edge means backwards edge. if (e < 0) { e=-e; delta = -1; } else { delta = 1; } if (!edge[e].count) { if (firstedge) edge[firstedge].prev = e; edge[e].next = firstedge; edge[e].prev = 0; firstedge = e; edge[e].count = delta; } else { edge[e].count += delta; if (!edge[e].count) //unlink { if (edge[e].next) { edge[edge[e].next].prev = edge[e].prev; } if (edge[e].prev) edge[edge[e].prev].next = edge[e].next; else firstedge = edge[e].next; } } } SHM_TriangleFan(surf->mesh->numvertexes, surf->mesh->xyz_array, dl->origin, PROJECTION_DISTANCE); } } } } // recurse down the back side SHM_RecursiveWorldNodeQ2_r (dl, node->children[!side]); } static void SHM_MarkLeavesQ2(dlight_t *dl, unsigned char *lvis) { mnode_t *node; int i; mleaf_t *leaf; int cluster; sh_shadowframe++; if (!dl->die) { //static //variation on mark leaves for (i=0,leaf=cl.worldmodel->leafs ; inumleafs ; i++, leaf++) { cluster = leaf->cluster; if (cluster == -1) continue; if (lvis[cluster>>3] & (1<<(cluster&7))) { node = (mnode_t *)leaf; do { if (node->shadowframe == sh_shadowframe) break; node->shadowframe = sh_shadowframe; node = node->parent; } while (node); } } } else { //dynamic lights will be discarded after this frame anyway, so only include leafs that are visible //variation on mark leaves for (i=0,leaf=cl.worldmodel->leafs ; inumleafs ; i++, leaf++) { cluster = leaf->cluster; if (cluster == -1) continue; if (lvis[cluster>>3] & (1<<(cluster&7))) { node = (mnode_t *)leaf; do { if (node->shadowframe == sh_shadowframe) break; node->shadowframe = sh_shadowframe; node = node->parent; } while (node); } } } } #endif static void SHM_MarkLeavesQ1(dlight_t *dl, unsigned char *lvis) { mnode_t *node; int i; sh_shadowframe++; if (!lvis) return; //variation on mark leaves for (i=0 ; inumclusters ; i++) { if (lvis[i>>3] & (1<<(i&7))) { node = (mnode_t *)&cl.worldmodel->leafs[i+1]; do { if (node->shadowframe == sh_shadowframe) break; node->shadowframe = sh_shadowframe; node = node->parent; } while (node); } } } #ifdef Q3BSPS static void SHM_RecursiveWorldNodeQ3_r (dlight_t *dl, mnode_t *node) { mplane_t *splitplane; float dist; msurface_t **msurf; msurface_t *surf; mleaf_t *leaf; int i; if (node->contents != -1) { leaf = (mleaf_t *)node; if (leaf->cluster >= 0) sh_shmesh->litleaves[leaf->cluster>>3] |= 1<<(leaf->cluster&7); // mark the polygons msurf = leaf->firstmarksurface; for (i=0 ; inummarksurfaces ; i++, msurf++) { surf = *msurf; //only check each surface once. it can appear in multiple leafs. if (surf->shadowframe == sh_shadowframe) continue; surf->shadowframe = sh_shadowframe; //FIXME: radius check SHM_Shadow_Cache_Surface(surf); if (sh_shmesh->type == SMT_SHADOWMAP) SHM_MeshFrontOnly(surf->mesh->numvertexes, surf->mesh->xyz_array, surf->mesh->numindexes, surf->mesh->indexes); } return; } splitplane = node->plane; dist = DotProduct (dl->origin, splitplane->normal) - splitplane->dist; if (dist > dl->radius) { SHM_RecursiveWorldNodeQ3_r (dl, node->children[0]); return; } if (dist < -dl->radius) { SHM_RecursiveWorldNodeQ3_r (dl, node->children[1]); return; } SHM_RecursiveWorldNodeQ3_r (dl, node->children[0]); SHM_RecursiveWorldNodeQ3_r (dl, node->children[1]); } #endif static struct { unsigned int numtris; unsigned int maxtris; struct { signed int edge[3]; } *tris; /*negative for reverse edge*/ unsigned int numedges; unsigned int maxedges; struct { unsigned int vert[2]; } *edges; unsigned int numpoints; unsigned int maxpoints; vec3_t *points; unsigned int maxedgeuses; int *edgeuses; /*negative for back sides, so 0 means unused or used equally on both sides*/ } cv; static void SHM_Shutdown(void) { SH_FreeShadowMesh_(&sh_tempshmesh); BZ_Free(sh_tempshmesh.litleaves); sh_tempshmesh.litleaves = NULL; sh_tempshmesh.leafbytes = 0; free(cv.tris); free(cv.edges); free(cv.points); memset(&cv, 0, sizeof(cv)); } #ifdef Q3BSPS #define VERT_POS_EPSILON (1.0f/32) static int SHM_ComposeVolume_FindVert(float *vert) { int i; for (i = 0; i < cv.numpoints; i++) { #if 1 if (cv.points[i][0] == vert[0] && cv.points[i][1] == vert[1] && cv.points[i][2] == vert[2]) #else vec3_t d; d[0] = cv.points[i][0]-vert[0]; d[1] = cv.points[i][1]-vert[1]; d[2] = cv.points[i][2]-vert[2]; if (d[0]*d[0] < VERT_POS_EPSILON && d[1]*d[1] < VERT_POS_EPSILON && d[2]*d[2] < VERT_POS_EPSILON) #endif return i; } VectorCopy(vert, cv.points[i]); cv.numpoints++; return i; } static int SHM_ComposeVolume_FindEdge(int v1, int v2) { int i; for (i = 0; i < cv.numedges; i++) { if (cv.edges[i].vert[0] == v1 && cv.edges[i].vert[1] == v2) return i; if (cv.edges[i].vert[0] == v2 && cv.edges[i].vert[1] == v1) return -(i+1); } cv.edges[i].vert[0] = v1; cv.edges[i].vert[1] = v2; cv.numedges++; return i; } /*each triangle is coplanar, and all face the light, and its a triangle fan. this is a special case that provides a slight speedup*/ static void SHM_ComposeVolume_Fan(vecV_t *points, int numpoints) { int newmax; int lastedge; int i; #define MAX_ARRAY_VERTS 65535 static index_t pointidx[MAX_ARRAY_VERTS]; /*make sure there's space*/ newmax = (cv.numpoints+numpoints + inc)&~(inc-1); if (cv.maxpoints < newmax) { cv.maxpoints = newmax; cv.points = BZ_Realloc(cv.points, newmax * sizeof(*cv.points)); } newmax = (cv.numedges+(numpoints-2)*3 + inc)&~(inc-1); if (cv.maxedges < newmax) { cv.maxedges = newmax; cv.edges = BZ_Realloc(cv.edges, newmax * sizeof(*cv.edges)); } newmax = (cv.numtris+(numpoints-2) + inc)&~(inc-1); if (cv.maxtris < newmax) { cv.maxtris = newmax; cv.tris = BZ_Realloc(cv.tris, newmax * sizeof(*cv.tris)); } for (i = 0; i < numpoints; i++) { pointidx[i] = SHM_ComposeVolume_FindVert(points[i]); } lastedge = SHM_ComposeVolume_FindEdge(pointidx[0], pointidx[1]); for (i = 2; i < numpoints; i++) { cv.tris[cv.numtris].edge[0] = lastedge; cv.tris[cv.numtris].edge[1] = SHM_ComposeVolume_FindEdge(pointidx[i-1], pointidx[i]); lastedge = SHM_ComposeVolume_FindEdge(pointidx[i], pointidx[0]); cv.tris[cv.numtris].edge[2] = lastedge; lastedge = -(lastedge+1); cv.numtris++; } } static void SHM_ComposeVolume_Soup(vecV_t *points, int numpoints, index_t *idx, int numidx) { int newmax; int i; #define MAX_ARRAY_VERTS 65535 static index_t pointidx[MAX_ARRAY_VERTS]; /*make sure there's space*/ newmax = (cv.numpoints+numpoints + inc)&~(inc-1); if (cv.maxpoints < newmax) { cv.maxpoints = newmax; cv.points = BZ_Realloc(cv.points, newmax * sizeof(*cv.points)); } newmax = (cv.numedges+numidx + inc)&~(inc-1); if (cv.maxedges < newmax) { cv.maxedges = newmax; cv.edges = BZ_Realloc(cv.edges, newmax * sizeof(*cv.edges)); } newmax = (cv.numtris+numidx/3 + inc)&~(inc-1); if (cv.maxtris < newmax) { cv.maxtris = newmax; cv.tris = BZ_Realloc(cv.tris, newmax * sizeof(*cv.tris)); } for (i = 0; i < numpoints; i++) { pointidx[i] = SHM_ComposeVolume_FindVert(points[i]); } for (i = 0; i < numidx; i+=3, idx+=3) { cv.tris[cv.numtris].edge[0] = SHM_ComposeVolume_FindEdge(pointidx[idx[0]], pointidx[idx[1]]); cv.tris[cv.numtris].edge[1] = SHM_ComposeVolume_FindEdge(pointidx[idx[1]], pointidx[idx[2]]); cv.tris[cv.numtris].edge[2] = SHM_ComposeVolume_FindEdge(pointidx[idx[2]], pointidx[idx[0]]); cv.numtris++; } } /*call this function after generating litsurfs meshes*/ static void SHM_ComposeVolume_BruteForce(dlight_t *dl) { shadowmeshbatch_t *sms; unsigned int tno; unsigned int sno; int i, e; mesh_t *sm; vec3_t ext; float sc; cv.numedges = 0; cv.numpoints = 0; cv.numtris = 0; for (tno = 0; tno < sh_shmesh->numbatches; tno++) { sms = &sh_shmesh->batches[tno]; if (!sms->count) continue; if ((cl.worldmodel->shadowbatches[tno].tex->shader->flags & (SHADER_BLEND|SHADER_NODRAW))) continue; for (sno = 0; sno < sms->count; sno++) { sm = sms->s[sno]; if (sm->istrifan) SHM_ComposeVolume_Fan(sm->xyz_array, sm->numvertexes); else SHM_ComposeVolume_Soup(sm->xyz_array, sm->numvertexes, sm->indexes, sm->numindexes); } } /*FIXME: clip away overlapping triangles*/ if (cv.maxedgeuses < cv.numedges) { BZ_Free(cv.edgeuses); cv.maxedgeuses = cv.numedges; cv.edgeuses = Z_Malloc(cv.maxedgeuses * sizeof(*cv.edgeuses)); } else memset(cv.edgeuses, 0, cv.numedges * sizeof(*cv.edgeuses)); i = (sh_shmesh->numverts+cv.numpoints*6+inc+5)&~(inc-1); //and a bit of padding if (sh_shmesh->maxverts < i) { sh_shmesh->maxverts = i; sh_shmesh->verts = BZ_Realloc(sh_shmesh->verts, i * sizeof(*sh_shmesh->verts)); } for (i = 0; i < cv.numpoints; i++) { /*front face*/ sh_shmesh->verts[(i * 2) + 0][0] = cv.points[i][0]; sh_shmesh->verts[(i * 2) + 0][1] = cv.points[i][1]; sh_shmesh->verts[(i * 2) + 0][2] = cv.points[i][2]; /*shadow direction*/ ext[0] = cv.points[i][0]-dl->origin[0]; ext[1] = cv.points[i][1]-dl->origin[1]; ext[2] = cv.points[i][2]-dl->origin[2]; sc = dl->radius * VectorNormalize(ext); /*back face*/ sh_shmesh->verts[(i * 2) + 1][0] = cv.points[i][0] + ext[0] * sc; sh_shmesh->verts[(i * 2) + 1][1] = cv.points[i][1] + ext[1] * sc; sh_shmesh->verts[(i * 2) + 1][2] = cv.points[i][2] + ext[2] * sc; } sh_shmesh->numverts = i*2; i = (sh_shmesh->numindicies+cv.numtris*6+cv.numedges*6+inc+5)&~(inc-1); //and a bit of padding if (sh_shmesh->maxindicies < i) { sh_shmesh->maxindicies = i; sh_shmesh->indicies = BZ_Realloc(sh_shmesh->indicies, i * sizeof(*sh_shmesh->indicies)); } for (tno = 0; tno < cv.numtris; tno++) { for (i = 0; i < 3; i++) { e = cv.tris[tno].edge[i]; if (e < 0) { e = -(e+1); cv.edgeuses[e]--; e = cv.edges[e].vert[1]; } else { cv.edgeuses[e]++; e = cv.edges[e].vert[0]; } sh_shmesh->indicies[sh_shmesh->numindicies+i] = e*2; sh_shmesh->indicies[sh_shmesh->numindicies+5-i] = e*2 + 1; } sh_shmesh->numindicies += 6; } for (i = 0; i < cv.numedges; i++) { if (cv.edgeuses[i] > 0) { sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[1]*2 + 0; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[0]*2 + 0; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[0]*2 + 1; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[0]*2 + 1; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[1]*2 + 1; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[1]*2 + 0; } else if (cv.edgeuses[i] < 0) { //generally should not happen... sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[1]*2 + 0; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[0]*2 + 1; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[0]*2 + 0; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[0]*2 + 1; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[1]*2 + 0; sh_shmesh->indicies[sh_shmesh->numindicies++] = cv.edges[i].vert[1]*2 + 1; } } } #endif static struct shadowmesh_s *SHM_BuildShadowMesh(dlight_t *dl, unsigned char *lvis, int type) { float *v1, *v2; vec3_t v3, v4; if (dl->worldshadowmesh && !dl->rebuildcache && dl->worldshadowmesh->type == type) return dl->worldshadowmesh; if (!lvis) { int clus; if (type == SMT_ORTHO) ; else if ((type == SMT_SHADOWLESS || dl->lightcolourscales[0]) && cl.worldmodel->funcs.ClustersInSphere) //shadowless lights don't cast shadows, so they're seen through everything - their vis must reflect that. lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb, NULL); else { clus = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, dl->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, clus, &lvisb, PVM_FAST); if (cl.worldmodel->funcs.ClustersInSphere) lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb2, lvis); } } firstedge=0; if (maxedge < cl.worldmodel->numedges) { maxedge = cl.worldmodel->numedges; Z_Free(edge); edge = Z_Malloc(sizeof(*edge) * maxedge); } if (cl.worldmodel->type == mod_brush) { switch(cl.worldmodel->fromgame) { case fg_quake: case fg_halflife: /*if (!dl->die) { SHM_BeginShadowMesh(dl, true); SHM_MarkLeavesQ1(dl, lvis); SHM_RecursiveWorldNodeQ1_r(dl, cl.worldmodel->nodes); if (!surfonly) SHM_ComposeVolume_BruteForce(dl); } else*/ { SHM_BeginShadowMesh(dl, type); if (type == SMT_ORTHO) SHM_OrthoWorldLeafsQ1(dl); else { SHM_MarkLeavesQ1(dl, lvis); SHM_RecursiveWorldNodeQ1_r(dl, cl.worldmodel->nodes); } } break; #ifdef Q2BSPS case fg_quake2: SHM_BeginShadowMesh(dl, type); SHM_MarkLeavesQ2(dl, lvis); SHM_RecursiveWorldNodeQ2_r(dl, cl.worldmodel->nodes); break; #endif #ifdef Q3BSPS case fg_quake3: /*q3 doesn't have edge info*/ SHM_BeginShadowMesh(dl, type); if (type == SMT_ORTHO) SHM_OrthoWorldLeafsQ3(dl); else { sh_shadowframe++; SHM_RecursiveWorldNodeQ3_r(dl, cl.worldmodel->nodes); } if (type == SMT_STENCILVOLUME) SHM_ComposeVolume_BruteForce(dl); break; #endif default: SHM_BeginShadowMesh(dl, type); sh_shadowframe++; { int cluster = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, dl->origin); if (cluster >= 0) sh_shmesh->litleaves[cluster>>3] |= 1<<(cluster&7); } break; } } else { SHM_BeginShadowMesh(dl, type); sh_shadowframe++; } /*generate edge polys for map types that need it (q1/q2)*/ switch (type) { case SMT_STENCILVOLUME: SHM_BeginQuads(); while(firstedge) { //border v1 = cl.worldmodel->vertexes[cl.worldmodel->edges[firstedge].v[0]].position; v2 = cl.worldmodel->vertexes[cl.worldmodel->edges[firstedge].v[1]].position; //get positions of v3 and v4 based on the light position v3[0] = v1[0] + ( v1[0]-dl->origin[0] )*PROJECTION_DISTANCE; v3[1] = v1[1] + ( v1[1]-dl->origin[1] )*PROJECTION_DISTANCE; v3[2] = v1[2] + ( v1[2]-dl->origin[2] )*PROJECTION_DISTANCE; v4[0] = v2[0] + ( v2[0]-dl->origin[0] )*PROJECTION_DISTANCE; v4[1] = v2[1] + ( v2[1]-dl->origin[1] )*PROJECTION_DISTANCE; v4[2] = v2[2] + ( v2[2]-dl->origin[2] )*PROJECTION_DISTANCE; if (edge[firstedge].count > 0) { SHM_Vertex3fv(v3); SHM_Vertex3fv(v4); SHM_Vertex3fv(v2); SHM_Vertex3fv(v1); } else { SHM_Vertex3fv(v1); SHM_Vertex3fv(v2); SHM_Vertex3fv(v4); SHM_Vertex3fv(v3); } edge[firstedge].count=0; firstedge = edge[firstedge].next; } SHM_End(); break; } return SHM_FinishShadowMesh(dl); } static qboolean Sh_VisOverlaps(qbyte *v1, qbyte *v2) { int i, m; if (!v2 || !v1) return true; m = (cl.worldmodel->numclusters+7)>>3; for (i=(m&~3) ; i>=2; for (i=0 ; inumvisleafs); mleaf_t *wl = cl.worldmodel->leafs; unsigned char lv; /*we can potentially walk off the end of the leafs, but lightvis shouldn't be set for those*/ for (i = 0; i < m; i += 1<<3) { lv = lightvis[i>>3];// & vvis[i>>3]; if (!lv) continue; if ((lv&0x01) && wl[i+0].visframe == r_visframecount) return true; if ((lv&0x02) && wl[i+1].visframe == r_visframecount) return true; if ((lv&0x04) && wl[i+2].visframe == r_visframecount) return true; if ((lv&0x08) && wl[i+3].visframe == r_visframecount) return true; if ((lv&0x10) && wl[i+4].visframe == r_visframecount) return true; if ((lv&0x20) && wl[i+5].visframe == r_visframecount) return true; if ((lv&0x40) && wl[i+6].visframe == r_visframecount) return true; if ((lv&0x80) && wl[i+7].visframe == r_visframecount) return true; } return false; } #endif /* static void Sh_Scissor (srect_t *r) { //float xs = vid.pixelwidth / (float)vid.width, ys = vid.pixelheight / (float)vid.height; switch(qrenderer) { case QR_NONE: case QR_SOFTWARE: case QR_DIRECT3D11: default: break; case QR_OPENGL: #ifdef GLQUAKE qglScissor( floor(r_refdef.pxrect.x + r->x*r_refdef.pxrect.width), floor((r_refdef.pxrect.y + r->y*r_refdef.pxrect.height) - r_refdef.pxrect.height), ceil(r->width * r_refdef.pxrect.width), ceil(r->height * r_refdef.pxrect.height)); qglEnable(GL_SCISSOR_TEST); if (qglDepthBoundsEXT) { qglDepthBoundsEXT(r->dmin, r->dmax); qglEnable(GL_DEPTH_BOUNDS_TEST_EXT); } #endif break; case QR_DIRECT3D9: #ifdef D3D9QUAKE { RECT rect; rect.left = r->x; rect.right = r->x + r->width; rect.top = r->y; rect.bottom = r->y + r->height; IDirect3DDevice9_SetScissorRect(pD3DDev9, &rect); } #endif break; } } static void Sh_ScissorOff (void) { switch(qrenderer) { default: break; case QR_OPENGL: #ifdef GLQUAKE qglDisable(GL_SCISSOR_TEST); if (qglDepthBoundsEXT) qglDisable(GL_DEPTH_BOUNDS_TEST_EXT); #endif break; case QR_DIRECT3D9: #ifdef D3D9QUAKE #endif break; } } */ #if 0 static qboolean Sh_ScissorForSphere(vec3_t center, float radius, vrect_t *rect) { /*return false to say that its fully offscreen*/ float v[4], tempv[4]; int i; vrect_t r; rect->x = 0; rect->y = 0; rect->width = vid.pixelwidth; rect->height = vid.pixelheight; /* for (i = 0; i < 4; i++) { v[3] = 1; VectorMA(center, radius, frustum[i].normal, v); tempv[0] = r_refdef.m_view[0]*v[0] + r_refdef.m_view[4]*v[1] + r_refdef.m_view[8]*v[2] + r_refdef.m_view[12]*v[3]; tempv[1] = r_refdef.m_view[1]*v[0] + r_refdef.m_view[5]*v[1] + r_refdef.m_view[9]*v[2] + r_refdef.m_view[13]*v[3]; tempv[2] = r_refdef.m_view[2]*v[0] + r_refdef.m_view[6]*v[1] + r_refdef.m_view[10]*v[2] + r_refdef.m_view[14]*v[3]; tempv[3] = r_refdef.m_view[3]*v[0] + r_refdef.m_view[7]*v[1] + r_refdef.m_view[11]*v[2] + r_refdef.m_view[15]*v[3]; product[0] = r_refdef.m_projection[0]*tempv[0] + r_refdef.m_projection[4]*tempv[1] + r_refdef.m_projection[8]*tempv[2] + r_refdef.m_projection[12]*tempv[3]; product[1] = r_refdef.m_projection[1]*tempv[0] + r_refdef.m_projection[5]*tempv[1] + r_refdef.m_projection[9]*tempv[2] + r_refdef.m_projection[13]*tempv[3]; product[2] = r_refdef.m_projection[2]*tempv[0] + r_refdef.m_projection[6]*tempv[1] + r_refdef.m_projection[10]*tempv[2] + r_refdef.m_projection[14]*tempv[3]; product[3] = r_refdef.m_projection[3]*tempv[0] + r_refdef.m_projection[7]*tempv[1] + r_refdef.m_projection[11]*tempv[2] + r_refdef.m_projection[15]*tempv[3]; v[0] /= v[3]; v[1] /= v[3]; v[2] /= v[3]; out[0] = (1+v[0])/2; out[1] = (1+v[1])/2; out[2] = (1+v[2])/2; r.x } */ return false; } #endif #define BoxesOverlap(a,b,c,d) ((a)[0] <= (d)[0] && (b)[0] >= (c)[0] && (a)[1] <= (d)[1] && (b)[1] >= (c)[1] && (a)[2] <= (d)[2] && (b)[2] >= (c)[2]) static qboolean Sh_ScissorForBox(vec3_t mins, vec3_t maxs, srect_t *r) { static const int edge[12][2] = { {0, 1}, {0, 2}, {1, 3}, {2, 3}, {4, 5}, {4, 6}, {5, 7}, {6, 