Files
test/source/blender/blenkernel/intern/material.c
Brecht Van Lommel 0b49dc77de 2.5: Bump Mapping
Patch by Alfredo de Greef. Considerably improves the quality of bump
mapping, and texture filtering for displacement and warp too. Mainly
this is achieved by getting the texture derivatives just right in
various cases, many thanks to Alfredo for figuring this one out, works
great.


This is enabled by default now, but disabled still for existing
textures to preserve backwards compatibility. Can be enabled with
the "New Bump" option in the material texture slot in the outliner.

Also, I made the range for the normal factor a bit smaller since this
gives stronger effects, but note that you can still type in larger
values than the slider allows.
2009-07-21 13:46:49 +00:00

1173 lines
25 KiB
C

/* material.c
*
*
* $Id$
*
* ***** BEGIN GPL LICENSE BLOCK *****
*
* 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.
*
* The Original Code is Copyright (C) 2001-2002 by NaN Holding BV.
* All rights reserved.
*
* The Original Code is: all of this file.
*
* Contributor(s): none yet.
*
* ***** END GPL LICENSE BLOCK *****
*/
#include <string.h>
#include <math.h>
#include "MEM_guardedalloc.h"
#include "DNA_curve_types.h"
#include "DNA_material_types.h"
#include "DNA_mesh_types.h"
#include "DNA_meta_types.h"
#include "DNA_node_types.h"
#include "DNA_object_types.h"
#include "DNA_scene_types.h"
#include "DNA_texture_types.h"
#include "DNA_userdef_types.h"
#include "BLI_blenlib.h"
#include "BLI_arithb.h"
#include "BKE_animsys.h"
#include "BKE_blender.h"
#include "BKE_displist.h"
#include "BKE_global.h"
#include "BKE_icons.h"
#include "BKE_library.h"
#include "BKE_main.h"
#include "BKE_material.h"
#include "BKE_mesh.h"
#include "BKE_node.h"
#include "BKE_utildefines.h"
#ifndef DISABLE_PYTHON
#include "BPY_extern.h"
#endif
#include "GPU_material.h"
/* used in UI and render */
Material defmaterial;
/* called on startup, creator.c */
void init_def_material(void)
{
init_material(&defmaterial);
}
/* not material itself */
void free_material(Material *ma)
{
MTex *mtex;
int a;
#ifndef DISABLE_PYTHON
BPY_free_scriptlink(&ma->scriptlink);
#endif
for(a=0; a<MAX_MTEX; a++) {
mtex= ma->mtex[a];
if(mtex && mtex->tex) mtex->tex->id.us--;
if(mtex) MEM_freeN(mtex);
}
if(ma->ramp_col) MEM_freeN(ma->ramp_col);
if(ma->ramp_spec) MEM_freeN(ma->ramp_spec);
BKE_free_animdata((ID *)ma);
BKE_previewimg_free(&ma->preview);
BKE_icon_delete((struct ID*)ma);
ma->id.icon_id = 0;
/* is no lib link block, but material extension */
if(ma->nodetree) {
ntreeFreeTree(ma->nodetree);
MEM_freeN(ma->nodetree);
}
if(ma->gpumaterial.first)
GPU_material_free(ma);
}
void init_material(Material *ma)
{
ma->r= ma->g= ma->b= ma->ref= 0.8;
ma->specr= ma->specg= ma->specb= 1.0;
ma->mirr= ma->mirg= ma->mirb= 1.0;
ma->spectra= 1.0;
ma->amb= 0.5;
ma->alpha= 1.0;
ma->spec= ma->hasize= 0.5;
ma->har= 50;
ma->starc= ma->ringc= 4;
ma->linec= 12;
ma->flarec= 1;
ma->flaresize= ma->subsize= 1.0;
ma->flareboost= 1;
ma->seed2= 6;
ma->friction= 0.5;
ma->refrac= 4.0;
ma->roughness= 0.5;
ma->param[0]= 0.5;
ma->param[1]= 0.1;
