1 #define __NR_FILTER_GAUSSIAN_CPP__
3 /*
4 * Gaussian blur renderer
5 *
6 * Authors:
7 * Niko Kiirala <niko@kiirala.com>
8 * bulia byak
9 * Jasper van de Gronde <th.v.d.gronde@hccnet.nl>
10 *
11 * Copyright (C) 2006 authors
12 *
13 * Released under GNU GPL, read the file 'COPYING' for more information
14 */
16 #include "config.h" // Needed for HAVE_OPENMP
18 #include <algorithm>
19 #include <cmath>
20 #include <complex>
21 #include <cstdlib>
22 #include <glib.h>
23 #include <limits>
24 #if HAVE_OPENMP
25 #include <omp.h>
26 #endif //HAVE_OPENMP
28 #include "2geom/isnan.h"
30 #include "display/nr-filter-primitive.h"
31 #include "display/nr-filter-gaussian.h"
32 #include "display/nr-filter-types.h"
33 #include "display/nr-filter-units.h"
34 #include "libnr/nr-blit.h"
35 #include "libnr/nr-pixblock.h"
36 #include <2geom/matrix.h>
37 #include "util/fixed_point.h"
38 #include "preferences.h"
40 #ifndef INK_UNUSED
41 #define INK_UNUSED(x) ((void)(x))
42 #endif
44 // IIR filtering method based on:
45 // L.J. van Vliet, I.T. Young, and P.W. Verbeek, Recursive Gaussian Derivative Filters,
46 // in: A.K. Jain, S. Venkatesh, B.C. Lovell (eds.),
47 // ICPR'98, Proc. 14th Int. Conference on Pattern Recognition (Brisbane, Aug. 16-20),
48 // IEEE Computer Society Press, Los Alamitos, 1998, 509-514.
49 //
50 // Using the backwards-pass initialization procedure from:
51 // Boundary Conditions for Young - van Vliet Recursive Filtering
52 // Bill Triggs, Michael Sdika
53 // IEEE Transactions on Signal Processing, Volume 54, Number 5 - may 2006
55 // Number of IIR filter coefficients used. Currently only 3 is supported.
56 // "Recursive Gaussian Derivative Filters" says this is enough though (and
57 // some testing indeed shows that the quality doesn't improve much if larger
58 // filters are used).
59 static size_t const N = 3;
61 template<typename InIt, typename OutIt, typename Size>
62 inline void copy_n(InIt beg_in, Size N, OutIt beg_out) {
63 std::copy(beg_in, beg_in+N, beg_out);
64 }
66 // Type used for IIR filter coefficients (can be 10.21 signed fixed point, see Anisotropic Gaussian Filtering Using Fixed Point Arithmetic, Christoph H. Lampert & Oliver Wirjadi, 2006)
67 typedef double IIRValue;
69 // Type used for FIR filter coefficients (can be 16.16 unsigned fixed point, should have 8 or more bits in the fractional part, the integer part should be capable of storing approximately 20*255)
70 typedef Inkscape::Util::FixedPoint<unsigned int,16> FIRValue;
72 template<typename T> static inline T sqr(T const& v) { return v*v; }
74 template<typename T> static inline T clip(T const& v, T const& a, T const& b) {
75 if ( v < a ) return a;
76 if ( v > b ) return b;
77 return v;
78 }
80 template<typename Tt, typename Ts>
81 static inline Tt round_cast(Ts const& v) {
82 static Ts const rndoffset(.5);
83 return static_cast<Tt>(v+rndoffset);
84 }
86 template<typename Tt, typename Ts>
87 static inline Tt clip_round_cast(Ts const& v, Tt const minval=std::numeric_limits<Tt>::min(), Tt const maxval=std::numeric_limits<Tt>::max()) {
88 if ( v < minval ) return minval;
89 if ( v > maxval ) return maxval;
90 return round_cast<Tt>(v);
91 }
93 namespace NR {
95 FilterGaussian::FilterGaussian()
96 {
97 _deviation_x = _deviation_y = 0.0;
98 }
100 FilterPrimitive *FilterGaussian::create()
101 {
102 return new FilterGaussian();
103 }
105 FilterGaussian::~FilterGaussian()
106 {
107 // Nothing to do here
108 }
110 static int
111 _effect_area_scr(double const deviation)
112 {
113 return (int)std::ceil(deviation * 3.0);
114 }
116 static void
117 _make_kernel(FIRValue *const kernel, double const deviation)
118 {
119 int const scr_len = _effect_area_scr(deviation);
120 double const d_sq = sqr(deviation) * 2;
121 double k[scr_len+1]; // This is only called for small kernel sizes (above approximately 10 coefficients the IIR filter is used)
123 // Compute kernel and sum of coefficients
124 // Note that actually only half the kernel is computed, as it is symmetric
125 double sum = 0;
126 for ( int i = scr_len; i >= 0 ; i-- ) {
127 k[i] = std::exp(-sqr(i) / d_sq);
128 if ( i > 0 ) sum += k[i];
129 }
130 // the sum of the complete kernel is twice as large (plus the center element which we skipped above to prevent counting it twice)
131 sum = 2*sum + k[0];
133 // Normalize kernel (making sure the sum is exactly 1)
134 double ksum = 0;
135 FIRValue kernelsum = 0;
136 for ( int i = scr_len; i >= 1 ; i-- ) {
137 ksum += k[i]/sum;
138 kernel[i] = ksum-static_cast<double>(kernelsum);
139 kernelsum += kernel[i];
140 }
141 kernel[0] = FIRValue(1)-2*kernelsum;
142 }
144 // Return value (v) should satisfy:
145 // 2^(2*v)*255<2^32
146 // 255<2^(32-2*v)
147 // 2^8<=2^(32-2*v)
148 // 8<=32-2*v
149 // 2*v<=24
150 // v<=12
151 static int
152 _effect_subsample_step_log2(double const deviation, int const quality)
153 {
154 // To make sure FIR will always be used (unless the kernel is VERY big):
155 // deviation/step <= 3
156 // deviation/3 <= step
157 // log(deviation/3) <= log(step)
158 // So when x below is >= 1/3 FIR will almost always be used.
