Welcome to mirror list, hosted at ThFree Co, Russian Federation.

git.blender.org/blender.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorCampbell Barton <ideasman42@gmail.com>2012-07-08 02:51:57 +0400
committerCampbell Barton <ideasman42@gmail.com>2012-07-08 02:51:57 +0400
commitcfb7aee017d95137e19b7b006d9393b5d6a935d4 (patch)
tree005a1981af6c5a0f36959d73479cc69137842eb4 /source/blender/render/intern/source/convertblender.c
parentd58ce290e1e4dcb8d0b96259fdf29c854bfaef49 (diff)
style cleanup
Diffstat (limited to 'source/blender/render/intern/source/convertblender.c')
-rw-r--r--source/blender/render/intern/source/convertblender.c76
1 files changed, 38 insertions, 38 deletions
diff --git a/source/blender/render/intern/source/convertblender.c b/source/blender/render/intern/source/convertblender.c
index d21a22f553b..cb6a0fd1198 100644
--- a/source/blender/render/intern/source/convertblender.c
+++ b/source/blender/render/intern/source/convertblender.c
@@ -525,7 +525,7 @@ typedef struct {
} SRenderMeshToTangent;
-// interface
+/* interface */
#include "mikktspace.h"
static int GetNumFaces(const SMikkTSpaceContext * pContext)
@@ -1029,7 +1029,7 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
static int second=0;
sub_v3_v3v3(nor, vec, vec1);
- normalize_v3(nor); // nor needed as tangent
+ normalize_v3(nor); /* nor needed as tangent */
cross_v3_v3v3(cross, vec, nor);
/* turn cross in pixelsize */
@@ -1089,7 +1089,7 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
add_v3_v3(vlr->v1->co, cross);
copy_v3_v3(vlr->v1->n, nor);
vlr->v1->orco= sd->orco;
- vlr->v1->accum= -1.0f; // accum abuse for strand texco
+ vlr->v1->accum = -1.0f; /* accum abuse for strand texco */
copy_v3_v3(vlr->v2->co, vec);
sub_v3_v3v3(vlr->v2->co, vlr->v2->co, cross);
@@ -1101,8 +1101,8 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
add_v3_v3(vlr->v4->co, cross);
copy_v3_v3(vlr->v4->n, nor);
vlr->v4->orco= sd->orco;
- vlr->v4->accum= 1.0f; // accum abuse for strand texco
-
+ vlr->v4->accum = 1.0f; /* accum abuse for strand texco */
+
copy_v3_v3(vlr->v3->co, vec1);
sub_v3_v3v3(vlr->v3->co, vlr->v3->co, cross);
copy_v3_v3(vlr->v3->n, nor);
@@ -1163,7 +1163,7 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
add_v3_v3(v1->co, cross);
copy_v3_v3(v1->n, nor);
v1->orco= sd->orco;
- v1->accum= -1.0f; // accum abuse for strand texco
+ v1->accum = -1.0f; /* accum abuse for strand texco */
copy_v3_v3(v2->co, vec);
sub_v3_v3v3(v2->co, v2->co, cross);
@@ -1180,11 +1180,11 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
vlr->v2= v2;
vlr->v3= RE_findOrAddVert(obr, obr->totvert++);
vlr->v4= RE_findOrAddVert(obr, obr->totvert++);
-
- v1= vlr->v4; // cycle
- v2= vlr->v3; // cycle
-
+ v1= vlr->v4; /* cycle */
+ v2= vlr->v3; /* cycle */
+
+
if (sd->adapt) {
second=0;
copy_v3_v3(anor, nor);
@@ -1209,9 +1209,9 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
vlr->v2= v2;
vlr->v3= RE_findOrAddVert(obr, obr->totvert++);
vlr->v4= RE_findOrAddVert(obr, obr->totvert++);
-
- v1= vlr->v4; // cycle
- v2= vlr->v3; // cycle
+
+ v1= vlr->v4; /* cycle */