7}, {0, 4}, {1, 5}, {2, 6}, {3, 7} }; //the box is a simple cube. //clip each vert to the near clip plane //insert a replacement vertex for edges that cross the nearclip plane where it crosses //calc the scissor rect from projecting the verts that survived, plus the clipped edge ones. float ncpdist; float dist[8]; int sign[8]; vec4_t vert[20]; vec3_t p[8]; int numverts = 0, i, v1, v2; vec4_t v,tv; float frac; float x,x1,x2,y,y1,y2; double z, z1, z2; r->x = 0; r->y = 0; r->width = 1; r->height = 1; r->dmin = 0; r->dmax = 1; if (!r_shadow_scissor.ival) { r->x = 0; r->y = 0; r->width = 1; r->height = 1; return false; } /*if view is inside the box, then skip this maths*/ // if (BoxesOverlap(r_refdef.vieworg, r_refdef.vieworg, mins, maxs)) // { // return false; // } ncpdist = DotProduct(r_refdef.vieworg, vpn) + r_refdef.mindist; for (i = 0; i < 8; i++) { p[i][0] = (i & 1) ? mins[0] : maxs[0]; p[i][1] = (i & 2) ? mins[1] : maxs[1]; p[i][2] = (i & 4) ? mins[2] : maxs[2]; dist[i] = ncpdist - DotProduct(p[i], vpn); sign[i] = (dist[i] > 0); if (!sign[i]) { VectorCopy(p[i], vert[numverts]); numverts++; } } /*fully clipped by near plane*/ if (!numverts) return true; if (numverts != 8) { /*crosses near clip plane somewhere*/ for (i = 0; i < 12; i++) { v1 = edge[i][0]; v2 = edge[i][1]; if (sign[v1] != sign[v2]) { frac = dist[v1] / (dist[v1] - dist[v2]); VectorInterpolate(p[v1], frac, p[v2], vert[numverts]); numverts++; } } } x1 = y1 = z1 = 1; x2 = y2 = z2 = -1; /*transform each vert to get the screen pos*/ for (i = 0; i < numverts; i++) { vert[i][3] = 1; Matrix4x4_CM_Transform4(r_refdef.m_view, vert[i], tv); Matrix4x4_CM_Transform4(r_refdef.m_projection_std, tv, v); x = v[0] / v[3]; y = v[1] / v[3]; z = (double)v[2] / v[3]; if (x < x1) x1 = x; if (x > x2) x2 = x; if (y < y1) y1 = y; if (y > y2) y2 = y; if (z < z1) z1 = z; if (z > z2) z2 = z; } x1 = (1+x1) / 2; x2 = (1+x2) / 2; y1 = (1+y1) / 2; y2 = (1+y2) / 2; z1 = (1+z1) / 2; z2 = (1+z2) / 2; if (x1 < 0) x1 = 0; if (y1 < 0) y1 = 0; if (x2 < 0) x2 = 0; if (y2 < 0) y2 = 0; if (x1 > 1) x1 = 1; if (y1 > 1) y1 = 1; if (x2 > 1) x2 = 1; if (y2 > 1) y2 = 1; r->x = x1; r->y = y1; r->width = x2 - r->x; r->height = y2 - r->y; if (r->width == 0 || r->height == 0) return true; //meh r->dmin = z1; r->dmax = z2; return false; } #if 0 static qboolean Sh_ScissorForBox(vec3_t mins, vec3_t maxs, vrect_t *r) { int i, ix1, iy1, ix2, iy2; float x1, y1, x2, y2, x, y, f; vec3_t smins, smaxs; vec4_t v, v2; r->x = 0; r->y = 0; r->width = vid.pixelwidth; r->height = vid.pixelheight; if (0)//!r_shadow_scissor.integer) { return false; } // if view is inside the box, just say yes it's fully visible if (BoxesOverlap(r_refdef.vieworg, r_refdef.vieworg, mins, maxs)) { return false; } for (i = 0;i < 3;i++) { if (vpn[i] >= 0) { v[i] = mins[i]; v2[i] = maxs[i]; } else { v[i] = maxs[i]; v2[i] = mins[i]; } } f = DotProduct(vpn, r_refdef.vieworg); if (DotProduct(vpn, v2) <= f) { // entirely behind nearclip plane, entirely obscured return true; } if (DotProduct(vpn, v) >= f) { // entirely infront of nearclip plane x1 = y1 = x2 = y2 = 0; for (i = 0;i < 8;i++) { v[0] = (i & 1) ? mins[0] : maxs[0]; v[1] = (i & 2) ? mins[1] : maxs[1]; v[2] = (i & 4) ? mins[2] : maxs[2]; v[3] = 1.0f; Matrix4x4_CM_Project(v, v2, r_refdef.viewangles, r_refdef.vieworg, r_refdef.fov_x, r_refdef.fov_y); v2[0]*=vid.pixelwidth; v2[1]*=vid.pixelheight; // GL_TransformToScreen(v, v2); //Con_Printf("%.3f %.3f %.3f %.3f transformed to %.3f %.3f %.3f %.3f\n", v[0], v[1], v[2], v[3], v2[0], v2[1], v2[2], v2[3]); x = v2[0]; y = v2[1]; if (i) { if (x1 > x) x1 = x; if (x2 < x) x2 = x; if (y1 > y) y1 = y; if (y2 < y) y2 = y; } else { x1 = x2 = x; y1 = y2 = y; } } } else { // clipped by nearclip plane // this is nasty and crude... // create viewspace bbox i = 0; /*unrolled the first iteration to avoid warnings*/ v[0] = ((i & 1) ? mins[0] : maxs[0]) - r_refdef.vieworg[0]; v[1] = ((i & 2) ? mins[1] : maxs[1]) - r_refdef.vieworg[1]; v[2] = ((i & 4) ? mins[2] : maxs[2]) - r_refdef.vieworg[2]; v2[0] = DotProduct(v, vright); v2[1] = DotProduct(v, vup); v2[2] = DotProduct(v, vpn); smins[0] = smaxs[0] = v2[0]; smins[1] = smaxs[1] = v2[1]; smins[2] = smaxs[2] = v2[2]; for (i = 1;i < 8;i++) { v[0] = ((i & 1) ? mins[0] : maxs[0]) - r_refdef.vieworg[0]; v[1] = ((i & 2) ? mins[1] : maxs[1]) - r_refdef.vieworg[1]; v[2] = ((i & 4) ? mins[2] : maxs[2]) - r_refdef.vieworg[2]; v2[0] = DotProduct(v, vright); v2[1] = DotProduct(v, vup); v2[2] = DotProduct(v, vpn); if (smins[0] > v2[0]) smins[0] = v2[0]; if (smaxs[0] < v2[0]) smaxs[0] = v2[0]; if (smins[1] > v2[1]) smins[1] = v2[1]; if (smaxs[1] < v2[1]) smaxs[1] = v2[1]; if (smins[2] > v2[2]) smins[2] = v2[2]; if (smaxs[2] < v2[2]) smaxs[2] = v2[2]; } // now we have a bbox in viewspace // clip it to the view plane if (smins[2] < 1) smins[2] = 1; // return true if that culled the box if (smins[2] >= smaxs[2]) return true; // ok some of it is infront of the view, transform each corner back to // worldspace and then to screenspace and make screen rect // initialize these variables just to avoid compiler warnings x1 = y1 = x2 = y2 = 0; for (i = 0;i < 8;i++) { v2[0] = (i & 1) ? smins[0] : smaxs[0]; v2[1] = (i & 2) ? smins[1] : smaxs[1]; v2[2] = (i & 4) ? smins[2] : smaxs[2]; v[0] = v2[0] * vright[0] + v2[1] * vup[0] + v2[2] * vpn[0] + r_refdef.vieworg[0]; v[1] = v2[0] * vright[1] + v2[1] * vup[1] + v2[2] * vpn[1] + r_refdef.vieworg[1]; v[2] = v2[0] * vright[2] + v2[1] * vup[2] + v2[2] * vpn[2] + r_refdef.vieworg[2]; v[3] = 1.0f; Matrix4x4_CM_Project(v, v2, r_refdef.viewangles, r_refdef.vieworg, r_refdef.fov_x, r_refdef.fov_y); v2[0]*=vid.pixelwidth; v2[1]*=vid.pixelheight; //Con_Printf("%.3f %.3f %.3f %.3f transformed to %.3f %.3f %.3f %.3f\n", v[0], v[1], v[2], v[3], v2[0], v2[1], v2[2], v2[3]); x = v2[0]; y = v2[1]; if (i) { if (x1 > x) x1 = x; if (x2 < x) x2 = x; if (y1 > y) y1 = y; if (y2 < y) y2 = y; } else { x1 = x2 = x; y1 = y2 = y; } } #if 1 // this code doesn't handle boxes with any points behind view properly x1 = 1000;x2 = -1000; y1 = 1000;y2 = -1000; for (i = 0;i < 8;i++) { v[0] = (i & 1) ? mins[0] : maxs[0]; v[1] = (i & 2) ? mins[1] : maxs[1]; v[2] = (i & 4) ? mins[2] : maxs[2]; v[3] = 1.0f; Matrix4x4_CM_Project(v, v2, r_refdef.viewangles, r_refdef.vieworg, r_refdef.fov_x, r_refdef.fov_y); v2[0]*=vid.pixelwidth; v2[1]*=vid.pixelheight; //Con_Printf("%.3f %.3f %.3f %.3f transformed to %.3f %.3f %.3f %.3f\n", v[0], v[1], v[2], v[3], v2[0], v2[1], v2[2], v2[3]); if (v2[2] > 0) { x = v2[0]; y = v2[1]; if (x1 > x) x1 = x; if (x2 < x) x2 = x; if (y1 > y) y1 = y; if (y2 < y) y2 = y; } } #endif } ix1 = x1 - 1.0f; iy1 = y1 - 1.0f; ix2 = x2 + 1.0f; iy2 = y2 + 1.0f; //Con_Printf("%f %f %f %f\n", x1, y1, x2, y2); if (ix1 < r->x) ix1 = r->x; if (iy1 < r->y) iy1 = r->y; if (ix2 > r->x + r->width) ix2 = r->x + r->width; if (iy2 > r->y + r->height) iy2 = r->y + r->height; if (ix2 <= ix1 || iy2 <= iy1) return true; // set up the scissor rectangle r->x = ix1; r->y = iy1; r->width = ix2 - ix1; r->height = iy2 - iy1; return false; } #endif void D3D11BE_BeginShadowmapFace(void); //determine the 5 bounding points of a shadowmap light projection side //needs to match Sh_GenShadowFace static void Sh_LightFrustumPlanes(dlight_t *l, vec3_t axis[3], vec4_t *planes, int face) { vec3_t tmp; int axis0, axis1, axis2; int dir; int i; //+x,+y,+z,-x,-y,-z axis0 = (face+0)%3; //our major axis axis1 = (face+1)%3; axis2 = (face+2)%3; dir = (face >= 3)?-1:1; //center point is always the same VectorCopy(l->origin, planes[4]); VectorScale(axis[axis0], dir, planes[4]); VectorNormalize(planes[4]); planes[4][3] = r_shadow_shadowmapping_nearclip.value + DotProduct(planes[4], l->origin); for (i = 0; i < 4; i++) { VectorScale(axis[axis0], dir, tmp); VectorMA(tmp, ((i&1)?1:-1), axis[axis1], tmp); VectorMA(tmp, ((i&2)?1:-1), axis[axis2], planes[i]); VectorNormalize(planes[i]); planes[i][3] = DotProduct(planes[i], l->origin); } } //culling for the face happens in the caller. //these faces should thus match Sh_LightFrustumPlanes static void Sh_GenShadowFace(dlight_t *l, vec3_t axis[3], int lighttype, shadowmesh_t *smesh, int face, int smsize, float proj[16]) { vec3_t t1,t2,t3; texture_t *tex; int tno; /* if (face >= 3) face -= 3; else face += 3; */ switch(face) { case 0: //down VectorCopy(axis[0], t1); VectorCopy(axis[1], t2); VectorCopy(axis[2], t3); Matrix4x4_CM_LightMatrixFromAxis(r_refdef.m_view, t1, t2, t3, l->origin); r_refdef.flipcull = 0; break; case 1: //back VectorCopy(axis[2], t1); VectorCopy(axis[1], t2); VectorCopy(axis[0], t3); Matrix4x4_CM_LightMatrixFromAxis(r_refdef.m_view, t1, t2, t3, l->origin); r_refdef.flipcull = SHADER_CULL_FLIP; break; case 2: //right VectorCopy(axis[0], t1); VectorCopy(axis[2], t2); VectorCopy(axis[1], t3); Matrix4x4_CM_LightMatrixFromAxis(r_refdef.m_view, t1, t2, t3, l->origin); r_refdef.flipcull = SHADER_CULL_FLIP; break; case 3: //up VectorCopy(axis[0], t1); VectorCopy(axis[1], t2); VectorCopy(axis[2], t3); VectorNegate(t3, t3); Matrix4x4_CM_LightMatrixFromAxis(r_refdef.m_view, t1, t2, t3, l->origin); r_refdef.flipcull = SHADER_CULL_FLIP; break; case 4: //forward VectorCopy(axis[2], t1); VectorCopy(axis[1], t2); VectorCopy(axis[0], t3); VectorNegate(t3, t3); Matrix4x4_CM_LightMatrixFromAxis(r_refdef.m_view, t1, t2, t3, l->origin); r_refdef.flipcull = 0; break; case 5: //left VectorCopy(axis[0], t1); VectorCopy(axis[2], t2); VectorCopy(axis[1], t3); VectorNegate(t3, t3); Matrix4x4_CM_LightMatrixFromAxis(r_refdef.m_view, t1, t2, t3, l->origin); r_refdef.flipcull = 0; break; } if (lighttype & (LSHADER_SPOT|LSHADER_ORTHO)) { r_refdef.pxrect.x = (SHADOWMAP_SIZE-smsize)/2; r_refdef.pxrect.width = smsize; r_refdef.pxrect.height = smsize; r_refdef.pxrect.y = (SHADOWMAP_SIZE-smsize)/2; r_refdef.pxrect.maxheight = SHADOWMAP_SIZE; } else { r_refdef.pxrect.x = (face%3 * SHADOWMAP_SIZE) + (SHADOWMAP_SIZE-smsize)/2; r_refdef.pxrect.width = smsize; r_refdef.pxrect.height = smsize; r_refdef.pxrect.y = (((face<3)*SHADOWMAP_SIZE) + (SHADOWMAP_SIZE-smsize)/2); r_refdef.pxrect.maxheight = SHADOWMAP_SIZE*2; } R_SetFrustum(proj, r_refdef.m_view); if (lighttype & LSHADER_ORTHO) r_refdef.frustum_numplanes = 4; //kill the near clip plane - we allow ANYTHING nearer through. #ifdef SHADOWDBG_COLOURNOTDEPTH BE_SelectMode(BEM_STANDARD); #else BE_SelectMode(BEM_DEPTHONLY); #endif BE_SelectEntity(&r_worldentity); switch(qrenderer) { #ifdef GLQUAKE case QR_OPENGL: GL_ViewportUpdate(); if (lighttype & LSHADER_ORTHO) qglEnable(GL_DEPTH_CLAMP_ARB); GL_CullFace(SHADER_CULL_FRONT); GLBE_RenderShadowBuffer(smesh->numverts, smesh->vebo[0], smesh->verts, smesh->numindicies, smesh->vebo[1], smesh->indicies); break; #endif #ifdef VKQUAKE case QR_VULKAN: //FIXME: generate a single commandbuffer (requires full separation of viewprojection matrix) VKBE_BeginShadowmapFace(); VKBE_RenderShadowBuffer(smesh->vkbuffer); break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: //opengl render targets are upside down - our code kinda assumes gl r_refdef.pxrect.y = r_refdef.pxrect.maxheight -(r_refdef.pxrect.y+r_refdef.pxrect.height); D3D11BE_BeginShadowmapFace(); D3D11BE_RenderShadowBuffer(smesh->numverts, smesh->d3d11_vbuffer, smesh->numindicies, smesh->d3d11_ibuffer); break; #endif default: //FIXME: should be able to merge batches between textures+lightmaps. for (tno = 0; tno < smesh->numbatches; tno++) { if (!smesh->batches[tno].count) continue; tex = cl.worldmodel->shadowbatches[tno].tex; if (tex->shader->flags & (SHADER_NOSHADOWS|SHADER_NODRAW)) //FIXME: shadows not lights continue; BE_DrawMesh_List(tex->shader, smesh->batches[tno].count, smesh->batches[tno].s, cl.worldmodel->shadowbatches[tno].vbo, NULL, 0); } break; } //fixme: this walks through the entity lists up to 6 times per frame per entity. switch(qrenderer) { default: break; #ifdef GLQUAKE case QR_OPENGL: GLBE_BaseEntTextures(); if (lighttype & LSHADER_ORTHO) qglDisable(GL_DEPTH_CLAMP_ARB); break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: D3D9BE_BaseEntTextures(); break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: D3D11BE_BaseEntTextures(); break; #endif #ifdef VKQUAKE case QR_VULKAN: VKBE_BaseEntTextures(); break; #endif } /* { int i; static float depth[SHADOWMAP_SIZE*SHADOWMAP_SIZE]; qglReadPixels(0, 0, smsize, smsize, GL_DEPTH_COMPONENT, GL_FLOAT, depth); for (i = SHADOWMAP_SIZE*SHADOWMAP_SIZE; i --> 0; ) { if (depth[i] == 1) *((unsigned int*)depth+i) = 0; else *((unsigned int*)depth+i) = 0xff000000|((((unsigned char)(int)(depth[i]*128)))*0x10101); } qglTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, smsize, smsize, 0, GL_RGBA, GL_UNSIGNED_BYTE, depth); qglTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR); qglTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR); qglTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP); qglTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP); } */ } qboolean Sh_GenShadowMap (dlight_t *l, int lighttype, vec3_t axis[3], qbyte *lvis, int smsize) { int restorefbo = 0; int f,lf; float oprojs[16], oprojv[16], oview[16]; pxrect_t oprect; shadowmesh_t *smesh; qboolean isspot = !!(lighttype & (LSHADER_SPOT|LSHADER_ORTHO)); int sidevisible; int oldflip = r_refdef.flipcull; int oldexternalview = r_refdef.externalview; if (isspot) { //spotlights only face forwards. which is side 4. which is annoying. f = 4; lf = f+1; sidevisible = 1<origin, 0)) { //if the light's center isn't onscreen, cull individual faces //FIXME: if the fov is < 90, we need to clip by the near lightplane first for (; f < lf; f++) { vec4_t planes[5]; float dist; int fp,lp; Sh_LightFrustumPlanes(l, axis, planes, f); for (fp = 0; fp < r_refdef.frustum_numplanes; fp++) { vec3_t nearest; //make a guess based upon the frustum plane VectorMA(l->origin, l->radius, r_refdef.frustum[fp].normal, nearest); //clip that point to the various planes for(lp = 0; lp < 5; lp++) { dist = DotProduct(nearest, planes[lp]) - planes[lp][3]; if (dist < 0) VectorMA(nearest, dist, planes[lp], nearest); } // P_RunParticleEffect(nearest, vec3_origin, 15, 1); //give up if the best point for any frustum plane is offscreen dist = DotProduct(r_refdef.frustum[fp].normal, nearest) - r_refdef.frustum[fp].dist; if (dist <= 0) break; } if (fp != r_refdef.frustum_numplanes) sidevisible &= ~(1u<fov, l->fov, r_shadow_shadowmapping_nearclip.value, l->radius, false); else if (lighttype & LSHADER_ORTHO) { float xmin = -l->radius; float ymin = -l->radius; float znear = -l->radius; float xmax = l->radius; float ymax = l->radius; float zfar = l->radius; Matrix4x4_CM_Orthographic(r_refdef.m_projection_std, xmin, xmax, ymax, ymin, znear, zfar); } else Matrix4x4_CM_Projection_Far(r_refdef.m_projection_std, 90, 90, r_shadow_shadowmapping_nearclip.value, l->radius, false); memcpy(r_refdef.m_projection_view, r_refdef.m_projection_std, sizeof(r_refdef.m_projection_view)); switch(qrenderer) { default: return false; #ifdef GLQUAKE case QR_OPENGL: if (!GLBE_BeginShadowMap(isspot, (isspot?SHADOWMAP_SIZE:(SHADOWMAP_SIZE*3)), (isspot?SHADOWMAP_SIZE:(SHADOWMAP_SIZE*2)), &restorefbo)) return false; break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: if (!D3D11_BeginShadowMap(isspot, (isspot?SHADOWMAP_SIZE:(SHADOWMAP_SIZE*3)), (isspot?SHADOWMAP_SIZE:(SHADOWMAP_SIZE*2)))) return false; break; #endif #ifdef VKQUAKE case QR_VULKAN: if (!VKBE_BeginShadowmap(isspot, (isspot?SHADOWMAP_SIZE:(SHADOWMAP_SIZE*3)), (isspot?SHADOWMAP_SIZE:(SHADOWMAP_SIZE*2)))) return false; break; #endif } r_refdef.externalview = true; //never any viewmodels /*generate faces*/ for (f = 0; f < 6; f++) { if (sidevisible & (1u<rebuildcache && l->worldshadowmesh) { lvis = l->worldshadowmesh->litleaves; //fixme: check head node first? if (!Sh_LeafInView(l->worldshadowmesh->litleaves, vvis)) { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return false; } } else { int clus; clus = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, l->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, clus, &lvisb, PVM_FAST); //FIXME: surely we can use the phs for this? if (!Sh_VisOverlaps(lvis, vvis)) //The two viewing areas do not intersect. { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return false; } } } else lvis = NULL; if (lighttype & (LSHADER_SPOT | LSHADER_ORTHO)) texwidth = texheight = smsize = SHADOWMAP_SIZE; else { //Stolen from DP. Actually, LH pasted it to me in IRC. vec3_t nearestpoint; vec3_t d; float distance, lodlinear; nearestpoint[0] = bound(l->origin[0]-l->radius, r_origin[0], l->origin[0]+l->radius); nearestpoint[1] = bound(l->origin[1]-l->radius, r_origin[1], l->origin[1]+l->radius); nearestpoint[2] = bound(l->origin[2]-l->radius, r_origin[2], l->origin[2]+l->radius); VectorSubtract(nearestpoint, r_origin, d); distance = VectorLength(d); lodlinear = (l->radius * r_shadow_shadowmapping_precision.value) / sqrt(max(1.0f, distance / l->radius)); smsize = bound(16, lodlinear, SHADOWMAP_SIZE); texwidth = smsize*3; texheight = smsize*2; } switch(qrenderer) { #ifdef GLQUAKE case QR_OPENGL: GLBE_SetupForShadowMap(l, texwidth, texheight, (smsize-4) / (float)SHADOWMAP_SIZE); break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: D3D11BE_SetupForShadowMap(l, texwidth, texheight, (smsize-4) / (float)SHADOWMAP_SIZE); break; #endif #ifdef VKQUAKE case QR_VULKAN: VKBE_SetupForShadowMap(l, texwidth, texheight, (smsize-4) / (float)SHADOWMAP_SIZE); break; #endif default: (void)texwidth; (void)texheight; break; } //fixme: light rotation if (!Sh_GenShadowMap(l, lighttype, l->axis, lvis, smsize)) return false; //didn't need to do anything return true; } static void Sh_DrawShadowMapLight(dlight_t *l, vec3_t colour, vec3_t axis[3], qbyte *vvis) { vec3_t mins, maxs; qbyte *lvis; srect_t rect; int smsize; int lighttype; int texwidth, texheight; if (l->fov != 0) lighttype = LSHADER_SMAP|LSHADER_SPOT; #ifdef LFLAG_ORTHO else if (l->flags & LFLAG_ORTHO) lighttype = LSHADER_SMAP|LSHADER_ORTHO; #endif else lighttype = LSHADER_SMAP; if (R_CullSphere(l->origin, l->radius)) { RQuantAdd(RQUANT_RTLIGHT_CULL_FRUSTUM, 1); return; //this should be the more common case } mins[0] = l->origin[0] - l->radius; mins[1] = l->origin[1] - l->radius; mins[2] = l->origin[2] - l->radius; maxs[0] = l->origin[0] + l->radius; maxs[1] = l->origin[1] + l->radius; maxs[2] = l->origin[2] + l->radius; if (Sh_ScissorForBox(mins, maxs, &rect)) { RQuantAdd(RQUANT_RTLIGHT_CULL_SCISSOR, 1); return; } if (vvis) { if (!l->rebuildcache && l->worldshadowmesh) { lvis = l->worldshadowmesh->litleaves; //fixme: check head node first? if (!Sh_LeafInView(l->worldshadowmesh->litleaves, vvis)) { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return; } } else { int clus; clus = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, l->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, clus, &lvisb, PVM_FAST); //FIXME: surely we can use the phs for this? if (cl.worldmodel->funcs.ClustersInSphere) lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, l->origin, l->radius, &lvisb2, lvis); if (!Sh_VisOverlaps(lvis, vvis)) //The two viewing areas do not intersect. { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return; } } } else lvis = NULL; if (lighttype & LSHADER_SPOT) { smsize = SHADOWMAP_SIZE; //spot lights or ortho lights can just use the full thing. texwidth = smsize; texheight = smsize; } else if (lighttype & LSHADER_ORTHO) { smsize = SHADOWMAP_SIZE; //spot lights or ortho lights can just use the full thing. texwidth = smsize; texheight = smsize; } else { //Stolen from DP. Actually, LH pasted it to me in IRC. vec3_t nearestpoint; vec3_t d; float distance, lodlinear; nearestpoint[0] = bound(l->origin[0]-l->radius, r_origin[0], l->origin[0]+l->radius); nearestpoint[1] = bound(l->origin[1]-l->radius, r_origin[1], l->origin[1]+l->radius); nearestpoint[2] = bound(l->origin[2]-l->radius, r_origin[2], l->origin[2]+l->radius); VectorSubtract(nearestpoint, r_origin, d); distance = VectorLength(d); lodlinear = (l->radius * r_shadow_shadowmapping_precision.value) / sqrt(max(1.0f, distance / l->radius)); smsize = bound(16, lodlinear, SHADOWMAP_SIZE); texwidth = smsize*3; texheight = smsize*2; } switch(qrenderer) { #ifdef GLQUAKE case QR_OPENGL: GLBE_SetupForShadowMap(l, texwidth, texheight, (smsize-4) / (float)SHADOWMAP_SIZE); break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: D3D11BE_SetupForShadowMap(l, texwidth, texheight, (smsize-4) / (float)SHADOWMAP_SIZE); break; #endif #ifdef VKQUAKE case QR_VULKAN: VKBE_SetupForShadowMap(l, texwidth, texheight, (smsize-4) / (float)SHADOWMAP_SIZE); break; #endif default: (void)texwidth; (void)texheight; break; } if (!BE_SelectDLight(l, colour, axis, lighttype)) return; if (!Sh_GenShadowMap(l, lighttype, axis, lvis, smsize)) return; RQuantAdd(RQUANT_RTLIGHT_DRAWN, 1); //may as well use scissors BE_Scissor(&rect); BE_SelectEntity(&r_worldentity); BE_SelectMode(BEM_LIGHT); Sh_DrawEntLighting(l, colour); } /* draws faces facing the light Note: Backend mode must have been selected in advance, as must the light to light from */ static void Sh_DrawEntLighting(dlight_t *light, vec3_t colour) { int tno; texture_t *tex; shader_t *shader; shadowmesh_t *sm; sm = light->worldshadowmesh; if (light->rebuildcache) sm = &sh_tempshmesh; if (sm) { for (tno = 0; tno < sm->numbatches; tno++) { if (!sm->batches[tno].count) continue; tex = cl.worldmodel->shadowbatches[tno].tex; if (cl.worldmodel->fromgame == fg_quake2) shader = R_TextureAnimation_Q2(tex)->shader; else shader = R_TextureAnimation(false, tex)->shader; if (shader->flags & (SHADER_NODLIGHT|SHADER_NODRAW|SHADER_SKY)) continue; //FIXME: it should be worth building a dedicated ebo, for static ones BE_DrawMesh_List(shader, sm->batches[tno].count, sm->batches[tno].s, cl.worldmodel->shadowbatches[tno].vbo, NULL, 0); RQuantAdd(RQUANT_LITFACES, sm->batches[tno].count); } switch(qrenderer) { default: break; #ifdef GLQUAKE case QR_OPENGL: GLBE_BaseEntTextures(); break; #endif #ifdef VKQUAKE case QR_VULKAN: VKBE_BaseEntTextures(); break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: D3D9BE_BaseEntTextures(); break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: D3D11BE_BaseEntTextures(); break; #endif } } } #ifdef GLQUAKE /*Fixme: this is brute forced*/ #ifdef warningmsg #pragma warningmsg("brush shadows are bruteforced") #endif static void Sh_DrawBrushModelShadow(dlight_t *dl, entity_t *e) { int v; float *v1, *v2; vec3_t v3, v4; vec3_t lightorg; int i; model_t *model; msurface_t *surf; if (qrenderer != QR_OPENGL) return; if (BE_LightCullModel(e->origin, e->model)) return; RotateLightVector((void *)e->axis, e->origin, dl->origin, lightorg); BE_SelectEntity(e); GL_DeselectVAO(); GL_SelectVBO(0); GL_SelectEBO(0); qglEnableClientState(GL_VERTEX_ARRAY); #ifdef BEF_PUSHDEPTH GLBE_PolyOffsetStencilShadow(r_pushdepth); #else GLBE_PolyOffsetStencilShadow(); #endif model = e->model; surf = model->surfaces+model->firstmodelsurface; for (i = 0; i < model->nummodelsurfaces; i++, surf++) { if (surf->flags & SURF_PLANEBACK) {//inverted normal. if (DotProduct(surf->plane->normal, lightorg)-surf->plane->dist >= -0.1) continue; } else { if (DotProduct(surf->plane->normal, lightorg)-surf->plane->dist <= 0.1) continue; } if (surf->flags & (SURF_DRAWALPHA | SURF_DRAWTILED)) { // no shadows continue; } if (!surf->mesh) continue; //front face qglVertexPointer(3, GL_FLOAT, sizeof(vecV_t), surf->mesh->xyz_array); qglDrawArrays(GL_POLYGON, 0, surf->mesh->numvertexes); // qglDrawRangeElements(GL_TRIANGLES, 0, surf->mesh->numvertexes, surf->mesh->numindexes, GL_INDEX_TYPE, surf->mesh->indexes); RQuantAdd(RQUANT_SHADOWINDICIES, surf->mesh->numvertexes); for (v = 0; v < surf->mesh->numvertexes; v++) { //border v1 = surf->mesh->xyz_array[v]; v2 = surf->mesh->xyz_array[( v+1 )%surf->mesh->numvertexes]; //get positions of v3 and v4 based on the light position v3[0] = ( v1[0]-lightorg[0] ); v3[1] = ( v1[1]-lightorg[1] ); v3[2] = ( v1[2]-lightorg[2] ); VectorNormalizeFast(v3); VectorScale(v3, PROJECTION_DISTANCE, v3); v4[0] = ( v2[0]-lightorg[0] ); v4[1] = ( v2[1]-lightorg[1] ); v4[2] = ( v2[2]-lightorg[2] ); VectorNormalizeFast(v4); VectorScale(v4, PROJECTION_DISTANCE, v4); //Now draw the quad from the two verts to the projected light //verts qglBegin( GL_QUAD_STRIP ); qglVertex3fv(v1); qglVertex3f (v1[0]+v3[0], v1[1]+v3[1], v1[2]+v3[2]); qglVertex3fv(v2); qglVertex3f (v2[0]+v4[0], v2[1]+v4[1], v2[2]+v4[2]); qglEnd(); } //back //the same applies as earlier qglBegin(GL_POLYGON); for (v = surf->mesh->numvertexes-1; v >=0; v--) { v1 = surf->mesh->xyz_array[v]; v3[0] = (v1[0]-lightorg[0]); v3[1] = (v1[1]-lightorg[1]); v3[2] = (v1[2]-lightorg[2]); VectorNormalizeFast(v3); VectorScale(v3, PROJECTION_DISTANCE, v3); qglVertex3f(v1[0]+v3[0], v1[1]+v3[1], v1[2]+v3[2]); } qglEnd(); } #ifdef BEF_PUSHDEPTH GLBE_PolyOffsetStencilShadow(false); #else GLBE_PolyOffsetStencilShadow(); #endif } #endif #if defined(GLQUAKE) || defined(D3D9QUAKE) /*when this is called, the gl state has been set up to draw the stencil volumes using whatever extensions we have if secondside is set, then the gpu sucks and we're drawing stuff the slow 2-pass way, and this is the second pass. */ static void Sh_DrawStencilLightShadows(dlight_t *dl, qbyte *lvis, qbyte *vvis, qboolean secondside) { struct shadowmesh_s *sm; #ifdef GLQUAKE extern cvar_t gl_part_flame; int i; entity_t *ent; model_t *emodel; #endif sm = SHM_BuildShadowMesh(dl, lvis, SMT_STENCILVOLUME); if (!sm) { #ifdef GLQUAKE Sh_DrawBrushModelShadow(dl, &r_worldentity); #endif } else { switch (qrenderer) { case QR_NONE: case QR_SOFTWARE: default: break; #ifdef D3D11QUAKE // case QR_DIRECT3D11: // D3D11BE_RenderShadowBuffer(sm->numverts, sm->d3d11_vbuffer, sm->numindicies, sm->d3d11_ibuffer); // break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: D3D9BE_RenderShadowBuffer(sm->numverts, sm->d3d9_vbuffer, sm->numindicies, sm->d3d9_ibuffer); break; #endif #ifdef GLQUAKE case QR_OPENGL: GLBE_RenderShadowBuffer(sm->numverts, sm->vebo[0], sm->verts, sm->numindicies, sm->vebo[1], sm->indicies); break; #endif #ifdef VKQUAKE // case QR_VULKAN: // VKBE_RenderShadowBuffer(sm->numverts, sm->vebo[0], sm->verts, sm->numindicies, sm->vebo[1], sm->indicies); // break; #endif } } if (!r_drawentities.value) return; #ifdef GLQUAKE if (qrenderer != QR_OPENGL) return; //FIXME: still uses glBegin specifics. if (gl_config_nofixedfunc) return; /*too lazy to use shaders*/ if (gl_config_gles) return; //FIXME: uses glBegin // draw sprites seperately, because of alpha blending for (i=0 ; irtype != RT_MODEL) continue; if (ent->flags & (RF_NOSHADOW|Q2RF_BEAM)) continue; if (ent->keynum == dl->key && ent->keynum) continue; emodel = ent->model; if (!emodel) continue; if (cls.allow_anyparticles) //allowed or static { if (emodel->engineflags & MDLF_EMITREPLACE) { if (gl_part_flame.value) continue; } } if (emodel->loadstate == MLS_NOTLOADED) { if (!Mod_LoadModel(emodel, MLV_WARN)) continue; } if (emodel->loadstate != MLS_LOADED) continue; switch (emodel->type) { case mod_alias: if (r_drawentities.ival == 3) continue; R_DrawGAliasShadowVolume (ent, dl->origin, dl->radius); break; case mod_brush: if (r_drawentities.ival == 2) continue; Sh_DrawBrushModelShadow (dl, ent); break; case mod_sprite: //never any shadows on sprites, it doesn't really make sense. break; default: break; } } BE_SelectEntity(&r_worldentity); #endif } //draws a light using stencil shadows. //redraws world geometry up to 3 times per light... static qboolean Sh_DrawStencilLight(dlight_t *dl, vec3_t colour, vec3_t axis[3], qbyte *vvis) { int sref; int clus; qbyte *lvis; srect_t rect; vec3_t mins; vec3_t maxs; if (R_CullSphere(dl->origin, dl->radius)) { RQuantAdd(RQUANT_RTLIGHT_CULL_FRUSTUM, 1); return false; //this should be the more common case } mins[0] = dl->origin[0] - dl->radius; mins[1] = dl->origin[1] - dl->radius; mins[2] = dl->origin[2] - dl->radius; maxs[0] = dl->origin[0] + dl->radius; maxs[1] = dl->origin[1] + dl->radius; maxs[2] = dl->origin[2] + dl->radius; if (!dl->rebuildcache) { //fixme: check head node first? if (!Sh_LeafInView(dl->worldshadowmesh->litleaves, vvis)) { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return false; } lvis = NULL; } else { clus = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, dl->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, clus, &lvisb, PVM_FAST); // if (cl.worldmodel->funcs.ClustersInSphere) // lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb2, lvis); if (!Sh_VisOverlaps(lvis, vvis)) //The two viewing areas do not intersect. { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return false; } } //sets up the gl scissor (and culls to view) if (Sh_ScissorForBox(mins, maxs, &rect)) { RQuantAdd(RQUANT_RTLIGHT_CULL_SCISSOR, 1); return false; //this doesn't cull often. } RQuantAdd(RQUANT_RTLIGHT_DRAWN, 1); BE_SelectDLight(dl, colour, axis, LSHADER_STANDARD); BE_SelectMode(BEM_STENCIL); //The backend doesn't maintain scissor state. //The backend doesn't maintain stencil test state either - it needs to be active for more than just stencils, or disabled. its awkward. BE_Scissor(&rect); switch(qrenderer) { default: (void)sref; break; #ifdef GLQUAKE case QR_OPENGL: { int sfrontfail; int sbackfail; qglEnable(GL_STENCIL_TEST); //FIXME: is it practical to test to see if scissors allow not clearing the stencil buffer? /*we don't need all that much stencil buffer depth, and if we don't get enough or have dodgy volumes, wrap if we can*/ #ifdef I_LIVE_IN_A_FREE_COUNTRY sref = 0; sbackfail = GL_INCR; sfrontfail = GL_DECR; if (gl_config.ext_stencil_wrap) { //minimise damage... sbackfail = GL_INCR_WRAP_EXT; sfrontfail = GL_DECR_WRAP_EXT; } #else sref = (1<color[0], dl->color[1], dl->color[2], 1); qglDisable(GL_STENCIL_TEST); // qglEnable(GL_POLYGON_OFFSET_FILL); // qglPolygonOffset(-1, -1); // qglPolygonMode(GL_FRONT_AND_BACK, GL_LINE); Sh_DrawStencilLightShadows(dl, lvis, vvis, false); // qglDisable(GL_POLYGON_OFFSET_FILL); // qglPolygonMode(GL_FRONT_AND_BACK, GL_FILL); } #endif if (qglStencilOpSeparateATI) { //ATI/GLES/ARB method sref/=2; qglClearStencil(sref); qglClear(GL_STENCIL_BUFFER_BIT); GL_CullFace(0); qglStencilFunc(GL_ALWAYS, 0, ~0); qglStencilOpSeparateATI(GL_BACK, GL_KEEP, sbackfail, GL_KEEP); qglStencilOpSeparateATI(GL_FRONT, GL_KEEP, sfrontfail, GL_KEEP); Sh_DrawStencilLightShadows(dl, lvis, vvis, false); qglStencilOpSeparateATI(GL_FRONT_AND_BACK, GL_KEEP, GL_KEEP, GL_KEEP); GL_CullFace(SHADER_CULL_FRONT); qglStencilFunc(GL_EQUAL, sref, ~0); } else if (qglActiveStencilFaceEXT) { //Nvidia-specific method. sref/=2; qglClearStencil(sref); qglClear(GL_STENCIL_BUFFER_BIT); GL_CullFace(0); qglEnable(GL_STENCIL_TEST_TWO_SIDE_EXT); qglActiveStencilFaceEXT(GL_BACK); qglStencilOp(GL_KEEP, sbackfail, GL_KEEP); qglStencilFunc(GL_ALWAYS, 0, ~0 ); qglActiveStencilFaceEXT(GL_FRONT); qglStencilOp(GL_KEEP, sfrontfail, GL_KEEP); qglStencilFunc(GL_ALWAYS, 0, ~0 ); Sh_DrawStencilLightShadows(dl, lvis, vvis, false); qglActiveStencilFaceEXT(GL_BACK); qglStencilOp(GL_KEEP, GL_KEEP, GL_KEEP); qglStencilFunc(GL_ALWAYS, 0, ~0 ); qglActiveStencilFaceEXT(GL_FRONT); qglStencilOp(GL_KEEP, GL_KEEP, GL_KEEP); qglStencilFunc(GL_EQUAL, sref, ~0 ); qglDisable(GL_STENCIL_TEST_TWO_SIDE_EXT); } else //your graphics card sucks and lacks efficient stencil shadow techniques. { //centered around 0. Will only be increased then decreased less. qglClearStencil(sref); qglClear(GL_STENCIL_BUFFER_BIT); qglStencilFunc(GL_ALWAYS, 0, ~0); GL_CullFace(SHADER_CULL_BACK); qglStencilOp(GL_KEEP, sbackfail, GL_KEEP); Sh_DrawStencilLightShadows(dl, lvis, vvis, false); GL_CullFace(SHADER_CULL_FRONT); qglStencilOp(GL_KEEP, sfrontfail, GL_KEEP); Sh_DrawStencilLightShadows(dl, lvis, vvis, true); qglStencilOp(GL_KEEP, GL_KEEP, GL_KEEP); qglStencilFunc(GL_EQUAL, sref, ~0); } if (gl_config.arb_depth_clamp) qglDisable(GL_DEPTH_CLAMP_ARB); //end stencil writing. BE_SelectMode(BEM_LIGHT); Sh_DrawEntLighting(dl, colour); qglDisable(GL_STENCIL_TEST); qglStencilFunc( GL_ALWAYS, 0, ~0 ); } break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: sref = (1<<8)-1; sref/=2; /*clear the stencil buffer*/ IDirect3DDevice9_Clear(pD3DDev9, 0, NULL, D3DCLEAR_STENCIL, D3DCOLOR_XRGB(0, 0, 0), 1.0f, sref); /*set up 2-sided stenciling*/ D3D9BE_Cull(0); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILENABLE, true); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILFUNC, D3DCMP_ALWAYS); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_TWOSIDEDSTENCILMODE, true); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILFAIL, D3DSTENCILOP_KEEP); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILZFAIL, D3DSTENCILOP_DECR); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILPASS, D3DSTENCILOP_KEEP); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILFUNC, D3DCMP_ALWAYS); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_CCW_STENCILFAIL, D3DSTENCILOP_KEEP); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_CCW_STENCILZFAIL, D3DSTENCILOP_INCR); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_CCW_STENCILPASS, D3DSTENCILOP_KEEP); /*draw the shadows*/ Sh_DrawStencilLightShadows(dl, lvis, vvis, false); //disable stencil writing IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILZFAIL, D3DSTENCILOP_KEEP); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_TWOSIDEDSTENCILMODE, false); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILFUNC, D3DCMP_EQUAL); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILREF, sref); IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILMASK, ~0); /*draw the light*/ BE_SelectMode(BEM_LIGHT); Sh_DrawEntLighting(dl, colour); /*okay, no more stencil stuff*/ IDirect3DDevice9_SetRenderState(pD3DDev9, D3DRS_STENCILENABLE, false); break; #endif } return true; } #else #define Sh_DrawStencilLight Sh_DrawShadowlessLight #endif qboolean Sh_CullLight(dlight_t *dl, qbyte *vvis) { if (R_CullSphere(dl->origin, dl->radius)) { RQuantAdd(RQUANT_RTLIGHT_CULL_FRUSTUM, 1); return true; //this should be the more common case } if (!dl->rebuildcache) { //fixme: check head node first? if (!Sh_LeafInView(dl->worldshadowmesh->litleaves, vvis)) { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return true; } } else { int clus; qbyte *lvis; clus = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, dl->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, clus, &lvisb, PVM_FAST); // if (cl.worldmodel->funcs.ClustersInSphere) // lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb2, lvis); SHM_BuildShadowMesh(dl, lvis, SMT_DEFERRED); if (!Sh_VisOverlaps(lvis, vvis)) //The two viewing areas do not intersect. { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return true; } } return false; //please draw this... } static void Sh_DrawShadowlessLight(dlight_t *dl, vec3_t colour, vec3_t axis[3], qbyte *vvis) { vec3_t mins, maxs; srect_t rect; if (R_CullSphere(dl->origin, dl->radius)) { RQuantAdd(RQUANT_RTLIGHT_CULL_FRUSTUM, 1); return; //this should be the more common case } if (!dl->rebuildcache) { //fixme: check head node first? if (!Sh_LeafInView(dl->worldshadowmesh->litleaves, vvis)) { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return; } } else { int clus; qbyte *lvis; if (cl.worldmodel->funcs.ClustersInSphere) lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb2, NULL); else { clus = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, dl->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, clus, &lvisb, PVM_FAST); } SHM_BuildShadowMesh(dl, lvis, SMT_SHADOWLESS); if (!Sh_VisOverlaps(lvis, vvis)) //The two viewing areas do not intersect. { RQuantAdd(RQUANT_RTLIGHT_CULL_PVS, 1); return; } } mins[0] = dl->origin[0] - dl->radius; mins[1] = dl->origin[1] - dl->radius; mins[2] = dl->origin[2] - dl->radius; maxs[0] = dl->origin[0] + dl->radius; maxs[1] = dl->origin[1] + dl->radius; maxs[2] = dl->origin[2] + dl->radius; //sets up the gl scissor (actually just culls to view) if (Sh_ScissorForBox(mins, maxs, &rect)) { RQuantAdd(RQUANT_RTLIGHT_CULL_SCISSOR, 1); return; //was culled. } //should we actually scissor here? there's not really much point I suppose. BE_Scissor(NULL); RQuantAdd(RQUANT_RTLIGHT_DRAWN, 1); BE_SelectDLight(dl, colour, axis, dl->fov?LSHADER_SPOT:LSHADER_STANDARD); BE_SelectMode(BEM_LIGHT); Sh_DrawEntLighting(dl, colour); } void Sh_DrawCrepuscularLight(dlight_t *dl, float *colours) { #ifdef GLQUAKE int oldfbo; static mesh_t mesh; image_t *oldsrccol; static vecV_t xyz[4] = { {-1,-1,-1}, {-1,1,-1}, {1,1,-1}, {1,-1,-1} }; static vec2_t tc[4] = { {0,0}, {0,1}, {1,1}, {1,0} }; static index_t idx[6] = { 0,1,2, 0,2,3 }; if (qrenderer != QR_OPENGL) return; mesh.numindexes = 6; mesh.numvertexes = 4; mesh.xyz_array = xyz; mesh.st_array = tc; mesh.indexes = idx; /* a crepuscular light (seriously, that's the correct spelling) is one that gives 'god rays', rather than regular light. our implementation doesn't cast shadows. this allows it to actually be outside the map, and to shine through cloud layers in the sky. we could cast shadows if the light was actually inside, I suppose. Anyway, its done using an FBO, where everything but the sky is black (stuff that occludes the sky is black too). which is then blitted onto the screen in 2d-space. */ /*requires an FBO, as stated above*/ if (!gl_config.ext_framebuffer_objects) return; //fixme: we should add an extra few pixels each side to the fbo, to avoid too much weirdness at screen edges. if (!crepuscular_texture_id) { /*FIXME: requires npot*/ crepuscular_shader = R_RegisterShader("crepuscular_screen", SUF_NONE, "{\n" "program crepuscular_rays\n" "{\n" "map $sourcecolour\n" "blend add\n" "}\n" "}\n" ); crepuscular_texture_id = Image_CreateTexture("***crepusculartexture***", NULL, IF_LINEAR|IF_NOMIPMAP|IF_CLAMP|IF_NOGAMMA); Image_Upload(crepuscular_texture_id, TF_RGBA32, NULL, NULL, vid.pixelwidth, vid.pixelheight, IF_LINEAR|IF_NOMIPMAP|IF_CLAMP|IF_NOGAMMA); } BE_Scissor(NULL); oldfbo = GLBE_FBO_Update(&crepuscular_fbo, FBO_RB_DEPTH, &crepuscular_texture_id, 1, r_nulltex, vid.pixelwidth, vid.pixelheight, 0); GL_ForceDepthWritable(); // qglClearColor(0, 0, 0, 1); qglClear(GL_DEPTH_BUFFER_BIT); BE_SelectMode(BEM_CREPUSCULAR); BE_SelectDLight(dl, colours, dl->axis, LSHADER_STANDARD); GLBE_SubmitMeshes(cl.worldmodel->batches, SHADER_SORT_PORTAL, SHADER_SORT_BLEND); GLBE_FBO_Pop(oldfbo); oldsrccol = NULL;//shaderstate.tex_sourcecol; GLBE_FBO_Sources(crepuscular_texture_id, NULL); // crepuscular_shader->defaulttextures.base = crepuscular_texture_id; //shaderstate.tex_sourcecol = oldsrccol; BE_SelectMode(BEM_STANDARD); BE_DrawMesh_Single(crepuscular_shader, &mesh, NULL, 0); GLBE_FBO_Sources(oldsrccol, NULL); #endif } void Sh_PurgeShadowMeshes(void) { dlight_t *dl; size_t i; for (dl = cl_dlights, i=0; iworldshadowmesh) { SH_FreeShadowMesh(dl->worldshadowmesh); dl->worldshadowmesh = NULL; dl->rebuildcache = true; } } Z_Free(edge); edge = NULL; maxedge = 0; } void R_StaticEntityToRTLight(int i); void Sh_PreGenerateLights(void) { unsigned int ignoreflags; dlight_t *dl; int shadowtype; int leaf; qbyte *lvis; int i; r_shadow_realtime_world_lightmaps.value = atof(r_shadow_realtime_world_lightmaps.string); if (!cl.worldmodel) return; if ((r_shadow_realtime_dlight.ival || r_shadow_realtime_world.ival) && rtlights_max == RTL_FIRST) { qboolean okay = false; if (!okay) okay |= R_LoadRTLights(); if (!okay) { for (i = 0; i < cl.num_statics; i++) R_StaticEntityToRTLight(i); okay |= rtlights_max != RTL_FIRST; } if (!okay) okay |= R_ImportRTLights(Mod_GetEntitiesString(cl.worldmodel)); if (!okay && r_shadow_realtime_world.ival && r_shadow_realtime_world_lightmaps.value != 1) { r_shadow_realtime_world_lightmaps.value = 1; Con_Printf(CON_WARNING "No lights detected in map.\n"); } for (i = 0; i < cl.num_statics; i++) { R_StaticEntityToRTLight(i); } } ignoreflags = (r_shadow_realtime_world.value?LFLAG_REALTIMEMODE:LFLAG_NORMALMODE); for (dl = cl_dlights+rtlights_first, i=rtlights_first; irebuildcache = true; if (dl->radius) { if (dl->flags & ignoreflags) { if (dl->flags & LFLAG_CREPUSCULAR) continue; if (((!dl->die)?!r_shadow_realtime_world_shadows.ival:!r_shadow_realtime_dlight_shadows.ival) || (dl->flags & LFLAG_NOSHADOWS)) shadowtype = SMT_SHADOWLESS; else if (dl->flags & LFLAG_SHADOWMAP || r_shadow_shadowmapping.ival) shadowtype = SMT_SHADOWMAP; else shadowtype = SMT_STENCILVOLUME; //shadowless and lights with an ambient term pass through walls, so need to affect EVERY leaf withing the sphere. if ((shadowtype == SMT_SHADOWLESS || dl->lightcolourscales[0]) && cl.worldmodel->funcs.ClustersInSphere) lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb2, NULL); else { //other lights only want to use the source leaf's pvs (clamped by the sphere) leaf = cl.worldmodel->funcs.ClusterForPoint(cl.worldmodel, dl->origin); lvis = cl.worldmodel->funcs.ClusterPVS(cl.worldmodel, leaf, &lvisb, PVM_FAST); if (cl.worldmodel->funcs.ClustersInSphere) lvis = cl.worldmodel->funcs.ClustersInSphere(cl.worldmodel, dl->origin, dl->radius, &lvisb2, lvis); } SHM_BuildShadowMesh(dl, lvis, shadowtype); continue; } } if (dl->worldshadowmesh) { SH_FreeShadowMesh(dl->worldshadowmesh); dl->worldshadowmesh = NULL; dl->rebuildcache = true; } } } void Com_ParseVector(char *str, vec3_t out) { str = COM_Parse(str); out[0] = atof(com_token); str = COM_Parse(str); out[1] = atof(com_token); str = COM_Parse(str); out[2] = atof(com_token); } void Sh_CheckSettings(void) { qboolean canstencil = false, cansmap = false, canshadowless = false; r_shadow_shadowmapping.ival = r_shadow_shadowmapping.value; r_shadow_realtime_world.ival = r_shadow_realtime_world.value; r_shadow_realtime_dlight.ival = r_shadow_realtime_dlight.value; r_shadow_realtime_world_shadows.ival = r_shadow_realtime_world_shadows.value; r_shadow_realtime_dlight_shadows.ival = r_shadow_realtime_dlight_shadows.value; switch(qrenderer) { #ifdef VKQUAKE case QR_VULKAN: canshadowless = true; cansmap = true; canstencil = false; break; #endif #ifdef GLQUAKE case QR_OPENGL: canshadowless = gl_config.arb_shader_objects || !gl_config_nofixedfunc; //falls back to crappy texture env if (gl_config.arb_shader_objects && gl_config.ext_framebuffer_objects && gl_config.arb_depth_texture)// && gl_config.arb_shadow) cansmap = true; else if ((r_shadow_realtime_world_shadows.ival || r_shadow_realtime_dlight_shadows.ival) && r_shadow_shadowmapping.ival) { if (!gl_config.arb_shader_objects) Con_DPrintf("Shadowmapping unsupported: No arb_shader_objects\n"); else if (!gl_config.ext_framebuffer_objects) Con_DPrintf("Shadowmapping unsupported: No ext_framebuffer_objects\n"); else if (!gl_config.arb_depth_texture) Con_DPrintf("Shadowmapping unsupported: No arb_depth_texture\n"); } if (gl_stencilbits) canstencil = true; break; #endif #ifdef D3D9QUAKE case QR_DIRECT3D9: canshadowless = true; //the code still has a lot of ifdefs, so will crash if you try it in a merged build. //its not really usable in d3d-only builds either, so no great loss. // canstencil = true; break; #endif #ifdef D3D11QUAKE case QR_DIRECT3D11: canshadowless = true; //all feature levels /* shadows are buggy right now. tbh they've always been buggy... rendering seems fine, its just the shadowmaps that are bad if (D3D11_BeginShadowMap(0, SHADOWMAP_SIZE*3, SHADOWMAP_SIZE*2)) { D3D11_EndShadowMap(); cansmap = true; //tends to not work properly until feature level 10 for one error or another. } */ break; #endif default: break; } if (!canstencil && !cansmap && !canshadowless) { //can't even do lighting if (r_shadow_realtime_world.ival || r_shadow_realtime_dlight.ival) Con_Printf("Missing rendering features: realtime lighting is not possible.\n"); r_shadow_realtime_world.ival = 0; r_shadow_realtime_dlight.ival = 0; } else if (!canstencil && !cansmap) { //no shadow methods available at all. if ((r_shadow_realtime_world.ival&&r_shadow_realtime_world_shadows.ival)||(r_shadow_realtime_dlight.ival&&r_shadow_realtime_dlight_shadows.ival)) Con_Printf("Missing rendering features: realtime shadows are not possible.