ma->param[2]= 0.5;
ma->param[3]= 0.1;
ma->rms= 0.1;
ma->darkness= 1.0;
ma->strand_sta= ma->strand_end= 1.0f;
ma->ang= 1.0;
ma->ray_depth= 2;
ma->ray_depth_tra= 2;
ma->fresnel_mir= 0.0;
ma->fresnel_tra= 0.0;
ma->fresnel_tra_i= 1.25;
ma->fresnel_mir_i= 1.25;
ma->tx_limit= 0.0;
ma->tx_falloff= 1.0;
ma->shad_alpha= 1.0f;
ma->gloss_mir = ma->gloss_tra= 1.0;
ma->samp_gloss_mir = ma->samp_gloss_tra= 18;
ma->adapt_thresh_mir = ma->adapt_thresh_tra = 0.005;
ma->dist_mir = 0.0;
ma->fadeto_mir = MA_RAYMIR_FADETOSKY;
ma->rampfac_col= 1.0;
ma->rampfac_spec= 1.0;
ma->pr_lamp= 3; /* two lamps, is bits */
ma->pr_type= MA_SPHERE;
ma->sss_radius[0]= 1.0f;
ma->sss_radius[1]= 1.0f;
ma->sss_radius[2]= 1.0f;
ma->sss_col[0]= 0.8f;
ma->sss_col[1]= 0.8f;
ma->sss_col[2]= 0.8f;
ma->sss_error= 0.05f;
ma->sss_scale= 0.1f;
ma->sss_ior= 1.3f;
ma->sss_colfac= 1.0f;
ma->sss_texfac= 0.0f;
ma->sss_front= 1.0f;
ma->sss_back= 1.0f;
ma->mode= MA_TRACEBLE|MA_SHADBUF|MA_SHADOW|MA_RADIO|MA_RAYBIAS|MA_TANGENT_STR;
ma->preview = NULL;
}
Material *add_material(char *name)
{
Material *ma;
ma= alloc_libblock(&G.main->mat, ID_MA, name);
init_material(ma);
return ma;
}
Material *copy_material(Material *ma)
{
Material *man;
int a;
man= copy_libblock(ma);
#if 0 // XXX old animation system
id_us_plus((ID *)man->ipo);
#endif // XXX old animation system
id_us_plus((ID *)man->group);
for(a=0; a<MAX_MTEX; a++) {
if(ma->mtex[a]) {
man->mtex[a]= MEM_mallocN(sizeof(MTex), "copymaterial");
memcpy(man->mtex[a], ma->mtex[a], sizeof(MTex));
id_us_plus((ID *)man->mtex[a]->tex);
}
}
#ifndef DISABLE_PYTHON
BPY_copy_scriptlink(&ma->scriptlink);
#endif
if(ma->ramp_col) man->ramp_col= MEM_dupallocN(ma->ramp_col);
if(ma->ramp_spec) man->ramp_spec= MEM_dupallocN(ma->ramp_spec);
if (ma->preview) man->preview = BKE_previewimg_copy(ma->preview);
if(ma->nodetree) {
man->nodetree= ntreeCopyTree(ma->nodetree, 0); /* 0 == full new tree */
}
man->gpumaterial.first= man->gpumaterial.last= NULL;
return man;
}
void make_local_material(Material *ma)
{
Object *ob;
Mesh *me;
Curve *cu;
MetaBall *mb;
Material *man;
int a, local=0, lib=0;
/* - only lib users: do nothing
* - only local users: set flag
* - mixed: make copy
*/
if(ma->id.lib==0) return;
if(ma->id.us==1) {
ma->id.lib= 0;
ma->id.flag= LIB_LOCAL;
new_id(0, (ID *)ma, 0);
for(a=0; a<MAX_MTEX; a++) {
if(ma->mtex[a]) id_lib_extern((ID *)ma->mtex[a]->tex);
}
return;
}
/* test objects */
ob= G.main->object.first;
while(ob) {
if(ob->mat) {
for(a=0; a<ob->totcol; a++) {
if(ob->mat[a]==ma) {
if(ob->id.lib) lib= 1;
else local= 1;
}
}
}
ob= ob->id.next;
}
/* test meshes */
me= G.main->mesh.first;
while(me) {
if(me->mat) {
for(a=0; a<me->totcol; a++) {
if(me->mat[a]==ma) {
if(me->id.lib) lib= 1;
else local= 1;
}
}
}
me= me->id.next;
}
/* test curves */
cu= G.main->curve.first;
while(cu) {
if(cu->mat) {
for(a=0; a<cu->totcol; a++) {
if(cu->mat[a]==ma) {
if(cu->id.lib) lib= 1;
else local= 1;
}
}
}
cu= cu->id.next;
}
/* test mballs */
mb= G.main->mball.first;
while(mb) {
if(mb->mat) {
for(a=0; a<mb->totcol; a++) {
if(mb->mat[a]==ma) {
if(mb->id.lib) lib= 1;
else local= 1;
}
}
}