159 // This means IIR is almost only used with the modes BETTER or BEST.
160 int stepsize_l2;
161 switch (quality) {
162 case BLUR_QUALITY_WORST:
163 // 2 == log(x*8/3))
164 // 2^2 == x*2^3/3
165 // x == 3/2
166 stepsize_l2 = clip(static_cast<int>(log(deviation*(3./2.))/log(2.)), 0, 12);
167 break;
168 case BLUR_QUALITY_WORSE:
169 // 2 == log(x*16/3))
170 // 2^2 == x*2^4/3
171 // x == 3/2^2
172 stepsize_l2 = clip(static_cast<int>(log(deviation*(3./4.))/log(2.)), 0, 12);
173 break;
174 case BLUR_QUALITY_BETTER:
175 // 2 == log(x*32/3))
176 // 2 == x*2^5/3
177 // x == 3/2^4
178 stepsize_l2 = clip(static_cast<int>(log(deviation*(3./16.))/log(2.)), 0, 12);
179 break;
180 case BLUR_QUALITY_BEST:
181 stepsize_l2 = 0; // no subsampling at all
182 break;
183 case BLUR_QUALITY_NORMAL:
184 default:
185 // 2 == log(x*16/3))
186 // 2 == x*2^4/3
187 // x == 3/2^3
188 stepsize_l2 = clip(static_cast<int>(log(deviation*(3./8.))/log(2.)), 0, 12);
189 break;
190 }
191 return stepsize_l2;
192 }
194 /**
195 * Sanity check function for indexing pixblocks.
196 * Catches reading and writing outside the pixblock area.
197 * When enabled, decreases filter rendering speed massively.
198 */
199 static inline void
200 _check_index(NRPixBlock const * const pb, int const location, int const line)
201 {
202 if (false) {
203 int max_loc = pb->rs * (pb->area.y1 - pb->area.y0);
204 if (location < 0 || location >= max_loc)
205 g_warning("Location %d out of bounds (0 ... %d) at line %d", location, max_loc, line);
206 }
207 }
209 static void calcFilter(double const sigma, double b[N]) {
210 assert(N==3);
211 std::complex<double> const d1_org(1.40098, 1.00236);
212 double const d3_org = 1.85132;
213 double qbeg = 1; // Don't go lower than sigma==2 (we'd probably want a normal convolution in that case anyway)
214 double qend = 2*sigma;
215 double const sigmasqr = sqr(sigma);
216 double s;
217 do { // Binary search for right q (a linear interpolation scheme is suggested, but this should work fine as well)
218 double const q = (qbeg+qend)/2;
219 // Compute scaled filter coefficients
220 std::complex<double> const d1 = pow(d1_org, 1.0/q);
221 double const d3 = pow(d3_org, 1.0/q);
222 // Compute actual sigma^2
223 double const ssqr = 2*(2*(d1/sqr(d1-1.)).real()+d3/sqr(d3-1.));
224 if ( ssqr < sigmasqr ) {
225 qbeg = q;
226 } else {
227 qend = q;
228 }
229 s = sqrt(ssqr);
230 } while(qend-qbeg>(sigma/(1<<30)));
231 // Compute filter coefficients
232 double const q = (qbeg+qend)/2;
233 std::complex<double> const d1 = pow(d1_org, 1.0/q);
234 double const d3 = pow(d3_org, 1.0/q);
235 double const absd1sqr = std::norm(d1); // d1*d2 = d1*conj(d1) = |d1|^2 = std::norm(d1)
236 double const re2d1 = 2*d1.real(); // d1+d2 = d1+conj(d1) = 2*real(d1)
237 double const bscale = 1.0/(absd1sqr*d3);
238 b[2] = -bscale;
239 b[1] = bscale*(d3+re2d1);
240 b[0] = -bscale*(absd1sqr+d3*re2d1);
241 }
243 static void calcTriggsSdikaM(double const b[N], double M[N*N]) {
244 assert(N==3);
245 double a1=b[0], a2=b[1], a3=b[2];
246 double const Mscale = 1.0/((1+a1-a2+a3)*(1-a1-a2-a3)*(1+a2+(a1-a3)*a3));
