/* Copyright (C) 1996-1997 Id Software, Inc. This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. */ // models.c -- model loading and caching // models are the only shared resource between a client and server running // on the same machine. #include "quakedef.h" #include "image.h" #include "r_shadow.h" cvar_t r_mipskins = {CVAR_SAVE, "r_mipskins", "0", "mipmaps model skins so they render faster in the distance and do not display noise artifacts, can cause discoloration of skins if they contain undesirable border colors"}; dp_model_t *loadmodel; static mempool_t *mod_mempool; static memexpandablearray_t models; static mempool_t* q3shaders_mem; typedef struct q3shader_hash_entry_s { q3shaderinfo_t shader; struct q3shader_hash_entry_s* chain; } q3shader_hash_entry_t; #define Q3SHADER_HASH_SIZE 1021 typedef struct q3shader_data_s { memexpandablearray_t hash_entries; q3shader_hash_entry_t hash[Q3SHADER_HASH_SIZE]; memexpandablearray_t char_ptrs; } q3shader_data_t; static q3shader_data_t* q3shader_data; static void mod_start(void) { int i, count; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; SCR_PushLoadingScreen(false, "Loading models", 1.0); count = 0; for (i = 0;i < nummodels;i++) if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && mod->name[0] != '*') if (mod->used) ++count; for (i = 0;i < nummodels;i++) if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && mod->name[0] != '*') if (mod->used) { SCR_PushLoadingScreen(true, mod->name, 1.0 / count); Mod_LoadModel(mod, true, false); SCR_PopLoadingScreen(false); } SCR_PopLoadingScreen(false); } static void mod_shutdown(void) { int i; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; for (i = 0;i < nummodels;i++) if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && (mod->loaded || mod->mempool)) Mod_UnloadModel(mod); Mod_FreeQ3Shaders(); } static void mod_newmap(void) { msurface_t *surface; int i, j, k, surfacenum, ssize, tsize; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; R_SkinFrame_PrepareForPurge(); for (i = 0;i < nummodels;i++) { if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->mempool && mod->data_textures) { for (j = 0;j < mod->num_textures;j++) { for (k = 0;k < mod->data_textures[j].numskinframes;k++) R_SkinFrame_MarkUsed(mod->data_textures[j].skinframes[k]); for (k = 0;k < mod->data_textures[j].backgroundnumskinframes;k++) R_SkinFrame_MarkUsed(mod->data_textures[j].backgroundskinframes[k]); } } } R_SkinFrame_Purge(); if (!cl_stainmaps_clearonload.integer) return; for (i = 0;i < nummodels;i++) { if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->mempool && mod->data_surfaces) { for (surfacenum = 0, surface = mod->data_surfaces;surfacenum < mod->num_surfaces;surfacenum++, surface++) { if (surface->lightmapinfo && surface->lightmapinfo->stainsamples) { ssize = (surface->lightmapinfo->extents[0] >> 4) + 1; tsize = (surface->lightmapinfo->extents[1] >> 4) + 1; memset(surface->lightmapinfo->stainsamples, 255, ssize * tsize * 3); mod->brushq1.lightmapupdateflags[surfacenum] = true; } } } } } /* =============== Mod_Init =============== */ static void Mod_Print(void); static void Mod_Precache (void); static void Mod_Decompile_f(void); static void Mod_BuildVBOs(void); void Mod_Init (void) { mod_mempool = Mem_AllocPool("modelinfo", 0, NULL); Mem_ExpandableArray_NewArray(&models, mod_mempool, sizeof(dp_model_t), 16); Mod_BrushInit(); Mod_AliasInit(); Mod_SpriteInit(); Cvar_RegisterVariable(&r_mipskins); Cmd_AddCommand ("modellist", Mod_Print, "prints a list of loaded models"); Cmd_AddCommand ("modelprecache", Mod_Precache, "load a model"); Cmd_AddCommand ("modeldecompile", Mod_Decompile_f, "exports a model in several formats for editing purposes"); } void Mod_RenderInit(void) { R_RegisterModule("Models", mod_start, mod_shutdown, mod_newmap); } void Mod_UnloadModel (dp_model_t *mod) { char name[MAX_QPATH]; qboolean used; dp_model_t *parentmodel; if (developer_loading.integer) Con_Printf("unloading model %s\n", mod->name); strlcpy(name, mod->name, sizeof(name)); parentmodel = mod->brush.parentmodel; used = mod->used; if (mod->surfmesh.ebo3i) R_Mesh_DestroyBufferObject(mod->surfmesh.ebo3i); if (mod->surfmesh.ebo3s) R_Mesh_DestroyBufferObject(mod->surfmesh.ebo3s); if (mod->surfmesh.vbo) R_Mesh_DestroyBufferObject(mod->surfmesh.vbo); // free textures/memory attached to the model R_FreeTexturePool(&mod->texturepool); Mem_FreePool(&mod->mempool); // clear the struct to make it available memset(mod, 0, sizeof(dp_model_t)); // restore the fields we want to preserve strlcpy(mod->name, name, sizeof(mod->name)); mod->brush.parentmodel = parentmodel; mod->used = used; mod->loaded = false; } void R_Model_Null_Draw(entity_render_t *ent) { return; } typedef void (*mod_framegroupify_parsegroups_t) (unsigned int i, int start, int len, float fps, qboolean loop, void *pass); int Mod_FrameGroupify_ParseGroups(const char *buf, mod_framegroupify_parsegroups_t cb, void *pass) { const char *bufptr; int start, len; float fps; unsigned int i; qboolean loop; bufptr = buf; i = 0; for(;;) { // an anim scene! if (!COM_ParseToken_Simple(&bufptr, true, false)) break; if (!strcmp(com_token, "\n")) continue; // empty line start = atoi(com_token); if (!COM_ParseToken_Simple(&bufptr, true, false)) break; if (!strcmp(com_token, "\n")) { Con_Printf("framegroups file: missing number of frames\n"); continue; } len = atoi(com_token); if (!COM_ParseToken_Simple(&bufptr, true, false)) break; // we default to looping as it's usually wanted, so to NOT loop you append a 0 if (strcmp(com_token, "\n")) { fps = atof(com_token); if (!COM_ParseToken_Simple(&bufptr, true, false)) break; if (strcmp(com_token, "\n")) loop = atoi(com_token); else loop = true; } else { fps = 20; loop = true; } if(cb) cb(i, start, len, fps, loop, pass); ++i; } return i; } void Mod_FrameGroupify_ParseGroups_Count (unsigned int i, int start, int len, float fps, qboolean loop, void *pass) { unsigned int *cnt = (unsigned int *) pass; ++*cnt; } void Mod_FrameGroupify_ParseGroups_Store (unsigned int i, int start, int len, float fps, qboolean loop, void *pass) { animscene_t *anim = (animscene_t *) pass; dpsnprintf(anim[i].name, sizeof(anim[i].name), "groupified_%d", i); anim[i].firstframe = start; anim[i].framecount = len; anim[i].framerate = fps; anim[i].loop = loop; //Con_Printf("frame group %d is %d %d %f %d\n", i, start, len, fps, loop); } void Mod_FrameGroupify(dp_model_t *mod, const char *buf) { unsigned int cnt; // 0. count cnt = Mod_FrameGroupify_ParseGroups(buf, NULL, NULL); // 1. reallocate if(mod->animscenes) Mem_Free(mod->animscenes); mod->animscenes = (animscene_t *) Mem_Alloc(tempmempool, sizeof(animscene_t) * cnt); // 2. parse Mod_FrameGroupify_ParseGroups(buf, Mod_FrameGroupify_ParseGroups_Store, mod->animscenes); } /* ================== Mod_LoadModel Loads a model ================== */ dp_model_t *Mod_LoadModel(dp_model_t *mod, qboolean crash, qboolean checkdisk) { int num; unsigned int crc; void *buf; fs_offset_t filesize; mod->used = true; if (mod->name[0] == '*') // submodel return mod; if (!strcmp(mod->name, "null")) { if(mod->loaded) return mod; if (mod->loaded || mod->mempool) Mod_UnloadModel(mod); if (developer_loading.integer) Con_Printf("loading model %s\n", mod->name); mod->used = true; mod->crc = -1; mod->loaded = false; VectorClear(mod->normalmins); VectorClear(mod->normalmaxs); VectorClear(mod->yawmins); VectorClear(mod->yawmaxs); VectorClear(mod->rotatedmins); VectorClear(mod->rotatedmaxs); mod->modeldatatypestring = "null"; mod->type = mod_null; mod->Draw = R_Model_Null_Draw; mod->numframes = 2; mod->numskins = 1; // no fatal errors occurred, so this model is ready to use. mod->loaded = true; return mod; } crc = 0; buf = NULL; // even if the model is loaded it still may need reloading... // if it is not loaded or checkdisk is true we need to calculate the crc if (!mod->loaded || checkdisk) { if (checkdisk && mod->loaded) Con_DPrintf("checking model %s\n", mod->name); buf = FS_LoadFile (mod->name, tempmempool, false, &filesize); if (buf) { crc = CRC_Block((unsigned char *)buf, filesize); // we need to reload the model if the crc does not match if (mod->crc != crc) mod->loaded = false; } } // if the model is already loaded and checks passed, just return if (mod->loaded) { if (buf) Mem_Free(buf); return mod; } if (developer_loading.integer) Con_Printf("loading model %s\n", mod->name); SCR_PushLoadingScreen(true, mod->name, 1); // LordHavoc: unload the existing model in this slot (if there is one) if (mod->loaded || mod->mempool) Mod_UnloadModel(mod); // load the model mod->used = true; mod->crc = crc; // errors can prevent the corresponding mod->loaded = true; mod->loaded = false; // default model radius and bounding box (mainly for missing models) mod->radius = 16; VectorSet(mod->normalmins, -mod->radius, -mod->radius, -mod->radius); VectorSet(mod->normalmaxs, mod->radius, mod->radius, mod->radius); VectorSet(mod->yawmins, -mod->radius, -mod->radius, -mod->radius); VectorSet(mod->yawmaxs, mod->radius, mod->radius, mod->radius); VectorSet(mod->rotatedmins, -mod->radius, -mod->radius, -mod->radius); VectorSet(mod->rotatedmaxs, mod->radius, mod->radius, mod->radius); if (!q3shaders_mem) { // load q3 shaders for the first time, or after a level change Mod_LoadQ3Shaders(); } if (buf) { char *bufend = (char *)buf + filesize; // all models use memory, so allocate a memory pool mod->mempool = Mem_AllocPool(mod->name, 0, NULL); num = LittleLong(*((int *)buf)); // call the apropriate loader loadmodel = mod; if (!strcasecmp(FS_FileExtension(mod->name), "obj")) Mod_OBJ_Load(mod, buf, bufend); else if (!memcmp(buf, "IDPO", 4)) Mod_IDP0_Load(mod, buf, bufend); else if (!memcmp(buf, "IDP2", 4)) Mod_IDP2_Load(mod, buf, bufend); else if (!memcmp(buf, "IDP3", 4)) Mod_IDP3_Load(mod, buf, bufend); else if (!memcmp(buf, "IDSP", 4)) Mod_IDSP_Load(mod, buf, bufend); else if (!memcmp(buf, "IDS2", 4)) Mod_IDS2_Load(mod, buf, bufend); else if (!memcmp(buf, "IBSP", 4)) Mod_IBSP_Load(mod, buf, bufend); else if (!memcmp(buf, "ZYMOTICMODEL", 12)) Mod_ZYMOTICMODEL_Load(mod, buf, bufend); else if (!memcmp(buf, "DARKPLACESMODEL", 16)) Mod_DARKPLACESMODEL_Load(mod, buf, bufend); else if (!memcmp(buf, "ACTRHEAD", 8)) Mod_PSKMODEL_Load(mod, buf, bufend); else if (strlen(mod->name) >= 4 && !strcmp(mod->name + strlen(mod->name) - 4, ".map")) Mod_MAP_Load(mod, buf, bufend); else if (num == BSPVERSION || num == 30) Mod_Q1BSP_Load(mod, buf, bufend); else Con_Printf("Mod_LoadModel: model \"%s\" is of unknown/unsupported type\n", mod->name); Mem_Free(buf); buf = FS_LoadFile (va("%s.framegroups", mod->name), tempmempool, false, &filesize); if(buf) Mod_FrameGroupify(mod, buf); Mem_Free(buf); Mod_BuildVBOs(); } else if (crash) { // LordHavoc: Sys_Error was *ANNOYING* Con_Printf ("Mod_LoadModel: %s not found\n", mod->name); } // no fatal errors occurred, so this model is ready to use. mod->loaded = true; SCR_PopLoadingScreen(false); return mod; } void Mod_ClearUsed(void) { int i; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; for (i = 0;i < nummodels;i++) if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0]) mod->used = false; } void Mod_PurgeUnused(void) { int i; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; for (i = 0;i < nummodels;i++) { if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && !mod->used) { Mod_UnloadModel(mod); Mem_ExpandableArray_FreeRecord(&models, mod); } } } /* ================== Mod_FindName ================== */ dp_model_t *Mod_FindName(const char *name, const char *parentname) { int i; int nummodels; dp_model_t *mod; if (!parentname) parentname = ""; // if we're not dedicatd, the renderer calls will crash without video Host_StartVideo(); nummodels = Mem_ExpandableArray_IndexRange(&models); if (!name[0]) Host_Error ("Mod_ForName: NULL name"); // search the currently loaded models for (i = 0;i < nummodels;i++) { if ((mod = (dp_model_t*) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && !strcmp(mod->name, name) && ((!mod->brush.parentmodel && !parentname[0]) || (mod->brush.parentmodel && parentname[0] && !strcmp(mod->brush.parentmodel->name, parentname)))) { mod->used = true; return mod; } } // no match found, create a new one mod = (dp_model_t *) Mem_ExpandableArray_AllocRecord(&models); strlcpy(mod->name, name, sizeof(mod->name)); if (parentname[0]) mod->brush.parentmodel = Mod_FindName(parentname, NULL); else mod->brush.parentmodel = NULL; mod->loaded = false; mod->used = true; return mod; } /* ================== Mod_ForName Loads in a model for the given name ================== */ dp_model_t *Mod_ForName(const char *name, qboolean crash, qboolean checkdisk, const char *parentname) { dp_model_t *model; model = Mod_FindName(name, parentname); if (!model->loaded || checkdisk) Mod_LoadModel(model, crash, checkdisk); return model; } /* ================== Mod_Reload Reloads all models if they have changed ================== */ void Mod_Reload(void) { int i, count; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; SCR_PushLoadingScreen(false, "Reloading models", 1.0); count = 0; for (i = 0;i < nummodels;i++) if ((mod = (dp_model_t *) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && mod->name[0] != '*' && mod->used) ++count; for (i = 0;i < nummodels;i++) if ((mod = (dp_model_t *) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && mod->name[0] != '*' && mod->used) { SCR_PushLoadingScreen(true, mod->name, 1.0 / count); Mod_LoadModel(mod, true, true); SCR_PopLoadingScreen(false); } SCR_PopLoadingScreen(false); } unsigned char *mod_base; //============================================================================= /* ================ Mod_Print ================ */ static void Mod_Print(void) { int i; int nummodels = Mem_ExpandableArray_IndexRange(&models); dp_model_t *mod; Con_Print("Loaded models:\n"); for (i = 0;i < nummodels;i++) { if ((mod = (dp_model_t *) Mem_ExpandableArray_RecordAtIndex(&models, i)) && mod->name[0] && mod->name[0] != '*') { if (mod->brush.numsubmodels) Con_Printf("%4iK %s (%i submodels)\n", mod->mempool ? (int)((mod->mempool->totalsize + 1023) / 1024) : 0, mod->name, mod->brush.numsubmodels); else Con_Printf("%4iK %s\n", mod->mempool ? (int)((mod->mempool->totalsize + 1023) / 1024) : 0, mod->name); } } } /* ================ Mod_Precache ================ */ static void Mod_Precache(void) { if (Cmd_Argc() == 2) Mod_ForName(Cmd_Argv(1), false, true, Cmd_Argv(1)[0] == '*' ? cl.model_name[1] : NULL); else Con_Print("usage: modelprecache \n"); } int Mod_BuildVertexRemapTableFromElements(int numelements, const int *elements, int numvertices, int *remapvertices) { int i, count; unsigned char *used; used = (unsigned char *)Mem_Alloc(tempmempool, numvertices); memset(used, 0, numvertices); for (i = 0;i < numelements;i++) used[elements[i]] = 1; for (i = 0, count = 0;i < numvertices;i++) remapvertices[i] = used[i] ? count++ : -1; Mem_Free(used); return count; } #if 1 // fast way, using an edge hash #define TRIANGLEEDGEHASH 8192 void Mod_BuildTriangleNeighbors(int *neighbors, const int *elements, int numtriangles) { int i, j, p, e1, e2, *n, hashindex, count, match; const int *e; typedef struct edgehashentry_s { struct edgehashentry_s *next; int triangle; int element[2]; } edgehashentry_t; edgehashentry_t *edgehash[TRIANGLEEDGEHASH], *edgehashentries, edgehashentriesbuffer[TRIANGLEEDGEHASH*3], *hash; memset(edgehash, 0, sizeof(edgehash)); edgehashentries = edgehashentriesbuffer; // if there are too many triangles for the stack array, allocate larger buffer if (numtriangles > TRIANGLEEDGEHASH) edgehashentries = (edgehashentry_t *)Mem_Alloc(tempmempool, numtriangles * 3 * sizeof(edgehashentry_t)); // find neighboring triangles for (i = 0, e = elements, n = neighbors;i < numtriangles;i++, e += 3, n += 3) { for (j = 0, p = 2;j < 3;p = j, j++) { e1 = e[p]; e2 = e[j]; // this hash index works for both forward and backward edges hashindex = (unsigned int)(e1 + e2) % TRIANGLEEDGEHASH; hash = edgehashentries + i * 3 + j; hash->next = edgehash[hashindex]; edgehash[hashindex] = hash; hash->triangle = i; hash->element[0] = e1; hash->element[1] = e2; } } for (i = 0, e = elements, n = neighbors;i < numtriangles;i++, e += 3, n += 3) { for (j = 0, p = 2;j < 3;p = j, j++) { e1 = e[p]; e2 = e[j]; // this hash index works for both forward and backward edges hashindex = (unsigned int)(e1 + e2) % TRIANGLEEDGEHASH; count = 0; match = -1; for (hash = edgehash[hashindex];hash;hash = hash->next) { if (hash->element[0] == e2 && hash->element[1] == e1) { if (hash->triangle != i) match = hash->triangle; count++; } else if ((hash->element[0] == e1 && hash->element[1] == e2)) count++; } // detect edges shared by three triangles and make them seams if (count > 2) match = -1; n[p] = match; } // also send a keepalive here (this can take a while too!) CL_KeepaliveMessage(false); } // free the allocated buffer if (edgehashentries != edgehashentriesbuffer) Mem_Free(edgehashentries); } #else // very slow but simple way static int Mod_FindTriangleWithEdge(const int *elements, int numtriangles, int start, int end, int ignore) { int i, match, count; count = 0; match = -1; for (i = 0;i < numtriangles;i++, elements += 3) { if ((elements[0] == start && elements[1] == end) || (elements[1] == start && elements[2] == end) || (elements[2] == start && elements[0] == end)) { if (i != ignore) match = i; count++; } else if ((elements[1] == start && elements[0] == end) || (elements[2] == start && elements[1] == end) || (elements[0] == start && elements[2] == end)) count++; } // detect edges shared by three triangles and make them seams if (count > 2) match = -1; return match; } void Mod_BuildTriangleNeighbors(int *neighbors, const int *elements, int numtriangles) { int i, *n; const int *e; for (i = 0, e = elements, n = neighbors;i < numtriangles;i++, e += 3, n += 3) { n[0] = Mod_FindTriangleWithEdge(elements, numtriangles, e[1], e[0], i); n[1] = Mod_FindTriangleWithEdge(elements, numtriangles, e[2], e[1], i); n[2] = Mod_FindTriangleWithEdge(elements, numtriangles, e[0], e[2], i); } } #endif void Mod_ValidateElements(int *elements, int numtriangles, int firstvertex, int numverts, const char *filename, int fileline) { int i, warned = false, endvertex = firstvertex + numverts; for (i = 0;i < numtriangles * 3;i++) { if (elements[i] < firstvertex || elements[i] >= endvertex) { if (!warned) { warned = true; Con_Printf("Mod_ValidateElements: out of bounds elements detected at %s:%d\n", filename, fileline); } elements[i] = firstvertex; } } } // warning: this is an expensive function! void Mod_BuildNormals(int firstvertex, int numvertices, int numtriangles, const float *vertex3f, const int *elements, float *normal3f, qboolean areaweighting) { int i, j; const int *element; float *vectorNormal; float areaNormal[3]; // clear the vectors memset(normal3f + 3 * firstvertex, 0, numvertices * sizeof(float[3])); // process each vertex of each triangle and accumulate the results // use area-averaging, to make triangles with a big area have a bigger // weighting on the vertex normal than triangles with a small area // to do so, just add the 'normals' together (the bigger the area // the greater the length of the normal is element = elements; for (i = 0; i < numtriangles; i++, element += 3) { TriangleNormal( vertex3f + element[0] * 3, vertex3f + element[1] * 3, vertex3f + element[2] * 3, areaNormal ); if (!areaweighting) VectorNormalize(areaNormal); for (j = 0;j < 3;j++) { vectorNormal = normal3f + element[j] * 3; vectorNormal[0] += areaNormal[0]; vectorNormal[1] += areaNormal[1]; vectorNormal[2] += areaNormal[2]; } } // and just normalize the accumulated vertex normal in the end vectorNormal = normal3f + 3 * firstvertex; for (i = 0; i < numvertices; i++, vectorNormal += 3) VectorNormalize(vectorNormal); } void Mod_BuildBumpVectors(const float *v0, const float *v1, const float *v2, const float *tc0, const float *tc1, const float *tc2, float *svector3f, float *tvector3f, float *normal3f) { float f, tangentcross[3], v10[3], v20[3], tc10[2], tc20[2]; // 79 add/sub/negate/multiply (1 cycle), 1 compare (3 cycle?), total cycles not counting load/store/exchange roughly 82 cycles // 6 add, 28 subtract, 39 multiply, 1 compare, 50% chance of 6 negates // 6 multiply, 9 subtract VectorSubtract(v1, v0, v10); VectorSubtract(v2, v0, v20); normal3f[0] = v20[1] * v10[2] - v20[2] * v10[1]; normal3f[1] = v20[2] * v10[0] - v20[0] * v10[2]; normal3f[2] = v20[0] * v10[1] - v20[1] * v10[0]; // 12 multiply, 10 subtract tc10[1] = tc1[1] - tc0[1]; tc20[1] = tc2[1] - tc0[1]; svector3f[0] = tc10[1] * v20[0] - tc20[1] * v10[0]; svector3f[1] = tc10[1] * v20[1] - tc20[1] * v10[1]; svector3f[2] = tc10[1] * v20[2] - tc20[1] * v10[2]; tc10[0] = tc1[0] - tc0[0]; tc20[0] = tc2[0] - tc0[0]; tvector3f[0] = tc10[0] * v20[0] - tc20[0] * v10[0]; tvector3f[1] = tc10[0] * v20[1] - tc20[0] * v10[1]; tvector3f[2] = tc10[0] * v20[2] - tc20[0] * v10[2]; // 12 multiply, 4 add, 6 subtract f = DotProduct(svector3f, normal3f); svector3f[0] -= f * normal3f[0]; svector3f[1] -= f * normal3f[1]; svector3f[2] -= f * normal3f[2]; f = DotProduct(tvector3f, normal3f); tvector3f[0] -= f * normal3f[0]; tvector3f[1] -= f * normal3f[1]; tvector3f[2] -= f * normal3f[2]; // if texture is mapped the wrong way (counterclockwise), the tangents // have to be flipped, this is detected by calculating a normal from the // two tangents, and seeing if it is opposite the surface normal // 9 multiply, 2 add, 3 subtract, 1 compare, 50% chance of: 6 negates CrossProduct(tvector3f, svector3f, tangentcross); if (DotProduct(tangentcross, normal3f) < 0) { VectorNegate(svector3f, svector3f); VectorNegate(tvector3f, tvector3f); } } // warning: this is a very expensive function! void Mod_BuildTextureVectorsFromNormals(int firstvertex, int numvertices, int numtriangles, const float *vertex3f, const float *texcoord2f, const float *normal3f, const int *elements, float *svector3f, float *tvector3f, qboolean areaweighting) { int i, tnum; float sdir[3], tdir[3], normal[3], *sv, *tv; const float *v0, *v1, *v2, *tc0, *tc1, *tc2, *n; float f, tangentcross[3], v10[3], v20[3], tc10[2], tc20[2]; const int *e; // clear the vectors memset(svector3f + 3 * firstvertex, 0, numvertices * sizeof(float[3])); memset(tvector3f + 3 * firstvertex, 0, numvertices * sizeof(float[3])); // process each vertex of each triangle and accumulate the results for (tnum = 0, e = elements;tnum < numtriangles;tnum++, e += 3) { v0 = vertex3f + e[0] * 3; v1 = vertex3f + e[1] * 3; v2 = vertex3f + e[2] * 3; tc0 = texcoord2f + e[0] * 2; tc1 = texcoord2f + e[1] * 2; tc2 = texcoord2f + e[2] * 2; // 79 add/sub/negate/multiply (1 cycle), 1 compare (3 cycle?), total cycles not counting load/store/exchange roughly 82 cycles // 6 add, 28 subtract, 39 multiply, 1 compare, 50% chance of 6 negates // calculate the edge directions and surface normal // 6 multiply, 9 subtract VectorSubtract(v1, v0, v10); VectorSubtract(v2, v0, v20); normal[0] = v20[1] * v10[2] - v20[2] * v10[1]; normal[1] = v20[2] * v10[0] - v20[0] * v10[2]; normal[2] = v20[0] * v10[1] - v20[1] * v10[0]; // calculate the tangents // 12 multiply, 10 subtract tc10[1] = tc1[1] - tc0[1]; tc20[1] = tc2[1] - tc0[1]; sdir[0] = tc10[1] * v20[0] - tc20[1] * v10[0]; sdir[1] = tc10[1] * v20[1] - tc20[1] * v10[1]; sdir[2] = tc10[1] * v20[2] - tc20[1] * v10[2]; tc10[0] = tc1[0] - tc0[0]; tc20[0] = tc2[0] - tc0[0]; tdir[0] = tc10[0] * v20[0] - tc20[0] * v10[0]; tdir[1] = tc10[0] * v20[1] - tc20[0] * v10[1]; tdir[2] = tc10[0] * v20[2] - tc20[0] * v10[2]; // if texture is mapped the wrong way (counterclockwise), the tangents // have to be flipped, this is detected by calculating a normal from the // two tangents, and seeing if it is opposite the surface normal // 9 multiply, 2 add, 3 subtract, 1 compare, 50% chance of: 6 negates CrossProduct(tdir, sdir, tangentcross); if (DotProduct(tangentcross, normal) < 0) { VectorNegate(sdir, sdir); VectorNegate(tdir, tdir); } if (!areaweighting) { VectorNormalize(sdir); VectorNormalize(tdir); } for (i = 0;i < 3;i++) { VectorAdd(svector3f + e[i]*3, sdir, svector3f + e[i]*3); VectorAdd(tvector3f + e[i]*3, tdir, tvector3f + e[i]*3); } } // make the tangents completely perpendicular to the surface normal, and // then normalize them // 16 assignments, 2 divide, 2 sqrt, 2 negates, 14 adds, 24 multiplies for (i = 0, sv = svector3f + 3 * firstvertex, tv = tvector3f + 3 * firstvertex, n = normal3f + 3 * firstvertex;i < numvertices;i++, sv += 3, tv += 3, n += 3) { f = -DotProduct(sv, n); VectorMA(sv, f, n, sv); VectorNormalize(sv); f = -DotProduct(tv, n); VectorMA(tv, f, n, tv); VectorNormalize(tv); } } void Mod_AllocSurfMesh(mempool_t *mempool, int numvertices, int numtriangles, qboolean lightmapoffsets, qboolean vertexcolors, qboolean neighbors) { unsigned char *data; data = (unsigned char *)Mem_Alloc(mempool, numvertices * (3 + 3 + 3 + 3 + 2 + 2 + (vertexcolors ? 