mb= mb->id.next;
}
if(local && lib==0) {
ma->id.lib= 0;
ma->id.flag= LIB_LOCAL;
for(a=0; a<MAX_MTEX; a++) {
if(ma->mtex[a]) id_lib_extern((ID *)ma->mtex[a]->tex);
}
new_id(0, (ID *)ma, 0);
}
else if(local && lib) {
man= copy_material(ma);
man->id.us= 0;
/* do objects */
ob= G.main->object.first;
while(ob) {
if(ob->mat) {
for(a=0; a<ob->totcol; a++) {
if(ob->mat[a]==ma) {
if(ob->id.lib==0) {
ob->mat[a]= man;
man->id.us++;
ma->id.us--;
}
}
}
}
ob= ob->id.next;
}
/* do meshes */
me= G.main->mesh.first;
while(me) {
if(me->mat) {
for(a=0; a<me->totcol; a++) {
if(me->mat[a]==ma) {
if(me->id.lib==0) {
me->mat[a]= man;
man->id.us++;
ma->id.us--;
}
}
}
}
me= me->id.next;
}
/* do curves */
cu= G.main->curve.first;
while(cu) {
if(cu->mat) {
for(a=0; a<cu->totcol; a++) {
if(cu->mat[a]==ma) {
if(cu->id.lib==0) {
cu->mat[a]= man;
man->id.us++;
ma->id.us--;
}
}
}
}
cu= cu->id.next;
}
/* do mballs */
mb= G.main->mball.first;
while(mb) {
if(mb->mat) {
for(a=0; a<mb->totcol; a++) {
if(mb->mat[a]==ma) {
if(mb->id.lib==0) {
mb->mat[a]= man;
man->id.us++;
ma->id.us--;
}
}
}
}
mb= mb->id.next;
}
}
}
Material ***give_matarar(Object *ob)
{
Mesh *me;
Curve *cu;
MetaBall *mb;
if(ob->type==OB_MESH) {
me= ob->data;
return &(me->mat);
}
else if ELEM3(ob->type, OB_CURVE, OB_FONT, OB_SURF) {
cu= ob->data;
return &(cu->mat);
}
else if(ob->type==OB_MBALL) {
mb= ob->data;
return &(mb->mat);
}
return NULL;
}
short *give_totcolp(Object *ob)
{
Mesh *me;
Curve *cu;
MetaBall *mb;
if(ob->type==OB_MESH) {
me= ob->data;
return &(me->totcol);
}
else if ELEM3(ob->type, OB_CURVE, OB_FONT, OB_SURF) {
cu= ob->data;
return &(cu->totcol);
}
else if(ob->type==OB_MBALL) {
mb= ob->data;
return &(mb->totcol);
}
return NULL;
}
Material *give_current_material(Object *ob, int act)
{
Material ***matarar, *ma;
short *totcolp;
if(ob==NULL) return NULL;
/* if object cannot have material, totcolp==NULL */
totcolp= give_totcolp(ob);
if(totcolp==NULL || ob->totcol==0) return NULL;
if(act>ob->totcol) act= ob->totcol;
else if(act<=0) act= 1;
if(ob->matbits[act-1]) { /* in object */
ma= ob->mat[act-1];
}
else { /* in data */
/* check for inconsistancy */
if(*totcolp < ob->totcol)
ob->totcol= *totcolp;
if(act>ob->totcol) act= ob->totcol;
matarar= give_matarar(ob);
if(matarar && *matarar) ma= (*matarar)[act-1];
else ma= 0;
}
return ma;
}
ID *material_from(Object *ob, int act)
{
if(ob==0) return 0;
if(ob->totcol==0) return ob->data;
if(act==0) act= 1;
if(ob->matbits[act-1]) return (ID *)ob;
else return ob->data;
}
/* GS reads the memory pointed at in a specific ordering. There are,
* however two definitions for it. I have jotted them down here, both,
* but I think the first one is actually used. The thing is that
* big-endian systems might read this the wrong way round. OTOH, we
* constructed the IDs that are read out with this macro explicitly as
* well. I expect we'll sort it out soon... */
/* from blendef: */
#define GS(a) (*((short *)(a)))
/* from misc_util: flip the bytes from x */
/* #define GS(x) (((unsigned char *)(x))[0] << 8 | ((unsigned char *)(x))[1]) */
void test_object_materials(ID *id)
{