+ v2= vlr->v3; /* cycle */
copy_v3_v3(anor, nor);
copy_v3_v3(avec, vec);
@@ -1225,8 +1225,8 @@ static void static_particle_strand(Render *re, ObjectRen *obr, Material *ma, Par
add_v3_v3(vlr->v4->co, cross);
copy_v3_v3(vlr->v4->n, nor);
vlr->v4->orco= sd->orco;
- vlr->v4->accum= -1.0f + 2.0f*sd->time; // accum abuse for strand texco
-
+ vlr->v4->accum= -1.0f + 2.0f * sd->time; /* accum abuse for strand texco */
+
copy_v3_v3(vlr->v3->co, vec);
sub_v3_v3v3(vlr->v3->co, vlr->v3->co, cross);
copy_v3_v3(vlr->v3->n, nor);
@@ -1309,7 +1309,7 @@ static void static_particle_wire(ObjectRen *obr, Material *ma, const float vec[3
vlr->v3= vlr->v2;
vlr->v4= NULL;
- v1= vlr->v2; // cycle
+ v1= vlr->v2; /* cycle */
copy_v3_v3(v1->co, vec);
sub_v3_v3v3(vlr->n, vec, vec1);
@@ -1630,12 +1630,12 @@ static int render_new_particle_system(Render *re, ObjectRen *obr, ParticleSystem
/* 2.1 setup material stff */
ma= give_render_material(re, ob, part->omat);
-#if 0 // XXX old animation system
+#if 0 /* XXX old animation system */
if (ma->ipo) {
calc_ipo(ma->ipo, cfra);
execute_ipo((ID *)ma, ma->ipo);
}
-#endif // XXX old animation system
+#endif /* XXX old animation system */
hasize = ma->hasize;
seed = ma->seed1;
@@ -2087,10 +2087,10 @@ static int render_new_particle_system(Render *re, ObjectRen *obr, ParticleSystem
strandbuf->surface= cache_strand_surface(re, obr, psmd->dm, mat, timeoffset);
/* 4. clean up */
-#if 0 // XXX old animation system
+#if 0 /* XXX old animation system */
if (ma) do_mat_ipo(re->scene, ma);
-#endif // XXX old animation system
-
+#endif /* XXX old animation system */
+
if (orco1)
MEM_freeN(sd.orco);
@@ -3267,7 +3267,7 @@ static void init_render_mesh(Render *re, ObjectRen *obr, int timeoffset)
int a, a1, ok, vertofs;
int end, do_autosmooth = FALSE, totvert = 0;
int use_original_normals = FALSE;
- int recalc_normals = 0; // false by default
+ int recalc_normals = 0; /* false by default */
int negative_scale;
me= ob->data;
@@ -3471,7 +3471,7 @@ static void init_render_mesh(Render *re, ObjectRen *obr, int timeoffset)
int t;
mtf= RE_vlakren_get_tface(obr, vlr, mtfn++, &name, 1);
mtface= (MTFace*)layer->data;
- *mtf= mtface[a]; // copy face info
+ *mtf = mtface[a]; /* copy face info */
for (vindex=0; vindex<nr_verts; vindex++)
for (t=0; t<2; t++)
mtf->uv[vindex][t]=mtface[a].uv[rev_tab[vindex]][t];
@@ -3733,7 +3733,7 @@ static GroupObject *add_render_lamp(Render *re, Object *ob)
lar->shdwb= la->shdwb;
lar->k= la->k;
- // area
+ /* area */
lar->ray_samp= la->ray_samp;
lar->ray_sampy= la->ray_sampy;
lar->ray_sampz= la->ray_sampz;
@@ -3784,7 +3784,7 @@ static GroupObject *add_render_lamp(Render *re, Object *ob)
}
area_lamp_vectors(lar);
- init_jitter_plane(lar); // subsamples
+ init_jitter_plane(lar); /* subsamples */
}
else if (lar->type==LA_SUN) {
lar->ray_totsamp= lar->ray_samp*lar->ray_samp;
@@ -3819,7 +3819,7 @@ static GroupObject *add_render_lamp(Render *re, Object *ob)
memcpy(lar->mtex, la->mtex, MAX_MTEX*sizeof(void *));
- lar->lay= ob->lay & 0xFFFFFF; // higher 8 bits are localview layers