4 : 0)) * sizeof(float) + numvertices * (lightmapoffsets ? 1 : 0) * sizeof(int) + numtriangles * (3 + (neighbors ? 3 : 0)) * sizeof(int) + (numvertices <= 65536 ? numtriangles * sizeof(unsigned short[3]) : 0)); loadmodel->surfmesh.num_vertices = numvertices; loadmodel->surfmesh.num_triangles = numtriangles; if (loadmodel->surfmesh.num_vertices) { loadmodel->surfmesh.data_vertex3f = (float *)data, data += sizeof(float[3]) * loadmodel->surfmesh.num_vertices; loadmodel->surfmesh.data_svector3f = (float *)data, data += sizeof(float[3]) * loadmodel->surfmesh.num_vertices; loadmodel->surfmesh.data_tvector3f = (float *)data, data += sizeof(float[3]) * loadmodel->surfmesh.num_vertices; loadmodel->surfmesh.data_normal3f = (float *)data, data += sizeof(float[3]) * loadmodel->surfmesh.num_vertices; loadmodel->surfmesh.data_texcoordtexture2f = (float *)data, data += sizeof(float[2]) * loadmodel->surfmesh.num_vertices; loadmodel->surfmesh.data_texcoordlightmap2f = (float *)data, data += sizeof(float[2]) * loadmodel->surfmesh.num_vertices; if (vertexcolors) loadmodel->surfmesh.data_lightmapcolor4f = (float *)data, data += sizeof(float[4]) * loadmodel->surfmesh.num_vertices; if (lightmapoffsets) loadmodel->surfmesh.data_lightmapoffsets = (int *)data, data += sizeof(int) * loadmodel->surfmesh.num_vertices; } if (loadmodel->surfmesh.num_triangles) { loadmodel->surfmesh.data_element3i = (int *)data, data += sizeof(int[3]) * loadmodel->surfmesh.num_triangles; if (neighbors) loadmodel->surfmesh.data_neighbor3i = (int *)data, data += sizeof(int[3]) * loadmodel->surfmesh.num_triangles; if (loadmodel->surfmesh.num_vertices <= 65536) loadmodel->surfmesh.data_element3s = (unsigned short *)data, data += sizeof(unsigned short[3]) * loadmodel->surfmesh.num_triangles; } } shadowmesh_t *Mod_ShadowMesh_Alloc(mempool_t *mempool, int maxverts, int maxtriangles, rtexture_t *map_diffuse, rtexture_t *map_specular, rtexture_t *map_normal, int light, int neighbors, int expandable) { shadowmesh_t *newmesh; unsigned char *data; int size; size = sizeof(shadowmesh_t); size += maxverts * sizeof(float[3]); if (light) size += maxverts * sizeof(float[11]); size += maxtriangles * sizeof(int[3]); if (maxverts <= 65536) size += maxtriangles * sizeof(unsigned short[3]); if (neighbors) size += maxtriangles * sizeof(int[3]); if (expandable) size += SHADOWMESHVERTEXHASH * sizeof(shadowmeshvertexhash_t *) + maxverts * sizeof(shadowmeshvertexhash_t); data = (unsigned char *)Mem_Alloc(mempool, size); newmesh = (shadowmesh_t *)data;data += sizeof(*newmesh); newmesh->map_diffuse = map_diffuse; newmesh->map_specular = map_specular; newmesh->map_normal = map_normal; newmesh->maxverts = maxverts; newmesh->maxtriangles = maxtriangles; newmesh->numverts = 0; newmesh->numtriangles = 0; newmesh->vertex3f = (float *)data;data += maxverts * sizeof(float[3]); if (light) { newmesh->svector3f = (float *)data;data += maxverts * sizeof(float[3]); newmesh->tvector3f = (float *)data;data += maxverts * sizeof(float[3]); newmesh->normal3f = (float *)data;data += maxverts * sizeof(float[3]); newmesh->texcoord2f = (float *)data;data += maxverts * sizeof(float[2]); } newmesh->element3i = (int *)data;data += maxtriangles * sizeof(int[3]); if (neighbors) { newmesh->neighbor3i = (int *)data;data += maxtriangles * sizeof(int[3]); } if (expandable) { newmesh->vertexhashtable = (shadowmeshvertexhash_t **)data;data += SHADOWMESHVERTEXHASH * sizeof(shadowmeshvertexhash_t *); newmesh->vertexhashentries = (shadowmeshvertexhash_t *)data;data += maxverts * sizeof(shadowmeshvertexhash_t); } if (maxverts <= 65536) newmesh->element3s = (unsigned short *)data;data += maxtriangles * sizeof(unsigned short[3]); return newmesh; } shadowmesh_t *Mod_ShadowMesh_ReAlloc(mempool_t *mempool, shadowmesh_t *oldmesh, int light, int neighbors) { shadowmesh_t *newmesh; newmesh = Mod_ShadowMesh_Alloc(mempool, oldmesh->numverts, oldmesh->numtriangles, oldmesh->map_diffuse, oldmesh->map_specular, oldmesh->map_normal, light, neighbors, false); newmesh->numverts = oldmesh->numverts; newmesh->numtriangles = oldmesh->numtriangles; memcpy(newmesh->vertex3f, oldmesh->vertex3f, oldmesh->numverts * sizeof(float[3])); if (newmesh->svector3f && oldmesh->svector3f) { memcpy(newmesh->svector3f, oldmesh->svector3f, oldmesh->numverts * sizeof(float[3])); memcpy(newmesh->tvector3f, oldmesh->tvector3f, oldmesh->numverts * sizeof(float[3])); memcpy(newmesh->normal3f, oldmesh->normal3f, oldmesh->numverts * sizeof(float[3])); memcpy(newmesh->texcoord2f, oldmesh->texcoord2f, oldmesh->numverts * sizeof(float[2])); } memcpy(newmesh->element3i, oldmesh->element3i, oldmesh->numtriangles * sizeof(int[3])); if (newmesh->neighbor3i && oldmesh->neighbor3i) memcpy(newmesh->neighbor3i, oldmesh->neighbor3i, oldmesh->numtriangles * sizeof(int[3])); return newmesh; } int Mod_ShadowMesh_AddVertex(shadowmesh_t *mesh, float *vertex14f) { int hashindex, vnum; shadowmeshvertexhash_t *hash; // this uses prime numbers intentionally hashindex = (unsigned int) (vertex14f[0] * 2003 + vertex14f[1] * 4001 + vertex14f[2] * 7919) % SHADOWMESHVERTEXHASH; for (hash = mesh->vertexhashtable[hashindex];hash;hash = hash->next) { vnum = (hash - mesh->vertexhashentries); if ((mesh->vertex3f == NULL || (mesh->vertex3f[vnum * 3 + 0] == vertex14f[0] && mesh->vertex3f[vnum * 3 + 1] == vertex14f[1] && mesh->vertex3f[vnum * 3 + 2] == vertex14f[2])) && (mesh->svector3f == NULL || (mesh->svector3f[vnum * 3 + 0] == vertex14f[3] && mesh->svector3f[vnum * 3 + 1] == vertex14f[4] && mesh->svector3f[vnum * 3 + 2] == vertex14f[5])) && (mesh->tvector3f == NULL || (mesh->tvector3f[vnum * 3 + 0] == vertex14f[6] && mesh->tvector3f[vnum * 3 + 1] == vertex14f[7] && mesh->tvector3f[vnum * 3 + 2] == vertex14f[8])) && (mesh->normal3f == NULL || (mesh->normal3f[vnum * 3 + 0] == vertex14f[9] && mesh->normal3f[vnum * 3 + 1] == vertex14f[10] && mesh->normal3f[vnum * 3 + 2] == vertex14f[11])) && (mesh->texcoord2f == NULL || (mesh->texcoord2f[vnum * 2 + 0] == vertex14f[12] && mesh->texcoord2f[vnum * 2 + 1] == vertex14f[13]))) return hash - mesh->vertexhashentries; } vnum = mesh->numverts++; hash = mesh->vertexhashentries + vnum; hash->next = mesh->vertexhashtable[hashindex]; mesh->vertexhashtable[hashindex] = hash; if (mesh->vertex3f) {mesh->vertex3f[vnum * 3 + 0] = vertex14f[0];mesh->vertex3f[vnum * 3 + 1] = vertex14f[1];mesh->vertex3f[vnum * 3 + 2] = vertex14f[2];} if (mesh->svector3f) {mesh->svector3f[vnum * 3 + 0] = vertex14f[3];mesh->svector3f[vnum * 3 + 1] = vertex14f[4];mesh->svector3f[vnum * 3 + 2] = vertex14f[5];} if (mesh->tvector3f) {mesh->tvector3f[vnum * 3 + 0] = vertex14f[6];mesh->tvector3f[vnum * 3 + 1] = vertex14f[7];mesh->tvector3f[vnum * 3 + 2] = vertex14f[8];} if (mesh->normal3f) {mesh->normal3f[vnum * 3 + 0] = vertex14f[9];mesh->normal3f[vnum * 3 + 1] = vertex14f[10];mesh->normal3f[vnum * 3 + 2] = vertex14f[11];} if (mesh->texcoord2f) {mesh->texcoord2f[vnum * 2 + 0] = vertex14f[12];mesh->texcoord2f[vnum * 2 + 1] = vertex14f[13];} return vnum; } void Mod_ShadowMesh_AddTriangle(mempool_t *mempool, shadowmesh_t *mesh, rtexture_t *map_diffuse, rtexture_t *map_specular, rtexture_t *map_normal, float *vertex14f) { if (mesh->numtriangles == 0) { // set the properties on this empty mesh to be more favorable... // (note: this case only occurs for the first triangle added to a new mesh chain) mesh->map_diffuse = map_diffuse; mesh->map_specular = map_specular; mesh->map_normal = map_normal; } while (mesh->map_diffuse != map_diffuse || mesh->map_specular != map_specular || mesh->map_normal != map_normal || mesh->numverts + 3 > mesh->maxverts || mesh->numtriangles + 1 > mesh->maxtriangles) { if (mesh->next == NULL) mesh->next = Mod_ShadowMesh_Alloc(mempool, max(mesh->maxverts, 300), max(mesh->maxtriangles, 100), map_diffuse, map_specular, map_normal, mesh->svector3f != NULL, mesh->neighbor3i != NULL, true); mesh = mesh->next; } mesh->element3i[mesh->numtriangles * 3 + 0] = Mod_ShadowMesh_AddVertex(mesh, vertex14f + 14 * 0); mesh->element3i[mesh->numtriangles * 3 + 1] = Mod_ShadowMesh_AddVertex(mesh, vertex14f + 14 * 1); mesh->element3i[mesh->numtriangles * 3 + 2] = Mod_ShadowMesh_AddVertex(mesh, vertex14f + 14 * 2); mesh->numtriangles++; } void Mod_ShadowMesh_AddMesh(mempool_t *mempool, shadowmesh_t *mesh, rtexture_t *map_diffuse, rtexture_t *map_specular, rtexture_t *map_normal, const float *vertex3f, const float *svector3f, const float *tvector3f, const float *normal3f, const float *texcoord2f, int numtris, const int *element3i) { int i, j, e; float vbuf[3*14], *v; memset(vbuf, 0, sizeof(vbuf)); for (i = 0;i < numtris;i++) { for (j = 0, v = vbuf;j < 3;j++, v += 14) { e = *element3i++; if (vertex3f) { v[0] = vertex3f[e * 3 + 0]; v[1] = vertex3f[e * 3 + 1]; v[2] = vertex3f[e * 3 + 2]; } if (svector3f) { v[3] = svector3f[e * 3 + 0]; v[4] = svector3f[e * 3 + 1]; v[5] = svector3f[e * 3 + 2]; } if (tvector3f) { v[6] = tvector3f[e * 3 + 0]; v[7] = tvector3f[e * 3 + 1]; v[8] = tvector3f[e * 3 + 2]; } if (normal3f) { v[9] = normal3f[e * 3 + 0]; v[10] = normal3f[e * 3 + 1]; v[11] = normal3f[e * 3 + 2]; } if (texcoord2f) { v[12] = texcoord2f[e * 2 + 0]; v[13] = texcoord2f[e * 2 + 1]; } } Mod_ShadowMesh_AddTriangle(mempool, mesh, map_diffuse, map_specular, map_normal, vbuf); } // the triangle calculation can take a while, so let's do a keepalive here CL_KeepaliveMessage(false); } shadowmesh_t *Mod_ShadowMesh_Begin(mempool_t *mempool, int maxverts, int maxtriangles, rtexture_t *map_diffuse, rtexture_t *map_specular, rtexture_t *map_normal, int light, int neighbors, int expandable) { // the preparation before shadow mesh initialization can take a while, so let's do a keepalive here CL_KeepaliveMessage(false); return Mod_ShadowMesh_Alloc(mempool, maxverts, maxtriangles, map_diffuse, map_specular, map_normal, light, neighbors, expandable); } static void Mod_ShadowMesh_CreateVBOs(shadowmesh_t *mesh) { if (!gl_support_arb_vertex_buffer_object) return; // element buffer is easy because it's just one array if (mesh->numtriangles) { if (mesh->element3s) mesh->ebo3s = R_Mesh_CreateStaticBufferObject(GL_ELEMENT_ARRAY_BUFFER_ARB, mesh->element3s, mesh->numtriangles * sizeof(unsigned short[3]), "shadowmesh"); else mesh->ebo3i = R_Mesh_CreateStaticBufferObject(GL_ELEMENT_ARRAY_BUFFER_ARB, mesh->element3i, mesh->numtriangles * sizeof(unsigned int[3]), "shadowmesh"); } // vertex buffer is several arrays and we put them in the same buffer // // is this wise? the texcoordtexture2f array is used with dynamic // vertex/svector/tvector/normal when rendering animated models, on the // other hand animated models don't use a lot of vertices anyway... if (mesh->numverts) { size_t size; unsigned char *mem; size = 0; mesh->vbooffset_vertex3f = size;if (mesh->vertex3f ) size += mesh->numverts * sizeof(float[3]); mesh->vbooffset_svector3f = size;if (mesh->svector3f ) size += mesh->numverts * sizeof(float[3]); mesh->vbooffset_tvector3f = size;if (mesh->tvector3f ) size += mesh->numverts * sizeof(float[3]); mesh->vbooffset_normal3f = size;if (mesh->normal3f ) size += mesh->numverts * sizeof(float[3]); mesh->vbooffset_texcoord2f = size;if (mesh->texcoord2f ) size += mesh->numverts * sizeof(float[2]); mem = (unsigned char *)Mem_Alloc(tempmempool, size); if (mesh->vertex3f ) memcpy(mem + mesh->vbooffset_vertex3f , mesh->vertex3f , mesh->numverts * sizeof(float[3])); if (mesh->svector3f ) memcpy(mem + mesh->vbooffset_svector3f , mesh->svector3f , mesh->numverts * sizeof(float[3])); if (mesh->tvector3f ) memcpy(mem + mesh->vbooffset_tvector3f , mesh->tvector3f , mesh->numverts * sizeof(float[3])); if (mesh->normal3f ) memcpy(mem + mesh->vbooffset_normal3f , mesh->normal3f , mesh->numverts * sizeof(float[3])); if (mesh->texcoord2f ) memcpy(mem + mesh->vbooffset_texcoord2f , mesh->texcoord2f , mesh->numverts * sizeof(float[2])); mesh->vbo = R_Mesh_CreateStaticBufferObject(GL_ARRAY_BUFFER_ARB, mem, size, "shadowmesh"); Mem_Free(mem); } } shadowmesh_t *Mod_ShadowMesh_Finish(mempool_t *mempool, shadowmesh_t *firstmesh, qboolean light, qboolean neighbors, qboolean createvbo) { shadowmesh_t *mesh, *newmesh, *nextmesh; // reallocate meshs to conserve space for (mesh = firstmesh, firstmesh = NULL;mesh;mesh = nextmesh) { nextmesh = mesh->next; if (mesh->numverts >= 3 && mesh->numtriangles >= 1) { newmesh = Mod_ShadowMesh_ReAlloc(mempool, mesh, light, neighbors); newmesh->next = firstmesh; firstmesh = newmesh; if (newmesh->element3s) { int i; for (i = 0;i < newmesh->numtriangles*3;i++) newmesh->element3s[i] = newmesh->element3i[i]; } if (createvbo) Mod_ShadowMesh_CreateVBOs(newmesh); } Mem_Free(mesh); } // this can take a while, so let's do a keepalive here CL_KeepaliveMessage(false); return firstmesh; } void Mod_ShadowMesh_CalcBBox(shadowmesh_t *firstmesh, vec3_t mins, vec3_t maxs, vec3_t center, float *radius) { int i; shadowmesh_t *mesh; vec3_t nmins, nmaxs, ncenter, temp; float nradius2, dist2, *v; VectorClear(nmins); VectorClear(nmaxs); // calculate bbox for (mesh = firstmesh;mesh;mesh = mesh->next) { if (mesh == firstmesh) { VectorCopy(mesh->vertex3f, nmins); VectorCopy(mesh->vertex3f, nmaxs); } for (i = 0, v = mesh->vertex3f;i < mesh->numverts;i++, v += 3) { if (nmins[0] > v[0]) nmins[0] = v[0];if (nmaxs[0] < v[0]) nmaxs[0] = v[0]; if (nmins[1] > v[1]) nmins[1] = v[1];if (nmaxs[1] < v[1]) nmaxs[1] = v[1]; if (nmins[2] > v[2]) nmins[2] = v[2];if (nmaxs[2] < v[2]) nmaxs[2] = v[2]; } } // calculate center and radius ncenter[0] = (nmins[0] + nmaxs[0]) * 0.5f; ncenter[1] = (nmins[1] + nmaxs[1]) * 0.5f; ncenter[2] = (nmins[2] + nmaxs[2]) * 0.5f; nradius2 = 0; for (mesh = firstmesh;mesh;mesh = mesh->next) { for (i = 0, v = mesh->vertex3f;i < mesh->numverts;i++, v += 3) { VectorSubtract(v, ncenter, temp); dist2 = DotProduct(temp, temp); if (nradius2 < dist2) nradius2 = dist2; } } // return data if (mins) VectorCopy(nmins, mins); if (maxs) VectorCopy(nmaxs, maxs); if (center) VectorCopy(ncenter, center); if (radius) *radius = sqrt(nradius2); } void Mod_ShadowMesh_Free(shadowmesh_t *mesh) { shadowmesh_t *nextmesh; for (;mesh;mesh = nextmesh) { if (mesh->ebo3i) R_Mesh_DestroyBufferObject(mesh->ebo3i); if (mesh->ebo3s) R_Mesh_DestroyBufferObject(mesh->ebo3s); if (mesh->vbo) R_Mesh_DestroyBufferObject(mesh->vbo); nextmesh = mesh->next; Mem_Free(mesh); } } void Mod_GetTerrainVertex3fTexCoord2fFromBGRA(const unsigned char *imagepixels, int imagewidth, int imageheight, int ix, int iy, float *vertex3f, float *texcoord2f, matrix4x4_t *pixelstepmatrix, matrix4x4_t *pixeltexturestepmatrix) { float v[3], tc[3]; v[0] = ix; v[1] = iy; if (ix >= 0 && iy >= 0 && ix < imagewidth && iy < imageheight) v[2] = (imagepixels[((iy*imagewidth)+ix)*4+0] + imagepixels[((iy*imagewidth)+ix)*4+1] + imagepixels[((iy*imagewidth)+ix)*4+2]) * (1.0f / 765.0f); else v[2] = 0; Matrix4x4_Transform(pixelstepmatrix, v, vertex3f); Matrix4x4_Transform(pixeltexturestepmatrix, v, tc); texcoord2f[0] = tc[0]; texcoord2f[1] = tc[1]; } void Mod_GetTerrainVertexFromBGRA(const unsigned char *imagepixels, int imagewidth, int imageheight, int ix, int iy, float *vertex3f, float *svector3f, float *tvector3f, float *normal3f, float *texcoord2f, matrix4x4_t *pixelstepmatrix, matrix4x4_t *pixeltexturestepmatrix) { float vup[3], vdown[3], vleft[3], vright[3]; float tcup[3], tcdown[3], tcleft[3], tcright[3]; float sv[3], tv[3], nl[3]; Mod_GetTerrainVertex3fTexCoord2fFromBGRA(imagepixels, imagewidth, imageheight, ix, iy, vertex3f, texcoord2f, pixelstepmatrix, pixeltexturestepmatrix); Mod_GetTerrainVertex3fTexCoord2fFromBGRA(imagepixels, imagewidth, imageheight, ix, iy - 1, vup, tcup, pixelstepmatrix, pixeltexturestepmatrix); Mod_GetTerrainVertex3fTexCoord2fFromBGRA(imagepixels, imagewidth, imageheight, ix, iy + 1, vdown, tcdown, pixelstepmatrix, pixeltexturestepmatrix); Mod_GetTerrainVertex3fTexCoord2fFromBGRA(imagepixels, imagewidth, imageheight, ix - 1, iy, vleft, tcleft, pixelstepmatrix, pixeltexturestepmatrix); Mod_GetTerrainVertex3fTexCoord2fFromBGRA(imagepixels, imagewidth, imageheight, ix + 1, iy, vright, tcright, pixelstepmatrix, pixeltexturestepmatrix); Mod_BuildBumpVectors(vertex3f, vup, vright, texcoord2f, tcup, tcright, svector3f, tvector3f, normal3f); Mod_BuildBumpVectors(vertex3f, vright, vdown, texcoord2f, tcright, tcdown, sv, tv, nl); VectorAdd(svector3f, sv, svector3f); VectorAdd(tvector3f, tv, tvector3f); VectorAdd(normal3f, nl, normal3f); Mod_BuildBumpVectors(vertex3f, vdown, vleft, texcoord2f, tcdown, tcleft, sv, tv, nl); VectorAdd(svector3f, sv, svector3f); VectorAdd(tvector3f, tv, tvector3f); VectorAdd(normal3f, nl, normal3f); Mod_BuildBumpVectors(vertex3f, vleft, vup, texcoord2f, tcleft, tcup, sv, tv, nl); VectorAdd(svector3f, sv, svector3f); VectorAdd(tvector3f, tv, tvector3f); VectorAdd(normal3f, nl, normal3f); } void Mod_ConstructTerrainPatchFromBGRA(const unsigned char *imagepixels, int imagewidth, int imageheight, int x1, int y1, int width, int height, int *element3i, int *neighbor3i, float *vertex3f, float *svector3f, float *tvector3f, float *normal3f, float *texcoord2f, matrix4x4_t *pixelstepmatrix, matrix4x4_t *pixeltexturestepmatrix) { int x, y, ix, iy, *e; e = element3i; for (y = 0;y < height;y++) { for (x = 0;x < width;x++) { e[0] = (y + 1) * (width + 1) + (x + 0); e[1] = (y + 0) * (width + 1) + (x + 0); e[2] = (y + 1) * (width + 1) + (x + 1); e[3] = (y + 0) * (width + 1) + (x + 0); e[4] = (y + 0) * (width + 1) + (x + 1); e[5] = (y + 1) * (width + 1) + (x + 1); e += 6; } } Mod_BuildTriangleNeighbors(neighbor3i, element3i, width*height*2); for (y = 0, iy = y1;y < height + 1;y++, iy++) for (x = 0, ix = x1;x < width + 1;x++, ix++, vertex3f += 3, texcoord2f += 2, svector3f += 3, tvector3f += 3, normal3f += 3) Mod_GetTerrainVertexFromBGRA(imagepixels, imagewidth, imageheight, ix, iy, vertex3f, texcoord2f, svector3f, tvector3f, normal3f, pixelstepmatrix, pixeltexturestepmatrix); } q3wavefunc_t Mod_LoadQ3Shaders_EnumerateWaveFunc(const char *s) { if (!strcasecmp(s, "sin")) return Q3WAVEFUNC_SIN; if (!strcasecmp(s, "square")) return Q3WAVEFUNC_SQUARE; if (!strcasecmp(s, "triangle")) return Q3WAVEFUNC_TRIANGLE; if (!strcasecmp(s, "sawtooth")) return Q3WAVEFUNC_SAWTOOTH; if (!strcasecmp(s, "inversesawtooth")) return Q3WAVEFUNC_INVERSESAWTOOTH; if (!strcasecmp(s, "noise")) return Q3WAVEFUNC_NOISE; Con_DPrintf("Mod_LoadQ3Shaders: unknown wavefunc %s\n", s); return Q3WAVEFUNC_NONE; } void Mod_FreeQ3Shaders(void) { Mem_FreePool(&q3shaders_mem); } static void Q3Shader_AddToHash (q3shaderinfo_t* shader) { unsigned short hash = CRC_Block_CaseInsensitive ((const unsigned char *)shader->name, strlen (shader->name)); q3shader_hash_entry_t* entry = q3shader_data->hash + (hash % Q3SHADER_HASH_SIZE); q3shader_hash_entry_t* lastEntry = NULL; while (entry != NULL) { if (strcasecmp (entry->shader.name, shader->name) == 0) { Con_Printf("Shader '%s' already defined\n", shader->name); return; } lastEntry = entry; entry = entry->chain; } if (entry == NULL) { if (lastEntry->shader.name[0] != 0) { /* Add to chain */ q3shader_hash_entry_t* newEntry = (q3shader_hash_entry_t*) Mem_ExpandableArray_AllocRecord (&q3shader_data->hash_entries); while (lastEntry->chain != NULL) lastEntry = lastEntry->chain; lastEntry->chain = newEntry; newEntry->chain = NULL; lastEntry = newEntry; } /* else: head of chain, in hash entry array */ entry = lastEntry; } memcpy (&entry->shader, shader, sizeof (q3shaderinfo_t)); } extern cvar_t r_picmipworld; void Mod_LoadQ3Shaders(void) { int j; int fileindex; fssearch_t *search; char *f; const char *text; q3shaderinfo_t shader; q3shaderinfo_layer_t *layer; int numparameters; char parameter[TEXTURE_MAXFRAMES + 4][Q3PATHLENGTH]; Mod_FreeQ3Shaders(); q3shaders_mem = Mem_AllocPool("q3shaders", 0, NULL); q3shader_data = (q3shader_data_t*)Mem_Alloc (q3shaders_mem, sizeof (q3shader_data_t)); Mem_ExpandableArray_NewArray (&q3shader_data->hash_entries, q3shaders_mem, sizeof (q3shader_hash_entry_t), 256); Mem_ExpandableArray_NewArray (&q3shader_data->char_ptrs, q3shaders_mem, sizeof (char**), 256); search = FS_Search("scripts/*.shader", true, false); if (!search) return; for (fileindex = 0;fileindex < search->numfilenames;fileindex++) { text = f = (char *)FS_LoadFile(search->filenames[fileindex], tempmempool, false, NULL); if (!f) continue; while (COM_ParseToken_QuakeC(&text, false)) { memset (&shader, 0, sizeof(shader)); shader.reflectmin = 0; shader.reflectmax = 1; shader.refractfactor = 1; Vector4Set(shader.refractcolor4f, 1, 1, 1, 1); shader.reflectfactor = 1; Vector4Set(shader.reflectcolor4f, 1, 1, 1, 1); shader.r_water_wateralpha = 1; strlcpy(shader.name, com_token, sizeof(shader.name)); if (!COM_ParseToken_QuakeC(&text, false) || strcasecmp(com_token, "{")) { Con_Printf("%s parsing error - expected \"{\", found \"%s\"\n", search->filenames[fileindex], com_token); break; } while (COM_ParseToken_QuakeC(&text, false)) { if (!strcasecmp(com_token, "}")) break; if (!strcasecmp(com_token, "{")) { static q3shaderinfo_layer_t dummy; if (shader.numlayers < Q3SHADER_MAXLAYERS) { layer = shader.layers + shader.numlayers++; } else { // parse and process it anyway, just don't store it (so a map $lightmap or such stuff still is found) memset(&dummy, 0, sizeof(dummy)); layer = &dummy; } layer->rgbgen.rgbgen = Q3RGBGEN_IDENTITY; layer->alphagen.alphagen = Q3ALPHAGEN_IDENTITY; layer->tcgen.tcgen = Q3TCGEN_TEXTURE; layer->blendfunc[0] = GL_ONE; layer->blendfunc[1] = GL_ZERO; while (COM_ParseToken_QuakeC(&text, false)) { if (!strcasecmp(com_token, "}")) break; if (!strcasecmp(com_token, "\n")) continue; numparameters = 0; for (j = 0;strcasecmp(com_token, "\n") && strcasecmp(com_token, "}");j++) { if (j < TEXTURE_MAXFRAMES + 4) { strlcpy(parameter[j], com_token, sizeof(parameter[j])); numparameters = j + 1; } if (!COM_ParseToken_QuakeC(&text, true)) break; } for (j = numparameters;j < TEXTURE_MAXFRAMES + 4;j++) parameter[j][0] = 0; if (developer.integer >= 100) { Con_Printf("%s %i: ", shader.name, shader.numlayers - 1); for (j = 0;j < numparameters;j++) Con_Printf(" %s", parameter[j]); Con_Print("\n"); } if (numparameters >= 2 && !strcasecmp(parameter[0], "blendfunc")) { if (numparameters == 2) { if (!strcasecmp(parameter[1], "add")) { layer->blendfunc[0] = GL_ONE; layer->blendfunc[1] = GL_ONE; } else if (!strcasecmp(parameter[1], "filter")) { layer->blendfunc[0] = GL_DST_COLOR; layer->blendfunc[1] = GL_ZERO; } else if (!strcasecmp(parameter[1], "blend")) { layer->blendfunc[0] = GL_SRC_ALPHA; layer->blendfunc[1] = GL_ONE_MINUS_SRC_ALPHA; } } else if (numparameters == 3) { int k; for (k = 0;k < 2;k++) { if (!strcasecmp(parameter[k+1], "GL_ONE")) layer->blendfunc[k] = GL_ONE; else if (!strcasecmp(parameter[k+1], "GL_ZERO")) layer->blendfunc[k] = GL_ZERO; else if (!strcasecmp(parameter[k+1], "GL_SRC_COLOR")) layer->blendfunc[k] = GL_SRC_COLOR; else if (!strcasecmp(parameter[k+1], "GL_SRC_ALPHA")) layer->blendfunc[k] = GL_SRC_ALPHA; else if (!strcasecmp(parameter[k+1], "GL_DST_COLOR")) layer->blendfunc[k] = GL_DST_COLOR; else if (!strcasecmp(parameter[k+1], "GL_DST_ALPHA")) layer->blendfunc[k] = GL_ONE_MINUS_DST_ALPHA; else if (!strcasecmp(parameter[k+1], "GL_ONE_MINUS_SRC_COLOR")) layer->blendfunc[k] = GL_ONE_MINUS_SRC_COLOR; else if (!strcasecmp(parameter[k+1], "GL_ONE_MINUS_SRC_ALPHA")) layer->blendfunc[k] = GL_ONE_MINUS_SRC_ALPHA; else if (!strcasecmp(parameter[k+1], "GL_ONE_MINUS_DST_COLOR")) layer->blendfunc[k] = GL_ONE_MINUS_DST_COLOR; else if (!strcasecmp(parameter[k+1], "GL_ONE_MINUS_DST_ALPHA")) layer->blendfunc[k] = GL_ONE_MINUS_DST_ALPHA; else layer->blendfunc[k] = GL_ONE; // default in case of parsing error } } } if (numparameters >= 2 && !strcasecmp(parameter[0], "alphafunc")) layer->alphatest = true; if (numparameters >= 2 && (!strcasecmp(parameter[0], "map") || !strcasecmp(parameter[0], "clampmap"))) { if (!strcasecmp(parameter[0], "clampmap")) layer->clampmap = true; layer->numframes = 1; layer->framerate = 1; layer->texturename = (char**)Mem_ExpandableArray_AllocRecord ( &q3shader_data->char_ptrs); layer->texturename[0] = Mem_strdup (q3shaders_mem, parameter[1]); if (!strcasecmp(parameter[1], "$lightmap")) shader.lighting = true; } else if (numparameters >= 3 && (!strcasecmp(parameter[0], "animmap") || !strcasecmp(parameter[0], "animclampmap"))) { int