247 M[0] = 1-a2-a1*a3-sqr(a3);
248 M[1] = (a1+a3)*(a2+a1*a3);
249 M[2] = a3*(a1+a2*a3);
250 M[3] = a1+a2*a3;
251 M[4] = (1-a2)*(a2+a1*a3);
252 M[5] = a3*(1-a2-a1*a3-sqr(a3));
253 M[6] = a1*(a1+a3)+a2*(1-a2);
254 M[7] = a1*(a2-sqr(a3))+a3*(1+a2*(a2-1)-sqr(a3));
255 M[8] = a3*(a1+a2*a3);
256 for(unsigned int i=0; i<9; i++) M[i] *= Mscale;
257 }
259 template<unsigned int SIZE>
260 static void calcTriggsSdikaInitialization(double const M[N*N], IIRValue const uold[N][SIZE], IIRValue const uplus[SIZE], IIRValue const vplus[SIZE], IIRValue const alpha, IIRValue vold[N][SIZE]) {
261 for(unsigned int c=0; c<SIZE; c++) {
262 double uminp[N];
263 for(unsigned int i=0; i<N; i++) uminp[i] = uold[i][c] - uplus[c];
264 for(unsigned int i=0; i<N; i++) {
265 double voldf = 0;
266 for(unsigned int j=0; j<N; j++) {
267 voldf += uminp[j]*M[i*N+j];
268 }
269 // Properly takes care of the scaling coefficient alpha and vplus (which is already appropriately scaled)
270 // This was arrived at by starting from a version of the blur filter that ignored the scaling coefficient
271 // (and scaled the final output by alpha^2) and then gradually reintroducing the scaling coefficient.
272 vold[i][c] = voldf*alpha;
273 vold[i][c] += vplus[c];
274 }
275 }
276 }
278 // Filters over 1st dimension
279 template<typename PT, unsigned int PC, bool PREMULTIPLIED_ALPHA>
280 static void
281 filter2D_IIR(PT *const dest, int const dstr1, int const dstr2,
282 PT const *const src, int const sstr1, int const sstr2,
283 int const n1, int const n2, IIRValue const b[N+1], double const M[N*N],
284 IIRValue *const tmpdata[], int const num_threads)
285 {
286 #if HAVE_OPENMP
287 #pragma omp parallel for num_threads(num_threads)
288 #else
289 INK_UNUSED(num_threads);
290 #endif // HAVE_OPENMP
291 for ( int c2 = 0 ; c2 < n2 ; c2++ ) {
292 #if HAVE_OPENMP
293 unsigned int tid = omp_get_thread_num();
294 #else
295 unsigned int tid = 0;
296 #endif // HAVE_OPENMP
297 // corresponding line in the source and output buffer
298 PT const * srcimg = src + c2*sstr2;
299 PT * dstimg = dest + c2*dstr2 + n1*dstr1;
300 // Border constants
301 IIRValue imin[PC]; copy_n(srcimg + (0)*sstr1, PC, imin);
302 IIRValue iplus[PC]; copy_n(srcimg + (n1-1)*sstr1, PC, iplus);
303 // Forward pass
304 IIRValue u[N+1][PC];
305 for(unsigned int i=0; i<N; i++) copy_n(imin, PC, u[i]);
306 for ( int c1 = 0 ; c1 < n1 ; c1++ ) {
307 for(unsigned int i=N; i>0; i--) copy_n(u[i-1], PC, u[i]);
308 copy_n(srcimg, PC, u[0]);
309 srcimg += sstr1;
310 for(unsigned int c=0; c<PC; c++) u[0][c] *= b[0];
311 for(unsigned int i=1; i<N+1; i++) {
312 for(unsigned int c=0; c<PC; c++) u[0][c] += u[i][c]*b[i];
313 }
314 copy_n(u[0], PC, tmpdata[tid]+c1*PC);
315 }
316 // Backward pass
317 IIRValue v[N+1][PC];
318 calcTriggsSdikaInitialization<PC>(M, u, iplus, iplus, b[0], v);
319 dstimg -= dstr1;
320 if ( PREMULTIPLIED_ALPHA ) {
321 dstimg[PC-1] = clip_round_cast<PT>(v[0][PC-1]);
322 for(unsigned int c=0; c<PC-1; c++) dstimg[c] = clip_round_cast<PT>(v[0][c], std::numeric_limits<PT>::min(), dstimg[PC-1]);
