|
1 /* |
|
2 * Copyright 1998-2003 Sun Microsystems, Inc. All Rights Reserved. |
|
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 |
|
7 * published by the Free Software Foundation. Sun designates this |
|
8 * particular file as subject to the "Classpath" exception as provided |
|
9 * by Sun in the LICENSE file that accompanied this code. |
|
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 * |
|
21 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
|
22 * CA 95054 USA or visit www.sun.com if you need additional information or |
|
23 * have any questions. |
|
24 */ |
|
25 |
|
26 |
|
27 |
|
28 #include "vis_proto.h" |
|
29 #include "mlib_image.h" |
|
30 #include "mlib_v_ImageLookUpFunc.h" |
|
31 |
|
32 /***************************************************************/ |
|
33 static void mlib_v_ImageLookUp_S32_S16_124_D1(const mlib_s32 * src, |
|
34 mlib_s16 * dst, |
|
35 mlib_s32 xsize, |
|
36 const mlib_s16 * table0, |
|
37 const mlib_s16 * table1, |
|
38 const mlib_s16 * table2, |
|
39 const mlib_s16 * table3); |
|
40 |
|
41 static void mlib_v_ImageLookUp_S32_S16_3_D1(const mlib_s32 * src, |
|
42 mlib_s16 * dst, |
|
43 mlib_s32 xsize, |
|
44 const mlib_s16 * table0, |
|
45 const mlib_s16 * table1, |
|
46 const mlib_s16 * table2); |
|
47 |
|
48 /***************************************************************/ |
|
49 |
|
50 #define VIS_LD_U16_I(X, Y) vis_ld_u16_i((void *)(X), (Y)) |
|
51 |
|
52 /***************************************************************/ |
|
53 void mlib_v_ImageLookUp_S32_S16_124_D1(const mlib_s32 * src, |
|
54 mlib_s16 * dst, |
|
55 mlib_s32 xsize, |
|
56 const mlib_s16 * table0, |
|
57 const mlib_s16 * table1, |
|
58 const mlib_s16 * table2, |
|
59 const mlib_s16 * table3) |
|
60 { |
|
61 mlib_s32 *sp; /* pointer to source data */ |
|
62 mlib_s32 s0, s1, s2, s3; /* source data */ |
|
63 mlib_s16 *dl; /* pointer to start of destination */ |
|
64 mlib_s16 *dend; /* pointer to end of destination */ |
|
65 mlib_d64 *dp; /* aligned pointer to destination */ |
|
66 mlib_d64 t0, t1, t2; /* destination data */ |
|
67 mlib_d64 t3, acc0; /* destination data */ |
|
68 mlib_s32 emask; /* edge mask */ |
|
69 mlib_s32 i, num; /* loop variable */ |
|
70 |
|
71 dl = dst; |
|
72 sp = (void *)src; |
|
73 dp = (mlib_d64 *) dl; |
|
74 dend = dl + xsize - 1; |
|
75 |
|
76 vis_alignaddr((void *)0, 6); |
|
77 |
|
78 if (xsize >= 4) { |
|
79 |
|
80 s0 = sp[0]; |
|
81 s1 = sp[1]; |
|
82 s2 = sp[2]; |
|
83 s3 = sp[3]; |
|
84 sp += 4; |
|
85 |
|
86 #pragma pipeloop(0) |
|
87 for (i = 0; i <= xsize - 8; i += 4, sp += 4) { |
|
88 t3 = VIS_LD_U16_I(table3, ((mlib_addr) 2 * s3)); |
|
89 t2 = VIS_LD_U16_I(table2, ((mlib_addr) 2 * s2)); |
|
90 t1 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s1)); |
|
91 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
92 acc0 = vis_faligndata(t3, acc0); |
|
93 acc0 = vis_faligndata(t2, acc0); |
|
94 acc0 = vis_faligndata(t1, acc0); |
|
