author | lana |
Thu, 26 Dec 2013 12:04:16 -0800 | |
changeset 23010 | 6dadb192ad81 |
parent 8939 | 04615dca2a76 |
permissions | -rw-r--r-- |
2 | 1 |
/* |
23010
6dadb192ad81
8029235: Update copyright year to match last edit in jdk8 jdk repository for 2013
lana
parents:
8939
diff
changeset
|
2 |
* Copyright (c) 2003, 2011, Oracle and/or its affiliates. All rights reserved. |
2 | 3 |
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 |
* |
|
5 |
* This code is free software; you can redistribute it and/or modify it |
|
6 |
* under the terms of the GNU General Public License version 2 only, as |
|
5506 | 7 |
* published by the Free Software Foundation. Oracle designates this |
2 | 8 |
* particular file as subject to the "Classpath" exception as provided |
5506 | 9 |
* by Oracle in the LICENSE file that accompanied this code. |
2 | 10 |
* |
11 |
* This code is distributed in the hope that it will be useful, but WITHOUT |
|
12 |
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
13 |
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
14 |
* version 2 for more details (a copy is included in the LICENSE file that |
|
15 |
* accompanied this code). |
|
16 |
* |
|
17 |
* You should have received a copy of the GNU General Public License version |
|
18 |
* 2 along with this work; if not, write to the Free Software Foundation, |
|
19 |
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
20 |
* |
|
5506 | 21 |
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
22 |
* or visit www.oracle.com if you need additional information or have any |
|
23 |
* questions. |
|
2 | 24 |
*/ |
25 |
||
26 |
||
27 |
/* |
|
28 |
* FUNCTION |
|
29 |
* Internal functions for mlib_ImageConv* on D64/F32 type and |
|
30 |
* MLIB_EDGE_DST_NO_WRITE mask |
|
31 |
* |
|
32 |
*/ |
|
33 |
||
34 |
#include "mlib_image.h" |
|
35 |
#include "mlib_ImageConv.h" |
|
36 |
||
37 |
/***************************************************************/ |
|
38 |
/* |
|
39 |
This define switches between functions of MLIB_DOUBLE and MLIB_FLOAT types: |
|
40 |
Files mlib_ImageConv_D64nw.c and mlib_ImageConv_F32nw.c |
|
41 |
*/ |
|
42 |
||
43 |
#define TYPE_DOUBLE |
|
44 |
||
45 |
/***************************************************************/ |
|
46 |
#ifdef TYPE_DOUBLE |
|
47 |
||
48 |
#define CONV_FUNC(KERN) mlib_conv##KERN##nw_d64 |
|
49 |
||
50 |
#define DTYPE mlib_d64 |
|
51 |
||
52 |
#else |
|
53 |
||
54 |
#define CONV_FUNC(KERN) mlib_conv##KERN##nw_f32 |
|
55 |
||
56 |
#define DTYPE mlib_f32 |
|
57 |
||
58 |
#endif /* TYPE_DOUBLE */ |
|
59 |
||
60 |
/***************************************************************/ |
|
61 |
#define GET_SRC_DST_PARAMETERS(type) \ |
|
62 |
mlib_s32 hgt = mlib_ImageGetHeight(src); \ |
|
63 |
mlib_s32 wid = mlib_ImageGetWidth(src); \ |
|
64 |
mlib_s32 sll = mlib_ImageGetStride(src) / sizeof(type); \ |
|
65 |
mlib_s32 dll = mlib_ImageGetStride(dst) / sizeof(type); \ |
|
66 |
type* adr_src = mlib_ImageGetData(src); \ |
|
67 |
type* adr_dst = mlib_ImageGetData(dst); \ |
|
68 |
mlib_s32 chan1 = mlib_ImageGetChannels(src) |
|
69 |
||
70 |
/***************************************************************/ |
|
71 |
#define DEF_VARS(type) \ |
|
72 |
GET_SRC_DST_PARAMETERS(type); \ |
|
73 |
type *sl; \ |
|
8939 | 74 |
type *dl, *dp = NULL; \ |
75 |
mlib_s32 i = 0, j, c |
|
2 | 76 |
|
77 |
/***************************************************************/ |
|
78 |
#undef KSIZE |
|
79 |
#define KSIZE 2 |
|
80 |
||
81 |
mlib_status CONV_FUNC(2x2)(mlib_image *dst, |
|
82 |
const mlib_image *src, |
|
83 |
const mlib_d64 *kern, |
|
84 |
mlib_s32 cmask) |
|
85 |
{ |
|
86 |
DEF_VARS(DTYPE); |
|
87 |
DTYPE *sp0, *sp1; |
|
88 |
mlib_s32 chan2 = chan1 + chan1; |
|
89 |
mlib_s32 chan3 = chan1 + chan2; |
|
90 |
mlib_s32 chan4 = chan3 + chan1; |
|
91 |
DTYPE k0, k1, k2, k3; |
|
92 |
DTYPE p00, p01, p02, p03, p04, |
|
93 |
p10, p11, p12, p13, p14; |
|
94 |
||
95 |
/* keep kernel in regs */ |
|
96 |
k0 = (DTYPE)kern[0]; k1 = (DTYPE)kern[1]; |
|
97 |
k2 = (DTYPE)kern[2]; k3 = (DTYPE)kern[3]; |
|
98 |
||
99 |
wid -= (KSIZE - 1); |
|
100 |