/* make the ob mat-array same size as 'ob->data' mat-array */
Object *ob;
Mesh *me;
Curve *cu;
MetaBall *mb;
Material **newmatar;
char *newmatbits;
int totcol=0;
if(id==0) return;
if( GS(id->name)==ID_ME ) {
me= (Mesh *)id;
totcol= me->totcol;
}
else if( GS(id->name)==ID_CU ) {
cu= (Curve *)id;
totcol= cu->totcol;
}
else if( GS(id->name)==ID_MB ) {
mb= (MetaBall *)id;
totcol= mb->totcol;
}
else return;
ob= G.main->object.first;
while(ob) {
if(ob->data==id) {
if(totcol==0) {
if(ob->totcol) {
MEM_freeN(ob->mat);
MEM_freeN(ob->matbits);
ob->mat= NULL;
ob->matbits= NULL;
}
}
else if(ob->totcol<totcol) {
newmatar= MEM_callocN(sizeof(void *)*totcol, "newmatar");
newmatbits= MEM_callocN(sizeof(char)*totcol, "newmatbits");
if(ob->totcol) {
memcpy(newmatar, ob->mat, sizeof(void *)*ob->totcol);
memcpy(newmatbits, ob->matbits, sizeof(char)*ob->totcol);
MEM_freeN(ob->mat);
MEM_freeN(ob->matbits);
}
ob->mat= newmatar;
ob->matbits= newmatbits;
}
ob->totcol= totcol;
if(ob->totcol && ob->actcol==0) ob->actcol= 1;
if(ob->actcol>ob->totcol) ob->actcol= ob->totcol;
}
ob= ob->id.next;
}
}
void assign_material(Object *ob, Material *ma, int act)
{
Material *mao, **matar, ***matarar;
char *matbits;
short *totcolp;
if(act>MAXMAT) return;
if(act<1) act= 1;
/* test arraylens */
totcolp= give_totcolp(ob);
matarar= give_matarar(ob);
if(totcolp==0 || matarar==0) return;
if(act > *totcolp) {
matar= MEM_callocN(sizeof(void *)*act, "matarray1");
if(*totcolp) {
memcpy(matar, *matarar, sizeof(void *)*(*totcolp));
MEM_freeN(*matarar);
}
*matarar= matar;
*totcolp= act;
}
if(act > ob->totcol) {
matar= MEM_callocN(sizeof(void *)*act, "matarray2");
matbits= MEM_callocN(sizeof(char)*act, "matbits1");
if( ob->totcol) {
memcpy(matar, ob->mat, sizeof(void *)*( ob->totcol ));
memcpy(matbits, ob->matbits, sizeof(char)*(*totcolp));
MEM_freeN(ob->mat);
MEM_freeN(ob->matbits);
}
ob->mat= matar;
ob->matbits= matbits;
ob->totcol= act;
/* copy object/mesh linking, or assign based on userpref */
if(ob->actcol)
ob->matbits[act-1]= ob->matbits[ob->actcol-1];
else
ob->matbits[act-1]= (U.flag & USER_MAT_ON_OB)? 1: 0;
}
/* do it */
if(ob->matbits[act-1]) { /* in object */
mao= ob->mat[act-1];
if(mao) mao->id.us--;
ob->mat[act-1]= ma;
}
else { /* in data */
mao= (*matarar)[act-1];
if(mao) mao->id.us--;
(*matarar)[act-1]= ma;
}
if(ma)
id_us_plus((ID *)ma);
test_object_materials(ob->data);
}
int find_material_index(Object *ob, Material *ma)
{
Material ***matarar;
short a, *totcolp;
if(ma==NULL) return 0;
totcolp= give_totcolp(ob);
matarar= give_matarar(ob);
if(totcolp==NULL || matarar==NULL) return 0;
for(a=0; a<*totcolp; a++)
if((*matarar)[a]==ma)
break;
if(a<*totcolp)
return a+1;
return 0;
}
void object_add_material_slot(Object *ob)
{
Material *ma;
if(ob==0) return;
if(ob->totcol>=MAXMAT) return;
ma= give_current_material(ob, ob->actcol);
assign_material(ob, ma, ob->totcol+1);
ob->actcol= ob->totcol;
}
static void do_init_render_material(Material *ma, int r_mode, float *amb)
{
MTex *mtex;
int a, needuv=0, needtang=0;
if(ma->flarec==0) ma->flarec= 1;
/* add all texcoflags from mtex, texco and mapto were cleared in advance */