323 } else {
324 for(unsigned int c=0; c<PC; c++) dstimg[c] = clip_round_cast<PT>(v[0][c]);
325 }
326 int c1=n1-1;
327 while(c1-->0) {
328 for(unsigned int i=N; i>0; i--) copy_n(v[i-1], PC, v[i]);
329 copy_n(tmpdata[tid]+c1*PC, PC, v[0]);
330 for(unsigned int c=0; c<PC; c++) v[0][c] *= b[0];
331 for(unsigned int i=1; i<N+1; i++) {
332 for(unsigned int c=0; c<PC; c++) v[0][c] += v[i][c]*b[i];
333 }
334 dstimg -= dstr1;
335 if ( PREMULTIPLIED_ALPHA ) {
336 dstimg[PC-1] = clip_round_cast<PT>(v[0][PC-1]);
337 for(unsigned int c=0; c<PC-1; c++) dstimg[c] = clip_round_cast<PT>(v[0][c], std::numeric_limits<PT>::min(), dstimg[PC-1]);
338 } else {
339 for(unsigned int c=0; c<PC; c++) dstimg[c] = clip_round_cast<PT>(v[0][c]);
340 }
341 }
342 }
343 }
345 // Filters over 1st dimension
346 // Assumes kernel is symmetric
347 // scr_len should be size of kernel - 1
348 template<typename PT, unsigned int PC>
349 static void
350 filter2D_FIR(PT *const dst, int const dstr1, int const dstr2,
351 PT const *const src, int const sstr1, int const sstr2,
352 int const n1, int const n2, FIRValue const *const kernel, int const scr_len, int const num_threads)
353 {
354 // Past pixels seen (to enable in-place operation)
355 PT history[scr_len+1][PC];
357 #if HAVE_OPENMP
358 #pragma omp parallel for num_threads(num_threads) private(history)
359 #else
360 INK_UNUSED(num_threads);
361 #endif // HAVE_OPENMP
362 for ( int c2 = 0 ; c2 < n2 ; c2++ ) {
364 // corresponding line in the source buffer
365 int const src_line = c2 * sstr2;
367 // current line in the output buffer
368 int const dst_line = c2 * dstr2;
370 int skipbuf[4] = {INT_MIN, INT_MIN, INT_MIN, INT_MIN};
372 // history initialization
373 PT imin[PC]; copy_n(src + src_line, PC, imin);
374 for(int i=0; i<scr_len; i++) copy_n(imin, PC, history[i]);
376 for ( int c1 = 0 ; c1 < n1 ; c1++ ) {
378 int const src_disp = src_line + c1 * sstr1;
379 int const dst_disp = dst_line + c1 * sstr1;
381 // update history
382 for(int i=scr_len; i>0; i--) copy_n(history[i-1], PC, history[i]);
383 copy_n(src + src_disp, PC, history[0]);
385 // for all bytes of the pixel
386 for ( unsigned int byte = 0 ; byte < PC ; byte++) {
388 if(skipbuf[byte] > c1) continue;
390 FIRValue sum = 0;
391 int last_in = -1;
392 int different_count = 0;
394 // go over our point's neighbours in the history
395 for ( int i = 0 ; i <= scr_len ; i++ ) {
396 // value at the pixel
397 PT in_byte = history[i][byte];
399 // is it the same as last one we saw?
400 if(in_byte != last_in) different_count++;
401 last_in = in_byte;
403 // sum pixels weighted by the kernel
404 sum += in_byte * kernel[i];
405 }
407 // go over our point's neighborhood on x axis in the in buffer
408 int nb_src_disp = src_disp + byte;
409 for ( int i = 1 ; i <= scr_len ; i++ ) {
410 // the pixel we're looking at
411 int c1_in = c1 + i;
412 if (c1_in >= n1) {
413 c1_in = n1 - 1;
414 } else {
415 nb_src_disp += sstr1;
416 }
418 // value at the pixel
419 PT in_byte = src[nb_src_disp];
421 // is it the same as last one we saw?
