1 |
#ifndef lint |
2 |
static const char RCSid[] = "$Id: rttree_reduce.c,v 2.3 2011/06/01 16:51:03 greg Exp $"; |
3 |
#endif |
4 |
/* |
5 |
* A utility called by genBSDF.pl to reduce tensor tree samples and output |
6 |
* in a standard format as required by XML specification for variable |
7 |
* resolution BSDF data. We are not meant to be run by the user directly. |
8 |
*/ |
9 |
|
10 |
#include "rtio.h" |
11 |
#include "rterror.h" |
12 |
#include "platform.h" |
13 |
#include <stdlib.h> |
14 |
#include <math.h> |
15 |
|
16 |
float *datarr; /* our loaded BSDF data array */ |
17 |
int ttrank = 4; /* tensor tree rank */ |
18 |
int log2g = 4; /* log2 of grid resolution */ |
19 |
int infmt = 'a'; /* input format ('a','f','d') */ |
20 |
double pctcull = 95.; /* target culling percentile */ |
21 |
|
22 |
#define dval3(ix,ox,oy) datarr[((((ix)<<log2g)+(ox))<<log2g)+(oy)] |
23 |
#define dval4(ix,iy,ox,oy) datarr[((((((ix)<<log2g)+(iy))<<log2g)+(ox))<<log2g)+(oy)] |
24 |
|
25 |
/* Tensor tree node */ |
26 |
typedef struct ttree_s { |
27 |
float vmin, vmax; /* value extrema */ |
28 |
float vavg; /* average */ |
29 |
struct ttree_s *kid; /* 2^ttrank children */ |
30 |
} TNODE; |
31 |
|
32 |
#define HISTLEN 300 /* histogram resolution */ |
33 |
#define HISTMAX 10. /* maximum recorded measure in histogram */ |
34 |
|
35 |
int histo[HISTLEN]; /* histogram freq. of variance measure */ |
36 |
|
37 |
double tthresh; /* acceptance threshold (TBD) */ |
38 |
|
39 |
#define var_measure(tp) ( ((tp)->vmax - (tp)->vmin) / \ |
40 |
(sqrt((tp)->vavg) + .03) ) |
41 |
#define above_threshold(tp) (var_measure(tp) > tthresh) |
42 |
|
43 |
/* Allocate a new set of children for the given node (no checks) */ |
44 |
static void |
45 |
new_kids(TNODE *pn) |
46 |
{ |
47 |
pn->kid = (TNODE *)calloc(1<<ttrank, sizeof(TNODE)); |
48 |
if (pn->kid == NULL) |
49 |
error(SYSTEM, "out of memory in new_kids"); |
50 |
} |
51 |
|
52 |
/* Free children for this node */ |
53 |
static void |
54 |
free_kids(TNODE *pn) |
55 |
{ |
56 |
int i; |
57 |
|
58 |
if (pn->kid == NULL) |
59 |
return; |
60 |
for (i = 1<<ttrank; i--; ) |
61 |
free_kids(pn->kid+i); |
62 |
free(pn->kid); |
63 |
pn->kid = NULL; |
64 |
} |
65 |
|
66 |
/* Build a tensor tree starting from the given hypercube */ |
67 |
static void |
68 |
build_tree(TNODE *tp, const int bmin[], int l2s) |
69 |
{ |
70 |
int bkmin[4]; |
71 |
int i, j; |
72 |
|
73 |
tp->vmin = 1e20; |
74 |
tp->vmax = 0; |
75 |
tp->vavg = 0; |
76 |
if (l2s <= 1) { /* reached upper leaves */ |
77 |
for (i = 1<<ttrank; i--; ) { |
78 |
float val; |
79 |
for (j = ttrank; j--; ) |
80 |
bkmin[j] = bmin[j] + (i>>j & 1); |
81 |
val = (ttrank == 3) ? dval3(bkmin[0],bkmin[1],bkmin[2]) |
82 |
