sh4: hpeldsp: Move half-pel assembly from dsputil to hpeldsp
[ffmpeg.git] / libavcodec / sh4 / dsputil_align.c
1 /*
2  * aligned/packed access motion
3  *
4  * Copyright (c) 2001-2003 BERO <bero@geocities.co.jp>
5  *
6  * This file is part of Libav.
7  *
8  * Libav is free software; you can redistribute it and/or
9  * modify it under the terms of the GNU Lesser General Public
10  * License as published by the Free Software Foundation; either
11  * version 2.1 of the License, or (at your option) any later version.
12  *
13  * Libav is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * Lesser General Public License for more details.
17  *
18  * You should have received a copy of the GNU Lesser General Public
19  * License along with Libav; if not, write to the Free Software
20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21  */
22
23 #include "libavutil/attributes.h"
24 #include "libavcodec/avcodec.h"
25 #include "libavcodec/dsputil.h"
26 #include "libavcodec/rnd_avg.h"
27 #include "dsputil_sh4.h"
28
29
30 #define         LP(p)           *(uint32_t*)(p)
31 #define         LPC(p)          *(const uint32_t*)(p)
32
33
34 #define         UNPACK(ph,pl,tt0,tt1) do { \
35         uint32_t t0,t1; t0=tt0;t1=tt1; \
36         ph = ( (t0 & ~BYTE_VEC32(0x03))>>2) + ( (t1 & ~BYTE_VEC32(0x03))>>2); \
37         pl = (t0 & BYTE_VEC32(0x03)) + (t1 & BYTE_VEC32(0x03)); } while(0)
38
39 #define         rnd_PACK(ph,pl,nph,npl) ph + nph + (((pl + npl + BYTE_VEC32(0x02))>>2) & BYTE_VEC32(0x03))
40 #define         no_rnd_PACK(ph,pl,nph,npl)      ph + nph + (((pl + npl + BYTE_VEC32(0x01))>>2) & BYTE_VEC32(0x03))
41
42 /* little-endian */
43 #define         MERGE1(a,b,ofs) (ofs==0)?a:( ((a)>>(8*ofs))|((b)<<(32-8*ofs)) )
44 #define         MERGE2(a,b,ofs) (ofs==3)?b:( ((a)>>(8*(ofs+1)))|((b)<<(32-8*(ofs+1))) )
45 /* big
46 #define         MERGE1(a,b,ofs) (ofs==0)?a:( ((a)<<(8*ofs))|((b)>>(32-8*ofs)) )
47 #define         MERGE2(a,b,ofs) (ofs==3)?b:( ((a)<<(8+8*ofs))|((b)>>(32-8-8*ofs)) )
48 */
49
50
51 #define         put(d,s)        d = s
52 #define         avg(d,s)        d = rnd_avg32(s,d)
53
54 #define         OP_C4(ofs) \
55         ref-=ofs; \
56         do { \
57                 OP(LP(dest),MERGE1(LPC(ref),LPC(ref+4),ofs)); \
58                 ref+=stride; \
59                 dest+=stride; \
60         } while(--height)
61
62 #define        OP_C40() \
63         do { \
64                 OP(LP(dest),LPC(ref)); \
65                 ref+=stride; \
66                 dest+=stride; \
67         } while(--height)
68
69 #define         OP_C(ofs,sz,avg2) \
70 { \
71         ref-=ofs; \
72         do { \
73                 uint32_t        t0,t1; \
74                 t0 = LPC(ref+0); \
75                 t1 = LPC(ref+4); \
76                 OP(LP(dest+0), MERGE1(t0,t1,ofs)); \
77                 t0 = LPC(ref+8); \
78                 OP(LP(dest+4), MERGE1(t1,t0,ofs)); \
79 if (sz==16) { \
80                 t1 = LPC(ref+12); \
81                 OP(LP(dest+8), MERGE1(t0,t1,ofs)); \
82                 t0 = LPC(ref+16); \
83                 OP(LP(dest+12), MERGE1(t1,t0,ofs)); \
84 } \
85                 ref+=stride; \
86                 dest+= stride; \
87         } while(--height); \
88 }
89
90 /* aligned */
91 #define         OP_C0(sz,avg2) \
92 { \
93         do { \
94                 OP(LP(dest+0), LPC(ref+0)); \
95                 OP(LP(dest+4), LPC(ref+4)); \
96 if (sz==16) { \
97                 OP(LP(dest+8), LPC(ref+8)); \
98                 OP(LP(dest+12), LPC(ref+12)); \
99 } \
100                 ref+=stride; \
101                 dest+= stride; \
102         } while(--height); \
103 }
104