422 if(in_byte != last_in) different_count++;
423 last_in = in_byte;
425 // sum pixels weighted by the kernel
426 sum += in_byte * kernel[i];
427 }
429 // store the result in bufx
430 dst[dst_disp + byte] = round_cast<PT>(sum);
432 // optimization: if there was no variation within this point's neighborhood,
433 // skip ahead while we keep seeing the same last_in byte:
434 // blurring flat color would not change it anyway
435 if (different_count <= 1) {
436 int pos = c1 + 1;
437 int nb_src_disp = src_disp + (1+scr_len)*sstr1 + byte; // src_line + (pos+scr_len) * sstr1 + byte
438 int nb_dst_disp = dst_disp + (1) *dstr1 + byte; // dst_line + (pos) * sstr1 + byte
439 while(pos + scr_len < n1 && src[nb_src_disp] == last_in) {
440 dst[nb_dst_disp] = last_in;
441 pos++;
442 nb_src_disp += sstr1;
443 nb_dst_disp += sstr1;
444 }
445 skipbuf[byte] = pos;
446 }
447 }
448 }
449 }
450 }
452 template<typename PT, unsigned int PC>
453 static void
454 downsample(PT *const dst, int const dstr1, int const dstr2, int const dn1, int const dn2,
455 PT const *const src, int const sstr1, int const sstr2, int const sn1, int const sn2,
456 int const step1_l2, int const step2_l2)
457 {
458 unsigned int const divisor_l2 = step1_l2+step2_l2; // step1*step2=2^(step1_l2+step2_l2)
459 unsigned int const round_offset = (1<<divisor_l2)/2;
460 int const step1 = 1<<step1_l2;
461 int const step2 = 1<<step2_l2;
462 int const step1_2 = step1/2;
463 int const step2_2 = step2/2;
464 for(int dc2 = 0 ; dc2 < dn2 ; dc2++) {
465 int const sc2_begin = (dc2<<step2_l2)-step2_2;
466 int const sc2_end = sc2_begin+step2;
467 for(int dc1 = 0 ; dc1 < dn1 ; dc1++) {
468 int const sc1_begin = (dc1<<step1_l2)-step1_2;
469 int const sc1_end = sc1_begin+step1;
470 unsigned int sum[PC];
471 std::fill_n(sum, PC, 0);
472 for(int sc2 = sc2_begin ; sc2 < sc2_end ; sc2++) {
473 for(int sc1 = sc1_begin ; sc1 < sc1_end ; sc1++) {
474 for(unsigned int ch = 0 ; ch < PC ; ch++) {
475 sum[ch] += src[clip(sc2,0,sn2-1)*sstr2+clip(sc1,0,sn1-1)*sstr1+ch];
476 }
477 }
478 }
479 for(unsigned int ch = 0 ; ch < PC ; ch++) {
480 dst[dc2*dstr2+dc1*dstr1+ch] = static_cast<PT>((sum[ch]+round_offset)>>divisor_l2);
481 }
482 }
483 }
484 }
486 template<typename PT, unsigned int PC>
487 static void
488 upsample(PT *const dst, int const dstr1, int const dstr2, unsigned int const dn1, unsigned int const dn2,
489 PT const *const src, int const sstr1, int const sstr2, unsigned int const sn1, unsigned int const sn2,
490 unsigned int const step1_l2, unsigned int const step2_l2)
491 {
492 assert(((sn1-1)<<step1_l2)>=dn1 && ((sn2-1)<<step2_l2)>=dn2); // The last pixel of the source image should fall outside the destination image
493 unsigned int const divisor_l2 = step1_l2+step2_l2; // step1*step2=2^(step1_l2+step2_l2)
494 unsigned int const round_offset = (1<<divisor_l2)/2;
495 unsigned int const step1 = 1<<step1_l2;
496 unsigned int const step2 = 1<<step2_l2;
497 for ( unsigned int sc2 = 0 ; sc2 < sn2-1 ; sc2++ ) {
498 unsigned int const dc2_begin = (sc2 << step2_l2);
499 unsigned int const dc2_end = std::min(dn2, dc2_begin+step2);
500 for ( unsigned int sc1 = 0 ; sc1 < sn1-1 ; sc1++ ) {
501 unsigned int const dc1_begin = (sc1 << step1_l2);
502 unsigned int const dc1_end = std::min(dn1, dc1_begin+step1);
503 for ( unsigned int byte = 0 ; byte < PC ; byte++) {
505 // get 4 values at the corners of the pixel from src
506 PT a00 = src[sstr2* sc2 + sstr1* sc1 + byte];
507 PT a10 = src[sstr2* sc2 + sstr1*(sc1+1) + byte];
508 PT a01 = src[sstr2*(sc2+1) + sstr1* sc1 + byte];
509 PT a11 = src[sstr2*(sc2+1) + sstr1*(sc1+1) + byte];
511 // initialize values for linear interpolation
512 unsigned int a0 = a00*step2/*+a01*0*/;
513 unsigned int a1 = a10*step2/*+a11*0*/;
515 // iterate over the rectangle to be interpolated
516 for ( unsigned int dc2 = dc2_begin ; dc2 < dc2_end ; dc2++ ) {
518 // prepare linear interpolation for this row
519 unsigned int a = a0*step1/*+a1*0*/+round_offset;
521 for ( unsigned int dc1 = dc1_begin ; dc1 < dc1_end ; dc1++ ) {
523 // simple linear interpolation
524 dst[dstr2*dc2 + dstr1*dc1 + byte] = static_cast<PT>(a>>divisor_l2);
526 // compute a = a0*(ix-1)+a1*(xi+1)+round_offset
527 a = a - a0 + a1;
528 }
530 // compute a0 = a00*(iy-1)+a01*(yi+1) and similar for a1
531 a0 = a0 - a00 + a01;
532 a1 = a1 - a10 + a11;
533 }
534 }
535 }
536 }
537 }