hgt -= (KSIZE - 1); |
|
101 |
||
102 |
for (c = 0; c < chan1; c++) { |
|
103 |
if (!(cmask & (1 << (chan1 - 1 - c)))) continue; |
|
104 |
||
105 |
dl = adr_dst + c; |
|
106 |
sl = adr_src + c; |
|
107 |
||
108 |
for (j = 0; j < hgt; j++) { |
|
109 |
dp = dl; |
|
110 |
sp0 = sl; |
|
111 |
sp1 = sp0 + sll; |
|
112 |
||
113 |
p04 = sp0[0]; |
|
114 |
p14 = sp1[0]; |
|
115 |
||
116 |
sp0 += chan1; |
|
117 |
sp1 += chan1; |
|
118 |
||
119 |
#ifdef __SUNPRO_C |
|
120 |
#pragma pipeloop(0) |
|
121 |
#endif /* __SUNPRO_C */ |
|
122 |
for (i = 0; i <= (wid - 4); i += 4) { |
|
123 |
p00 = p04; p10 = p14; |
|
124 |
||
125 |
p01 = sp0[0]; p11 = sp1[0]; |
|
126 |
p02 = sp0[chan1]; p12 = sp1[chan1]; |
|
127 |
p03 = sp0[chan2]; p13 = sp1[chan2]; |
|
128 |
p04 = sp0[chan3]; p14 = sp1[chan3]; |
|
129 |
||
130 |
dp[0 ] = p00 * k0 + p01 * k1 + p10 * k2 + p11 * k3; |
|
131 |
dp[chan1] = p01 * k0 + p02 * k1 + p11 * k2 + p12 * k3; |
|
132 |
dp[chan2] = p02 * k0 + p03 * k1 + p12 * k2 + p13 * k3; |
|
133 |
dp[chan3] = p03 * k0 + p04 * k1 + p13 * k2 + p14 * k3; |
|
134 |
||
135 |
dp += chan4; |
|
136 |
sp0 += chan4; |
|
137 |
sp1 += chan4; |
|
138 |
} |
|
139 |
||
140 |
if (i < wid) { |
|
141 |
p00 = p04; p10 = p14; |
|
142 |
p01 = sp0[0]; p11 = sp1[0]; |
|
143 |
dp[0] = p00 * k0 + p01 * k1 + p10 * k2 + p11 * k3; |
|
144 |
||
145 |
if ((i + 1) < wid) { |
|
146 |
p02 = sp0[chan1]; p12 = sp1[chan1]; |
|
147 |
dp[chan1] = p01 * k0 + p02 * k1 + p11 * k2 + p12 * k3; |
|
148 |
||
149 |
if ((i + 2) < wid) { |
|
150 |
p03 = sp0[chan2]; p13 = sp1[chan2]; |
|
151 |
dp[chan2] = p02 * k0 + p03 * k1 + p12 * k2 + p13 * k3; |
|
152 |
} |
|
153 |
} |
|
154 |
} |
|
155 |
||
156 |
sl += sll; |
|
157 |
dl += dll; |
|
158 |
} |
|
159 |
} |
|
160 |
||
161 |
return MLIB_SUCCESS; |
|
162 |
} |
|
163 |
||
164 |
/***************************************************************/ |
|
165 |
#undef KSIZE |
|
166 |
#define KSIZE 3 |
|
167 |
||
168 |
mlib_status CONV_FUNC(3x3)(mlib_image *dst, |
|
169 |
const mlib_image *src, |
|
170 |
const mlib_d64 *kern, |
|
171 |
mlib_s32 cmask) |
|
172 |
{ |
|
173 |
DEF_VARS(DTYPE); |
|
174 |
mlib_s32 chan2 = chan1 + chan1; |
|
175 |
DTYPE *sp0, *sp1; |
|
176 |
DTYPE *sp2; |
|
177 |
DTYPE k0, k1, k2, k3, k4, k5, k6, k7, k8; |
|
178 |
DTYPE p02, p03, p12, p13, p22, p23; |
|
179 |
||
180 |
/* keep kernel in regs */ |
|
181 |
k0 = (DTYPE)kern[0]; k1 = (DTYPE)kern[1]; k2 = (DTYPE)kern[2]; |
|
182 |
k3 = (DTYPE)kern[3]; k4 = (DTYPE)kern[4]; k5 = (DTYPE)kern[5]; |
|
183 |
k6 = (DTYPE)kern[6]; k7 = (DTYPE)kern[7]; k8 = (DTYPE)kern[8]; |
|
184 |
||
185 |
wid -= (KSIZE - 1); |
|
186 |
hgt -= (KSIZE - 1); |
|
187 |
||
188 |
adr_dst += ((KSIZE - 1)/2)*(dll + chan1); |
|
189 |
||
190 |
for (c = 0; c < chan1; c++) { |
|
191 |
if (!(cmask & (1 << (chan1 - 1 - c)))) continue; |
|
192 |
||
193 |
sl = adr_src + c; |
|
194 |
dl = adr_dst + c; |
|
195 |
||
196 |
for (j = 0; j < hgt; j++) { |
|
197 |
DTYPE s0, s1; |
|
198 |
||
199 |
dp = dl; |
|
200 |
sp0 = sl; |
|
201 |
sp1 = sp0 + sll; |
|
202 |
sp2 = sp1 + sll; |
|
203 |
||
204 |
p02 = sp0[0]; |
|
205 |
p12 = sp1[0]; |
|
206 |
p22 = sp2[0]; |
|
207 |
||
208 |
p03 = sp0[chan1]; |
|
209 |
p13 = sp1[chan1]; |
|
210 |
p23 = sp2[chan1]; |
|
211 |
||
212 |
s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7; |
|
213 |
s1 = p03 * k0 + p13 * k3 + p23 * k6; |
|
214 |
||
215 |
sp0 += chan2; |
|
216 |
sp1 += chan2; |
|
217 |
sp2 += chan2; |
|
218 |
||
219 |
#ifdef __SUNPRO_C |
|
220 |
#pragma pipeloop(0) |
|
221 |
#endif /* __SUNPRO_C */ |
|
222 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
223 |
p02 = sp0[0]; p12 = sp1[0]; p22 = sp2[0]; |
|
224 |
p03 = sp0[chan1]; p13 = sp1[chan1]; p23 = sp2[chan1]; |
|
225 |
||
226 |
dp[0 ] = s0 + p02 * k2 + p12 * k5 + p22 * k8; |
|
227 |
dp[chan1] = s1 + p02 * k1 + p03 * k2 + p12 * k4 + p13 * k5 + p22 * k7 + p23 * k8; |
|
228 |
||
229 |
s0 = p02 * k0 + p03 * k1 + p12 * k3 + p13 * k4 + p22 * k6 + p23 * k7; |
|
230 |
s1 = p03 * k0 + p13 * k3 + p23 * k6; |
|
231 |
||
232 |
sp0 += chan2; |
|
233 |
sp1 += chan2; |
|
234 |
sp2 += chan2; |
|
235 |
dp += chan2; |