for(a=0; a<MAX_MTEX; a++) {
/* separate tex switching */
if(ma->septex & (1<<a)) continue;
mtex= ma->mtex[a];
if(mtex && mtex->tex && (mtex->tex->type | (mtex->tex->use_nodes && mtex->tex->nodetree) )) {
ma->texco |= mtex->texco;
ma->mapto |= mtex->mapto;
if(r_mode & R_OSA) {
if ELEM3(mtex->tex->type, TEX_IMAGE, TEX_PLUGIN, TEX_ENVMAP) ma->texco |= TEXCO_OSA;
else if(mtex->texflag & MTEX_NEW_BUMP) ma->texco |= TEXCO_OSA; // NEWBUMP: need texture derivatives for procedurals as well
}
if(ma->texco & (TEXCO_ORCO|TEXCO_REFL|TEXCO_NORM|TEXCO_STRAND|TEXCO_STRESS)) needuv= 1;
else if(ma->texco & (TEXCO_GLOB|TEXCO_UV|TEXCO_OBJECT|TEXCO_SPEED)) needuv= 1;
else if(ma->texco & (TEXCO_LAVECTOR|TEXCO_VIEW|TEXCO_STICKY)) needuv= 1;
if((ma->mapto & MAP_NORM) && (mtex->normapspace == MTEX_NSPACE_TANGENT))
needtang= 1;
}
}
if(needtang) ma->mode |= MA_NORMAP_TANG;
else ma->mode &= ~MA_NORMAP_TANG;
if(r_mode & R_RADIO)
if(ma->mode & MA_RADIO) needuv= 1;
if(ma->mode & (MA_VERTEXCOL|MA_VERTEXCOLP|MA_FACETEXTURE)) {
needuv= 1;
if(r_mode & R_OSA) ma->texco |= TEXCO_OSA; /* for texfaces */
}
if(needuv) ma->texco |= NEED_UV;
/* since the raytracer doesnt recalc O structs for each ray, we have to preset them all */
if(r_mode & R_RAYTRACE) {
if(ma->mode & (MA_RAYMIRROR|MA_RAYTRANSP|MA_SHADOW_TRA)) {
ma->texco |= NEED_UV|TEXCO_ORCO|TEXCO_REFL|TEXCO_NORM;
if(r_mode & R_OSA) ma->texco |= TEXCO_OSA;
}
}
if(amb) {
ma->ambr= ma->amb*amb[0];
ma->ambg= ma->amb*amb[1];
ma->ambb= ma->amb*amb[2];
}
/* will become or-ed result of all node modes */
ma->mode_l= ma->mode;
ma->mode_l &= ~MA_SHLESS;
if(ma->strand_surfnor > 0.0f)
ma->mode_l |= MA_STR_SURFDIFF;
}
static void init_render_nodetree(bNodeTree *ntree, Material *basemat, int r_mode, float *amb)
{
bNode *node;
for(node=ntree->nodes.first; node; node= node->next) {
if(node->id) {
if(GS(node->id->name)==ID_MA) {
Material *ma= (Material *)node->id;
if(ma!=basemat) {
do_init_render_material(ma, r_mode, amb);
basemat->texco |= ma->texco;
basemat->mode_l |= ma->mode_l;
}
}
else if(node->type==NODE_GROUP)
init_render_nodetree((bNodeTree *)node->id, basemat, r_mode, amb);
}
}
/* parses the geom+tex nodes */
ntreeShaderGetTexcoMode(ntree, r_mode, &basemat->texco, &basemat->mode_l);
}
void init_render_material(Material *mat, int r_mode, float *amb)
{
do_init_render_material(mat, r_mode, amb);
if(mat->nodetree && mat->use_nodes) {
init_render_nodetree(mat->nodetree, mat, r_mode, amb);
ntreeBeginExecTree(mat->nodetree); /* has internal flag to detect it only does it once */
}
}
void init_render_materials(int r_mode, float *amb)
{
Material *ma;
/* clear these flags before going over materials, to make sure they
* are cleared only once, otherwise node materials contained in other
* node materials can go wrong */
for(ma= G.main->mat.first; ma; ma= ma->id.next) {
if(ma->id.us) {
ma->texco= 0;
ma->mapto= 0;
}
}
/* two steps, first initialize, then or the flags for layers */
for(ma= G.main->mat.first; ma; ma= ma->id.next) {
/* is_used flag comes back in convertblender.c */
ma->flag &= ~MA_IS_USED;
if(ma->id.us)
init_render_material(ma, r_mode, amb);