539 int FilterGaussian::render(FilterSlot &slot, FilterUnits const &units)
540 {
541 /* in holds the input pixblock */
542 NRPixBlock *in = slot.get(_input);
543 if (!in) {
544 g_warning("Missing source image for feGaussianBlur (in=%d)", _input);
545 return 1;
546 }
548 Geom::Matrix trans = units.get_matrix_primitiveunits2pb();
550 /* If to either direction, the standard deviation is zero or
551 * input image is not defined,
552 * a transparent black image should be returned. */
553 if (_deviation_x <= 0 || _deviation_y <= 0 || in == NULL) {
554 NRPixBlock *out = new NRPixBlock;
555 if (in == NULL) {
556 // A bit guessing here, but source graphic is likely to be of
557 // right size
558 in = slot.get(NR_FILTER_SOURCEGRAPHIC);
559 }
560 nr_pixblock_setup_fast(out, in->mode, in->area.x0, in->area.y0,
561 in->area.x1, in->area.y1, true);
562 if (out->data.px != NULL) {
563 out->empty = false;
564 slot.set(_output, out);
565 }
566 return 0;
567 }
569 // Some common constants
570 Inkscape::Preferences *prefs = Inkscape::Preferences::get();
571 int const width_org = in->area.x1-in->area.x0, height_org = in->area.y1-in->area.y0;
572 double const deviation_x_org = _deviation_x * trans.expansionX();
573 double const deviation_y_org = _deviation_y * trans.expansionY();
574 int const PC = NR_PIXBLOCK_BPP(in);
575 #if HAVE_OPENMP
576 int const NTHREADS = std::max(1,std::min(8,prefs->getInt("/options/threading/numthreads",omp_get_num_procs())));
577 #else
578 int const NTHREADS = 1;
579 #endif // HAVE_OPENMP
581 // Subsampling constants
582 int const quality = prefs->getInt("/options/blurquality/value");
583 int const x_step_l2 = _effect_subsample_step_log2(deviation_x_org, quality);
584 int const y_step_l2 = _effect_subsample_step_log2(deviation_y_org, quality);
585 int const x_step = 1<<x_step_l2;
586 int const y_step = 1<<y_step_l2;
587 bool const resampling = x_step > 1 || y_step > 1;
588 int const width = resampling ? static_cast<int>(ceil(static_cast<double>(width_org)/x_step))+1 : width_org;
589 int const height = resampling ? static_cast<int>(ceil(static_cast<double>(height_org)/y_step))+1 : height_org;
590 double const deviation_x = deviation_x_org / x_step;
591 double const deviation_y = deviation_y_org / y_step;
592 int const scr_len_x = _effect_area_scr(deviation_x);
593 int const scr_len_y = _effect_area_scr(deviation_y);
595 // Decide which filter to use for X and Y
596 // This threshold was determined by trial-and-error for one specific machine,
597 // so there's a good chance that it's not optimal.
598 // Whatever you do, don't go below 1 (and preferrably not even below 2), as
599 // the IIR filter gets unstable there.
600 bool const use_IIR_x = deviation_x > 3;
601 bool const use_IIR_y = deviation_y > 3;
603 // new buffer for the subsampled output
604 NRPixBlock *out = new NRPixBlock;
605 nr_pixblock_setup_fast(out, in->mode, in->area.x0/x_step, in->area.y0/y_step,
606 in->area.x0/x_step+width, in->area.y0/y_step+height, true);
607 if (out->size != NR_PIXBLOCK_SIZE_TINY && out->data.px == NULL) {
608 // alas, we've accomplished a lot, but ran out of memory - so abort
609 return 0;
610 }
611 // Temporary storage for IIR filter
612 // NOTE: This can be eliminated, but it reduces the precision a bit
613 IIRValue * tmpdata[NTHREADS];
614 std::fill_n(tmpdata, NTHREADS, (IIRValue*)0);
615 if ( use_IIR_x || use_IIR_y ) {
616 for(int i=0; i<NTHREADS; i++) {
617 tmpdata[i] = new IIRValue[std::max(width,height)*PC];
618 if (tmpdata[i] == NULL) {
619 nr_pixblock_release(out);
620 while(i-->0) {
621 delete[] tmpdata[i];
622 }
623 delete out;
624 return 0;
625 }
626 }
627 }
628 NRPixBlock *ssin = in;
629 if ( resampling ) {
630 ssin = out;
631 // Downsample
632 switch(in->mode) {
633 case NR_PIXBLOCK_MODE_A8: ///< Grayscale
634 downsample<unsigned char,1>(NR_PIXBLOCK_PX(out), 1, out->rs, width, height, NR_PIXBLOCK_PX(in), 1, in->rs, width_org, height_org, x_step_l2, y_step_l2);
635 break;
636 case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB
637 downsample<unsigned char,3>(NR_PIXBLOCK_PX(out), 3, out->rs, width, height, NR_PIXBLOCK_PX(in), 3, in->rs, width_org, height_org, x_step_l2, y_step_l2);
638 break;
639 case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA
640 downsample<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, width, height, NR_PIXBLOCK_PX(in), 4, in->rs, width_org, height_org, x_step_l2, y_step_l2);
