60ee0969d52921a303fbab86104e76571c75ef49
[pdfium.git] / core / src / fxcodec / fx_libopenjpeg / libopenjpeg20 / mct.c
1 /*
2  * The copyright in this software is being made available under the 2-clauses 
3  * BSD License, included below. This software may be subject to other third 
4  * party and contributor rights, including patent rights, and no such rights
5  * are granted under this license.
6  *
7  * Copyright (c) 2002-2014, Universite catholique de Louvain (UCL), Belgium
8  * Copyright (c) 2002-2014, Professor Benoit Macq
9  * Copyright (c) 2001-2003, David Janssens
10  * Copyright (c) 2002-2003, Yannick Verschueren
11  * Copyright (c) 2003-2007, Francois-Olivier Devaux 
12  * Copyright (c) 2003-2014, Antonin Descampe
13  * Copyright (c) 2005, Herve Drolon, FreeImage Team
14  * Copyright (c) 2008, 2011-2012, Centre National d'Etudes Spatiales (CNES), FR 
15  * Copyright (c) 2012, CS Systemes d'Information, France
16  * All rights reserved.
17  *
18  * Redistribution and use in source and binary forms, with or without
19  * modification, are permitted provided that the following conditions
20  * are met:
21  * 1. Redistributions of source code must retain the above copyright
22  *    notice, this list of conditions and the following disclaimer.
23  * 2. Redistributions in binary form must reproduce the above copyright
24  *    notice, this list of conditions and the following disclaimer in the
25  *    documentation and/or other materials provided with the distribution.
26  *
27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS `AS IS'
28  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30  * ARE DISCLAIMED.  IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
31  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
32  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
33  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
34  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
35  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
36  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
37  * POSSIBILITY OF SUCH DAMAGE.
38  */
39
40 #ifdef __SSE__
41 #include <xmmintrin.h>
42 #endif
43
44 #include "opj_includes.h"
45
46 /* <summary> */
47 /* This table contains the norms of the basis function of the reversible MCT. */
48 /* </summary> */
49 static const OPJ_FLOAT64 opj_mct_norms[3] = { 1.732, .8292, .8292 };
50
51 /* <summary> */
52 /* This table contains the norms of the basis function of the irreversible MCT. */
53 /* </summary> */
54 static const OPJ_FLOAT64 opj_mct_norms_real[3] = { 1.732, 1.805, 1.573 };
55
56 const OPJ_FLOAT64 * opj_mct_get_mct_norms ()
57 {
58         return opj_mct_norms;
59 }
60
61 const OPJ_FLOAT64 * opj_mct_get_mct_norms_real ()
62 {
63         return opj_mct_norms_real;
64 }
65
66 /* <summary> */
67 /* Foward reversible MCT. */
68 /* </summary> */
69 void opj_mct_encode(
70                 OPJ_INT32* restrict c0,
71                 OPJ_INT32* restrict c1,
72                 OPJ_INT32* restrict c2,
73                 OPJ_UINT32 n)
74 {
75         OPJ_UINT32 i;
76         for(i = 0; i < n; ++i) {
77                 OPJ_INT32 r = c0[i];
78                 OPJ_INT32 g = c1[i];
79                 OPJ_INT32 b = c2[i];
80                 OPJ_INT32 y = (r + (g * 2) + b) >> 2;
81                 OPJ_INT32 u = b - g;
82                 OPJ_INT32 v = r - g;
83                 c0[i] = y;
84                 c1[i] = u;
85                 c2[i] = v;
86         }
87 }
88
89 /* <summary> */
90 /* Inverse reversible MCT. */
91 /* </summary> */
92 void opj_mct_decode(
93                 OPJ_INT32* restrict c0,
94                 OPJ_INT32* restrict c1, 
95                 OPJ_INT32* restrict c2, 
96                 OPJ_UINT32 n)
97 {
98         OPJ_UINT32 i;
99         for (i = 0; i < n; ++i) {
100                 OPJ_INT32 y = c0[i];
101                 OPJ_INT32 u = c1[i];
102                 OPJ_INT32 v = c2[i];
103                 OPJ_INT32 g = y - ((u + v) >> 2);
104                 OPJ_INT32 r = v + g;
105                 OPJ_INT32 b = u + g;