}
do_init_render_material(&defmaterial, r_mode, amb);
}
/* only needed for nodes now */
void end_render_material(Material *mat)
{
if(mat && mat->nodetree && mat->use_nodes)
ntreeEndExecTree(mat->nodetree); /* has internal flag to detect it only does it once */
}
void end_render_materials(void)
{
Material *ma;
for(ma= G.main->mat.first; ma; ma= ma->id.next)
if(ma->id.us)
end_render_material(ma);
}
static int material_in_nodetree(bNodeTree *ntree, Material *mat)
{
bNode *node;
for(node=ntree->nodes.first; node; node= node->next) {
if(node->id && GS(node->id->name)==ID_MA) {
if(node->id==(ID*)mat)
return 1;
}
else if(node->type==NODE_GROUP)
if(material_in_nodetree((bNodeTree*)node->id, mat))
return 1;
}
return 0;
}
int material_in_material(Material *parmat, Material *mat)
{
if(parmat==mat)
return 1;
else if(parmat->nodetree && parmat->use_nodes)
return material_in_nodetree(parmat->nodetree, mat);
else
return 0;
}
/* ****************** */
char colname_array[125][20]= {
"Black","DarkRed","HalfRed","Red","Red",
"DarkGreen","DarkOlive","Brown","Chocolate","OrangeRed",
"HalfGreen","GreenOlive","DryOlive","Goldenrod","DarkOrange",
"LightGreen","Chartreuse","YellowGreen","Yellow","Gold",
"Green","LawnGreen","GreenYellow","LightOlive","Yellow",
"DarkBlue","DarkPurple","HotPink","VioletPink","RedPink",
"SlateGray","DarkGrey","PalePurple","IndianRed","Tomato",
"SeaGreen","PaleGreen","GreenKhaki","LightBrown","LightSalmon",
"SpringGreen","PaleGreen","MediumOlive","YellowBrown","LightGold",
"LightGreen","LightGreen","LightGreen","GreenYellow","PaleYellow",
"HalfBlue","DarkSky","HalfMagenta","VioletRed","DeepPink",
"SteelBlue","SkyBlue","Orchid","LightHotPink","HotPink",
"SeaGreen","SlateGray","MediumGrey","Burlywood","LightPink",
"SpringGreen","Aquamarine","PaleGreen","Khaki","PaleOrange",
"SpringGreen","SeaGreen","PaleGreen","PaleWhite","YellowWhite",
"LightBlue","Purple","MediumOrchid","Magenta","Magenta",
"RoyalBlue","SlateBlue","MediumOrchid","Orchid","Magenta",
"DeepSkyBlue","LightSteelBlue","LightSkyBlue","Violet","LightPink",
"Cyan","DarkTurquoise","SkyBlue","Grey","Snow",
"Mint","Mint","Aquamarine","MintCream","Ivory",
"Blue","Blue","DarkMagenta","DarkOrchid","Magenta",
"SkyBlue","RoyalBlue","LightSlateBlue","MediumOrchid","Magenta",
"DodgerBlue","SteelBlue","MediumPurple","PalePurple","Plum",
"DeepSkyBlue","PaleBlue","LightSkyBlue","PalePurple","Thistle",
"Cyan","ColdBlue","PaleTurquoise","GhostWhite","White"
};
void automatname(Material *ma)
{
int nr, r, g, b;
float ref;
if(ma==0) return;
if(ma->mode & MA_SHLESS) ref= 1.0;
else ref= ma->ref;
r= (int)(4.99*(ref*ma->r));
g= (int)(4.99*(ref*ma->g));
b= (int)(4.99*(ref*ma->b));
nr= r + 5*g + 25*b;
if(nr>124) nr= 124;
new_id(&G.main->mat, (ID *)ma, colname_array[nr]);
}
void object_remove_material_slot(Object *ob)
{
Material *mao, ***matarar;
Object *obt;
Curve *cu;
Nurb *nu;
short *totcolp;
int a, actcol;
if(ob==NULL || ob->totcol==0) return;
/* take a mesh/curve/mball as starting point, remove 1 index,
* AND with all objects that share the ob->data
*
* after that check indices in mesh/curve/mball!!!