|
236 |
} |
|
237 |
||
238 |
if (wid & 1) { |
|
239 |
p02 = sp0[0]; p12 = sp1[0]; p22 = sp2[0]; |
|
240 |
dp[0] = s0 + p02 * k2 + p12 * k5 + p22 * k8; |
|
241 |
} |
|
242 |
||
243 |
sl += sll; |
|
244 |
dl += dll; |
|
245 |
} |
|
246 |
} |
|
247 |
||
248 |
return MLIB_SUCCESS; |
|
249 |
} |
|
250 |
||
251 |
/***************************************************************/ |
|
252 |
#undef KSIZE |
|
253 |
#define KSIZE 4 |
|
254 |
||
255 |
mlib_status CONV_FUNC(4x4)(mlib_image *dst, |
|
256 |
const mlib_image *src, |
|
257 |
const mlib_d64 *k, |
|
258 |
mlib_s32 cmask) |
|
259 |
{ |
|
260 |
DTYPE k0, k1, k2, k3, k4, k5, k6, k7; |
|
261 |
DTYPE p00, p01, p02, p03, p04, |
|
262 |
p10, p11, p12, p13, p14; |
|
263 |
DEF_VARS(DTYPE); |
|
264 |
DTYPE *sp0, *sp1; |
|
265 |
mlib_s32 chan2 = chan1 + chan1; |
|
266 |
mlib_s32 chan3 = chan1 + chan2; |
|
267 |
||
268 |
wid -= (KSIZE - 1); |
|
269 |
hgt -= (KSIZE - 1); |
|
270 |
||
271 |
adr_dst += ((KSIZE - 1)/2)*(dll + chan1); |
|
272 |
||
273 |
for (c = 0; c < chan1; c++) { |
|
274 |
if (!(cmask & (1 << (chan1 - 1 - c)))) continue; |
|
275 |
||
276 |
sl = adr_src + c; |
|
277 |
dl = adr_dst + c; |
|
278 |
||
279 |
for (j = 0; j < hgt; j++) { |
|
280 |
/* |
|
281 |
* First loop on two first lines of kernel |
|
282 |
*/ |
|
283 |
sp0 = sl; |
|
284 |
sp1 = sp0 + sll; |
|
285 |
dp = dl; |
|
286 |
||
287 |
k0 = (DTYPE)k[0]; k1 = (DTYPE)k[1]; k2 = (DTYPE)k[2]; k3 = (DTYPE)k[3]; |
|
288 |
k4 = (DTYPE)k[4]; k5 = (DTYPE)k[5]; k6 = (DTYPE)k[6]; k7 = (DTYPE)k[7]; |
|
289 |
||
290 |
p02 = sp0[0]; p12 = sp1[0]; |
|
291 |
p03 = sp0[chan1]; p13 = sp1[chan1]; |
|
292 |
p04 = sp0[chan2]; p14 = sp1[chan2]; |
|
293 |
||
294 |
sp0 += chan3; |
|
295 |
sp1 += chan3; |
|
296 |
||
297 |
#ifdef __SUNPRO_C |
|
298 |
#pragma pipeloop(0) |
|
299 |
#endif /* __SUNPRO_C */ |
|
300 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
301 |
p00 = p02; p10 = p12; |
|
302 |
p01 = p03; p11 = p13; |
|
303 |
p02 = p04; p12 = p14; |
|
304 |
||
305 |
p03 = sp0[0]; p13 = sp1[0]; |
|
306 |
p04 = sp0[chan1]; p14 = sp1[chan1]; |
|
307 |
||
308 |
dp[0 ] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + |
|
309 |
p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7); |
|
310 |
dp[chan1] = (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + |
|
311 |
p11 * k4 + p12 * k5 + p13 * k6 + p14 * k7); |
|
312 |
||
313 |
sp0 += chan2; |
|
314 |
sp1 += chan2; |
|
315 |
dp += chan2; |
|
316 |
} |
|
317 |
||
318 |
if (wid & 1) { |
|
319 |
p00 = p02; p10 = p12; |
|
320 |
p01 = p03; p11 = p13; |
|
321 |
p02 = p04; p12 = p14; |
|
322 |
p03 = sp0[0]; p13 = sp1[0]; |
|
323 |
||
324 |
dp[0] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + |
|
325 |
p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7); |
|
326 |
} |
|
327 |
||
328 |
/* |
|
329 |
* Second loop on two last lines of kernel |
|
330 |
*/ |
|
331 |
sp0 = sl + 2*sll; |
|
332 |
sp1 = sp0 + sll; |
|
333 |
dp = dl; |
|
334 |
||
335 |
k0 = (DTYPE)k[ 8]; k1 = (DTYPE)k[ 9]; k2 = (DTYPE)k[10]; k3 = (DTYPE)k[11]; |
|
336 |
k4 = (DTYPE)k[12]; k5 = (DTYPE)k[13]; k6 = (DTYPE)k[14]; k7 = (DTYPE)k[15]; |
|
337 |
||
338 |
p02 = sp0[0]; p12 = sp1[0]; |
|
339 |
p03 = sp0[chan1]; p13 = sp1[chan1]; |
|
340 |
p04 = sp0[chan2]; p14 = sp1[chan2]; |
|
341 |
||
342 |
sp0 += chan3; |
|
343 |
sp1 += chan3; |
|
344 |
||
345 |
#ifdef __SUNPRO_C |
|
346 |
#pragma pipeloop(0) |
|
347 |
#endif /* __SUNPRO_C */ |
|
348 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
349 |
p00 = p02; p10 = p12; |
|
350 |
p01 = p03; p11 = p13; |
|
351 |
p02 = p04; p12 = p14; |
|
352 |
||
353 |
p03 = sp0[0]; p13 = sp1[0]; |
|
354 |
p04 = sp0[chan1]; p14 = sp1[chan1]; |
|
355 |
||
356 |
dp[0 ] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + |
|
357 |
p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7); |
|
358 |
dp[chan1] += (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + |
|
359 |
p11 * k4 + p12 * k5 + p13 * k6 + p14 * k7); |
|
360 |
||
361 |
sp0 += chan2; |
|
362 |
sp1 += chan2; |
|
363 |
dp += chan2; |
|
364 |
} |
|
365 |
||
366 |
if (wid & 1) { |
|
367 |