95 acc0 = vis_faligndata(t0, acc0); |
|
96 s0 = sp[0]; |
|
97 s1 = sp[1]; |
|
98 s2 = sp[2]; |
|
99 s3 = sp[3]; |
|
100 *dp++ = acc0; |
|
101 } |
|
102 |
|
103 t3 = VIS_LD_U16_I(table3, ((mlib_addr) 2 * s3)); |
|
104 t2 = VIS_LD_U16_I(table2, ((mlib_addr) 2 * s2)); |
|
105 t1 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s1)); |
|
106 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
107 acc0 = vis_faligndata(t3, acc0); |
|
108 acc0 = vis_faligndata(t2, acc0); |
|
109 acc0 = vis_faligndata(t1, acc0); |
|
110 acc0 = vis_faligndata(t0, acc0); |
|
111 *dp++ = acc0; |
|
112 } |
|
113 |
|
114 if ((mlib_addr) dp <= (mlib_addr) dend) { |
|
115 |
|
116 num = (mlib_s32) ((mlib_s16 *) dend - (mlib_s16 *) dp); |
|
117 sp += num; |
|
118 num++; |
|
119 |
|
120 if (num == 1) { |
|
121 s0 = *sp; |
|
122 |
|
123 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
124 acc0 = vis_faligndata(t0, acc0); |
|
125 } |
|
126 else if (num == 2) { |
|
127 s0 = *sp; |
|
128 sp--; |
|
129 |
|
130 t0 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s0)); |
|
131 acc0 = vis_faligndata(t0, acc0); |
|
132 |
|
133 s0 = *sp; |
|
134 |
|
135 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
136 acc0 = vis_faligndata(t0, acc0); |
|
137 } |
|
138 else if (num == 3) { |
|
139 s0 = *sp; |
|
140 sp--; |
|
141 |
|
142 t0 = VIS_LD_U16_I(table2, ((mlib_addr) 2 * s0)); |
|
143 acc0 = vis_faligndata(t0, acc0); |
|
144 |
|
145 s0 = *sp; |
|
146 sp--; |
|
147 |
|
148 t0 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s0)); |
|
149 acc0 = vis_faligndata(t0, acc0); |
|
150 |
|
151 s0 = *sp; |
|
152 |
|
153 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
154 acc0 = vis_faligndata(t0, acc0); |
|
155 } |
|
156 |
|
157 emask = vis_edge16(dp, dend); |
|
158 vis_pst_16(acc0, dp, emask); |
|
159 } |
|
160 } |
|
161 |
|
162 /***************************************************************/ |
|
163 void mlib_v_ImageLookUp_S32_S16_1(const mlib_s32 * src, |
|
164 mlib_s32 slb, |
|
165 mlib_s16 * dst, |
|
166 mlib_s32 dlb, |
|
167 mlib_s32 xsize, |
|
168 mlib_s32 ysize, const mlib_s16 ** table) |
|
169 { |
|
170 mlib_s32 *sl; |
|
171 mlib_s16 *dl; |
|
172 const mlib_s16 *tab = &table[0][(mlib_u32) 2147483648u]; |
|
173 mlib_s32 j, i; |
|
174 |
|
175 sl = (void *)src; |
|
176 dl = dst; |
|
177 |
|
178 /* row loop */ |
|
179 for (j = 0; j < ysize; j++) { |
|
180 mlib_s32 *sp = sl; |
|
181 mlib_s16 *dp = dl; |
|
182 mlib_s32 off, size = xsize; |
|
183 |
|
184 off = (mlib_s32) (((8 - ((mlib_addr) dp & 7)) & 7) >> 1); |
|
185 |
|
186 off = (off < size) ? off : size; |
|
187 |
|
188 for (i = 0; i < off; i++, sp++) { |
|
189 *dp++ = tab[sp[0]]; |
|
190 size--; |
|
191 } |
|
192 |
|
193 if (size > 0) { |
|
194 mlib_v_ImageLookUp_S32_S16_124_D1(sp, dp, size, tab, tab, tab, tab); |
|
195 } |
|
196 |
|
197 sl = (mlib_s32 *) ((mlib_u8 *) sl + slb); |
|
198 dl = (mlib_s16 *) ((mlib_u8 *) dl + dlb); |
|
199 } |
|
200 } |
|
201 |
|