106                 c0[i] = r;
107                 c1[i] = g;
108                 c2[i] = b;
109         }
110 }
111
112 /* <summary> */
113 /* Get norm of basis function of reversible MCT. */
114 /* </summary> */
115 OPJ_FLOAT64 opj_mct_getnorm(OPJ_UINT32 compno) {
116         return opj_mct_norms[compno];
117 }
118
119 /* <summary> */
120 /* Foward irreversible MCT. */
121 /* </summary> */
122 void opj_mct_encode_real(
123                 OPJ_INT32* restrict c0,
124                 OPJ_INT32* restrict c1,
125                 OPJ_INT32* restrict c2,
126                 OPJ_UINT32 n)
127 {
128         OPJ_UINT32 i;
129         for(i = 0; i < n; ++i) {
130                 OPJ_INT32 r = c0[i];
131                 OPJ_INT32 g = c1[i];
132                 OPJ_INT32 b = c2[i];
133                 OPJ_INT32 y =  opj_int_fix_mul(r, 2449) + opj_int_fix_mul(g, 4809) + opj_int_fix_mul(b, 934);
134                 OPJ_INT32 u = -opj_int_fix_mul(r, 1382) - opj_int_fix_mul(g, 2714) + opj_int_fix_mul(b, 4096);
135                 OPJ_INT32 v =  opj_int_fix_mul(r, 4096) - opj_int_fix_mul(g, 3430) - opj_int_fix_mul(b, 666);
136                 c0[i] = y;
137                 c1[i] = u;
138                 c2[i] = v;
139         }
140 }
141
142 /* <summary> */
143 /* Inverse irreversible MCT. */
144 /* </summary> */
145 void opj_mct_decode_real(
146                 OPJ_FLOAT32* restrict c0,
147                 OPJ_FLOAT32* restrict c1,
148                 OPJ_FLOAT32* restrict c2,
149                 OPJ_UINT32 n)
150 {
151         OPJ_UINT32 i;
152 #ifdef __SSE__
153         __m128 vrv, vgu, vgv, vbu;
154         vrv = _mm_set1_ps(1.402f);
155         vgu = _mm_set1_ps(0.34413f);
156         vgv = _mm_set1_ps(0.71414f);
157         vbu = _mm_set1_ps(1.772f);
158         for (i = 0; i < (n >> 3); ++i) {
159                 __m128 vy, vu, vv;
160                 __m128 vr, vg, vb;
161
162                 vy = _mm_load_ps(c0);
163                 vu = _mm_load_ps(c1);
164                 vv = _mm_load_ps(c2);
165                 vr = _mm_add_ps(vy, _mm_mul_ps(vv, vrv));
166                 vg = _mm_sub_ps(_mm_sub_ps(vy, _mm_mul_ps(vu, vgu)), _mm_mul_ps(vv, vgv));
167                 vb = _mm_add_ps(vy, _mm_mul_ps(vu, vbu));
168                 _mm_store_ps(c0, vr);
169                 _mm_store_ps(c1, vg);
170                 _mm_store_ps(c2, vb);
171                 c0 += 4;
172                 c1 += 4;
173                 c2 += 4;
174
175                 vy = _mm_load_ps(c0);
176                 vu = _mm_load_ps(c1);
177                 vv = _mm_load_ps(c2);
178                 vr = _mm_add_ps(vy, _mm_mul_ps(vv, vrv));
179                 vg = _mm_sub_ps(_mm_sub_ps(vy, _mm_mul_ps(vu, vgu)), _mm_mul_ps(vv, vgv));
180                 vb = _mm_add_ps(vy, _mm_mul_ps(vu, vbu));
181                 _mm_store_ps(c0, vr);
182                 _mm_store_ps(c1, vg);
183                 _mm_store_ps(c2, vb);
184                 c0 += 4;
185                 c1 += 4;
186                 c2 += 4;
187         }
188         n &= 7;
189 #endif
190         for(i = 0; i < n; ++i) {
191                 OPJ_FLOAT32 y = c0[i];
192                 OPJ_FLOAT32 u = c1[i];
193                 OPJ_FLOAT32 v = c2[i];
194                 OPJ_FLOAT32 r = y + (v * 1.402f);
195                 OPJ_FLOAT32 g = y - (u * 0.34413f) - (v * (0.71414f));
196                 OPJ_FLOAT32 b = y + (u * 1.772f);
197                 c0[i] = r;
198                 c1[i] = g;
199                 c2[i] = b;
200         }
201 }
202
203 /* <summary> */
204 /* Get norm of basis function of irreversible MCT. */
205 /* </summary> */
206 OPJ_FLOAT64 opj_mct_getnorm_real(OPJ_UINT32 compno) {
207         return opj_mct_norms_real[compno];
208 }
209
210
211 OPJ_BOOL opj_mct_encode_custom(
212                                            OPJ_BYTE * pCodingdata,
213                                            OPJ_UINT32 n,
214                                            OPJ_BYTE ** pData,
215                                            OPJ_UINT32 pNbComp,
216                                            OPJ_UINT32 isSigned)