p00 = p02; p10 = p12; |
|
368 |
p01 = p03; p11 = p13; |
|
369 |
p02 = p04; p12 = p14; |
|
370 |
p03 = sp0[0]; p13 = sp1[0]; |
|
371 |
||
372 |
dp[0] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + |
|
373 |
p10 * k4 + p11 * k5 + p12 * k6 + p13 * k7); |
|
374 |
} |
|
375 |
||
376 |
/* next line */ |
|
377 |
sl += sll; |
|
378 |
dl += dll; |
|
379 |
} |
|
380 |
} |
|
381 |
||
382 |
return MLIB_SUCCESS; |
|
383 |
} |
|
384 |
||
385 |
/***************************************************************/ |
|
386 |
#undef KSIZE |
|
387 |
#define KSIZE 5 |
|
388 |
||
389 |
mlib_status CONV_FUNC(5x5)(mlib_image *dst, |
|
390 |
const mlib_image *src, |
|
391 |
const mlib_d64 *k, |
|
392 |
mlib_s32 cmask) |
|
393 |
{ |
|
394 |
DTYPE k0, k1, k2, k3, k4, k5, k6, k7, k8, k9; |
|
395 |
DTYPE p00, p01, p02, p03, p04, p05, |
|
396 |
p10, p11, p12, p13, p14, p15; |
|
397 |
DEF_VARS(DTYPE); |
|
398 |
DTYPE *sp0, *sp1; |
|
399 |
mlib_s32 chan2 = chan1 + chan1; |
|
400 |
mlib_s32 chan3 = chan1 + chan2; |
|
401 |
mlib_s32 chan4 = chan3 + chan1; |
|
402 |
||
403 |
wid -= (KSIZE - 1); |
|
404 |
hgt -= (KSIZE - 1); |
|
405 |
||
406 |
adr_dst += ((KSIZE - 1)/2)*(dll + chan1); |
|
407 |
||
408 |
for (c = 0; c < chan1; c++) { |
|
409 |
if (!(cmask & (1 << (chan1 - 1 - c)))) continue; |
|
410 |
||
411 |
sl = adr_src + c; |
|
412 |
dl = adr_dst + c; |
|
413 |
||
414 |
for (j = 0; j < hgt; j++) { |
|
415 |
/* |
|
416 |
* First loop |
|
417 |
*/ |
|
418 |
sp0 = sl; |
|
419 |
sp1 = sp0 + sll; |
|
420 |
dp = dl; |
|
421 |
||
422 |
k0 = (DTYPE)k[0]; k1 = (DTYPE)k[1]; k2 = (DTYPE)k[2]; k3 = (DTYPE)k[3]; k4 = (DTYPE)k[4]; |
|
423 |
k5 = (DTYPE)k[5]; k6 = (DTYPE)k[6]; k7 = (DTYPE)k[7]; k8 = (DTYPE)k[8]; k9 = (DTYPE)k[9]; |
|
424 |
||
425 |
p02 = sp0[0]; p12 = sp1[0]; |
|
426 |
p03 = sp0[chan1]; p13 = sp1[chan1]; |
|
427 |
p04 = sp0[chan2]; p14 = sp1[chan2]; |
|
428 |
p05 = sp0[chan3]; p15 = sp1[chan3]; |
|
429 |
||
430 |
sp0 += chan4; |
|
431 |
sp1 += chan4; |
|
432 |
||
433 |
#ifdef __SUNPRO_C |
|
434 |
#pragma pipeloop(0) |
|
435 |
#endif /* __SUNPRO_C */ |
|
436 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
437 |
p00 = p02; p10 = p12; |
|
438 |
p01 = p03; p11 = p13; |
|
439 |
p02 = p04; p12 = p14; |
|
440 |
p03 = p05; p13 = p15; |
|
441 |
||
442 |
p04 = sp0[0]; p14 = sp1[0]; |
|
443 |
p05 = sp0[chan1]; p15 = sp1[chan1]; |
|
444 |
||
445 |
dp[ 0] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 + |
|
446 |
p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9); |
|
447 |
dp[chan1] = (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 + |
|
448 |
p11 * k5 + p12 * k6 + p13 * k7 + p14 * k8 + p15 * k9); |
|
449 |
||
450 |
sp0 += chan2; |
|
451 |
sp1 += chan2; |
|
452 |
dp += chan2; |
|
453 |
} |
|
454 |
||
455 |
if (wid & 1) { |
|
456 |
p00 = p02; p10 = p12; |
|
457 |
p01 = p03; p11 = p13; |
|
458 |
p02 = p04; p12 = p14; |
|
459 |
p03 = p05; p13 = p15; |
|
460 |
||
461 |
p04 = sp0[0]; p14 = sp1[0]; |
|
462 |
||
463 |
dp[0] = (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 + |
|
464 |
p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9); |
|
465 |
} |
|
466 |
||
467 |
/* |
|
468 |
* Second loop |
|
469 |
*/ |
|
470 |
sp0 = sl + 2*sll; |
|
471 |
sp1 = sp0 + sll; |
|
472 |
dp = dl; |
|
473 |
||
474 |
k0 = (DTYPE)k[10]; k1 = (DTYPE)k[11]; k2 = (DTYPE)k[12]; k3 = (DTYPE)k[13]; k4 = (DTYPE)k[14]; |
|
475 |
k5 = (DTYPE)k[15]; k6 = (DTYPE)k[16]; k7 = (DTYPE)k[17]; k8 = (DTYPE)k[18]; k9 = (DTYPE)k[19]; |
|
476 |
||
477 |
p02 = sp0[0]; p12 = sp1[0]; |
|
478 |
p03 = sp0[chan1]; p13 = sp1[chan1]; |
|
479 |
p04 = sp0[chan2]; p14 = sp1[chan2]; |
|
480 |
p05 = sp0[chan3]; p15 = sp1[chan3]; |
|
481 |
||
482 |
sp0 += chan4; |
|
483 |
sp1 += chan4; |
|
484 |
||
485 |
#ifdef __SUNPRO_C |
|
486 |
#pragma pipeloop(0) |
|
487 |
#endif /* __SUNPRO_C */ |
|
488 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
489 |
p00 = p02; p10 = p12; |
|
490 |
p01 = p03; p11 = p13; |
|
491 |
p02 = p04; p12 = p14; |
|
492 |
p03 = p05; p13 = p15; |
|
493 |
||
494 |
p04 = sp0[0]; p14 = sp1[0]; |
|
495 |
p05 = sp0[chan1]; p15 = sp1[chan1]; |
|
496 |
||