105 #define         OP_X(ofs,sz,avg2) \
106 { \
107         ref-=ofs; \
108         do { \
109                 uint32_t        t0,t1; \
110                 t0 = LPC(ref+0); \
111                 t1 = LPC(ref+4); \
112                 OP(LP(dest+0), avg2(MERGE1(t0,t1,ofs),MERGE2(t0,t1,ofs))); \
113                 t0 = LPC(ref+8); \
114                 OP(LP(dest+4), avg2(MERGE1(t1,t0,ofs),MERGE2(t1,t0,ofs))); \
115 if (sz==16) { \
116                 t1 = LPC(ref+12); \
117                 OP(LP(dest+8), avg2(MERGE1(t0,t1,ofs),MERGE2(t0,t1,ofs))); \
118                 t0 = LPC(ref+16); \
119                 OP(LP(dest+12), avg2(MERGE1(t1,t0,ofs),MERGE2(t1,t0,ofs))); \
120 } \
121                 ref+=stride; \
122                 dest+= stride; \
123         } while(--height); \
124 }
125
126 /* aligned */
127 #define         OP_Y0(sz,avg2) \
128 { \
129         uint32_t t0,t1,t2,t3,t; \
130 \
131         t0 = LPC(ref+0); \
132         t1 = LPC(ref+4); \
133 if (sz==16) { \
134         t2 = LPC(ref+8); \
135         t3 = LPC(ref+12); \
136 } \
137         do { \
138                 ref += stride; \
139 \
140                 t = LPC(ref+0); \
141                 OP(LP(dest+0), avg2(t0,t)); t0 = t; \
142                 t = LPC(ref+4); \
143                 OP(LP(dest+4), avg2(t1,t)); t1 = t; \
144 if (sz==16) { \
145                 t = LPC(ref+8); \
146                 OP(LP(dest+8), avg2(t2,t)); t2 = t; \
147                 t = LPC(ref+12); \
148                 OP(LP(dest+12), avg2(t3,t)); t3 = t; \
149 } \
150                 dest+= stride; \
151         } while(--height); \
152 }
153
154 #define         OP_Y(ofs,sz,avg2) \
155 { \
156         uint32_t t0,t1,t2,t3,t,w0,w1; \
157 \
158         ref-=ofs; \
159         w0 = LPC(ref+0); \
160         w1 = LPC(ref+4); \
161         t0 = MERGE1(w0,w1,ofs); \
162         w0 = LPC(ref+8); \
163         t1 = MERGE1(w1,w0,ofs); \
164 if (sz==16) { \
165         w1 = LPC(ref+12); \
166         t2 = MERGE1(w0,w1,ofs); \
167         w0 = LPC(ref+16); \
168         t3 = MERGE1(w1,w0,ofs); \
169 } \
170         do { \
171                 ref += stride; \
172 \
173                 w0 = LPC(ref+0); \
174                 w1 = LPC(ref+4); \
175                 t = MERGE1(w0,w1,ofs); \
176                 OP(LP(dest+0), avg2(t0,t)); t0 = t; \
177                 w0 = LPC(ref+8); \
178                 t = MERGE1(w1,w0,ofs); \
179                 OP(LP(dest+4), avg2(t1,t)); t1 = t; \
180 if (sz==16) { \
181                 w1 = LPC(ref+12); \
182                 t = MERGE1(w0,w1,ofs); \
183                 OP(LP(dest+8), avg2(t2,t)); t2 = t; \
184                 w0 = LPC(ref+16); \
185                 t = MERGE1(w1,w0,ofs); \
186                 OP(LP(dest+12), avg2(t3,t)); t3 = t; \
187 } \
188                 dest+=stride; \
189         } while(--height); \
190 }
191
192 #define OP_X0(sz,avg2) OP_X(0,sz,avg2)
193 #define OP_XY0(sz,PACK) OP_XY(0,sz,PACK)
194 #define         OP_XY(ofs,sz,PACK) \
195 { \
196         uint32_t        t2,t3,w0,w1; \
197         uint32_t        a0,a1,a2,a3,a4,a5,a6,a7; \
198 \
199         ref -= ofs; \
200         w0 = LPC(ref+0); \
201         w1 = LPC(ref+4); \
202         UNPACK(a0,a1,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
203         w0 = LPC(ref+8); \
204         UNPACK(a2,a3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
205 if (sz==16) { \
206         w1 = LPC(ref+12); \
207         UNPACK(a4,a5,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
208         w0 = LPC(ref+16); \
209         UNPACK(a6,a7,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
210 } \
211         do { \