: dval4(bkmin[0],bkmin[1],bkmin[2],bkmin[3]); |
83 |
if (val < tp->vmin) |
84 |
tp->vmin = val; |
85 |
if (val > tp->vmax) |
86 |
tp->vmax = val; |
87 |
tp->vavg += val; |
88 |
} |
89 |
tp->vavg /= (float)(1<<ttrank); |
90 |
/* record stats */ |
91 |
i = (HISTLEN/HISTMAX) * var_measure(tp); |
92 |
if (i >= HISTLEN) i = HISTLEN-1; |
93 |
++histo[i]; |
94 |
return; |
95 |
} |
96 |
--l2s; /* else still branching */ |
97 |
new_kids(tp); /* grow recursively */ |
98 |
for (i = 1<<ttrank; i--; ) { |
99 |
for (j = ttrank; j--; ) |
100 |
bkmin[j] = bmin[j] + ((i>>j & 1)<<l2s); |
101 |
build_tree(tp->kid+i, bkmin, l2s); |
102 |
if (tp->kid[i].vmin < tp->vmin) |
103 |
tp->vmin = tp->kid[i].vmin; |
104 |
if (tp->kid[i].vmax > tp->vmax) |
105 |
tp->vmax = tp->kid[i].vmax; |
106 |
tp->vavg += tp->kid[i].vavg; |
107 |
} |
108 |
tp->vavg /= (float)(1<<ttrank); |
109 |
} |
110 |
|
111 |
/* Set our trimming threshold */ |
112 |
static void |
113 |
set_threshold() |
114 |
{ |
115 |
int hsum = 0; |
116 |
int i; |
117 |
|
118 |
for (i = HISTLEN; i--; ) |
119 |
hsum += histo[i]; |
120 |
hsum = pctcull*.01 * (double)hsum; |
121 |
for (i = 0; hsum > 0; i++) |
122 |
hsum -= histo[i]; |
123 |
tthresh = (HISTMAX/HISTLEN) * i; |
124 |
} |
125 |
|
126 |
/* Trim our tree according to the current threshold */ |
127 |
static void |
128 |
trim_tree(TNODE *tp) |
129 |
{ |
130 |
if (tp->kid == NULL) |
131 |
return; |
132 |
if (above_threshold(tp)) { /* keeping branches? */ |
133 |
int i = 1<<ttrank; |
134 |
while (i--) |
135 |
trim_tree(tp->kid+i); |
136 |
return; |
137 |
} |
138 |
free_kids(tp); /* else trim at this point */ |
139 |
} |
140 |
|
141 |
/* Print a tensor tree from the given hypercube */ |
142 |
static void |
143 |
print_tree(const TNODE *tp, const int bmin[], int l2s) |
144 |
{ |
145 |
int bkmin[4]; |
146 |
int i, j; |
147 |
|
148 |
for (j = log2g-l2s; j--; ) /* indent based on branch level */ |
149 |
fputs(" ", stdout); |
150 |
fputc('{', stdout); /* special case for upper leaves */ |
151 |
if (l2s <= 1 && above_threshold(tp)) { |
152 |
for (i = 0; i < 1<<ttrank; i++) { |
153 |
float val; |
154 |
for (j = ttrank; j--; ) |
155 |
bkmin[j] = bmin[j] + (i>>j & 1); |
156 |
val = (ttrank == 3) ? dval3(bkmin[0],bkmin[1],bkmin[2]) |
157 |
: dval4(bkmin[0],bkmin[1],bkmin[2],bkmin[3]); |
158 |
printf(" %.4e", val); |
159 |
} |
160 |
fputs(" }\n", stdout); |
161 |
return; |
162 |
} |
163 |
if (tp->kid == NULL) { /* trimmed limb */ |
164 |
printf(" %.6e }\n", tp->vavg); |
165 |
return; |
166 |
} |
167 |
--l2s; /* else still branching */ |
168 |
fputc('\n', stdout); |
169 |
for (i = 0; i < 1<<ttrank; i++) { |
170 |
for (j = ttrank; j--; ) |
171 |
bkmin[j] = bmin[j] + ((i>>j & 1)<<l2s); |