497 |
dp[ 0] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 + |
|
498 |
p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9); |
|
499 |
dp[chan1] += (p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4 + |
|
500 |
p11 * k5 + p12 * k6 + p13 * k7 + p14 * k8 + p15 * k9); |
|
501 |
||
502 |
sp0 += chan2; |
|
503 |
sp1 += chan2; |
|
504 |
dp += chan2; |
|
505 |
} |
|
506 |
||
507 |
if (wid & 1) { |
|
508 |
p00 = p02; p10 = p12; |
|
509 |
p01 = p03; p11 = p13; |
|
510 |
p02 = p04; p12 = p14; |
|
511 |
p03 = p05; p13 = p15; |
|
512 |
||
513 |
p04 = sp0[0]; p14 = sp1[0]; |
|
514 |
||
515 |
dp[0] += (p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4 + |
|
516 |
p10 * k5 + p11 * k6 + p12 * k7 + p13 * k8 + p14 * k9); |
|
517 |
} |
|
518 |
||
519 |
/* |
|
520 |
* 3 loop |
|
521 |
*/ |
|
522 |
dp = dl; |
|
523 |
sp0 = sl + 4*sll; |
|
524 |
||
525 |
k0 = (DTYPE)k[20]; k1 = (DTYPE)k[21]; k2 = (DTYPE)k[22]; k3 = (DTYPE)k[23]; k4 = (DTYPE)k[24]; |
|
526 |
||
527 |
p02 = sp0[0]; |
|
528 |
p03 = sp0[chan1]; |
|
529 |
p04 = sp0[chan2]; |
|
530 |
p05 = sp0[chan3]; |
|
531 |
||
532 |
sp0 += chan2 + chan2; |
|
533 |
||
534 |
#ifdef __SUNPRO_C |
|
535 |
#pragma pipeloop(0) |
|
536 |
#endif /* __SUNPRO_C */ |
|
537 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
538 |
p00 = p02; p01 = p03; p02 = p04; p03 = p05; |
|
539 |
||
540 |
p04 = sp0[0]; p05 = sp0[chan1]; |
|
541 |
||
542 |
dp[0 ] += p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4; |
|
543 |
dp[chan1] += p01 * k0 + p02 * k1 + p03 * k2 + p04 * k3 + p05 * k4; |
|
544 |
||
545 |
dp += chan2; |
|
546 |
sp0 += chan2; |
|
547 |
} |
|
548 |
||
549 |
if (wid & 1) { |
|
550 |
p00 = p02; p01 = p03; p02 = p04; p03 = p05; |
|
551 |
||
552 |
p04 = sp0[0]; |
|
553 |
||
554 |
dp[0] += p00 * k0 + p01 * k1 + p02 * k2 + p03 * k3 + p04 * k4; |
|
555 |
} |
|
556 |
||
557 |
/* next line */ |
|
558 |
sl += sll; |
|
559 |
dl += dll; |
|
560 |
} |
|
561 |
} |
|
562 |
||
563 |
return MLIB_SUCCESS; |
|
564 |
} |
|
565 |
||
566 |
/***************************************************************/ |
|
567 |
#define BUFF_SIZE 1600 |
|
568 |
||
569 |
#define CACHE_SIZE (64*1024) |
|
570 |
||
571 |
static mlib_status mlib_ImageConv1xN(mlib_image *dst, |
|
572 |
const mlib_image *src, |
|
573 |
const DTYPE *k, |
|
574 |
mlib_s32 n, |
|
575 |
mlib_s32 dn, |
|
576 |
mlib_s32 cmask) |
|
577 |
{ |
|
578 |
DTYPE buff[BUFF_SIZE], *pbuff = buff; |
|
579 |
const DTYPE *pk; |
|
580 |
DTYPE k0, k1, k2, k3; |
|
581 |
DTYPE p0, p1, p2, p3, p4; |
|
582 |
DTYPE *sp, *sl_c, *dl_c, *sl0; |
|
583 |
DEF_VARS(DTYPE); |
|
584 |
mlib_s32 off, kh; |
|
585 |
mlib_s32 l, hsize, max_hsize; |
|
586 |
||
587 |
hgt -= (n - 1); |
|
588 |
adr_dst += dn*dll; |
|
589 |
||
590 |
max_hsize = (CACHE_SIZE/sizeof(DTYPE))/sll; |
|
591 |
||
592 |
if (!max_hsize) max_hsize = 1; |
|
593 |
||
594 |
if (max_hsize > BUFF_SIZE) { |
|
595 |
pbuff = mlib_malloc(sizeof(DTYPE)*max_hsize); |
|
596 |
} |
|
597 |
||
598 |
sl_c = adr_src; |
|
599 |
dl_c = adr_dst; |
|
600 |
||
601 |
for (l = 0; l < hgt; l += hsize) { |
|
602 |
hsize = hgt - l; |
|
603 |
||
604 |
if (hsize > max_hsize) hsize = max_hsize; |
|
605 |
||
606 |
for (c = 0; c < chan1; c++) { |
|
607 |
if (!(cmask & (1 << (chan1 - 1 - c)))) continue; |
|
608 |
||
609 |
sl = sl_c + c; |
|
610 |
dl = dl_c + c; |
|
611 |
||
612 |
#ifdef __SUNPRO_C |
|
613 |
#pragma pipeloop(0) |
|
614 |
#endif /* __SUNPRO_C */ |
|
615 |
for (j = 0; j < hsize; j++) pbuff[j] = 0.0; |
|
616 |
||
617 |
for (i = 0; i < wid; i++) { |
|
618 |
sl0 = sl; |
|
619 |
||
620 |
for (off = 0; off < (n - 4); off += 4) { |
|
621 |
pk = k + off; |
|
622 |
sp = sl0; |
|
623 |
||
624 |
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; |
|
625 |
p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll]; |
|
626 |
sp += 3*sll; |
|
627 |
||
628 |
#ifdef __SUNPRO_C |
|
629 |
#pragma pipeloop(0) |
|
630 |
#endif /* __SUNPRO_C */ |
|
631 |
for (j = 0; j < hsize; j += 2) { |
|
632 |
p0 = p2; p1 = p3; p2 = p4; |
|
633 |
p3 = sp[0]; |
|
634 |
p4 = sp[sll]; |
|
635 |
||
636 |
pbuff[j ] += p0*k0 + p1*k1 + p2*k2 + p3*k3; |