\n"); r_shadow_realtime_world_shadows.ival = 0; r_shadow_realtime_dlight_shadows.ival = 0; } else if (!canstencil || !cansmap) { //only one shadow method if (!!r_shadow_shadowmapping.ival != cansmap) { if (r_shadow_shadowmapping.ival && ((r_shadow_realtime_world.ival&&r_shadow_realtime_world_shadows.ival)||(r_shadow_realtime_dlight.ival&&r_shadow_realtime_dlight_shadows.ival))) Con_Printf("Missing rendering features: forcing shadowmapping %s.\n", cansmap?"on":"off"); r_shadow_shadowmapping.ival = cansmap; } } else { //both shadow methods available. } } void Sh_CalcPointLight(vec3_t point, vec3_t light) { vec3_t colour; dlight_t *dl; vec3_t disp; float dist; float frac; int i; unsigned int ignoreflags; vec3_t norm, impact; ignoreflags = (r_shadow_realtime_world.value?LFLAG_REALTIMEMODE:LFLAG_NORMALMODE); VectorClear(light); if (ignoreflags) for (dl = cl_dlights+rtlights_first, i=rtlights_first; iflags & ignoreflags)) continue; if (dl->key == cl.playerview[0].viewentity) //ignore the light if its emitting from the player. generally the player can't *SEE* that light so it still counts. continue; //disable this check if this function gets used for anything other than iris adaptation colour[0] = dl->color[0]; colour[1] = dl->color[1]; colour[2] = dl->color[2]; if (dl->style) { colour[0] *= cl_lightstyle[dl->style-1].colours[0] * d_lightstylevalue[dl->style-1]/255.0f; colour[1] *= cl_lightstyle[dl->style-1].colours[1] * d_lightstylevalue[dl->style-1]/255.0f; colour[2] *= cl_lightstyle[dl->style-1].colours[2] * d_lightstylevalue[dl->style-1]/255.0f; } else { colour[0] *= r_lightstylescale.value; colour[1] *= r_lightstylescale.value; colour[2] *= r_lightstylescale.value; } if (colour[0] < 0.001 && colour[1] < 0.001 && colour[2] < 0.001) continue; //just switch these off. VectorSubtract(dl->origin, point, disp); dist = VectorLength(disp); frac = dist / dl->radius; if (frac >= 1) continue; //FIXME: this should be affected by the direction. if (CL_TraceLine(point, dl->origin, impact, norm, NULL)>=1) VectorMA(light, 1-frac, colour, light); } } int drawdlightnum; void Sh_DrawLights(qbyte *vis) { vec3_t rotated[3]; vec3_t *axis; vec3_t colour; dlight_t *dl; int i; unsigned int ignoreflags; if (r_shadow_realtime_world.modified || r_shadow_realtime_world_shadows.modified || r_shadow_realtime_dlight.modified || r_shadow_realtime_dlight_shadows.modified || r_shadow_shadowmapping.modified) { r_shadow_realtime_world.modified = r_shadow_realtime_world_shadows.modified = r_shadow_realtime_dlight.modified = r_shadow_realtime_dlight_shadows.modified = r_shadow_shadowmapping.modified = false; Sh_CheckSettings(); //make sure the lighting is reloaded Sh_PreGenerateLights(); } if (r_lightprepass) return; if (!r_shadow_realtime_world.ival && !r_shadow_realtime_dlight.ival) { return; } ignoreflags = (r_shadow_realtime_world.value?LFLAG_REALTIMEMODE:LFLAG_NORMALMODE); // if (r_refdef.recurse) for (dl = cl_dlights+rtlights_first, i=rtlights_first; iradius) continue; //dead if (!(dl->flags & ignoreflags)) continue; colour[0] = dl->color[0]; colour[1] = dl->color[1]; colour[2] = dl->color[2]; if (dl->customstyle) { const char *map = dl->customstyle; int maplen = strlen(map); int idx, v1, v2, vd; float frac, strength; if (!maplen) { strength = ('m'-'a')*22 * r_lightstylescale.value/255.0; } else if (map[0] == '=') { strength = atof(map+1)*r_lightstylescale.value; } else { frac = (cl.time*r_lightstylespeed.value); if (*map == '?' && maplen>1) { map++; maplen--; frac += i*M_PI; } frac += i*M_PI; if (frac < 0) frac = 0; idx = (int)frac; frac -= idx; //this can require updates at 1000 times a second.. Depends on your framerate of course v1 = idx % maplen; v1 = map[v1] - 'a'; v2 = (idx+1) % maplen; v2 = map[v2] - 'a'; vd = v1 - v2; if (/*!r_lightstylesmooth.ival ||*/ vd < -r_lightstylesmooth_limit.ival || vd > r_lightstylesmooth_limit.ival) strength = v1*(22/255.0)*r_lightstylescale.value; else strength = (v1*(1-frac) + v2*(frac))*(22/255.0)*r_lightstylescale.value; } strength *= d_lightstylevalue[0]/256.0f; //a lot of QW mods use lightstyle 0 for a global darkening fade-in thing, so be sure to respect that. colour[0] *= strength; colour[1] *= strength; colour[2] *= strength; } if (dl->style) { colour[0] *= cl_lightstyle[dl->style-1].colours[0] * d_lightstylevalue[dl->style-1]/255.0f; colour[1] *= cl_lightstyle[dl->style-1].colours[1] * d_lightstylevalue[dl->style-1]/255.0f; colour[2] *= cl_lightstyle[dl->style-1].colours[2] * d_lightstylevalue[dl->style-1]/255.0f; } else { colour[0] *= r_lightstylescale.value; colour[1] *= r_lightstylescale.value; colour[2] *= r_lightstylescale.value; } colour[0] *= r_refdef.hdr_value; colour[1] *= r_refdef.hdr_value; colour[2] *= r_refdef.hdr_value; if (colour[0] < 0.001 && colour[1] < 0.001 && colour[2] < 0.001) continue; //just switch these off. if (!dl->lightcolourscales[0] && !dl->lightcolourscales[1] && !dl->lightcolourscales[2]) continue; //these lights are just coronas. if (dl->rotation[0] || dl->rotation[1] || dl->rotation[2]) { //auto-rotating (static) rtlights vec3_t rot; vec3_t rotationaxis[3]; VectorScale(dl->rotation, cl.time, rot); AngleVectorsFLU(rot, rotationaxis[0], rotationaxis[1], rotationaxis[2]); Matrix3_Multiply(dl->axis, rotationaxis, rotated); axis = rotated; } else axis = dl->axis; drawdlightnum++; if (dl->flags & LFLAG_ORTHO) { vec3_t saveorg = {dl->origin[0], dl->origin[1], dl->origin[2]}, neworg; vec3_t saveaxis[3]; memcpy(saveaxis, dl->axis, sizeof(saveaxis)); memcpy(dl->axis, axis, sizeof(saveaxis)); VectorMA(r_origin, dl->radius/3, vpn, neworg); VectorCopy(neworg, dl->origin); dl->rebuildcache = true; Sh_DrawShadowMapLight(dl, colour, axis, NULL); VectorCopy(saveorg, dl->origin); memcpy(dl->axis, saveaxis, sizeof(saveaxis)); } else if (dl->flags & LFLAG_CREPUSCULAR) Sh_DrawCrepuscularLight(dl, colour); else if (((i >= RTL_FIRST)?!r_shadow_realtime_world_shadows.ival:!r_shadow_realtime_dlight_shadows.ival) || dl->flags & LFLAG_NOSHADOWS) { Sh_DrawShadowlessLight(dl, colour, axis, vis); } else if ((dl->flags & LFLAG_SHADOWMAP) || r_shadow_shadowmapping.ival) { Sh_DrawShadowMapLight(dl, colour, axis, vis); } else { Sh_DrawStencilLight(dl, colour, axis, vis); } } #ifdef GLQUAKE if (gl_config.arb_shader_objects) { dlight_t sun = {0}; vec3_t sundir; float dot; Com_ParseVector(r_sun_dir.string, sundir); Com_ParseVector(r_sun_colour.string, colour); //fade it out if we're looking at an angle parallel to it (to avoid nasty visible graduations or backwards rays!) dot = DotProduct(vpn, sundir); dot = 1-dot; dot *= dot; dot = 1-dot; VectorScale(colour, dot, colour); if (colour[0] > 0.001 || colour[1] > 0.001 || colour[2] > 0.001) { //only do this if we can see some sky surfaces. pointless otherwise batch_t *b; for (b = cl.worldmodel->batches[SHADER_SORT_SKY]; b; b = b->next) { if (b->meshes) break; } if (b) { VectorNormalize(sundir); VectorMA(r_origin, 1000, sundir, sun.origin); Sh_DrawCrepuscularLight(&sun, colour); } } } #endif BE_Scissor(NULL); BE_SelectMode(BEM_STANDARD); // if (developer.value) // Con_Printf("%i lights drawn, %i frustum culled, %i pvs culled, %i scissor culled\n", bench.numlights, bench.numfrustumculled, bench.numpvsculled, bench.numscissorculled); // memset(&bench, 0, sizeof(bench)); drawdlightnum = -1; } #endif //stencil shadows generally require that the farclip distance is really really far away //so this little function is used to check if its needed or not. qboolean Sh_StencilShadowsActive(void) { #if defined(RTLIGHTS) && !defined(SERVERONLY) //if shadowmapping is forced on all lights then we don't need special depth stuff // if (r_shadow_shadowmapping.ival) // return false; if (isDedicated) return false; return (r_shadow_realtime_dlight.ival && r_shadow_realtime_dlight_shadows.ival) || (r_shadow_realtime_world.ival && r_shadow_realtime_world_shadows.ival); #else return false; #endif } void Sh_RegisterCvars(void) { #if defined(RTLIGHTS) && !defined(SERVERONLY) #define REALTIMELIGHTING "Realtime Lighting" Cvar_Register (&r_shadow_scissor, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_world, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_world_shadows, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_dlight, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_dlight_ambient, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_dlight_diffuse, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_dlight_specular, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_dlight_shadows, REALTIMELIGHTING); Cvar_Register (&r_shadow_realtime_world_lightmaps, REALTIMELIGHTING); Cvar_Register (&r_shadow_playershadows, REALTIMELIGHTING); Cvar_Register (&r_shadow_shadowmapping, REALTIMELIGHTING); Cvar_Register (&r_shadow_shadowmapping_precision, REALTIMELIGHTING); Cvar_Register (&r_shadow_shadowmapping_nearclip, REALTIMELIGHTING); Cvar_Register (&r_shadow_shadowmapping_bias, REALTIMELIGHTING); Cvar_Register (&r_sun_dir, REALTIMELIGHTING); Cvar_Register (&r_sun_colour, REALTIMELIGHTING); #endif }