*/
totcolp= give_totcolp(ob);
matarar= give_matarar(ob);
/* we delete the actcol */
if(ob->totcol) {
mao= (*matarar)[ob->actcol-1];
if(mao) mao->id.us--;
}
for(a=ob->actcol; a<ob->totcol; a++)
(*matarar)[a-1]= (*matarar)[a];
(*totcolp)--;
if(*totcolp==0) {
MEM_freeN(*matarar);
*matarar= 0;
}
actcol= ob->actcol;
obt= G.main->object.first;
while(obt) {
if(obt->data==ob->data) {
/* WATCH IT: do not use actcol from ob or from obt (can become zero) */
mao= obt->mat[actcol-1];
if(mao) mao->id.us--;
for(a=actcol; a<obt->totcol; a++) {
obt->mat[a-1]= obt->mat[a];
obt->matbits[a-1]= obt->matbits[a];
}
obt->totcol--;
if(obt->actcol > obt->totcol) obt->actcol= obt->totcol;
if(obt->totcol==0) {
MEM_freeN(obt->mat);
MEM_freeN(obt->matbits);
obt->mat= 0;
obt->matbits= NULL;
}
}
obt= obt->id.next;
}
/* check indices from mesh */
if(ob->type==OB_MESH) {
Mesh *me= get_mesh(ob);
mesh_delete_material_index(me, actcol-1);
freedisplist(&ob->disp);
}
else if ELEM(ob->type, OB_CURVE, OB_SURF) {
cu= ob->data;
nu= cu->nurb.first;
while(nu) {
if(nu->mat_nr && nu->mat_nr>=actcol-1) {
nu->mat_nr--;
if (ob->type == OB_CURVE) nu->charidx--;
}
nu= nu->next;
}
freedisplist(&ob->disp);
}
}
/* r g b = current value, col = new value, fac==0 is no change */
/* if g==NULL, it only does r channel */
void ramp_blend(int type, float *r, float *g, float *b, float fac, float *col)
{
float tmp, facm= 1.0-fac;
switch (type) {
case MA_RAMP_BLEND:
*r = facm*(*r) + fac*col[0];
if(g) {
*g = facm*(*g) + fac*col[1];
*b = facm*(*b) + fac*col[2];
}
break;
case MA_RAMP_ADD:
*r += fac*col[0];
if(g) {
*g += fac*col[1];
*b += fac*col[2];
}
break;
case MA_RAMP_MULT:
*r *= (facm + fac*col[0]);
if(g) {
*g *= (facm + fac*col[1]);
*b *= (facm + fac*col[2]);
}
break;
case MA_RAMP_SCREEN:
*r = 1.0 - (facm + fac*(1.0 - col[0])) * (1.0 - *r);
if(g) {
*g = 1.0 - (facm + fac*(1.0 - col[1])) * (1.0 - *g);
*b = 1.0 - (facm + fac*(1.0 - col[2])) * (1.0 - *b);
}
break;
case MA_RAMP_OVERLAY:
if(*r < 0.5f)
*r *= (facm + 2.0f*fac*col[0]);
else
*r = 1.0 - (facm + 2.0f*fac*(1.0 - col[0])) * (1.0 - *r);
if(g) {
if(*g < 0.5f)
*g *= (facm + 2.0f*fac*col[1]);
else
*g = 1.0 - (facm + 2.0f*fac*(1.0 - col[1])) * (1.0 - *g);
if(*b < 0.5f)
*b *= (facm + 2.0f*fac*col[2]);
else
*b = 1.0 - (facm + 2.0f*fac*(1.0 - col[2])) * (1.0 - *b);
}
break;
case MA_RAMP_SUB:
*r -= fac*col[0];
if(g) {
*g -= fac*col[1];
*b -= fac*col[2];
}
break;
case MA_RAMP_DIV:
if(col[0]!=0.0)
*r = facm*(*r) + fac*(*r)/col[0];
if(g) {
if(col[1]!=0.0)
*g = facm*(*g) + fac*(*g)/col[1];