|
637 |
pbuff[j + 1] += p1*k0 + p2*k1 + p3*k2 + p4*k3; |
|
638 |
||
639 |
sp += 2*sll; |
|
640 |
} |
|
641 |
||
642 |
sl0 += 4*sll; |
|
643 |
} |
|
644 |
||
645 |
pk = k + off; |
|
646 |
sp = sl0; |
|
647 |
||
648 |
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; |
|
649 |
p2 = sp[0]; p3 = sp[sll]; p4 = sp[2*sll]; |
|
650 |
||
651 |
dp = dl; |
|
652 |
kh = n - off; |
|
653 |
||
654 |
if (kh == 4) { |
|
655 |
sp += 3*sll; |
|
656 |
||
657 |
#ifdef __SUNPRO_C |
|
658 |
#pragma pipeloop(0) |
|
659 |
#endif /* __SUNPRO_C */ |
|
660 |
for (j = 0; j <= (hsize - 2); j += 2) { |
|
661 |
p0 = p2; p1 = p3; p2 = p4; |
|
662 |
p3 = sp[0]; |
|
663 |
p4 = sp[sll]; |
|
664 |
||
665 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]; |
|
666 |
dp[dll] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + pbuff[j + 1]; |
|
667 |
||
668 |
pbuff[j] = 0; |
|
669 |
pbuff[j + 1] = 0; |
|
670 |
||
671 |
sp += 2*sll; |
|
672 |
dp += 2*dll; |
|
673 |
} |
|
674 |
||
675 |
if (j < hsize) { |
|
676 |
p0 = p2; p1 = p3; p2 = p4; |
|
677 |
p3 = sp[0]; |
|
678 |
||
679 |
dp[0] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + pbuff[j]; |
|
680 |
||
681 |
pbuff[j] = 0; |
|
682 |
} |
|
683 |
||
684 |
} else if (kh == 3) { |
|
685 |
sp += 2*sll; |
|
686 |
||
687 |
#ifdef __SUNPRO_C |
|
688 |
#pragma pipeloop(0) |
|
689 |
#endif /* __SUNPRO_C */ |
|
690 |
for (j = 0; j <= (hsize - 2); j += 2) { |
|
691 |
p0 = p2; p1 = p3; |
|
692 |
p2 = sp[0]; |
|
693 |
p3 = sp[sll]; |
|
694 |
||
695 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + pbuff[j]; |
|
696 |
dp[dll] = p1*k0 + p2*k1 + p3*k2 + pbuff[j + 1]; |
|
697 |
||
698 |
pbuff[j] = 0; |
|
699 |
pbuff[j + 1] = 0; |
|
700 |
||
701 |
sp += 2*sll; |
|
702 |
dp += 2*dll; |
|
703 |
} |
|
704 |
||
705 |
if (j < hsize) { |
|
706 |
p0 = p2; p1 = p3; |
|
707 |
p2 = sp[0]; |
|
708 |
||
709 |
dp[0] = p0*k0 + p1*k1 + p2*k2 + pbuff[j]; |
|
710 |
||
711 |
pbuff[j] = 0; |
|
712 |
} |
|
713 |
||
714 |
} else if (kh == 2) { |
|
715 |
sp += sll; |
|
716 |
||
717 |
#ifdef __SUNPRO_C |
|
718 |
#pragma pipeloop(0) |
|
719 |
#endif /* __SUNPRO_C */ |
|
720 |
for (j = 0; j <= (hsize - 2); j += 2) { |
|
721 |
p0 = p2; |
|
722 |
p1 = sp[0]; |
|
723 |
p2 = sp[sll]; |
|
724 |
||
725 |
dp[0 ] = p0*k0 + p1*k1 + pbuff[j]; |
|
726 |
dp[dll] = p1*k0 + p2*k1 + pbuff[j + 1]; |
|
727 |
||
728 |
pbuff[j] = 0; |
|
729 |
pbuff[j + 1] = 0; |
|
730 |
||
731 |
sp += 2*sll; |
|
732 |
dp += 2*dll; |
|
733 |
} |
|
734 |
||
735 |
if (j < hsize) { |
|
736 |
p0 = p2; |
|
737 |
p1 = sp[0]; |
|
738 |
||
739 |
dp[0] = p0*k0 + p1*k1 + pbuff[j]; |
|
740 |
||
741 |
pbuff[j] = 0; |
|
742 |
} |
|
743 |
||
744 |
} else /* if (kh == 1) */ { |
|
745 |
#ifdef __SUNPRO_C |
|
746 |
#pragma pipeloop(0) |
|
747 |
#endif /* __SUNPRO_C */ |
|
748 |
for (j = 0; j < hsize; j++) { |
|
749 |
p0 = sp[0]; |
|
750 |
||
751 |
dp[0] = p0*k0 + pbuff[j]; |
|
752 |
||
753 |
pbuff[j] = 0; |
|
754 |
||
755 |
sp += sll; |
|
756 |
dp += dll; |
|
757 |
} |
|
758 |
} |
|
759 |
||
760 |
sl += chan1; |
|
761 |
dl += chan1; |
|
762 |
} |
|
763 |
} |
|
764 |
||
765 |
sl_c += max_hsize*sll; |
|
766 |
dl_c += max_hsize*dll; |
|
767 |
} |
|
768 |
||
769 |
if (pbuff != buff) mlib_free(pbuff); |
|
770 |
||
771 |
return MLIB_SUCCESS; |
|
772 |
} |
|
773 |
||
774 |
/***************************************************************/ |
|
775 |
#define MAX_KER 7 |
|
776 |
#define MAX_NM 81 |
|
777 |
||
778 |
mlib_status CONV_FUNC(MxN)(mlib_image *dst, |
|
779 |
const mlib_image *src, |
|
780 |
const mlib_d64 *ker, |
|
781 |
mlib_s32 m, |
|
782 |
mlib_s32 n, |
|
783 |
mlib_s32 dm, |
|
784 |
mlib_s32 dn, |
|
785 |
mlib_s32 cmask) |
|
786 |
{ |
|
787 |
DTYPE k0, k1, k2, k3, k4, k5, k6, *sp; |
|
788 |
DTYPE p0, p1, p2, p3, p4, p5, p6, p7; |
|
789 |
mlib_s32 l, off, kw; |
|
790 |
DEF_VARS(DTYPE); |
|
791 |
mlib_s32 chan2 = chan1 + chan1; |
|
792 |
mlib_s32 chan3 = chan1 + chan2; |
|
793 |
||
794 |
#ifdef TYPE_DOUBLE |
|
795 |
const mlib_d64 *k = ker; |
|
796 |
#else |
|
797 |
mlib_f32 k_arr[MAX_NM], *k = k_arr; |
|
798 |
||
799 |
if (n*m > MAX_NM) { |
|