202 /***************************************************************/ |
|
203 void mlib_v_ImageLookUp_S32_S16_2(const mlib_s32 * src, |
|
204 mlib_s32 slb, |
|
205 mlib_s16 * dst, |
|
206 mlib_s32 dlb, |
|
207 mlib_s32 xsize, |
|
208 mlib_s32 ysize, const mlib_s16 ** table) |
|
209 { |
|
210 mlib_s32 *sl; |
|
211 mlib_s16 *dl; |
|
212 const mlib_s16 *tab; |
|
213 mlib_s32 j, i; |
|
214 |
|
215 sl = (void *)src; |
|
216 dl = dst; |
|
217 |
|
218 /* row loop */ |
|
219 for (j = 0; j < ysize; j++) { |
|
220 mlib_s32 *sp = sl; |
|
221 mlib_s16 *dp = dl; |
|
222 mlib_s32 off, size = xsize * 2; |
|
223 const mlib_s16 *tab0 = &table[0][(mlib_u32) 2147483648u]; |
|
224 const mlib_s16 *tab1 = &table[1][(mlib_u32) 2147483648u]; |
|
225 |
|
226 off = (mlib_s32) (((8 - ((mlib_addr) dp & 7)) & 7) >> 1); |
|
227 |
|
228 off = (off < size) ? off : size; |
|
229 |
|
230 for (i = 0; i < off - 1; i += 2, sp += 2) { |
|
231 *dp++ = tab0[sp[0]]; |
|
232 *dp++ = tab1[sp[1]]; |
|
233 size -= 2; |
|
234 } |
|
235 |
|
236 if ((off & 1) != 0) { |
|
237 *dp++ = tab0[sp[0]]; |
|
238 size--; |
|
239 sp++; |
|
240 tab = tab0; |
|
241 tab0 = tab1; |
|
242 tab1 = tab; |
|
243 } |
|
244 |
|
245 if (size > 0) { |
|
246 mlib_v_ImageLookUp_S32_S16_124_D1(sp, dp, size, tab0, tab1, tab0, tab1); |
|
247 } |
|
248 |
|
249 sl = (mlib_s32 *) ((mlib_u8 *) sl + slb); |
|
250 dl = (mlib_s16 *) ((mlib_u8 *) dl + dlb); |
|
251 } |
|
252 } |
|
253 |
|
254 /***************************************************************/ |
|
255 void mlib_v_ImageLookUp_S32_S16_4(const mlib_s32 * src, |
|
256 mlib_s32 slb, |
|
257 mlib_s16 * dst, |
|
258 mlib_s32 dlb, |
|
259 mlib_s32 xsize, |
|
260 mlib_s32 ysize, const mlib_s16 ** table) |
|
261 { |
|
262 mlib_s32 *sl; |
|
263 mlib_s16 *dl; |
|
264 const mlib_s16 *tab; |
|
265 mlib_s32 j; |
|
266 |
|
267 sl = (void *)src; |
|
268 dl = dst; |
|
269 |
|
270 /* row loop */ |
|
271 for (j = 0; j < ysize; j++) { |
|
272 mlib_s32 *sp = sl; |
|
273 mlib_s16 *dp = dl; |
|
274 const mlib_s16 *tab0 = &table[0][(mlib_u32) 2147483648u]; |
|
275 const mlib_s16 *tab1 = &table[1][(mlib_u32) 2147483648u]; |
|
276 const mlib_s16 *tab2 = &table[2][(mlib_u32) 2147483648u]; |
|
277 const mlib_s16 *tab3 = &table[3][(mlib_u32) 2147483648u]; |
|
278 mlib_s32 off, size = xsize * 4; |
|
279 |
|
280 off = (mlib_s32) (((8 - ((mlib_addr) dp & 7)) & 7) >> 1); |
|
281 |
|
282 off = (off < size) ? off : size; |
|
283 |
|
284 if (off == 1) { |
|
285 *dp++ = tab0[sp[0]]; |
|
286 tab = tab0; |
|
287 tab0 = tab1; |
|
288 tab1 = tab2; |
|
289 tab2 = tab3; |
|
290 tab3 = tab; |
|
291 size--; |
|
292 sp++; |
|
293 } |
|
294 else if (off == 2) { |
|
295 *dp++ = tab0[sp[0]]; |
|
296 *dp++ = tab1[sp[1]]; |
|
297 tab = tab0; |
|
298 tab0 = tab2; |
|
299 tab2 = tab; |
|
300 tab = tab1; |
|
301 tab1 = tab3; |
|
302 tab3 = tab; |
|
303 size -= 2; |
|
304 sp += 2; |
|
305 } |
|
306 else if (off == 3) { |
|
307 *dp++ = tab0[sp[0]]; |
|
308 *dp++ = tab1[sp[1]]; |