172 |
print_tree(tp->kid+i, bkmin, l2s); |
173 |
} |
174 |
++l2s; |
175 |
for (j = log2g-l2s; j--; ) |
176 |
fputs(" ", stdout); |
177 |
fputs("}\n", stdout); |
178 |
} |
179 |
|
180 |
/* Read a row of data in ASCII */ |
181 |
static int |
182 |
read_ascii(float *rowp, int n) |
183 |
{ |
184 |
int n2go; |
185 |
|
186 |
if ((rowp == NULL) | (n <= 0)) |
187 |
return(0); |
188 |
for (n2go = n; n2go; n2go--) |
189 |
if (scanf("%f", rowp++) != 1) |
190 |
break; |
191 |
if (n2go) |
192 |
error(USER, "unexpected EOD on ascii input"); |
193 |
return(n-n2go); |
194 |
} |
195 |
|
196 |
/* Read a row of float data */ |
197 |
static int |
198 |
read_float(float *rowp, int n) |
199 |
{ |
200 |
int nread; |
201 |
|
202 |
if ((rowp == NULL) | (n <= 0)) |
203 |
return(0); |
204 |
nread = fread(rowp, sizeof(float), n, stdin); |
205 |
if (nread != n) |
206 |
error(USER, "unexpected EOF on float input"); |
207 |
return(nread); |
208 |
} |
209 |
|
210 |
/* Read a row of double data */ |
211 |
static int |
212 |
read_double(float *rowp, int n) |
213 |
{ |
214 |
static double *rowbuf = NULL; |
215 |
static int rblen = 0; |
216 |
int nread, i; |
217 |
|
218 |
if ((rowp == NULL) | (n <= 0)) { |
219 |
if (rblen) { |
220 |
free(rowbuf); |
221 |
rowbuf = NULL; rblen = 0; |
222 |
} |
223 |
return(0); |
224 |
} |
225 |
if (rblen < n) { |
226 |
rowbuf = (double *)realloc(rowbuf, sizeof(double)*(rblen=n)); |
227 |
if (rowbuf == NULL) |
228 |
error(SYSTEM, "out of memory in read_double"); |
229 |
} |
230 |
nread = fread(rowbuf, sizeof(double), n, stdin); |
231 |
if (nread != n) |
232 |
error(USER, "unexpected EOF on double input"); |
233 |
for (i = 0; i < nread; i++) |
234 |
*rowp++ = rowbuf[i]; |
235 |
return(nread); |
236 |
} |
237 |
|
238 |
/* Load data array, filling zeroes for rank 3 demi-tensor */ |
239 |
static void |
240 |
load_data() |
241 |
{ |
242 |
int (*readf)(float *, int) = NULL; |
243 |
|
244 |
switch (infmt) { |
245 |
case 'a': |
246 |
readf = &read_ascii; |
247 |
break; |
248 |
case 'f': |
249 |
readf = &read_float; |
250 |
break; |
251 |
case 'd': |
252 |
readf = &read_double; |
253 |
break; |
254 |
default: |
255 |
error(COMMAND, "unsupported input format"); |
256 |
break; |
257 |
} |
258 |
datarr = (float *)calloc(1<<(log2g*ttrank), sizeof(float)); |
259 |
if (datarr == NULL) |
260 |
error(SYSTEM, "out of memory in load_data"); |
261 |
if (ttrank == 3) { |
262 |
int ix, ox; |
263 |
for (ix = 0; ix < 1<<(log2g-1); ix++) |
264 |
for (ox = 0; ox < 1<<log2g; ox++) |
265 |
(*readf)(datarr+((((ix)<<log2g)+(ox))<<log2g), |
266 |
1<<log2g); |
267 |
} else /* ttrank == 4 */ { |
268 |
int ix, iy, ox; |
269 |
for (ix = 0; ix < 1<<log2g; ix++) |
270 |
for (iy = 0; iy < 1<<log2g; iy++) |