800 |
k = mlib_malloc(n*m*sizeof(mlib_f32)); |
|
801 |
||
802 |
if (k == NULL) return MLIB_FAILURE; |
|
803 |
} |
|
804 |
||
805 |
for (i = 0; i < n*m; i++) k[i] = (mlib_f32)ker[i]; |
|
806 |
#endif /* TYPE_DOUBLE */ |
|
807 |
||
808 |
if (m == 1) return mlib_ImageConv1xN(dst, src, k, n, dn, cmask); |
|
809 |
||
810 |
wid -= (m - 1); |
|
811 |
hgt -= (n - 1); |
|
812 |
adr_dst += dn*dll + dm*chan1; |
|
813 |
||
814 |
for (c = 0; c < chan1; c++) { |
|
815 |
if (!(cmask & (1 << (chan1 - 1 - c)))) continue; |
|
816 |
||
817 |
sl = adr_src + c; |
|
818 |
dl = adr_dst + c; |
|
819 |
||
820 |
for (j = 0; j < hgt; j++) { |
|
821 |
const DTYPE *pk = k; |
|
822 |
||
823 |
for (l = 0; l < n; l++) { |
|
824 |
DTYPE *sp0 = sl + l*sll; |
|
825 |
||
826 |
for (off = 0; off < m; off += kw, pk += kw, sp0 += chan1) { |
|
827 |
kw = m - off; |
|
828 |
||
829 |
if (kw > 2*MAX_KER) kw = MAX_KER; else |
|
830 |
if (kw > MAX_KER) kw = kw/2; |
|
831 |
||
832 |
p2 = sp0[0]; p3 = sp0[chan1]; p4 = sp0[chan2]; |
|
833 |
sp0 += chan3; |
|
834 |
p5 = sp0[0]; p6 = sp0[chan1]; p7 = sp0[chan2]; |
|
835 |
||
836 |
k0 = pk[0]; k1 = pk[1]; k2 = pk[2]; k3 = pk[3]; |
|
837 |
k4 = pk[4]; k5 = pk[5]; k6 = pk[6]; |
|
838 |
||
839 |
dp = dl; |
|
840 |
||
841 |
if (kw == 7) { |
|
842 |
sp = sp0 += chan3; |
|
843 |
||
844 |
if (pk == k) { |
|
845 |
#ifdef __SUNPRO_C |
|
846 |
#pragma pipeloop(0) |
|
847 |
#endif /* __SUNPRO_C */ |
|
848 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
849 |
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; |
|
850 |
||
851 |
p5 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1]; |
|
852 |
||
853 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6; |
|
854 |
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6; |
|
855 |
||
856 |
sp += chan2; |
|
857 |
dp += chan2; |
|
858 |
} |
|
859 |
||
860 |
} else { |
|
861 |
#ifdef __SUNPRO_C |
|
862 |
#pragma pipeloop(0) |
|
863 |
#endif /* __SUNPRO_C */ |
|
864 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
865 |
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; |
|
866 |
||
867 |
p5 = sp[- chan1]; p6 = sp[0]; p7 = sp[chan1]; |
|
868 |
||
869 |
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5 + p6*k6; |
|
870 |
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5 + p7*k6; |
|
871 |
||
872 |
sp += chan2; |
|
873 |
dp += chan2; |
|
874 |
} |
|
875 |
} |
|
876 |
||
877 |
} else if (kw == 6) { |
|
878 |
sp = sp0 += chan2; |
|
879 |
||
880 |
if (pk == k) { |
|
881 |
#ifdef __SUNPRO_C |
|
882 |
#pragma pipeloop(0) |
|
883 |
#endif /* __SUNPRO_C */ |
|
884 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
885 |
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; |
|
886 |
||
887 |
p5 = sp[0]; p6 = sp[chan1]; |
|
888 |
||
889 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5; |
|
890 |
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5; |
|
891 |
||
892 |
sp += chan2; |
|
893 |
dp += chan2; |
|
894 |
} |
|
895 |
||
896 |
} else { |
|
897 |
#ifdef __SUNPRO_C |
|
898 |
#pragma pipeloop(0) |
|
899 |
#endif /* __SUNPRO_C */ |
|
900 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
901 |
p0 = p2; p1 = p3; p2 = p4; p3 = p5; p4 = p6; |
|
902 |
||
903 |
p5 = sp[0]; p6 = sp[chan1]; |
|
904 |
||
905 |
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4 + p5*k5; |
|
906 |
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4 + p6*k5; |
|
907 |
||
908 |
sp += chan2; |
|
909 |
dp += chan2; |
|
910 |
} |
|
911 |
} |
|
912 |
||
913 |
} else if (kw == 5) { |
|
914 |
sp = sp0 += chan1; |
|
915 |
||
916 |
if (pk == k) { |
|
917 |
#ifdef __SUNPRO_C |
|
918 |
#pragma pipeloop(0) |
|
919 |
#endif /* __SUNPRO_C */ |
|
920 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
921 |
p0 = p2; p1 = p3; p2 = p4; p3 = p5; |
|
922 |
||
923 |
p4 = sp[0]; p5 = sp[chan1]; |
|
924 |
||
925 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4; |
|
926 |
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4; |
|
927 |
||
928 |
sp += chan2; |
|
929 |
dp += chan2; |
|
930 |
} |
|
931 |
||
932 |
} else { |
|
933 |
#ifdef __SUNPRO_C |
|
934 |
