IWMMXT configure support + runtime selection patch by (Gildas Bazin, gbazin : altern...
[ffmpeg.git] / libavcodec / armv4l / dsputil_arm.c
1 /*
2  * ARMv4L optimized DSP utils
3  * Copyright (c) 2001 Lionel Ulmer.
4  *
5  * This library is free software; you can redistribute it and/or
6  * modify it under the terms of the GNU Lesser General Public
7  * License as published by the Free Software Foundation; either
8  * version 2 of the License, or (at your option) any later version.
9  *
10  * This library is distributed in the hope that it will be useful,
11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13  * Lesser General Public License for more details.
14  *
15  * You should have received a copy of the GNU Lesser General Public
16  * License along with this library; if not, write to the Free Software
17  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
18  */
19
20 #include "../dsputil.h"
21 #ifdef HAVE_IPP
22 #include "ipp.h"
23 #endif
24
25 extern void dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx);
26
27 extern void j_rev_dct_ARM(DCTELEM *data);
28 extern void simple_idct_ARM(DCTELEM *data);
29
30 /* XXX: local hack */
31 static void (*ff_put_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
32 static void (*ff_add_pixels_clamped)(const DCTELEM *block, uint8_t *pixels, int line_size);
33
34 void put_pixels8_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
35 void put_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
36 void put_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
37 void put_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
38
39 void put_no_rnd_pixels8_x2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
40 void put_no_rnd_pixels8_y2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
41 void put_no_rnd_pixels8_xy2_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
42
43 void put_pixels16_arm(uint8_t *block, const uint8_t *pixels, int line_size, int h);
44
45 CALL_2X_PIXELS(put_pixels16_x2_arm , put_pixels8_x2_arm , 8)
46 CALL_2X_PIXELS(put_pixels16_y2_arm , put_pixels8_y2_arm , 8)
47 CALL_2X_PIXELS(put_pixels16_xy2_arm, put_pixels8_xy2_arm, 8)
48 CALL_2X_PIXELS(put_no_rnd_pixels16_x2_arm , put_no_rnd_pixels8_x2_arm , 8)
49 CALL_2X_PIXELS(put_no_rnd_pixels16_y2_arm , put_no_rnd_pixels8_y2_arm , 8)
50 CALL_2X_PIXELS(put_no_rnd_pixels16_xy2_arm, put_no_rnd_pixels8_xy2_arm, 8)
51
52 static void add_pixels_clamped_ARM(short *block, unsigned char *dest, int line_size)
53 {
54     asm volatile (
55                   "mov r10, #8 \n\t"
56
57                   "1: \n\t"
58
59                   /* load dest */
60                   "ldr r4, [%1] \n\t"
61                   /* block[0] and block[1]*/
62                   "ldrsh r5, [%0] \n\t"
63                   "ldrsh r7, [%0, #2] \n\t"
64                   "and r6, r4, #0xFF \n\t"
65                   "and r8, r4, #0xFF00 \n\t"
66                   "add r6, r5, r6 \n\t"
67                   "add r8, r7, r8, lsr #8 \n\t"
68                   "mvn r5, r5 \n\t"
69                   "mvn r7, r7 \n\t"
70                   "tst r6, #0x100 \n\t"
71                   "movne r6, r5, lsr #24 \n\t"
72                   "tst r8, #0x100 \n\t"
73                   "movne r8, r7, lsr #24 \n\t"
74                   "mov r9, r6 \n\t"
75                   "ldrsh r5, [%0, #4] \n\t" /* moved form [A] */
76                   "orr r9, r9, r8, lsl #8 \n\t"
77                   /* block[2] and block[3] */
78                   /* [A] */
79                   "ldrsh r7, [%0, #6] \n\t"
80                   "and r6, r4, #0xFF0000 \n\t"
81                   "and r8, r4, #0xFF000000 \n\t"
82                   "add r6, r5, r6, lsr #16 \n\t"
83                   "add r8, r7, r8, lsr #24 \n\t"
84                   "mvn r5, r5 \n\t"
85                   "mvn r7, r7 \n\t"
86                   "tst r6, #0x100 \n\t"
87                   "movne r6, r5, lsr #24 \n\t"
88                   "tst r8, #0x100 \n\t"
89                   "movne r8, r7, lsr #24 \n\t"
90                   "orr r9, r9, r6, lsl #16 \n\t"
91                   "ldr r4, [%1, #4] \n\t"       /* moved form [B] */
92                   "orr r9, r9, r8, lsl #24 \n\t"
93                   /* store dest */
94                   "ldrsh r5, [%0, #8] \n\t" /* moved form [C] */
95                   "str r9, [%1] \n\t"
96
97                   /* load dest */
98                   /* [B] */
99                   /* block[4] and block[5] */
100                   /* [C] */
101                   "ldrsh r7, [%0, #10] \n\t"
102                   "and r6, r4, #0xFF \n\t"
103                   "and r8, r4, #0xFF00 \n\t"
104                   "add r6, r5, r6 \n\t"
105                   "add r8, r7, r8, lsr #8 \n\t"
106                   "mvn r5, r5 \n\t"
107                   "mvn r7, r7 \n\t"
108                   "tst r6, #0x100 \n\t"
109                   "movne r6, r5, lsr #24 \n\t"
110                   "tst r8, #0x100 \n\t"
111                   "movne r8, r7, lsr #24 \n\t"
112                   "mov r9, r6 \n\t"
113                   "ldrsh r5, [%0, #12] \n\t" /* moved from [D] */
114                   "orr r9, r9, r8, lsl #8 \n\t"
115                   /* block[6] and block[7] */