+ lar->lay = ob->lay & 0xFFFFFF; /* higher 8 bits are localview layers */
lar->falloff_type = la->falloff_type;
lar->ld1= la->att1;
@@ -3904,7 +3904,7 @@ static GroupObject *add_render_lamp(Render *re, Object *ob)
/* Per lamp, one shadow buffer is made. */
lar->bufflag= la->bufflag;
copy_m4_m4(mat, ob->obmat);
- initshadowbuf(re, lar, mat); // mat is altered
+ initshadowbuf(re, lar, mat); /* mat is altered */
}
@@ -4247,8 +4247,8 @@ static void check_non_flat_quads(ObjectRen *obr)
xn = dot_v3v3(nor, vlr->n);
- if (ABS(xn) < 0.999995f ) { // checked on noisy fractal grid
-
+ if (ABS(xn) < 0.999995f ) { /* checked on noisy fractal grid */
+
float d1, d2;
vlr1= RE_vlakren_copy(obr, vlr);
@@ -5478,7 +5478,7 @@ static int load_fluidsimspeedvectors(Render *re, ObjectInstanceRen *obi, float *
else
ver++;
- // get fluid velocity
+ /* get fluid velocity */
fsvec[3] = 0.0f;
//fsvec[0] = fsvec[1] = fsvec[2] = fsvec[3] = 0.0; fsvec[2] = 2.0f; // NT fixed test
for (j=0;j<3;j++) fsvec[j] = velarray[a].vel[j];
@@ -5490,31 +5490,31 @@ static int load_fluidsimspeedvectors(Render *re, ObjectInstanceRen *obi, float *
fsvec[2] = avgvel[2];
}
- // transform (=rotate) to cam space
+ /* transform (=rotate) to cam space */
camco[0] = dot_v3v3(imat[0], fsvec);
camco[1] = dot_v3v3(imat[1], fsvec);
camco[2] = dot_v3v3(imat[2], fsvec);
- // get homogenous coordinates
+ /* get homogenous coordinates */
projectvert(camco, winmat, hoco);
projectvert(ver->co, winmat, ho);
/* now map hocos to screenspace, uses very primitive clip still */
- // use ho[3] of original vertex, xy component of vel. direction
+ /* use ho[3] of original vertex, xy component of vel. direction */
if (ho[3]<0.1f) div= 10.0f;
else div= 1.0f/ho[3];
zco[0]= zmulx*hoco[0]*div;
zco[1]= zmuly*hoco[1]*div;
- // maximize speed as usual
+ /* maximize speed as usual */
len= zco[0]*zco[0] + zco[1]*zco[1];
if (len > winsq) {
len= winroot/sqrtf(len);
zco[0]*= len; zco[1]*= len;
}
-
+
speed= RE_vertren_get_winspeed(obi, ver, 1);
- // set both to the same value
+ /* set both to the same value */
speed[0]= speed[2]= zco[0];
speed[1]= speed[3]= zco[1];
//if (a < 20) fprintf(stderr,"speed %d %f,%f | camco %f,%f,%f | hoco %f,%f,%f,%f\n", a, speed[0], speed[1], camco[0],camco[1], camco[2], hoco[0],hoco[1], hoco[2],hoco[3]); // NT DEBUG
@@ -5651,7 +5651,7 @@ void RE_Database_FromScene_Vectors(Render *re, Main *bmain, Scene *sce, unsigned
continue;
}
- // NT check for fluidsim special treatment
+ /* NT check for fluidsim special treatment */
fluidmd = (FluidsimModifierData *)modifiers_findByType(obi->ob, eModifierType_Fluidsim);
if (fluidmd && fluidmd->fss && (fluidmd->fss->type & OB_FLUIDSIM_DOMAIN)) {
/* use preloaded per vertex simulation data, only does calculation for step=1 */
@@ -5664,7 +5664,7 @@ void RE_Database_FromScene_Vectors(Render *re, Main *bmain, Scene *sce, unsigned
calculate_speedvectors(re, obi, oldobi->vectors, step);
else
printf("Warning: object %s has different amount of vertices or strands on other frame\n", obi->ob->id.name+2);
- } // not fluidsim
+ } /* not fluidsim */
oldobi= oldobi->next;
}