i; layer->numframes = min(numparameters - 2, TEXTURE_MAXFRAMES); layer->framerate = atof(parameter[1]); layer->texturename = (char **) Mem_Alloc (q3shaders_mem, sizeof (char*) * layer->numframes); for (i = 0;i < layer->numframes;i++) layer->texturename[i] = Mem_strdup (q3shaders_mem, parameter[i + 2]); } else if (numparameters >= 2 && !strcasecmp(parameter[0], "rgbgen")) { int i; for (i = 0;i < numparameters - 2 && i < Q3RGBGEN_MAXPARMS;i++) layer->rgbgen.parms[i] = atof(parameter[i+2]); if (!strcasecmp(parameter[1], "identity")) layer->rgbgen.rgbgen = Q3RGBGEN_IDENTITY; else if (!strcasecmp(parameter[1], "const")) layer->rgbgen.rgbgen = Q3RGBGEN_CONST; else if (!strcasecmp(parameter[1], "entity")) layer->rgbgen.rgbgen = Q3RGBGEN_ENTITY; else if (!strcasecmp(parameter[1], "exactvertex")) layer->rgbgen.rgbgen = Q3RGBGEN_EXACTVERTEX; else if (!strcasecmp(parameter[1], "identitylighting")) layer->rgbgen.rgbgen = Q3RGBGEN_IDENTITYLIGHTING; else if (!strcasecmp(parameter[1], "lightingdiffuse")) layer->rgbgen.rgbgen = Q3RGBGEN_LIGHTINGDIFFUSE; else if (!strcasecmp(parameter[1], "oneminusentity")) layer->rgbgen.rgbgen = Q3RGBGEN_ONEMINUSENTITY; else if (!strcasecmp(parameter[1], "oneminusvertex")) layer->rgbgen.rgbgen = Q3RGBGEN_ONEMINUSVERTEX; else if (!strcasecmp(parameter[1], "vertex")) layer->rgbgen.rgbgen = Q3RGBGEN_VERTEX; else if (!strcasecmp(parameter[1], "wave")) { layer->rgbgen.rgbgen = Q3RGBGEN_WAVE; layer->rgbgen.wavefunc = Mod_LoadQ3Shaders_EnumerateWaveFunc(parameter[2]); for (i = 0;i < numparameters - 3 && i < Q3WAVEPARMS;i++) layer->rgbgen.waveparms[i] = atof(parameter[i+3]); } else Con_DPrintf("%s parsing warning: unknown rgbgen %s\n", search->filenames[fileindex], parameter[1]); } else if (numparameters >= 2 && !strcasecmp(parameter[0], "alphagen")) { int i; for (i = 0;i < numparameters - 2 && i < Q3ALPHAGEN_MAXPARMS;i++) layer->alphagen.parms[i] = atof(parameter[i+2]); if (!strcasecmp(parameter[1], "identity")) layer->alphagen.alphagen = Q3ALPHAGEN_IDENTITY; else if (!strcasecmp(parameter[1], "const")) layer->alphagen.alphagen = Q3ALPHAGEN_CONST; else if (!strcasecmp(parameter[1], "entity")) layer->alphagen.alphagen = Q3ALPHAGEN_ENTITY; else if (!strcasecmp(parameter[1], "lightingspecular")) layer->alphagen.alphagen = Q3ALPHAGEN_LIGHTINGSPECULAR; else if (!strcasecmp(parameter[1], "oneminusentity")) layer->alphagen.alphagen = Q3ALPHAGEN_ONEMINUSENTITY; else if (!strcasecmp(parameter[1], "oneminusvertex")) layer->alphagen.alphagen = Q3ALPHAGEN_ONEMINUSVERTEX; else if (!strcasecmp(parameter[1], "portal")) layer->alphagen.alphagen = Q3ALPHAGEN_PORTAL; else if (!strcasecmp(parameter[1], "vertex")) layer->alphagen.alphagen = Q3ALPHAGEN_VERTEX; else if (!strcasecmp(parameter[1], "wave")) { layer->alphagen.alphagen = Q3ALPHAGEN_WAVE; layer->alphagen.wavefunc = Mod_LoadQ3Shaders_EnumerateWaveFunc(parameter[2]); for (i = 0;i < numparameters - 3 && i < Q3WAVEPARMS;i++) layer->alphagen.waveparms[i] = atof(parameter[i+3]); } else Con_DPrintf("%s parsing warning: unknown alphagen %s\n", search->filenames[fileindex], parameter[1]); } else if (numparameters >= 2 && (!strcasecmp(parameter[0], "texgen") || !strcasecmp(parameter[0], "tcgen"))) { int i; // observed values: tcgen environment // no other values have been observed in real shaders for (i = 0;i < numparameters - 2 && i < Q3TCGEN_MAXPARMS;i++) layer->tcgen.parms[i] = atof(parameter[i+2]); if (!strcasecmp(parameter[1], "base")) layer->tcgen.tcgen = Q3TCGEN_TEXTURE; else if (!strcasecmp(parameter[1], "texture")) layer->tcgen.tcgen = Q3TCGEN_TEXTURE; else if (!strcasecmp(parameter[1], "environment")) layer->tcgen.tcgen = Q3TCGEN_ENVIRONMENT; else if (!strcasecmp(parameter[1], "lightmap")) layer->tcgen.tcgen = Q3TCGEN_LIGHTMAP; else if (!strcasecmp(parameter[1], "vector")) layer->tcgen.tcgen = Q3TCGEN_VECTOR; else Con_DPrintf("%s parsing warning: unknown tcgen mode %s\n", search->filenames[fileindex], parameter[1]); } else if (numparameters >= 2 && !strcasecmp(parameter[0], "tcmod")) { int i, tcmodindex; // observed values: // tcmod rotate # // tcmod scale # # // tcmod scroll # # // tcmod stretch sin # # # # // tcmod stretch triangle # # # # // tcmod transform # # # # # # // tcmod turb # # # # // tcmod turb sin # # # # (this is bogus) // no other values have been observed in real shaders for (tcmodindex = 0;tcmodindex < Q3MAXTCMODS;tcmodindex++) if (!layer->tcmods[tcmodindex].tcmod) break; if (tcmodindex < Q3MAXTCMODS) { for (i = 0;i < numparameters - 2 && i < Q3TCMOD_MAXPARMS;i++) layer->tcmods[tcmodindex].parms[i] = atof(parameter[i+2]); if (!strcasecmp(parameter[1], "entitytranslate")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_ENTITYTRANSLATE; else if (!strcasecmp(parameter[1], "rotate")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_ROTATE; else if (!strcasecmp(parameter[1], "scale")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_SCALE; else if (!strcasecmp(parameter[1], "scroll")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_SCROLL; else if (!strcasecmp(parameter[1], "page")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_PAGE; else if (!strcasecmp(parameter[1], "stretch")) { layer->tcmods[tcmodindex].tcmod = Q3TCMOD_STRETCH; layer->tcmods[tcmodindex].wavefunc = Mod_LoadQ3Shaders_EnumerateWaveFunc(parameter[2]); for (i = 0;i < numparameters - 3 && i < Q3WAVEPARMS;i++) layer->tcmods[tcmodindex].waveparms[i] = atof(parameter[i+3]); } else if (!strcasecmp(parameter[1], "transform")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_TRANSFORM; else if (!strcasecmp(parameter[1], "turb")) layer->tcmods[tcmodindex].tcmod = Q3TCMOD_TURBULENT; else Con_DPrintf("%s parsing warning: unknown tcmod mode %s\n", search->filenames[fileindex], parameter[1]); } else Con_DPrintf("%s parsing warning: too many tcmods on one layer\n", search->filenames[fileindex]); } // break out a level if it was a closing brace (not using the character here to not confuse vim) if (!strcasecmp(com_token, "}")) break; } if (layer->rgbgen.rgbgen == Q3RGBGEN_LIGHTINGDIFFUSE || layer->rgbgen.rgbgen == Q3RGBGEN_VERTEX) shader.lighting = true; if (layer->alphagen.alphagen == Q3ALPHAGEN_VERTEX) { if (layer == shader.layers + 0) { // vertex controlled transparency shader.vertexalpha = true; } else { // multilayer terrain shader or similar shader.textureblendalpha = true; } } layer->texflags = TEXF_ALPHA | TEXF_PRECACHE; if (!(shader.surfaceparms & Q3SURFACEPARM_NOMIPMAPS)) layer->texflags |= TEXF_MIPMAP; if (!(shader.textureflags & Q3TEXTUREFLAG_NOPICMIP)) layer->texflags |= TEXF_PICMIP | TEXF_COMPRESS; if (layer->clampmap) layer->texflags |= TEXF_CLAMP; continue; } numparameters = 0; for (j = 0;strcasecmp(com_token, "\n") && strcasecmp(com_token, "}");j++) { if (j < TEXTURE_MAXFRAMES + 4) { strlcpy(parameter[j], com_token, sizeof(parameter[j])); numparameters = j + 1; } if (!COM_ParseToken_QuakeC(&text, true)) break; } for (j = numparameters;j < TEXTURE_MAXFRAMES + 4;j++) parameter[j][0] = 0; if (fileindex == 0 && !strcasecmp(com_token, "}")) break; if (developer.integer >= 100) { Con_Printf("%s: ", shader.name); for (j = 0;j < numparameters;j++) Con_Printf(" %s", parameter[j]); Con_Print("\n"); } if (numparameters < 1) continue; if (!strcasecmp(parameter[0], "surfaceparm") && numparameters >= 2) { if (!strcasecmp(parameter[1], "alphashadow")) shader.surfaceparms |= Q3SURFACEPARM_ALPHASHADOW; else if (!strcasecmp(parameter[1], "areaportal")) shader.surfaceparms |= Q3SURFACEPARM_AREAPORTAL; else if (!strcasecmp(parameter[1], "botclip")) shader.surfaceparms |= Q3SURFACEPARM_BOTCLIP; else if (!strcasecmp(parameter[1], "clusterportal")) shader.surfaceparms |= Q3SURFACEPARM_CLUSTERPORTAL; else if (!strcasecmp(parameter[1], "detail")) shader.surfaceparms |= Q3SURFACEPARM_DETAIL; else if (!strcasecmp(parameter[1], "donotenter")) shader.surfaceparms |= Q3SURFACEPARM_DONOTENTER; else if (!strcasecmp(parameter[1], "dust")) shader.surfaceparms |= Q3SURFACEPARM_DUST; else if (!strcasecmp(parameter[1], "hint")) shader.surfaceparms |= Q3SURFACEPARM_HINT; else if (!strcasecmp(parameter[1], "fog")) shader.surfaceparms |= Q3SURFACEPARM_FOG; else if (!strcasecmp(parameter[1], "lava")) shader.surfaceparms |= Q3SURFACEPARM_LAVA; else if (!strcasecmp(parameter[1], "lightfilter")) shader.surfaceparms |= Q3SURFACEPARM_LIGHTFILTER; else if (!strcasecmp(parameter[1], "lightgrid")) shader.surfaceparms |= Q3SURFACEPARM_LIGHTGRID; else if (!strcasecmp(parameter[1], "metalsteps")) shader.surfaceparms |= Q3SURFACEPARM_METALSTEPS; else if (!strcasecmp(parameter[1], "nodamage")) shader.surfaceparms |= Q3SURFACEPARM_NODAMAGE; else if (!strcasecmp(parameter[1], "nodlight")) shader.surfaceparms |= Q3SURFACEPARM_NODLIGHT; else if (!strcasecmp(parameter[1], "nodraw")) shader.surfaceparms |= Q3SURFACEPARM_NODRAW; else if (!strcasecmp(parameter[1], "nodrop")) shader.surfaceparms |= Q3SURFACEPARM_NODROP; else if (!strcasecmp(parameter[1], "noimpact")) shader.surfaceparms |= Q3SURFACEPARM_NOIMPACT; else if (!strcasecmp(parameter[1], "nolightmap")) shader.surfaceparms |= Q3SURFACEPARM_NOLIGHTMAP; else if (!strcasecmp(parameter[1], "nomarks")) shader.surfaceparms |= Q3SURFACEPARM_NOMARKS; else if (!strcasecmp(parameter[1], "nomipmaps")) shader.surfaceparms |= Q3SURFACEPARM_NOMIPMAPS; else if (!strcasecmp(parameter[1], "nonsolid")) shader.surfaceparms |= Q3SURFACEPARM_NONSOLID; else if (!strcasecmp(parameter[1], "origin")) shader.surfaceparms |= Q3SURFACEPARM_ORIGIN; else if (!strcasecmp(parameter[1], "playerclip")) shader.surfaceparms |= Q3SURFACEPARM_PLAYERCLIP; else if (!strcasecmp(parameter[1], "sky")) shader.surfaceparms |= Q3SURFACEPARM_SKY; else if (!strcasecmp(parameter[1], "slick")) shader.surfaceparms |= Q3SURFACEPARM_SLICK; else if (!strcasecmp(parameter[1], "slime")) shader.surfaceparms |= Q3SURFACEPARM_SLIME; else if (!strcasecmp(parameter[1], "structural")) shader.surfaceparms |= Q3SURFACEPARM_STRUCTURAL; else if (!strcasecmp(parameter[1], "trans")) shader.surfaceparms |= Q3SURFACEPARM_TRANS; else if (!strcasecmp(parameter[1], "water")) shader.surfaceparms |= Q3SURFACEPARM_WATER; else if (!strcasecmp(parameter[1], "pointlight")) shader.surfaceparms |= Q3SURFACEPARM_POINTLIGHT; else if (!strcasecmp(parameter[1], "antiportal")) shader.surfaceparms |= Q3SURFACEPARM_ANTIPORTAL; else Con_DPrintf("%s parsing warning: unknown surfaceparm \"%s\"\n", search->filenames[fileindex], parameter[1]); } else if (!strcasecmp(parameter[0], "dpshadow")) shader.dpshadow = true; else if (!strcasecmp(parameter[0], "dpnoshadow")) shader.dpnoshadow = true; else if (!strcasecmp(parameter[0], "sky") && numparameters >= 2) { // some q3 skies don't have the sky parm set shader.surfaceparms |= Q3SURFACEPARM_SKY; strlcpy(shader.skyboxname, parameter[1], sizeof(shader.skyboxname)); } else if (!strcasecmp(parameter[0], "skyparms") && numparameters >= 2) { // some q3 skies don't have the sky parm set shader.surfaceparms |= Q3SURFACEPARM_SKY; if (!atoi(parameter[1]) && strcasecmp(parameter[1], "-")) strlcpy(shader.skyboxname, parameter[1], sizeof(shader.skyboxname)); } else if (!strcasecmp(parameter[0], "cull") && numparameters >= 2) { if (!strcasecmp(parameter[1], "disable") || !strcasecmp(parameter[1], "none") || !strcasecmp(parameter[1], "twosided")) shader.textureflags |= Q3TEXTUREFLAG_TWOSIDED; } else if (!strcasecmp(parameter[0], "nomipmaps")) shader.surfaceparms |= Q3SURFACEPARM_NOMIPMAPS; else if (!strcasecmp(parameter[0], "nopicmip")) shader.textureflags |= Q3TEXTUREFLAG_NOPICMIP; else if (!strcasecmp(parameter[0], "polygonoffset")) shader.textureflags |= Q3TEXTUREFLAG_POLYGONOFFSET; else if (!strcasecmp(parameter[0], "dp_refract") && numparameters >= 5) { shader.textureflags |= Q3TEXTUREFLAG_REFRACTION; shader.refractfactor = atof(parameter[1]); Vector4Set(shader.refractcolor4f, atof(parameter[2]), atof(parameter[3]), atof(parameter[4]), 1); } else if (!strcasecmp(parameter[0], "dp_reflect") && numparameters >= 6) { shader.textureflags |= Q3TEXTUREFLAG_REFLECTION; shader.reflectfactor = atof(parameter[1]); Vector4Set(shader.reflectcolor4f, atof(parameter[2]), atof(parameter[3]), atof(parameter[4]), atof(parameter[5])); } else if (!strcasecmp(parameter[0], "dp_water") && numparameters >= 12) { shader.textureflags |= Q3TEXTUREFLAG_WATERSHADER; shader.reflectmin = atof(parameter[1]); shader.reflectmax = atof(parameter[2]); shader.refractfactor = atof(parameter[3]); shader.reflectfactor = atof(parameter[4]); Vector4Set(shader.refractcolor4f, atof(parameter[5]), atof(parameter[6]), atof(parameter[7]), 1); Vector4Set(shader.reflectcolor4f, atof(parameter[8]), atof(parameter[9]), atof(parameter[10]), 1); shader.r_water_wateralpha = atof(parameter[11]); } else if (!strcasecmp(parameter[0], "deformvertexes") && numparameters >= 2) { int i, deformindex; for (deformindex = 0;deformindex < Q3MAXDEFORMS;deformindex++) if (!shader.deforms[deformindex].deform) break; if (deformindex < Q3MAXDEFORMS) { for (i = 0;i < numparameters - 2 && i < Q3DEFORM_MAXPARMS;i++) shader.deforms[deformindex].parms[i] = atof(parameter[i+2]); if (!strcasecmp(parameter[1], "projectionshadow")) shader.deforms[deformindex].deform = Q3DEFORM_PROJECTIONSHADOW; else if (!strcasecmp(parameter[1], "autosprite" )) shader.deforms[deformindex].deform = Q3DEFORM_AUTOSPRITE; else if (!strcasecmp(parameter[1], "autosprite2" )) shader.deforms[deformindex].deform = Q3DEFORM_AUTOSPRITE2; else if (!strcasecmp(parameter[1], "text0" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT0; else if (!strcasecmp(parameter[1], "text1" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT1; else if (!strcasecmp(parameter[1], "text2" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT2; else if (!strcasecmp(parameter[1], "text3" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT3; else if (!strcasecmp(parameter[1], "text4" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT4; else if (!strcasecmp(parameter[1], "text5" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT5; else if (!strcasecmp(parameter[1], "text6" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT6; else if (!strcasecmp(parameter[1], "text7" )) shader.deforms[deformindex].deform = Q3DEFORM_TEXT7; else if (!strcasecmp(parameter[1], "bulge" )) shader.deforms[deformindex].deform = Q3DEFORM_BULGE; else if (!strcasecmp(parameter[1], "normal" )) shader.deforms[deformindex].deform = Q3DEFORM_NORMAL; else if (!strcasecmp(parameter[1], "wave" )) { shader.deforms[deformindex].deform = Q3DEFORM_WAVE; shader.deforms[deformindex].wavefunc = Mod_LoadQ3Shaders_EnumerateWaveFunc(parameter[3]); for (i = 0;i < numparameters - 4 && i < Q3WAVEPARMS;i++) shader.deforms[deformindex].waveparms[i] = atof(parameter[i+4]); } else if (!strcasecmp(parameter[1], "move" )) { shader.deforms[deformindex].deform = Q3DEFORM_MOVE; shader.deforms[deformindex].wavefunc = Mod_LoadQ3Shaders_EnumerateWaveFunc(parameter[5]); for (i = 0;i < numparameters - 6 && i < Q3WAVEPARMS;i++) shader.deforms[deformindex].waveparms[i] = atof(parameter[i+6]); } } } } // pick the primary layer to render with if (shader.numlayers) { shader.backgroundlayer = -1; shader.primarylayer = 0; // if lightmap comes first this is definitely an ordinary texture // if the first two layers have the correct blendfuncs and use vertex alpha, it is a blended terrain shader if ((shader.layers[shader.primarylayer].texturename != NULL) && !strcasecmp(shader.layers[shader.primarylayer].texturename[0], "$lightmap")) { shader.backgroundlayer = -1; shader.primarylayer = 1; } else if (shader.numlayers >= 2 && shader.layers[1].alphagen.alphagen == Q3ALPHAGEN_VERTEX && (shader.layers[0].blendfunc[0] == GL_ONE && shader.layers[0].blendfunc[1] == GL_ZERO && !shader.layers[0].alphatest) && ((shader.layers[1].blendfunc[0] == GL_SRC_ALPHA && shader.layers[1].blendfunc[1] == GL_ONE_MINUS_SRC_ALPHA) || (shader.layers[1].blendfunc[0] == GL_ONE && shader.layers[1].blendfunc[1] == GL_ZERO && shader.layers[1].alphatest))) { // terrain blending or other effects shader.backgroundlayer = 0; shader.primarylayer = 1; } } // fix up multiple reflection types if(shader.textureflags & Q3TEXTUREFLAG_WATERSHADER) shader.textureflags &= ~(Q3TEXTUREFLAG_REFRACTION | Q3TEXTUREFLAG_REFLECTION); Q3Shader_AddToHash (&shader); } Mem_Free(f); } } q3shaderinfo_t *Mod_LookupQ3Shader(const char *name) { unsigned short hash; q3shader_hash_entry_t* entry; if (!q3shaders_mem) Mod_LoadQ3Shaders(); hash = CRC_Block_CaseInsensitive ((const unsigned char *)name, strlen (name)); entry = q3shader_data->hash + (hash % Q3SHADER_HASH_SIZE); while (entry != NULL) { if (strcasecmp (entry->shader.name, name) == 0) return &entry->shader; entry = entry->chain; } return NULL; } qboolean Mod_LoadTextureFromQ3Shader(texture_t *texture, const char *name, qboolean warnmissing, qboolean fallback, int defaulttexflags) { int j; int texflagsmask; qboolean success = true; q3shaderinfo_t *shader; if (!name) name = ""; strlcpy(texture->name, name, sizeof(texture->name)); shader = name[0] ? Mod_LookupQ3Shader(name) : NULL; texflagsmask = ~0; if(!(defaulttexflags & TEXF_PICMIP)) texflagsmask &= ~TEXF_PICMIP; if(!(defaulttexflags & TEXF_COMPRESS)) texflagsmask &= ~TEXF_COMPRESS; if (shader) { if (developer_loading.integer) Con_Printf("%s: loaded shader for %s\n", loadmodel->name, name); texture->surfaceparms = shader->surfaceparms; // allow disabling of picmip or compression by defaulttexflags texture->textureflags = shader->textureflags & texflagsmask; if (shader->surfaceparms & Q3SURFACEPARM_SKY) { texture->basematerialflags = MATERIALFLAG_SKY | MATERIALFLAG_NOSHADOW; if (shader->skyboxname[0]) { // quake3 seems to append a _ to the skybox name, so this must do so as well dpsnprintf(loadmodel->brush.skybox, sizeof(loadmodel->brush.skybox), "%s_", shader->skyboxname); } } else if ((texture->surfaceflags & Q3SURFACEFLAG_NODRAW) || shader->numlayers == 0) texture->basematerialflags = MATERIALFLAG_NODRAW | MATERIALFLAG_NOSHADOW; else texture->basematerialflags = MATERIALFLAG_WALL; if (shader->layers[0].alphatest) texture->basematerialflags |= MATERIALFLAG_ALPHATEST | MATERIALFLAG_NOSHADOW; if (shader->textureflags & Q3TEXTUREFLAG_TWOSIDED) texture->basematerialflags |= MATERIALFLAG_NOSHADOW | MATERIALFLAG_NOCULLFACE; if (shader->textureflags & Q3TEXTUREFLAG_POLYGONOFFSET) texture->biaspolygonoffset -= 2; if (shader->textureflags & Q3TEXTUREFLAG_REFRACTION) texture->basematerialflags |= MATERIALFLAG_REFRACTION; if (shader->textureflags & Q3TEXTUREFLAG_REFLECTION) texture->basematerialflags |= MATERIALFLAG_REFLECTION; if (shader->textureflags & Q3TEXTUREFLAG_WATERSHADER) texture->basematerialflags |= MATERIALFLAG_WATERSHADER; texture->customblendfunc[0] = GL_ONE; texture->customblendfunc[1] = GL_ZERO; if (shader->numlayers > 0) { texture->customblendfunc[0] = shader->layers[0].blendfunc[0]; texture->customblendfunc[1] = shader->layers[0].blendfunc[1]; /* Q3 shader blendfuncs actually used in the game (* = supported by DP) * additive GL_ONE GL_ONE additive weird GL_ONE GL_SRC_ALPHA additive weird 2 GL_ONE GL_ONE_MINUS_SRC_ALPHA * alpha GL_SRC_ALPHA GL_ONE_MINUS_SRC_ALPHA alpha inverse GL_ONE_MINUS_SRC_ALPHA GL_SRC_ALPHA brighten GL_DST_COLOR GL_ONE brighten GL_ONE GL_SRC_COLOR brighten weird GL_DST_COLOR GL_ONE_MINUS_DST_ALPHA brighten weird 2 GL_DST_COLOR GL_SRC_ALPHA * modulate GL_DST_COLOR GL_ZERO * modulate GL_ZERO GL_SRC_COLOR modulate inverse GL_ZERO GL_ONE_MINUS_SRC_COLOR modulate inverse alpha GL_ZERO GL_SRC_ALPHA modulate weird inverse GL_ONE_MINUS_DST_COLOR GL_ZERO * modulate x2 GL_DST_COLOR GL_SRC_COLOR * no blend GL_ONE GL_ZERO nothing GL_ZERO GL_ONE */ // if not opaque, figure out what blendfunc to use if (shader->layers[0].blendfunc[0] != GL_ONE || shader->layers[0].blendfunc[1] != GL_ZERO) { if (shader->layers[0].blendfunc[0] == GL_ONE && shader->layers[0].blendfunc[1] == GL_ONE) texture->basematerialflags |= MATERIALFLAG_ADD | MATERIALFLAG_BLENDED | MATERIALFLAG_NOSHADOW; else if (shader->layers[0].blendfunc[0] == GL_SRC_ALPHA && shader->layers[0].blendfunc[1] == GL_ONE) texture->basematerialflags |= MATERIALFLAG_ADD | MATERIALFLAG_BLENDED | MATERIALFLAG_NOSHADOW; else if (shader->layers[0].blendfunc[0] == GL_SRC_ALPHA && shader->layers[0].blendfunc[1] == GL_ONE_MINUS_SRC_ALPHA) texture->basematerialflags |= MATERIALFLAG_ALPHA | MATERIALFLAG_BLENDED | MATERIALFLAG_NOSHADOW; else texture->basematerialflags |= MATERIALFLAG_CUSTOMBLEND | MATERIALFLAG_FULLBRIGHT | MATERIALFLAG_BLENDED | MATERIALFLAG_NOSHADOW; } } if (!shader->lighting) texture->basematerialflags |= MATERIALFLAG_FULLBRIGHT; if (shader->primarylayer >= 0) { q3shaderinfo_layer_t* primarylayer = shader->layers + shader->primarylayer; // copy over many primarylayer parameters texture->rgbgen = primarylayer->rgbgen; texture->alphagen = primarylayer->alphagen; texture->tcgen = primarylayer->tcgen; memcpy(texture->tcmods, primarylayer->tcmods, sizeof(texture->tcmods)); // load the textures texture->numskinframes = primarylayer->numframes; texture->skinframerate = primarylayer->framerate; for (j = 0;j < primarylayer->numframes;j++) { if(cls.state == ca_dedicated) { texture->skinframes[j] = NULL; } else if (!(texture->skinframes[j] = R_SkinFrame_LoadExternal(primarylayer->texturename[j], primarylayer->texflags & texflagsmask, false))) { Con_Printf("^1%s:^7 could not load texture ^3\"%s\"^7 (frame %i) for shader ^2\"%s\"\n", loadmodel->name, primarylayer->texturename[j], j, texture->name); texture->skinframes[j] = R_SkinFrame_LoadMissing(); } } } if (shader->backgroundlayer >= 0) { q3shaderinfo_layer_t* backgroundlayer = shader->layers + shader->backgroundlayer; texture->backgroundnumskinframes = backgroundlayer->numframes; texture->backgroundskinframerate = backgroundlayer->framerate; for (j = 0;j < backgroundlayer->numframes;j++) { if(cls.state == ca_dedicated) { texture->skinframes[j] = NULL; } else if (!(texture->backgroundskinframes[j] = R_SkinFrame_LoadExternal(backgroundlayer->texturename[j], backgroundlayer->texflags & texflagsmask, false))) { Con_Printf("^1%s:^7 could not load texture ^3\"%s\"^7 (background frame %i) for shader ^2\"%s\"\n", loadmodel->name, backgroundlayer->texturename[j], j, texture->name); texture->backgroundskinframes[j] = R_SkinFrame_LoadMissing(); } } } if (shader->dpshadow) texture->basematerialflags &= ~MATERIALFLAG_NOSHADOW; if (shader->dpnoshadow) texture->basematerialflags |= MATERIALFLAG_NOSHADOW; memcpy(texture->deforms, shader->deforms, sizeof(texture->deforms)); texture->reflectmin = shader->reflectmin; texture->reflectmax = shader->reflectmax; texture->refractfactor = shader->refractfactor; Vector4Copy(shader->refractcolor4f, texture->refractcolor4f); texture->reflectfactor = shader->reflectfactor; Vector4Copy(shader->reflectcolor4f, texture->reflectcolor4f); texture->r_water_wateralpha = shader->r_water_wateralpha; } else if (!strcmp(texture->name, "noshader") || !texture->name[0]) { if (developer.integer >= 100) Con_Printf("^1%s:^7 using fallback noshader material for ^3\"%s\"\n", loadmodel->name, name); texture->surfaceparms = 0; } else if (!strcmp(texture->name, "common/nodraw") || !strcmp(texture->name, "textures/common/nodraw")) { if (developer.integer >= 100) Con_Printf("^1%s:^7 using fallback nodraw material for ^3\"%s\"\n", loadmodel->name, name); texture->surfaceparms = 0; texture->basematerialflags = MATERIALFLAG_NODRAW | MATERIALFLAG_NOSHADOW; } else { if (developer.integer >= 100) Con_Printf("^1%s:^7 No shader found for texture ^3\"%s\"\n", loadmodel->name, texture->name); texture->surfaceparms = 0; if (texture->surfaceflags & Q3SURFACEFLAG_NODRAW) texture->basematerialflags |= MATERIALFLAG_NODRAW | MATERIALFLAG_NOSHADOW; else if (texture->surfaceflags & Q3SURFACEFLAG_SKY) texture->basematerialflags |= MATERIALFLAG_SKY | MATERIALFLAG_NOSHADOW; else texture->basematerialflags |= MATERIALFLAG_WALL; texture->numskinframes = 1; if(cls.state == ca_dedicated) { texture->skinframes[0] = NULL; } else { if (fallback) { qboolean has_alpha; if ((texture->skinframes[0] = R_SkinFrame_LoadExternal_CheckAlpha(texture->name, defaulttexflags, false, &has_alpha))) { if(has_alpha && (defaulttexflags & TEXF_ALPHA)) texture->basematerialflags |= MATERIALFLAG_ALPHA | MATERIALFLAG_BLENDED | MATERIALFLAG_NOSHADOW; } else success = false; } else success = false; if (!success && warnmissing) Con_Printf("^1%s:^7 could not load texture ^3\"%s\"\n", loadmodel->name, texture->name); } } // init the animation variables texture->currentframe = texture; if (texture->numskinframes < 1) texture->numskinframes = 1; if (!texture->skinframes[0]) texture->skinframes[0] = R_SkinFrame_LoadMissing(); texture->currentskinframe = texture->skinframes[0]; texture->backgroundcurrentskinframe = texture->backgroundskinframes[0]; return success; } skinfile_t *Mod_LoadSkinFiles(void) { int i, words, line, wordsoverflow; char *text; const char *data; skinfile_t *skinfile = NULL, *first = NULL; skinfileitem_t *skinfileitem; char word[10][MAX_QPATH]; /* sample file: U_bodyBox,models/players/Legoman/BikerA2.tga U_RArm,models/players/Legoman/BikerA1.tga U_LArm,models/players/Legoman/BikerA1.tga U_armor,common/nodraw U_sword,common/nodraw U_shield,common/nodraw U_homb,common/nodraw U_backpack,common/nodraw U_colcha,common/nodraw tag_head, tag_weapon, tag_torso, */ memset(word, 0, sizeof(word)); for (i = 0;i < MAX_SKINS && (data = text = (char *)FS_LoadFile(va("%s_%i.skin", loadmodel->name, i), tempmempool, true, NULL));i++) { // If it's the first file we parse if (skinfile == NULL) { skinfile = (skinfile_t *)Mem_Alloc(loadmodel->mempool, sizeof(skinfile_t)); first = skinfile; } else { skinfile->next = (skinfile_t *)Mem_Alloc(loadmodel->mempool, sizeof(skinfile_t)); skinfile = skinfile->next; } skinfile->next = NULL; for(line = 0;;line++) { // parse line if (!COM_ParseToken_QuakeC(&data, true)) break; if (!strcmp(com_token, "\n")) continue; words = 0; wordsoverflow = false; do { if (words < 10) strlcpy(word[words++], com_token, sizeof (word[0])); else wordsoverflow = true; } while (COM_ParseToken_QuakeC(&data, true) && strcmp(com_token, "\n")); if (wordsoverflow) { Con_Printf("Mod_LoadSkinFiles: parsing error in file \"%s_%i.skin\" on line #%i: line with too many statements, skipping\n", loadmodel->name, i, line); continue; } // words is always >= 1 if (!strcmp(word[0], "replace")) { if (words == 3) { if (developer_loading.integer) Con_Printf("Mod_LoadSkinFiles: parsed mesh \"%s\" shader replacement \"%s\"\n", word[1], word[2]); skinfileitem = (skinfileitem_t *)Mem_Alloc(loadmodel->mempool, sizeof(skinfileitem_t)); skinfileitem->next = skinfile->items; skinfile->items = skinfileitem; strlcpy (skinfileitem->name, word[1], sizeof (skinfileitem->name)); strlcpy (skinfileitem->replacement, word[2], sizeof (skinfileitem->replacement)); } else Con_Printf("Mod_LoadSkinFiles: parsing error in file \"%s_%i.skin\" on line #%i: wrong number of parameters to command \"%s\", see documentation in DP_GFX_SKINFILES extension in dpextensions.qc\n", loadmodel->name, i, line, word[0]); } else if (words >= 2 && !strncmp(word[0], "tag_", 4)) { // tag name, like "tag_weapon," // not used for anything (not even in Quake3) } else if (words >= 2 && !strcmp(word[1], ",")) { // mesh shader name, like "U_RArm,models/players/Legoman/BikerA1.tga" if (developer_loading.integer) Con_Printf("Mod_LoadSkinFiles: parsed mesh \"%s\" shader replacement \"%s\"\n", word[0], word[2]); skinfileitem = (skinfileitem_t *)Mem_Alloc(loadmodel->mempool, sizeof(skinfileitem_t)); skinfileitem->next = skinfile->items; skinfile->items = skinfileitem; strlcpy (skinfileitem->name, word[0], sizeof (skinfileitem->name)); strlcpy (skinfileitem->replacement, word[2], sizeof (skinfileitem->replacement)); } else Con_Printf("Mod_LoadSkinFiles: parsing error in file \"%s_%i.skin\" on line #%i: does not look like tag or mesh specification, or replace command, see documentation in DP_GFX_SKINFILES extension in dpextensions.qc\n", loadmodel->name, i, line); } Mem_Free(text); } if (i) loadmodel->numskins = i; return first; } void Mod_FreeSkinFiles(skinfile_t *skinfile) { skinfile_t *next; skinfileitem_t *skinfileitem, *nextitem; for (;skinfile;skinfile = next) { next = skinfile->next; for (skinfileitem = skinfile->items;skinfileitem;skinfileitem = nextitem) { nextitem = skinfileitem->next; Mem_Free(skinfileitem); } Mem_Free(skinfile); } } int Mod_CountSkinFiles(skinfile_t *skinfile) { int i; for (i = 0;skinfile;skinfile = skinfile->next, i++); return i; } void Mod_SnapVertices(int numcomponents, int numvertices, float *vertices, float snap) { int i; double isnap = 1.0 / snap; for (i = 0;i < numvertices*numcomponents;i++) vertices[i] = floor(vertices[i]*isnap)*snap; } int Mod_RemoveDegenerateTriangles(int numtriangles, const int *inelement3i, int *outelement3i, const float *vertex3f) { int i, outtriangles; float edgedir1[3], edgedir2[3], temp[3]; // a degenerate triangle is one with no width (thickness, surface area) // these are characterized by having all 3 points colinear (along a line) // or having two points identical // the simplest check is to calculate the triangle's area for (i = 0, outtriangles = 0;i < numtriangles;i++, inelement3i += 3) { // calculate first edge VectorSubtract(vertex3f + inelement3i[1] * 3, vertex3f + inelement3i[0] * 3, edgedir1); VectorSubtract(vertex3f + inelement3i[2] * 3, vertex3f + inelement3i[0] * 3, edgedir2); CrossProduct(edgedir1, edgedir2, temp); if (VectorLength2(temp) < 0.001f) continue; // degenerate triangle (no area) // valid triangle (has area) VectorCopy(inelement3i, outelement3i); outelement3i += 3; outtriangles++; } return outtriangles; } void Mod_VertexRangeFromElements(int numelements, const int *elements, int *firstvertexpointer, int *lastvertexpointer) { int i, e; int firstvertex, lastvertex; if (numelements > 0 && elements) { firstvertex = lastvertex = elements[0]; for (i = 1;i < numelements;i++) { e = elements[i]; firstvertex = min(firstvertex, e); lastvertex = max(lastvertex, e); } } else firstvertex = lastvertex = 0; if (firstvertexpointer) *firstvertexpointer = firstvertex; if (lastvertexpointer) *lastvertexpointer = lastvertex; } void Mod_MakeSortedSurfaces(dp_model_t *mod) { // make an optimal set of texture-sorted batches to draw... int j, t; int *firstsurfacefortexture; int *numsurfacesfortexture; if (!mod->sortedmodelsurfaces) mod->sortedmodelsurfaces = (int *) Mem_Alloc(loadmodel->mempool, mod->nummodelsurfaces * sizeof(*mod->sortedmodelsurfaces)); firstsurfacefortexture = (int *) Mem_Alloc(tempmempool, mod->num_textures * sizeof(*firstsurfacefortexture)); numsurfacesfortexture = (int *) Mem_Alloc(tempmempool, mod->num_textures * sizeof(*numsurfacesfortexture)); memset(numsurfacesfortexture, 0, mod->num_textures * sizeof(*numsurfacesfortexture)); for (j = 0;j < mod->nummodelsurfaces;j++) { const msurface_t *surface = mod->data_surfaces + j + mod->firstmodelsurface; int t = (int)(surface->texture - mod->data_textures); numsurfacesfortexture[t]++; } j = 0; for (t = 0;t < mod->num_textures;t++) { firstsurfacefortexture[t] = j; j += numsurfacesfortexture[t]; } for (j = 0;j < mod->nummodelsurfaces;j++) { const msurface_t *surface = mod->data_surfaces + j + mod->firstmodelsurface; int t = (int)(surface->texture - mod->data_textures); mod->sortedmodelsurfaces[firstsurfacefortexture[t]++] = j + mod->firstmodelsurface; } Mem_Free(firstsurfacefortexture); Mem_Free(numsurfacesfortexture); } static void Mod_BuildVBOs(void) { if (!gl_support_arb_vertex_buffer_object) return; // element buffer is easy because it's just one array if (loadmodel->surfmesh.num_triangles) { if (loadmodel->surfmesh.data_element3s) { int i; for (i = 0;i < loadmodel->surfmesh.num_triangles*3;i++) loadmodel->surfmesh.data_element3s[i] = loadmodel->surfmesh.data_element3i[i]; loadmodel->surfmesh.ebo3s = R_Mesh_CreateStaticBufferObject(GL_ELEMENT_ARRAY_BUFFER_ARB, loadmodel->surfmesh.data_element3s, loadmodel->surfmesh.num_triangles * sizeof(unsigned short[3]), loadmodel->name); } else loadmodel->surfmesh.ebo3i = R_Mesh_CreateStaticBufferObject(GL_ELEMENT_ARRAY_BUFFER_ARB, loadmodel->surfmesh.data_element3i, loadmodel->surfmesh.num_triangles * sizeof(unsigned int[3]), loadmodel->name); } // vertex buffer is several arrays and we put them in the same buffer // // is this wise? the texcoordtexture2f array is used with dynamic // vertex/svector/tvector/normal when rendering animated models, on the // other hand animated models don't use a lot of vertices anyway... if (loadmodel->surfmesh.num_vertices) { size_t size; unsigned char *mem; size = 0; loadmodel->surfmesh.vbooffset_vertex3f = size;if (loadmodel->surfmesh.data_vertex3f ) size += loadmodel->surfmesh.num_vertices * sizeof(float[3]); loadmodel->surfmesh.vbooffset_svector3f = size;if (loadmodel->surfmesh.data_svector3f ) size += loadmodel->surfmesh.num_vertices * sizeof(float[3]); loadmodel->surfmesh.vbooffset_tvector3f = size;if (loadmodel->surfmesh.data_tvector3f ) size += loadmodel->surfmesh.num_vertices * sizeof(float[3]); loadmodel->surfmesh.vbooffset_normal3f = size;if (loadmodel->surfmesh.data_normal3f ) size += loadmodel->surfmesh.num_vertices * sizeof(float[3]); loadmodel->surfmesh.vbooffset_texcoordtexture2f = size;if (loadmodel->surfmesh.data_texcoordtexture2f ) size += loadmodel->surfmesh.num_vertices * sizeof(float[2]); loadmodel->surfmesh.vbooffset_texcoordlightmap2f = size;if (loadmodel->surfmesh.data_texcoordlightmap2f) size += loadmodel->surfmesh.num_vertices * sizeof(float[2]); loadmodel->surfmesh.vbooffset_lightmapcolor4f = size;if (loadmodel->surfmesh.data_lightmapcolor4f ) size += loadmodel->surfmesh.num_vertices * sizeof(float[4]); mem = (unsigned char *)Mem_Alloc(tempmempool, size); if (loadmodel->surfmesh.data_vertex3f ) memcpy(mem + loadmodel->surfmesh.vbooffset_vertex3f , loadmodel->surfmesh.data_vertex3f , loadmodel->surfmesh.num_vertices * sizeof(float[3])); if (loadmodel->surfmesh.data_svector3f ) memcpy(mem + loadmodel->surfmesh.vbooffset_svector3f , loadmodel->surfmesh.data_svector3f , loadmodel->surfmesh.num_vertices * sizeof(float[3])); if (loadmodel->surfmesh.data_tvector3f ) memcpy(mem + loadmodel->surfmesh.vbooffset_tvector3f , loadmodel->surfmesh.data_tvector3f , loadmodel->surfmesh.num_vertices * sizeof(float[3])); if (loadmodel->surfmesh.data_normal3f ) memcpy(mem + loadmodel->surfmesh.vbooffset_normal3f , loadmodel->surfmesh.data_normal3f , loadmodel->surfmesh.num_vertices * sizeof(float[3])); if (loadmodel->surfmesh.data_texcoordtexture2f ) memcpy(mem + loadmodel->surfmesh.vbooffset_texcoordtexture2f , loadmodel->surfmesh.data_texcoordtexture2f , loadmodel->surfmesh.num_vertices * sizeof(float[2])); if (loadmodel->surfmesh.data_texcoordlightmap2f) memcpy(mem + loadmodel->surfmesh.vbooffset_texcoordlightmap2f, loadmodel->surfmesh.data_texcoordlightmap2f, loadmodel->surfmesh.num_vertices * sizeof(float[2])); if (loadmodel->surfmesh.data_lightmapcolor4f ) memcpy(mem + loadmodel->surfmesh.vbooffset_lightmapcolor4f , loadmodel->surfmesh.data_lightmapcolor4f , loadmodel->surfmesh.num_vertices * sizeof(float[4])); loadmodel->surfmesh.vbo = R_Mesh_CreateStaticBufferObject(GL_ARRAY_BUFFER_ARB, mem, size, loadmodel->name); Mem_Free(mem); } } static void Mod_Decompile_OBJ(dp_model_t *model, const char *filename, const char *mtlfilename, const char *originalfilename) { int vertexindex, surfaceindex, triangleindex, textureindex, countvertices = 0, countsurfaces = 0, countfaces = 0, counttextures = 0; int a, b, c; const char *texname; const int *e; const float *v, *vn, *vt; size_t l; size_t outbufferpos = 0; size_t outbuffermax = 0x100000; char *outbuffer = (char *) Z_Malloc(outbuffermax), *oldbuffer; const msurface_t *surface; const int maxtextures = 256; char *texturenames = (char *) Z_Malloc(maxtextures * MAX_QPATH); // construct the mtllib file l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "# mtllib for %s exported by darkplaces engine\n", originalfilename); if (l > 0) outbufferpos += l; for (surfaceindex = 0, surface = model->data_surfaces;surfaceindex < model->num_surfaces;surfaceindex++, surface++) { countsurfaces++; countvertices += surface->num_vertices; countfaces += surface->num_triangles; texname = (surface->texture && surface->texture->name[0]) ? surface->texture->name : "default"; for (textureindex = 0;textureindex < maxtextures && texturenames[textureindex*MAX_QPATH];textureindex++) if (!strcmp(texturenames + textureindex * MAX_QPATH, texname)) break; if (textureindex >= maxtextures) continue; // just a precaution if (counttextures < textureindex + 1) counttextures = textureindex + 1; strlcpy(texturenames + textureindex * MAX_QPATH, texname, MAX_QPATH); if (outbufferpos >= outbuffermax >> 1) { outbuffermax *= 2; oldbuffer = outbuffer; outbuffer = (char *) Z_Malloc(outbuffermax); memcpy(outbuffer, oldbuffer, outbufferpos); Z_Free(oldbuffer); } l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "newmtl %s\nNs 96.078431\nKa 0 0 0\nKd 0.64 0.64 0.64\nKs 0.5 0.5 0.5\nNi 1\nd 1\nillum 2\nmap_Kd %s%s\n\n", texname, texname, strstr(texname, ".tga") ? "" : ".tga"); if (l > 0) outbufferpos += l; } // write the mtllib file FS_WriteFile(mtlfilename, outbuffer, outbufferpos); outbufferpos = 0; // construct the obj file l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "# model exported from %s by darkplaces engine\n# %i vertices, %i faces, %i surfaces\nmtllib %s\n", originalfilename, countvertices, countfaces, countsurfaces, mtlfilename); if (l > 0) outbufferpos += l; for (vertexindex = 0, v = model->surfmesh.data_vertex3f, vn = model->surfmesh.data_normal3f, vt = model->surfmesh.data_texcoordtexture2f;vertexindex < model->surfmesh.num_vertices;vertexindex++, v += 3, vn += 3, vt += 2) { if (outbufferpos >= outbuffermax >> 1) { outbuffermax *= 2; oldbuffer = outbuffer; outbuffer = (char *) Z_Malloc(outbuffermax); memcpy(outbuffer, oldbuffer, outbufferpos); Z_Free(oldbuffer); } l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "v %f %f %f\nvn %f %f %f\nvt %f %f\n", v[0], v[2], -v[1], vn[0], vn[2], -vn[1], vt[0], 1-vt[1]); if (l > 0) outbufferpos += l; } for (surfaceindex = 0, surface = model->data_surfaces;surfaceindex < model->num_surfaces;surfaceindex++, surface++) { l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "usemtl %s\n", (surface->texture && surface->texture->name[0]) ? surface->texture->name : "default"); if (l > 0) outbufferpos += l; for (triangleindex = 0, e = model->surfmesh.data_element3i + surface->num_firsttriangle * 3;triangleindex < surface->num_triangles;triangleindex++, e += 3) { if (outbufferpos >= outbuffermax >> 1) { outbuffermax *= 2; oldbuffer = outbuffer; outbuffer = (char *) Z_Malloc(outbuffermax); memcpy(outbuffer, oldbuffer, outbufferpos); Z_Free(oldbuffer); } a = e[0]+1; b = e[2]+1; c = e[1]+1; l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "f %i/%i/%i %i/%i/%i %i/%i/%i\n", a,a,a,b,b,b,c,c,c); if (l > 0) outbufferpos += l; } } // write the obj file FS_WriteFile(filename, outbuffer, outbufferpos); // clean up Z_Free(outbuffer); Z_Free(texturenames); // print some stats Con_Printf("Wrote %s (%i bytes, %i vertices, %i faces, %i surfaces with %i distinct textures)\n", filename, (int)outbufferpos, countvertices, countfaces, countsurfaces, counttextures); } static void Mod_Decompile_SMD(dp_model_t *model, const char *filename, int firstpose, int numposes, qboolean writetriangles) { int countnodes = 0, counttriangles = 0, countframes = 0; int surfaceindex; int triangleindex; int transformindex; int poseindex; int cornerindex; float modelscale; const int *e; const float *pose; size_t l; size_t outbufferpos = 0; size_t outbuffermax = 0x100000; char *outbuffer = (char *) Z_Malloc(outbuffermax), *oldbuffer; const msurface_t *surface; l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "version 1\nnodes\n"); if (l > 0) outbufferpos += l; modelscale = 1; if(model->num_poses >= 0) modelscale = sqrt(model->data_poses[0] * model->data_poses[0] + model->data_poses[1] * model->data_poses[1] + model->data_poses[2] * model->data_poses[2]); if(fabs(modelscale - 1) > 1e-4) { if(firstpose == 0) // only print the when writing the reference pose Con_Printf("The model has an old-style model scale of %f\n", modelscale); } else modelscale = 1; for (transformindex = 0;transformindex < model->num_bones;transformindex++) { if (outbufferpos >= outbuffermax >> 1) { outbuffermax *= 2; oldbuffer = outbuffer; outbuffer = (char *) Z_Malloc(outbuffermax); memcpy(outbuffer, oldbuffer, outbufferpos); Z_Free(oldbuffer); } countnodes++; l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%3i \"%s\" %3i\n", transformindex, model->data_bones[transformindex].name, model->data_bones[transformindex].parent); if (l > 0) outbufferpos += l; } l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "end\nskeleton\n"); if (l > 0) outbufferpos += l; for (poseindex = 0, pose = model->data_poses + model->num_bones * 12 * firstpose;poseindex < numposes;poseindex++) { countframes++; l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "time %i\n", poseindex); if (l > 0) outbufferpos += l; for (transformindex = 0;transformindex < model->num_bones;transformindex++, pose += 12) { float a, b, c; float angles[3]; float mtest[3][4]; if (outbufferpos >= outbuffermax >> 1) { outbuffermax *= 2; oldbuffer = outbuffer; outbuffer = (char *) Z_Malloc(outbuffermax); memcpy(outbuffer, oldbuffer, outbufferpos); Z_Free(oldbuffer); } // strangely the smd angles are for a transposed matrix, so we // have to generate a transposed matrix, then convert that... mtest[0][0] = pose[ 0]; mtest[0][1] = pose[ 4]; mtest[0][2] = pose[ 8]; mtest[0][3] = pose[ 3]; mtest[1][0] = pose[ 1]; mtest[1][1] = pose[ 5]; mtest[1][2] = pose[ 9]; mtest[1][3] = pose[ 7]; mtest[2][0] = pose[ 2]; mtest[2][1] = pose[ 6]; mtest[2][2] = pose[10]; mtest[2][3] = pose[11]; AnglesFromVectors(angles, mtest[0], mtest[2], false); if (angles[0] >= 180) angles[0] -= 360; if (angles[1] >= 180) angles[1] -= 360; if (angles[2] >= 180) angles[2] -= 360; a = DEG2RAD(angles[ROLL]); b = DEG2RAD(angles[PITCH]); c = DEG2RAD(angles[YAW]); #if 0 { float cy, sy, cp, sp, cr, sr; float test[3][4]; // smd matrix construction, for comparing to non-transposed m sy = sin(c); cy = cos(c); sp = sin(b); cp = cos(b); sr = sin(a); cr = cos(a); test[0][0] = cp*cy; test[1][0] = cp*sy; test[2][0] = -sp; test[0][1] = sr*sp*cy+cr*-sy; test[1][1] = sr*sp*sy+cr*cy; test[2][1] = sr*cp; test[0][2] = (cr*sp*cy+-sr*-sy); test[1][2] = (cr*sp*sy+-sr*cy); test[2][2] = cr*cp; test[0][3] = pose[3]; test[1][3] = pose[7]; test[2][3] = pose[11]; } #endif l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%3i %f %f %f %f %f %f\n", transformindex, pose[3] * modelscale, pose[7] * modelscale, pose[11] * modelscale, DEG2RAD(angles[ROLL]), DEG2RAD(angles[PITCH]), DEG2RAD(angles[YAW])); if (l > 0) outbufferpos += l; } } l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "end\n"); if (l > 0) outbufferpos += l; if (writetriangles) { l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "triangles\n"); if (l > 0) outbufferpos += l; for (surfaceindex = 0, surface = model->data_surfaces;surfaceindex < model->num_surfaces;surfaceindex++, surface++) { for (triangleindex = 0, e = model->surfmesh.data_element3i + surface->num_firsttriangle * 3;triangleindex < surface->num_triangles;triangleindex++, e += 3) { counttriangles++; if (outbufferpos >= outbuffermax >> 1) { outbuffermax *= 2; oldbuffer = outbuffer; outbuffer = (char *) Z_Malloc(outbuffermax); memcpy(outbuffer, oldbuffer, outbufferpos); Z_Free(oldbuffer); } l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%s\n", surface->texture && surface->texture->name[0] ? surface->texture->name : "default.bmp"); if (l > 0) outbufferpos += l; for (cornerindex = 0;cornerindex < 3;cornerindex++) { const int index = e[2-cornerindex]; const float *v = model->surfmesh.data_vertex3f + index * 3; const float *vn = model->surfmesh.data_normal3f + index * 3; const float *vt = model->surfmesh.data_texcoordtexture2f + index * 2; const int *wi = model->surfmesh.data_vertexweightindex4i + index * 4; const float *wf = model->surfmesh.data_vertexweightinfluence4f + index * 4; if (wf[3]) l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%3i %f %f %f %f %f %f %f %f 4 %i %f %i %f %i %f %i %f\n", wi[0], v[0], v[1], v[2], vn[0], vn[1], vn[2], vt[0], 1 - vt[1], wi[0], wf[0], wi[1], wf[1], wi[2], wf[2], wi[3], wf[3]); else if (wf[2]) l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%3i %f %f %f %f %f %f %f %f 3 %i %f %i %f %i %f\n" , wi[0], v[0], v[1], v[2], vn[0], vn[1], vn[2], vt[0], 1 - vt[1], wi[0], wf[0], wi[1], wf[1], wi[2], wf[2]); else if (wf[1]) l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%3i %f %f %f %f %f %f %f %f 2 %i %f %i %f\n" , wi[0], v[0], v[1], v[2], vn[0], vn[1], vn[2], vt[0], 1 - vt[1], wi[0], wf[0], wi[1], wf[1]); else l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "%3i %f %f %f %f %f %f %f %f\n" , wi[0], v[0], v[1], v[2], vn[0], vn[1], vn[2], vt[0], 1 - vt[1]); if (l > 0) outbufferpos += l; } } } l = dpsnprintf(outbuffer + outbufferpos, outbuffermax - outbufferpos, "end\n"); if (l > 0) outbufferpos += l; } FS_WriteFile(filename, outbuffer, outbufferpos); Z_Free(outbuffer); Con_Printf("Wrote %s (%i bytes, %i nodes, %i frames, %i triangles)\n", filename, (int)outbufferpos, countnodes, countframes, counttriangles); } /* ================ Mod_Decompile_f decompiles a model to editable files ================ */ static void Mod_Decompile_f(void) { int i, j, k, l, first, count; dp_model_t *mod; char inname[MAX_QPATH]; char outname[MAX_QPATH]; char mtlname[MAX_QPATH]; char basename[MAX_QPATH]; char animname[MAX_QPATH]; char animname2[MAX_QPATH]; char zymtextbuffer[16384]; char dpmtextbuffer[16384]; int zymtextsize = 0; int dpmtextsize = 0; if (Cmd_Argc() != 2) { Con_Print("usage: modeldecompile \n"); return; } strlcpy(inname, Cmd_Argv(1), sizeof(inname)); FS_StripExtension(inname, basename, sizeof(basename)); mod = Mod_ForName(inname, false, true, inname[0] == '*' ? cl.model_name[1] : NULL); if (mod->brush.submodel) { // if we're decompiling a submodel, be sure to give it a proper name based on its parent FS_StripExtension(cl.model_name[1], outname, sizeof(outname)); dpsnprintf(basename, sizeof(basename), "%s/%s", outname, mod->name); outname[0] = 0; } if (!mod) { Con_Print("No such model\n"); return; } if (!mod->surfmesh.num_triangles) { Con_Print("Empty model (or sprite)\n"); return; } // export OBJ if possible (not on sprites) if (mod->surfmesh.num_triangles) { dpsnprintf(outname, sizeof(outname), "%s_decompiled.obj", basename); dpsnprintf(mtlname, sizeof(mtlname), "%s_decompiled.mtl", basename); Mod_Decompile_OBJ(mod, outname, mtlname, inname); } // export SMD if possible (only for skeletal models) if (mod->surfmesh.num_triangles && mod->num_bones) { dpsnprintf(outname, sizeof(outname), "%s_decompiled/ref1.smd", basename); Mod_Decompile_SMD(mod, outname, 0, 1, true); l = dpsnprintf(zymtextbuffer + zymtextsize, sizeof(zymtextbuffer) - zymtextsize, "output out.zym\nscale 1\norigin 0 0 0\nmesh ref1.smd\n"); if (l > 0) zymtextsize += l; l = dpsnprintf(dpmtextbuffer + dpmtextsize, sizeof(dpmtextbuffer) - dpmtextsize, "outputdir .\nmodel out\nscale 1\norigin 0 0 0\nscene ref1.smd\n"); if (l > 0) dpmtextsize += l; for (i = 0;i < mod->numframes;i = j) { strlcpy(animname, mod->animscenes[i].name, sizeof(animname)); first = mod->animscenes[i].firstframe; if (mod->animscenes[i].framecount > 1) { // framegroup anim count = mod->animscenes[i].framecount; j = i + 1; } else { // individual frame // check for additional frames with same name for (l = 0, k = strlen(animname);animname[l];l++) if ((animname[l] < '0' || animname[l] > '9') && animname[l] != '_') k = l + 1; animname[k] = 0; count = (mod->num_poses / mod->num_bones) - first; for (j = i + 1;j < mod->numframes;j++) { strlcpy(animname2, mod->animscenes[j].name, sizeof(animname2)); for (l = 0, k = strlen(animname2);animname2[l];l++) if ((animname2[l] < '0' || animname2[l] > '9') && animname2[l] != '_') k = l + 1; animname2[k] = 0; if (strcmp(animname2, animname) || mod->animscenes[j].framecount > 1) { count = mod->animscenes[j].firstframe - first; break; } } // if it's only one frame, use the original frame name if (j == i + 1) strlcpy(animname, mod->animscenes[i].name, sizeof(animname)); } dpsnprintf(outname, sizeof(outname), "%s_decompiled/%s.smd", basename, animname); Mod_Decompile_SMD(mod, outname, first, count, false); if (zymtextsize < (int)sizeof(zymtextbuffer) - 100) { l = dpsnprintf(zymtextbuffer + zymtextsize, sizeof(zymtextbuffer) - zymtextsize, "scene %s.smd fps %g\n", animname, mod->animscenes[i].framerate); if (l > 0) zymtextsize += l; } if (dpmtextsize < (int)sizeof(dpmtextbuffer) - 100) { l = dpsnprintf(dpmtextbuffer + dpmtextsize, sizeof(dpmtextbuffer) - dpmtextsize, "scene %s.smd\n", animname); if (l > 0) dpmtextsize += l; } } if (zymtextsize) FS_WriteFile(va("%s_decompiled/out_zym.txt", basename), zymtextbuffer, (fs_offset_t)zymtextsize); if (dpmtextsize) FS_WriteFile(va("%s_decompiled/out_dpm.txt", basename), dpmtextbuffer, (fs_offset_t)dpmtextsize); } }