116                   /* [D] */
117                   "ldrsh r7, [%0, #14] \n\t"
118                   "and r6, r4, #0xFF0000 \n\t"
119                   "and r8, r4, #0xFF000000 \n\t"
120                   "add r6, r5, r6, lsr #16 \n\t"
121                   "add r8, r7, r8, lsr #24 \n\t"
122                   "mvn r5, r5 \n\t"
123                   "mvn r7, r7 \n\t"
124                   "tst r6, #0x100 \n\t"
125                   "movne r6, r5, lsr #24 \n\t"
126                   "tst r8, #0x100 \n\t"
127                   "movne r8, r7, lsr #24 \n\t"
128                   "orr r9, r9, r6, lsl #16 \n\t"
129                   "add %0, %0, #16 \n\t" /* moved from [E] */
130                   "orr r9, r9, r8, lsl #24 \n\t"
131                   "subs r10, r10, #1 \n\t" /* moved from [F] */
132                   /* store dest */
133                   "str r9, [%1, #4] \n\t"
134
135                   /* [E] */
136                   /* [F] */
137                   "add %1, %1, %2 \n\t"
138                   "bne 1b \n\t"
139                   : "+r"(block),
140                     "+r"(dest)
141                   : "r"(line_size)
142                   : "r4", "r5", "r6", "r7", "r8", "r9", "r10", "cc", "memory" );
143 }
144
145 /* XXX: those functions should be suppressed ASAP when all IDCTs are
146    converted */
147 static void j_rev_dct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
148 {
149     j_rev_dct_ARM (block);
150     ff_put_pixels_clamped(block, dest, line_size);
151 }
152 static void j_rev_dct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
153 {
154     j_rev_dct_ARM (block);
155     ff_add_pixels_clamped(block, dest, line_size);
156 }
157 static void simple_idct_ARM_put(uint8_t *dest, int line_size, DCTELEM *block)
158 {
159     simple_idct_ARM (block);
160     ff_put_pixels_clamped(block, dest, line_size);
161 }
162 static void simple_idct_ARM_add(uint8_t *dest, int line_size, DCTELEM *block)
163 {
164     simple_idct_ARM (block);
165     ff_add_pixels_clamped(block, dest, line_size);
166 }
167 static void simple_idct_ipp(DCTELEM *block)
168 {
169 #ifdef HAVE_IPP
170     ippiDCT8x8Inv_Video_16s_C1I(block);
171 #endif
172 }
173 static void simple_idct_ipp_put(uint8_t *dest, int line_size, DCTELEM *block)
174 {
175 #ifdef HAVE_IPP
176     ippiDCT8x8Inv_Video_16s8u_C1R(block, dest, line_size);
177 #endif
178 }
179
180 void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size);
181
182 static void simple_idct_ipp_add(uint8_t *dest, int line_size, DCTELEM *block)
183 {
184 #ifdef HAVE_IPP
185     ippiDCT8x8Inv_Video_16s_C1I(block);
186 #ifdef HAVE_IWMMXT
187     add_pixels_clamped_iwmmxt(block, dest, line_size);
188 #else
189     add_pixels_clamped_ARM(block, dest, line_size);
190 #endif
191 #endif
192 }
193
194 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx)
195 {
196     const int idct_algo= avctx->idct_algo;
197
198     ff_put_pixels_clamped = c->put_pixels_clamped;
199     ff_add_pixels_clamped = c->add_pixels_clamped;
200
201 #ifdef HAVE_IPP
202     if(idct_algo==FF_IDCT_ARM){
203 #else
204     if(idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_ARM){
205 #endif
206         c->idct_put= j_rev_dct_ARM_put;
207         c->idct_add= j_rev_dct_ARM_add;
208         c->idct    = j_rev_dct_ARM;
209         c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;/* FF_NO_IDCT_PERM */
210     } else if (idct_algo==FF_IDCT_SIMPLEARM){
211         c->idct_put= simple_idct_ARM_put;
212         c->idct_add= simple_idct_ARM_add;
213         c->idct    = simple_idct_ARM;
214         c->idct_permutation_type= FF_NO_IDCT_PERM;
215 #ifdef HAVE_IPP
216     } else if (idct_algo==FF_IDCT_AUTO || idct_algo==FF_IDCT_IPP){
217 #else
218     } else if (idct_algo==FF_IDCT_IPP){
219 #endif
220         c->idct_put= simple_idct_ipp_put;
221         c->idct_add= simple_idct_ipp_add;
222         c->idct    = simple_idct_ipp;
223         c->idct_permutation_type= FF_NO_IDCT_PERM;
224     }
225
226 /*     c->put_pixels_tab[0][0] = put_pixels16_arm; */ // NG!
227     c->put_pixels_tab[0][1] = put_pixels16_x2_arm; //OK!
228     c->put_pixels_tab[0][2] = put_pixels16_y2_arm; //OK!
229 /*     c->put_pixels_tab[0][3] = put_pixels16_xy2_arm; /\* NG *\/ */
230 /*     c->put_no_rnd_pixels_tab[0][0] = put_pixels16_arm; // ?(»È¤ï¤ì¤Ê¤¤) */
231     c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_arm; // OK
232     c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_arm; //OK
233 /*     c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_arm; //NG */
234     c->put_pixels_tab[1][0] = put_pixels8_arm; //OK
235     c->put_pixels_tab[1][1] = put_pixels8_x2_arm; //OK
236 /*     c->put_pixels_tab[1][2] = put_pixels8_y2_arm; //NG */
237 /*     c->put_pixels_tab[1][3] = put_pixels8_xy2_arm; //NG */
238     c->put_no_rnd_pixels_tab[1][0] = put_pixels8_arm;//OK
239     c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_arm; //OK
240     c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_arm; //OK
241 /*     c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_arm;//NG */
242
243 #ifdef HAVE_IWMMXT
244     dsputil_init_iwmmxt(c, avctx);
245 #endif
246 }