|
309 *dp++ = tab2[sp[2]]; |
|
310 tab = tab3; |
|
311 tab3 = tab2; |
|
312 tab2 = tab1; |
|
313 tab1 = tab0; |
|
314 tab0 = tab; |
|
315 size -= 3; |
|
316 sp += 3; |
|
317 } |
|
318 |
|
319 if (size > 0) { |
|
320 mlib_v_ImageLookUp_S32_S16_124_D1(sp, dp, size, tab0, tab1, tab2, tab3); |
|
321 } |
|
322 |
|
323 sl = (mlib_s32 *) ((mlib_u8 *) sl + slb); |
|
324 dl = (mlib_s16 *) ((mlib_u8 *) dl + dlb); |
|
325 } |
|
326 } |
|
327 |
|
328 /***************************************************************/ |
|
329 void mlib_v_ImageLookUp_S32_S16_3_D1(const mlib_s32 * src, |
|
330 mlib_s16 * dst, |
|
331 mlib_s32 xsize, |
|
332 const mlib_s16 * table0, |
|
333 const mlib_s16 * table1, |
|
334 const mlib_s16 * table2) |
|
335 { |
|
336 mlib_s32 *sp; /* pointer to source data */ |
|
337 mlib_s32 s0, s1, s2, s3; /* source data */ |
|
338 mlib_s16 *dl; /* pointer to start of destination */ |
|
339 mlib_s16 *dend; /* pointer to end of destination */ |
|
340 mlib_d64 *dp; /* aligned pointer to destination */ |
|
341 mlib_d64 t0, t1, t2, t3; /* destination data */ |
|
342 mlib_d64 acc0; /* destination data */ |
|
343 mlib_s32 emask; /* edge mask */ |
|
344 mlib_s32 i, num; /* loop variable */ |
|
345 const mlib_s16 *table; |
|
346 |
|
347 dl = dst; |
|
348 sp = (void *)src; |
|
349 dp = (mlib_d64 *) dl; |
|
350 dend = dl + xsize - 1; |
|
351 |
|
352 vis_alignaddr((void *)0, 6); |
|
353 |
|
354 if (xsize >= 4) { |
|
355 |
|
356 s0 = sp[0]; |
|
357 s1 = sp[1]; |
|
358 s2 = sp[2]; |
|
359 s3 = sp[3]; |
|
360 sp += 4; |
|
361 |
|
362 #pragma pipeloop(0) |
|
363 for (i = 0; i <= xsize - 8; i += 4, sp += 4) { |
|
364 t3 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s3)); |
|
365 t2 = VIS_LD_U16_I(table2, ((mlib_addr) 2 * s2)); |
|
366 t1 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s1)); |
|
367 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
368 acc0 = vis_faligndata(t3, acc0); |
|
369 acc0 = vis_faligndata(t2, acc0); |
|
370 acc0 = vis_faligndata(t1, acc0); |
|
371 acc0 = vis_faligndata(t0, acc0); |
|
372 s0 = sp[0]; |
|
373 s1 = sp[1]; |
|
374 s2 = sp[2]; |
|
375 s3 = sp[3]; |
|
376 *dp++ = acc0; |
|
377 table = table0; |
|
378 table0 = table1; |
|
379 table1 = table2; |
|
380 table2 = table; |
|
381 } |
|
382 |
|
383 t3 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s3)); |
|
384 t2 = VIS_LD_U16_I(table2, ((mlib_addr) 2 * s2)); |
|
385 t1 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s1)); |
|
386 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
387 acc0 = vis_faligndata(t3, acc0); |
|
388 acc0 = vis_faligndata(t2, acc0); |
|
389 acc0 = vis_faligndata(t1, acc0); |
|
390 acc0 = vis_faligndata(t0, acc0); |
|
391 *dp++ = acc0; |
|
392 table = table0; |
|
393 table0 = table1; |
|
394 table1 = table2; |
|
395 table2 = table; |
|
396 } |
|
397 |
|
398 if ((mlib_addr) dp <= (mlib_addr) dend) { |
|
399 |
|
400 num = (mlib_s32) ((mlib_s16 *) dend - (mlib_s16 *) dp); |