212                 ref+=stride; \
213                 w0 = LPC(ref+0); \
214                 w1 = LPC(ref+4); \
215                 UNPACK(t2,t3,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
216                 OP(LP(dest+0),PACK(a0,a1,t2,t3)); \
217                 a0 = t2; a1 = t3; \
218                 w0 = LPC(ref+8); \
219                 UNPACK(t2,t3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
220                 OP(LP(dest+4),PACK(a2,a3,t2,t3)); \
221                 a2 = t2; a3 = t3; \
222 if (sz==16) { \
223                 w1 = LPC(ref+12); \
224                 UNPACK(t2,t3,MERGE1(w0,w1,ofs),MERGE2(w0,w1,ofs)); \
225                 OP(LP(dest+8),PACK(a4,a5,t2,t3)); \
226                 a4 = t2; a5 = t3; \
227                 w0 = LPC(ref+16); \
228                 UNPACK(t2,t3,MERGE1(w1,w0,ofs),MERGE2(w1,w0,ofs)); \
229                 OP(LP(dest+12),PACK(a6,a7,t2,t3)); \
230                 a6 = t2; a7 = t3; \
231 } \
232                 dest+=stride; \
233         } while(--height); \
234 }
235
236 #define         put_pixels8_c            ff_put_rnd_pixels8_o
237 #define         put_pixels16_c           ff_put_rnd_pixels16_o
238 #define         avg_pixels8_c            ff_avg_rnd_pixels8_o
239 #define         avg_pixels16_c           ff_avg_rnd_pixels16_o
240 #define         put_no_rnd_pixels8_c     ff_put_rnd_pixels8_o
241 #define         put_no_rnd_pixels16_c    ff_put_rnd_pixels16_o
242 #define         avg_no_rnd_pixels16_c    ff_avg_rnd_pixels16_o
243
244 #if CONFIG_HPELDSP
245
246 #include "qpel.c"
247
248 #endif
249
250 av_cold void ff_dsputil_init_align(DSPContext *c, AVCodecContext *avctx)
251 {
252 #if CONFIG_HPELDSP
253
254 #define dspfunc(PFX, IDX, NUM) \
255     c->PFX ## _pixels_tab[IDX][ 0] = PFX ## NUM ## _mc00_sh4; \
256     c->PFX ## _pixels_tab[IDX][ 1] = PFX ## NUM ## _mc10_sh4; \
257     c->PFX ## _pixels_tab[IDX][ 2] = PFX ## NUM ## _mc20_sh4; \
258     c->PFX ## _pixels_tab[IDX][ 3] = PFX ## NUM ## _mc30_sh4; \
259     c->PFX ## _pixels_tab[IDX][ 4] = PFX ## NUM ## _mc01_sh4; \
260     c->PFX ## _pixels_tab[IDX][ 5] = PFX ## NUM ## _mc11_sh4; \
261     c->PFX ## _pixels_tab[IDX][ 6] = PFX ## NUM ## _mc21_sh4; \
262     c->PFX ## _pixels_tab[IDX][ 7] = PFX ## NUM ## _mc31_sh4; \
263     c->PFX ## _pixels_tab[IDX][ 8] = PFX ## NUM ## _mc02_sh4; \
264     c->PFX ## _pixels_tab[IDX][ 9] = PFX ## NUM ## _mc12_sh4; \
265     c->PFX ## _pixels_tab[IDX][10] = PFX ## NUM ## _mc22_sh4; \
266     c->PFX ## _pixels_tab[IDX][11] = PFX ## NUM ## _mc32_sh4; \
267     c->PFX ## _pixels_tab[IDX][12] = PFX ## NUM ## _mc03_sh4; \
268     c->PFX ## _pixels_tab[IDX][13] = PFX ## NUM ## _mc13_sh4; \
269     c->PFX ## _pixels_tab[IDX][14] = PFX ## NUM ## _mc23_sh4; \
270     c->PFX ## _pixels_tab[IDX][15] = PFX ## NUM ## _mc33_sh4
271
272     dspfunc(put_qpel, 0, 16);
273     dspfunc(put_no_rnd_qpel, 0, 16);
274
275     dspfunc(avg_qpel, 0, 16);
276     /* dspfunc(avg_no_rnd_qpel, 0, 16); */
277
278     dspfunc(put_qpel, 1, 8);
279     dspfunc(put_no_rnd_qpel, 1, 8);
280
281     dspfunc(avg_qpel, 1, 8);
282     /* dspfunc(avg_no_rnd_qpel, 1, 8); */
283
284 #undef dspfunc
285
286     c->put_mspel_pixels_tab[0]= put_mspel8_mc00_sh4;
287     c->put_mspel_pixels_tab[1]= put_mspel8_mc10_sh4;
288     c->put_mspel_pixels_tab[2]= put_mspel8_mc20_sh4;
289     c->put_mspel_pixels_tab[3]= put_mspel8_mc30_sh4;
290     c->put_mspel_pixels_tab[4]= put_mspel8_mc02_sh4;
291     c->put_mspel_pixels_tab[5]= put_mspel8_mc12_sh4;
292     c->put_mspel_pixels_tab[6]= put_mspel8_mc22_sh4;
293     c->put_mspel_pixels_tab[7]= put_mspel8_mc32_sh4;
294
295     c->gmc1 = gmc1_c;
296
297 #endif
298 }