#pragma pipeloop(0) |
|
935 |
#endif /* __SUNPRO_C */ |
|
936 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
937 |
p0 = p2; p1 = p3; p2 = p4; p3 = p5; |
|
938 |
||
939 |
p4 = sp[0]; p5 = sp[chan1]; |
|
940 |
||
941 |
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3 + p4*k4; |
|
942 |
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3 + p5*k4; |
|
943 |
||
944 |
sp += chan2; |
|
945 |
dp += chan2; |
|
946 |
} |
|
947 |
} |
|
948 |
||
949 |
} else if (kw == 4) { |
|
950 |
||
951 |
sp = sp0; |
|
952 |
||
953 |
if (pk == k) { |
|
954 |
#ifdef __SUNPRO_C |
|
955 |
#pragma pipeloop(0) |
|
956 |
#endif /* __SUNPRO_C */ |
|
957 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
958 |
p0 = p2; p1 = p3; p2 = p4; |
|
959 |
||
960 |
p3 = sp[0]; p4 = sp[chan1]; |
|
961 |
||
962 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2 + p3*k3; |
|
963 |
dp[chan1] = p1*k0 + p2*k1 + p3*k2 + p4*k3; |
|
964 |
||
965 |
sp += chan2; |
|
966 |
dp += chan2; |
|
967 |
} |
|
968 |
||
969 |
} else { |
|
970 |
#ifdef __SUNPRO_C |
|
971 |
#pragma pipeloop(0) |
|
972 |
#endif /* __SUNPRO_C */ |
|
973 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
974 |
p0 = p2; p1 = p3; p2 = p4; |
|
975 |
||
976 |
p3 = sp[0]; p4 = sp[chan1]; |
|
977 |
||
978 |
dp[0 ] += p0*k0 + p1*k1 + p2*k2 + p3*k3; |
|
979 |
dp[chan1] += p1*k0 + p2*k1 + p3*k2 + p4*k3; |
|
980 |
||
981 |
sp += chan2; |
|
982 |
dp += chan2; |
|
983 |
} |
|
984 |
} |
|
985 |
||
986 |
} else if (kw == 3) { |
|
987 |
sp = sp0 -= chan1; |
|
988 |
||
989 |
if (pk == k) { |
|
990 |
#ifdef __SUNPRO_C |
|
991 |
#pragma pipeloop(0) |
|
992 |
#endif /* __SUNPRO_C */ |
|
993 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
994 |
p0 = p2; p1 = p3; |
|
995 |
||
996 |
p2 = sp[0]; p3 = sp[chan1]; |
|
997 |
||
998 |
dp[0 ] = p0*k0 + p1*k1 + p2*k2; |
|
999 |
dp[chan1] = p1*k0 + p2*k1 + p3*k2; |
|
1000 |
||
1001 |
sp += chan2; |
|
1002 |
dp += chan2; |
|
1003 |
} |
|
1004 |
||
1005 |
} else { |
|
1006 |
#ifdef __SUNPRO_C |
|
1007 |
#pragma pipeloop(0) |
|
1008 |
#endif /* __SUNPRO_C */ |
|
1009 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
1010 |
p0 = p2; p1 = p3; |
|
1011 |
||
1012 |
p2 = sp[0]; p3 = sp[chan1]; |
|
1013 |
||
1014 |
dp[0 ] += p0*k0 + p1*k1 + p2*k2; |
|
1015 |
dp[chan1] += p1*k0 + p2*k1 + p3*k2; |
|
1016 |
||
1017 |
sp += chan2; |
|
1018 |
dp += chan2; |
|
1019 |
} |
|
1020 |
} |
|
1021 |
||
1022 |
} else { /* kw == 2 */ |
|
1023 |
sp = sp0 -= chan2; |
|
1024 |
||
1025 |
if (pk == k) { |
|
1026 |
#ifdef __SUNPRO_C |
|
1027 |
#pragma pipeloop(0) |
|
1028 |
#endif /* __SUNPRO_C */ |
|
1029 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
1030 |
p0 = p2; |
|
1031 |
||
1032 |
p1 = sp[0]; p2 = sp[chan1]; |
|
1033 |
||
1034 |
dp[0 ] = p0*k0 + p1*k1; |
|
1035 |
dp[chan1] = p1*k0 + p2*k1; |
|
1036 |
||
1037 |
sp += chan2; |
|
1038 |
dp += chan2; |
|
1039 |
} |
|
1040 |
||
1041 |
} else { |
|
1042 |
#ifdef __SUNPRO_C |
|
1043 |
#pragma pipeloop(0) |
|
1044 |
#endif /* __SUNPRO_C */ |
|
1045 |
for (i = 0; i <= (wid - 2); i += 2) { |
|
1046 |
p0 = p2; |
|
1047 |
||
1048 |
p1 = sp[0]; p2 = sp[chan1]; |
|
1049 |
||
1050 |
dp[0 ] += p0*k0 + p1*k1; |
|
1051 |
dp[chan1] += p1*k0 + p2*k1; |
|
1052 |
||
1053 |
sp += chan2; |
|
1054 |
dp += chan2; |
|
1055 |
} |
|
1056 |
} |
|
1057 |
} |
|
1058 |
} |
|
1059 |
} |
|
1060 |
||
1061 |
/* last pixels */ |
|
1062 |
||
1063 |
if (wid & 1) { |
|
1064 |
DTYPE *sp0 = sl + i*chan1, s = 0; |
|
1065 |
const DTYPE *pk = k; |
|
1066 |
mlib_s32 x; |
|
1067 |
||
1068 |
for (l = 0; l < n; l++) { |
|
1069 |
DTYPE *sp = sp0 + l*sll; |
|
1070 |
||
1071 |
for (x = 0; x < m; x++) s += sp[x*chan1] * (*pk++); |
|
1072 |
} |
|
1073 |
||
1074 |
dp[0] = s; |
|
1075 |
} |
|
1076 |
||
1077 |
/* next line */ |
|
1078 |
sl += sll; |
|
1079 |
dl += dll; |
|
1080 |
} |
|
1081 |
} |
|
1082 |
||
1083 |
#ifndef TYPE_DOUBLE |
|
1084 |
||
1085 |
if (k != k_arr) mlib_free(k); |
|
1086 |
#endif /* TYPE_DOUBLE */ |
|
1087 |
||
1088 |
return MLIB_SUCCESS; |
|
1089 |
} |
|
1090 |
||
1091 |
/***************************************************************/ |