271 |
for (ox = 0; ox < 1<<log2g; ox++) |
272 |
(*readf)(datarr + |
273 |
((((((ix)<<log2g)+(iy))<<log2g)+(ox))<<log2g), |
274 |
1<<log2g); |
275 |
} |
276 |
(*readf)(NULL, 0); /* releases any buffers */ |
277 |
if (infmt == 'a') { |
278 |
int c; |
279 |
while ((c = getc(stdin)) != EOF) { |
280 |
switch (c) { |
281 |
case ' ': |
282 |
case '\t': |
283 |
case '\r': |
284 |
case '\n': |
285 |
continue; |
286 |
} |
287 |
error(WARNING, "data past end of expected input"); |
288 |
break; |
289 |
} |
290 |
} else if (getc(stdin) != EOF) |
291 |
error(WARNING, "binary data past end of expected input"); |
292 |
} |
293 |
|
294 |
/* Load BSDF array, coalesce uniform regions and format as tensor tree */ |
295 |
int |
296 |
main(int argc, char *argv[]) |
297 |
{ |
298 |
int doheader = 1; |
299 |
int bmin[4]; |
300 |
TNODE gtree; |
301 |
int i; |
302 |
/* get options and parameters */ |
303 |
for (i = 1; i < argc && argv[i][0] == '-'; i++) |
304 |
switch (argv[i][1]) { |
305 |
case 'h': |
306 |
doheader = !doheader; |
307 |
break; |
308 |
case 'r': |
309 |
ttrank = atoi(argv[++i]); |
310 |
if (ttrank != 3 && ttrank != 4) |
311 |
goto userr; |
312 |
break; |
313 |
case 'g': |
314 |
log2g = atoi(argv[++i]); |
315 |
if (log2g <= 1) |
316 |
goto userr; |
317 |
break; |
318 |
case 't': |
319 |
pctcull = atof(argv[++i]); |
320 |
if ((pctcull < 0) | (pctcull >= 100.)) |
321 |
goto userr; |
322 |
break; |
323 |
case 'f': |
324 |
infmt = argv[i][2]; |
325 |
if (!infmt || strchr("afd", infmt) == NULL) |
326 |
goto userr; |
327 |
break; |
328 |
default: |
329 |
goto userr; |
330 |
} |
331 |
if (i < argc-1) |
332 |
goto userr; |
333 |
/* load input data */ |
334 |
if (i == argc-1 && freopen(argv[i], "rb", stdin) == NULL) { |
335 |
sprintf(errmsg, "cannot open input file '%s'", argv[i]); |
336 |
error(SYSTEM, errmsg); |
337 |
} |
338 |
if (infmt != 'a') |
339 |
SET_FILE_BINARY(stdin); |
340 |
load_data(); |
341 |
if (doheader) { |
342 |
for (i = 0; i < argc; i++) { |
343 |
fputs(argv[i], stdout); |
344 |
fputc(i < argc-1 ? ' ' : '\n', stdout); |
345 |
} |
346 |
fputc('\n', stdout); |
347 |
} |
348 |
gtree.kid = NULL; /* create our tree */ |
349 |
bmin[0] = bmin[1] = bmin[2] = bmin[3] = 0; |
350 |
build_tree(>ree, bmin, log2g); |
351 |
/* compute threshold & trim tree */ |
352 |
set_threshold(); |
353 |
trim_tree(>ree); |
354 |
/* format to stdout */ |
355 |
print_tree(>ree, bmin, log2g); |
356 |
/* Clean up isn't necessary for main()... |
357 |
free_kids(>ree); |
358 |
free(datarr); |
359 |
*/ |
360 |
return(0); |
361 |
userr: |
362 |
fprintf(stderr, "Usage: %s [-h][-f{a|f|d}][-r {3|4}][-g log2grid][-t trim%%] [input]\n", |
363 |
argv[0]); |
364 |
return(1); |
365 |
} |