217 {
218         OPJ_FLOAT32 * lMct = (OPJ_FLOAT32 *) pCodingdata;
219         OPJ_UINT32 i;
220         OPJ_UINT32 j;
221         OPJ_UINT32 k;
222         OPJ_UINT32 lNbMatCoeff = pNbComp * pNbComp;
223         OPJ_INT32 * lCurrentData = 00;
224         OPJ_INT32 * lCurrentMatrix = 00;
225         OPJ_INT32 ** lData = (OPJ_INT32 **) pData;
226         OPJ_UINT32 lMultiplicator = 1 << 13;
227         OPJ_INT32 * lMctPtr;
228
229     OPJ_ARG_NOT_USED(isSigned);
230
231         lCurrentData = (OPJ_INT32 *) opj_malloc((pNbComp + lNbMatCoeff) * sizeof(OPJ_INT32));
232         if (! lCurrentData) {
233                 return OPJ_FALSE;
234         }
235
236         lCurrentMatrix = lCurrentData + pNbComp;
237
238         for (i =0;i<lNbMatCoeff;++i) {
239                 lCurrentMatrix[i] = (OPJ_INT32) (*(lMct++) * (OPJ_FLOAT32)lMultiplicator);
240         }
241
242         for (i = 0; i < n; ++i)  {
243                 lMctPtr = lCurrentMatrix;
244                 for (j=0;j<pNbComp;++j) {
245                         lCurrentData[j] = (*(lData[j]));
246                 }
247
248                 for (j=0;j<pNbComp;++j) {
249                         *(lData[j]) = 0;
250                         for (k=0;k<pNbComp;++k) {
251                                 *(lData[j]) += opj_int_fix_mul(*lMctPtr, lCurrentData[k]);
252                                 ++lMctPtr;
253                         }
254
255                         ++lData[j];
256                 }
257         }
258
259         opj_free(lCurrentData);
260
261         return OPJ_TRUE;
262 }
263
264 OPJ_BOOL opj_mct_decode_custom(
265                                            OPJ_BYTE * pDecodingData,
266                                            OPJ_UINT32 n,
267                                            OPJ_BYTE ** pData,
268                                            OPJ_UINT32 pNbComp,
269                                            OPJ_UINT32 isSigned)
270 {
271         OPJ_FLOAT32 * lMct;
272         OPJ_UINT32 i;
273         OPJ_UINT32 j;
274         OPJ_UINT32 k;
275
276         OPJ_FLOAT32 * lCurrentData = 00;
277         OPJ_FLOAT32 * lCurrentResult = 00;
278         OPJ_FLOAT32 ** lData = (OPJ_FLOAT32 **) pData;
279
280     OPJ_ARG_NOT_USED(isSigned);
281
282         lCurrentData = (OPJ_FLOAT32 *) opj_malloc (2 * pNbComp * sizeof(OPJ_FLOAT32));
283         if (! lCurrentData) {
284                 return OPJ_FALSE;
285         }
286         lCurrentResult = lCurrentData + pNbComp;
287
288         for (i = 0; i < n; ++i) {
289                 lMct = (OPJ_FLOAT32 *) pDecodingData;
290                 for (j=0;j<pNbComp;++j) {
291                         lCurrentData[j] = (OPJ_FLOAT32) (*(lData[j]));
292                 }
293                 for (j=0;j<pNbComp;++j) {
294                         lCurrentResult[j] = 0;
295                         for     (k=0;k<pNbComp;++k)     {
296                                 lCurrentResult[j] += *(lMct++) * lCurrentData[k];
297                         }
298                         *(lData[j]++) = (OPJ_FLOAT32) (lCurrentResult[j]);
299                 }
300         }
301         opj_free(lCurrentData);
302         return OPJ_TRUE;
303 }
304
305 void opj_calculate_norms(       OPJ_FLOAT64 * pNorms,
306                                                         OPJ_UINT32 pNbComps,
307                                                         OPJ_FLOAT32 * pMatrix)
308 {
309         OPJ_UINT32 i,j,lIndex;
310         OPJ_FLOAT32 lCurrentValue;
311         OPJ_FLOAT64 * lNorms = (OPJ_FLOAT64 *) pNorms;
312         OPJ_FLOAT32 * lMatrix = (OPJ_FLOAT32 *) pMatrix;
313
314         for     (i=0;i<pNbComps;++i) {
315                 lNorms[i] = 0;
316                 lIndex = i;
317
318                 for     (j=0;j<pNbComps;++j) {
319                         lCurrentValue = lMatrix[lIndex];
320                         lIndex += pNbComps;
321                         lNorms[i] += lCurrentValue * lCurrentValue;
322                 }
323                 lNorms[i] = sqrt(lNorms[i]);
324         }
325 }