641 break;
642 case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA
643 downsample<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, width, height, NR_PIXBLOCK_PX(in), 4, in->rs, width_org, height_org, x_step_l2, y_step_l2);
644 break;
645 default:
646 assert(false);
647 };
648 }
650 if (use_IIR_x) {
651 // Filter variables
652 IIRValue b[N+1]; // scaling coefficient + filter coefficients (can be 10.21 fixed point)
653 double bf[N]; // computed filter coefficients
654 double M[N*N]; // matrix used for initialization procedure (has to be double)
656 // Compute filter (x)
657 calcFilter(deviation_x, bf);
658 for(size_t i=0; i<N; i++) bf[i] = -bf[i];
659 b[0] = 1; // b[0] == alpha (scaling coefficient)
660 for(size_t i=0; i<N; i++) {
661 b[i+1] = bf[i];
662 b[0] -= b[i+1];
663 }
665 // Compute initialization matrix (x)
666 calcTriggsSdikaM(bf, M);
668 // Filter (x)
669 switch(in->mode) {
670 case NR_PIXBLOCK_MODE_A8: ///< Grayscale
671 filter2D_IIR<unsigned char,1,false>(NR_PIXBLOCK_PX(out), 1, out->rs, NR_PIXBLOCK_PX(ssin), 1, ssin->rs, width, height, b, M, tmpdata, NTHREADS);
672 break;
673 case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB
674 filter2D_IIR<unsigned char,3,false>(NR_PIXBLOCK_PX(out), 3, out->rs, NR_PIXBLOCK_PX(ssin), 3, ssin->rs, width, height, b, M, tmpdata, NTHREADS);
675 break;
676 case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA
677 filter2D_IIR<unsigned char,4,false>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, b, M, tmpdata, NTHREADS);
678 break;
679 case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA
680 filter2D_IIR<unsigned char,4,true >(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, b, M, tmpdata, NTHREADS);
681 break;
682 default:
683 assert(false);
684 };
685 } else if ( scr_len_x > 1 ) { // !use_IIR_x
686 // Filter kernel for x direction
687 FIRValue kernel[scr_len_x];
688 _make_kernel(kernel, deviation_x);
690 // Filter (x)
691 switch(in->mode) {
692 case NR_PIXBLOCK_MODE_A8: ///< Grayscale
693 filter2D_FIR<unsigned char,1>(NR_PIXBLOCK_PX(out), 1, out->rs, NR_PIXBLOCK_PX(ssin), 1, ssin->rs, width, height, kernel, scr_len_x, NTHREADS);
694 break;
695 case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB
696 filter2D_FIR<unsigned char,3>(NR_PIXBLOCK_PX(out), 3, out->rs, NR_PIXBLOCK_PX(ssin), 3, ssin->rs, width, height, kernel, scr_len_x, NTHREADS);
697 break;
698 case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA
699 filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, kernel, scr_len_x, NTHREADS);
700 break;
701 case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA
702 filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), 4, out->rs, NR_PIXBLOCK_PX(ssin), 4, ssin->rs, width, height, kernel, scr_len_x, NTHREADS);
703 break;
704 default:
705 assert(false);
706 };
707 } else if ( out != ssin ) { // out can be equal to ssin if resampling is used
708 nr_blit_pixblock_pixblock(out, ssin);
709 }
711 if (use_IIR_y) {
712 // Filter variables
713 IIRValue b[N+1]; // scaling coefficient + filter coefficients (can be 10.21 fixed point)
714 double bf[N]; // computed filter coefficients
715 double M[N*N]; // matrix used for initialization procedure (has to be double)
717 // Compute filter (y)
718 calcFilter(deviation_y, bf);
719 for(size_t i=0; i<N; i++) bf[i] = -bf[i];
720 b[0] = 1; // b[0] == alpha (scaling coefficient)
721 for(size_t i=0; i<N; i++) {
722 b[i+1] = bf[i];
723 b[0] -= b[i+1];
724 }
726 // Compute initialization matrix (y)
727 calcTriggsSdikaM(bf, M);
729 // Filter (y)
730 switch(in->mode) {
731 case NR_PIXBLOCK_MODE_A8: ///< Grayscale
732 filter2D_IIR<unsigned char,1,false>(NR_PIXBLOCK_PX(out), out->rs, 1, NR_PIXBLOCK_PX(out), out->rs, 1, height, width, b, M, tmpdata, NTHREADS);
733 break;
734 case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB
735 filter2D_IIR<unsigned char,3,false>(NR_PIXBLOCK_PX(out), out->rs, 3, NR_PIXBLOCK_PX(out), out->rs, 3, height, width, b, M, tmpdata, NTHREADS);
736 break;
737 case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA
738 filter2D_IIR<unsigned char,4,false>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, b, M, tmpdata, NTHREADS);
739 break;
740 case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA
741 filter2D_IIR<unsigned char,4,true >(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, b, M, tmpdata, NTHREADS);
742 break;
743 default:
744 assert(false);
745 };
746 } else if ( scr_len_y > 1 ) { // !use_IIR_y
747 // Filter kernel for y direction
748 FIRValue kernel[scr_len_y];
749 _make_kernel(kernel, deviation_y);
751 // Filter (y)
752 switch(in->mode) {
753 case NR_PIXBLOCK_MODE_A8: ///< Grayscale
754 filter2D_FIR<unsigned char,1>(NR_PIXBLOCK_PX(out), out->rs, 1, NR_PIXBLOCK_PX(out), out->rs, 1, height, width, kernel, scr_len_y, NTHREADS);
755 break;
756 case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB
757 filter2D_FIR<unsigned char,3>(NR_PIXBLOCK_PX(out), out->rs, 3, NR_PIXBLOCK_PX(out), out->rs, 3, height, width, kernel, scr_len_y, NTHREADS);
758 break;
759 case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA
760 filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, kernel, scr_len_y, NTHREADS);
761 break;
762 case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA
763 filter2D_FIR<unsigned char,4>(NR_PIXBLOCK_PX(out), out->rs, 4, NR_PIXBLOCK_PX(out), out->rs, 4, height, width, kernel, scr_len_y, NTHREADS);
764 break;
765 default:
766 assert(false);
767 };
768 }
770 for(int i=0; i<NTHREADS; i++) {
771 delete[] tmpdata[i]; // deleting a nullptr has no effect, so this is safe
772 }
774 if ( !resampling ) {
775 // No upsampling needed
776 out->empty = FALSE;
777 slot.set(_output, out);
778 } else {
779 // New buffer for the final output, same resolution as the in buffer
780 NRPixBlock *finalout = new NRPixBlock;
781 nr_pixblock_setup_fast(finalout, in->mode, in->area.x0, in->area.y0,
782 in->area.x1, in->area.y1, true);
783 if (finalout->size != NR_PIXBLOCK_SIZE_TINY && finalout->data.px == NULL) {
784 // alas, we've accomplished a lot, but ran out of memory - so abort
785 nr_pixblock_release(out);
786 delete out;
787 return 0;
788 }
790 // Upsample
791 switch(in->mode) {
792 case NR_PIXBLOCK_MODE_A8: ///< Grayscale
793 upsample<unsigned char,1>(NR_PIXBLOCK_PX(finalout), 1, finalout->rs, width_org, height_org, NR_PIXBLOCK_PX(out), 1, out->rs, width, height, x_step_l2, y_step_l2);
794 break;
795 case NR_PIXBLOCK_MODE_R8G8B8: ///< 8 bit RGB
796 upsample<unsigned char,3>(NR_PIXBLOCK_PX(finalout), 3, finalout->rs, width_org, height_org, NR_PIXBLOCK_PX(out), 3, out->rs, width, height, x_step_l2, y_step_l2);
797 break;
798 case NR_PIXBLOCK_MODE_R8G8B8A8N: ///< Normal 8 bit RGBA
799 upsample<unsigned char,4>(NR_PIXBLOCK_PX(finalout), 4, finalout->rs, width_org, height_org, NR_PIXBLOCK_PX(out), 4, out->rs, width, height, x_step_l2, y_step_l2);
800 break;
801 case NR_PIXBLOCK_MODE_R8G8B8A8P: ///< Premultiplied 8 bit RGBA
802 upsample<unsigned char,4>(NR_PIXBLOCK_PX(finalout), 4, finalout->rs, width_org, height_org, NR_PIXBLOCK_PX(out), 4, out->rs, width, height, x_step_l2, y_step_l2);
803 break;
804 default:
805 assert(false);
806 };
808 // We don't need the out buffer anymore
809 nr_pixblock_release(out);
810 delete out;
812 // The final out buffer gets returned
813 finalout->empty = FALSE;
814 slot.set(_output, finalout);
815 }
817 return 0;
818 }
820 void FilterGaussian::area_enlarge(NRRectL &area, Geom::Matrix const &trans)
821 {
822 int area_x = _effect_area_scr(_deviation_x * trans.expansionX());
823 int area_y = _effect_area_scr(_deviation_y * trans.expansionY());
824 // maximum is used because rotations can mix up these directions
825 // TODO: calculate a more tight-fitting rendering area
826 int area_max = std::max(area_x, area_y);
827 area.x0 -= area_max;
828 area.x1 += area_max;
829 area.y0 -= area_max;
830 area.y1 += area_max;
831 }
833 FilterTraits FilterGaussian::get_input_traits() {
834 return TRAIT_PARALLER;
835 }
837 void FilterGaussian::set_deviation(double deviation)
838 {
839 if(IS_FINITE(deviation) && deviation >= 0) {
840 _deviation_x = _deviation_y = deviation;
841 }
842 }
844 void FilterGaussian::set_deviation(double x, double y)
845 {
846 if(IS_FINITE(x) && x >= 0 && IS_FINITE(y) && y >= 0) {
847 _deviation_x = x;
848 _deviation_y = y;
849 }
850 }
852 } /* namespace NR */
854 /*
855 Local Variables:
856 mode:c++
857 c-file-style:"stroustrup"
858 c-file-offsets:((innamespace . 0)(inline-open . 0)(case-label . +))
859 indent-tabs-mode:nil
860 fill-column:99
861 End:
862 */
863 // vim: filetype=cpp:expandtab:shiftwidth=4:tabstop=8:softtabstop=4:encoding=utf-8:textwidth=99 :