if(col[2]!=0.0)
*b = facm*(*b) + fac*(*b)/col[2];
}
break;
case MA_RAMP_DIFF:
*r = facm*(*r) + fac*fabs(*r-col[0]);
if(g) {
*g = facm*(*g) + fac*fabs(*g-col[1]);
*b = facm*(*b) + fac*fabs(*b-col[2]);
}
break;
case MA_RAMP_DARK:
tmp= fac*col[0];
if(tmp < *r) *r= tmp;
if(g) {
tmp= fac*col[1];
if(tmp < *g) *g= tmp;
tmp= fac*col[2];
if(tmp < *b) *b= tmp;
}
break;
case MA_RAMP_LIGHT:
tmp= fac*col[0];
if(tmp > *r) *r= tmp;
if(g) {
tmp= fac*col[1];
if(tmp > *g) *g= tmp;
tmp= fac*col[2];
if(tmp > *b) *b= tmp;
}
break;
case MA_RAMP_DODGE:
if(*r !=0.0){
tmp = 1.0 - fac*col[0];
if(tmp <= 0.0)
*r = 1.0;
else if ((tmp = (*r) / tmp)> 1.0)
*r = 1.0;
else
*r = tmp;
}
if(g) {
if(*g !=0.0){
tmp = 1.0 - fac*col[1];
if(tmp <= 0.0 )
*g = 1.0;
else if ((tmp = (*g) / tmp) > 1.0 )
*g = 1.0;
else
*g = tmp;
}
if(*b !=0.0){
tmp = 1.0 - fac*col[2];
if(tmp <= 0.0)
*b = 1.0;
else if ((tmp = (*b) / tmp) > 1.0 )
*b = 1.0;
else
*b = tmp;
}
}
break;
case MA_RAMP_BURN:
tmp = facm + fac*col[0];
if(tmp <= 0.0)
*r = 0.0;
else if (( tmp = (1.0 - (1.0 - (*r)) / tmp )) < 0.0)
*r = 0.0;
else if (tmp > 1.0)
*r=1.0;
else
*r = tmp;
if(g) {
tmp = facm + fac*col[1];
if(tmp <= 0.0)
*g = 0.0;
else if (( tmp = (1.0 - (1.0 - (*g)) / tmp )) < 0.0 )
*g = 0.0;
else if(tmp >1.0)
*g=1.0;
else
*g = tmp;
tmp = facm + fac*col[2];
if(tmp <= 0.0)
*b = 0.0;
else if (( tmp = (1.0 - (1.0 - (*b)) / tmp )) < 0.0 )
*b = 0.0;
else if(tmp >1.0)
*b= 1.0;
else
*b = tmp;
}
break;
case MA_RAMP_HUE:
if(g){
float rH,rS,rV;
float colH,colS,colV;
float tmpr,tmpg,tmpb;
rgb_to_hsv(col[0],col[1],col[2],&colH,&colS,&colV);
if(colS!=0 ){
rgb_to_hsv(*r,*g,*b,&rH,&rS,&rV);
hsv_to_rgb( colH , rS, rV, &tmpr, &tmpg, &tmpb);
*r = facm*(*r) + fac*tmpr;
*g = facm*(*g) + fac*tmpg;
*b = facm*(*b) + fac*tmpb;
}
}
break;
case MA_RAMP_SAT:
if(g){
float rH,rS,rV;
float colH,colS,colV;
rgb_to_hsv(*r,*g,*b,&rH,&rS,&rV);
if(rS!=0){
rgb_to_hsv(col[0],col[1],col[2],&colH,&colS,&colV);
hsv_to_rgb( rH, (facm*rS +fac*colS), rV, r, g, b);
}
}
break;
case MA_RAMP_VAL:
if(g){
float rH,rS,rV;
float colH,colS,colV;
rgb_to_hsv(*r,*g,*b,&rH,&rS,&rV);
rgb_to_hsv(col[0],col[1],col[2],&colH,&colS,&colV);
hsv_to_rgb( rH, rS, (facm*rV +fac*colV), r, g, b);
}
break;
case MA_RAMP_COLOR:
if(g){
float rH,rS,rV;
float colH,colS,colV;
float tmpr,tmpg,tmpb;
rgb_to_hsv(col[0],col[1],col[2],&colH,&colS,&colV);
if(colS!=0){
rgb_to_hsv(*r,*g,*b,&rH,&rS,&rV);
hsv_to_rgb( colH, colS, rV, &tmpr, &tmpg, &tmpb);
*r = facm*(*r) + fac*tmpr;
*g = facm*(*g) + fac*tmpg;
*b = facm*(*b) + fac*tmpb;
}
}
break;
}
}