|
401 sp += num; |
|
402 num++; |
|
403 |
|
404 if (num == 1) { |
|
405 s0 = *sp; |
|
406 |
|
407 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
408 acc0 = vis_faligndata(t0, acc0); |
|
409 } |
|
410 else if (num == 2) { |
|
411 s0 = *sp; |
|
412 sp--; |
|
413 |
|
414 t0 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s0)); |
|
415 acc0 = vis_faligndata(t0, acc0); |
|
416 |
|
417 s0 = *sp; |
|
418 |
|
419 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
420 acc0 = vis_faligndata(t0, acc0); |
|
421 } |
|
422 else if (num == 3) { |
|
423 s0 = *sp; |
|
424 sp--; |
|
425 |
|
426 t0 = VIS_LD_U16_I(table2, ((mlib_addr) 2 * s0)); |
|
427 acc0 = vis_faligndata(t0, acc0); |
|
428 |
|
429 s0 = *sp; |
|
430 sp--; |
|
431 |
|
432 t0 = VIS_LD_U16_I(table1, ((mlib_addr) 2 * s0)); |
|
433 acc0 = vis_faligndata(t0, acc0); |
|
434 |
|
435 s0 = *sp; |
|
436 |
|
437 t0 = VIS_LD_U16_I(table0, ((mlib_addr) 2 * s0)); |
|
438 acc0 = vis_faligndata(t0, acc0); |
|
439 } |
|
440 |
|
441 emask = vis_edge16(dp, dend); |
|
442 vis_pst_16(acc0, dp, emask); |
|
443 } |
|
444 } |
|
445 |
|
446 /***************************************************************/ |
|
447 void mlib_v_ImageLookUp_S32_S16_3(const mlib_s32 * src, |
|
448 mlib_s32 slb, |
|
449 mlib_s16 * dst, |
|
450 mlib_s32 dlb, |
|
451 mlib_s32 xsize, |
|
452 mlib_s32 ysize, const mlib_s16 ** table) |
|
453 { |
|
454 mlib_s32 *sl; |
|
455 mlib_s16 *dl; |
|
456 const mlib_s16 *tab; |
|
457 mlib_s32 j, i; |
|
458 |
|
459 sl = (void *)src; |
|
460 dl = dst; |
|
461 |
|
462 /* row loop */ |
|
463 for (j = 0; j < ysize; j++) { |
|
464 mlib_s32 *sp = sl; |
|
465 mlib_s16 *dp = dl; |
|
466 const mlib_s16 *tab0 = &table[0][(mlib_u32) 2147483648u]; |
|
467 const mlib_s16 *tab1 = &table[1][(mlib_u32) 2147483648u]; |
|
468 const mlib_s16 *tab2 = &table[2][(mlib_u32) 2147483648u]; |
|
469 mlib_s32 off, size = xsize * 3; |
|
470 |
|
471 off = (mlib_s32) (((8 - ((mlib_addr) dp & 7)) & 7) >> 1); |
|
472 |
|
473 off = (off < size) ? off : size; |
|
474 |
|
475 for (i = 0; i < off - 2; i += 3, sp += 3) { |
|
476 *dp++ = tab0[sp[0]]; |
|
477 *dp++ = tab1[sp[1]]; |
|
478 *dp++ = tab2[sp[2]]; |
|
479 size -= 3; |
|
480 } |
|
481 |
|
482 off -= i; |
|
483 |
|
484 if (off == 1) { |
|
485 *dp++ = tab0[sp[0]]; |
|
486 tab = tab0; |
|
487 tab0 = tab1; |
|
488 tab1 = tab2; |
|
489 tab2 = tab; |
|
490 size--; |
|
491 sp++; |
|
492 } |
|
493 else if (off == 2) { |
|
494 *dp++ = tab0[sp[0]]; |
|
495 *dp++ = tab1[sp[1]]; |
|
496 tab = tab2; |
|
497 tab2 = tab1; |
|
498 tab1 = tab0; |
|
499 tab0 = tab; |
|
500 size -= 2; |
|
501 sp += 2; |
|
502 } |
|
503 |
|
504 if (size > 0) { |
|
505 mlib_v_ImageLookUp_S32_S16_3_D1(sp, dp, size, tab0, tab1, tab2); |
|
506 } |
|
507 |
|
508 sl = (mlib_s32 *) ((mlib_u8 *) sl + slb); |
|
509 dl = (mlib_s16 *) ((mlib_u8 *) dl + dlb); |
|
510 } |
|
511 } |
|
512 |
|
513 /***************************************************************/ |